15d4f98a2SYan Zheng /* 25d4f98a2SYan Zheng * Copyright (C) 2009 Oracle. All rights reserved. 35d4f98a2SYan Zheng * 45d4f98a2SYan Zheng * This program is free software; you can redistribute it and/or 55d4f98a2SYan Zheng * modify it under the terms of the GNU General Public 65d4f98a2SYan Zheng * License v2 as published by the Free Software Foundation. 75d4f98a2SYan Zheng * 85d4f98a2SYan Zheng * This program is distributed in the hope that it will be useful, 95d4f98a2SYan Zheng * but WITHOUT ANY WARRANTY; without even the implied warranty of 105d4f98a2SYan Zheng * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 115d4f98a2SYan Zheng * General Public License for more details. 125d4f98a2SYan Zheng * 135d4f98a2SYan Zheng * You should have received a copy of the GNU General Public 145d4f98a2SYan Zheng * License along with this program; if not, write to the 155d4f98a2SYan Zheng * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 165d4f98a2SYan Zheng * Boston, MA 021110-1307, USA. 175d4f98a2SYan Zheng */ 185d4f98a2SYan Zheng 195d4f98a2SYan Zheng #include <linux/sched.h> 205d4f98a2SYan Zheng #include <linux/pagemap.h> 215d4f98a2SYan Zheng #include <linux/writeback.h> 225d4f98a2SYan Zheng #include <linux/blkdev.h> 235d4f98a2SYan Zheng #include <linux/rbtree.h> 245a0e3ad6STejun Heo #include <linux/slab.h> 255d4f98a2SYan Zheng #include "ctree.h" 265d4f98a2SYan Zheng #include "disk-io.h" 275d4f98a2SYan Zheng #include "transaction.h" 285d4f98a2SYan Zheng #include "volumes.h" 295d4f98a2SYan Zheng #include "locking.h" 305d4f98a2SYan Zheng #include "btrfs_inode.h" 315d4f98a2SYan Zheng #include "async-thread.h" 320af3d00bSJosef Bacik #include "free-space-cache.h" 33581bb050SLi Zefan #include "inode-map.h" 3462b99540SQu Wenruo #include "qgroup.h" 35cdccee99SLiu Bo #include "print-tree.h" 365d4f98a2SYan Zheng 375d4f98a2SYan Zheng /* 385d4f98a2SYan Zheng * backref_node, mapping_node and tree_block start with this 395d4f98a2SYan Zheng */ 405d4f98a2SYan Zheng struct tree_entry { 415d4f98a2SYan Zheng struct rb_node rb_node; 425d4f98a2SYan Zheng u64 bytenr; 435d4f98a2SYan Zheng }; 445d4f98a2SYan Zheng 455d4f98a2SYan Zheng /* 465d4f98a2SYan Zheng * present a tree block in the backref cache 475d4f98a2SYan Zheng */ 485d4f98a2SYan Zheng struct backref_node { 495d4f98a2SYan Zheng struct rb_node rb_node; 505d4f98a2SYan Zheng u64 bytenr; 513fd0a558SYan, Zheng 523fd0a558SYan, Zheng u64 new_bytenr; 533fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 545d4f98a2SYan Zheng u64 owner; 553fd0a558SYan, Zheng /* link to pending, changed or detached list */ 563fd0a558SYan, Zheng struct list_head list; 575d4f98a2SYan Zheng /* list of upper level blocks reference this block */ 585d4f98a2SYan Zheng struct list_head upper; 595d4f98a2SYan Zheng /* list of child blocks in the cache */ 605d4f98a2SYan Zheng struct list_head lower; 615d4f98a2SYan Zheng /* NULL if this node is not tree root */ 625d4f98a2SYan Zheng struct btrfs_root *root; 635d4f98a2SYan Zheng /* extent buffer got by COW the block */ 645d4f98a2SYan Zheng struct extent_buffer *eb; 655d4f98a2SYan Zheng /* level of tree block */ 665d4f98a2SYan Zheng unsigned int level:8; 673fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 683fd0a558SYan, Zheng unsigned int cowonly:1; 693fd0a558SYan, Zheng /* 1 if no child node in the cache */ 705d4f98a2SYan Zheng unsigned int lowest:1; 715d4f98a2SYan Zheng /* is the extent buffer locked */ 725d4f98a2SYan Zheng unsigned int locked:1; 735d4f98a2SYan Zheng /* has the block been processed */ 745d4f98a2SYan Zheng unsigned int processed:1; 755d4f98a2SYan Zheng /* have backrefs of this block been checked */ 765d4f98a2SYan Zheng unsigned int checked:1; 773fd0a558SYan, Zheng /* 783fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 793fd0a558SYan, Zheng * level block pointers may not point to the new location 803fd0a558SYan, Zheng */ 813fd0a558SYan, Zheng unsigned int pending:1; 823fd0a558SYan, Zheng /* 833fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 843fd0a558SYan, Zheng * backref node. 853fd0a558SYan, Zheng */ 863fd0a558SYan, Zheng unsigned int detached:1; 875d4f98a2SYan Zheng }; 885d4f98a2SYan Zheng 895d4f98a2SYan Zheng /* 905d4f98a2SYan Zheng * present a block pointer in the backref cache 915d4f98a2SYan Zheng */ 925d4f98a2SYan Zheng struct backref_edge { 935d4f98a2SYan Zheng struct list_head list[2]; 945d4f98a2SYan Zheng struct backref_node *node[2]; 955d4f98a2SYan Zheng }; 965d4f98a2SYan Zheng 975d4f98a2SYan Zheng #define LOWER 0 985d4f98a2SYan Zheng #define UPPER 1 990647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 1005d4f98a2SYan Zheng 1015d4f98a2SYan Zheng struct backref_cache { 1025d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 1035d4f98a2SYan Zheng struct rb_root rb_root; 1043fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 1053fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 1063fd0a558SYan, Zheng /* 1073fd0a558SYan, Zheng * list of blocks that have been cowed but some block 1083fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 1093fd0a558SYan, Zheng * new location 1103fd0a558SYan, Zheng */ 1115d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1123fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1133fd0a558SYan, Zheng struct list_head leaves; 1143fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1153fd0a558SYan, Zheng struct list_head changed; 1163fd0a558SYan, Zheng /* list of detached backref node. */ 1173fd0a558SYan, Zheng struct list_head detached; 1183fd0a558SYan, Zheng 1193fd0a558SYan, Zheng u64 last_trans; 1203fd0a558SYan, Zheng 1213fd0a558SYan, Zheng int nr_nodes; 1223fd0a558SYan, Zheng int nr_edges; 1235d4f98a2SYan Zheng }; 1245d4f98a2SYan Zheng 1255d4f98a2SYan Zheng /* 1265d4f98a2SYan Zheng * map address of tree root to tree 1275d4f98a2SYan Zheng */ 1285d4f98a2SYan Zheng struct mapping_node { 1295d4f98a2SYan Zheng struct rb_node rb_node; 1305d4f98a2SYan Zheng u64 bytenr; 1315d4f98a2SYan Zheng void *data; 1325d4f98a2SYan Zheng }; 1335d4f98a2SYan Zheng 1345d4f98a2SYan Zheng struct mapping_tree { 1355d4f98a2SYan Zheng struct rb_root rb_root; 1365d4f98a2SYan Zheng spinlock_t lock; 1375d4f98a2SYan Zheng }; 1385d4f98a2SYan Zheng 1395d4f98a2SYan Zheng /* 1405d4f98a2SYan Zheng * present a tree block to process 1415d4f98a2SYan Zheng */ 1425d4f98a2SYan Zheng struct tree_block { 1435d4f98a2SYan Zheng struct rb_node rb_node; 1445d4f98a2SYan Zheng u64 bytenr; 1455d4f98a2SYan Zheng struct btrfs_key key; 1465d4f98a2SYan Zheng unsigned int level:8; 1475d4f98a2SYan Zheng unsigned int key_ready:1; 1485d4f98a2SYan Zheng }; 1495d4f98a2SYan Zheng 1500257bb82SYan, Zheng #define MAX_EXTENTS 128 1510257bb82SYan, Zheng 1520257bb82SYan, Zheng struct file_extent_cluster { 1530257bb82SYan, Zheng u64 start; 1540257bb82SYan, Zheng u64 end; 1550257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1560257bb82SYan, Zheng unsigned int nr; 1570257bb82SYan, Zheng }; 1580257bb82SYan, Zheng 1595d4f98a2SYan Zheng struct reloc_control { 1605d4f98a2SYan Zheng /* block group to relocate */ 1615d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group; 1625d4f98a2SYan Zheng /* extent tree */ 1635d4f98a2SYan Zheng struct btrfs_root *extent_root; 1645d4f98a2SYan Zheng /* inode for moving data */ 1655d4f98a2SYan Zheng struct inode *data_inode; 1663fd0a558SYan, Zheng 1673fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1683fd0a558SYan, Zheng 1693fd0a558SYan, Zheng struct backref_cache backref_cache; 1703fd0a558SYan, Zheng 1713fd0a558SYan, Zheng struct file_extent_cluster cluster; 1725d4f98a2SYan Zheng /* tree blocks have been processed */ 1735d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1745d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1755d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1765d4f98a2SYan Zheng /* list of reloc trees */ 1775d4f98a2SYan Zheng struct list_head reloc_roots; 1783fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1793fd0a558SYan, Zheng u64 merging_rsv_size; 1803fd0a558SYan, Zheng /* size of relocated tree nodes */ 1813fd0a558SYan, Zheng u64 nodes_relocated; 1820647bf56SWang Shilong /* reserved size for block group relocation*/ 1830647bf56SWang Shilong u64 reserved_bytes; 1843fd0a558SYan, Zheng 1855d4f98a2SYan Zheng u64 search_start; 1865d4f98a2SYan Zheng u64 extents_found; 1873fd0a558SYan, Zheng 1883fd0a558SYan, Zheng unsigned int stage:8; 1893fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1903fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1915d4f98a2SYan Zheng unsigned int found_file_extent:1; 1925d4f98a2SYan Zheng }; 1935d4f98a2SYan Zheng 1945d4f98a2SYan Zheng /* stages of data relocation */ 1955d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1965d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1975d4f98a2SYan Zheng 1983fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 1993fd0a558SYan, Zheng struct backref_node *node); 2003fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 2013fd0a558SYan, Zheng struct backref_node *node); 2025d4f98a2SYan Zheng 2035d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 2045d4f98a2SYan Zheng { 2056bef4d31SEric Paris tree->rb_root = RB_ROOT; 2065d4f98a2SYan Zheng spin_lock_init(&tree->lock); 2075d4f98a2SYan Zheng } 2085d4f98a2SYan Zheng 2095d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 2105d4f98a2SYan Zheng { 2115d4f98a2SYan Zheng int i; 2126bef4d31SEric Paris cache->rb_root = RB_ROOT; 2135d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2145d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2153fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2163fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2173fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2185d4f98a2SYan Zheng } 2195d4f98a2SYan Zheng 2203fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2215d4f98a2SYan Zheng { 2223fd0a558SYan, Zheng struct backref_node *node; 2233fd0a558SYan, Zheng int i; 2243fd0a558SYan, Zheng 2253fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2263fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2273fd0a558SYan, Zheng struct backref_node, list); 2283fd0a558SYan, Zheng remove_backref_node(cache, node); 2293fd0a558SYan, Zheng } 2303fd0a558SYan, Zheng 2313fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2323fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2333fd0a558SYan, Zheng struct backref_node, lower); 2343fd0a558SYan, Zheng remove_backref_node(cache, node); 2353fd0a558SYan, Zheng } 2363fd0a558SYan, Zheng 2373fd0a558SYan, Zheng cache->last_trans = 0; 2383fd0a558SYan, Zheng 2393fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 240f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 241f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 242f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 243f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 244f4907095SLiu Bo ASSERT(!cache->nr_nodes); 245f4907095SLiu Bo ASSERT(!cache->nr_edges); 2463fd0a558SYan, Zheng } 2473fd0a558SYan, Zheng 2483fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2493fd0a558SYan, Zheng { 2503fd0a558SYan, Zheng struct backref_node *node; 2513fd0a558SYan, Zheng 2523fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2533fd0a558SYan, Zheng if (node) { 2543fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2555d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2565d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2575d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2583fd0a558SYan, Zheng cache->nr_nodes++; 2593fd0a558SYan, Zheng } 2603fd0a558SYan, Zheng return node; 2613fd0a558SYan, Zheng } 2623fd0a558SYan, Zheng 2633fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 2643fd0a558SYan, Zheng struct backref_node *node) 2653fd0a558SYan, Zheng { 2663fd0a558SYan, Zheng if (node) { 2673fd0a558SYan, Zheng cache->nr_nodes--; 2683fd0a558SYan, Zheng kfree(node); 2693fd0a558SYan, Zheng } 2703fd0a558SYan, Zheng } 2713fd0a558SYan, Zheng 2723fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 2733fd0a558SYan, Zheng { 2743fd0a558SYan, Zheng struct backref_edge *edge; 2753fd0a558SYan, Zheng 2763fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 2773fd0a558SYan, Zheng if (edge) 2783fd0a558SYan, Zheng cache->nr_edges++; 2793fd0a558SYan, Zheng return edge; 2803fd0a558SYan, Zheng } 2813fd0a558SYan, Zheng 2823fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 2833fd0a558SYan, Zheng struct backref_edge *edge) 2843fd0a558SYan, Zheng { 2853fd0a558SYan, Zheng if (edge) { 2863fd0a558SYan, Zheng cache->nr_edges--; 2873fd0a558SYan, Zheng kfree(edge); 2883fd0a558SYan, Zheng } 2895d4f98a2SYan Zheng } 2905d4f98a2SYan Zheng 2915d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 2925d4f98a2SYan Zheng struct rb_node *node) 2935d4f98a2SYan Zheng { 2945d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 2955d4f98a2SYan Zheng struct rb_node *parent = NULL; 2965d4f98a2SYan Zheng struct tree_entry *entry; 2975d4f98a2SYan Zheng 2985d4f98a2SYan Zheng while (*p) { 2995d4f98a2SYan Zheng parent = *p; 3005d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 3015d4f98a2SYan Zheng 3025d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3035d4f98a2SYan Zheng p = &(*p)->rb_left; 3045d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3055d4f98a2SYan Zheng p = &(*p)->rb_right; 3065d4f98a2SYan Zheng else 3075d4f98a2SYan Zheng return parent; 3085d4f98a2SYan Zheng } 3095d4f98a2SYan Zheng 3105d4f98a2SYan Zheng rb_link_node(node, parent, p); 3115d4f98a2SYan Zheng rb_insert_color(node, root); 3125d4f98a2SYan Zheng return NULL; 3135d4f98a2SYan Zheng } 3145d4f98a2SYan Zheng 3155d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3165d4f98a2SYan Zheng { 3175d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3185d4f98a2SYan Zheng struct tree_entry *entry; 3195d4f98a2SYan Zheng 3205d4f98a2SYan Zheng while (n) { 3215d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3225d4f98a2SYan Zheng 3235d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3245d4f98a2SYan Zheng n = n->rb_left; 3255d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3265d4f98a2SYan Zheng n = n->rb_right; 3275d4f98a2SYan Zheng else 3285d4f98a2SYan Zheng return n; 3295d4f98a2SYan Zheng } 3305d4f98a2SYan Zheng return NULL; 3315d4f98a2SYan Zheng } 3325d4f98a2SYan Zheng 33348a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 33443c04fb1SJeff Mahoney { 33543c04fb1SJeff Mahoney 33643c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 33743c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 33843c04fb1SJeff Mahoney rb_node); 33943c04fb1SJeff Mahoney if (bnode->root) 34043c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 3415d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 3425d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 3435d163e0eSJeff Mahoney bytenr); 34443c04fb1SJeff Mahoney } 34543c04fb1SJeff Mahoney 3465d4f98a2SYan Zheng /* 3475d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3485d4f98a2SYan Zheng */ 3495d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3505d4f98a2SYan Zheng struct backref_edge *edges[], 3515d4f98a2SYan Zheng int *index) 3525d4f98a2SYan Zheng { 3535d4f98a2SYan Zheng struct backref_edge *edge; 3545d4f98a2SYan Zheng int idx = *index; 3555d4f98a2SYan Zheng 3565d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3575d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3585d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3595d4f98a2SYan Zheng edges[idx++] = edge; 3605d4f98a2SYan Zheng node = edge->node[UPPER]; 3615d4f98a2SYan Zheng } 3623fd0a558SYan, Zheng BUG_ON(node->detached); 3635d4f98a2SYan Zheng *index = idx; 3645d4f98a2SYan Zheng return node; 3655d4f98a2SYan Zheng } 3665d4f98a2SYan Zheng 3675d4f98a2SYan Zheng /* 3685d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 3695d4f98a2SYan Zheng */ 3705d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 3715d4f98a2SYan Zheng int *index) 3725d4f98a2SYan Zheng { 3735d4f98a2SYan Zheng struct backref_edge *edge; 3745d4f98a2SYan Zheng struct backref_node *lower; 3755d4f98a2SYan Zheng int idx = *index; 3765d4f98a2SYan Zheng 3775d4f98a2SYan Zheng while (idx > 0) { 3785d4f98a2SYan Zheng edge = edges[idx - 1]; 3795d4f98a2SYan Zheng lower = edge->node[LOWER]; 3805d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 3815d4f98a2SYan Zheng idx--; 3825d4f98a2SYan Zheng continue; 3835d4f98a2SYan Zheng } 3845d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 3855d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3865d4f98a2SYan Zheng edges[idx - 1] = edge; 3875d4f98a2SYan Zheng *index = idx; 3885d4f98a2SYan Zheng return edge->node[UPPER]; 3895d4f98a2SYan Zheng } 3905d4f98a2SYan Zheng *index = 0; 3915d4f98a2SYan Zheng return NULL; 3925d4f98a2SYan Zheng } 3935d4f98a2SYan Zheng 3943fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 3955d4f98a2SYan Zheng { 3965d4f98a2SYan Zheng if (node->locked) { 3975d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 3985d4f98a2SYan Zheng node->locked = 0; 3995d4f98a2SYan Zheng } 4003fd0a558SYan, Zheng } 4013fd0a558SYan, Zheng 4023fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 4033fd0a558SYan, Zheng { 4043fd0a558SYan, Zheng if (node->eb) { 4053fd0a558SYan, Zheng unlock_node_buffer(node); 4065d4f98a2SYan Zheng free_extent_buffer(node->eb); 4075d4f98a2SYan Zheng node->eb = NULL; 4085d4f98a2SYan Zheng } 4095d4f98a2SYan Zheng } 4105d4f98a2SYan Zheng 4115d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4125d4f98a2SYan Zheng struct backref_node *node) 4135d4f98a2SYan Zheng { 4145d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4155d4f98a2SYan Zheng 4165d4f98a2SYan Zheng drop_node_buffer(node); 4173fd0a558SYan, Zheng list_del(&node->list); 4185d4f98a2SYan Zheng list_del(&node->lower); 4193fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4205d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4213fd0a558SYan, Zheng free_backref_node(tree, node); 4225d4f98a2SYan Zheng } 4235d4f98a2SYan Zheng 4245d4f98a2SYan Zheng /* 4255d4f98a2SYan Zheng * remove a backref node from the backref cache 4265d4f98a2SYan Zheng */ 4275d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4285d4f98a2SYan Zheng struct backref_node *node) 4295d4f98a2SYan Zheng { 4305d4f98a2SYan Zheng struct backref_node *upper; 4315d4f98a2SYan Zheng struct backref_edge *edge; 4325d4f98a2SYan Zheng 4335d4f98a2SYan Zheng if (!node) 4345d4f98a2SYan Zheng return; 4355d4f98a2SYan Zheng 4363fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4375d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4385d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4395d4f98a2SYan Zheng list[LOWER]); 4405d4f98a2SYan Zheng upper = edge->node[UPPER]; 4415d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4425d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4433fd0a558SYan, Zheng free_backref_edge(cache, edge); 4443fd0a558SYan, Zheng 4453fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4463fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4473fd0a558SYan, Zheng drop_backref_node(cache, node); 4483fd0a558SYan, Zheng node = upper; 4493fd0a558SYan, Zheng node->lowest = 1; 4503fd0a558SYan, Zheng continue; 4513fd0a558SYan, Zheng } 4525d4f98a2SYan Zheng /* 4533fd0a558SYan, Zheng * add the node to leaf node list if no other 4545d4f98a2SYan Zheng * child block cached. 4555d4f98a2SYan Zheng */ 4565d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4573fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4585d4f98a2SYan Zheng upper->lowest = 1; 4595d4f98a2SYan Zheng } 4605d4f98a2SYan Zheng } 4613fd0a558SYan, Zheng 4625d4f98a2SYan Zheng drop_backref_node(cache, node); 4635d4f98a2SYan Zheng } 4645d4f98a2SYan Zheng 4653fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 4663fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 4673fd0a558SYan, Zheng { 4683fd0a558SYan, Zheng struct rb_node *rb_node; 4693fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 4703fd0a558SYan, Zheng node->bytenr = bytenr; 4713fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 47243c04fb1SJeff Mahoney if (rb_node) 47343c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 4743fd0a558SYan, Zheng } 4753fd0a558SYan, Zheng 4763fd0a558SYan, Zheng /* 4773fd0a558SYan, Zheng * update backref cache after a transaction commit 4783fd0a558SYan, Zheng */ 4793fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 4803fd0a558SYan, Zheng struct backref_cache *cache) 4813fd0a558SYan, Zheng { 4823fd0a558SYan, Zheng struct backref_node *node; 4833fd0a558SYan, Zheng int level = 0; 4843fd0a558SYan, Zheng 4853fd0a558SYan, Zheng if (cache->last_trans == 0) { 4863fd0a558SYan, Zheng cache->last_trans = trans->transid; 4873fd0a558SYan, Zheng return 0; 4883fd0a558SYan, Zheng } 4893fd0a558SYan, Zheng 4903fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 4913fd0a558SYan, Zheng return 0; 4923fd0a558SYan, Zheng 4933fd0a558SYan, Zheng /* 4943fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 4953fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 4963fd0a558SYan, Zheng * so the detached nodes are no longer useful. 4973fd0a558SYan, Zheng */ 4983fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 4993fd0a558SYan, Zheng node = list_entry(cache->detached.next, 5003fd0a558SYan, Zheng struct backref_node, list); 5013fd0a558SYan, Zheng remove_backref_node(cache, node); 5023fd0a558SYan, Zheng } 5033fd0a558SYan, Zheng 5043fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 5053fd0a558SYan, Zheng node = list_entry(cache->changed.next, 5063fd0a558SYan, Zheng struct backref_node, list); 5073fd0a558SYan, Zheng list_del_init(&node->list); 5083fd0a558SYan, Zheng BUG_ON(node->pending); 5093fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5103fd0a558SYan, Zheng } 5113fd0a558SYan, Zheng 5123fd0a558SYan, Zheng /* 5133fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5143fd0a558SYan, Zheng * errors during processing the pending nodes. 5153fd0a558SYan, Zheng */ 5163fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5173fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5183fd0a558SYan, Zheng BUG_ON(!node->pending); 5193fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5203fd0a558SYan, Zheng continue; 5213fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5223fd0a558SYan, Zheng } 5233fd0a558SYan, Zheng } 5243fd0a558SYan, Zheng 5253fd0a558SYan, Zheng cache->last_trans = 0; 5263fd0a558SYan, Zheng return 1; 5273fd0a558SYan, Zheng } 5283fd0a558SYan, Zheng 529f2a97a9dSDavid Sterba 5303fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5313fd0a558SYan, Zheng { 5323fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5333fd0a558SYan, Zheng 53427cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 5353fd0a558SYan, Zheng return 0; 5363fd0a558SYan, Zheng 5373fd0a558SYan, Zheng reloc_root = root->reloc_root; 5383fd0a558SYan, Zheng if (!reloc_root) 5393fd0a558SYan, Zheng return 0; 5403fd0a558SYan, Zheng 5413fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5423fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5433fd0a558SYan, Zheng return 0; 5443fd0a558SYan, Zheng /* 5453fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5463fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5473fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5483fd0a558SYan, Zheng * relocation. 5493fd0a558SYan, Zheng */ 5503fd0a558SYan, Zheng return 1; 5513fd0a558SYan, Zheng } 5525d4f98a2SYan Zheng /* 5535d4f98a2SYan Zheng * find reloc tree by address of tree root 5545d4f98a2SYan Zheng */ 5555d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5565d4f98a2SYan Zheng u64 bytenr) 5575d4f98a2SYan Zheng { 5585d4f98a2SYan Zheng struct rb_node *rb_node; 5595d4f98a2SYan Zheng struct mapping_node *node; 5605d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5615d4f98a2SYan Zheng 5625d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5635d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5645d4f98a2SYan Zheng if (rb_node) { 5655d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5665d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5675d4f98a2SYan Zheng } 5685d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5695d4f98a2SYan Zheng return root; 5705d4f98a2SYan Zheng } 5715d4f98a2SYan Zheng 5725d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5735d4f98a2SYan Zheng { 5745d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5755d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5765d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 5775d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 5785d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 57966463748SWang Shilong root_objectid == BTRFS_CSUM_TREE_OBJECTID || 58066463748SWang Shilong root_objectid == BTRFS_UUID_TREE_OBJECTID || 5813e4c5efbSDavid Sterba root_objectid == BTRFS_QUOTA_TREE_OBJECTID || 5823e4c5efbSDavid Sterba root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 5835d4f98a2SYan Zheng return 1; 5845d4f98a2SYan Zheng return 0; 5855d4f98a2SYan Zheng } 5865d4f98a2SYan Zheng 5875d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 5885d4f98a2SYan Zheng u64 root_objectid) 5895d4f98a2SYan Zheng { 5905d4f98a2SYan Zheng struct btrfs_key key; 5915d4f98a2SYan Zheng 5925d4f98a2SYan Zheng key.objectid = root_objectid; 5935d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 5945d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 5955d4f98a2SYan Zheng key.offset = 0; 5965d4f98a2SYan Zheng else 5975d4f98a2SYan Zheng key.offset = (u64)-1; 5985d4f98a2SYan Zheng 599c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 6005d4f98a2SYan Zheng } 6015d4f98a2SYan Zheng 6025d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 6035d4f98a2SYan Zheng static noinline_for_stack 6045d4f98a2SYan Zheng struct btrfs_root *find_tree_root(struct reloc_control *rc, 6055d4f98a2SYan Zheng struct extent_buffer *leaf, 6065d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0) 6075d4f98a2SYan Zheng { 6085d4f98a2SYan Zheng struct btrfs_root *root; 6095d4f98a2SYan Zheng u64 root_objectid = btrfs_ref_root_v0(leaf, ref0); 6105d4f98a2SYan Zheng u64 generation = btrfs_ref_generation_v0(leaf, ref0); 6115d4f98a2SYan Zheng 6125d4f98a2SYan Zheng BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID); 6135d4f98a2SYan Zheng 6145d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, root_objectid); 6155d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 6165d4f98a2SYan Zheng 61727cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && 6185d4f98a2SYan Zheng generation != btrfs_root_generation(&root->root_item)) 6195d4f98a2SYan Zheng return NULL; 6205d4f98a2SYan Zheng 6215d4f98a2SYan Zheng return root; 6225d4f98a2SYan Zheng } 6235d4f98a2SYan Zheng #endif 6245d4f98a2SYan Zheng 6255d4f98a2SYan Zheng static noinline_for_stack 6265d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 6275d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 6285d4f98a2SYan Zheng { 6293173a18fSJosef Bacik struct btrfs_key key; 6305d4f98a2SYan Zheng struct btrfs_extent_item *ei; 6315d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6325d4f98a2SYan Zheng u32 item_size; 6335d4f98a2SYan Zheng 6343173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6353173a18fSJosef Bacik 6365d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 6375d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 6385d4f98a2SYan Zheng if (item_size < sizeof(*ei)) { 6395d4f98a2SYan Zheng WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 6405d4f98a2SYan Zheng return 1; 6415d4f98a2SYan Zheng } 6425d4f98a2SYan Zheng #endif 6435d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6445d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6455d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6465d4f98a2SYan Zheng 6473173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6483173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6495d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6505d4f98a2SYan Zheng return 1; 6515d4f98a2SYan Zheng } 652d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 653d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 654d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 655d062d13cSJosef Bacik return 1; 656d062d13cSJosef Bacik } 6575d4f98a2SYan Zheng 6583173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6595d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6605d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6613173a18fSJosef Bacik } else { 6623173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6633173a18fSJosef Bacik } 6645d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6655d4f98a2SYan Zheng return 0; 6665d4f98a2SYan Zheng } 6675d4f98a2SYan Zheng 6685d4f98a2SYan Zheng /* 6695d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6705d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6715d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6725d4f98a2SYan Zheng * 6735d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 67401327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 67501327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 6765d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6775d4f98a2SYan Zheng * 6785d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6795d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6805d4f98a2SYan Zheng * block are also cached. 6815d4f98a2SYan Zheng */ 6823fd0a558SYan, Zheng static noinline_for_stack 6833fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6845d4f98a2SYan Zheng struct btrfs_key *node_key, 6855d4f98a2SYan Zheng int level, u64 bytenr) 6865d4f98a2SYan Zheng { 6873fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 6885d4f98a2SYan Zheng struct btrfs_path *path1; 6895d4f98a2SYan Zheng struct btrfs_path *path2; 6905d4f98a2SYan Zheng struct extent_buffer *eb; 6915d4f98a2SYan Zheng struct btrfs_root *root; 6925d4f98a2SYan Zheng struct backref_node *cur; 6935d4f98a2SYan Zheng struct backref_node *upper; 6945d4f98a2SYan Zheng struct backref_node *lower; 6955d4f98a2SYan Zheng struct backref_node *node = NULL; 6965d4f98a2SYan Zheng struct backref_node *exist = NULL; 6975d4f98a2SYan Zheng struct backref_edge *edge; 6985d4f98a2SYan Zheng struct rb_node *rb_node; 6995d4f98a2SYan Zheng struct btrfs_key key; 7005d4f98a2SYan Zheng unsigned long end; 7015d4f98a2SYan Zheng unsigned long ptr; 7025d4f98a2SYan Zheng LIST_HEAD(list); 7033fd0a558SYan, Zheng LIST_HEAD(useless); 7043fd0a558SYan, Zheng int cowonly; 7055d4f98a2SYan Zheng int ret; 7065d4f98a2SYan Zheng int err = 0; 707b6c60c80SJosef Bacik bool need_check = true; 7085d4f98a2SYan Zheng 7095d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 7105d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 7115d4f98a2SYan Zheng if (!path1 || !path2) { 7125d4f98a2SYan Zheng err = -ENOMEM; 7135d4f98a2SYan Zheng goto out; 7145d4f98a2SYan Zheng } 715e4058b54SDavid Sterba path1->reada = READA_FORWARD; 716e4058b54SDavid Sterba path2->reada = READA_FORWARD; 7175d4f98a2SYan Zheng 7183fd0a558SYan, Zheng node = alloc_backref_node(cache); 7195d4f98a2SYan Zheng if (!node) { 7205d4f98a2SYan Zheng err = -ENOMEM; 7215d4f98a2SYan Zheng goto out; 7225d4f98a2SYan Zheng } 7235d4f98a2SYan Zheng 7245d4f98a2SYan Zheng node->bytenr = bytenr; 7255d4f98a2SYan Zheng node->level = level; 7265d4f98a2SYan Zheng node->lowest = 1; 7275d4f98a2SYan Zheng cur = node; 7285d4f98a2SYan Zheng again: 7295d4f98a2SYan Zheng end = 0; 7305d4f98a2SYan Zheng ptr = 0; 7315d4f98a2SYan Zheng key.objectid = cur->bytenr; 7323173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7335d4f98a2SYan Zheng key.offset = (u64)-1; 7345d4f98a2SYan Zheng 7355d4f98a2SYan Zheng path1->search_commit_root = 1; 7365d4f98a2SYan Zheng path1->skip_locking = 1; 7375d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7385d4f98a2SYan Zheng 0, 0); 7395d4f98a2SYan Zheng if (ret < 0) { 7405d4f98a2SYan Zheng err = ret; 7415d4f98a2SYan Zheng goto out; 7425d4f98a2SYan Zheng } 74375bfb9afSJosef Bacik ASSERT(ret); 74475bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7455d4f98a2SYan Zheng 7465d4f98a2SYan Zheng path1->slots[0]--; 7475d4f98a2SYan Zheng 7485d4f98a2SYan Zheng WARN_ON(cur->checked); 7495d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7505d4f98a2SYan Zheng /* 75170f23fd6SJustin P. Mattock * the backref was added previously when processing 7525d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7535d4f98a2SYan Zheng */ 75475bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7555d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7565d4f98a2SYan Zheng list[LOWER]); 75775bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7585d4f98a2SYan Zheng exist = edge->node[UPPER]; 7595d4f98a2SYan Zheng /* 7605d4f98a2SYan Zheng * add the upper level block to pending list if we need 7615d4f98a2SYan Zheng * check its backrefs 7625d4f98a2SYan Zheng */ 7635d4f98a2SYan Zheng if (!exist->checked) 7645d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7655d4f98a2SYan Zheng } else { 7665d4f98a2SYan Zheng exist = NULL; 7675d4f98a2SYan Zheng } 7685d4f98a2SYan Zheng 7695d4f98a2SYan Zheng while (1) { 7705d4f98a2SYan Zheng cond_resched(); 7715d4f98a2SYan Zheng eb = path1->nodes[0]; 7725d4f98a2SYan Zheng 7735d4f98a2SYan Zheng if (ptr >= end) { 7745d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7755d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7765d4f98a2SYan Zheng if (ret < 0) { 7775d4f98a2SYan Zheng err = ret; 7785d4f98a2SYan Zheng goto out; 7795d4f98a2SYan Zheng } 7805d4f98a2SYan Zheng if (ret > 0) 7815d4f98a2SYan Zheng break; 7825d4f98a2SYan Zheng eb = path1->nodes[0]; 7835d4f98a2SYan Zheng } 7845d4f98a2SYan Zheng 7855d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7865d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7875d4f98a2SYan Zheng WARN_ON(exist); 7885d4f98a2SYan Zheng break; 7895d4f98a2SYan Zheng } 7905d4f98a2SYan Zheng 7913173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7923173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7935d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7945d4f98a2SYan Zheng &ptr, &end); 7955d4f98a2SYan Zheng if (ret) 7965d4f98a2SYan Zheng goto next; 7975d4f98a2SYan Zheng } 7985d4f98a2SYan Zheng } 7995d4f98a2SYan Zheng 8005d4f98a2SYan Zheng if (ptr < end) { 8015d4f98a2SYan Zheng /* update key for inline back ref */ 8025d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 8033de28d57SLiu Bo int type; 8045d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 8053de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 8063de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 8073de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 8083de28d57SLiu Bo err = -EINVAL; 8093de28d57SLiu Bo goto out; 8103de28d57SLiu Bo } 8113de28d57SLiu Bo key.type = type; 8125d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 8133de28d57SLiu Bo 8145d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 8155d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 8165d4f98a2SYan Zheng } 8175d4f98a2SYan Zheng 8185d4f98a2SYan Zheng if (exist && 8195d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8205d4f98a2SYan Zheng exist->owner == key.offset) || 8215d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8225d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8235d4f98a2SYan Zheng exist = NULL; 8245d4f98a2SYan Zheng goto next; 8255d4f98a2SYan Zheng } 8265d4f98a2SYan Zheng 8275d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 8285d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY || 8295d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 8303fd0a558SYan, Zheng if (key.type == BTRFS_EXTENT_REF_V0_KEY) { 8315d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 8325d4f98a2SYan Zheng ref0 = btrfs_item_ptr(eb, path1->slots[0], 8335d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 8343fd0a558SYan, Zheng if (key.objectid == key.offset) { 835046f264fSYan, Zheng root = find_tree_root(rc, eb, ref0); 8363fd0a558SYan, Zheng if (root && !should_ignore_root(root)) 8375d4f98a2SYan Zheng cur->root = root; 8385d4f98a2SYan Zheng else 8393fd0a558SYan, Zheng list_add(&cur->list, &useless); 8405d4f98a2SYan Zheng break; 8415d4f98a2SYan Zheng } 842046f264fSYan, Zheng if (is_cowonly_root(btrfs_ref_root_v0(eb, 843046f264fSYan, Zheng ref0))) 844046f264fSYan, Zheng cur->cowonly = 1; 8453fd0a558SYan, Zheng } 8465d4f98a2SYan Zheng #else 84775bfb9afSJosef Bacik ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY); 8485d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8495d4f98a2SYan Zheng #endif 8505d4f98a2SYan Zheng if (key.objectid == key.offset) { 8515d4f98a2SYan Zheng /* 8525d4f98a2SYan Zheng * only root blocks of reloc trees use 8535d4f98a2SYan Zheng * backref of this type. 8545d4f98a2SYan Zheng */ 8555d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 85675bfb9afSJosef Bacik ASSERT(root); 8575d4f98a2SYan Zheng cur->root = root; 8585d4f98a2SYan Zheng break; 8595d4f98a2SYan Zheng } 8605d4f98a2SYan Zheng 8613fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8625d4f98a2SYan Zheng if (!edge) { 8635d4f98a2SYan Zheng err = -ENOMEM; 8645d4f98a2SYan Zheng goto out; 8655d4f98a2SYan Zheng } 8665d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8675d4f98a2SYan Zheng if (!rb_node) { 8683fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8695d4f98a2SYan Zheng if (!upper) { 8703fd0a558SYan, Zheng free_backref_edge(cache, edge); 8715d4f98a2SYan Zheng err = -ENOMEM; 8725d4f98a2SYan Zheng goto out; 8735d4f98a2SYan Zheng } 8745d4f98a2SYan Zheng upper->bytenr = key.offset; 8755d4f98a2SYan Zheng upper->level = cur->level + 1; 8765d4f98a2SYan Zheng /* 8775d4f98a2SYan Zheng * backrefs for the upper level block isn't 8785d4f98a2SYan Zheng * cached, add the block to pending list 8795d4f98a2SYan Zheng */ 8805d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8815d4f98a2SYan Zheng } else { 8825d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8835d4f98a2SYan Zheng rb_node); 88475bfb9afSJosef Bacik ASSERT(upper->checked); 8855d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8865d4f98a2SYan Zheng } 8873fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8885d4f98a2SYan Zheng edge->node[LOWER] = cur; 8893fd0a558SYan, Zheng edge->node[UPPER] = upper; 8905d4f98a2SYan Zheng 8915d4f98a2SYan Zheng goto next; 8925d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8935d4f98a2SYan Zheng goto next; 8945d4f98a2SYan Zheng } 8955d4f98a2SYan Zheng 8965d4f98a2SYan Zheng /* key.type == BTRFS_TREE_BLOCK_REF_KEY */ 8975d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8985d4f98a2SYan Zheng if (IS_ERR(root)) { 8995d4f98a2SYan Zheng err = PTR_ERR(root); 9005d4f98a2SYan Zheng goto out; 9015d4f98a2SYan Zheng } 9025d4f98a2SYan Zheng 90327cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 9043fd0a558SYan, Zheng cur->cowonly = 1; 9053fd0a558SYan, Zheng 9065d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 9075d4f98a2SYan Zheng /* tree root */ 90875bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9095d4f98a2SYan Zheng cur->bytenr); 9103fd0a558SYan, Zheng if (should_ignore_root(root)) 9113fd0a558SYan, Zheng list_add(&cur->list, &useless); 9123fd0a558SYan, Zheng else 9135d4f98a2SYan Zheng cur->root = root; 9145d4f98a2SYan Zheng break; 9155d4f98a2SYan Zheng } 9165d4f98a2SYan Zheng 9175d4f98a2SYan Zheng level = cur->level + 1; 9185d4f98a2SYan Zheng 9195d4f98a2SYan Zheng /* 9205d4f98a2SYan Zheng * searching the tree to find upper level blocks 9215d4f98a2SYan Zheng * reference the block. 9225d4f98a2SYan Zheng */ 9235d4f98a2SYan Zheng path2->search_commit_root = 1; 9245d4f98a2SYan Zheng path2->skip_locking = 1; 9255d4f98a2SYan Zheng path2->lowest_level = level; 9265d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 9275d4f98a2SYan Zheng path2->lowest_level = 0; 9285d4f98a2SYan Zheng if (ret < 0) { 9295d4f98a2SYan Zheng err = ret; 9305d4f98a2SYan Zheng goto out; 9315d4f98a2SYan Zheng } 93233c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 93333c66f43SYan Zheng path2->slots[level]--; 9345d4f98a2SYan Zheng 9355d4f98a2SYan Zheng eb = path2->nodes[level]; 9363561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 9373561b9dbSLiu Bo cur->bytenr) { 9383561b9dbSLiu Bo btrfs_err(root->fs_info, 9393561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 9403561b9dbSLiu Bo cur->bytenr, level - 1, root->objectid, 9413561b9dbSLiu Bo node_key->objectid, node_key->type, 9423561b9dbSLiu Bo node_key->offset); 9433561b9dbSLiu Bo err = -ENOENT; 9443561b9dbSLiu Bo goto out; 9453561b9dbSLiu Bo } 9465d4f98a2SYan Zheng lower = cur; 947b6c60c80SJosef Bacik need_check = true; 9485d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9495d4f98a2SYan Zheng if (!path2->nodes[level]) { 95075bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9515d4f98a2SYan Zheng lower->bytenr); 9523fd0a558SYan, Zheng if (should_ignore_root(root)) 9533fd0a558SYan, Zheng list_add(&lower->list, &useless); 9543fd0a558SYan, Zheng else 9555d4f98a2SYan Zheng lower->root = root; 9565d4f98a2SYan Zheng break; 9575d4f98a2SYan Zheng } 9585d4f98a2SYan Zheng 9593fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9605d4f98a2SYan Zheng if (!edge) { 9615d4f98a2SYan Zheng err = -ENOMEM; 9625d4f98a2SYan Zheng goto out; 9635d4f98a2SYan Zheng } 9645d4f98a2SYan Zheng 9655d4f98a2SYan Zheng eb = path2->nodes[level]; 9665d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9675d4f98a2SYan Zheng if (!rb_node) { 9683fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9695d4f98a2SYan Zheng if (!upper) { 9703fd0a558SYan, Zheng free_backref_edge(cache, edge); 9715d4f98a2SYan Zheng err = -ENOMEM; 9725d4f98a2SYan Zheng goto out; 9735d4f98a2SYan Zheng } 9745d4f98a2SYan Zheng upper->bytenr = eb->start; 9755d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9765d4f98a2SYan Zheng upper->level = lower->level + 1; 97727cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 97827cdeb70SMiao Xie &root->state)) 9793fd0a558SYan, Zheng upper->cowonly = 1; 9805d4f98a2SYan Zheng 9815d4f98a2SYan Zheng /* 9825d4f98a2SYan Zheng * if we know the block isn't shared 9835d4f98a2SYan Zheng * we can void checking its backrefs. 9845d4f98a2SYan Zheng */ 9855d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9865d4f98a2SYan Zheng upper->checked = 0; 9875d4f98a2SYan Zheng else 9885d4f98a2SYan Zheng upper->checked = 1; 9895d4f98a2SYan Zheng 9905d4f98a2SYan Zheng /* 9915d4f98a2SYan Zheng * add the block to pending list if we 992b6c60c80SJosef Bacik * need check its backrefs, we only do this once 993b6c60c80SJosef Bacik * while walking up a tree as we will catch 994b6c60c80SJosef Bacik * anything else later on. 9955d4f98a2SYan Zheng */ 996b6c60c80SJosef Bacik if (!upper->checked && need_check) { 997b6c60c80SJosef Bacik need_check = false; 9985d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9995d4f98a2SYan Zheng &list); 1000bbe90514SJosef Bacik } else { 1001bbe90514SJosef Bacik if (upper->checked) 1002bbe90514SJosef Bacik need_check = true; 10035d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 1004bbe90514SJosef Bacik } 10055d4f98a2SYan Zheng } else { 10065d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 10075d4f98a2SYan Zheng rb_node); 100875bfb9afSJosef Bacik ASSERT(upper->checked); 10095d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 10103fd0a558SYan, Zheng if (!upper->owner) 10113fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 10125d4f98a2SYan Zheng } 10135d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 10145d4f98a2SYan Zheng edge->node[LOWER] = lower; 10153fd0a558SYan, Zheng edge->node[UPPER] = upper; 10165d4f98a2SYan Zheng 10175d4f98a2SYan Zheng if (rb_node) 10185d4f98a2SYan Zheng break; 10195d4f98a2SYan Zheng lower = upper; 10205d4f98a2SYan Zheng upper = NULL; 10215d4f98a2SYan Zheng } 1022b3b4aa74SDavid Sterba btrfs_release_path(path2); 10235d4f98a2SYan Zheng next: 10245d4f98a2SYan Zheng if (ptr < end) { 10255d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 10265d4f98a2SYan Zheng if (ptr >= end) { 10275d4f98a2SYan Zheng WARN_ON(ptr > end); 10285d4f98a2SYan Zheng ptr = 0; 10295d4f98a2SYan Zheng end = 0; 10305d4f98a2SYan Zheng } 10315d4f98a2SYan Zheng } 10325d4f98a2SYan Zheng if (ptr >= end) 10335d4f98a2SYan Zheng path1->slots[0]++; 10345d4f98a2SYan Zheng } 1035b3b4aa74SDavid Sterba btrfs_release_path(path1); 10365d4f98a2SYan Zheng 10375d4f98a2SYan Zheng cur->checked = 1; 10385d4f98a2SYan Zheng WARN_ON(exist); 10395d4f98a2SYan Zheng 10405d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10415d4f98a2SYan Zheng if (!list_empty(&list)) { 10425d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10435d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10445d4f98a2SYan Zheng cur = edge->node[UPPER]; 10455d4f98a2SYan Zheng goto again; 10465d4f98a2SYan Zheng } 10475d4f98a2SYan Zheng 10485d4f98a2SYan Zheng /* 10495d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10505d4f98a2SYan Zheng * into the cache. 10515d4f98a2SYan Zheng */ 105275bfb9afSJosef Bacik ASSERT(node->checked); 10533fd0a558SYan, Zheng cowonly = node->cowonly; 10543fd0a558SYan, Zheng if (!cowonly) { 10553fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10563fd0a558SYan, Zheng &node->rb_node); 105743c04fb1SJeff Mahoney if (rb_node) 105843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10593fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10603fd0a558SYan, Zheng } 10615d4f98a2SYan Zheng 10625d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10635d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10645d4f98a2SYan Zheng 10655d4f98a2SYan Zheng while (!list_empty(&list)) { 10665d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10675d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10685d4f98a2SYan Zheng upper = edge->node[UPPER]; 10693fd0a558SYan, Zheng if (upper->detached) { 10703fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10713fd0a558SYan, Zheng lower = edge->node[LOWER]; 10723fd0a558SYan, Zheng free_backref_edge(cache, edge); 10733fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10743fd0a558SYan, Zheng list_add(&lower->list, &useless); 10753fd0a558SYan, Zheng continue; 10763fd0a558SYan, Zheng } 10775d4f98a2SYan Zheng 10785d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10795d4f98a2SYan Zheng if (upper->lowest) { 10805d4f98a2SYan Zheng list_del_init(&upper->lower); 10815d4f98a2SYan Zheng upper->lowest = 0; 10825d4f98a2SYan Zheng } 10835d4f98a2SYan Zheng 10845d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10855d4f98a2SYan Zheng continue; 10865d4f98a2SYan Zheng } 10875d4f98a2SYan Zheng 108875bfb9afSJosef Bacik if (!upper->checked) { 108975bfb9afSJosef Bacik /* 109075bfb9afSJosef Bacik * Still want to blow up for developers since this is a 109175bfb9afSJosef Bacik * logic bug. 109275bfb9afSJosef Bacik */ 109375bfb9afSJosef Bacik ASSERT(0); 109475bfb9afSJosef Bacik err = -EINVAL; 109575bfb9afSJosef Bacik goto out; 109675bfb9afSJosef Bacik } 109775bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 109875bfb9afSJosef Bacik ASSERT(0); 109975bfb9afSJosef Bacik err = -EINVAL; 110075bfb9afSJosef Bacik goto out; 110175bfb9afSJosef Bacik } 110275bfb9afSJosef Bacik 11033fd0a558SYan, Zheng if (!cowonly) { 11045d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 11055d4f98a2SYan Zheng &upper->rb_node); 110643c04fb1SJeff Mahoney if (rb_node) 110743c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 110843c04fb1SJeff Mahoney upper->bytenr); 11093fd0a558SYan, Zheng } 11105d4f98a2SYan Zheng 11115d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11125d4f98a2SYan Zheng 11135d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 11145d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 11155d4f98a2SYan Zheng } 11163fd0a558SYan, Zheng /* 11173fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 11183fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 11193fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 11203fd0a558SYan, Zheng * lookup. 11213fd0a558SYan, Zheng */ 11223fd0a558SYan, Zheng while (!list_empty(&useless)) { 11233fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 11243fd0a558SYan, Zheng list_del_init(&upper->list); 112575bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 11263fd0a558SYan, Zheng if (upper == node) 11273fd0a558SYan, Zheng node = NULL; 11283fd0a558SYan, Zheng if (upper->lowest) { 11293fd0a558SYan, Zheng list_del_init(&upper->lower); 11303fd0a558SYan, Zheng upper->lowest = 0; 11313fd0a558SYan, Zheng } 11323fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 11333fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 11343fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 11353fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11363fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11373fd0a558SYan, Zheng lower = edge->node[LOWER]; 11383fd0a558SYan, Zheng free_backref_edge(cache, edge); 11393fd0a558SYan, Zheng 11403fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11413fd0a558SYan, Zheng list_add(&lower->list, &useless); 11423fd0a558SYan, Zheng } 11433fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11443fd0a558SYan, Zheng if (upper->level > 0) { 11453fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11463fd0a558SYan, Zheng upper->detached = 1; 11473fd0a558SYan, Zheng } else { 11483fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11493fd0a558SYan, Zheng free_backref_node(cache, upper); 11503fd0a558SYan, Zheng } 11513fd0a558SYan, Zheng } 11525d4f98a2SYan Zheng out: 11535d4f98a2SYan Zheng btrfs_free_path(path1); 11545d4f98a2SYan Zheng btrfs_free_path(path2); 11555d4f98a2SYan Zheng if (err) { 11563fd0a558SYan, Zheng while (!list_empty(&useless)) { 11573fd0a558SYan, Zheng lower = list_entry(useless.next, 115875bfb9afSJosef Bacik struct backref_node, list); 115975bfb9afSJosef Bacik list_del_init(&lower->list); 11603fd0a558SYan, Zheng } 116175bfb9afSJosef Bacik while (!list_empty(&list)) { 116275bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 116375bfb9afSJosef Bacik list[UPPER]); 116475bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11653fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 116675bfb9afSJosef Bacik lower = edge->node[LOWER]; 11675d4f98a2SYan Zheng upper = edge->node[UPPER]; 11683fd0a558SYan, Zheng free_backref_edge(cache, edge); 116975bfb9afSJosef Bacik 117075bfb9afSJosef Bacik /* 117175bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 117275bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 117375bfb9afSJosef Bacik */ 117475bfb9afSJosef Bacik if (list_empty(&lower->upper) && 117575bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 117675bfb9afSJosef Bacik list_add(&lower->list, &useless); 117775bfb9afSJosef Bacik 117875bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 117975bfb9afSJosef Bacik continue; 118075bfb9afSJosef Bacik 118101327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 118275bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 118375bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 118475bfb9afSJosef Bacik if (list_empty(&upper->upper)) 118575bfb9afSJosef Bacik list_add(&upper->list, &useless); 118675bfb9afSJosef Bacik } 118775bfb9afSJosef Bacik 118875bfb9afSJosef Bacik while (!list_empty(&useless)) { 118975bfb9afSJosef Bacik lower = list_entry(useless.next, 119075bfb9afSJosef Bacik struct backref_node, list); 119175bfb9afSJosef Bacik list_del_init(&lower->list); 11920fd8c3daSLiu Bo if (lower == node) 11930fd8c3daSLiu Bo node = NULL; 119475bfb9afSJosef Bacik free_backref_node(cache, lower); 11955d4f98a2SYan Zheng } 11960fd8c3daSLiu Bo 11970fd8c3daSLiu Bo free_backref_node(cache, node); 11985d4f98a2SYan Zheng return ERR_PTR(err); 11995d4f98a2SYan Zheng } 120075bfb9afSJosef Bacik ASSERT(!node || !node->detached); 12015d4f98a2SYan Zheng return node; 12025d4f98a2SYan Zheng } 12035d4f98a2SYan Zheng 12045d4f98a2SYan Zheng /* 12053fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 12063fd0a558SYan, Zheng * the backref node is created by cloning backref node that 12073fd0a558SYan, Zheng * corresponds to root of source tree 12083fd0a558SYan, Zheng */ 12093fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 12103fd0a558SYan, Zheng struct reloc_control *rc, 12113fd0a558SYan, Zheng struct btrfs_root *src, 12123fd0a558SYan, Zheng struct btrfs_root *dest) 12133fd0a558SYan, Zheng { 12143fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 12153fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 12163fd0a558SYan, Zheng struct backref_node *node = NULL; 12173fd0a558SYan, Zheng struct backref_node *new_node; 12183fd0a558SYan, Zheng struct backref_edge *edge; 12193fd0a558SYan, Zheng struct backref_edge *new_edge; 12203fd0a558SYan, Zheng struct rb_node *rb_node; 12213fd0a558SYan, Zheng 12223fd0a558SYan, Zheng if (cache->last_trans > 0) 12233fd0a558SYan, Zheng update_backref_cache(trans, cache); 12243fd0a558SYan, Zheng 12253fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 12263fd0a558SYan, Zheng if (rb_node) { 12273fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 12283fd0a558SYan, Zheng if (node->detached) 12293fd0a558SYan, Zheng node = NULL; 12303fd0a558SYan, Zheng else 12313fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 12323fd0a558SYan, Zheng } 12333fd0a558SYan, Zheng 12343fd0a558SYan, Zheng if (!node) { 12353fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 12363fd0a558SYan, Zheng reloc_root->commit_root->start); 12373fd0a558SYan, Zheng if (rb_node) { 12383fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 12393fd0a558SYan, Zheng rb_node); 12403fd0a558SYan, Zheng BUG_ON(node->detached); 12413fd0a558SYan, Zheng } 12423fd0a558SYan, Zheng } 12433fd0a558SYan, Zheng 12443fd0a558SYan, Zheng if (!node) 12453fd0a558SYan, Zheng return 0; 12463fd0a558SYan, Zheng 12473fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12483fd0a558SYan, Zheng if (!new_node) 12493fd0a558SYan, Zheng return -ENOMEM; 12503fd0a558SYan, Zheng 12513fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12523fd0a558SYan, Zheng new_node->level = node->level; 12533fd0a558SYan, Zheng new_node->lowest = node->lowest; 12546848ad64SYan, Zheng new_node->checked = 1; 12553fd0a558SYan, Zheng new_node->root = dest; 12563fd0a558SYan, Zheng 12573fd0a558SYan, Zheng if (!node->lowest) { 12583fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12593fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12603fd0a558SYan, Zheng if (!new_edge) 12613fd0a558SYan, Zheng goto fail; 12623fd0a558SYan, Zheng 12633fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12643fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12653fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12663fd0a558SYan, Zheng &new_node->lower); 12673fd0a558SYan, Zheng } 126876b9e23dSMiao Xie } else { 126976b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12703fd0a558SYan, Zheng } 12713fd0a558SYan, Zheng 12723fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12733fd0a558SYan, Zheng &new_node->rb_node); 127443c04fb1SJeff Mahoney if (rb_node) 127543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12763fd0a558SYan, Zheng 12773fd0a558SYan, Zheng if (!new_node->lowest) { 12783fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12793fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12803fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12813fd0a558SYan, Zheng } 12823fd0a558SYan, Zheng } 12833fd0a558SYan, Zheng return 0; 12843fd0a558SYan, Zheng fail: 12853fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12863fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12873fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12883fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12893fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12903fd0a558SYan, Zheng } 12913fd0a558SYan, Zheng free_backref_node(cache, new_node); 12923fd0a558SYan, Zheng return -ENOMEM; 12933fd0a558SYan, Zheng } 12943fd0a558SYan, Zheng 12953fd0a558SYan, Zheng /* 12965d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12975d4f98a2SYan Zheng */ 1298ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12995d4f98a2SYan Zheng { 13000b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13015d4f98a2SYan Zheng struct rb_node *rb_node; 13025d4f98a2SYan Zheng struct mapping_node *node; 13030b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 13045d4f98a2SYan Zheng 13055d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1306ffd7b339SJeff Mahoney if (!node) 1307ffd7b339SJeff Mahoney return -ENOMEM; 13085d4f98a2SYan Zheng 13095d4f98a2SYan Zheng node->bytenr = root->node->start; 13105d4f98a2SYan Zheng node->data = root; 13115d4f98a2SYan Zheng 13125d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13135d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13145d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13155d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1316ffd7b339SJeff Mahoney if (rb_node) { 13170b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 13185d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 13195d163e0eSJeff Mahoney node->bytenr); 1320ffd7b339SJeff Mahoney } 13215d4f98a2SYan Zheng 13225d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 13235d4f98a2SYan Zheng return 0; 13245d4f98a2SYan Zheng } 13255d4f98a2SYan Zheng 13265d4f98a2SYan Zheng /* 1327c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 13285d4f98a2SYan Zheng * mapping 13295d4f98a2SYan Zheng */ 1330c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 13315d4f98a2SYan Zheng { 13320b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13335d4f98a2SYan Zheng struct rb_node *rb_node; 13345d4f98a2SYan Zheng struct mapping_node *node = NULL; 13350b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 13365d4f98a2SYan Zheng 13375d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13385d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1339c974c464SWang Shilong root->node->start); 1340c974c464SWang Shilong if (rb_node) { 1341c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1342c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1343c974c464SWang Shilong } 1344c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1345c974c464SWang Shilong 1346c974c464SWang Shilong if (!node) 1347c974c464SWang Shilong return; 1348c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1349c974c464SWang Shilong 13500b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1351c974c464SWang Shilong list_del_init(&root->root_list); 13520b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1353c974c464SWang Shilong kfree(node); 1354c974c464SWang Shilong } 1355c974c464SWang Shilong 1356c974c464SWang Shilong /* 1357c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1358c974c464SWang Shilong * mapping 1359c974c464SWang Shilong */ 1360c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1361c974c464SWang Shilong { 13620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1363c974c464SWang Shilong struct rb_node *rb_node; 1364c974c464SWang Shilong struct mapping_node *node = NULL; 13650b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1366c974c464SWang Shilong 1367c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1368c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1369c974c464SWang Shilong root->node->start); 13705d4f98a2SYan Zheng if (rb_node) { 13715d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13725d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13735d4f98a2SYan Zheng } 13745d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13755d4f98a2SYan Zheng 13768f71f3e0SLiu Bo if (!node) 13778f71f3e0SLiu Bo return 0; 13785d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13795d4f98a2SYan Zheng 13805d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1381c974c464SWang Shilong node->bytenr = new_bytenr; 13825d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13835d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13845d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 138543c04fb1SJeff Mahoney if (rb_node) 138643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13875d4f98a2SYan Zheng return 0; 13885d4f98a2SYan Zheng } 13895d4f98a2SYan Zheng 13903fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13913fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13925d4f98a2SYan Zheng { 13930b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13945d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13955d4f98a2SYan Zheng struct extent_buffer *eb; 13965d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13975d4f98a2SYan Zheng struct btrfs_key root_key; 13985d4f98a2SYan Zheng int ret; 13995d4f98a2SYan Zheng 14005d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 14015d4f98a2SYan Zheng BUG_ON(!root_item); 14025d4f98a2SYan Zheng 14035d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 14045d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 14053fd0a558SYan, Zheng root_key.offset = objectid; 14065d4f98a2SYan Zheng 14073fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1408054570a1SFilipe Manana u64 commit_root_gen; 1409054570a1SFilipe Manana 14103fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 14115d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 14125d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 14135d4f98a2SYan Zheng BUG_ON(ret); 1414054570a1SFilipe Manana /* 1415054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1416054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1417054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1418054570a1SFilipe Manana * either inside the critical section of a transaction commit 1419054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1420054570a1SFilipe Manana * it's created before the transaction commit is started. 1421054570a1SFilipe Manana */ 1422054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1423054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 14243fd0a558SYan, Zheng } else { 14253fd0a558SYan, Zheng /* 14263fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 14273fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 14283fd0a558SYan, Zheng * modified after it was created have RELOC flag 14293fd0a558SYan, Zheng * set in their headers. so it's OK to not update 14303fd0a558SYan, Zheng * the 'last_snapshot'. 14313fd0a558SYan, Zheng */ 14323fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 14333fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 14343fd0a558SYan, Zheng BUG_ON(ret); 14353fd0a558SYan, Zheng } 14363fd0a558SYan, Zheng 14375d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 14385d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14395d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14405d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14413fd0a558SYan, Zheng 14423fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14433fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14443fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14453fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14465d4f98a2SYan Zheng root_item->drop_level = 0; 14473fd0a558SYan, Zheng } 14485d4f98a2SYan Zheng 14495d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14505d4f98a2SYan Zheng free_extent_buffer(eb); 14515d4f98a2SYan Zheng 14520b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 14535d4f98a2SYan Zheng &root_key, root_item); 14545d4f98a2SYan Zheng BUG_ON(ret); 14555d4f98a2SYan Zheng kfree(root_item); 14565d4f98a2SYan Zheng 14570b246afaSJeff Mahoney reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key); 14585d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14595d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14603fd0a558SYan, Zheng return reloc_root; 14613fd0a558SYan, Zheng } 14623fd0a558SYan, Zheng 14633fd0a558SYan, Zheng /* 14643fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 14653fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 14663fd0a558SYan, Zheng */ 14673fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 14683fd0a558SYan, Zheng struct btrfs_root *root) 14693fd0a558SYan, Zheng { 14700b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14713fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14720b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 147320dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14743fd0a558SYan, Zheng int clear_rsv = 0; 1475ffd7b339SJeff Mahoney int ret; 14763fd0a558SYan, Zheng 14773fd0a558SYan, Zheng if (root->reloc_root) { 14783fd0a558SYan, Zheng reloc_root = root->reloc_root; 14793fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14803fd0a558SYan, Zheng return 0; 14813fd0a558SYan, Zheng } 14823fd0a558SYan, Zheng 14833fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14843fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14853fd0a558SYan, Zheng return 0; 14863fd0a558SYan, Zheng 148720dd2cbfSMiao Xie if (!trans->reloc_reserved) { 148820dd2cbfSMiao Xie rsv = trans->block_rsv; 14893fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14903fd0a558SYan, Zheng clear_rsv = 1; 14913fd0a558SYan, Zheng } 14923fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14933fd0a558SYan, Zheng if (clear_rsv) 149420dd2cbfSMiao Xie trans->block_rsv = rsv; 14955d4f98a2SYan Zheng 1496ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1497ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14985d4f98a2SYan Zheng root->reloc_root = reloc_root; 14995d4f98a2SYan Zheng return 0; 15005d4f98a2SYan Zheng } 15015d4f98a2SYan Zheng 15025d4f98a2SYan Zheng /* 15035d4f98a2SYan Zheng * update root item of reloc tree 15045d4f98a2SYan Zheng */ 15055d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 15065d4f98a2SYan Zheng struct btrfs_root *root) 15075d4f98a2SYan Zheng { 15080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15095d4f98a2SYan Zheng struct btrfs_root *reloc_root; 15105d4f98a2SYan Zheng struct btrfs_root_item *root_item; 15115d4f98a2SYan Zheng int ret; 15125d4f98a2SYan Zheng 15135d4f98a2SYan Zheng if (!root->reloc_root) 15147585717fSChris Mason goto out; 15155d4f98a2SYan Zheng 15165d4f98a2SYan Zheng reloc_root = root->reloc_root; 15175d4f98a2SYan Zheng root_item = &reloc_root->root_item; 15185d4f98a2SYan Zheng 15190b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 15203fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 15215d4f98a2SYan Zheng root->reloc_root = NULL; 1522c974c464SWang Shilong __del_reloc_root(reloc_root); 15235d4f98a2SYan Zheng } 15245d4f98a2SYan Zheng 15255d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 15265d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 15275d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 15285d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 15295d4f98a2SYan Zheng } 15305d4f98a2SYan Zheng 15310b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 15325d4f98a2SYan Zheng &reloc_root->root_key, root_item); 15335d4f98a2SYan Zheng BUG_ON(ret); 15347585717fSChris Mason 15357585717fSChris Mason out: 15365d4f98a2SYan Zheng return 0; 15375d4f98a2SYan Zheng } 15385d4f98a2SYan Zheng 15395d4f98a2SYan Zheng /* 15405d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15415d4f98a2SYan Zheng * in a subvolume 15425d4f98a2SYan Zheng */ 15435d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15445d4f98a2SYan Zheng { 15455d4f98a2SYan Zheng struct rb_node *node; 15465d4f98a2SYan Zheng struct rb_node *prev; 15475d4f98a2SYan Zheng struct btrfs_inode *entry; 15485d4f98a2SYan Zheng struct inode *inode; 15495d4f98a2SYan Zheng 15505d4f98a2SYan Zheng spin_lock(&root->inode_lock); 15515d4f98a2SYan Zheng again: 15525d4f98a2SYan Zheng node = root->inode_tree.rb_node; 15535d4f98a2SYan Zheng prev = NULL; 15545d4f98a2SYan Zheng while (node) { 15555d4f98a2SYan Zheng prev = node; 15565d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15575d4f98a2SYan Zheng 15584a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 15595d4f98a2SYan Zheng node = node->rb_left; 15604a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 15615d4f98a2SYan Zheng node = node->rb_right; 15625d4f98a2SYan Zheng else 15635d4f98a2SYan Zheng break; 15645d4f98a2SYan Zheng } 15655d4f98a2SYan Zheng if (!node) { 15665d4f98a2SYan Zheng while (prev) { 15675d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 15684a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 15695d4f98a2SYan Zheng node = prev; 15705d4f98a2SYan Zheng break; 15715d4f98a2SYan Zheng } 15725d4f98a2SYan Zheng prev = rb_next(prev); 15735d4f98a2SYan Zheng } 15745d4f98a2SYan Zheng } 15755d4f98a2SYan Zheng while (node) { 15765d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15775d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15785d4f98a2SYan Zheng if (inode) { 15795d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15805d4f98a2SYan Zheng return inode; 15815d4f98a2SYan Zheng } 15825d4f98a2SYan Zheng 15834a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 15845d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15855d4f98a2SYan Zheng goto again; 15865d4f98a2SYan Zheng 15875d4f98a2SYan Zheng node = rb_next(node); 15885d4f98a2SYan Zheng } 15895d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15905d4f98a2SYan Zheng return NULL; 15915d4f98a2SYan Zheng } 15925d4f98a2SYan Zheng 15935d4f98a2SYan Zheng static int in_block_group(u64 bytenr, 15945d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group) 15955d4f98a2SYan Zheng { 15965d4f98a2SYan Zheng if (bytenr >= block_group->key.objectid && 15975d4f98a2SYan Zheng bytenr < block_group->key.objectid + block_group->key.offset) 15985d4f98a2SYan Zheng return 1; 15995d4f98a2SYan Zheng return 0; 16005d4f98a2SYan Zheng } 16015d4f98a2SYan Zheng 16025d4f98a2SYan Zheng /* 16035d4f98a2SYan Zheng * get new location of data 16045d4f98a2SYan Zheng */ 16055d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 16065d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 16075d4f98a2SYan Zheng { 16085d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 16095d4f98a2SYan Zheng struct btrfs_path *path; 16105d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16115d4f98a2SYan Zheng struct extent_buffer *leaf; 16125d4f98a2SYan Zheng int ret; 16135d4f98a2SYan Zheng 16145d4f98a2SYan Zheng path = btrfs_alloc_path(); 16155d4f98a2SYan Zheng if (!path) 16165d4f98a2SYan Zheng return -ENOMEM; 16175d4f98a2SYan Zheng 16185d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1619f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1620f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 16215d4f98a2SYan Zheng if (ret < 0) 16225d4f98a2SYan Zheng goto out; 16235d4f98a2SYan Zheng if (ret > 0) { 16245d4f98a2SYan Zheng ret = -ENOENT; 16255d4f98a2SYan Zheng goto out; 16265d4f98a2SYan Zheng } 16275d4f98a2SYan Zheng 16285d4f98a2SYan Zheng leaf = path->nodes[0]; 16295d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 16305d4f98a2SYan Zheng struct btrfs_file_extent_item); 16315d4f98a2SYan Zheng 16325d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 16335d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 16345d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 16355d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 16365d4f98a2SYan Zheng 16375d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 163883d4cfd4SJosef Bacik ret = -EINVAL; 16395d4f98a2SYan Zheng goto out; 16405d4f98a2SYan Zheng } 16415d4f98a2SYan Zheng 16425d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16435d4f98a2SYan Zheng ret = 0; 16445d4f98a2SYan Zheng out: 16455d4f98a2SYan Zheng btrfs_free_path(path); 16465d4f98a2SYan Zheng return ret; 16475d4f98a2SYan Zheng } 16485d4f98a2SYan Zheng 16495d4f98a2SYan Zheng /* 16505d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16515d4f98a2SYan Zheng * the new locations. 16525d4f98a2SYan Zheng */ 16533fd0a558SYan, Zheng static noinline_for_stack 16543fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 16555d4f98a2SYan Zheng struct reloc_control *rc, 16565d4f98a2SYan Zheng struct btrfs_root *root, 16573fd0a558SYan, Zheng struct extent_buffer *leaf) 16585d4f98a2SYan Zheng { 16590b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 16605d4f98a2SYan Zheng struct btrfs_key key; 16615d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16625d4f98a2SYan Zheng struct inode *inode = NULL; 16635d4f98a2SYan Zheng u64 parent; 16645d4f98a2SYan Zheng u64 bytenr; 16653fd0a558SYan, Zheng u64 new_bytenr = 0; 16665d4f98a2SYan Zheng u64 num_bytes; 16675d4f98a2SYan Zheng u64 end; 16685d4f98a2SYan Zheng u32 nritems; 16695d4f98a2SYan Zheng u32 i; 167083d4cfd4SJosef Bacik int ret = 0; 16715d4f98a2SYan Zheng int first = 1; 16725d4f98a2SYan Zheng int dirty = 0; 16735d4f98a2SYan Zheng 16745d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16755d4f98a2SYan Zheng return 0; 16765d4f98a2SYan Zheng 16775d4f98a2SYan Zheng /* reloc trees always use full backref */ 16785d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16795d4f98a2SYan Zheng parent = leaf->start; 16805d4f98a2SYan Zheng else 16815d4f98a2SYan Zheng parent = 0; 16825d4f98a2SYan Zheng 16835d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16845d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 16855d4f98a2SYan Zheng cond_resched(); 16865d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16875d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16885d4f98a2SYan Zheng continue; 16895d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16905d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16915d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16925d4f98a2SYan Zheng continue; 16935d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16945d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16955d4f98a2SYan Zheng if (bytenr == 0) 16965d4f98a2SYan Zheng continue; 16975d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16985d4f98a2SYan Zheng continue; 16995d4f98a2SYan Zheng 17005d4f98a2SYan Zheng /* 17015d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 17025d4f98a2SYan Zheng * to complete and drop the extent cache 17035d4f98a2SYan Zheng */ 17045d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 17055d4f98a2SYan Zheng if (first) { 17065d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17075d4f98a2SYan Zheng first = 0; 17084a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 17093fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 17105d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17115d4f98a2SYan Zheng } 17124a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 17135d4f98a2SYan Zheng end = key.offset + 17145d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 17155d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 17160b246afaSJeff Mahoney fs_info->sectorsize)); 17170b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 17185d4f98a2SYan Zheng end--; 17195d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1720d0082371SJeff Mahoney key.offset, end); 17215d4f98a2SYan Zheng if (!ret) 17225d4f98a2SYan Zheng continue; 17235d4f98a2SYan Zheng 1724dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1725dcdbc059SNikolay Borisov key.offset, end, 1); 17265d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1727d0082371SJeff Mahoney key.offset, end); 17285d4f98a2SYan Zheng } 17295d4f98a2SYan Zheng } 17305d4f98a2SYan Zheng 17315d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 17325d4f98a2SYan Zheng bytenr, num_bytes); 173383d4cfd4SJosef Bacik if (ret) { 173483d4cfd4SJosef Bacik /* 173583d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 173683d4cfd4SJosef Bacik * in the file extent yet. 173783d4cfd4SJosef Bacik */ 173883d4cfd4SJosef Bacik break; 17393fd0a558SYan, Zheng } 17405d4f98a2SYan Zheng 17415d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17425d4f98a2SYan Zheng dirty = 1; 17435d4f98a2SYan Zheng 17445d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 174584f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, root, new_bytenr, 17465d4f98a2SYan Zheng num_bytes, parent, 17475d4f98a2SYan Zheng btrfs_header_owner(leaf), 1748b06c4bf5SFilipe Manana key.objectid, key.offset); 174983d4cfd4SJosef Bacik if (ret) { 175066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 175183d4cfd4SJosef Bacik break; 175283d4cfd4SJosef Bacik } 17535d4f98a2SYan Zheng 175484f7d8e6SJosef Bacik ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 17555d4f98a2SYan Zheng parent, btrfs_header_owner(leaf), 1756b06c4bf5SFilipe Manana key.objectid, key.offset); 175783d4cfd4SJosef Bacik if (ret) { 175866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 175983d4cfd4SJosef Bacik break; 176083d4cfd4SJosef Bacik } 17615d4f98a2SYan Zheng } 17625d4f98a2SYan Zheng if (dirty) 17635d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17643fd0a558SYan, Zheng if (inode) 17653fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 176683d4cfd4SJosef Bacik return ret; 17675d4f98a2SYan Zheng } 17685d4f98a2SYan Zheng 17695d4f98a2SYan Zheng static noinline_for_stack 17705d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17715d4f98a2SYan Zheng struct btrfs_path *path, int level) 17725d4f98a2SYan Zheng { 17735d4f98a2SYan Zheng struct btrfs_disk_key key1; 17745d4f98a2SYan Zheng struct btrfs_disk_key key2; 17755d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17765d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17775d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17785d4f98a2SYan Zheng } 17795d4f98a2SYan Zheng 17805d4f98a2SYan Zheng /* 17815d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17825d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17835d4f98a2SYan Zheng * reloc tree was create can be replaced. 17845d4f98a2SYan Zheng * 17855d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17865d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17875d4f98a2SYan Zheng * errors, a negative error number is returned. 17885d4f98a2SYan Zheng */ 17893fd0a558SYan, Zheng static noinline_for_stack 17903fd0a558SYan, Zheng int replace_path(struct btrfs_trans_handle *trans, 17915d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17925d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17935d4f98a2SYan Zheng int lowest_level, int max_level) 17945d4f98a2SYan Zheng { 17950b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 17965d4f98a2SYan Zheng struct extent_buffer *eb; 17975d4f98a2SYan Zheng struct extent_buffer *parent; 17985d4f98a2SYan Zheng struct btrfs_key key; 17995d4f98a2SYan Zheng u64 old_bytenr; 18005d4f98a2SYan Zheng u64 new_bytenr; 18015d4f98a2SYan Zheng u64 old_ptr_gen; 18025d4f98a2SYan Zheng u64 new_ptr_gen; 18035d4f98a2SYan Zheng u64 last_snapshot; 18045d4f98a2SYan Zheng u32 blocksize; 18053fd0a558SYan, Zheng int cow = 0; 18065d4f98a2SYan Zheng int level; 18075d4f98a2SYan Zheng int ret; 18085d4f98a2SYan Zheng int slot; 18095d4f98a2SYan Zheng 18105d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 18115d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 18125d4f98a2SYan Zheng 18135d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 18143fd0a558SYan, Zheng again: 18155d4f98a2SYan Zheng slot = path->slots[lowest_level]; 18165d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 18175d4f98a2SYan Zheng 18185d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 18195d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 18205d4f98a2SYan Zheng level = btrfs_header_level(eb); 18215d4f98a2SYan Zheng 18225d4f98a2SYan Zheng if (level < lowest_level) { 18235d4f98a2SYan Zheng btrfs_tree_unlock(eb); 18245d4f98a2SYan Zheng free_extent_buffer(eb); 18255d4f98a2SYan Zheng return 0; 18265d4f98a2SYan Zheng } 18275d4f98a2SYan Zheng 18283fd0a558SYan, Zheng if (cow) { 18295d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18305d4f98a2SYan Zheng BUG_ON(ret); 18313fd0a558SYan, Zheng } 18325d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 18335d4f98a2SYan Zheng 18345d4f98a2SYan Zheng if (next_key) { 18355d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18365d4f98a2SYan Zheng next_key->type = (u8)-1; 18375d4f98a2SYan Zheng next_key->offset = (u64)-1; 18385d4f98a2SYan Zheng } 18395d4f98a2SYan Zheng 18405d4f98a2SYan Zheng parent = eb; 18415d4f98a2SYan Zheng while (1) { 18425d4f98a2SYan Zheng level = btrfs_header_level(parent); 18435d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18445d4f98a2SYan Zheng 18455d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 18465d4f98a2SYan Zheng if (ret && slot > 0) 18475d4f98a2SYan Zheng slot--; 18485d4f98a2SYan Zheng 18495d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18505d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18515d4f98a2SYan Zheng 18525d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 18530b246afaSJeff Mahoney blocksize = fs_info->nodesize; 18545d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 18555d4f98a2SYan Zheng 18565d4f98a2SYan Zheng if (level <= max_level) { 18575d4f98a2SYan Zheng eb = path->nodes[level]; 18585d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18595d4f98a2SYan Zheng path->slots[level]); 18605d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18615d4f98a2SYan Zheng path->slots[level]); 18625d4f98a2SYan Zheng } else { 18635d4f98a2SYan Zheng new_bytenr = 0; 18645d4f98a2SYan Zheng new_ptr_gen = 0; 18655d4f98a2SYan Zheng } 18665d4f98a2SYan Zheng 1867fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18685d4f98a2SYan Zheng ret = level; 18695d4f98a2SYan Zheng break; 18705d4f98a2SYan Zheng } 18715d4f98a2SYan Zheng 18725d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18735d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18743fd0a558SYan, Zheng if (level <= lowest_level) { 18755d4f98a2SYan Zheng ret = 0; 18765d4f98a2SYan Zheng break; 18775d4f98a2SYan Zheng } 18785d4f98a2SYan Zheng 18792ff7e61eSJeff Mahoney eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen); 188064c043deSLiu Bo if (IS_ERR(eb)) { 188164c043deSLiu Bo ret = PTR_ERR(eb); 1882264813acSLiu Bo break; 188364c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 188464c043deSLiu Bo ret = -EIO; 1885416bc658SJosef Bacik free_extent_buffer(eb); 1886379cde74SStefan Behrens break; 1887416bc658SJosef Bacik } 18885d4f98a2SYan Zheng btrfs_tree_lock(eb); 18893fd0a558SYan, Zheng if (cow) { 18905d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18915d4f98a2SYan Zheng slot, &eb); 18925d4f98a2SYan Zheng BUG_ON(ret); 18935d4f98a2SYan Zheng } 18943fd0a558SYan, Zheng btrfs_set_lock_blocking(eb); 18955d4f98a2SYan Zheng 18965d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18975d4f98a2SYan Zheng free_extent_buffer(parent); 18985d4f98a2SYan Zheng 18995d4f98a2SYan Zheng parent = eb; 19005d4f98a2SYan Zheng continue; 19015d4f98a2SYan Zheng } 19025d4f98a2SYan Zheng 19033fd0a558SYan, Zheng if (!cow) { 19043fd0a558SYan, Zheng btrfs_tree_unlock(parent); 19053fd0a558SYan, Zheng free_extent_buffer(parent); 19063fd0a558SYan, Zheng cow = 1; 19073fd0a558SYan, Zheng goto again; 19083fd0a558SYan, Zheng } 19093fd0a558SYan, Zheng 19105d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 19115d4f98a2SYan Zheng path->slots[level]); 1912b3b4aa74SDavid Sterba btrfs_release_path(path); 19135d4f98a2SYan Zheng 19145d4f98a2SYan Zheng path->lowest_level = level; 19155d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 19165d4f98a2SYan Zheng path->lowest_level = 0; 19175d4f98a2SYan Zheng BUG_ON(ret); 19185d4f98a2SYan Zheng 19195d4f98a2SYan Zheng /* 1920824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1921824d8dffSQu Wenruo * 1922824d8dffSQu Wenruo * We must trace both trees. 1923824d8dffSQu Wenruo * 1) Tree reloc subtree 1924824d8dffSQu Wenruo * If not traced, we will leak data numbers 1925824d8dffSQu Wenruo * 2) Fs subtree 1926824d8dffSQu Wenruo * If not traced, we will double count old data 1927824d8dffSQu Wenruo * and tree block numbers, if current trans doesn't free 1928824d8dffSQu Wenruo * data reloc tree inode. 1929824d8dffSQu Wenruo */ 1930824d8dffSQu Wenruo ret = btrfs_qgroup_trace_subtree(trans, src, parent, 1931824d8dffSQu Wenruo btrfs_header_generation(parent), 1932824d8dffSQu Wenruo btrfs_header_level(parent)); 1933824d8dffSQu Wenruo if (ret < 0) 1934824d8dffSQu Wenruo break; 1935824d8dffSQu Wenruo ret = btrfs_qgroup_trace_subtree(trans, dest, 1936824d8dffSQu Wenruo path->nodes[level], 1937824d8dffSQu Wenruo btrfs_header_generation(path->nodes[level]), 1938824d8dffSQu Wenruo btrfs_header_level(path->nodes[level])); 1939824d8dffSQu Wenruo if (ret < 0) 1940824d8dffSQu Wenruo break; 1941824d8dffSQu Wenruo 1942824d8dffSQu Wenruo /* 19435d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 19445d4f98a2SYan Zheng */ 19455d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 19465d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19475d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19485d4f98a2SYan Zheng 19495d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19505d4f98a2SYan Zheng path->slots[level], old_bytenr); 19515d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19525d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19535d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19545d4f98a2SYan Zheng 195584f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, src, old_bytenr, 19562ff7e61eSJeff Mahoney blocksize, path->nodes[level]->start, 1957b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19585d4f98a2SYan Zheng BUG_ON(ret); 195984f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, 19602ff7e61eSJeff Mahoney blocksize, 0, dest->root_key.objectid, 19612ff7e61eSJeff Mahoney level - 1, 0); 19625d4f98a2SYan Zheng BUG_ON(ret); 19635d4f98a2SYan Zheng 196484f7d8e6SJosef Bacik ret = btrfs_free_extent(trans, src, new_bytenr, blocksize, 19655d4f98a2SYan Zheng path->nodes[level]->start, 1966b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19675d4f98a2SYan Zheng BUG_ON(ret); 19685d4f98a2SYan Zheng 196984f7d8e6SJosef Bacik ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize, 19705d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 1971b06c4bf5SFilipe Manana 0); 19725d4f98a2SYan Zheng BUG_ON(ret); 19735d4f98a2SYan Zheng 19745d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19755d4f98a2SYan Zheng 19765d4f98a2SYan Zheng ret = level; 19775d4f98a2SYan Zheng break; 19785d4f98a2SYan Zheng } 19795d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19805d4f98a2SYan Zheng free_extent_buffer(parent); 19815d4f98a2SYan Zheng return ret; 19825d4f98a2SYan Zheng } 19835d4f98a2SYan Zheng 19845d4f98a2SYan Zheng /* 19855d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19865d4f98a2SYan Zheng */ 19875d4f98a2SYan Zheng static noinline_for_stack 19885d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19895d4f98a2SYan Zheng int *level) 19905d4f98a2SYan Zheng { 19915d4f98a2SYan Zheng struct extent_buffer *eb; 19925d4f98a2SYan Zheng int i; 19935d4f98a2SYan Zheng u64 last_snapshot; 19945d4f98a2SYan Zheng u32 nritems; 19955d4f98a2SYan Zheng 19965d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19975d4f98a2SYan Zheng 19985d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19995d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20005d4f98a2SYan Zheng path->nodes[i] = NULL; 20015d4f98a2SYan Zheng } 20025d4f98a2SYan Zheng 20035d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 20045d4f98a2SYan Zheng eb = path->nodes[i]; 20055d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20065d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 20075d4f98a2SYan Zheng path->slots[i]++; 20085d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 20095d4f98a2SYan Zheng last_snapshot) 20105d4f98a2SYan Zheng continue; 20115d4f98a2SYan Zheng 20125d4f98a2SYan Zheng *level = i; 20135d4f98a2SYan Zheng return 0; 20145d4f98a2SYan Zheng } 20155d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20165d4f98a2SYan Zheng path->nodes[i] = NULL; 20175d4f98a2SYan Zheng } 20185d4f98a2SYan Zheng return 1; 20195d4f98a2SYan Zheng } 20205d4f98a2SYan Zheng 20215d4f98a2SYan Zheng /* 20225d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 20235d4f98a2SYan Zheng */ 20245d4f98a2SYan Zheng static noinline_for_stack 20255d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20265d4f98a2SYan Zheng int *level) 20275d4f98a2SYan Zheng { 20282ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20295d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 20305d4f98a2SYan Zheng int i; 20315d4f98a2SYan Zheng u64 bytenr; 20325d4f98a2SYan Zheng u64 ptr_gen = 0; 20335d4f98a2SYan Zheng u64 last_snapshot; 20345d4f98a2SYan Zheng u32 nritems; 20355d4f98a2SYan Zheng 20365d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20375d4f98a2SYan Zheng 20385d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 20395d4f98a2SYan Zheng eb = path->nodes[i]; 20405d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20415d4f98a2SYan Zheng while (path->slots[i] < nritems) { 20425d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 20435d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 20445d4f98a2SYan Zheng break; 20455d4f98a2SYan Zheng path->slots[i]++; 20465d4f98a2SYan Zheng } 20475d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20485d4f98a2SYan Zheng if (i == *level) 20495d4f98a2SYan Zheng break; 20505d4f98a2SYan Zheng *level = i + 1; 20515d4f98a2SYan Zheng return 0; 20525d4f98a2SYan Zheng } 20535d4f98a2SYan Zheng if (i == 1) { 20545d4f98a2SYan Zheng *level = i; 20555d4f98a2SYan Zheng return 0; 20565d4f98a2SYan Zheng } 20575d4f98a2SYan Zheng 20585d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 20592ff7e61eSJeff Mahoney eb = read_tree_block(fs_info, bytenr, ptr_gen); 206064c043deSLiu Bo if (IS_ERR(eb)) { 206164c043deSLiu Bo return PTR_ERR(eb); 206264c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2063416bc658SJosef Bacik free_extent_buffer(eb); 2064416bc658SJosef Bacik return -EIO; 2065416bc658SJosef Bacik } 20665d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20675d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20685d4f98a2SYan Zheng path->slots[i - 1] = 0; 20695d4f98a2SYan Zheng } 20705d4f98a2SYan Zheng return 1; 20715d4f98a2SYan Zheng } 20725d4f98a2SYan Zheng 20735d4f98a2SYan Zheng /* 20745d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 20755d4f98a2SYan Zheng * [min_key, max_key) 20765d4f98a2SYan Zheng */ 20775d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 20785d4f98a2SYan Zheng struct btrfs_key *min_key, 20795d4f98a2SYan Zheng struct btrfs_key *max_key) 20805d4f98a2SYan Zheng { 20810b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20825d4f98a2SYan Zheng struct inode *inode = NULL; 20835d4f98a2SYan Zheng u64 objectid; 20845d4f98a2SYan Zheng u64 start, end; 208533345d01SLi Zefan u64 ino; 20865d4f98a2SYan Zheng 20875d4f98a2SYan Zheng objectid = min_key->objectid; 20885d4f98a2SYan Zheng while (1) { 20895d4f98a2SYan Zheng cond_resched(); 20905d4f98a2SYan Zheng iput(inode); 20915d4f98a2SYan Zheng 20925d4f98a2SYan Zheng if (objectid > max_key->objectid) 20935d4f98a2SYan Zheng break; 20945d4f98a2SYan Zheng 20955d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20965d4f98a2SYan Zheng if (!inode) 20975d4f98a2SYan Zheng break; 20984a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 20995d4f98a2SYan Zheng 210033345d01SLi Zefan if (ino > max_key->objectid) { 21015d4f98a2SYan Zheng iput(inode); 21025d4f98a2SYan Zheng break; 21035d4f98a2SYan Zheng } 21045d4f98a2SYan Zheng 210533345d01SLi Zefan objectid = ino + 1; 21065d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 21075d4f98a2SYan Zheng continue; 21085d4f98a2SYan Zheng 210933345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 21105d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 21115d4f98a2SYan Zheng continue; 21125d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 21135d4f98a2SYan Zheng start = 0; 21145d4f98a2SYan Zheng else { 21155d4f98a2SYan Zheng start = min_key->offset; 21160b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 21175d4f98a2SYan Zheng } 21185d4f98a2SYan Zheng } else { 21195d4f98a2SYan Zheng start = 0; 21205d4f98a2SYan Zheng } 21215d4f98a2SYan Zheng 212233345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 21235d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 21245d4f98a2SYan Zheng continue; 21255d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 21265d4f98a2SYan Zheng end = (u64)-1; 21275d4f98a2SYan Zheng } else { 21285d4f98a2SYan Zheng if (max_key->offset == 0) 21295d4f98a2SYan Zheng continue; 21305d4f98a2SYan Zheng end = max_key->offset; 21310b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 21325d4f98a2SYan Zheng end--; 21335d4f98a2SYan Zheng } 21345d4f98a2SYan Zheng } else { 21355d4f98a2SYan Zheng end = (u64)-1; 21365d4f98a2SYan Zheng } 21375d4f98a2SYan Zheng 21385d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2139d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2140dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2141d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 21425d4f98a2SYan Zheng } 21435d4f98a2SYan Zheng return 0; 21445d4f98a2SYan Zheng } 21455d4f98a2SYan Zheng 21465d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 21475d4f98a2SYan Zheng struct btrfs_key *key) 21485d4f98a2SYan Zheng 21495d4f98a2SYan Zheng { 21505d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21515d4f98a2SYan Zheng if (!path->nodes[level]) 21525d4f98a2SYan Zheng break; 21535d4f98a2SYan Zheng if (path->slots[level] + 1 < 21545d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21555d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21565d4f98a2SYan Zheng path->slots[level] + 1); 21575d4f98a2SYan Zheng return 0; 21585d4f98a2SYan Zheng } 21595d4f98a2SYan Zheng level++; 21605d4f98a2SYan Zheng } 21615d4f98a2SYan Zheng return 1; 21625d4f98a2SYan Zheng } 21635d4f98a2SYan Zheng 21645d4f98a2SYan Zheng /* 21655d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 21665d4f98a2SYan Zheng * fs tree. 21675d4f98a2SYan Zheng */ 21685d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 21695d4f98a2SYan Zheng struct btrfs_root *root) 21705d4f98a2SYan Zheng { 21710b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 21725d4f98a2SYan Zheng LIST_HEAD(inode_list); 21735d4f98a2SYan Zheng struct btrfs_key key; 21745d4f98a2SYan Zheng struct btrfs_key next_key; 21759e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 21765d4f98a2SYan Zheng struct btrfs_root *reloc_root; 21775d4f98a2SYan Zheng struct btrfs_root_item *root_item; 21785d4f98a2SYan Zheng struct btrfs_path *path; 21793fd0a558SYan, Zheng struct extent_buffer *leaf; 21805d4f98a2SYan Zheng int level; 21815d4f98a2SYan Zheng int max_level; 21825d4f98a2SYan Zheng int replaced = 0; 21835d4f98a2SYan Zheng int ret; 21845d4f98a2SYan Zheng int err = 0; 21853fd0a558SYan, Zheng u32 min_reserved; 21865d4f98a2SYan Zheng 21875d4f98a2SYan Zheng path = btrfs_alloc_path(); 21885d4f98a2SYan Zheng if (!path) 21895d4f98a2SYan Zheng return -ENOMEM; 2190e4058b54SDavid Sterba path->reada = READA_FORWARD; 21915d4f98a2SYan Zheng 21925d4f98a2SYan Zheng reloc_root = root->reloc_root; 21935d4f98a2SYan Zheng root_item = &reloc_root->root_item; 21945d4f98a2SYan Zheng 21955d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 21965d4f98a2SYan Zheng level = btrfs_root_level(root_item); 21975d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 21985d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 21995d4f98a2SYan Zheng path->slots[level] = 0; 22005d4f98a2SYan Zheng } else { 22015d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 22025d4f98a2SYan Zheng 22035d4f98a2SYan Zheng level = root_item->drop_level; 22045d4f98a2SYan Zheng BUG_ON(level == 0); 22055d4f98a2SYan Zheng path->lowest_level = level; 22065d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 220733c66f43SYan Zheng path->lowest_level = 0; 22085d4f98a2SYan Zheng if (ret < 0) { 22095d4f98a2SYan Zheng btrfs_free_path(path); 22105d4f98a2SYan Zheng return ret; 22115d4f98a2SYan Zheng } 22125d4f98a2SYan Zheng 22135d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 22145d4f98a2SYan Zheng path->slots[level]); 22155d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 22165d4f98a2SYan Zheng 22175d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 22185d4f98a2SYan Zheng } 22195d4f98a2SYan Zheng 22200b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22215d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 22225d4f98a2SYan Zheng 22235d4f98a2SYan Zheng while (1) { 222408e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 222508e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22263fd0a558SYan, Zheng if (ret) { 22279e6a0c52SJosef Bacik err = ret; 22289e6a0c52SJosef Bacik goto out; 22293fd0a558SYan, Zheng } 22309e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 22319e6a0c52SJosef Bacik if (IS_ERR(trans)) { 22329e6a0c52SJosef Bacik err = PTR_ERR(trans); 22339e6a0c52SJosef Bacik trans = NULL; 22349e6a0c52SJosef Bacik goto out; 22359e6a0c52SJosef Bacik } 22369e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 22373fd0a558SYan, Zheng 22383fd0a558SYan, Zheng replaced = 0; 22395d4f98a2SYan Zheng max_level = level; 22405d4f98a2SYan Zheng 22415d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 22425d4f98a2SYan Zheng if (ret < 0) { 22435d4f98a2SYan Zheng err = ret; 22445d4f98a2SYan Zheng goto out; 22455d4f98a2SYan Zheng } 22465d4f98a2SYan Zheng if (ret > 0) 22475d4f98a2SYan Zheng break; 22485d4f98a2SYan Zheng 22495d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 22505d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 22515d4f98a2SYan Zheng ret = 0; 22525d4f98a2SYan Zheng } else { 22533fd0a558SYan, Zheng ret = replace_path(trans, root, reloc_root, path, 22543fd0a558SYan, Zheng &next_key, level, max_level); 22555d4f98a2SYan Zheng } 22565d4f98a2SYan Zheng if (ret < 0) { 22575d4f98a2SYan Zheng err = ret; 22585d4f98a2SYan Zheng goto out; 22595d4f98a2SYan Zheng } 22605d4f98a2SYan Zheng 22615d4f98a2SYan Zheng if (ret > 0) { 22625d4f98a2SYan Zheng level = ret; 22635d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 22645d4f98a2SYan Zheng path->slots[level]); 22655d4f98a2SYan Zheng replaced = 1; 22665d4f98a2SYan Zheng } 22675d4f98a2SYan Zheng 22685d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 22695d4f98a2SYan Zheng if (ret > 0) 22705d4f98a2SYan Zheng break; 22715d4f98a2SYan Zheng 22725d4f98a2SYan Zheng BUG_ON(level == 0); 22735d4f98a2SYan Zheng /* 22745d4f98a2SYan Zheng * save the merging progress in the drop_progress. 22755d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 22765d4f98a2SYan Zheng */ 22775d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 22785d4f98a2SYan Zheng path->slots[level]); 22795d4f98a2SYan Zheng root_item->drop_level = level; 22805d4f98a2SYan Zheng 22813a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 22829e6a0c52SJosef Bacik trans = NULL; 22835d4f98a2SYan Zheng 22842ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 22855d4f98a2SYan Zheng 22865d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22875d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22885d4f98a2SYan Zheng } 22895d4f98a2SYan Zheng 22905d4f98a2SYan Zheng /* 22915d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 22925d4f98a2SYan Zheng * relocated and the block is tree root. 22935d4f98a2SYan Zheng */ 22945d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 22955d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 22965d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 22975d4f98a2SYan Zheng free_extent_buffer(leaf); 22985d4f98a2SYan Zheng if (ret < 0) 22995d4f98a2SYan Zheng err = ret; 23005d4f98a2SYan Zheng out: 23015d4f98a2SYan Zheng btrfs_free_path(path); 23025d4f98a2SYan Zheng 23035d4f98a2SYan Zheng if (err == 0) { 23045d4f98a2SYan Zheng memset(&root_item->drop_progress, 0, 23055d4f98a2SYan Zheng sizeof(root_item->drop_progress)); 23065d4f98a2SYan Zheng root_item->drop_level = 0; 23075d4f98a2SYan Zheng btrfs_set_root_refs(root_item, 0); 23083fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 23095d4f98a2SYan Zheng } 23105d4f98a2SYan Zheng 23119e6a0c52SJosef Bacik if (trans) 23123a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23135d4f98a2SYan Zheng 23142ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23155d4f98a2SYan Zheng 23165d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23175d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23185d4f98a2SYan Zheng 23195d4f98a2SYan Zheng return err; 23205d4f98a2SYan Zheng } 23215d4f98a2SYan Zheng 23223fd0a558SYan, Zheng static noinline_for_stack 23233fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 23245d4f98a2SYan Zheng { 23253fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 23260b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23273fd0a558SYan, Zheng struct btrfs_root *reloc_root; 23285d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 23293fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23303fd0a558SYan, Zheng u64 num_bytes = 0; 23313fd0a558SYan, Zheng int ret; 23323fd0a558SYan, Zheng 23330b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23340b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23353fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 23360b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 23377585717fSChris Mason 23383fd0a558SYan, Zheng again: 23393fd0a558SYan, Zheng if (!err) { 23403fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 234108e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 234208e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23433fd0a558SYan, Zheng if (ret) 23443fd0a558SYan, Zheng err = ret; 23453fd0a558SYan, Zheng } 23463fd0a558SYan, Zheng 23477a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 23483612b495STsutomu Itoh if (IS_ERR(trans)) { 23493612b495STsutomu Itoh if (!err) 23502ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23512ff7e61eSJeff Mahoney num_bytes); 23523612b495STsutomu Itoh return PTR_ERR(trans); 23533612b495STsutomu Itoh } 23543fd0a558SYan, Zheng 23553fd0a558SYan, Zheng if (!err) { 23563fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 23573a45bb20SJeff Mahoney btrfs_end_transaction(trans); 23582ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23592ff7e61eSJeff Mahoney num_bytes); 23603fd0a558SYan, Zheng goto again; 23613fd0a558SYan, Zheng } 23623fd0a558SYan, Zheng } 23633fd0a558SYan, Zheng 23643fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 23653fd0a558SYan, Zheng 23663fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 23673fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 23683fd0a558SYan, Zheng struct btrfs_root, root_list); 23693fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 23703fd0a558SYan, Zheng 23710b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 23723fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 23733fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 23743fd0a558SYan, Zheng 23753fd0a558SYan, Zheng /* 23763fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 23773fd0a558SYan, Zheng * knows it should resumes merging 23783fd0a558SYan, Zheng */ 23793fd0a558SYan, Zheng if (!err) 23803fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 23813fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 23823fd0a558SYan, Zheng 23833fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 23843fd0a558SYan, Zheng } 23853fd0a558SYan, Zheng 23863fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 23873fd0a558SYan, Zheng 23883fd0a558SYan, Zheng if (!err) 23893a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 23903fd0a558SYan, Zheng else 23913a45bb20SJeff Mahoney btrfs_end_transaction(trans); 23923fd0a558SYan, Zheng return err; 23933fd0a558SYan, Zheng } 23943fd0a558SYan, Zheng 23953fd0a558SYan, Zheng static noinline_for_stack 2396aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2397aca1bba6SLiu Bo { 2398aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2399aca1bba6SLiu Bo 2400aca1bba6SLiu Bo while (!list_empty(list)) { 2401aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2402aca1bba6SLiu Bo root_list); 2403bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 24046bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 24056bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 24066bdf131fSJosef Bacik reloc_root->node = NULL; 24076bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2408aca1bba6SLiu Bo } 2409aca1bba6SLiu Bo } 2410aca1bba6SLiu Bo 2411aca1bba6SLiu Bo static noinline_for_stack 241294404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 24133fd0a558SYan, Zheng { 24140b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 24155d4f98a2SYan Zheng struct btrfs_root *root; 24165d4f98a2SYan Zheng struct btrfs_root *reloc_root; 24173fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24183fd0a558SYan, Zheng int found = 0; 2419aca1bba6SLiu Bo int ret = 0; 24203fd0a558SYan, Zheng again: 24213fd0a558SYan, Zheng root = rc->extent_root; 24227585717fSChris Mason 24237585717fSChris Mason /* 24247585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 24257585717fSChris Mason * we have to make sure nobody is in the middle of 24267585717fSChris Mason * adding their roots to the list while we are 24277585717fSChris Mason * doing this splice 24287585717fSChris Mason */ 24290b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24303fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 24310b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24325d4f98a2SYan Zheng 24333fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 24343fd0a558SYan, Zheng found = 1; 24353fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 24363fd0a558SYan, Zheng struct btrfs_root, root_list); 24375d4f98a2SYan Zheng 24385d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 24390b246afaSJeff Mahoney root = read_fs_root(fs_info, 24405d4f98a2SYan Zheng reloc_root->root_key.offset); 24415d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24425d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24435d4f98a2SYan Zheng 24443fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2445b37b39cdSJosef Bacik if (ret) { 244625e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 244725e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 244825e293c2SWang Shilong &reloc_roots); 2449aca1bba6SLiu Bo goto out; 2450b37b39cdSJosef Bacik } 24513fd0a558SYan, Zheng } else { 24523fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24533fd0a558SYan, Zheng } 24545bc7247aSMiao Xie 24552c536799SJeff Mahoney ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1); 2456aca1bba6SLiu Bo if (ret < 0) { 2457aca1bba6SLiu Bo if (list_empty(&reloc_root->root_list)) 2458aca1bba6SLiu Bo list_add_tail(&reloc_root->root_list, 2459aca1bba6SLiu Bo &reloc_roots); 2460aca1bba6SLiu Bo goto out; 2461aca1bba6SLiu Bo } 24625d4f98a2SYan Zheng } 24635d4f98a2SYan Zheng 24643fd0a558SYan, Zheng if (found) { 24653fd0a558SYan, Zheng found = 0; 24663fd0a558SYan, Zheng goto again; 24675d4f98a2SYan Zheng } 2468aca1bba6SLiu Bo out: 2469aca1bba6SLiu Bo if (ret) { 24700b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2471aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2472aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2473467bb1d2SWang Shilong 2474467bb1d2SWang Shilong /* new reloc root may be added */ 24750b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2476467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 24770b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2478467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2479467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2480aca1bba6SLiu Bo } 2481aca1bba6SLiu Bo 24825d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 24835d4f98a2SYan Zheng } 24845d4f98a2SYan Zheng 24855d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 24865d4f98a2SYan Zheng { 24875d4f98a2SYan Zheng struct tree_block *block; 24885d4f98a2SYan Zheng struct rb_node *rb_node; 24895d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 24905d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 24915d4f98a2SYan Zheng rb_erase(rb_node, blocks); 24925d4f98a2SYan Zheng kfree(block); 24935d4f98a2SYan Zheng } 24945d4f98a2SYan Zheng } 24955d4f98a2SYan Zheng 24965d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 24975d4f98a2SYan Zheng struct btrfs_root *reloc_root) 24985d4f98a2SYan Zheng { 24990b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 25005d4f98a2SYan Zheng struct btrfs_root *root; 25015d4f98a2SYan Zheng 25025d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 25035d4f98a2SYan Zheng return 0; 25045d4f98a2SYan Zheng 25050b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25065d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25075d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25085d4f98a2SYan Zheng 25095d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 25105d4f98a2SYan Zheng } 25115d4f98a2SYan Zheng 25123fd0a558SYan, Zheng static noinline_for_stack 25133fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 25143fd0a558SYan, Zheng struct reloc_control *rc, 25155d4f98a2SYan Zheng struct backref_node *node, 2516dc4103f9SWang Shilong struct backref_edge *edges[]) 25175d4f98a2SYan Zheng { 25185d4f98a2SYan Zheng struct backref_node *next; 25195d4f98a2SYan Zheng struct btrfs_root *root; 25203fd0a558SYan, Zheng int index = 0; 25213fd0a558SYan, Zheng 25225d4f98a2SYan Zheng next = node; 25235d4f98a2SYan Zheng while (1) { 25245d4f98a2SYan Zheng cond_resched(); 25255d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 25265d4f98a2SYan Zheng root = next->root; 25273fd0a558SYan, Zheng BUG_ON(!root); 252827cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 25295d4f98a2SYan Zheng 25305d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 25315d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 25325d4f98a2SYan Zheng break; 25335d4f98a2SYan Zheng } 25345d4f98a2SYan Zheng 25355d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 25363fd0a558SYan, Zheng root = root->reloc_root; 25373fd0a558SYan, Zheng 25383fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 25393fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 25403fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 25413fd0a558SYan, Zheng next->new_bytenr = root->node->start; 25423fd0a558SYan, Zheng next->root = root; 25433fd0a558SYan, Zheng list_add_tail(&next->list, 25443fd0a558SYan, Zheng &rc->backref_cache.changed); 25453fd0a558SYan, Zheng __mark_block_processed(rc, next); 25465d4f98a2SYan Zheng break; 25475d4f98a2SYan Zheng } 25485d4f98a2SYan Zheng 25493fd0a558SYan, Zheng WARN_ON(1); 25505d4f98a2SYan Zheng root = NULL; 25515d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25525d4f98a2SYan Zheng if (!next || next->level <= node->level) 25535d4f98a2SYan Zheng break; 25545d4f98a2SYan Zheng } 25553fd0a558SYan, Zheng if (!root) 25563fd0a558SYan, Zheng return NULL; 25575d4f98a2SYan Zheng 25583fd0a558SYan, Zheng next = node; 25593fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 25603fd0a558SYan, Zheng while (1) { 25613fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 25623fd0a558SYan, Zheng if (--index < 0) 25633fd0a558SYan, Zheng break; 25643fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 25653fd0a558SYan, Zheng } 25665d4f98a2SYan Zheng return root; 25675d4f98a2SYan Zheng } 25685d4f98a2SYan Zheng 25693fd0a558SYan, Zheng /* 25703fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 25713fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 25723fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 25733fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 25743fd0a558SYan, Zheng */ 25755d4f98a2SYan Zheng static noinline_for_stack 2576147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 25775d4f98a2SYan Zheng { 25783fd0a558SYan, Zheng struct backref_node *next; 25793fd0a558SYan, Zheng struct btrfs_root *root; 25803fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 25815d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25823fd0a558SYan, Zheng int index = 0; 25833fd0a558SYan, Zheng 25843fd0a558SYan, Zheng next = node; 25853fd0a558SYan, Zheng while (1) { 25863fd0a558SYan, Zheng cond_resched(); 25873fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 25883fd0a558SYan, Zheng root = next->root; 25893fd0a558SYan, Zheng BUG_ON(!root); 25903fd0a558SYan, Zheng 259125985edcSLucas De Marchi /* no other choice for non-references counted tree */ 259227cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 25933fd0a558SYan, Zheng return root; 25943fd0a558SYan, Zheng 25953fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 25963fd0a558SYan, Zheng fs_root = root; 25973fd0a558SYan, Zheng 25983fd0a558SYan, Zheng if (next != node) 25993fd0a558SYan, Zheng return NULL; 26003fd0a558SYan, Zheng 26013fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26023fd0a558SYan, Zheng if (!next || next->level <= node->level) 26033fd0a558SYan, Zheng break; 26043fd0a558SYan, Zheng } 26053fd0a558SYan, Zheng 26063fd0a558SYan, Zheng if (!fs_root) 26073fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 26083fd0a558SYan, Zheng return fs_root; 26095d4f98a2SYan Zheng } 26105d4f98a2SYan Zheng 26115d4f98a2SYan Zheng static noinline_for_stack 26123fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 26133fd0a558SYan, Zheng struct backref_node *node, int reserve) 26145d4f98a2SYan Zheng { 26150b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 26163fd0a558SYan, Zheng struct backref_node *next = node; 26173fd0a558SYan, Zheng struct backref_edge *edge; 26183fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26193fd0a558SYan, Zheng u64 num_bytes = 0; 26203fd0a558SYan, Zheng int index = 0; 26215d4f98a2SYan Zheng 26223fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 26233fd0a558SYan, Zheng 26243fd0a558SYan, Zheng while (next) { 26253fd0a558SYan, Zheng cond_resched(); 26265d4f98a2SYan Zheng while (1) { 26273fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 26285d4f98a2SYan Zheng break; 26295d4f98a2SYan Zheng 26300b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 26313fd0a558SYan, Zheng 26323fd0a558SYan, Zheng if (list_empty(&next->upper)) 26333fd0a558SYan, Zheng break; 26343fd0a558SYan, Zheng 26353fd0a558SYan, Zheng edge = list_entry(next->upper.next, 26363fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 26373fd0a558SYan, Zheng edges[index++] = edge; 26383fd0a558SYan, Zheng next = edge->node[UPPER]; 26395d4f98a2SYan Zheng } 26403fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26413fd0a558SYan, Zheng } 26423fd0a558SYan, Zheng return num_bytes; 26433fd0a558SYan, Zheng } 26443fd0a558SYan, Zheng 26453fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 26463fd0a558SYan, Zheng struct reloc_control *rc, 26473fd0a558SYan, Zheng struct backref_node *node) 26483fd0a558SYan, Zheng { 26493fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2650da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 26513fd0a558SYan, Zheng u64 num_bytes; 26523fd0a558SYan, Zheng int ret; 26530647bf56SWang Shilong u64 tmp; 26543fd0a558SYan, Zheng 26553fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 26563fd0a558SYan, Zheng 26573fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 26580647bf56SWang Shilong rc->reserved_bytes += num_bytes; 26598ca17f0fSJosef Bacik 26608ca17f0fSJosef Bacik /* 26618ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 26628ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 26638ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 26648ca17f0fSJosef Bacik */ 26650647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 26668ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 26673fd0a558SYan, Zheng if (ret) { 2668da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 26690647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 26700647bf56SWang Shilong tmp <<= 1; 26710647bf56SWang Shilong /* 26720647bf56SWang Shilong * only one thread can access block_rsv at this point, 26730647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 26740647bf56SWang Shilong * we expand more reservation size here to allow enough 26758ca17f0fSJosef Bacik * space for relocation and we will return eailer in 26760647bf56SWang Shilong * enospc case. 26770647bf56SWang Shilong */ 2678da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 26790647bf56SWang Shilong RELOCATION_RESERVED_NODES; 26808ca17f0fSJosef Bacik return -EAGAIN; 26813fd0a558SYan, Zheng } 26823fd0a558SYan, Zheng 26833fd0a558SYan, Zheng return 0; 26843fd0a558SYan, Zheng } 26853fd0a558SYan, Zheng 26865d4f98a2SYan Zheng /* 26875d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 26885d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 26895d4f98a2SYan Zheng * 26905d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 26915d4f98a2SYan Zheng * in that case this function just updates pointers. 26925d4f98a2SYan Zheng */ 26935d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 26943fd0a558SYan, Zheng struct reloc_control *rc, 26955d4f98a2SYan Zheng struct backref_node *node, 26965d4f98a2SYan Zheng struct btrfs_key *key, 26975d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 26985d4f98a2SYan Zheng { 26992ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27005d4f98a2SYan Zheng struct backref_node *upper; 27015d4f98a2SYan Zheng struct backref_edge *edge; 27025d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27035d4f98a2SYan Zheng struct btrfs_root *root; 27045d4f98a2SYan Zheng struct extent_buffer *eb; 27055d4f98a2SYan Zheng u32 blocksize; 27065d4f98a2SYan Zheng u64 bytenr; 27075d4f98a2SYan Zheng u64 generation; 27085d4f98a2SYan Zheng int slot; 27095d4f98a2SYan Zheng int ret; 27105d4f98a2SYan Zheng int err = 0; 27115d4f98a2SYan Zheng 27125d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 27135d4f98a2SYan Zheng 27145d4f98a2SYan Zheng path->lowest_level = node->level + 1; 27153fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 27165d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 27175d4f98a2SYan Zheng cond_resched(); 27185d4f98a2SYan Zheng 27195d4f98a2SYan Zheng upper = edge->node[UPPER]; 2720dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 27213fd0a558SYan, Zheng BUG_ON(!root); 27225d4f98a2SYan Zheng 27233fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 27243fd0a558SYan, Zheng if (!lowest) { 27253fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 27263fd0a558SYan, Zheng upper->level, &slot); 27273fd0a558SYan, Zheng BUG_ON(ret); 27283fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27293fd0a558SYan, Zheng if (node->eb->start == bytenr) 27303fd0a558SYan, Zheng goto next; 27313fd0a558SYan, Zheng } 27325d4f98a2SYan Zheng drop_node_buffer(upper); 27333fd0a558SYan, Zheng } 27345d4f98a2SYan Zheng 27355d4f98a2SYan Zheng if (!upper->eb) { 27365d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 27373561b9dbSLiu Bo if (ret) { 27383561b9dbSLiu Bo if (ret < 0) 27395d4f98a2SYan Zheng err = ret; 27403561b9dbSLiu Bo else 27413561b9dbSLiu Bo err = -ENOENT; 27423561b9dbSLiu Bo 27433561b9dbSLiu Bo btrfs_release_path(path); 27445d4f98a2SYan Zheng break; 27455d4f98a2SYan Zheng } 27465d4f98a2SYan Zheng 27473fd0a558SYan, Zheng if (!upper->eb) { 27483fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 27493fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 27503fd0a558SYan, Zheng } else { 27513fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 27523fd0a558SYan, Zheng } 27533fd0a558SYan, Zheng 27543fd0a558SYan, Zheng upper->locked = 1; 27553fd0a558SYan, Zheng path->locks[upper->level] = 0; 27563fd0a558SYan, Zheng 27575d4f98a2SYan Zheng slot = path->slots[upper->level]; 2758b3b4aa74SDavid Sterba btrfs_release_path(path); 27595d4f98a2SYan Zheng } else { 27605d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 27615d4f98a2SYan Zheng &slot); 27625d4f98a2SYan Zheng BUG_ON(ret); 27635d4f98a2SYan Zheng } 27645d4f98a2SYan Zheng 27655d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27663fd0a558SYan, Zheng if (lowest) { 27674547f4d8SLiu Bo if (bytenr != node->bytenr) { 27684547f4d8SLiu Bo btrfs_err(root->fs_info, 27694547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 27704547f4d8SLiu Bo bytenr, node->bytenr, slot, 27714547f4d8SLiu Bo upper->eb->start); 27724547f4d8SLiu Bo err = -EIO; 27734547f4d8SLiu Bo goto next; 27744547f4d8SLiu Bo } 27755d4f98a2SYan Zheng } else { 27763fd0a558SYan, Zheng if (node->eb->start == bytenr) 27773fd0a558SYan, Zheng goto next; 27785d4f98a2SYan Zheng } 27795d4f98a2SYan Zheng 2780da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 27815d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 27822ff7e61eSJeff Mahoney eb = read_tree_block(fs_info, bytenr, generation); 278364c043deSLiu Bo if (IS_ERR(eb)) { 278464c043deSLiu Bo err = PTR_ERR(eb); 278564c043deSLiu Bo goto next; 278664c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2787416bc658SJosef Bacik free_extent_buffer(eb); 278897d9a8a4STsutomu Itoh err = -EIO; 278997d9a8a4STsutomu Itoh goto next; 279097d9a8a4STsutomu Itoh } 27915d4f98a2SYan Zheng btrfs_tree_lock(eb); 27925d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 27935d4f98a2SYan Zheng 27945d4f98a2SYan Zheng if (!node->eb) { 27955d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 27965d4f98a2SYan Zheng slot, &eb); 27973fd0a558SYan, Zheng btrfs_tree_unlock(eb); 27983fd0a558SYan, Zheng free_extent_buffer(eb); 27995d4f98a2SYan Zheng if (ret < 0) { 28005d4f98a2SYan Zheng err = ret; 28013fd0a558SYan, Zheng goto next; 28025d4f98a2SYan Zheng } 28033fd0a558SYan, Zheng BUG_ON(node->eb != eb); 28045d4f98a2SYan Zheng } else { 28055d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 28065d4f98a2SYan Zheng node->eb->start); 28075d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 28085d4f98a2SYan Zheng trans->transid); 28095d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 28105d4f98a2SYan Zheng 281184f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, root, 28125d4f98a2SYan Zheng node->eb->start, blocksize, 28135d4f98a2SYan Zheng upper->eb->start, 28145d4f98a2SYan Zheng btrfs_header_owner(upper->eb), 2815b06c4bf5SFilipe Manana node->level, 0); 28165d4f98a2SYan Zheng BUG_ON(ret); 28175d4f98a2SYan Zheng 28185d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 28195d4f98a2SYan Zheng BUG_ON(ret); 28205d4f98a2SYan Zheng } 28213fd0a558SYan, Zheng next: 28223fd0a558SYan, Zheng if (!upper->pending) 28233fd0a558SYan, Zheng drop_node_buffer(upper); 28243fd0a558SYan, Zheng else 28253fd0a558SYan, Zheng unlock_node_buffer(upper); 28263fd0a558SYan, Zheng if (err) 28273fd0a558SYan, Zheng break; 28285d4f98a2SYan Zheng } 28293fd0a558SYan, Zheng 28303fd0a558SYan, Zheng if (!err && node->pending) { 28313fd0a558SYan, Zheng drop_node_buffer(node); 28323fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 28333fd0a558SYan, Zheng node->pending = 0; 28345d4f98a2SYan Zheng } 28353fd0a558SYan, Zheng 28365d4f98a2SYan Zheng path->lowest_level = 0; 28373fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 28385d4f98a2SYan Zheng return err; 28395d4f98a2SYan Zheng } 28405d4f98a2SYan Zheng 28415d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 28423fd0a558SYan, Zheng struct reloc_control *rc, 28435d4f98a2SYan Zheng struct backref_node *node, 28445d4f98a2SYan Zheng struct btrfs_path *path) 28455d4f98a2SYan Zheng { 28465d4f98a2SYan Zheng struct btrfs_key key; 28475d4f98a2SYan Zheng 28485d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 28493fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 28505d4f98a2SYan Zheng } 28515d4f98a2SYan Zheng 28525d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 28533fd0a558SYan, Zheng struct reloc_control *rc, 28543fd0a558SYan, Zheng struct btrfs_path *path, int err) 28555d4f98a2SYan Zheng { 28563fd0a558SYan, Zheng LIST_HEAD(list); 28573fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 28585d4f98a2SYan Zheng struct backref_node *node; 28595d4f98a2SYan Zheng int level; 28605d4f98a2SYan Zheng int ret; 28615d4f98a2SYan Zheng 28625d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 28635d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 28645d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 28653fd0a558SYan, Zheng struct backref_node, list); 28663fd0a558SYan, Zheng list_move_tail(&node->list, &list); 28673fd0a558SYan, Zheng BUG_ON(!node->pending); 28685d4f98a2SYan Zheng 28693fd0a558SYan, Zheng if (!err) { 28703fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 28715d4f98a2SYan Zheng if (ret < 0) 28725d4f98a2SYan Zheng err = ret; 28735d4f98a2SYan Zheng } 28745d4f98a2SYan Zheng } 28753fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 28763fd0a558SYan, Zheng } 28775d4f98a2SYan Zheng return err; 28785d4f98a2SYan Zheng } 28795d4f98a2SYan Zheng 28805d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 28813fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 28823fd0a558SYan, Zheng { 28833fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2884ceeb0ae7SDavid Sterba EXTENT_DIRTY); 28853fd0a558SYan, Zheng } 28863fd0a558SYan, Zheng 28873fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 28885d4f98a2SYan Zheng struct backref_node *node) 28895d4f98a2SYan Zheng { 28905d4f98a2SYan Zheng u32 blocksize; 28915d4f98a2SYan Zheng if (node->level == 0 || 28925d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2893da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 28943fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 28955d4f98a2SYan Zheng } 28965d4f98a2SYan Zheng node->processed = 1; 28975d4f98a2SYan Zheng } 28985d4f98a2SYan Zheng 28995d4f98a2SYan Zheng /* 29005d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 29015d4f98a2SYan Zheng * as processed. 29025d4f98a2SYan Zheng */ 29035d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 29045d4f98a2SYan Zheng struct backref_node *node) 29055d4f98a2SYan Zheng { 29065d4f98a2SYan Zheng struct backref_node *next = node; 29075d4f98a2SYan Zheng struct backref_edge *edge; 29085d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 29095d4f98a2SYan Zheng int index = 0; 29105d4f98a2SYan Zheng 29115d4f98a2SYan Zheng while (next) { 29125d4f98a2SYan Zheng cond_resched(); 29135d4f98a2SYan Zheng while (1) { 29145d4f98a2SYan Zheng if (next->processed) 29155d4f98a2SYan Zheng break; 29165d4f98a2SYan Zheng 29173fd0a558SYan, Zheng __mark_block_processed(rc, next); 29185d4f98a2SYan Zheng 29195d4f98a2SYan Zheng if (list_empty(&next->upper)) 29205d4f98a2SYan Zheng break; 29215d4f98a2SYan Zheng 29225d4f98a2SYan Zheng edge = list_entry(next->upper.next, 29235d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 29245d4f98a2SYan Zheng edges[index++] = edge; 29255d4f98a2SYan Zheng next = edge->node[UPPER]; 29265d4f98a2SYan Zheng } 29275d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 29285d4f98a2SYan Zheng } 29295d4f98a2SYan Zheng } 29305d4f98a2SYan Zheng 29317476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 29325d4f98a2SYan Zheng { 2933da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 29347476dfdaSDavid Sterba 29355d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 29369655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 29375d4f98a2SYan Zheng return 1; 29385d4f98a2SYan Zheng return 0; 29395d4f98a2SYan Zheng } 29405d4f98a2SYan Zheng 29412ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 29425d4f98a2SYan Zheng struct tree_block *block) 29435d4f98a2SYan Zheng { 29445d4f98a2SYan Zheng struct extent_buffer *eb; 29455d4f98a2SYan Zheng 29465d4f98a2SYan Zheng BUG_ON(block->key_ready); 29472ff7e61eSJeff Mahoney eb = read_tree_block(fs_info, block->bytenr, block->key.offset); 294864c043deSLiu Bo if (IS_ERR(eb)) { 294964c043deSLiu Bo return PTR_ERR(eb); 295064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2951416bc658SJosef Bacik free_extent_buffer(eb); 2952416bc658SJosef Bacik return -EIO; 2953416bc658SJosef Bacik } 29545d4f98a2SYan Zheng WARN_ON(btrfs_header_level(eb) != block->level); 29555d4f98a2SYan Zheng if (block->level == 0) 29565d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 29575d4f98a2SYan Zheng else 29585d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 29595d4f98a2SYan Zheng free_extent_buffer(eb); 29605d4f98a2SYan Zheng block->key_ready = 1; 29615d4f98a2SYan Zheng return 0; 29625d4f98a2SYan Zheng } 29635d4f98a2SYan Zheng 29645d4f98a2SYan Zheng /* 29655d4f98a2SYan Zheng * helper function to relocate a tree block 29665d4f98a2SYan Zheng */ 29675d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 29685d4f98a2SYan Zheng struct reloc_control *rc, 29695d4f98a2SYan Zheng struct backref_node *node, 29705d4f98a2SYan Zheng struct btrfs_key *key, 29715d4f98a2SYan Zheng struct btrfs_path *path) 29725d4f98a2SYan Zheng { 29735d4f98a2SYan Zheng struct btrfs_root *root; 29743fd0a558SYan, Zheng int ret = 0; 29755d4f98a2SYan Zheng 29763fd0a558SYan, Zheng if (!node) 29775d4f98a2SYan Zheng return 0; 29783fd0a558SYan, Zheng 29793fd0a558SYan, Zheng BUG_ON(node->processed); 2980147d256eSZhaolei root = select_one_root(node); 29813fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 29823fd0a558SYan, Zheng update_processed_blocks(rc, node); 29833fd0a558SYan, Zheng goto out; 29845d4f98a2SYan Zheng } 29855d4f98a2SYan Zheng 298627cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29873fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 29883fd0a558SYan, Zheng if (ret) 29895d4f98a2SYan Zheng goto out; 29905d4f98a2SYan Zheng } 29913fd0a558SYan, Zheng 29923fd0a558SYan, Zheng if (root) { 299327cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29943fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 29953fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 29963fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 29973fd0a558SYan, Zheng root = root->reloc_root; 29983fd0a558SYan, Zheng node->new_bytenr = root->node->start; 29993fd0a558SYan, Zheng node->root = root; 30003fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 30013fd0a558SYan, Zheng } else { 30025d4f98a2SYan Zheng path->lowest_level = node->level; 30035d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3004b3b4aa74SDavid Sterba btrfs_release_path(path); 30053fd0a558SYan, Zheng if (ret > 0) 30065d4f98a2SYan Zheng ret = 0; 30073fd0a558SYan, Zheng } 30083fd0a558SYan, Zheng if (!ret) 30093fd0a558SYan, Zheng update_processed_blocks(rc, node); 30103fd0a558SYan, Zheng } else { 30113fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 30123fd0a558SYan, Zheng } 30135d4f98a2SYan Zheng out: 30140647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 30153fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 30165d4f98a2SYan Zheng return ret; 30175d4f98a2SYan Zheng } 30185d4f98a2SYan Zheng 30195d4f98a2SYan Zheng /* 30205d4f98a2SYan Zheng * relocate a list of blocks 30215d4f98a2SYan Zheng */ 30225d4f98a2SYan Zheng static noinline_for_stack 30235d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 30245d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 30255d4f98a2SYan Zheng { 30262ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 30275d4f98a2SYan Zheng struct backref_node *node; 30285d4f98a2SYan Zheng struct btrfs_path *path; 30295d4f98a2SYan Zheng struct tree_block *block; 30305d4f98a2SYan Zheng struct rb_node *rb_node; 30315d4f98a2SYan Zheng int ret; 30325d4f98a2SYan Zheng int err = 0; 30335d4f98a2SYan Zheng 30345d4f98a2SYan Zheng path = btrfs_alloc_path(); 3035e1a12670SLiu Bo if (!path) { 3036e1a12670SLiu Bo err = -ENOMEM; 303734c2b290SDavid Sterba goto out_free_blocks; 3038e1a12670SLiu Bo } 30395d4f98a2SYan Zheng 30405d4f98a2SYan Zheng rb_node = rb_first(blocks); 30415d4f98a2SYan Zheng while (rb_node) { 30425d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 30435d4f98a2SYan Zheng if (!block->key_ready) 30442ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 30455d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30465d4f98a2SYan Zheng } 30475d4f98a2SYan Zheng 30485d4f98a2SYan Zheng rb_node = rb_first(blocks); 30495d4f98a2SYan Zheng while (rb_node) { 30505d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 305134c2b290SDavid Sterba if (!block->key_ready) { 30522ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 305334c2b290SDavid Sterba if (err) 305434c2b290SDavid Sterba goto out_free_path; 305534c2b290SDavid Sterba } 30565d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30575d4f98a2SYan Zheng } 30585d4f98a2SYan Zheng 30595d4f98a2SYan Zheng rb_node = rb_first(blocks); 30605d4f98a2SYan Zheng while (rb_node) { 30615d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 30625d4f98a2SYan Zheng 30633fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 30645d4f98a2SYan Zheng block->level, block->bytenr); 30655d4f98a2SYan Zheng if (IS_ERR(node)) { 30665d4f98a2SYan Zheng err = PTR_ERR(node); 30675d4f98a2SYan Zheng goto out; 30685d4f98a2SYan Zheng } 30695d4f98a2SYan Zheng 30705d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 30715d4f98a2SYan Zheng path); 30725d4f98a2SYan Zheng if (ret < 0) { 30733fd0a558SYan, Zheng if (ret != -EAGAIN || rb_node == rb_first(blocks)) 30745d4f98a2SYan Zheng err = ret; 30755d4f98a2SYan Zheng goto out; 30765d4f98a2SYan Zheng } 30775d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30785d4f98a2SYan Zheng } 30795d4f98a2SYan Zheng out: 30803fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 30815d4f98a2SYan Zheng 308234c2b290SDavid Sterba out_free_path: 30835d4f98a2SYan Zheng btrfs_free_path(path); 308434c2b290SDavid Sterba out_free_blocks: 3085e1a12670SLiu Bo free_block_list(blocks); 30865d4f98a2SYan Zheng return err; 30875d4f98a2SYan Zheng } 30885d4f98a2SYan Zheng 30895d4f98a2SYan Zheng static noinline_for_stack 3090efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3091efa56464SYan, Zheng struct file_extent_cluster *cluster) 3092efa56464SYan, Zheng { 3093efa56464SYan, Zheng u64 alloc_hint = 0; 3094efa56464SYan, Zheng u64 start; 3095efa56464SYan, Zheng u64 end; 3096efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3097efa56464SYan, Zheng u64 num_bytes; 3098efa56464SYan, Zheng int nr = 0; 3099efa56464SYan, Zheng int ret = 0; 3100dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3101dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 310218513091SWang Xiaoguang u64 cur_offset; 3103364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3104efa56464SYan, Zheng 3105efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 31065955102cSAl Viro inode_lock(inode); 3107efa56464SYan, Zheng 3108364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3109dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3110efa56464SYan, Zheng if (ret) 3111efa56464SYan, Zheng goto out; 3112efa56464SYan, Zheng 311318513091SWang Xiaoguang cur_offset = prealloc_start; 3114efa56464SYan, Zheng while (nr < cluster->nr) { 3115efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3116efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3117efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3118efa56464SYan, Zheng else 3119efa56464SYan, Zheng end = cluster->end - offset; 3120efa56464SYan, Zheng 3121d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3122efa56464SYan, Zheng num_bytes = end + 1 - start; 312318513091SWang Xiaoguang if (cur_offset < start) 3124bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3125bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3126efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3127efa56464SYan, Zheng num_bytes, num_bytes, 3128efa56464SYan, Zheng end + 1, &alloc_hint); 312918513091SWang Xiaoguang cur_offset = end + 1; 3130d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3131efa56464SYan, Zheng if (ret) 3132efa56464SYan, Zheng break; 3133efa56464SYan, Zheng nr++; 3134efa56464SYan, Zheng } 313518513091SWang Xiaoguang if (cur_offset < prealloc_end) 3136bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3137bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3138efa56464SYan, Zheng out: 31395955102cSAl Viro inode_unlock(inode); 3140364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3141efa56464SYan, Zheng return ret; 3142efa56464SYan, Zheng } 3143efa56464SYan, Zheng 3144efa56464SYan, Zheng static noinline_for_stack 31450257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 31460257bb82SYan, Zheng u64 block_start) 31475d4f98a2SYan Zheng { 31480b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 31495d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 31505d4f98a2SYan Zheng struct extent_map *em; 31510257bb82SYan, Zheng int ret = 0; 31525d4f98a2SYan Zheng 3153172ddd60SDavid Sterba em = alloc_extent_map(); 31540257bb82SYan, Zheng if (!em) 31550257bb82SYan, Zheng return -ENOMEM; 31560257bb82SYan, Zheng 31575d4f98a2SYan Zheng em->start = start; 31580257bb82SYan, Zheng em->len = end + 1 - start; 31590257bb82SYan, Zheng em->block_len = em->len; 31600257bb82SYan, Zheng em->block_start = block_start; 31610b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 31625d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 31635d4f98a2SYan Zheng 3164d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 31655d4f98a2SYan Zheng while (1) { 3166890871beSChris Mason write_lock(&em_tree->lock); 316709a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3168890871beSChris Mason write_unlock(&em_tree->lock); 31695d4f98a2SYan Zheng if (ret != -EEXIST) { 31705d4f98a2SYan Zheng free_extent_map(em); 31715d4f98a2SYan Zheng break; 31725d4f98a2SYan Zheng } 3173dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 31745d4f98a2SYan Zheng } 3175d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 31760257bb82SYan, Zheng return ret; 31770257bb82SYan, Zheng } 31785d4f98a2SYan Zheng 31790257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 31800257bb82SYan, Zheng struct file_extent_cluster *cluster) 31810257bb82SYan, Zheng { 31822ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 31830257bb82SYan, Zheng u64 page_start; 31840257bb82SYan, Zheng u64 page_end; 31850257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 31860257bb82SYan, Zheng unsigned long index; 31870257bb82SYan, Zheng unsigned long last_index; 31880257bb82SYan, Zheng struct page *page; 31890257bb82SYan, Zheng struct file_ra_state *ra; 31903b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 31910257bb82SYan, Zheng int nr = 0; 31920257bb82SYan, Zheng int ret = 0; 31930257bb82SYan, Zheng 31940257bb82SYan, Zheng if (!cluster->nr) 31950257bb82SYan, Zheng return 0; 31960257bb82SYan, Zheng 31970257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 31980257bb82SYan, Zheng if (!ra) 31990257bb82SYan, Zheng return -ENOMEM; 32000257bb82SYan, Zheng 3201efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 32020257bb82SYan, Zheng if (ret) 3203efa56464SYan, Zheng goto out; 32040257bb82SYan, Zheng 32050257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 32060257bb82SYan, Zheng 3207efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3208efa56464SYan, Zheng cluster->end - offset, cluster->start); 3209efa56464SYan, Zheng if (ret) 3210efa56464SYan, Zheng goto out; 3211efa56464SYan, Zheng 321209cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 321309cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 32140257bb82SYan, Zheng while (index <= last_index) { 32159f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 32169f3db423SNikolay Borisov PAGE_SIZE); 3217efa56464SYan, Zheng if (ret) 3218efa56464SYan, Zheng goto out; 3219efa56464SYan, Zheng 32200257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 32210257bb82SYan, Zheng if (!page) { 32220257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 32230257bb82SYan, Zheng ra, NULL, index, 32240257bb82SYan, Zheng last_index + 1 - index); 3225a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 32263b16a4e3SJosef Bacik mask); 32270257bb82SYan, Zheng if (!page) { 3228691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 322909cbfeafSKirill A. Shutemov PAGE_SIZE); 32300257bb82SYan, Zheng ret = -ENOMEM; 3231efa56464SYan, Zheng goto out; 32320257bb82SYan, Zheng } 32330257bb82SYan, Zheng } 32340257bb82SYan, Zheng 32350257bb82SYan, Zheng if (PageReadahead(page)) { 32360257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 32370257bb82SYan, Zheng ra, NULL, page, index, 32380257bb82SYan, Zheng last_index + 1 - index); 32390257bb82SYan, Zheng } 32400257bb82SYan, Zheng 32410257bb82SYan, Zheng if (!PageUptodate(page)) { 32420257bb82SYan, Zheng btrfs_readpage(NULL, page); 32430257bb82SYan, Zheng lock_page(page); 32440257bb82SYan, Zheng if (!PageUptodate(page)) { 32450257bb82SYan, Zheng unlock_page(page); 324609cbfeafSKirill A. Shutemov put_page(page); 3247691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 324809cbfeafSKirill A. Shutemov PAGE_SIZE); 32498b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 32508b62f87bSJosef Bacik PAGE_SIZE); 32510257bb82SYan, Zheng ret = -EIO; 3252efa56464SYan, Zheng goto out; 32530257bb82SYan, Zheng } 32540257bb82SYan, Zheng } 32550257bb82SYan, Zheng 32564eee4fa4SMiao Xie page_start = page_offset(page); 325709cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 32580257bb82SYan, Zheng 3259d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 32600257bb82SYan, Zheng 32610257bb82SYan, Zheng set_page_extent_mapped(page); 32620257bb82SYan, Zheng 32630257bb82SYan, Zheng if (nr < cluster->nr && 32640257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 32650257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 32660257bb82SYan, Zheng page_start, page_end, 3267ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 32680257bb82SYan, Zheng nr++; 32690257bb82SYan, Zheng } 32700257bb82SYan, Zheng 3271ba8b04c1SQu Wenruo btrfs_set_extent_delalloc(inode, page_start, page_end, NULL, 0); 32720257bb82SYan, Zheng set_page_dirty(page); 32730257bb82SYan, Zheng 32740257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3275d0082371SJeff Mahoney page_start, page_end); 32760257bb82SYan, Zheng unlock_page(page); 327709cbfeafSKirill A. Shutemov put_page(page); 32780257bb82SYan, Zheng 32790257bb82SYan, Zheng index++; 32808b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); 3281efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 32822ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 32830257bb82SYan, Zheng } 32840257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3285efa56464SYan, Zheng out: 32860257bb82SYan, Zheng kfree(ra); 32870257bb82SYan, Zheng return ret; 32880257bb82SYan, Zheng } 32890257bb82SYan, Zheng 32900257bb82SYan, Zheng static noinline_for_stack 32910257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 32920257bb82SYan, Zheng struct file_extent_cluster *cluster) 32930257bb82SYan, Zheng { 32940257bb82SYan, Zheng int ret; 32950257bb82SYan, Zheng 32960257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 32970257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 32980257bb82SYan, Zheng if (ret) 32990257bb82SYan, Zheng return ret; 33000257bb82SYan, Zheng cluster->nr = 0; 33010257bb82SYan, Zheng } 33020257bb82SYan, Zheng 33030257bb82SYan, Zheng if (!cluster->nr) 33040257bb82SYan, Zheng cluster->start = extent_key->objectid; 33050257bb82SYan, Zheng else 33060257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 33070257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 33080257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 33090257bb82SYan, Zheng cluster->nr++; 33100257bb82SYan, Zheng 33110257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 33120257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33130257bb82SYan, Zheng if (ret) 33140257bb82SYan, Zheng return ret; 33150257bb82SYan, Zheng cluster->nr = 0; 33160257bb82SYan, Zheng } 33170257bb82SYan, Zheng return 0; 33185d4f98a2SYan Zheng } 33195d4f98a2SYan Zheng 33205d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 33215d4f98a2SYan Zheng static int get_ref_objectid_v0(struct reloc_control *rc, 33225d4f98a2SYan Zheng struct btrfs_path *path, 33235d4f98a2SYan Zheng struct btrfs_key *extent_key, 33245d4f98a2SYan Zheng u64 *ref_objectid, int *path_change) 33255d4f98a2SYan Zheng { 33265d4f98a2SYan Zheng struct btrfs_key key; 33275d4f98a2SYan Zheng struct extent_buffer *leaf; 33285d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 33295d4f98a2SYan Zheng int ret; 33305d4f98a2SYan Zheng int slot; 33315d4f98a2SYan Zheng 33325d4f98a2SYan Zheng leaf = path->nodes[0]; 33335d4f98a2SYan Zheng slot = path->slots[0]; 33345d4f98a2SYan Zheng while (1) { 33355d4f98a2SYan Zheng if (slot >= btrfs_header_nritems(leaf)) { 33365d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 33375d4f98a2SYan Zheng if (ret < 0) 33385d4f98a2SYan Zheng return ret; 33395d4f98a2SYan Zheng BUG_ON(ret > 0); 33405d4f98a2SYan Zheng leaf = path->nodes[0]; 33415d4f98a2SYan Zheng slot = path->slots[0]; 33425d4f98a2SYan Zheng if (path_change) 33435d4f98a2SYan Zheng *path_change = 1; 33445d4f98a2SYan Zheng } 33455d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 33465d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 33475d4f98a2SYan Zheng return -ENOENT; 33485d4f98a2SYan Zheng 33495d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_REF_V0_KEY) { 33505d4f98a2SYan Zheng slot++; 33515d4f98a2SYan Zheng continue; 33525d4f98a2SYan Zheng } 33535d4f98a2SYan Zheng ref0 = btrfs_item_ptr(leaf, slot, 33545d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 33555d4f98a2SYan Zheng *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0); 33565d4f98a2SYan Zheng break; 33575d4f98a2SYan Zheng } 33585d4f98a2SYan Zheng return 0; 33595d4f98a2SYan Zheng } 33605d4f98a2SYan Zheng #endif 33615d4f98a2SYan Zheng 33625d4f98a2SYan Zheng /* 33635d4f98a2SYan Zheng * helper to add a tree block to the list. 33645d4f98a2SYan Zheng * the major work is getting the generation and level of the block 33655d4f98a2SYan Zheng */ 33665d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 33675d4f98a2SYan Zheng struct btrfs_key *extent_key, 33685d4f98a2SYan Zheng struct btrfs_path *path, 33695d4f98a2SYan Zheng struct rb_root *blocks) 33705d4f98a2SYan Zheng { 33715d4f98a2SYan Zheng struct extent_buffer *eb; 33725d4f98a2SYan Zheng struct btrfs_extent_item *ei; 33735d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 33745d4f98a2SYan Zheng struct tree_block *block; 33755d4f98a2SYan Zheng struct rb_node *rb_node; 33765d4f98a2SYan Zheng u32 item_size; 33775d4f98a2SYan Zheng int level = -1; 33787fdf4b60SWang Shilong u64 generation; 33795d4f98a2SYan Zheng 33805d4f98a2SYan Zheng eb = path->nodes[0]; 33815d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 33825d4f98a2SYan Zheng 33833173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 33843173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 33855d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 33865d4f98a2SYan Zheng struct btrfs_extent_item); 33873173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 33885d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 33895d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 33905d4f98a2SYan Zheng } else { 33913173a18fSJosef Bacik level = (int)extent_key->offset; 33923173a18fSJosef Bacik } 33933173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 33943173a18fSJosef Bacik } else { 33955d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 33965d4f98a2SYan Zheng u64 ref_owner; 33975d4f98a2SYan Zheng int ret; 33985d4f98a2SYan Zheng 33995d4f98a2SYan Zheng BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 34005d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, extent_key, 34015d4f98a2SYan Zheng &ref_owner, NULL); 3402411fc6bcSAndi Kleen if (ret < 0) 3403411fc6bcSAndi Kleen return ret; 34045d4f98a2SYan Zheng BUG_ON(ref_owner >= BTRFS_MAX_LEVEL); 34055d4f98a2SYan Zheng level = (int)ref_owner; 34065d4f98a2SYan Zheng /* FIXME: get real generation */ 34075d4f98a2SYan Zheng generation = 0; 34085d4f98a2SYan Zheng #else 34095d4f98a2SYan Zheng BUG(); 34105d4f98a2SYan Zheng #endif 34115d4f98a2SYan Zheng } 34125d4f98a2SYan Zheng 3413b3b4aa74SDavid Sterba btrfs_release_path(path); 34145d4f98a2SYan Zheng 34155d4f98a2SYan Zheng BUG_ON(level == -1); 34165d4f98a2SYan Zheng 34175d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 34185d4f98a2SYan Zheng if (!block) 34195d4f98a2SYan Zheng return -ENOMEM; 34205d4f98a2SYan Zheng 34215d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3422da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 34235d4f98a2SYan Zheng block->key.offset = generation; 34245d4f98a2SYan Zheng block->level = level; 34255d4f98a2SYan Zheng block->key_ready = 0; 34265d4f98a2SYan Zheng 34275d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 342843c04fb1SJeff Mahoney if (rb_node) 342943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 34305d4f98a2SYan Zheng 34315d4f98a2SYan Zheng return 0; 34325d4f98a2SYan Zheng } 34335d4f98a2SYan Zheng 34345d4f98a2SYan Zheng /* 34355d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 34365d4f98a2SYan Zheng */ 34375d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 34385d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 34395d4f98a2SYan Zheng struct rb_root *blocks) 34405d4f98a2SYan Zheng { 34410b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34425d4f98a2SYan Zheng struct btrfs_path *path; 34435d4f98a2SYan Zheng struct btrfs_key key; 34445d4f98a2SYan Zheng int ret; 34450b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 34465d4f98a2SYan Zheng 34477476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 34485d4f98a2SYan Zheng return 0; 34495d4f98a2SYan Zheng 34505d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 34515d4f98a2SYan Zheng return 0; 34525d4f98a2SYan Zheng 34535d4f98a2SYan Zheng path = btrfs_alloc_path(); 34545d4f98a2SYan Zheng if (!path) 34555d4f98a2SYan Zheng return -ENOMEM; 3456aee68ee5SJosef Bacik again: 34575d4f98a2SYan Zheng key.objectid = bytenr; 3458aee68ee5SJosef Bacik if (skinny) { 3459aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3460aee68ee5SJosef Bacik key.offset = (u64)-1; 3461aee68ee5SJosef Bacik } else { 34625d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34635d4f98a2SYan Zheng key.offset = blocksize; 3464aee68ee5SJosef Bacik } 34655d4f98a2SYan Zheng 34665d4f98a2SYan Zheng path->search_commit_root = 1; 34675d4f98a2SYan Zheng path->skip_locking = 1; 34685d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 34695d4f98a2SYan Zheng if (ret < 0) 34705d4f98a2SYan Zheng goto out; 34715d4f98a2SYan Zheng 3472aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3473aee68ee5SJosef Bacik if (path->slots[0]) { 3474aee68ee5SJosef Bacik path->slots[0]--; 3475aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3476aee68ee5SJosef Bacik path->slots[0]); 34773173a18fSJosef Bacik if (key.objectid == bytenr && 3478aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3479aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3480aee68ee5SJosef Bacik key.offset == blocksize))) 34813173a18fSJosef Bacik ret = 0; 34823173a18fSJosef Bacik } 3483aee68ee5SJosef Bacik 3484aee68ee5SJosef Bacik if (ret) { 3485aee68ee5SJosef Bacik skinny = false; 3486aee68ee5SJosef Bacik btrfs_release_path(path); 3487aee68ee5SJosef Bacik goto again; 3488aee68ee5SJosef Bacik } 3489aee68ee5SJosef Bacik } 3490cdccee99SLiu Bo if (ret) { 3491cdccee99SLiu Bo ASSERT(ret == 1); 3492cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3493cdccee99SLiu Bo btrfs_err(fs_info, 3494cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3495cdccee99SLiu Bo bytenr); 3496cdccee99SLiu Bo WARN_ON(1); 3497cdccee99SLiu Bo ret = -EINVAL; 3498cdccee99SLiu Bo goto out; 3499cdccee99SLiu Bo } 35003173a18fSJosef Bacik 35015d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 35025d4f98a2SYan Zheng out: 35035d4f98a2SYan Zheng btrfs_free_path(path); 35045d4f98a2SYan Zheng return ret; 35055d4f98a2SYan Zheng } 35065d4f98a2SYan Zheng 35075d4f98a2SYan Zheng /* 35085d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 35095d4f98a2SYan Zheng */ 35105d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 35115d4f98a2SYan Zheng struct extent_buffer *eb) 35125d4f98a2SYan Zheng { 35135d4f98a2SYan Zheng u64 flags; 35145d4f98a2SYan Zheng int ret; 35155d4f98a2SYan Zheng 35165d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 35175d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 35185d4f98a2SYan Zheng return 1; 35195d4f98a2SYan Zheng 35202ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info, 35213173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 35223173a18fSJosef Bacik NULL, &flags); 35235d4f98a2SYan Zheng BUG_ON(ret); 35245d4f98a2SYan Zheng 35255d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 35265d4f98a2SYan Zheng ret = 1; 35275d4f98a2SYan Zheng else 35285d4f98a2SYan Zheng ret = 0; 35295d4f98a2SYan Zheng return ret; 35305d4f98a2SYan Zheng } 35315d4f98a2SYan Zheng 35320af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 35331bbc621eSChris Mason struct btrfs_block_group_cache *block_group, 35341bbc621eSChris Mason struct inode *inode, 35351bbc621eSChris Mason u64 ino) 35360af3d00bSJosef Bacik { 35370af3d00bSJosef Bacik struct btrfs_key key; 35380af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 35390af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 35400af3d00bSJosef Bacik int ret = 0; 35410af3d00bSJosef Bacik 35420af3d00bSJosef Bacik if (inode) 35430af3d00bSJosef Bacik goto truncate; 35440af3d00bSJosef Bacik 35450af3d00bSJosef Bacik key.objectid = ino; 35460af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 35470af3d00bSJosef Bacik key.offset = 0; 35480af3d00bSJosef Bacik 35490af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 3550f54fb859STsutomu Itoh if (IS_ERR(inode) || is_bad_inode(inode)) { 3551f54fb859STsutomu Itoh if (!IS_ERR(inode)) 35520af3d00bSJosef Bacik iput(inode); 35530af3d00bSJosef Bacik return -ENOENT; 35540af3d00bSJosef Bacik } 35550af3d00bSJosef Bacik 35560af3d00bSJosef Bacik truncate: 35572ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 35587b61cd92SMiao Xie &fs_info->global_block_rsv); 35597b61cd92SMiao Xie if (ret) 35607b61cd92SMiao Xie goto out; 35617b61cd92SMiao Xie 35627a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 35630af3d00bSJosef Bacik if (IS_ERR(trans)) { 35643612b495STsutomu Itoh ret = PTR_ERR(trans); 35650af3d00bSJosef Bacik goto out; 35660af3d00bSJosef Bacik } 35670af3d00bSJosef Bacik 356877ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 35690af3d00bSJosef Bacik 35703a45bb20SJeff Mahoney btrfs_end_transaction(trans); 35712ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 35720af3d00bSJosef Bacik out: 35730af3d00bSJosef Bacik iput(inode); 35740af3d00bSJosef Bacik return ret; 35750af3d00bSJosef Bacik } 35760af3d00bSJosef Bacik 35775d4f98a2SYan Zheng /* 35785d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 35795d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 35805d4f98a2SYan Zheng */ 35815d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 35825d4f98a2SYan Zheng struct btrfs_key *extent_key, 35835d4f98a2SYan Zheng struct extent_buffer *leaf, 35845d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 35855d4f98a2SYan Zheng struct rb_root *blocks) 35865d4f98a2SYan Zheng { 35870b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35885d4f98a2SYan Zheng struct btrfs_path *path; 35895d4f98a2SYan Zheng struct tree_block *block; 35905d4f98a2SYan Zheng struct btrfs_root *root; 35915d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 35925d4f98a2SYan Zheng struct rb_node *rb_node; 35935d4f98a2SYan Zheng struct btrfs_key key; 35945d4f98a2SYan Zheng u64 ref_root; 35955d4f98a2SYan Zheng u64 ref_objectid; 35965d4f98a2SYan Zheng u64 ref_offset; 35975d4f98a2SYan Zheng u32 ref_count; 35985d4f98a2SYan Zheng u32 nritems; 35995d4f98a2SYan Zheng int err = 0; 36005d4f98a2SYan Zheng int added = 0; 36015d4f98a2SYan Zheng int counted; 36025d4f98a2SYan Zheng int ret; 36035d4f98a2SYan Zheng 36045d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 36055d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 36065d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 36075d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 36085d4f98a2SYan Zheng 36090af3d00bSJosef Bacik /* 36100af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 36110af3d00bSJosef Bacik * it and redo the search. 36120af3d00bSJosef Bacik */ 36130af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 36140b246afaSJeff Mahoney ret = delete_block_group_cache(fs_info, rc->block_group, 36150af3d00bSJosef Bacik NULL, ref_objectid); 36160af3d00bSJosef Bacik if (ret != -ENOENT) 36170af3d00bSJosef Bacik return ret; 36180af3d00bSJosef Bacik ret = 0; 36190af3d00bSJosef Bacik } 36200af3d00bSJosef Bacik 36210af3d00bSJosef Bacik path = btrfs_alloc_path(); 36220af3d00bSJosef Bacik if (!path) 36230af3d00bSJosef Bacik return -ENOMEM; 3624e4058b54SDavid Sterba path->reada = READA_FORWARD; 36250af3d00bSJosef Bacik 36260b246afaSJeff Mahoney root = read_fs_root(fs_info, ref_root); 36275d4f98a2SYan Zheng if (IS_ERR(root)) { 36285d4f98a2SYan Zheng err = PTR_ERR(root); 36295d4f98a2SYan Zheng goto out; 36305d4f98a2SYan Zheng } 36315d4f98a2SYan Zheng 36325d4f98a2SYan Zheng key.objectid = ref_objectid; 36335d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 363484850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 363584850e8dSYan, Zheng key.offset = 0; 363684850e8dSYan, Zheng else 363784850e8dSYan, Zheng key.offset = ref_offset; 36385d4f98a2SYan Zheng 36395d4f98a2SYan Zheng path->search_commit_root = 1; 36405d4f98a2SYan Zheng path->skip_locking = 1; 36415d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 36425d4f98a2SYan Zheng if (ret < 0) { 36435d4f98a2SYan Zheng err = ret; 36445d4f98a2SYan Zheng goto out; 36455d4f98a2SYan Zheng } 36465d4f98a2SYan Zheng 36475d4f98a2SYan Zheng leaf = path->nodes[0]; 36485d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36495d4f98a2SYan Zheng /* 36505d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 36515d4f98a2SYan Zheng * are not counted in 36525d4f98a2SYan Zheng */ 36535d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36545d4f98a2SYan Zheng counted = 0; 36555d4f98a2SYan Zheng else 36565d4f98a2SYan Zheng counted = 1; 36575d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36585d4f98a2SYan Zheng if (rb_node) { 36595d4f98a2SYan Zheng if (counted) 36605d4f98a2SYan Zheng added = 1; 36615d4f98a2SYan Zheng else 36625d4f98a2SYan Zheng path->slots[0] = nritems; 36635d4f98a2SYan Zheng } 36645d4f98a2SYan Zheng 36655d4f98a2SYan Zheng while (ref_count > 0) { 36665d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 36675d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 36685d4f98a2SYan Zheng if (ret < 0) { 36695d4f98a2SYan Zheng err = ret; 36705d4f98a2SYan Zheng goto out; 36715d4f98a2SYan Zheng } 3672fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 36735d4f98a2SYan Zheng goto out; 36745d4f98a2SYan Zheng 36755d4f98a2SYan Zheng leaf = path->nodes[0]; 36765d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36775d4f98a2SYan Zheng added = 0; 36785d4f98a2SYan Zheng 36795d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36805d4f98a2SYan Zheng counted = 0; 36815d4f98a2SYan Zheng else 36825d4f98a2SYan Zheng counted = 1; 36835d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36845d4f98a2SYan Zheng if (rb_node) { 36855d4f98a2SYan Zheng if (counted) 36865d4f98a2SYan Zheng added = 1; 36875d4f98a2SYan Zheng else 36885d4f98a2SYan Zheng path->slots[0] = nritems; 36895d4f98a2SYan Zheng } 36905d4f98a2SYan Zheng } 36915d4f98a2SYan Zheng 36925d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3693fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3694fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 36955d4f98a2SYan Zheng break; 36965d4f98a2SYan Zheng 36975d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 36985d4f98a2SYan Zheng struct btrfs_file_extent_item); 36995d4f98a2SYan Zheng 37005d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 37015d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 37025d4f98a2SYan Zheng goto next; 37035d4f98a2SYan Zheng 37045d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 37055d4f98a2SYan Zheng extent_key->objectid) 37065d4f98a2SYan Zheng goto next; 37075d4f98a2SYan Zheng 37085d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 37095d4f98a2SYan Zheng if (key.offset != ref_offset) 37105d4f98a2SYan Zheng goto next; 37115d4f98a2SYan Zheng 37125d4f98a2SYan Zheng if (counted) 37135d4f98a2SYan Zheng ref_count--; 37145d4f98a2SYan Zheng if (added) 37155d4f98a2SYan Zheng goto next; 37165d4f98a2SYan Zheng 37177476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 37185d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 37195d4f98a2SYan Zheng if (!block) { 37205d4f98a2SYan Zheng err = -ENOMEM; 37215d4f98a2SYan Zheng break; 37225d4f98a2SYan Zheng } 37235d4f98a2SYan Zheng block->bytenr = leaf->start; 37245d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 37255d4f98a2SYan Zheng block->level = 0; 37265d4f98a2SYan Zheng block->key_ready = 1; 37275d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 37285d4f98a2SYan Zheng &block->rb_node); 372943c04fb1SJeff Mahoney if (rb_node) 373043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 373143c04fb1SJeff Mahoney block->bytenr); 37325d4f98a2SYan Zheng } 37335d4f98a2SYan Zheng if (counted) 37345d4f98a2SYan Zheng added = 1; 37355d4f98a2SYan Zheng else 37365d4f98a2SYan Zheng path->slots[0] = nritems; 37375d4f98a2SYan Zheng next: 37385d4f98a2SYan Zheng path->slots[0]++; 37395d4f98a2SYan Zheng 37405d4f98a2SYan Zheng } 37415d4f98a2SYan Zheng out: 37425d4f98a2SYan Zheng btrfs_free_path(path); 37435d4f98a2SYan Zheng return err; 37445d4f98a2SYan Zheng } 37455d4f98a2SYan Zheng 37465d4f98a2SYan Zheng /* 37472c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37485d4f98a2SYan Zheng */ 37495d4f98a2SYan Zheng static noinline_for_stack 37505d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 37515d4f98a2SYan Zheng struct btrfs_key *extent_key, 37525d4f98a2SYan Zheng struct btrfs_path *path, 37535d4f98a2SYan Zheng struct rb_root *blocks) 37545d4f98a2SYan Zheng { 37555d4f98a2SYan Zheng struct btrfs_key key; 37565d4f98a2SYan Zheng struct extent_buffer *eb; 37575d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 37585d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 37595d4f98a2SYan Zheng unsigned long ptr; 37605d4f98a2SYan Zheng unsigned long end; 3761da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 3762647f63bdSFilipe David Borba Manana int ret = 0; 37635d4f98a2SYan Zheng int err = 0; 37645d4f98a2SYan Zheng 37655d4f98a2SYan Zheng eb = path->nodes[0]; 37665d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 37675d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 37685d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 37695d4f98a2SYan Zheng if (ptr + sizeof(struct btrfs_extent_item_v0) == end) 37705d4f98a2SYan Zheng ptr = end; 37715d4f98a2SYan Zheng else 37725d4f98a2SYan Zheng #endif 37735d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 37745d4f98a2SYan Zheng 37755d4f98a2SYan Zheng while (ptr < end) { 37765d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 37773de28d57SLiu Bo key.type = btrfs_get_extent_inline_ref_type(eb, iref, 37783de28d57SLiu Bo BTRFS_REF_TYPE_DATA); 37795d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37805d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 37815d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37825d4f98a2SYan Zheng blocks); 37835d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37845d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 37855d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37865d4f98a2SYan Zheng eb, dref, blocks); 37875d4f98a2SYan Zheng } else { 3788b14c55a1SLiu Bo ret = -EINVAL; 3789b14c55a1SLiu Bo btrfs_err(rc->extent_root->fs_info, 3790b14c55a1SLiu Bo "extent %llu slot %d has an invalid inline ref type", 3791b14c55a1SLiu Bo eb->start, path->slots[0]); 37925d4f98a2SYan Zheng } 3793647f63bdSFilipe David Borba Manana if (ret) { 3794647f63bdSFilipe David Borba Manana err = ret; 3795647f63bdSFilipe David Borba Manana goto out; 3796647f63bdSFilipe David Borba Manana } 37975d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 37985d4f98a2SYan Zheng } 37995d4f98a2SYan Zheng WARN_ON(ptr > end); 38005d4f98a2SYan Zheng 38015d4f98a2SYan Zheng while (1) { 38025d4f98a2SYan Zheng cond_resched(); 38035d4f98a2SYan Zheng eb = path->nodes[0]; 38045d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 38055d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38065d4f98a2SYan Zheng if (ret < 0) { 38075d4f98a2SYan Zheng err = ret; 38085d4f98a2SYan Zheng break; 38095d4f98a2SYan Zheng } 38105d4f98a2SYan Zheng if (ret > 0) 38115d4f98a2SYan Zheng break; 38125d4f98a2SYan Zheng eb = path->nodes[0]; 38135d4f98a2SYan Zheng } 38145d4f98a2SYan Zheng 38155d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 38165d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 38175d4f98a2SYan Zheng break; 38185d4f98a2SYan Zheng 38195d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 38205d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY || 38215d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 38225d4f98a2SYan Zheng #else 38235d4f98a2SYan Zheng BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); 38245d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 38255d4f98a2SYan Zheng #endif 38265d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 38275d4f98a2SYan Zheng blocks); 38285d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 38295d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 38305d4f98a2SYan Zheng struct btrfs_extent_data_ref); 38315d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 38325d4f98a2SYan Zheng eb, dref, blocks); 38335d4f98a2SYan Zheng } else { 38345d4f98a2SYan Zheng ret = 0; 38355d4f98a2SYan Zheng } 38365d4f98a2SYan Zheng if (ret) { 38375d4f98a2SYan Zheng err = ret; 38385d4f98a2SYan Zheng break; 38395d4f98a2SYan Zheng } 38405d4f98a2SYan Zheng path->slots[0]++; 38415d4f98a2SYan Zheng } 3842647f63bdSFilipe David Borba Manana out: 3843b3b4aa74SDavid Sterba btrfs_release_path(path); 38445d4f98a2SYan Zheng if (err) 38455d4f98a2SYan Zheng free_block_list(blocks); 38465d4f98a2SYan Zheng return err; 38475d4f98a2SYan Zheng } 38485d4f98a2SYan Zheng 38495d4f98a2SYan Zheng /* 38502c016dc2SLiu Bo * helper to find next unprocessed extent 38515d4f98a2SYan Zheng */ 38525d4f98a2SYan Zheng static noinline_for_stack 3853147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 38543fd0a558SYan, Zheng struct btrfs_key *extent_key) 38555d4f98a2SYan Zheng { 38560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38575d4f98a2SYan Zheng struct btrfs_key key; 38585d4f98a2SYan Zheng struct extent_buffer *leaf; 38595d4f98a2SYan Zheng u64 start, end, last; 38605d4f98a2SYan Zheng int ret; 38615d4f98a2SYan Zheng 38625d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 38635d4f98a2SYan Zheng while (1) { 38645d4f98a2SYan Zheng cond_resched(); 38655d4f98a2SYan Zheng if (rc->search_start >= last) { 38665d4f98a2SYan Zheng ret = 1; 38675d4f98a2SYan Zheng break; 38685d4f98a2SYan Zheng } 38695d4f98a2SYan Zheng 38705d4f98a2SYan Zheng key.objectid = rc->search_start; 38715d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 38725d4f98a2SYan Zheng key.offset = 0; 38735d4f98a2SYan Zheng 38745d4f98a2SYan Zheng path->search_commit_root = 1; 38755d4f98a2SYan Zheng path->skip_locking = 1; 38765d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 38775d4f98a2SYan Zheng 0, 0); 38785d4f98a2SYan Zheng if (ret < 0) 38795d4f98a2SYan Zheng break; 38805d4f98a2SYan Zheng next: 38815d4f98a2SYan Zheng leaf = path->nodes[0]; 38825d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38835d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38845d4f98a2SYan Zheng if (ret != 0) 38855d4f98a2SYan Zheng break; 38865d4f98a2SYan Zheng leaf = path->nodes[0]; 38875d4f98a2SYan Zheng } 38885d4f98a2SYan Zheng 38895d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38905d4f98a2SYan Zheng if (key.objectid >= last) { 38915d4f98a2SYan Zheng ret = 1; 38925d4f98a2SYan Zheng break; 38935d4f98a2SYan Zheng } 38945d4f98a2SYan Zheng 38953173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38963173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38973173a18fSJosef Bacik path->slots[0]++; 38983173a18fSJosef Bacik goto next; 38993173a18fSJosef Bacik } 39003173a18fSJosef Bacik 39013173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 39025d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 39035d4f98a2SYan Zheng path->slots[0]++; 39045d4f98a2SYan Zheng goto next; 39055d4f98a2SYan Zheng } 39065d4f98a2SYan Zheng 39073173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 39080b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 39093173a18fSJosef Bacik rc->search_start) { 39103173a18fSJosef Bacik path->slots[0]++; 39113173a18fSJosef Bacik goto next; 39123173a18fSJosef Bacik } 39133173a18fSJosef Bacik 39145d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 39155d4f98a2SYan Zheng key.objectid, &start, &end, 3916e6138876SJosef Bacik EXTENT_DIRTY, NULL); 39175d4f98a2SYan Zheng 39185d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3919b3b4aa74SDavid Sterba btrfs_release_path(path); 39205d4f98a2SYan Zheng rc->search_start = end + 1; 39215d4f98a2SYan Zheng } else { 39223173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 39235d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 39243173a18fSJosef Bacik else 39253173a18fSJosef Bacik rc->search_start = key.objectid + 39260b246afaSJeff Mahoney fs_info->nodesize; 39273fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 39285d4f98a2SYan Zheng return 0; 39295d4f98a2SYan Zheng } 39305d4f98a2SYan Zheng } 3931b3b4aa74SDavid Sterba btrfs_release_path(path); 39325d4f98a2SYan Zheng return ret; 39335d4f98a2SYan Zheng } 39345d4f98a2SYan Zheng 39355d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 39365d4f98a2SYan Zheng { 39375d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39387585717fSChris Mason 39397585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39405d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 39417585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39425d4f98a2SYan Zheng } 39435d4f98a2SYan Zheng 39445d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 39455d4f98a2SYan Zheng { 39465d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39477585717fSChris Mason 39487585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39495d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 39507585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39515d4f98a2SYan Zheng } 39525d4f98a2SYan Zheng 39535d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 39545d4f98a2SYan Zheng { 39555d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39565d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39575d4f98a2SYan Zheng return 1; 39585d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 39595d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39605d4f98a2SYan Zheng return 1; 39615d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39625d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 39635d4f98a2SYan Zheng return 1; 39645d4f98a2SYan Zheng return 0; 39655d4f98a2SYan Zheng } 39665d4f98a2SYan Zheng 39673fd0a558SYan, Zheng static noinline_for_stack 39683fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 39693fd0a558SYan, Zheng { 39703fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3971ac2fabacSJosef Bacik int ret; 39723fd0a558SYan, Zheng 39732ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 397466d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 39753fd0a558SYan, Zheng if (!rc->block_rsv) 39763fd0a558SYan, Zheng return -ENOMEM; 39773fd0a558SYan, Zheng 39783fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 39793fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 39803fd0a558SYan, Zheng rc->extents_found = 0; 39813fd0a558SYan, Zheng rc->nodes_relocated = 0; 39823fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39830647bf56SWang Shilong rc->reserved_bytes = 0; 3984da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 39850647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3986ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3987ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3988ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3989ac2fabacSJosef Bacik if (ret) 3990ac2fabacSJosef Bacik return ret; 39913fd0a558SYan, Zheng 39923fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39933fd0a558SYan, Zheng set_reloc_control(rc); 39943fd0a558SYan, Zheng 39957a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 399628818947SLiu Bo if (IS_ERR(trans)) { 399728818947SLiu Bo unset_reloc_control(rc); 399828818947SLiu Bo /* 399928818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 400028818947SLiu Bo * cleanup. And callers are responsible to free the above 400128818947SLiu Bo * block rsv. 400228818947SLiu Bo */ 400328818947SLiu Bo return PTR_ERR(trans); 400428818947SLiu Bo } 40053a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 40063fd0a558SYan, Zheng return 0; 40073fd0a558SYan, Zheng } 400876dda93cSYan, Zheng 40095d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 40105d4f98a2SYan Zheng { 40112ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40125d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 40135d4f98a2SYan Zheng struct btrfs_key key; 40145d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 40155d4f98a2SYan Zheng struct btrfs_path *path; 40165d4f98a2SYan Zheng struct btrfs_extent_item *ei; 40175d4f98a2SYan Zheng u64 flags; 40185d4f98a2SYan Zheng u32 item_size; 40195d4f98a2SYan Zheng int ret; 40205d4f98a2SYan Zheng int err = 0; 4021c87f08caSChris Mason int progress = 0; 40225d4f98a2SYan Zheng 40235d4f98a2SYan Zheng path = btrfs_alloc_path(); 40243fd0a558SYan, Zheng if (!path) 40255d4f98a2SYan Zheng return -ENOMEM; 4026e4058b54SDavid Sterba path->reada = READA_FORWARD; 40273fd0a558SYan, Zheng 40283fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 40293fd0a558SYan, Zheng if (ret) { 40303fd0a558SYan, Zheng err = ret; 40313fd0a558SYan, Zheng goto out_free; 40322423fdfbSJiri Slaby } 40335d4f98a2SYan Zheng 40345d4f98a2SYan Zheng while (1) { 40350647bf56SWang Shilong rc->reserved_bytes = 0; 40360647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40370647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40380647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40390647bf56SWang Shilong if (ret) { 40400647bf56SWang Shilong err = ret; 40410647bf56SWang Shilong break; 40420647bf56SWang Shilong } 4043c87f08caSChris Mason progress++; 4044a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40450f788c58SLiu Bo if (IS_ERR(trans)) { 40460f788c58SLiu Bo err = PTR_ERR(trans); 40470f788c58SLiu Bo trans = NULL; 40480f788c58SLiu Bo break; 40490f788c58SLiu Bo } 4050c87f08caSChris Mason restart: 40513fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40523a45bb20SJeff Mahoney btrfs_end_transaction(trans); 40533fd0a558SYan, Zheng continue; 40543fd0a558SYan, Zheng } 40553fd0a558SYan, Zheng 4056147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40575d4f98a2SYan Zheng if (ret < 0) 40585d4f98a2SYan Zheng err = ret; 40595d4f98a2SYan Zheng if (ret != 0) 40605d4f98a2SYan Zheng break; 40615d4f98a2SYan Zheng 40625d4f98a2SYan Zheng rc->extents_found++; 40635d4f98a2SYan Zheng 40645d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40655d4f98a2SYan Zheng struct btrfs_extent_item); 40663fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40675d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40685d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40695d4f98a2SYan Zheng ret = check_extent_flags(flags); 40705d4f98a2SYan Zheng BUG_ON(ret); 40715d4f98a2SYan Zheng 40725d4f98a2SYan Zheng } else { 40735d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 40745d4f98a2SYan Zheng u64 ref_owner; 40755d4f98a2SYan Zheng int path_change = 0; 40765d4f98a2SYan Zheng 40775d4f98a2SYan Zheng BUG_ON(item_size != 40785d4f98a2SYan Zheng sizeof(struct btrfs_extent_item_v0)); 40795d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, &key, &ref_owner, 40805d4f98a2SYan Zheng &path_change); 40814b3576e4SZhaolei if (ret < 0) { 40824b3576e4SZhaolei err = ret; 40834b3576e4SZhaolei break; 40844b3576e4SZhaolei } 40855d4f98a2SYan Zheng if (ref_owner < BTRFS_FIRST_FREE_OBJECTID) 40865d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_TREE_BLOCK; 40875d4f98a2SYan Zheng else 40885d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_DATA; 40895d4f98a2SYan Zheng 40905d4f98a2SYan Zheng if (path_change) { 4091b3b4aa74SDavid Sterba btrfs_release_path(path); 40925d4f98a2SYan Zheng 40935d4f98a2SYan Zheng path->search_commit_root = 1; 40945d4f98a2SYan Zheng path->skip_locking = 1; 40955d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, 40965d4f98a2SYan Zheng &key, path, 0, 0); 40975d4f98a2SYan Zheng if (ret < 0) { 40985d4f98a2SYan Zheng err = ret; 40995d4f98a2SYan Zheng break; 41005d4f98a2SYan Zheng } 41015d4f98a2SYan Zheng BUG_ON(ret > 0); 41025d4f98a2SYan Zheng } 41035d4f98a2SYan Zheng #else 41045d4f98a2SYan Zheng BUG(); 41055d4f98a2SYan Zheng #endif 41065d4f98a2SYan Zheng } 41075d4f98a2SYan Zheng 41085d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 41095d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 41105d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 41115d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41125d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 41135d4f98a2SYan Zheng } else { 4114b3b4aa74SDavid Sterba btrfs_release_path(path); 41155d4f98a2SYan Zheng ret = 0; 41165d4f98a2SYan Zheng } 41175d4f98a2SYan Zheng if (ret < 0) { 41183fd0a558SYan, Zheng err = ret; 41195d4f98a2SYan Zheng break; 41205d4f98a2SYan Zheng } 41215d4f98a2SYan Zheng 41225d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 41235d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 41245d4f98a2SYan Zheng if (ret < 0) { 41251708cc57SWang Shilong /* 41261708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 41271708cc57SWang Shilong * backref cache when committing transaction. 41281708cc57SWang Shilong */ 41291708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 41301708cc57SWang Shilong 41313fd0a558SYan, Zheng if (ret != -EAGAIN) { 41325d4f98a2SYan Zheng err = ret; 41335d4f98a2SYan Zheng break; 41345d4f98a2SYan Zheng } 41353fd0a558SYan, Zheng rc->extents_found--; 41363fd0a558SYan, Zheng rc->search_start = key.objectid; 41373fd0a558SYan, Zheng } 41385d4f98a2SYan Zheng } 41395d4f98a2SYan Zheng 41403a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41412ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41423fd0a558SYan, Zheng trans = NULL; 41435d4f98a2SYan Zheng 41445d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 41455d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41465d4f98a2SYan Zheng rc->found_file_extent = 1; 41470257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 41483fd0a558SYan, Zheng &key, &rc->cluster); 41495d4f98a2SYan Zheng if (ret < 0) { 41505d4f98a2SYan Zheng err = ret; 41515d4f98a2SYan Zheng break; 41525d4f98a2SYan Zheng } 41535d4f98a2SYan Zheng } 41545d4f98a2SYan Zheng } 4155c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 41562ff7e61eSJeff Mahoney ret = btrfs_force_chunk_alloc(trans, fs_info, 4157c87f08caSChris Mason rc->block_group->flags); 41589689457bSShilong Wang if (ret == 1) { 4159c87f08caSChris Mason err = 0; 4160c87f08caSChris Mason progress = 0; 4161c87f08caSChris Mason goto restart; 4162c87f08caSChris Mason } 4163c87f08caSChris Mason } 41643fd0a558SYan, Zheng 4165b3b4aa74SDavid Sterba btrfs_release_path(path); 416691166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41675d4f98a2SYan Zheng 41685d4f98a2SYan Zheng if (trans) { 41693a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41702ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41715d4f98a2SYan Zheng } 41725d4f98a2SYan Zheng 41730257bb82SYan, Zheng if (!err) { 41743fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 41753fd0a558SYan, Zheng &rc->cluster); 41760257bb82SYan, Zheng if (ret < 0) 41770257bb82SYan, Zheng err = ret; 41780257bb82SYan, Zheng } 41790257bb82SYan, Zheng 41803fd0a558SYan, Zheng rc->create_reloc_tree = 0; 41813fd0a558SYan, Zheng set_reloc_control(rc); 41820257bb82SYan, Zheng 41833fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 41842ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41855d4f98a2SYan Zheng 41863fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 41875d4f98a2SYan Zheng 41885d4f98a2SYan Zheng merge_reloc_roots(rc); 41895d4f98a2SYan Zheng 41903fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 41915d4f98a2SYan Zheng unset_reloc_control(rc); 41922ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41935d4f98a2SYan Zheng 41945d4f98a2SYan Zheng /* get rid of pinned extents */ 41957a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 419662b99540SQu Wenruo if (IS_ERR(trans)) { 41973612b495STsutomu Itoh err = PTR_ERR(trans); 419862b99540SQu Wenruo goto out_free; 419962b99540SQu Wenruo } 42003a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 42013fd0a558SYan, Zheng out_free: 42022ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 42033fd0a558SYan, Zheng btrfs_free_path(path); 42045d4f98a2SYan Zheng return err; 42055d4f98a2SYan Zheng } 42065d4f98a2SYan Zheng 42075d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 42080257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 42095d4f98a2SYan Zheng { 42105d4f98a2SYan Zheng struct btrfs_path *path; 42115d4f98a2SYan Zheng struct btrfs_inode_item *item; 42125d4f98a2SYan Zheng struct extent_buffer *leaf; 42135d4f98a2SYan Zheng int ret; 42145d4f98a2SYan Zheng 42155d4f98a2SYan Zheng path = btrfs_alloc_path(); 42165d4f98a2SYan Zheng if (!path) 42175d4f98a2SYan Zheng return -ENOMEM; 42185d4f98a2SYan Zheng 42195d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 42205d4f98a2SYan Zheng if (ret) 42215d4f98a2SYan Zheng goto out; 42225d4f98a2SYan Zheng 42235d4f98a2SYan Zheng leaf = path->nodes[0]; 42245d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4225b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 42265d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 42270257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 42285d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 42293fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 42303fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 42315d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 42325d4f98a2SYan Zheng out: 42335d4f98a2SYan Zheng btrfs_free_path(path); 42345d4f98a2SYan Zheng return ret; 42355d4f98a2SYan Zheng } 42365d4f98a2SYan Zheng 42375d4f98a2SYan Zheng /* 42385d4f98a2SYan Zheng * helper to create inode for data relocation. 42395d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 42405d4f98a2SYan Zheng */ 42413fd0a558SYan, Zheng static noinline_for_stack 42423fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 42435d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 42445d4f98a2SYan Zheng { 42455d4f98a2SYan Zheng struct inode *inode = NULL; 42465d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42475d4f98a2SYan Zheng struct btrfs_root *root; 42485d4f98a2SYan Zheng struct btrfs_key key; 42494624900dSZhaolei u64 objectid; 42505d4f98a2SYan Zheng int err = 0; 42515d4f98a2SYan Zheng 42525d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 42535d4f98a2SYan Zheng if (IS_ERR(root)) 42545d4f98a2SYan Zheng return ERR_CAST(root); 42555d4f98a2SYan Zheng 4256a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 42573fd0a558SYan, Zheng if (IS_ERR(trans)) 42583fd0a558SYan, Zheng return ERR_CAST(trans); 42595d4f98a2SYan Zheng 4260581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42615d4f98a2SYan Zheng if (err) 42625d4f98a2SYan Zheng goto out; 42635d4f98a2SYan Zheng 42640257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42655d4f98a2SYan Zheng BUG_ON(err); 42665d4f98a2SYan Zheng 42675d4f98a2SYan Zheng key.objectid = objectid; 42685d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 42695d4f98a2SYan Zheng key.offset = 0; 42700b246afaSJeff Mahoney inode = btrfs_iget(fs_info->sb, &key, root, NULL); 42715d4f98a2SYan Zheng BUG_ON(IS_ERR(inode) || is_bad_inode(inode)); 42725d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 42735d4f98a2SYan Zheng 427473f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 42755d4f98a2SYan Zheng out: 42763a45bb20SJeff Mahoney btrfs_end_transaction(trans); 42772ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42785d4f98a2SYan Zheng if (err) { 42795d4f98a2SYan Zheng if (inode) 42805d4f98a2SYan Zheng iput(inode); 42815d4f98a2SYan Zheng inode = ERR_PTR(err); 42825d4f98a2SYan Zheng } 42835d4f98a2SYan Zheng return inode; 42845d4f98a2SYan Zheng } 42855d4f98a2SYan Zheng 4286a9995eecSJosef Bacik static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 42873fd0a558SYan, Zheng { 42883fd0a558SYan, Zheng struct reloc_control *rc; 42893fd0a558SYan, Zheng 42903fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 42913fd0a558SYan, Zheng if (!rc) 42923fd0a558SYan, Zheng return NULL; 42933fd0a558SYan, Zheng 42943fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 42953fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 42963fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 4297c6100a4bSJosef Bacik extent_io_tree_init(&rc->processed_blocks, NULL); 42983fd0a558SYan, Zheng return rc; 42993fd0a558SYan, Zheng } 43003fd0a558SYan, Zheng 43015d4f98a2SYan Zheng /* 4302ebce0e01SAdam Borowski * Print the block group being relocated 4303ebce0e01SAdam Borowski */ 4304ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 4305ebce0e01SAdam Borowski struct btrfs_block_group_cache *block_group) 4306ebce0e01SAdam Borowski { 4307ebce0e01SAdam Borowski char buf[128]; /* prefixed by a '|' that'll be dropped */ 4308ebce0e01SAdam Borowski u64 flags = block_group->flags; 4309ebce0e01SAdam Borowski 4310ebce0e01SAdam Borowski /* Shouldn't happen */ 4311ebce0e01SAdam Borowski if (!flags) { 4312ebce0e01SAdam Borowski strcpy(buf, "|NONE"); 4313ebce0e01SAdam Borowski } else { 4314ebce0e01SAdam Borowski char *bp = buf; 4315ebce0e01SAdam Borowski 4316ebce0e01SAdam Borowski #define DESCRIBE_FLAG(f, d) \ 4317ebce0e01SAdam Borowski if (flags & BTRFS_BLOCK_GROUP_##f) { \ 4318ebce0e01SAdam Borowski bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \ 4319ebce0e01SAdam Borowski flags &= ~BTRFS_BLOCK_GROUP_##f; \ 4320ebce0e01SAdam Borowski } 4321ebce0e01SAdam Borowski DESCRIBE_FLAG(DATA, "data"); 4322ebce0e01SAdam Borowski DESCRIBE_FLAG(SYSTEM, "system"); 4323ebce0e01SAdam Borowski DESCRIBE_FLAG(METADATA, "metadata"); 4324ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID0, "raid0"); 4325ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID1, "raid1"); 4326ebce0e01SAdam Borowski DESCRIBE_FLAG(DUP, "dup"); 4327ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID10, "raid10"); 4328ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID5, "raid5"); 4329ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID6, "raid6"); 4330ebce0e01SAdam Borowski if (flags) 4331ebce0e01SAdam Borowski snprintf(buf, buf - bp + sizeof(buf), "|0x%llx", flags); 4332ebce0e01SAdam Borowski #undef DESCRIBE_FLAG 4333ebce0e01SAdam Borowski } 4334ebce0e01SAdam Borowski 4335ebce0e01SAdam Borowski btrfs_info(fs_info, 4336ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4337ebce0e01SAdam Borowski block_group->key.objectid, buf + 1); 4338ebce0e01SAdam Borowski } 4339ebce0e01SAdam Borowski 4340ebce0e01SAdam Borowski /* 43415d4f98a2SYan Zheng * function to relocate all extents in a block group. 43425d4f98a2SYan Zheng */ 43436bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 43445d4f98a2SYan Zheng { 43456bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 43465d4f98a2SYan Zheng struct reloc_control *rc; 43470af3d00bSJosef Bacik struct inode *inode; 43480af3d00bSJosef Bacik struct btrfs_path *path; 43495d4f98a2SYan Zheng int ret; 4350f0486c68SYan, Zheng int rw = 0; 43515d4f98a2SYan Zheng int err = 0; 43525d4f98a2SYan Zheng 4353a9995eecSJosef Bacik rc = alloc_reloc_control(fs_info); 43545d4f98a2SYan Zheng if (!rc) 43555d4f98a2SYan Zheng return -ENOMEM; 43565d4f98a2SYan Zheng 4357f0486c68SYan, Zheng rc->extent_root = extent_root; 43583fd0a558SYan, Zheng 43595d4f98a2SYan Zheng rc->block_group = btrfs_lookup_block_group(fs_info, group_start); 43605d4f98a2SYan Zheng BUG_ON(!rc->block_group); 43615d4f98a2SYan Zheng 43625e00f193SJeff Mahoney ret = btrfs_inc_block_group_ro(fs_info, rc->block_group); 4363f0486c68SYan, Zheng if (ret) { 4364f0486c68SYan, Zheng err = ret; 4365f0486c68SYan, Zheng goto out; 4366f0486c68SYan, Zheng } 4367f0486c68SYan, Zheng rw = 1; 4368f0486c68SYan, Zheng 43690af3d00bSJosef Bacik path = btrfs_alloc_path(); 43700af3d00bSJosef Bacik if (!path) { 43710af3d00bSJosef Bacik err = -ENOMEM; 43720af3d00bSJosef Bacik goto out; 43730af3d00bSJosef Bacik } 43740af3d00bSJosef Bacik 437577ab86bfSJeff Mahoney inode = lookup_free_space_inode(fs_info, rc->block_group, path); 43760af3d00bSJosef Bacik btrfs_free_path(path); 43770af3d00bSJosef Bacik 43780af3d00bSJosef Bacik if (!IS_ERR(inode)) 43791bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43800af3d00bSJosef Bacik else 43810af3d00bSJosef Bacik ret = PTR_ERR(inode); 43820af3d00bSJosef Bacik 43830af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 43840af3d00bSJosef Bacik err = ret; 43850af3d00bSJosef Bacik goto out; 43860af3d00bSJosef Bacik } 43870af3d00bSJosef Bacik 43885d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 43895d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 43905d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 43915d4f98a2SYan Zheng rc->data_inode = NULL; 43925d4f98a2SYan Zheng goto out; 43935d4f98a2SYan Zheng } 43945d4f98a2SYan Zheng 43950b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 43965d4f98a2SYan Zheng 43979cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4398f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 43996374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4400578def7cSFilipe Manana rc->block_group->key.objectid, 4401578def7cSFilipe Manana rc->block_group->key.offset); 44025d4f98a2SYan Zheng 44035d4f98a2SYan Zheng while (1) { 440476dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 44055d4f98a2SYan Zheng ret = relocate_block_group(rc); 440676dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 44075d4f98a2SYan Zheng if (ret < 0) { 44085d4f98a2SYan Zheng err = ret; 44093fd0a558SYan, Zheng goto out; 44105d4f98a2SYan Zheng } 44115d4f98a2SYan Zheng 44125d4f98a2SYan Zheng if (rc->extents_found == 0) 44135d4f98a2SYan Zheng break; 44145d4f98a2SYan Zheng 44150b246afaSJeff Mahoney btrfs_info(fs_info, "found %llu extents", rc->extents_found); 44165d4f98a2SYan Zheng 44175d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 44180ef8b726SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 44190ef8b726SJosef Bacik (u64)-1); 44200ef8b726SJosef Bacik if (ret) { 44210ef8b726SJosef Bacik err = ret; 44220ef8b726SJosef Bacik goto out; 44230ef8b726SJosef Bacik } 44245d4f98a2SYan Zheng invalidate_mapping_pages(rc->data_inode->i_mapping, 44255d4f98a2SYan Zheng 0, -1); 44265d4f98a2SYan Zheng rc->stage = UPDATE_DATA_PTRS; 44275d4f98a2SYan Zheng } 44285d4f98a2SYan Zheng } 44295d4f98a2SYan Zheng 44305d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 44315d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 44325d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 44335d4f98a2SYan Zheng out: 4434f0486c68SYan, Zheng if (err && rw) 44352ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 44365d4f98a2SYan Zheng iput(rc->data_inode); 44375d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 44385d4f98a2SYan Zheng kfree(rc); 44395d4f98a2SYan Zheng return err; 44405d4f98a2SYan Zheng } 44415d4f98a2SYan Zheng 444276dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 444376dda93cSYan, Zheng { 44440b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 444576dda93cSYan, Zheng struct btrfs_trans_handle *trans; 444679787eaaSJeff Mahoney int ret, err; 444776dda93cSYan, Zheng 44480b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 444979787eaaSJeff Mahoney if (IS_ERR(trans)) 445079787eaaSJeff Mahoney return PTR_ERR(trans); 445176dda93cSYan, Zheng 445276dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 445376dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 445476dda93cSYan, Zheng root->root_item.drop_level = 0; 445576dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 44560b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 445776dda93cSYan, Zheng &root->root_key, &root->root_item); 445876dda93cSYan, Zheng 44593a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 446079787eaaSJeff Mahoney if (err) 446179787eaaSJeff Mahoney return err; 446279787eaaSJeff Mahoney return ret; 446376dda93cSYan, Zheng } 446476dda93cSYan, Zheng 44655d4f98a2SYan Zheng /* 44665d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44675d4f98a2SYan Zheng * 44685d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44695d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44705d4f98a2SYan Zheng */ 44715d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44725d4f98a2SYan Zheng { 44730b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 44745d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 44755d4f98a2SYan Zheng struct btrfs_key key; 44765d4f98a2SYan Zheng struct btrfs_root *fs_root; 44775d4f98a2SYan Zheng struct btrfs_root *reloc_root; 44785d4f98a2SYan Zheng struct btrfs_path *path; 44795d4f98a2SYan Zheng struct extent_buffer *leaf; 44805d4f98a2SYan Zheng struct reloc_control *rc = NULL; 44815d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 44825d4f98a2SYan Zheng int ret; 44835d4f98a2SYan Zheng int err = 0; 44845d4f98a2SYan Zheng 44855d4f98a2SYan Zheng path = btrfs_alloc_path(); 44865d4f98a2SYan Zheng if (!path) 44875d4f98a2SYan Zheng return -ENOMEM; 4488e4058b54SDavid Sterba path->reada = READA_BACK; 44895d4f98a2SYan Zheng 44905d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 44915d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 44925d4f98a2SYan Zheng key.offset = (u64)-1; 44935d4f98a2SYan Zheng 44945d4f98a2SYan Zheng while (1) { 44950b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 44965d4f98a2SYan Zheng path, 0, 0); 44975d4f98a2SYan Zheng if (ret < 0) { 44985d4f98a2SYan Zheng err = ret; 44995d4f98a2SYan Zheng goto out; 45005d4f98a2SYan Zheng } 45015d4f98a2SYan Zheng if (ret > 0) { 45025d4f98a2SYan Zheng if (path->slots[0] == 0) 45035d4f98a2SYan Zheng break; 45045d4f98a2SYan Zheng path->slots[0]--; 45055d4f98a2SYan Zheng } 45065d4f98a2SYan Zheng leaf = path->nodes[0]; 45075d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4508b3b4aa74SDavid Sterba btrfs_release_path(path); 45095d4f98a2SYan Zheng 45105d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 45115d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 45125d4f98a2SYan Zheng break; 45135d4f98a2SYan Zheng 4514cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 45155d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 45165d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 45175d4f98a2SYan Zheng goto out; 45185d4f98a2SYan Zheng } 45195d4f98a2SYan Zheng 45205d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 45215d4f98a2SYan Zheng 45225d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 45230b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 45245d4f98a2SYan Zheng reloc_root->root_key.offset); 45255d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 452676dda93cSYan, Zheng ret = PTR_ERR(fs_root); 452776dda93cSYan, Zheng if (ret != -ENOENT) { 452876dda93cSYan, Zheng err = ret; 45295d4f98a2SYan Zheng goto out; 45305d4f98a2SYan Zheng } 453179787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 453279787eaaSJeff Mahoney if (ret < 0) { 453379787eaaSJeff Mahoney err = ret; 453479787eaaSJeff Mahoney goto out; 453579787eaaSJeff Mahoney } 453676dda93cSYan, Zheng } 45375d4f98a2SYan Zheng } 45385d4f98a2SYan Zheng 45395d4f98a2SYan Zheng if (key.offset == 0) 45405d4f98a2SYan Zheng break; 45415d4f98a2SYan Zheng 45425d4f98a2SYan Zheng key.offset--; 45435d4f98a2SYan Zheng } 4544b3b4aa74SDavid Sterba btrfs_release_path(path); 45455d4f98a2SYan Zheng 45465d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 45475d4f98a2SYan Zheng goto out; 45485d4f98a2SYan Zheng 45490b246afaSJeff Mahoney rc = alloc_reloc_control(fs_info); 45505d4f98a2SYan Zheng if (!rc) { 45515d4f98a2SYan Zheng err = -ENOMEM; 45525d4f98a2SYan Zheng goto out; 45535d4f98a2SYan Zheng } 45545d4f98a2SYan Zheng 45550b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 45565d4f98a2SYan Zheng 45575d4f98a2SYan Zheng set_reloc_control(rc); 45585d4f98a2SYan Zheng 45597a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45603612b495STsutomu Itoh if (IS_ERR(trans)) { 45613612b495STsutomu Itoh unset_reloc_control(rc); 45623612b495STsutomu Itoh err = PTR_ERR(trans); 45633612b495STsutomu Itoh goto out_free; 45643612b495STsutomu Itoh } 45653fd0a558SYan, Zheng 45663fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45673fd0a558SYan, Zheng 45685d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45695d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45705d4f98a2SYan Zheng struct btrfs_root, root_list); 45715d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45725d4f98a2SYan Zheng 45735d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 45745d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 45755d4f98a2SYan Zheng &rc->reloc_roots); 45765d4f98a2SYan Zheng continue; 45775d4f98a2SYan Zheng } 45785d4f98a2SYan Zheng 45790b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 458079787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 458179787eaaSJeff Mahoney err = PTR_ERR(fs_root); 458279787eaaSJeff Mahoney goto out_free; 458379787eaaSJeff Mahoney } 45845d4f98a2SYan Zheng 4585ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 458679787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 45875d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 45885d4f98a2SYan Zheng } 45895d4f98a2SYan Zheng 45903a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 459179787eaaSJeff Mahoney if (err) 459279787eaaSJeff Mahoney goto out_free; 45935d4f98a2SYan Zheng 45945d4f98a2SYan Zheng merge_reloc_roots(rc); 45955d4f98a2SYan Zheng 45965d4f98a2SYan Zheng unset_reloc_control(rc); 45975d4f98a2SYan Zheng 45987a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 459962b99540SQu Wenruo if (IS_ERR(trans)) { 46003612b495STsutomu Itoh err = PTR_ERR(trans); 460162b99540SQu Wenruo goto out_free; 460262b99540SQu Wenruo } 46033a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 46043612b495STsutomu Itoh out_free: 46055d4f98a2SYan Zheng kfree(rc); 46063612b495STsutomu Itoh out: 4607aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4608aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4609aca1bba6SLiu Bo 46105d4f98a2SYan Zheng btrfs_free_path(path); 46115d4f98a2SYan Zheng 46125d4f98a2SYan Zheng if (err == 0) { 46135d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 46140b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 46155d4f98a2SYan Zheng if (IS_ERR(fs_root)) 46165d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4617d7ce5843SMiao Xie else 461866b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 46195d4f98a2SYan Zheng } 46205d4f98a2SYan Zheng return err; 46215d4f98a2SYan Zheng } 46225d4f98a2SYan Zheng 46235d4f98a2SYan Zheng /* 46245d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 46255d4f98a2SYan Zheng * 46265d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 46275d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 46285d4f98a2SYan Zheng */ 46295d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 46305d4f98a2SYan Zheng { 46310b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 46325d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 46335d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 46345d4f98a2SYan Zheng int ret; 46355d4f98a2SYan Zheng u64 disk_bytenr; 46364577b014SJosef Bacik u64 new_bytenr; 46375d4f98a2SYan Zheng LIST_HEAD(list); 46385d4f98a2SYan Zheng 46395d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 46405d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 46415d4f98a2SYan Zheng 46425d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46430b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4644a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 464579787eaaSJeff Mahoney if (ret) 464679787eaaSJeff Mahoney goto out; 46475d4f98a2SYan Zheng 46485d4f98a2SYan Zheng while (!list_empty(&list)) { 46495d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46505d4f98a2SYan Zheng list_del_init(&sums->list); 46515d4f98a2SYan Zheng 46524577b014SJosef Bacik /* 46534577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46544577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46554577b014SJosef Bacik * extents but we may have written to say the middle of the 46564577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46574577b014SJosef Bacik * the right disk offset. 46584577b014SJosef Bacik * 46594577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46604577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46614577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46624577b014SJosef Bacik * disk length. 46634577b014SJosef Bacik */ 46644577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 46654577b014SJosef Bacik sums->bytenr = new_bytenr; 46665d4f98a2SYan Zheng 46675d4f98a2SYan Zheng btrfs_add_ordered_sum(inode, ordered, sums); 46685d4f98a2SYan Zheng } 466979787eaaSJeff Mahoney out: 46705d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4671411fc6bcSAndi Kleen return ret; 46725d4f98a2SYan Zheng } 46733fd0a558SYan, Zheng 467483d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 46753fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 46763fd0a558SYan, Zheng struct extent_buffer *cow) 46773fd0a558SYan, Zheng { 46780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46793fd0a558SYan, Zheng struct reloc_control *rc; 46803fd0a558SYan, Zheng struct backref_node *node; 46813fd0a558SYan, Zheng int first_cow = 0; 46823fd0a558SYan, Zheng int level; 468383d4cfd4SJosef Bacik int ret = 0; 46843fd0a558SYan, Zheng 46850b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 46863fd0a558SYan, Zheng if (!rc) 468783d4cfd4SJosef Bacik return 0; 46883fd0a558SYan, Zheng 46893fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46903fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46913fd0a558SYan, Zheng 4692c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4693c974c464SWang Shilong if (buf == root->node) 4694c974c464SWang Shilong __update_reloc_root(root, cow->start); 4695c974c464SWang Shilong } 4696c974c464SWang Shilong 46973fd0a558SYan, Zheng level = btrfs_header_level(buf); 46983fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46993fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 47003fd0a558SYan, Zheng first_cow = 1; 47013fd0a558SYan, Zheng 47023fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 47033fd0a558SYan, Zheng rc->create_reloc_tree) { 47043fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 47053fd0a558SYan, Zheng 47063fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 47073fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 47083fd0a558SYan, Zheng node->new_bytenr != buf->start); 47093fd0a558SYan, Zheng 47103fd0a558SYan, Zheng drop_node_buffer(node); 47113fd0a558SYan, Zheng extent_buffer_get(cow); 47123fd0a558SYan, Zheng node->eb = cow; 47133fd0a558SYan, Zheng node->new_bytenr = cow->start; 47143fd0a558SYan, Zheng 47153fd0a558SYan, Zheng if (!node->pending) { 47163fd0a558SYan, Zheng list_move_tail(&node->list, 47173fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 47183fd0a558SYan, Zheng node->pending = 1; 47193fd0a558SYan, Zheng } 47203fd0a558SYan, Zheng 47213fd0a558SYan, Zheng if (first_cow) 47223fd0a558SYan, Zheng __mark_block_processed(rc, node); 47233fd0a558SYan, Zheng 47243fd0a558SYan, Zheng if (first_cow && level > 0) 47253fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 47263fd0a558SYan, Zheng } 47273fd0a558SYan, Zheng 472883d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 47293fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 473083d4cfd4SJosef Bacik return ret; 47313fd0a558SYan, Zheng } 47323fd0a558SYan, Zheng 47333fd0a558SYan, Zheng /* 47343fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 473501327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 47363fd0a558SYan, Zheng */ 4737147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 47383fd0a558SYan, Zheng u64 *bytes_to_reserve) 47393fd0a558SYan, Zheng { 47403fd0a558SYan, Zheng struct btrfs_root *root; 47413fd0a558SYan, Zheng struct reloc_control *rc; 47423fd0a558SYan, Zheng 47433fd0a558SYan, Zheng root = pending->root; 47443fd0a558SYan, Zheng if (!root->reloc_root) 47453fd0a558SYan, Zheng return; 47463fd0a558SYan, Zheng 47473fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47483fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47493fd0a558SYan, Zheng return; 47503fd0a558SYan, Zheng 47513fd0a558SYan, Zheng root = root->reloc_root; 47523fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47533fd0a558SYan, Zheng /* 47543fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47553fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47563fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47573fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47583fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47593fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 47603fd0a558SYan, Zheng * tree may use more space than it frees. so we need 47613fd0a558SYan, Zheng * reserve extra space. 47623fd0a558SYan, Zheng */ 47633fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47643fd0a558SYan, Zheng } 47653fd0a558SYan, Zheng 47663fd0a558SYan, Zheng /* 47673fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47683fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 47693fd0a558SYan, Zheng */ 477049b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 47713fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 47723fd0a558SYan, Zheng { 47733fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 47743fd0a558SYan, Zheng struct btrfs_root *reloc_root; 47753fd0a558SYan, Zheng struct btrfs_root *new_root; 47763fd0a558SYan, Zheng struct reloc_control *rc; 47773fd0a558SYan, Zheng int ret; 47783fd0a558SYan, Zheng 47793fd0a558SYan, Zheng if (!root->reloc_root) 478049b25e05SJeff Mahoney return 0; 47813fd0a558SYan, Zheng 47823fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47833fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 47843fd0a558SYan, Zheng 47853fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 47863fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 47873fd0a558SYan, Zheng rc->block_rsv, 478825d609f8SJosef Bacik rc->nodes_relocated, 1); 478949b25e05SJeff Mahoney if (ret) 479049b25e05SJeff Mahoney return ret; 47913fd0a558SYan, Zheng } 47923fd0a558SYan, Zheng 47933fd0a558SYan, Zheng new_root = pending->snap; 47943fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47953fd0a558SYan, Zheng new_root->root_key.objectid); 479649b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 479749b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47983fd0a558SYan, Zheng 4799ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4800ffd7b339SJeff Mahoney BUG_ON(ret < 0); 48013fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 48023fd0a558SYan, Zheng 480349b25e05SJeff Mahoney if (rc->create_reloc_tree) 48043fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 480549b25e05SJeff Mahoney return ret; 48063fd0a558SYan, Zheng } 4807