1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 25d4f98a2SYan Zheng /* 35d4f98a2SYan Zheng * Copyright (C) 2009 Oracle. All rights reserved. 45d4f98a2SYan Zheng */ 55d4f98a2SYan Zheng 65d4f98a2SYan Zheng #include <linux/sched.h> 75d4f98a2SYan Zheng #include <linux/pagemap.h> 85d4f98a2SYan Zheng #include <linux/writeback.h> 95d4f98a2SYan Zheng #include <linux/blkdev.h> 105d4f98a2SYan Zheng #include <linux/rbtree.h> 115a0e3ad6STejun Heo #include <linux/slab.h> 125d4f98a2SYan Zheng #include "ctree.h" 135d4f98a2SYan Zheng #include "disk-io.h" 145d4f98a2SYan Zheng #include "transaction.h" 155d4f98a2SYan Zheng #include "volumes.h" 165d4f98a2SYan Zheng #include "locking.h" 175d4f98a2SYan Zheng #include "btrfs_inode.h" 185d4f98a2SYan Zheng #include "async-thread.h" 190af3d00bSJosef Bacik #include "free-space-cache.h" 20581bb050SLi Zefan #include "inode-map.h" 2162b99540SQu Wenruo #include "qgroup.h" 22cdccee99SLiu Bo #include "print-tree.h" 2386736342SJosef Bacik #include "delalloc-space.h" 24aac0023cSJosef Bacik #include "block-group.h" 255d4f98a2SYan Zheng 265d4f98a2SYan Zheng /* 275d4f98a2SYan Zheng * backref_node, mapping_node and tree_block start with this 285d4f98a2SYan Zheng */ 295d4f98a2SYan Zheng struct tree_entry { 305d4f98a2SYan Zheng struct rb_node rb_node; 315d4f98a2SYan Zheng u64 bytenr; 325d4f98a2SYan Zheng }; 335d4f98a2SYan Zheng 345d4f98a2SYan Zheng /* 355d4f98a2SYan Zheng * present a tree block in the backref cache 365d4f98a2SYan Zheng */ 375d4f98a2SYan Zheng struct backref_node { 385d4f98a2SYan Zheng struct rb_node rb_node; 395d4f98a2SYan Zheng u64 bytenr; 403fd0a558SYan, Zheng 413fd0a558SYan, Zheng u64 new_bytenr; 423fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 435d4f98a2SYan Zheng u64 owner; 443fd0a558SYan, Zheng /* link to pending, changed or detached list */ 453fd0a558SYan, Zheng struct list_head list; 465d4f98a2SYan Zheng /* list of upper level blocks reference this block */ 475d4f98a2SYan Zheng struct list_head upper; 485d4f98a2SYan Zheng /* list of child blocks in the cache */ 495d4f98a2SYan Zheng struct list_head lower; 505d4f98a2SYan Zheng /* NULL if this node is not tree root */ 515d4f98a2SYan Zheng struct btrfs_root *root; 525d4f98a2SYan Zheng /* extent buffer got by COW the block */ 535d4f98a2SYan Zheng struct extent_buffer *eb; 545d4f98a2SYan Zheng /* level of tree block */ 555d4f98a2SYan Zheng unsigned int level:8; 563fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 573fd0a558SYan, Zheng unsigned int cowonly:1; 583fd0a558SYan, Zheng /* 1 if no child node in the cache */ 595d4f98a2SYan Zheng unsigned int lowest:1; 605d4f98a2SYan Zheng /* is the extent buffer locked */ 615d4f98a2SYan Zheng unsigned int locked:1; 625d4f98a2SYan Zheng /* has the block been processed */ 635d4f98a2SYan Zheng unsigned int processed:1; 645d4f98a2SYan Zheng /* have backrefs of this block been checked */ 655d4f98a2SYan Zheng unsigned int checked:1; 663fd0a558SYan, Zheng /* 673fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 683fd0a558SYan, Zheng * level block pointers may not point to the new location 693fd0a558SYan, Zheng */ 703fd0a558SYan, Zheng unsigned int pending:1; 713fd0a558SYan, Zheng /* 723fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 733fd0a558SYan, Zheng * backref node. 743fd0a558SYan, Zheng */ 753fd0a558SYan, Zheng unsigned int detached:1; 765d4f98a2SYan Zheng }; 775d4f98a2SYan Zheng 785d4f98a2SYan Zheng /* 795d4f98a2SYan Zheng * present a block pointer in the backref cache 805d4f98a2SYan Zheng */ 815d4f98a2SYan Zheng struct backref_edge { 825d4f98a2SYan Zheng struct list_head list[2]; 835d4f98a2SYan Zheng struct backref_node *node[2]; 845d4f98a2SYan Zheng }; 855d4f98a2SYan Zheng 865d4f98a2SYan Zheng #define LOWER 0 875d4f98a2SYan Zheng #define UPPER 1 880647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 895d4f98a2SYan Zheng 905d4f98a2SYan Zheng struct backref_cache { 915d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 925d4f98a2SYan Zheng struct rb_root rb_root; 933fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 943fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 953fd0a558SYan, Zheng /* 963fd0a558SYan, Zheng * list of blocks that have been cowed but some block 973fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 983fd0a558SYan, Zheng * new location 993fd0a558SYan, Zheng */ 1005d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1013fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1023fd0a558SYan, Zheng struct list_head leaves; 1033fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1043fd0a558SYan, Zheng struct list_head changed; 1053fd0a558SYan, Zheng /* list of detached backref node. */ 1063fd0a558SYan, Zheng struct list_head detached; 1073fd0a558SYan, Zheng 1083fd0a558SYan, Zheng u64 last_trans; 1093fd0a558SYan, Zheng 1103fd0a558SYan, Zheng int nr_nodes; 1113fd0a558SYan, Zheng int nr_edges; 1125d4f98a2SYan Zheng }; 1135d4f98a2SYan Zheng 1145d4f98a2SYan Zheng /* 1155d4f98a2SYan Zheng * map address of tree root to tree 1165d4f98a2SYan Zheng */ 1175d4f98a2SYan Zheng struct mapping_node { 1185d4f98a2SYan Zheng struct rb_node rb_node; 1195d4f98a2SYan Zheng u64 bytenr; 1205d4f98a2SYan Zheng void *data; 1215d4f98a2SYan Zheng }; 1225d4f98a2SYan Zheng 1235d4f98a2SYan Zheng struct mapping_tree { 1245d4f98a2SYan Zheng struct rb_root rb_root; 1255d4f98a2SYan Zheng spinlock_t lock; 1265d4f98a2SYan Zheng }; 1275d4f98a2SYan Zheng 1285d4f98a2SYan Zheng /* 1295d4f98a2SYan Zheng * present a tree block to process 1305d4f98a2SYan Zheng */ 1315d4f98a2SYan Zheng struct tree_block { 1325d4f98a2SYan Zheng struct rb_node rb_node; 1335d4f98a2SYan Zheng u64 bytenr; 1345d4f98a2SYan Zheng struct btrfs_key key; 1355d4f98a2SYan Zheng unsigned int level:8; 1365d4f98a2SYan Zheng unsigned int key_ready:1; 1375d4f98a2SYan Zheng }; 1385d4f98a2SYan Zheng 1390257bb82SYan, Zheng #define MAX_EXTENTS 128 1400257bb82SYan, Zheng 1410257bb82SYan, Zheng struct file_extent_cluster { 1420257bb82SYan, Zheng u64 start; 1430257bb82SYan, Zheng u64 end; 1440257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1450257bb82SYan, Zheng unsigned int nr; 1460257bb82SYan, Zheng }; 1470257bb82SYan, Zheng 1485d4f98a2SYan Zheng struct reloc_control { 1495d4f98a2SYan Zheng /* block group to relocate */ 15032da5386SDavid Sterba struct btrfs_block_group *block_group; 1515d4f98a2SYan Zheng /* extent tree */ 1525d4f98a2SYan Zheng struct btrfs_root *extent_root; 1535d4f98a2SYan Zheng /* inode for moving data */ 1545d4f98a2SYan Zheng struct inode *data_inode; 1553fd0a558SYan, Zheng 1563fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1573fd0a558SYan, Zheng 1583fd0a558SYan, Zheng struct backref_cache backref_cache; 1593fd0a558SYan, Zheng 1603fd0a558SYan, Zheng struct file_extent_cluster cluster; 1615d4f98a2SYan Zheng /* tree blocks have been processed */ 1625d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1635d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1645d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1655d4f98a2SYan Zheng /* list of reloc trees */ 1665d4f98a2SYan Zheng struct list_head reloc_roots; 167d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 168d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 1693fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1703fd0a558SYan, Zheng u64 merging_rsv_size; 1713fd0a558SYan, Zheng /* size of relocated tree nodes */ 1723fd0a558SYan, Zheng u64 nodes_relocated; 1730647bf56SWang Shilong /* reserved size for block group relocation*/ 1740647bf56SWang Shilong u64 reserved_bytes; 1753fd0a558SYan, Zheng 1765d4f98a2SYan Zheng u64 search_start; 1775d4f98a2SYan Zheng u64 extents_found; 1783fd0a558SYan, Zheng 1793fd0a558SYan, Zheng unsigned int stage:8; 1803fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1813fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1825d4f98a2SYan Zheng unsigned int found_file_extent:1; 1835d4f98a2SYan Zheng }; 1845d4f98a2SYan Zheng 1855d4f98a2SYan Zheng /* stages of data relocation */ 1865d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1875d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1885d4f98a2SYan Zheng 1893fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 1903fd0a558SYan, Zheng struct backref_node *node); 1913fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 1923fd0a558SYan, Zheng struct backref_node *node); 1935d4f98a2SYan Zheng 1945d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1955d4f98a2SYan Zheng { 1966bef4d31SEric Paris tree->rb_root = RB_ROOT; 1975d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1985d4f98a2SYan Zheng } 1995d4f98a2SYan Zheng 2005d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 2015d4f98a2SYan Zheng { 2025d4f98a2SYan Zheng int i; 2036bef4d31SEric Paris cache->rb_root = RB_ROOT; 2045d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2055d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2063fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2073fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2083fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2095d4f98a2SYan Zheng } 2105d4f98a2SYan Zheng 2113fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2125d4f98a2SYan Zheng { 2133fd0a558SYan, Zheng struct backref_node *node; 2143fd0a558SYan, Zheng int i; 2153fd0a558SYan, Zheng 2163fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2173fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2183fd0a558SYan, Zheng struct backref_node, list); 2193fd0a558SYan, Zheng remove_backref_node(cache, node); 2203fd0a558SYan, Zheng } 2213fd0a558SYan, Zheng 2223fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2233fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2243fd0a558SYan, Zheng struct backref_node, lower); 2253fd0a558SYan, Zheng remove_backref_node(cache, node); 2263fd0a558SYan, Zheng } 2273fd0a558SYan, Zheng 2283fd0a558SYan, Zheng cache->last_trans = 0; 2293fd0a558SYan, Zheng 2303fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 231f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 232f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 233f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 234f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 235f4907095SLiu Bo ASSERT(!cache->nr_nodes); 236f4907095SLiu Bo ASSERT(!cache->nr_edges); 2373fd0a558SYan, Zheng } 2383fd0a558SYan, Zheng 2393fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2403fd0a558SYan, Zheng { 2413fd0a558SYan, Zheng struct backref_node *node; 2423fd0a558SYan, Zheng 2433fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2443fd0a558SYan, Zheng if (node) { 2453fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2465d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2475d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2485d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2493fd0a558SYan, Zheng cache->nr_nodes++; 2503fd0a558SYan, Zheng } 2513fd0a558SYan, Zheng return node; 2523fd0a558SYan, Zheng } 2533fd0a558SYan, Zheng 2543fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 2553fd0a558SYan, Zheng struct backref_node *node) 2563fd0a558SYan, Zheng { 2573fd0a558SYan, Zheng if (node) { 2583fd0a558SYan, Zheng cache->nr_nodes--; 2593fd0a558SYan, Zheng kfree(node); 2603fd0a558SYan, Zheng } 2613fd0a558SYan, Zheng } 2623fd0a558SYan, Zheng 2633fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 2643fd0a558SYan, Zheng { 2653fd0a558SYan, Zheng struct backref_edge *edge; 2663fd0a558SYan, Zheng 2673fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 2683fd0a558SYan, Zheng if (edge) 2693fd0a558SYan, Zheng cache->nr_edges++; 2703fd0a558SYan, Zheng return edge; 2713fd0a558SYan, Zheng } 2723fd0a558SYan, Zheng 2733fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 2743fd0a558SYan, Zheng struct backref_edge *edge) 2753fd0a558SYan, Zheng { 2763fd0a558SYan, Zheng if (edge) { 2773fd0a558SYan, Zheng cache->nr_edges--; 2783fd0a558SYan, Zheng kfree(edge); 2793fd0a558SYan, Zheng } 2805d4f98a2SYan Zheng } 2815d4f98a2SYan Zheng 2825d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 2835d4f98a2SYan Zheng struct rb_node *node) 2845d4f98a2SYan Zheng { 2855d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 2865d4f98a2SYan Zheng struct rb_node *parent = NULL; 2875d4f98a2SYan Zheng struct tree_entry *entry; 2885d4f98a2SYan Zheng 2895d4f98a2SYan Zheng while (*p) { 2905d4f98a2SYan Zheng parent = *p; 2915d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 2925d4f98a2SYan Zheng 2935d4f98a2SYan Zheng if (bytenr < entry->bytenr) 2945d4f98a2SYan Zheng p = &(*p)->rb_left; 2955d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 2965d4f98a2SYan Zheng p = &(*p)->rb_right; 2975d4f98a2SYan Zheng else 2985d4f98a2SYan Zheng return parent; 2995d4f98a2SYan Zheng } 3005d4f98a2SYan Zheng 3015d4f98a2SYan Zheng rb_link_node(node, parent, p); 3025d4f98a2SYan Zheng rb_insert_color(node, root); 3035d4f98a2SYan Zheng return NULL; 3045d4f98a2SYan Zheng } 3055d4f98a2SYan Zheng 3065d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3075d4f98a2SYan Zheng { 3085d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3095d4f98a2SYan Zheng struct tree_entry *entry; 3105d4f98a2SYan Zheng 3115d4f98a2SYan Zheng while (n) { 3125d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3135d4f98a2SYan Zheng 3145d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3155d4f98a2SYan Zheng n = n->rb_left; 3165d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3175d4f98a2SYan Zheng n = n->rb_right; 3185d4f98a2SYan Zheng else 3195d4f98a2SYan Zheng return n; 3205d4f98a2SYan Zheng } 3215d4f98a2SYan Zheng return NULL; 3225d4f98a2SYan Zheng } 3235d4f98a2SYan Zheng 32448a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 32543c04fb1SJeff Mahoney { 32643c04fb1SJeff Mahoney 32743c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 32843c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 32943c04fb1SJeff Mahoney rb_node); 33043c04fb1SJeff Mahoney if (bnode->root) 33143c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 3325d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 3335d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 3345d163e0eSJeff Mahoney bytenr); 33543c04fb1SJeff Mahoney } 33643c04fb1SJeff Mahoney 3375d4f98a2SYan Zheng /* 3385d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3395d4f98a2SYan Zheng */ 3405d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3415d4f98a2SYan Zheng struct backref_edge *edges[], 3425d4f98a2SYan Zheng int *index) 3435d4f98a2SYan Zheng { 3445d4f98a2SYan Zheng struct backref_edge *edge; 3455d4f98a2SYan Zheng int idx = *index; 3465d4f98a2SYan Zheng 3475d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3485d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3495d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3505d4f98a2SYan Zheng edges[idx++] = edge; 3515d4f98a2SYan Zheng node = edge->node[UPPER]; 3525d4f98a2SYan Zheng } 3533fd0a558SYan, Zheng BUG_ON(node->detached); 3545d4f98a2SYan Zheng *index = idx; 3555d4f98a2SYan Zheng return node; 3565d4f98a2SYan Zheng } 3575d4f98a2SYan Zheng 3585d4f98a2SYan Zheng /* 3595d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 3605d4f98a2SYan Zheng */ 3615d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 3625d4f98a2SYan Zheng int *index) 3635d4f98a2SYan Zheng { 3645d4f98a2SYan Zheng struct backref_edge *edge; 3655d4f98a2SYan Zheng struct backref_node *lower; 3665d4f98a2SYan Zheng int idx = *index; 3675d4f98a2SYan Zheng 3685d4f98a2SYan Zheng while (idx > 0) { 3695d4f98a2SYan Zheng edge = edges[idx - 1]; 3705d4f98a2SYan Zheng lower = edge->node[LOWER]; 3715d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 3725d4f98a2SYan Zheng idx--; 3735d4f98a2SYan Zheng continue; 3745d4f98a2SYan Zheng } 3755d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 3765d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3775d4f98a2SYan Zheng edges[idx - 1] = edge; 3785d4f98a2SYan Zheng *index = idx; 3795d4f98a2SYan Zheng return edge->node[UPPER]; 3805d4f98a2SYan Zheng } 3815d4f98a2SYan Zheng *index = 0; 3825d4f98a2SYan Zheng return NULL; 3835d4f98a2SYan Zheng } 3845d4f98a2SYan Zheng 3853fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 3865d4f98a2SYan Zheng { 3875d4f98a2SYan Zheng if (node->locked) { 3885d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 3895d4f98a2SYan Zheng node->locked = 0; 3905d4f98a2SYan Zheng } 3913fd0a558SYan, Zheng } 3923fd0a558SYan, Zheng 3933fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 3943fd0a558SYan, Zheng { 3953fd0a558SYan, Zheng if (node->eb) { 3963fd0a558SYan, Zheng unlock_node_buffer(node); 3975d4f98a2SYan Zheng free_extent_buffer(node->eb); 3985d4f98a2SYan Zheng node->eb = NULL; 3995d4f98a2SYan Zheng } 4005d4f98a2SYan Zheng } 4015d4f98a2SYan Zheng 4025d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4035d4f98a2SYan Zheng struct backref_node *node) 4045d4f98a2SYan Zheng { 4055d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4065d4f98a2SYan Zheng 4075d4f98a2SYan Zheng drop_node_buffer(node); 4083fd0a558SYan, Zheng list_del(&node->list); 4095d4f98a2SYan Zheng list_del(&node->lower); 4103fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4115d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4123fd0a558SYan, Zheng free_backref_node(tree, node); 4135d4f98a2SYan Zheng } 4145d4f98a2SYan Zheng 4155d4f98a2SYan Zheng /* 4165d4f98a2SYan Zheng * remove a backref node from the backref cache 4175d4f98a2SYan Zheng */ 4185d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4195d4f98a2SYan Zheng struct backref_node *node) 4205d4f98a2SYan Zheng { 4215d4f98a2SYan Zheng struct backref_node *upper; 4225d4f98a2SYan Zheng struct backref_edge *edge; 4235d4f98a2SYan Zheng 4245d4f98a2SYan Zheng if (!node) 4255d4f98a2SYan Zheng return; 4265d4f98a2SYan Zheng 4273fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4285d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4295d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4305d4f98a2SYan Zheng list[LOWER]); 4315d4f98a2SYan Zheng upper = edge->node[UPPER]; 4325d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4335d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4343fd0a558SYan, Zheng free_backref_edge(cache, edge); 4353fd0a558SYan, Zheng 4363fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4373fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4383fd0a558SYan, Zheng drop_backref_node(cache, node); 4393fd0a558SYan, Zheng node = upper; 4403fd0a558SYan, Zheng node->lowest = 1; 4413fd0a558SYan, Zheng continue; 4423fd0a558SYan, Zheng } 4435d4f98a2SYan Zheng /* 4443fd0a558SYan, Zheng * add the node to leaf node list if no other 4455d4f98a2SYan Zheng * child block cached. 4465d4f98a2SYan Zheng */ 4475d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4483fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4495d4f98a2SYan Zheng upper->lowest = 1; 4505d4f98a2SYan Zheng } 4515d4f98a2SYan Zheng } 4523fd0a558SYan, Zheng 4535d4f98a2SYan Zheng drop_backref_node(cache, node); 4545d4f98a2SYan Zheng } 4555d4f98a2SYan Zheng 4563fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 4573fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 4583fd0a558SYan, Zheng { 4593fd0a558SYan, Zheng struct rb_node *rb_node; 4603fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 4613fd0a558SYan, Zheng node->bytenr = bytenr; 4623fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 46343c04fb1SJeff Mahoney if (rb_node) 46443c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 4653fd0a558SYan, Zheng } 4663fd0a558SYan, Zheng 4673fd0a558SYan, Zheng /* 4683fd0a558SYan, Zheng * update backref cache after a transaction commit 4693fd0a558SYan, Zheng */ 4703fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 4713fd0a558SYan, Zheng struct backref_cache *cache) 4723fd0a558SYan, Zheng { 4733fd0a558SYan, Zheng struct backref_node *node; 4743fd0a558SYan, Zheng int level = 0; 4753fd0a558SYan, Zheng 4763fd0a558SYan, Zheng if (cache->last_trans == 0) { 4773fd0a558SYan, Zheng cache->last_trans = trans->transid; 4783fd0a558SYan, Zheng return 0; 4793fd0a558SYan, Zheng } 4803fd0a558SYan, Zheng 4813fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 4823fd0a558SYan, Zheng return 0; 4833fd0a558SYan, Zheng 4843fd0a558SYan, Zheng /* 4853fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 4863fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 4873fd0a558SYan, Zheng * so the detached nodes are no longer useful. 4883fd0a558SYan, Zheng */ 4893fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 4903fd0a558SYan, Zheng node = list_entry(cache->detached.next, 4913fd0a558SYan, Zheng struct backref_node, list); 4923fd0a558SYan, Zheng remove_backref_node(cache, node); 4933fd0a558SYan, Zheng } 4943fd0a558SYan, Zheng 4953fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 4963fd0a558SYan, Zheng node = list_entry(cache->changed.next, 4973fd0a558SYan, Zheng struct backref_node, list); 4983fd0a558SYan, Zheng list_del_init(&node->list); 4993fd0a558SYan, Zheng BUG_ON(node->pending); 5003fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5013fd0a558SYan, Zheng } 5023fd0a558SYan, Zheng 5033fd0a558SYan, Zheng /* 5043fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5053fd0a558SYan, Zheng * errors during processing the pending nodes. 5063fd0a558SYan, Zheng */ 5073fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5083fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5093fd0a558SYan, Zheng BUG_ON(!node->pending); 5103fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5113fd0a558SYan, Zheng continue; 5123fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5133fd0a558SYan, Zheng } 5143fd0a558SYan, Zheng } 5153fd0a558SYan, Zheng 5163fd0a558SYan, Zheng cache->last_trans = 0; 5173fd0a558SYan, Zheng return 1; 5183fd0a558SYan, Zheng } 5193fd0a558SYan, Zheng 520f2a97a9dSDavid Sterba 5213fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5223fd0a558SYan, Zheng { 5233fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5243fd0a558SYan, Zheng 52527cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 5263fd0a558SYan, Zheng return 0; 5273fd0a558SYan, Zheng 5283fd0a558SYan, Zheng reloc_root = root->reloc_root; 5293fd0a558SYan, Zheng if (!reloc_root) 5303fd0a558SYan, Zheng return 0; 5313fd0a558SYan, Zheng 5323fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5333fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5343fd0a558SYan, Zheng return 0; 5353fd0a558SYan, Zheng /* 5363fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5373fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5383fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5393fd0a558SYan, Zheng * relocation. 5403fd0a558SYan, Zheng */ 5413fd0a558SYan, Zheng return 1; 5423fd0a558SYan, Zheng } 5435d4f98a2SYan Zheng /* 5445d4f98a2SYan Zheng * find reloc tree by address of tree root 5455d4f98a2SYan Zheng */ 5465d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5475d4f98a2SYan Zheng u64 bytenr) 5485d4f98a2SYan Zheng { 5495d4f98a2SYan Zheng struct rb_node *rb_node; 5505d4f98a2SYan Zheng struct mapping_node *node; 5515d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5525d4f98a2SYan Zheng 5535d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5545d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5555d4f98a2SYan Zheng if (rb_node) { 5565d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5575d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5585d4f98a2SYan Zheng } 5595d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5605d4f98a2SYan Zheng return root; 5615d4f98a2SYan Zheng } 5625d4f98a2SYan Zheng 5635d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5645d4f98a2SYan Zheng { 5655d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5665d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5675d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 5685d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 5695d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 57066463748SWang Shilong root_objectid == BTRFS_CSUM_TREE_OBJECTID || 57166463748SWang Shilong root_objectid == BTRFS_UUID_TREE_OBJECTID || 5723e4c5efbSDavid Sterba root_objectid == BTRFS_QUOTA_TREE_OBJECTID || 5733e4c5efbSDavid Sterba root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 5745d4f98a2SYan Zheng return 1; 5755d4f98a2SYan Zheng return 0; 5765d4f98a2SYan Zheng } 5775d4f98a2SYan Zheng 5785d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 5795d4f98a2SYan Zheng u64 root_objectid) 5805d4f98a2SYan Zheng { 5815d4f98a2SYan Zheng struct btrfs_key key; 5825d4f98a2SYan Zheng 5835d4f98a2SYan Zheng key.objectid = root_objectid; 5845d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 5855d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 5865d4f98a2SYan Zheng key.offset = 0; 5875d4f98a2SYan Zheng else 5885d4f98a2SYan Zheng key.offset = (u64)-1; 5895d4f98a2SYan Zheng 590c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 5915d4f98a2SYan Zheng } 5925d4f98a2SYan Zheng 5935d4f98a2SYan Zheng static noinline_for_stack 5945d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 5955d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 5965d4f98a2SYan Zheng { 5973173a18fSJosef Bacik struct btrfs_key key; 5985d4f98a2SYan Zheng struct btrfs_extent_item *ei; 5995d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6005d4f98a2SYan Zheng u32 item_size; 6015d4f98a2SYan Zheng 6023173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6033173a18fSJosef Bacik 6045d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 605ba3c2b19SNikolay Borisov if (item_size < sizeof(*ei)) { 606ba3c2b19SNikolay Borisov btrfs_print_v0_err(leaf->fs_info); 607ba3c2b19SNikolay Borisov btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL); 608ba3c2b19SNikolay Borisov return 1; 609ba3c2b19SNikolay Borisov } 6105d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6115d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6125d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6135d4f98a2SYan Zheng 6143173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6153173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6165d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6175d4f98a2SYan Zheng return 1; 6185d4f98a2SYan Zheng } 619d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 620d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 621d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 622d062d13cSJosef Bacik return 1; 623d062d13cSJosef Bacik } 6245d4f98a2SYan Zheng 6253173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6265d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6275d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6283173a18fSJosef Bacik } else { 6293173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6303173a18fSJosef Bacik } 6315d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6325d4f98a2SYan Zheng return 0; 6335d4f98a2SYan Zheng } 6345d4f98a2SYan Zheng 6355d4f98a2SYan Zheng /* 6365d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6375d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6385d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6395d4f98a2SYan Zheng * 6405d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 64101327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 64201327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 6435d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6445d4f98a2SYan Zheng * 6455d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6465d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6475d4f98a2SYan Zheng * block are also cached. 6485d4f98a2SYan Zheng */ 6493fd0a558SYan, Zheng static noinline_for_stack 6503fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6515d4f98a2SYan Zheng struct btrfs_key *node_key, 6525d4f98a2SYan Zheng int level, u64 bytenr) 6535d4f98a2SYan Zheng { 6543fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 655fa6ac715SQu Wenruo struct btrfs_path *path1; /* For searching extent root */ 656fa6ac715SQu Wenruo struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */ 6575d4f98a2SYan Zheng struct extent_buffer *eb; 6585d4f98a2SYan Zheng struct btrfs_root *root; 6595d4f98a2SYan Zheng struct backref_node *cur; 6605d4f98a2SYan Zheng struct backref_node *upper; 6615d4f98a2SYan Zheng struct backref_node *lower; 6625d4f98a2SYan Zheng struct backref_node *node = NULL; 6635d4f98a2SYan Zheng struct backref_node *exist = NULL; 6645d4f98a2SYan Zheng struct backref_edge *edge; 6655d4f98a2SYan Zheng struct rb_node *rb_node; 6665d4f98a2SYan Zheng struct btrfs_key key; 6675d4f98a2SYan Zheng unsigned long end; 6685d4f98a2SYan Zheng unsigned long ptr; 669fa6ac715SQu Wenruo LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */ 6703fd0a558SYan, Zheng LIST_HEAD(useless); 6713fd0a558SYan, Zheng int cowonly; 6725d4f98a2SYan Zheng int ret; 6735d4f98a2SYan Zheng int err = 0; 674b6c60c80SJosef Bacik bool need_check = true; 6755d4f98a2SYan Zheng 6765d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 6775d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 6785d4f98a2SYan Zheng if (!path1 || !path2) { 6795d4f98a2SYan Zheng err = -ENOMEM; 6805d4f98a2SYan Zheng goto out; 6815d4f98a2SYan Zheng } 682e4058b54SDavid Sterba path1->reada = READA_FORWARD; 683e4058b54SDavid Sterba path2->reada = READA_FORWARD; 6845d4f98a2SYan Zheng 6853fd0a558SYan, Zheng node = alloc_backref_node(cache); 6865d4f98a2SYan Zheng if (!node) { 6875d4f98a2SYan Zheng err = -ENOMEM; 6885d4f98a2SYan Zheng goto out; 6895d4f98a2SYan Zheng } 6905d4f98a2SYan Zheng 6915d4f98a2SYan Zheng node->bytenr = bytenr; 6925d4f98a2SYan Zheng node->level = level; 6935d4f98a2SYan Zheng node->lowest = 1; 6945d4f98a2SYan Zheng cur = node; 6955d4f98a2SYan Zheng again: 6965d4f98a2SYan Zheng end = 0; 6975d4f98a2SYan Zheng ptr = 0; 6985d4f98a2SYan Zheng key.objectid = cur->bytenr; 6993173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7005d4f98a2SYan Zheng key.offset = (u64)-1; 7015d4f98a2SYan Zheng 7025d4f98a2SYan Zheng path1->search_commit_root = 1; 7035d4f98a2SYan Zheng path1->skip_locking = 1; 7045d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7055d4f98a2SYan Zheng 0, 0); 7065d4f98a2SYan Zheng if (ret < 0) { 7075d4f98a2SYan Zheng err = ret; 7085d4f98a2SYan Zheng goto out; 7095d4f98a2SYan Zheng } 71075bfb9afSJosef Bacik ASSERT(ret); 71175bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7125d4f98a2SYan Zheng 7135d4f98a2SYan Zheng path1->slots[0]--; 7145d4f98a2SYan Zheng 7155d4f98a2SYan Zheng WARN_ON(cur->checked); 7165d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7175d4f98a2SYan Zheng /* 71870f23fd6SJustin P. Mattock * the backref was added previously when processing 7195d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7205d4f98a2SYan Zheng */ 72175bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7225d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7235d4f98a2SYan Zheng list[LOWER]); 72475bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7255d4f98a2SYan Zheng exist = edge->node[UPPER]; 7265d4f98a2SYan Zheng /* 7275d4f98a2SYan Zheng * add the upper level block to pending list if we need 7285d4f98a2SYan Zheng * check its backrefs 7295d4f98a2SYan Zheng */ 7305d4f98a2SYan Zheng if (!exist->checked) 7315d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7325d4f98a2SYan Zheng } else { 7335d4f98a2SYan Zheng exist = NULL; 7345d4f98a2SYan Zheng } 7355d4f98a2SYan Zheng 7365d4f98a2SYan Zheng while (1) { 7375d4f98a2SYan Zheng cond_resched(); 7385d4f98a2SYan Zheng eb = path1->nodes[0]; 7395d4f98a2SYan Zheng 7405d4f98a2SYan Zheng if (ptr >= end) { 7415d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7425d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7435d4f98a2SYan Zheng if (ret < 0) { 7445d4f98a2SYan Zheng err = ret; 7455d4f98a2SYan Zheng goto out; 7465d4f98a2SYan Zheng } 7475d4f98a2SYan Zheng if (ret > 0) 7485d4f98a2SYan Zheng break; 7495d4f98a2SYan Zheng eb = path1->nodes[0]; 7505d4f98a2SYan Zheng } 7515d4f98a2SYan Zheng 7525d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7535d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7545d4f98a2SYan Zheng WARN_ON(exist); 7555d4f98a2SYan Zheng break; 7565d4f98a2SYan Zheng } 7575d4f98a2SYan Zheng 7583173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7593173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7605d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7615d4f98a2SYan Zheng &ptr, &end); 7625d4f98a2SYan Zheng if (ret) 7635d4f98a2SYan Zheng goto next; 7645d4f98a2SYan Zheng } 7655d4f98a2SYan Zheng } 7665d4f98a2SYan Zheng 7675d4f98a2SYan Zheng if (ptr < end) { 7685d4f98a2SYan Zheng /* update key for inline back ref */ 7695d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 7703de28d57SLiu Bo int type; 7715d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 7723de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 7733de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 7743de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 775af431dcbSSu Yue err = -EUCLEAN; 7763de28d57SLiu Bo goto out; 7773de28d57SLiu Bo } 7783de28d57SLiu Bo key.type = type; 7795d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 7803de28d57SLiu Bo 7815d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 7825d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 7835d4f98a2SYan Zheng } 7845d4f98a2SYan Zheng 785fa6ac715SQu Wenruo /* 786fa6ac715SQu Wenruo * Parent node found and matches current inline ref, no need to 787fa6ac715SQu Wenruo * rebuild this node for this inline ref. 788fa6ac715SQu Wenruo */ 7895d4f98a2SYan Zheng if (exist && 7905d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 7915d4f98a2SYan Zheng exist->owner == key.offset) || 7925d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 7935d4f98a2SYan Zheng exist->bytenr == key.offset))) { 7945d4f98a2SYan Zheng exist = NULL; 7955d4f98a2SYan Zheng goto next; 7965d4f98a2SYan Zheng } 7975d4f98a2SYan Zheng 798fa6ac715SQu Wenruo /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ 7995d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8005d4f98a2SYan Zheng if (key.objectid == key.offset) { 8015d4f98a2SYan Zheng /* 802fa6ac715SQu Wenruo * Only root blocks of reloc trees use backref 803fa6ac715SQu Wenruo * pointing to itself. 8045d4f98a2SYan Zheng */ 8055d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 80675bfb9afSJosef Bacik ASSERT(root); 8075d4f98a2SYan Zheng cur->root = root; 8085d4f98a2SYan Zheng break; 8095d4f98a2SYan Zheng } 8105d4f98a2SYan Zheng 8113fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8125d4f98a2SYan Zheng if (!edge) { 8135d4f98a2SYan Zheng err = -ENOMEM; 8145d4f98a2SYan Zheng goto out; 8155d4f98a2SYan Zheng } 8165d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8175d4f98a2SYan Zheng if (!rb_node) { 8183fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8195d4f98a2SYan Zheng if (!upper) { 8203fd0a558SYan, Zheng free_backref_edge(cache, edge); 8215d4f98a2SYan Zheng err = -ENOMEM; 8225d4f98a2SYan Zheng goto out; 8235d4f98a2SYan Zheng } 8245d4f98a2SYan Zheng upper->bytenr = key.offset; 8255d4f98a2SYan Zheng upper->level = cur->level + 1; 8265d4f98a2SYan Zheng /* 8275d4f98a2SYan Zheng * backrefs for the upper level block isn't 8285d4f98a2SYan Zheng * cached, add the block to pending list 8295d4f98a2SYan Zheng */ 8305d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8315d4f98a2SYan Zheng } else { 8325d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8335d4f98a2SYan Zheng rb_node); 83475bfb9afSJosef Bacik ASSERT(upper->checked); 8355d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8365d4f98a2SYan Zheng } 8373fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8385d4f98a2SYan Zheng edge->node[LOWER] = cur; 8393fd0a558SYan, Zheng edge->node[UPPER] = upper; 8405d4f98a2SYan Zheng 8415d4f98a2SYan Zheng goto next; 8426d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 843ba3c2b19SNikolay Borisov err = -EINVAL; 844ba3c2b19SNikolay Borisov btrfs_print_v0_err(rc->extent_root->fs_info); 845ba3c2b19SNikolay Borisov btrfs_handle_fs_error(rc->extent_root->fs_info, err, 846ba3c2b19SNikolay Borisov NULL); 847ba3c2b19SNikolay Borisov goto out; 8485d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8495d4f98a2SYan Zheng goto next; 8505d4f98a2SYan Zheng } 8515d4f98a2SYan Zheng 852fa6ac715SQu Wenruo /* 853fa6ac715SQu Wenruo * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset 854fa6ac715SQu Wenruo * means the root objectid. We need to search the tree to get 855fa6ac715SQu Wenruo * its parent bytenr. 856fa6ac715SQu Wenruo */ 8575d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8585d4f98a2SYan Zheng if (IS_ERR(root)) { 8595d4f98a2SYan Zheng err = PTR_ERR(root); 8605d4f98a2SYan Zheng goto out; 8615d4f98a2SYan Zheng } 8625d4f98a2SYan Zheng 86327cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 8643fd0a558SYan, Zheng cur->cowonly = 1; 8653fd0a558SYan, Zheng 8665d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 8675d4f98a2SYan Zheng /* tree root */ 86875bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 8695d4f98a2SYan Zheng cur->bytenr); 8703fd0a558SYan, Zheng if (should_ignore_root(root)) 8713fd0a558SYan, Zheng list_add(&cur->list, &useless); 8723fd0a558SYan, Zheng else 8735d4f98a2SYan Zheng cur->root = root; 8745d4f98a2SYan Zheng break; 8755d4f98a2SYan Zheng } 8765d4f98a2SYan Zheng 8775d4f98a2SYan Zheng level = cur->level + 1; 8785d4f98a2SYan Zheng 879fa6ac715SQu Wenruo /* Search the tree to find parent blocks referring the block. */ 8805d4f98a2SYan Zheng path2->search_commit_root = 1; 8815d4f98a2SYan Zheng path2->skip_locking = 1; 8825d4f98a2SYan Zheng path2->lowest_level = level; 8835d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 8845d4f98a2SYan Zheng path2->lowest_level = 0; 8855d4f98a2SYan Zheng if (ret < 0) { 8865d4f98a2SYan Zheng err = ret; 8875d4f98a2SYan Zheng goto out; 8885d4f98a2SYan Zheng } 88933c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 89033c66f43SYan Zheng path2->slots[level]--; 8915d4f98a2SYan Zheng 8925d4f98a2SYan Zheng eb = path2->nodes[level]; 8933561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 8943561b9dbSLiu Bo cur->bytenr) { 8953561b9dbSLiu Bo btrfs_err(root->fs_info, 8963561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 8974fd786e6SMisono Tomohiro cur->bytenr, level - 1, 8984fd786e6SMisono Tomohiro root->root_key.objectid, 8993561b9dbSLiu Bo node_key->objectid, node_key->type, 9003561b9dbSLiu Bo node_key->offset); 9013561b9dbSLiu Bo err = -ENOENT; 9023561b9dbSLiu Bo goto out; 9033561b9dbSLiu Bo } 9045d4f98a2SYan Zheng lower = cur; 905b6c60c80SJosef Bacik need_check = true; 906fa6ac715SQu Wenruo 907fa6ac715SQu Wenruo /* Add all nodes and edges in the path */ 9085d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9095d4f98a2SYan Zheng if (!path2->nodes[level]) { 91075bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9115d4f98a2SYan Zheng lower->bytenr); 9123fd0a558SYan, Zheng if (should_ignore_root(root)) 9133fd0a558SYan, Zheng list_add(&lower->list, &useless); 9143fd0a558SYan, Zheng else 9155d4f98a2SYan Zheng lower->root = root; 9165d4f98a2SYan Zheng break; 9175d4f98a2SYan Zheng } 9185d4f98a2SYan Zheng 9193fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9205d4f98a2SYan Zheng if (!edge) { 9215d4f98a2SYan Zheng err = -ENOMEM; 9225d4f98a2SYan Zheng goto out; 9235d4f98a2SYan Zheng } 9245d4f98a2SYan Zheng 9255d4f98a2SYan Zheng eb = path2->nodes[level]; 9265d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9275d4f98a2SYan Zheng if (!rb_node) { 9283fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9295d4f98a2SYan Zheng if (!upper) { 9303fd0a558SYan, Zheng free_backref_edge(cache, edge); 9315d4f98a2SYan Zheng err = -ENOMEM; 9325d4f98a2SYan Zheng goto out; 9335d4f98a2SYan Zheng } 9345d4f98a2SYan Zheng upper->bytenr = eb->start; 9355d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9365d4f98a2SYan Zheng upper->level = lower->level + 1; 93727cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 93827cdeb70SMiao Xie &root->state)) 9393fd0a558SYan, Zheng upper->cowonly = 1; 9405d4f98a2SYan Zheng 9415d4f98a2SYan Zheng /* 9425d4f98a2SYan Zheng * if we know the block isn't shared 9435d4f98a2SYan Zheng * we can void checking its backrefs. 9445d4f98a2SYan Zheng */ 9455d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9465d4f98a2SYan Zheng upper->checked = 0; 9475d4f98a2SYan Zheng else 9485d4f98a2SYan Zheng upper->checked = 1; 9495d4f98a2SYan Zheng 9505d4f98a2SYan Zheng /* 9515d4f98a2SYan Zheng * add the block to pending list if we 952b6c60c80SJosef Bacik * need check its backrefs, we only do this once 953b6c60c80SJosef Bacik * while walking up a tree as we will catch 954b6c60c80SJosef Bacik * anything else later on. 9555d4f98a2SYan Zheng */ 956b6c60c80SJosef Bacik if (!upper->checked && need_check) { 957b6c60c80SJosef Bacik need_check = false; 9585d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9595d4f98a2SYan Zheng &list); 960bbe90514SJosef Bacik } else { 961bbe90514SJosef Bacik if (upper->checked) 962bbe90514SJosef Bacik need_check = true; 9635d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 964bbe90514SJosef Bacik } 9655d4f98a2SYan Zheng } else { 9665d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9675d4f98a2SYan Zheng rb_node); 96875bfb9afSJosef Bacik ASSERT(upper->checked); 9695d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9703fd0a558SYan, Zheng if (!upper->owner) 9713fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 9725d4f98a2SYan Zheng } 9735d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 9745d4f98a2SYan Zheng edge->node[LOWER] = lower; 9753fd0a558SYan, Zheng edge->node[UPPER] = upper; 9765d4f98a2SYan Zheng 9775d4f98a2SYan Zheng if (rb_node) 9785d4f98a2SYan Zheng break; 9795d4f98a2SYan Zheng lower = upper; 9805d4f98a2SYan Zheng upper = NULL; 9815d4f98a2SYan Zheng } 982b3b4aa74SDavid Sterba btrfs_release_path(path2); 9835d4f98a2SYan Zheng next: 9845d4f98a2SYan Zheng if (ptr < end) { 9855d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 9865d4f98a2SYan Zheng if (ptr >= end) { 9875d4f98a2SYan Zheng WARN_ON(ptr > end); 9885d4f98a2SYan Zheng ptr = 0; 9895d4f98a2SYan Zheng end = 0; 9905d4f98a2SYan Zheng } 9915d4f98a2SYan Zheng } 9925d4f98a2SYan Zheng if (ptr >= end) 9935d4f98a2SYan Zheng path1->slots[0]++; 9945d4f98a2SYan Zheng } 995b3b4aa74SDavid Sterba btrfs_release_path(path1); 9965d4f98a2SYan Zheng 9975d4f98a2SYan Zheng cur->checked = 1; 9985d4f98a2SYan Zheng WARN_ON(exist); 9995d4f98a2SYan Zheng 10005d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10015d4f98a2SYan Zheng if (!list_empty(&list)) { 10025d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10035d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10045d4f98a2SYan Zheng cur = edge->node[UPPER]; 10055d4f98a2SYan Zheng goto again; 10065d4f98a2SYan Zheng } 10075d4f98a2SYan Zheng 10085d4f98a2SYan Zheng /* 10095d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10105d4f98a2SYan Zheng * into the cache. 10115d4f98a2SYan Zheng */ 101275bfb9afSJosef Bacik ASSERT(node->checked); 10133fd0a558SYan, Zheng cowonly = node->cowonly; 10143fd0a558SYan, Zheng if (!cowonly) { 10153fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10163fd0a558SYan, Zheng &node->rb_node); 101743c04fb1SJeff Mahoney if (rb_node) 101843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10193fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10203fd0a558SYan, Zheng } 10215d4f98a2SYan Zheng 10225d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10235d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10245d4f98a2SYan Zheng 10255d4f98a2SYan Zheng while (!list_empty(&list)) { 10265d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10275d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10285d4f98a2SYan Zheng upper = edge->node[UPPER]; 10293fd0a558SYan, Zheng if (upper->detached) { 10303fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10313fd0a558SYan, Zheng lower = edge->node[LOWER]; 10323fd0a558SYan, Zheng free_backref_edge(cache, edge); 10333fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10343fd0a558SYan, Zheng list_add(&lower->list, &useless); 10353fd0a558SYan, Zheng continue; 10363fd0a558SYan, Zheng } 10375d4f98a2SYan Zheng 10385d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10395d4f98a2SYan Zheng if (upper->lowest) { 10405d4f98a2SYan Zheng list_del_init(&upper->lower); 10415d4f98a2SYan Zheng upper->lowest = 0; 10425d4f98a2SYan Zheng } 10435d4f98a2SYan Zheng 10445d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10455d4f98a2SYan Zheng continue; 10465d4f98a2SYan Zheng } 10475d4f98a2SYan Zheng 104875bfb9afSJosef Bacik if (!upper->checked) { 104975bfb9afSJosef Bacik /* 105075bfb9afSJosef Bacik * Still want to blow up for developers since this is a 105175bfb9afSJosef Bacik * logic bug. 105275bfb9afSJosef Bacik */ 105375bfb9afSJosef Bacik ASSERT(0); 105475bfb9afSJosef Bacik err = -EINVAL; 105575bfb9afSJosef Bacik goto out; 105675bfb9afSJosef Bacik } 105775bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 105875bfb9afSJosef Bacik ASSERT(0); 105975bfb9afSJosef Bacik err = -EINVAL; 106075bfb9afSJosef Bacik goto out; 106175bfb9afSJosef Bacik } 106275bfb9afSJosef Bacik 10633fd0a558SYan, Zheng if (!cowonly) { 10645d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10655d4f98a2SYan Zheng &upper->rb_node); 106643c04fb1SJeff Mahoney if (rb_node) 106743c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 106843c04fb1SJeff Mahoney upper->bytenr); 10693fd0a558SYan, Zheng } 10705d4f98a2SYan Zheng 10715d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10725d4f98a2SYan Zheng 10735d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 10745d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10755d4f98a2SYan Zheng } 10763fd0a558SYan, Zheng /* 10773fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 10783fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 10793fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 10803fd0a558SYan, Zheng * lookup. 10813fd0a558SYan, Zheng */ 10823fd0a558SYan, Zheng while (!list_empty(&useless)) { 10833fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 10843fd0a558SYan, Zheng list_del_init(&upper->list); 108575bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 10863fd0a558SYan, Zheng if (upper == node) 10873fd0a558SYan, Zheng node = NULL; 10883fd0a558SYan, Zheng if (upper->lowest) { 10893fd0a558SYan, Zheng list_del_init(&upper->lower); 10903fd0a558SYan, Zheng upper->lowest = 0; 10913fd0a558SYan, Zheng } 10923fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 10933fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 10943fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 10953fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 10963fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10973fd0a558SYan, Zheng lower = edge->node[LOWER]; 10983fd0a558SYan, Zheng free_backref_edge(cache, edge); 10993fd0a558SYan, Zheng 11003fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11013fd0a558SYan, Zheng list_add(&lower->list, &useless); 11023fd0a558SYan, Zheng } 11033fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11043fd0a558SYan, Zheng if (upper->level > 0) { 11053fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11063fd0a558SYan, Zheng upper->detached = 1; 11073fd0a558SYan, Zheng } else { 11083fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11093fd0a558SYan, Zheng free_backref_node(cache, upper); 11103fd0a558SYan, Zheng } 11113fd0a558SYan, Zheng } 11125d4f98a2SYan Zheng out: 11135d4f98a2SYan Zheng btrfs_free_path(path1); 11145d4f98a2SYan Zheng btrfs_free_path(path2); 11155d4f98a2SYan Zheng if (err) { 11163fd0a558SYan, Zheng while (!list_empty(&useless)) { 11173fd0a558SYan, Zheng lower = list_entry(useless.next, 111875bfb9afSJosef Bacik struct backref_node, list); 111975bfb9afSJosef Bacik list_del_init(&lower->list); 11203fd0a558SYan, Zheng } 112175bfb9afSJosef Bacik while (!list_empty(&list)) { 112275bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 112375bfb9afSJosef Bacik list[UPPER]); 112475bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11253fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 112675bfb9afSJosef Bacik lower = edge->node[LOWER]; 11275d4f98a2SYan Zheng upper = edge->node[UPPER]; 11283fd0a558SYan, Zheng free_backref_edge(cache, edge); 112975bfb9afSJosef Bacik 113075bfb9afSJosef Bacik /* 113175bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 113275bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 113375bfb9afSJosef Bacik */ 113475bfb9afSJosef Bacik if (list_empty(&lower->upper) && 113575bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 113675bfb9afSJosef Bacik list_add(&lower->list, &useless); 113775bfb9afSJosef Bacik 113875bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 113975bfb9afSJosef Bacik continue; 114075bfb9afSJosef Bacik 114101327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 114275bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 114375bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 114475bfb9afSJosef Bacik if (list_empty(&upper->upper)) 114575bfb9afSJosef Bacik list_add(&upper->list, &useless); 114675bfb9afSJosef Bacik } 114775bfb9afSJosef Bacik 114875bfb9afSJosef Bacik while (!list_empty(&useless)) { 114975bfb9afSJosef Bacik lower = list_entry(useless.next, 115075bfb9afSJosef Bacik struct backref_node, list); 115175bfb9afSJosef Bacik list_del_init(&lower->list); 11520fd8c3daSLiu Bo if (lower == node) 11530fd8c3daSLiu Bo node = NULL; 115475bfb9afSJosef Bacik free_backref_node(cache, lower); 11555d4f98a2SYan Zheng } 11560fd8c3daSLiu Bo 11570fd8c3daSLiu Bo free_backref_node(cache, node); 11585d4f98a2SYan Zheng return ERR_PTR(err); 11595d4f98a2SYan Zheng } 116075bfb9afSJosef Bacik ASSERT(!node || !node->detached); 11615d4f98a2SYan Zheng return node; 11625d4f98a2SYan Zheng } 11635d4f98a2SYan Zheng 11645d4f98a2SYan Zheng /* 11653fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11663fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11673fd0a558SYan, Zheng * corresponds to root of source tree 11683fd0a558SYan, Zheng */ 11693fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 11703fd0a558SYan, Zheng struct reloc_control *rc, 11713fd0a558SYan, Zheng struct btrfs_root *src, 11723fd0a558SYan, Zheng struct btrfs_root *dest) 11733fd0a558SYan, Zheng { 11743fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 11753fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 11763fd0a558SYan, Zheng struct backref_node *node = NULL; 11773fd0a558SYan, Zheng struct backref_node *new_node; 11783fd0a558SYan, Zheng struct backref_edge *edge; 11793fd0a558SYan, Zheng struct backref_edge *new_edge; 11803fd0a558SYan, Zheng struct rb_node *rb_node; 11813fd0a558SYan, Zheng 11823fd0a558SYan, Zheng if (cache->last_trans > 0) 11833fd0a558SYan, Zheng update_backref_cache(trans, cache); 11843fd0a558SYan, Zheng 11853fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 11863fd0a558SYan, Zheng if (rb_node) { 11873fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 11883fd0a558SYan, Zheng if (node->detached) 11893fd0a558SYan, Zheng node = NULL; 11903fd0a558SYan, Zheng else 11913fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 11923fd0a558SYan, Zheng } 11933fd0a558SYan, Zheng 11943fd0a558SYan, Zheng if (!node) { 11953fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 11963fd0a558SYan, Zheng reloc_root->commit_root->start); 11973fd0a558SYan, Zheng if (rb_node) { 11983fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 11993fd0a558SYan, Zheng rb_node); 12003fd0a558SYan, Zheng BUG_ON(node->detached); 12013fd0a558SYan, Zheng } 12023fd0a558SYan, Zheng } 12033fd0a558SYan, Zheng 12043fd0a558SYan, Zheng if (!node) 12053fd0a558SYan, Zheng return 0; 12063fd0a558SYan, Zheng 12073fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12083fd0a558SYan, Zheng if (!new_node) 12093fd0a558SYan, Zheng return -ENOMEM; 12103fd0a558SYan, Zheng 12113fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12123fd0a558SYan, Zheng new_node->level = node->level; 12133fd0a558SYan, Zheng new_node->lowest = node->lowest; 12146848ad64SYan, Zheng new_node->checked = 1; 12153fd0a558SYan, Zheng new_node->root = dest; 12163fd0a558SYan, Zheng 12173fd0a558SYan, Zheng if (!node->lowest) { 12183fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12193fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12203fd0a558SYan, Zheng if (!new_edge) 12213fd0a558SYan, Zheng goto fail; 12223fd0a558SYan, Zheng 12233fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12243fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12253fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12263fd0a558SYan, Zheng &new_node->lower); 12273fd0a558SYan, Zheng } 122876b9e23dSMiao Xie } else { 122976b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12303fd0a558SYan, Zheng } 12313fd0a558SYan, Zheng 12323fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12333fd0a558SYan, Zheng &new_node->rb_node); 123443c04fb1SJeff Mahoney if (rb_node) 123543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12363fd0a558SYan, Zheng 12373fd0a558SYan, Zheng if (!new_node->lowest) { 12383fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12393fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12403fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12413fd0a558SYan, Zheng } 12423fd0a558SYan, Zheng } 12433fd0a558SYan, Zheng return 0; 12443fd0a558SYan, Zheng fail: 12453fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12463fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12473fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12483fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12493fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12503fd0a558SYan, Zheng } 12513fd0a558SYan, Zheng free_backref_node(cache, new_node); 12523fd0a558SYan, Zheng return -ENOMEM; 12533fd0a558SYan, Zheng } 12543fd0a558SYan, Zheng 12553fd0a558SYan, Zheng /* 12565d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12575d4f98a2SYan Zheng */ 1258ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12595d4f98a2SYan Zheng { 12600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12615d4f98a2SYan Zheng struct rb_node *rb_node; 12625d4f98a2SYan Zheng struct mapping_node *node; 12630b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 12645d4f98a2SYan Zheng 12655d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1266ffd7b339SJeff Mahoney if (!node) 1267ffd7b339SJeff Mahoney return -ENOMEM; 12685d4f98a2SYan Zheng 12695d4f98a2SYan Zheng node->bytenr = root->node->start; 12705d4f98a2SYan Zheng node->data = root; 12715d4f98a2SYan Zheng 12725d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12735d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 12745d4f98a2SYan Zheng node->bytenr, &node->rb_node); 12755d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1276ffd7b339SJeff Mahoney if (rb_node) { 12770b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 12785d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 12795d163e0eSJeff Mahoney node->bytenr); 1280ffd7b339SJeff Mahoney } 12815d4f98a2SYan Zheng 12825d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 12835d4f98a2SYan Zheng return 0; 12845d4f98a2SYan Zheng } 12855d4f98a2SYan Zheng 12865d4f98a2SYan Zheng /* 1287c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 12885d4f98a2SYan Zheng * mapping 12895d4f98a2SYan Zheng */ 1290c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 12915d4f98a2SYan Zheng { 12920b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12935d4f98a2SYan Zheng struct rb_node *rb_node; 12945d4f98a2SYan Zheng struct mapping_node *node = NULL; 12950b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 12965d4f98a2SYan Zheng 129765c6e82bSQu Wenruo if (rc && root->node) { 12985d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12995d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1300c974c464SWang Shilong root->node->start); 1301c974c464SWang Shilong if (rb_node) { 1302c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1303c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1304c974c464SWang Shilong } 1305c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1306c974c464SWang Shilong if (!node) 1307c974c464SWang Shilong return; 1308c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1309389305b2SQu Wenruo } 1310c974c464SWang Shilong 13110b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1312c974c464SWang Shilong list_del_init(&root->root_list); 13130b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1314c974c464SWang Shilong kfree(node); 1315c974c464SWang Shilong } 1316c974c464SWang Shilong 1317c974c464SWang Shilong /* 1318c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1319c974c464SWang Shilong * mapping 1320c974c464SWang Shilong */ 1321c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1322c974c464SWang Shilong { 13230b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1324c974c464SWang Shilong struct rb_node *rb_node; 1325c974c464SWang Shilong struct mapping_node *node = NULL; 13260b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1327c974c464SWang Shilong 1328c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1329c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1330c974c464SWang Shilong root->node->start); 13315d4f98a2SYan Zheng if (rb_node) { 13325d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13335d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13345d4f98a2SYan Zheng } 13355d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13365d4f98a2SYan Zheng 13378f71f3e0SLiu Bo if (!node) 13388f71f3e0SLiu Bo return 0; 13395d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13405d4f98a2SYan Zheng 13415d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1342c974c464SWang Shilong node->bytenr = new_bytenr; 13435d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13445d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13455d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 134643c04fb1SJeff Mahoney if (rb_node) 134743c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13485d4f98a2SYan Zheng return 0; 13495d4f98a2SYan Zheng } 13505d4f98a2SYan Zheng 13513fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13523fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13535d4f98a2SYan Zheng { 13540b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13555d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13565d4f98a2SYan Zheng struct extent_buffer *eb; 13575d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13585d4f98a2SYan Zheng struct btrfs_key root_key; 13595d4f98a2SYan Zheng int ret; 13605d4f98a2SYan Zheng 13615d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13625d4f98a2SYan Zheng BUG_ON(!root_item); 13635d4f98a2SYan Zheng 13645d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13655d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13663fd0a558SYan, Zheng root_key.offset = objectid; 13675d4f98a2SYan Zheng 13683fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1369054570a1SFilipe Manana u64 commit_root_gen; 1370054570a1SFilipe Manana 13713fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 13725d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 13735d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 13745d4f98a2SYan Zheng BUG_ON(ret); 1375054570a1SFilipe Manana /* 1376054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1377054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1378054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1379054570a1SFilipe Manana * either inside the critical section of a transaction commit 1380054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1381054570a1SFilipe Manana * it's created before the transaction commit is started. 1382054570a1SFilipe Manana */ 1383054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1384054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 13853fd0a558SYan, Zheng } else { 13863fd0a558SYan, Zheng /* 13873fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 13883fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 13893fd0a558SYan, Zheng * modified after it was created have RELOC flag 13903fd0a558SYan, Zheng * set in their headers. so it's OK to not update 13913fd0a558SYan, Zheng * the 'last_snapshot'. 13923fd0a558SYan, Zheng */ 13933fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 13943fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 13953fd0a558SYan, Zheng BUG_ON(ret); 13963fd0a558SYan, Zheng } 13973fd0a558SYan, Zheng 13985d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 13995d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14005d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14015d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14023fd0a558SYan, Zheng 14033fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14043fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14053fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14063fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14075d4f98a2SYan Zheng root_item->drop_level = 0; 14083fd0a558SYan, Zheng } 14095d4f98a2SYan Zheng 14105d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14115d4f98a2SYan Zheng free_extent_buffer(eb); 14125d4f98a2SYan Zheng 14130b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 14145d4f98a2SYan Zheng &root_key, root_item); 14155d4f98a2SYan Zheng BUG_ON(ret); 14165d4f98a2SYan Zheng kfree(root_item); 14175d4f98a2SYan Zheng 14180b246afaSJeff Mahoney reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key); 14195d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14205d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14213fd0a558SYan, Zheng return reloc_root; 14223fd0a558SYan, Zheng } 14233fd0a558SYan, Zheng 14243fd0a558SYan, Zheng /* 14253fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 14263fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 14273fd0a558SYan, Zheng */ 14283fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 14293fd0a558SYan, Zheng struct btrfs_root *root) 14303fd0a558SYan, Zheng { 14310b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14323fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14330b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 143420dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14353fd0a558SYan, Zheng int clear_rsv = 0; 1436ffd7b339SJeff Mahoney int ret; 14373fd0a558SYan, Zheng 14381fac4a54SQu Wenruo /* 14391fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 14401fac4a54SQu Wenruo * create/update the dead reloc tree 14411fac4a54SQu Wenruo */ 14421fac4a54SQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 14431fac4a54SQu Wenruo return 0; 14441fac4a54SQu Wenruo 14453fd0a558SYan, Zheng if (root->reloc_root) { 14463fd0a558SYan, Zheng reloc_root = root->reloc_root; 14473fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14483fd0a558SYan, Zheng return 0; 14493fd0a558SYan, Zheng } 14503fd0a558SYan, Zheng 14513fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14523fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14533fd0a558SYan, Zheng return 0; 14543fd0a558SYan, Zheng 145520dd2cbfSMiao Xie if (!trans->reloc_reserved) { 145620dd2cbfSMiao Xie rsv = trans->block_rsv; 14573fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14583fd0a558SYan, Zheng clear_rsv = 1; 14593fd0a558SYan, Zheng } 14603fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14613fd0a558SYan, Zheng if (clear_rsv) 146220dd2cbfSMiao Xie trans->block_rsv = rsv; 14635d4f98a2SYan Zheng 1464ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1465ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14665d4f98a2SYan Zheng root->reloc_root = reloc_root; 14675d4f98a2SYan Zheng return 0; 14685d4f98a2SYan Zheng } 14695d4f98a2SYan Zheng 14705d4f98a2SYan Zheng /* 14715d4f98a2SYan Zheng * update root item of reloc tree 14725d4f98a2SYan Zheng */ 14735d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 14745d4f98a2SYan Zheng struct btrfs_root *root) 14755d4f98a2SYan Zheng { 14760b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14775d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14785d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14795d4f98a2SYan Zheng int ret; 14805d4f98a2SYan Zheng 1481d2311e69SQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state) || 1482d2311e69SQu Wenruo !root->reloc_root) 14837585717fSChris Mason goto out; 14845d4f98a2SYan Zheng 14855d4f98a2SYan Zheng reloc_root = root->reloc_root; 14865d4f98a2SYan Zheng root_item = &reloc_root->root_item; 14875d4f98a2SYan Zheng 1488d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 14890b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 14903fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 1491d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 1492c974c464SWang Shilong __del_reloc_root(reloc_root); 14935d4f98a2SYan Zheng } 14945d4f98a2SYan Zheng 14955d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 14965d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 14975d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 14985d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 14995d4f98a2SYan Zheng } 15005d4f98a2SYan Zheng 15010b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 15025d4f98a2SYan Zheng &reloc_root->root_key, root_item); 15035d4f98a2SYan Zheng BUG_ON(ret); 15047585717fSChris Mason 15057585717fSChris Mason out: 15065d4f98a2SYan Zheng return 0; 15075d4f98a2SYan Zheng } 15085d4f98a2SYan Zheng 15095d4f98a2SYan Zheng /* 15105d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15115d4f98a2SYan Zheng * in a subvolume 15125d4f98a2SYan Zheng */ 15135d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15145d4f98a2SYan Zheng { 15155d4f98a2SYan Zheng struct rb_node *node; 15165d4f98a2SYan Zheng struct rb_node *prev; 15175d4f98a2SYan Zheng struct btrfs_inode *entry; 15185d4f98a2SYan Zheng struct inode *inode; 15195d4f98a2SYan Zheng 15205d4f98a2SYan Zheng spin_lock(&root->inode_lock); 15215d4f98a2SYan Zheng again: 15225d4f98a2SYan Zheng node = root->inode_tree.rb_node; 15235d4f98a2SYan Zheng prev = NULL; 15245d4f98a2SYan Zheng while (node) { 15255d4f98a2SYan Zheng prev = node; 15265d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15275d4f98a2SYan Zheng 15284a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 15295d4f98a2SYan Zheng node = node->rb_left; 15304a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 15315d4f98a2SYan Zheng node = node->rb_right; 15325d4f98a2SYan Zheng else 15335d4f98a2SYan Zheng break; 15345d4f98a2SYan Zheng } 15355d4f98a2SYan Zheng if (!node) { 15365d4f98a2SYan Zheng while (prev) { 15375d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 15384a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 15395d4f98a2SYan Zheng node = prev; 15405d4f98a2SYan Zheng break; 15415d4f98a2SYan Zheng } 15425d4f98a2SYan Zheng prev = rb_next(prev); 15435d4f98a2SYan Zheng } 15445d4f98a2SYan Zheng } 15455d4f98a2SYan Zheng while (node) { 15465d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15475d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15485d4f98a2SYan Zheng if (inode) { 15495d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15505d4f98a2SYan Zheng return inode; 15515d4f98a2SYan Zheng } 15525d4f98a2SYan Zheng 15534a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 15545d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15555d4f98a2SYan Zheng goto again; 15565d4f98a2SYan Zheng 15575d4f98a2SYan Zheng node = rb_next(node); 15585d4f98a2SYan Zheng } 15595d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15605d4f98a2SYan Zheng return NULL; 15615d4f98a2SYan Zheng } 15625d4f98a2SYan Zheng 156332da5386SDavid Sterba static int in_block_group(u64 bytenr, struct btrfs_block_group *block_group) 15645d4f98a2SYan Zheng { 1565b3470b5dSDavid Sterba if (bytenr >= block_group->start && 1566b3470b5dSDavid Sterba bytenr < block_group->start + block_group->length) 15675d4f98a2SYan Zheng return 1; 15685d4f98a2SYan Zheng return 0; 15695d4f98a2SYan Zheng } 15705d4f98a2SYan Zheng 15715d4f98a2SYan Zheng /* 15725d4f98a2SYan Zheng * get new location of data 15735d4f98a2SYan Zheng */ 15745d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 15755d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 15765d4f98a2SYan Zheng { 15775d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 15785d4f98a2SYan Zheng struct btrfs_path *path; 15795d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15805d4f98a2SYan Zheng struct extent_buffer *leaf; 15815d4f98a2SYan Zheng int ret; 15825d4f98a2SYan Zheng 15835d4f98a2SYan Zheng path = btrfs_alloc_path(); 15845d4f98a2SYan Zheng if (!path) 15855d4f98a2SYan Zheng return -ENOMEM; 15865d4f98a2SYan Zheng 15875d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1588f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1589f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 15905d4f98a2SYan Zheng if (ret < 0) 15915d4f98a2SYan Zheng goto out; 15925d4f98a2SYan Zheng if (ret > 0) { 15935d4f98a2SYan Zheng ret = -ENOENT; 15945d4f98a2SYan Zheng goto out; 15955d4f98a2SYan Zheng } 15965d4f98a2SYan Zheng 15975d4f98a2SYan Zheng leaf = path->nodes[0]; 15985d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 15995d4f98a2SYan Zheng struct btrfs_file_extent_item); 16005d4f98a2SYan Zheng 16015d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 16025d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 16035d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 16045d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 16055d4f98a2SYan Zheng 16065d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 160783d4cfd4SJosef Bacik ret = -EINVAL; 16085d4f98a2SYan Zheng goto out; 16095d4f98a2SYan Zheng } 16105d4f98a2SYan Zheng 16115d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16125d4f98a2SYan Zheng ret = 0; 16135d4f98a2SYan Zheng out: 16145d4f98a2SYan Zheng btrfs_free_path(path); 16155d4f98a2SYan Zheng return ret; 16165d4f98a2SYan Zheng } 16175d4f98a2SYan Zheng 16185d4f98a2SYan Zheng /* 16195d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16205d4f98a2SYan Zheng * the new locations. 16215d4f98a2SYan Zheng */ 16223fd0a558SYan, Zheng static noinline_for_stack 16233fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 16245d4f98a2SYan Zheng struct reloc_control *rc, 16255d4f98a2SYan Zheng struct btrfs_root *root, 16263fd0a558SYan, Zheng struct extent_buffer *leaf) 16275d4f98a2SYan Zheng { 16280b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 16295d4f98a2SYan Zheng struct btrfs_key key; 16305d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16315d4f98a2SYan Zheng struct inode *inode = NULL; 16325d4f98a2SYan Zheng u64 parent; 16335d4f98a2SYan Zheng u64 bytenr; 16343fd0a558SYan, Zheng u64 new_bytenr = 0; 16355d4f98a2SYan Zheng u64 num_bytes; 16365d4f98a2SYan Zheng u64 end; 16375d4f98a2SYan Zheng u32 nritems; 16385d4f98a2SYan Zheng u32 i; 163983d4cfd4SJosef Bacik int ret = 0; 16405d4f98a2SYan Zheng int first = 1; 16415d4f98a2SYan Zheng int dirty = 0; 16425d4f98a2SYan Zheng 16435d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16445d4f98a2SYan Zheng return 0; 16455d4f98a2SYan Zheng 16465d4f98a2SYan Zheng /* reloc trees always use full backref */ 16475d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16485d4f98a2SYan Zheng parent = leaf->start; 16495d4f98a2SYan Zheng else 16505d4f98a2SYan Zheng parent = 0; 16515d4f98a2SYan Zheng 16525d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16535d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 165482fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 165582fa113fSQu Wenruo 16565d4f98a2SYan Zheng cond_resched(); 16575d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16585d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16595d4f98a2SYan Zheng continue; 16605d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16615d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16625d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16635d4f98a2SYan Zheng continue; 16645d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16655d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16665d4f98a2SYan Zheng if (bytenr == 0) 16675d4f98a2SYan Zheng continue; 16685d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16695d4f98a2SYan Zheng continue; 16705d4f98a2SYan Zheng 16715d4f98a2SYan Zheng /* 16725d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16735d4f98a2SYan Zheng * to complete and drop the extent cache 16745d4f98a2SYan Zheng */ 16755d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16765d4f98a2SYan Zheng if (first) { 16775d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16785d4f98a2SYan Zheng first = 0; 16794a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 16803fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16815d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16825d4f98a2SYan Zheng } 16834a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 16845d4f98a2SYan Zheng end = key.offset + 16855d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 16865d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 16870b246afaSJeff Mahoney fs_info->sectorsize)); 16880b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 16895d4f98a2SYan Zheng end--; 16905d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1691d0082371SJeff Mahoney key.offset, end); 16925d4f98a2SYan Zheng if (!ret) 16935d4f98a2SYan Zheng continue; 16945d4f98a2SYan Zheng 1695dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1696dcdbc059SNikolay Borisov key.offset, end, 1); 16975d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1698d0082371SJeff Mahoney key.offset, end); 16995d4f98a2SYan Zheng } 17005d4f98a2SYan Zheng } 17015d4f98a2SYan Zheng 17025d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 17035d4f98a2SYan Zheng bytenr, num_bytes); 170483d4cfd4SJosef Bacik if (ret) { 170583d4cfd4SJosef Bacik /* 170683d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 170783d4cfd4SJosef Bacik * in the file extent yet. 170883d4cfd4SJosef Bacik */ 170983d4cfd4SJosef Bacik break; 17103fd0a558SYan, Zheng } 17115d4f98a2SYan Zheng 17125d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17135d4f98a2SYan Zheng dirty = 1; 17145d4f98a2SYan Zheng 17155d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 171682fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 171782fa113fSQu Wenruo num_bytes, parent); 171882fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 171982fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1720b06c4bf5SFilipe Manana key.objectid, key.offset); 172182fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 172283d4cfd4SJosef Bacik if (ret) { 172366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 172483d4cfd4SJosef Bacik break; 172583d4cfd4SJosef Bacik } 17265d4f98a2SYan Zheng 1727ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1728ffd4bb2aSQu Wenruo num_bytes, parent); 1729ffd4bb2aSQu Wenruo ref.real_root = root->root_key.objectid; 1730ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1731b06c4bf5SFilipe Manana key.objectid, key.offset); 1732ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 173383d4cfd4SJosef Bacik if (ret) { 173466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 173583d4cfd4SJosef Bacik break; 173683d4cfd4SJosef Bacik } 17375d4f98a2SYan Zheng } 17385d4f98a2SYan Zheng if (dirty) 17395d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17403fd0a558SYan, Zheng if (inode) 17413fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 174283d4cfd4SJosef Bacik return ret; 17435d4f98a2SYan Zheng } 17445d4f98a2SYan Zheng 17455d4f98a2SYan Zheng static noinline_for_stack 17465d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17475d4f98a2SYan Zheng struct btrfs_path *path, int level) 17485d4f98a2SYan Zheng { 17495d4f98a2SYan Zheng struct btrfs_disk_key key1; 17505d4f98a2SYan Zheng struct btrfs_disk_key key2; 17515d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17525d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17535d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17545d4f98a2SYan Zheng } 17555d4f98a2SYan Zheng 17565d4f98a2SYan Zheng /* 17575d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17585d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17595d4f98a2SYan Zheng * reloc tree was create can be replaced. 17605d4f98a2SYan Zheng * 17615d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17625d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17635d4f98a2SYan Zheng * errors, a negative error number is returned. 17645d4f98a2SYan Zheng */ 17653fd0a558SYan, Zheng static noinline_for_stack 17663d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 17675d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17685d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17695d4f98a2SYan Zheng int lowest_level, int max_level) 17705d4f98a2SYan Zheng { 17710b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 17725d4f98a2SYan Zheng struct extent_buffer *eb; 17735d4f98a2SYan Zheng struct extent_buffer *parent; 177482fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 17755d4f98a2SYan Zheng struct btrfs_key key; 17765d4f98a2SYan Zheng u64 old_bytenr; 17775d4f98a2SYan Zheng u64 new_bytenr; 17785d4f98a2SYan Zheng u64 old_ptr_gen; 17795d4f98a2SYan Zheng u64 new_ptr_gen; 17805d4f98a2SYan Zheng u64 last_snapshot; 17815d4f98a2SYan Zheng u32 blocksize; 17823fd0a558SYan, Zheng int cow = 0; 17835d4f98a2SYan Zheng int level; 17845d4f98a2SYan Zheng int ret; 17855d4f98a2SYan Zheng int slot; 17865d4f98a2SYan Zheng 17875d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17885d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17895d4f98a2SYan Zheng 17905d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 17913fd0a558SYan, Zheng again: 17925d4f98a2SYan Zheng slot = path->slots[lowest_level]; 17935d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 17945d4f98a2SYan Zheng 17955d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 17968bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 17975d4f98a2SYan Zheng level = btrfs_header_level(eb); 17985d4f98a2SYan Zheng 17995d4f98a2SYan Zheng if (level < lowest_level) { 18005d4f98a2SYan Zheng btrfs_tree_unlock(eb); 18015d4f98a2SYan Zheng free_extent_buffer(eb); 18025d4f98a2SYan Zheng return 0; 18035d4f98a2SYan Zheng } 18045d4f98a2SYan Zheng 18053fd0a558SYan, Zheng if (cow) { 18065d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18075d4f98a2SYan Zheng BUG_ON(ret); 18083fd0a558SYan, Zheng } 18098bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18105d4f98a2SYan Zheng 18115d4f98a2SYan Zheng if (next_key) { 18125d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18135d4f98a2SYan Zheng next_key->type = (u8)-1; 18145d4f98a2SYan Zheng next_key->offset = (u64)-1; 18155d4f98a2SYan Zheng } 18165d4f98a2SYan Zheng 18175d4f98a2SYan Zheng parent = eb; 18185d4f98a2SYan Zheng while (1) { 1819581c1760SQu Wenruo struct btrfs_key first_key; 1820581c1760SQu Wenruo 18215d4f98a2SYan Zheng level = btrfs_header_level(parent); 18225d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18235d4f98a2SYan Zheng 18245d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 1825cbca7d59SFilipe Manana if (ret < 0) 1826cbca7d59SFilipe Manana break; 18275d4f98a2SYan Zheng if (ret && slot > 0) 18285d4f98a2SYan Zheng slot--; 18295d4f98a2SYan Zheng 18305d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18315d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18325d4f98a2SYan Zheng 18335d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 18340b246afaSJeff Mahoney blocksize = fs_info->nodesize; 18355d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 183617515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 18375d4f98a2SYan Zheng 18385d4f98a2SYan Zheng if (level <= max_level) { 18395d4f98a2SYan Zheng eb = path->nodes[level]; 18405d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18415d4f98a2SYan Zheng path->slots[level]); 18425d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18435d4f98a2SYan Zheng path->slots[level]); 18445d4f98a2SYan Zheng } else { 18455d4f98a2SYan Zheng new_bytenr = 0; 18465d4f98a2SYan Zheng new_ptr_gen = 0; 18475d4f98a2SYan Zheng } 18485d4f98a2SYan Zheng 1849fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18505d4f98a2SYan Zheng ret = level; 18515d4f98a2SYan Zheng break; 18525d4f98a2SYan Zheng } 18535d4f98a2SYan Zheng 18545d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18555d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18563fd0a558SYan, Zheng if (level <= lowest_level) { 18575d4f98a2SYan Zheng ret = 0; 18585d4f98a2SYan Zheng break; 18595d4f98a2SYan Zheng } 18605d4f98a2SYan Zheng 1861581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 1862581c1760SQu Wenruo level - 1, &first_key); 186364c043deSLiu Bo if (IS_ERR(eb)) { 186464c043deSLiu Bo ret = PTR_ERR(eb); 1865264813acSLiu Bo break; 186664c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 186764c043deSLiu Bo ret = -EIO; 1868416bc658SJosef Bacik free_extent_buffer(eb); 1869379cde74SStefan Behrens break; 1870416bc658SJosef Bacik } 18715d4f98a2SYan Zheng btrfs_tree_lock(eb); 18723fd0a558SYan, Zheng if (cow) { 18735d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18745d4f98a2SYan Zheng slot, &eb); 18755d4f98a2SYan Zheng BUG_ON(ret); 18765d4f98a2SYan Zheng } 18778bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18785d4f98a2SYan Zheng 18795d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18805d4f98a2SYan Zheng free_extent_buffer(parent); 18815d4f98a2SYan Zheng 18825d4f98a2SYan Zheng parent = eb; 18835d4f98a2SYan Zheng continue; 18845d4f98a2SYan Zheng } 18855d4f98a2SYan Zheng 18863fd0a558SYan, Zheng if (!cow) { 18873fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18883fd0a558SYan, Zheng free_extent_buffer(parent); 18893fd0a558SYan, Zheng cow = 1; 18903fd0a558SYan, Zheng goto again; 18913fd0a558SYan, Zheng } 18923fd0a558SYan, Zheng 18935d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18945d4f98a2SYan Zheng path->slots[level]); 1895b3b4aa74SDavid Sterba btrfs_release_path(path); 18965d4f98a2SYan Zheng 18975d4f98a2SYan Zheng path->lowest_level = level; 18985d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 18995d4f98a2SYan Zheng path->lowest_level = 0; 19005d4f98a2SYan Zheng BUG_ON(ret); 19015d4f98a2SYan Zheng 19025d4f98a2SYan Zheng /* 1903824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1904824d8dffSQu Wenruo * 1905824d8dffSQu Wenruo * We must trace both trees. 1906824d8dffSQu Wenruo * 1) Tree reloc subtree 1907824d8dffSQu Wenruo * If not traced, we will leak data numbers 1908824d8dffSQu Wenruo * 2) Fs subtree 1909824d8dffSQu Wenruo * If not traced, we will double count old data 1910f616f5cdSQu Wenruo * 1911f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1912f616f5cdSQu Wenruo * the swapped tree blocks. 1913f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1914f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1915824d8dffSQu Wenruo */ 1916370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1917370a11b8SQu Wenruo rc->block_group, parent, slot, 1918370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1919370a11b8SQu Wenruo last_snapshot); 1920370a11b8SQu Wenruo if (ret < 0) 1921370a11b8SQu Wenruo break; 1922824d8dffSQu Wenruo /* 19235d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 19245d4f98a2SYan Zheng */ 19255d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 19265d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19275d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19285d4f98a2SYan Zheng 19295d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19305d4f98a2SYan Zheng path->slots[level], old_bytenr); 19315d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19325d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19335d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19345d4f98a2SYan Zheng 193582fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 193682fa113fSQu Wenruo blocksize, path->nodes[level]->start); 193782fa113fSQu Wenruo ref.skip_qgroup = true; 193882fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 193982fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 19405d4f98a2SYan Zheng BUG_ON(ret); 194182fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 194282fa113fSQu Wenruo blocksize, 0); 194382fa113fSQu Wenruo ref.skip_qgroup = true; 194482fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 194582fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 19465d4f98a2SYan Zheng BUG_ON(ret); 19475d4f98a2SYan Zheng 1948ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1949ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1950ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 1951ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1952ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 19535d4f98a2SYan Zheng BUG_ON(ret); 19545d4f98a2SYan Zheng 1955ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1956ffd4bb2aSQu Wenruo blocksize, 0); 1957ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 1958ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1959ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 19605d4f98a2SYan Zheng BUG_ON(ret); 19615d4f98a2SYan Zheng 19625d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19635d4f98a2SYan Zheng 19645d4f98a2SYan Zheng ret = level; 19655d4f98a2SYan Zheng break; 19665d4f98a2SYan Zheng } 19675d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19685d4f98a2SYan Zheng free_extent_buffer(parent); 19695d4f98a2SYan Zheng return ret; 19705d4f98a2SYan Zheng } 19715d4f98a2SYan Zheng 19725d4f98a2SYan Zheng /* 19735d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19745d4f98a2SYan Zheng */ 19755d4f98a2SYan Zheng static noinline_for_stack 19765d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19775d4f98a2SYan Zheng int *level) 19785d4f98a2SYan Zheng { 19795d4f98a2SYan Zheng struct extent_buffer *eb; 19805d4f98a2SYan Zheng int i; 19815d4f98a2SYan Zheng u64 last_snapshot; 19825d4f98a2SYan Zheng u32 nritems; 19835d4f98a2SYan Zheng 19845d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19855d4f98a2SYan Zheng 19865d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19875d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19885d4f98a2SYan Zheng path->nodes[i] = NULL; 19895d4f98a2SYan Zheng } 19905d4f98a2SYan Zheng 19915d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 19925d4f98a2SYan Zheng eb = path->nodes[i]; 19935d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19945d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 19955d4f98a2SYan Zheng path->slots[i]++; 19965d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 19975d4f98a2SYan Zheng last_snapshot) 19985d4f98a2SYan Zheng continue; 19995d4f98a2SYan Zheng 20005d4f98a2SYan Zheng *level = i; 20015d4f98a2SYan Zheng return 0; 20025d4f98a2SYan Zheng } 20035d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20045d4f98a2SYan Zheng path->nodes[i] = NULL; 20055d4f98a2SYan Zheng } 20065d4f98a2SYan Zheng return 1; 20075d4f98a2SYan Zheng } 20085d4f98a2SYan Zheng 20095d4f98a2SYan Zheng /* 20105d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 20115d4f98a2SYan Zheng */ 20125d4f98a2SYan Zheng static noinline_for_stack 20135d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20145d4f98a2SYan Zheng int *level) 20155d4f98a2SYan Zheng { 20162ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20175d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 20185d4f98a2SYan Zheng int i; 20195d4f98a2SYan Zheng u64 bytenr; 20205d4f98a2SYan Zheng u64 ptr_gen = 0; 20215d4f98a2SYan Zheng u64 last_snapshot; 20225d4f98a2SYan Zheng u32 nritems; 20235d4f98a2SYan Zheng 20245d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20255d4f98a2SYan Zheng 20265d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2027581c1760SQu Wenruo struct btrfs_key first_key; 2028581c1760SQu Wenruo 20295d4f98a2SYan Zheng eb = path->nodes[i]; 20305d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20315d4f98a2SYan Zheng while (path->slots[i] < nritems) { 20325d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 20335d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 20345d4f98a2SYan Zheng break; 20355d4f98a2SYan Zheng path->slots[i]++; 20365d4f98a2SYan Zheng } 20375d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20385d4f98a2SYan Zheng if (i == *level) 20395d4f98a2SYan Zheng break; 20405d4f98a2SYan Zheng *level = i + 1; 20415d4f98a2SYan Zheng return 0; 20425d4f98a2SYan Zheng } 20435d4f98a2SYan Zheng if (i == 1) { 20445d4f98a2SYan Zheng *level = i; 20455d4f98a2SYan Zheng return 0; 20465d4f98a2SYan Zheng } 20475d4f98a2SYan Zheng 20485d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2049581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2050581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2051581c1760SQu Wenruo &first_key); 205264c043deSLiu Bo if (IS_ERR(eb)) { 205364c043deSLiu Bo return PTR_ERR(eb); 205464c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2055416bc658SJosef Bacik free_extent_buffer(eb); 2056416bc658SJosef Bacik return -EIO; 2057416bc658SJosef Bacik } 20585d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20595d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20605d4f98a2SYan Zheng path->slots[i - 1] = 0; 20615d4f98a2SYan Zheng } 20625d4f98a2SYan Zheng return 1; 20635d4f98a2SYan Zheng } 20645d4f98a2SYan Zheng 20655d4f98a2SYan Zheng /* 20665d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 20675d4f98a2SYan Zheng * [min_key, max_key) 20685d4f98a2SYan Zheng */ 20695d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 20705d4f98a2SYan Zheng struct btrfs_key *min_key, 20715d4f98a2SYan Zheng struct btrfs_key *max_key) 20725d4f98a2SYan Zheng { 20730b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20745d4f98a2SYan Zheng struct inode *inode = NULL; 20755d4f98a2SYan Zheng u64 objectid; 20765d4f98a2SYan Zheng u64 start, end; 207733345d01SLi Zefan u64 ino; 20785d4f98a2SYan Zheng 20795d4f98a2SYan Zheng objectid = min_key->objectid; 20805d4f98a2SYan Zheng while (1) { 20815d4f98a2SYan Zheng cond_resched(); 20825d4f98a2SYan Zheng iput(inode); 20835d4f98a2SYan Zheng 20845d4f98a2SYan Zheng if (objectid > max_key->objectid) 20855d4f98a2SYan Zheng break; 20865d4f98a2SYan Zheng 20875d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20885d4f98a2SYan Zheng if (!inode) 20895d4f98a2SYan Zheng break; 20904a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 20915d4f98a2SYan Zheng 209233345d01SLi Zefan if (ino > max_key->objectid) { 20935d4f98a2SYan Zheng iput(inode); 20945d4f98a2SYan Zheng break; 20955d4f98a2SYan Zheng } 20965d4f98a2SYan Zheng 209733345d01SLi Zefan objectid = ino + 1; 20985d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 20995d4f98a2SYan Zheng continue; 21005d4f98a2SYan Zheng 210133345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 21025d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 21035d4f98a2SYan Zheng continue; 21045d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 21055d4f98a2SYan Zheng start = 0; 21065d4f98a2SYan Zheng else { 21075d4f98a2SYan Zheng start = min_key->offset; 21080b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 21095d4f98a2SYan Zheng } 21105d4f98a2SYan Zheng } else { 21115d4f98a2SYan Zheng start = 0; 21125d4f98a2SYan Zheng } 21135d4f98a2SYan Zheng 211433345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 21155d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 21165d4f98a2SYan Zheng continue; 21175d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 21185d4f98a2SYan Zheng end = (u64)-1; 21195d4f98a2SYan Zheng } else { 21205d4f98a2SYan Zheng if (max_key->offset == 0) 21215d4f98a2SYan Zheng continue; 21225d4f98a2SYan Zheng end = max_key->offset; 21230b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 21245d4f98a2SYan Zheng end--; 21255d4f98a2SYan Zheng } 21265d4f98a2SYan Zheng } else { 21275d4f98a2SYan Zheng end = (u64)-1; 21285d4f98a2SYan Zheng } 21295d4f98a2SYan Zheng 21305d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2131d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2132dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2133d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 21345d4f98a2SYan Zheng } 21355d4f98a2SYan Zheng return 0; 21365d4f98a2SYan Zheng } 21375d4f98a2SYan Zheng 21385d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 21395d4f98a2SYan Zheng struct btrfs_key *key) 21405d4f98a2SYan Zheng 21415d4f98a2SYan Zheng { 21425d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21435d4f98a2SYan Zheng if (!path->nodes[level]) 21445d4f98a2SYan Zheng break; 21455d4f98a2SYan Zheng if (path->slots[level] + 1 < 21465d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21475d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21485d4f98a2SYan Zheng path->slots[level] + 1); 21495d4f98a2SYan Zheng return 0; 21505d4f98a2SYan Zheng } 21515d4f98a2SYan Zheng level++; 21525d4f98a2SYan Zheng } 21535d4f98a2SYan Zheng return 1; 21545d4f98a2SYan Zheng } 21555d4f98a2SYan Zheng 21565d4f98a2SYan Zheng /* 2157d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 2158d2311e69SQu Wenruo */ 2159d2311e69SQu Wenruo static void insert_dirty_subvol(struct btrfs_trans_handle *trans, 2160d2311e69SQu Wenruo struct reloc_control *rc, 2161d2311e69SQu Wenruo struct btrfs_root *root) 2162d2311e69SQu Wenruo { 2163d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2164d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 2165d2311e69SQu Wenruo 2166d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 2167d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 2168d2311e69SQu Wenruo ASSERT(reloc_root); 2169d2311e69SQu Wenruo 2170d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 2171d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 2172d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 2173d2311e69SQu Wenruo reloc_root_item->drop_level = 0; 2174d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 2175d2311e69SQu Wenruo btrfs_update_reloc_root(trans, root); 2176d2311e69SQu Wenruo 2177d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 2178d2311e69SQu Wenruo btrfs_grab_fs_root(root); 2179d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 2180d2311e69SQu Wenruo } 2181d2311e69SQu Wenruo } 2182d2311e69SQu Wenruo 2183d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 2184d2311e69SQu Wenruo { 2185d2311e69SQu Wenruo struct btrfs_root *root; 2186d2311e69SQu Wenruo struct btrfs_root *next; 2187d2311e69SQu Wenruo int ret = 0; 218830d40577SQu Wenruo int ret2; 2189d2311e69SQu Wenruo 2190d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 2191d2311e69SQu Wenruo reloc_dirty_list) { 219230d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 219330d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 2194d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2195d2311e69SQu Wenruo 2196d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 2197d2311e69SQu Wenruo root->reloc_root = NULL; 2198d2311e69SQu Wenruo if (reloc_root) { 2199d2311e69SQu Wenruo 2200d2311e69SQu Wenruo ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1); 2201d2311e69SQu Wenruo if (ret2 < 0 && !ret) 2202d2311e69SQu Wenruo ret = ret2; 2203d2311e69SQu Wenruo } 22041fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 2205d2311e69SQu Wenruo btrfs_put_fs_root(root); 220630d40577SQu Wenruo } else { 220730d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 220830d40577SQu Wenruo ret2 = btrfs_drop_snapshot(root, NULL, 0, 1); 220930d40577SQu Wenruo if (ret2 < 0 && !ret) 221030d40577SQu Wenruo ret = ret2; 221130d40577SQu Wenruo } 2212d2311e69SQu Wenruo } 2213d2311e69SQu Wenruo return ret; 2214d2311e69SQu Wenruo } 2215d2311e69SQu Wenruo 2216d2311e69SQu Wenruo /* 22175d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 22185d4f98a2SYan Zheng * fs tree. 22195d4f98a2SYan Zheng */ 22205d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 22215d4f98a2SYan Zheng struct btrfs_root *root) 22225d4f98a2SYan Zheng { 22230b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 22245d4f98a2SYan Zheng struct btrfs_key key; 22255d4f98a2SYan Zheng struct btrfs_key next_key; 22269e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 22275d4f98a2SYan Zheng struct btrfs_root *reloc_root; 22285d4f98a2SYan Zheng struct btrfs_root_item *root_item; 22295d4f98a2SYan Zheng struct btrfs_path *path; 22303fd0a558SYan, Zheng struct extent_buffer *leaf; 22315d4f98a2SYan Zheng int level; 22325d4f98a2SYan Zheng int max_level; 22335d4f98a2SYan Zheng int replaced = 0; 22345d4f98a2SYan Zheng int ret; 22355d4f98a2SYan Zheng int err = 0; 22363fd0a558SYan, Zheng u32 min_reserved; 22375d4f98a2SYan Zheng 22385d4f98a2SYan Zheng path = btrfs_alloc_path(); 22395d4f98a2SYan Zheng if (!path) 22405d4f98a2SYan Zheng return -ENOMEM; 2241e4058b54SDavid Sterba path->reada = READA_FORWARD; 22425d4f98a2SYan Zheng 22435d4f98a2SYan Zheng reloc_root = root->reloc_root; 22445d4f98a2SYan Zheng root_item = &reloc_root->root_item; 22455d4f98a2SYan Zheng 22465d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 22475d4f98a2SYan Zheng level = btrfs_root_level(root_item); 224867439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 22495d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 22505d4f98a2SYan Zheng path->slots[level] = 0; 22515d4f98a2SYan Zheng } else { 22525d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 22535d4f98a2SYan Zheng 22545d4f98a2SYan Zheng level = root_item->drop_level; 22555d4f98a2SYan Zheng BUG_ON(level == 0); 22565d4f98a2SYan Zheng path->lowest_level = level; 22575d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 225833c66f43SYan Zheng path->lowest_level = 0; 22595d4f98a2SYan Zheng if (ret < 0) { 22605d4f98a2SYan Zheng btrfs_free_path(path); 22615d4f98a2SYan Zheng return ret; 22625d4f98a2SYan Zheng } 22635d4f98a2SYan Zheng 22645d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 22655d4f98a2SYan Zheng path->slots[level]); 22665d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 22675d4f98a2SYan Zheng 22685d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 22695d4f98a2SYan Zheng } 22705d4f98a2SYan Zheng 22710b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22725d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 22735d4f98a2SYan Zheng 22745d4f98a2SYan Zheng while (1) { 227508e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 227608e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22773fd0a558SYan, Zheng if (ret) { 22789e6a0c52SJosef Bacik err = ret; 22799e6a0c52SJosef Bacik goto out; 22803fd0a558SYan, Zheng } 22819e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 22829e6a0c52SJosef Bacik if (IS_ERR(trans)) { 22839e6a0c52SJosef Bacik err = PTR_ERR(trans); 22849e6a0c52SJosef Bacik trans = NULL; 22859e6a0c52SJosef Bacik goto out; 22869e6a0c52SJosef Bacik } 22879e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 22883fd0a558SYan, Zheng 22893fd0a558SYan, Zheng replaced = 0; 22905d4f98a2SYan Zheng max_level = level; 22915d4f98a2SYan Zheng 22925d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 22935d4f98a2SYan Zheng if (ret < 0) { 22945d4f98a2SYan Zheng err = ret; 22955d4f98a2SYan Zheng goto out; 22965d4f98a2SYan Zheng } 22975d4f98a2SYan Zheng if (ret > 0) 22985d4f98a2SYan Zheng break; 22995d4f98a2SYan Zheng 23005d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 23015d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 23025d4f98a2SYan Zheng ret = 0; 23035d4f98a2SYan Zheng } else { 23043d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 23053fd0a558SYan, Zheng &next_key, level, max_level); 23065d4f98a2SYan Zheng } 23075d4f98a2SYan Zheng if (ret < 0) { 23085d4f98a2SYan Zheng err = ret; 23095d4f98a2SYan Zheng goto out; 23105d4f98a2SYan Zheng } 23115d4f98a2SYan Zheng 23125d4f98a2SYan Zheng if (ret > 0) { 23135d4f98a2SYan Zheng level = ret; 23145d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 23155d4f98a2SYan Zheng path->slots[level]); 23165d4f98a2SYan Zheng replaced = 1; 23175d4f98a2SYan Zheng } 23185d4f98a2SYan Zheng 23195d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 23205d4f98a2SYan Zheng if (ret > 0) 23215d4f98a2SYan Zheng break; 23225d4f98a2SYan Zheng 23235d4f98a2SYan Zheng BUG_ON(level == 0); 23245d4f98a2SYan Zheng /* 23255d4f98a2SYan Zheng * save the merging progress in the drop_progress. 23265d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 23275d4f98a2SYan Zheng */ 23285d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 23295d4f98a2SYan Zheng path->slots[level]); 23305d4f98a2SYan Zheng root_item->drop_level = level; 23315d4f98a2SYan Zheng 23323a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23339e6a0c52SJosef Bacik trans = NULL; 23345d4f98a2SYan Zheng 23352ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23365d4f98a2SYan Zheng 23375d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23385d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23395d4f98a2SYan Zheng } 23405d4f98a2SYan Zheng 23415d4f98a2SYan Zheng /* 23425d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 23435d4f98a2SYan Zheng * relocated and the block is tree root. 23445d4f98a2SYan Zheng */ 23455d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 23465d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 23475d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 23485d4f98a2SYan Zheng free_extent_buffer(leaf); 23495d4f98a2SYan Zheng if (ret < 0) 23505d4f98a2SYan Zheng err = ret; 23515d4f98a2SYan Zheng out: 23525d4f98a2SYan Zheng btrfs_free_path(path); 23535d4f98a2SYan Zheng 2354d2311e69SQu Wenruo if (err == 0) 2355d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 23565d4f98a2SYan Zheng 23579e6a0c52SJosef Bacik if (trans) 23583a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23595d4f98a2SYan Zheng 23602ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23615d4f98a2SYan Zheng 23625d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23635d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23645d4f98a2SYan Zheng 23655d4f98a2SYan Zheng return err; 23665d4f98a2SYan Zheng } 23675d4f98a2SYan Zheng 23683fd0a558SYan, Zheng static noinline_for_stack 23693fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 23705d4f98a2SYan Zheng { 23713fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 23720b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23733fd0a558SYan, Zheng struct btrfs_root *reloc_root; 23745d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 23753fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23763fd0a558SYan, Zheng u64 num_bytes = 0; 23773fd0a558SYan, Zheng int ret; 23783fd0a558SYan, Zheng 23790b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23800b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23813fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 23820b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 23837585717fSChris Mason 23843fd0a558SYan, Zheng again: 23853fd0a558SYan, Zheng if (!err) { 23863fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 238708e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 238808e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23893fd0a558SYan, Zheng if (ret) 23903fd0a558SYan, Zheng err = ret; 23913fd0a558SYan, Zheng } 23923fd0a558SYan, Zheng 23937a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 23943612b495STsutomu Itoh if (IS_ERR(trans)) { 23953612b495STsutomu Itoh if (!err) 23962ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23972ff7e61eSJeff Mahoney num_bytes); 23983612b495STsutomu Itoh return PTR_ERR(trans); 23993612b495STsutomu Itoh } 24003fd0a558SYan, Zheng 24013fd0a558SYan, Zheng if (!err) { 24023fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 24033a45bb20SJeff Mahoney btrfs_end_transaction(trans); 24042ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 24052ff7e61eSJeff Mahoney num_bytes); 24063fd0a558SYan, Zheng goto again; 24073fd0a558SYan, Zheng } 24083fd0a558SYan, Zheng } 24093fd0a558SYan, Zheng 24103fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 24113fd0a558SYan, Zheng 24123fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 24133fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 24143fd0a558SYan, Zheng struct btrfs_root, root_list); 24153fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24163fd0a558SYan, Zheng 24170b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 24183fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 24193fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 24203fd0a558SYan, Zheng 24213fd0a558SYan, Zheng /* 24223fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 24233fd0a558SYan, Zheng * knows it should resumes merging 24243fd0a558SYan, Zheng */ 24253fd0a558SYan, Zheng if (!err) 24263fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 24273fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 24283fd0a558SYan, Zheng 24293fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 24303fd0a558SYan, Zheng } 24313fd0a558SYan, Zheng 24323fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 24333fd0a558SYan, Zheng 24343fd0a558SYan, Zheng if (!err) 24353a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 24363fd0a558SYan, Zheng else 24373a45bb20SJeff Mahoney btrfs_end_transaction(trans); 24383fd0a558SYan, Zheng return err; 24393fd0a558SYan, Zheng } 24403fd0a558SYan, Zheng 24413fd0a558SYan, Zheng static noinline_for_stack 2442aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2443aca1bba6SLiu Bo { 2444aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2445aca1bba6SLiu Bo 2446aca1bba6SLiu Bo while (!list_empty(list)) { 2447aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2448aca1bba6SLiu Bo root_list); 2449bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 24506bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 24516bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 24526bdf131fSJosef Bacik reloc_root->node = NULL; 24536bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2454aca1bba6SLiu Bo } 2455aca1bba6SLiu Bo } 2456aca1bba6SLiu Bo 2457aca1bba6SLiu Bo static noinline_for_stack 245894404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 24593fd0a558SYan, Zheng { 24600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 24615d4f98a2SYan Zheng struct btrfs_root *root; 24625d4f98a2SYan Zheng struct btrfs_root *reloc_root; 24633fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24643fd0a558SYan, Zheng int found = 0; 2465aca1bba6SLiu Bo int ret = 0; 24663fd0a558SYan, Zheng again: 24673fd0a558SYan, Zheng root = rc->extent_root; 24687585717fSChris Mason 24697585717fSChris Mason /* 24707585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 24717585717fSChris Mason * we have to make sure nobody is in the middle of 24727585717fSChris Mason * adding their roots to the list while we are 24737585717fSChris Mason * doing this splice 24747585717fSChris Mason */ 24750b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24763fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 24770b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24785d4f98a2SYan Zheng 24793fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 24803fd0a558SYan, Zheng found = 1; 24813fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 24823fd0a558SYan, Zheng struct btrfs_root, root_list); 24835d4f98a2SYan Zheng 24845d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 24850b246afaSJeff Mahoney root = read_fs_root(fs_info, 24865d4f98a2SYan Zheng reloc_root->root_key.offset); 24875d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24885d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24895d4f98a2SYan Zheng 24903fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2491b37b39cdSJosef Bacik if (ret) { 249225e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 249325e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 249425e293c2SWang Shilong &reloc_roots); 2495aca1bba6SLiu Bo goto out; 2496b37b39cdSJosef Bacik } 24973fd0a558SYan, Zheng } else { 24983fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 249930d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 250030d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 250130d40577SQu Wenruo &rc->dirty_subvol_roots); 25023fd0a558SYan, Zheng } 25035d4f98a2SYan Zheng } 25045d4f98a2SYan Zheng 25053fd0a558SYan, Zheng if (found) { 25063fd0a558SYan, Zheng found = 0; 25073fd0a558SYan, Zheng goto again; 25085d4f98a2SYan Zheng } 2509aca1bba6SLiu Bo out: 2510aca1bba6SLiu Bo if (ret) { 25110b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2512aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2513aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2514467bb1d2SWang Shilong 2515467bb1d2SWang Shilong /* new reloc root may be added */ 25160b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2517467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 25180b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2519467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2520467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2521aca1bba6SLiu Bo } 2522aca1bba6SLiu Bo 25235d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 25245d4f98a2SYan Zheng } 25255d4f98a2SYan Zheng 25265d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 25275d4f98a2SYan Zheng { 25285d4f98a2SYan Zheng struct tree_block *block; 25295d4f98a2SYan Zheng struct rb_node *rb_node; 25305d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 25315d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 25325d4f98a2SYan Zheng rb_erase(rb_node, blocks); 25335d4f98a2SYan Zheng kfree(block); 25345d4f98a2SYan Zheng } 25355d4f98a2SYan Zheng } 25365d4f98a2SYan Zheng 25375d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 25385d4f98a2SYan Zheng struct btrfs_root *reloc_root) 25395d4f98a2SYan Zheng { 25400b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 25415d4f98a2SYan Zheng struct btrfs_root *root; 25425d4f98a2SYan Zheng 25435d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 25445d4f98a2SYan Zheng return 0; 25455d4f98a2SYan Zheng 25460b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25475d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25485d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25495d4f98a2SYan Zheng 25505d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 25515d4f98a2SYan Zheng } 25525d4f98a2SYan Zheng 25533fd0a558SYan, Zheng static noinline_for_stack 25543fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 25553fd0a558SYan, Zheng struct reloc_control *rc, 25565d4f98a2SYan Zheng struct backref_node *node, 2557dc4103f9SWang Shilong struct backref_edge *edges[]) 25585d4f98a2SYan Zheng { 25595d4f98a2SYan Zheng struct backref_node *next; 25605d4f98a2SYan Zheng struct btrfs_root *root; 25613fd0a558SYan, Zheng int index = 0; 25623fd0a558SYan, Zheng 25635d4f98a2SYan Zheng next = node; 25645d4f98a2SYan Zheng while (1) { 25655d4f98a2SYan Zheng cond_resched(); 25665d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 25675d4f98a2SYan Zheng root = next->root; 25683fd0a558SYan, Zheng BUG_ON(!root); 256927cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 25705d4f98a2SYan Zheng 25715d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 25725d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 25735d4f98a2SYan Zheng break; 25745d4f98a2SYan Zheng } 25755d4f98a2SYan Zheng 25765d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 25773fd0a558SYan, Zheng root = root->reloc_root; 25783fd0a558SYan, Zheng 25793fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 25803fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 25813fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 25823fd0a558SYan, Zheng next->new_bytenr = root->node->start; 25833fd0a558SYan, Zheng next->root = root; 25843fd0a558SYan, Zheng list_add_tail(&next->list, 25853fd0a558SYan, Zheng &rc->backref_cache.changed); 25863fd0a558SYan, Zheng __mark_block_processed(rc, next); 25875d4f98a2SYan Zheng break; 25885d4f98a2SYan Zheng } 25895d4f98a2SYan Zheng 25903fd0a558SYan, Zheng WARN_ON(1); 25915d4f98a2SYan Zheng root = NULL; 25925d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25935d4f98a2SYan Zheng if (!next || next->level <= node->level) 25945d4f98a2SYan Zheng break; 25955d4f98a2SYan Zheng } 25963fd0a558SYan, Zheng if (!root) 25973fd0a558SYan, Zheng return NULL; 25985d4f98a2SYan Zheng 25993fd0a558SYan, Zheng next = node; 26003fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 26013fd0a558SYan, Zheng while (1) { 26023fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 26033fd0a558SYan, Zheng if (--index < 0) 26043fd0a558SYan, Zheng break; 26053fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 26063fd0a558SYan, Zheng } 26075d4f98a2SYan Zheng return root; 26085d4f98a2SYan Zheng } 26095d4f98a2SYan Zheng 26103fd0a558SYan, Zheng /* 26113fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 26123fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 26133fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 26143fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 26153fd0a558SYan, Zheng */ 26165d4f98a2SYan Zheng static noinline_for_stack 2617147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 26185d4f98a2SYan Zheng { 26193fd0a558SYan, Zheng struct backref_node *next; 26203fd0a558SYan, Zheng struct btrfs_root *root; 26213fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 26225d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26233fd0a558SYan, Zheng int index = 0; 26243fd0a558SYan, Zheng 26253fd0a558SYan, Zheng next = node; 26263fd0a558SYan, Zheng while (1) { 26273fd0a558SYan, Zheng cond_resched(); 26283fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 26293fd0a558SYan, Zheng root = next->root; 26303fd0a558SYan, Zheng BUG_ON(!root); 26313fd0a558SYan, Zheng 263225985edcSLucas De Marchi /* no other choice for non-references counted tree */ 263327cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 26343fd0a558SYan, Zheng return root; 26353fd0a558SYan, Zheng 26363fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 26373fd0a558SYan, Zheng fs_root = root; 26383fd0a558SYan, Zheng 26393fd0a558SYan, Zheng if (next != node) 26403fd0a558SYan, Zheng return NULL; 26413fd0a558SYan, Zheng 26423fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26433fd0a558SYan, Zheng if (!next || next->level <= node->level) 26443fd0a558SYan, Zheng break; 26453fd0a558SYan, Zheng } 26463fd0a558SYan, Zheng 26473fd0a558SYan, Zheng if (!fs_root) 26483fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 26493fd0a558SYan, Zheng return fs_root; 26505d4f98a2SYan Zheng } 26515d4f98a2SYan Zheng 26525d4f98a2SYan Zheng static noinline_for_stack 26533fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 26543fd0a558SYan, Zheng struct backref_node *node, int reserve) 26555d4f98a2SYan Zheng { 26560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 26573fd0a558SYan, Zheng struct backref_node *next = node; 26583fd0a558SYan, Zheng struct backref_edge *edge; 26593fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26603fd0a558SYan, Zheng u64 num_bytes = 0; 26613fd0a558SYan, Zheng int index = 0; 26625d4f98a2SYan Zheng 26633fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 26643fd0a558SYan, Zheng 26653fd0a558SYan, Zheng while (next) { 26663fd0a558SYan, Zheng cond_resched(); 26675d4f98a2SYan Zheng while (1) { 26683fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 26695d4f98a2SYan Zheng break; 26705d4f98a2SYan Zheng 26710b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 26723fd0a558SYan, Zheng 26733fd0a558SYan, Zheng if (list_empty(&next->upper)) 26743fd0a558SYan, Zheng break; 26753fd0a558SYan, Zheng 26763fd0a558SYan, Zheng edge = list_entry(next->upper.next, 26773fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 26783fd0a558SYan, Zheng edges[index++] = edge; 26793fd0a558SYan, Zheng next = edge->node[UPPER]; 26805d4f98a2SYan Zheng } 26813fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26823fd0a558SYan, Zheng } 26833fd0a558SYan, Zheng return num_bytes; 26843fd0a558SYan, Zheng } 26853fd0a558SYan, Zheng 26863fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 26873fd0a558SYan, Zheng struct reloc_control *rc, 26883fd0a558SYan, Zheng struct backref_node *node) 26893fd0a558SYan, Zheng { 26903fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2691da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 26923fd0a558SYan, Zheng u64 num_bytes; 26933fd0a558SYan, Zheng int ret; 26940647bf56SWang Shilong u64 tmp; 26953fd0a558SYan, Zheng 26963fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 26973fd0a558SYan, Zheng 26983fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 26990647bf56SWang Shilong rc->reserved_bytes += num_bytes; 27008ca17f0fSJosef Bacik 27018ca17f0fSJosef Bacik /* 27028ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 27038ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 27048ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 27058ca17f0fSJosef Bacik */ 27060647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 27078ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 27083fd0a558SYan, Zheng if (ret) { 2709da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 27100647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 27110647bf56SWang Shilong tmp <<= 1; 27120647bf56SWang Shilong /* 27130647bf56SWang Shilong * only one thread can access block_rsv at this point, 27140647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 27150647bf56SWang Shilong * we expand more reservation size here to allow enough 271652042d8eSAndrea Gelmini * space for relocation and we will return earlier in 27170647bf56SWang Shilong * enospc case. 27180647bf56SWang Shilong */ 2719da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 27200647bf56SWang Shilong RELOCATION_RESERVED_NODES; 27218ca17f0fSJosef Bacik return -EAGAIN; 27223fd0a558SYan, Zheng } 27233fd0a558SYan, Zheng 27243fd0a558SYan, Zheng return 0; 27253fd0a558SYan, Zheng } 27263fd0a558SYan, Zheng 27275d4f98a2SYan Zheng /* 27285d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 27295d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 27305d4f98a2SYan Zheng * 27315d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 27325d4f98a2SYan Zheng * in that case this function just updates pointers. 27335d4f98a2SYan Zheng */ 27345d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 27353fd0a558SYan, Zheng struct reloc_control *rc, 27365d4f98a2SYan Zheng struct backref_node *node, 27375d4f98a2SYan Zheng struct btrfs_key *key, 27385d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 27395d4f98a2SYan Zheng { 27402ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27415d4f98a2SYan Zheng struct backref_node *upper; 27425d4f98a2SYan Zheng struct backref_edge *edge; 27435d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27445d4f98a2SYan Zheng struct btrfs_root *root; 27455d4f98a2SYan Zheng struct extent_buffer *eb; 27465d4f98a2SYan Zheng u32 blocksize; 27475d4f98a2SYan Zheng u64 bytenr; 27485d4f98a2SYan Zheng u64 generation; 27495d4f98a2SYan Zheng int slot; 27505d4f98a2SYan Zheng int ret; 27515d4f98a2SYan Zheng int err = 0; 27525d4f98a2SYan Zheng 27535d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 27545d4f98a2SYan Zheng 27555d4f98a2SYan Zheng path->lowest_level = node->level + 1; 27563fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 27575d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2758581c1760SQu Wenruo struct btrfs_key first_key; 275982fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2760581c1760SQu Wenruo 27615d4f98a2SYan Zheng cond_resched(); 27625d4f98a2SYan Zheng 27635d4f98a2SYan Zheng upper = edge->node[UPPER]; 2764dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 27653fd0a558SYan, Zheng BUG_ON(!root); 27665d4f98a2SYan Zheng 27673fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 27683fd0a558SYan, Zheng if (!lowest) { 27693fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 27703fd0a558SYan, Zheng upper->level, &slot); 2771cbca7d59SFilipe Manana if (ret < 0) { 2772cbca7d59SFilipe Manana err = ret; 2773cbca7d59SFilipe Manana goto next; 2774cbca7d59SFilipe Manana } 27753fd0a558SYan, Zheng BUG_ON(ret); 27763fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27773fd0a558SYan, Zheng if (node->eb->start == bytenr) 27783fd0a558SYan, Zheng goto next; 27793fd0a558SYan, Zheng } 27805d4f98a2SYan Zheng drop_node_buffer(upper); 27813fd0a558SYan, Zheng } 27825d4f98a2SYan Zheng 27835d4f98a2SYan Zheng if (!upper->eb) { 27845d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 27853561b9dbSLiu Bo if (ret) { 27863561b9dbSLiu Bo if (ret < 0) 27875d4f98a2SYan Zheng err = ret; 27883561b9dbSLiu Bo else 27893561b9dbSLiu Bo err = -ENOENT; 27903561b9dbSLiu Bo 27913561b9dbSLiu Bo btrfs_release_path(path); 27925d4f98a2SYan Zheng break; 27935d4f98a2SYan Zheng } 27945d4f98a2SYan Zheng 27953fd0a558SYan, Zheng if (!upper->eb) { 27963fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 27973fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 27983fd0a558SYan, Zheng } else { 27993fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 28003fd0a558SYan, Zheng } 28013fd0a558SYan, Zheng 28023fd0a558SYan, Zheng upper->locked = 1; 28033fd0a558SYan, Zheng path->locks[upper->level] = 0; 28043fd0a558SYan, Zheng 28055d4f98a2SYan Zheng slot = path->slots[upper->level]; 2806b3b4aa74SDavid Sterba btrfs_release_path(path); 28075d4f98a2SYan Zheng } else { 28085d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 28095d4f98a2SYan Zheng &slot); 2810cbca7d59SFilipe Manana if (ret < 0) { 2811cbca7d59SFilipe Manana err = ret; 2812cbca7d59SFilipe Manana goto next; 2813cbca7d59SFilipe Manana } 28145d4f98a2SYan Zheng BUG_ON(ret); 28155d4f98a2SYan Zheng } 28165d4f98a2SYan Zheng 28175d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 28183fd0a558SYan, Zheng if (lowest) { 28194547f4d8SLiu Bo if (bytenr != node->bytenr) { 28204547f4d8SLiu Bo btrfs_err(root->fs_info, 28214547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 28224547f4d8SLiu Bo bytenr, node->bytenr, slot, 28234547f4d8SLiu Bo upper->eb->start); 28244547f4d8SLiu Bo err = -EIO; 28254547f4d8SLiu Bo goto next; 28264547f4d8SLiu Bo } 28275d4f98a2SYan Zheng } else { 28283fd0a558SYan, Zheng if (node->eb->start == bytenr) 28293fd0a558SYan, Zheng goto next; 28305d4f98a2SYan Zheng } 28315d4f98a2SYan Zheng 2832da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 28335d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2834581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2835581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2836581c1760SQu Wenruo upper->level - 1, &first_key); 283764c043deSLiu Bo if (IS_ERR(eb)) { 283864c043deSLiu Bo err = PTR_ERR(eb); 283964c043deSLiu Bo goto next; 284064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2841416bc658SJosef Bacik free_extent_buffer(eb); 284297d9a8a4STsutomu Itoh err = -EIO; 284397d9a8a4STsutomu Itoh goto next; 284497d9a8a4STsutomu Itoh } 28455d4f98a2SYan Zheng btrfs_tree_lock(eb); 28468bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 28475d4f98a2SYan Zheng 28485d4f98a2SYan Zheng if (!node->eb) { 28495d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 28505d4f98a2SYan Zheng slot, &eb); 28513fd0a558SYan, Zheng btrfs_tree_unlock(eb); 28523fd0a558SYan, Zheng free_extent_buffer(eb); 28535d4f98a2SYan Zheng if (ret < 0) { 28545d4f98a2SYan Zheng err = ret; 28553fd0a558SYan, Zheng goto next; 28565d4f98a2SYan Zheng } 28573fd0a558SYan, Zheng BUG_ON(node->eb != eb); 28585d4f98a2SYan Zheng } else { 28595d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 28605d4f98a2SYan Zheng node->eb->start); 28615d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 28625d4f98a2SYan Zheng trans->transid); 28635d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 28645d4f98a2SYan Zheng 286582fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 28665d4f98a2SYan Zheng node->eb->start, blocksize, 286782fa113fSQu Wenruo upper->eb->start); 286882fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 286982fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 287082fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 287182fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 28725d4f98a2SYan Zheng BUG_ON(ret); 28735d4f98a2SYan Zheng 28745d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 28755d4f98a2SYan Zheng BUG_ON(ret); 28765d4f98a2SYan Zheng } 28773fd0a558SYan, Zheng next: 28783fd0a558SYan, Zheng if (!upper->pending) 28793fd0a558SYan, Zheng drop_node_buffer(upper); 28803fd0a558SYan, Zheng else 28813fd0a558SYan, Zheng unlock_node_buffer(upper); 28823fd0a558SYan, Zheng if (err) 28833fd0a558SYan, Zheng break; 28845d4f98a2SYan Zheng } 28853fd0a558SYan, Zheng 28863fd0a558SYan, Zheng if (!err && node->pending) { 28873fd0a558SYan, Zheng drop_node_buffer(node); 28883fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 28893fd0a558SYan, Zheng node->pending = 0; 28905d4f98a2SYan Zheng } 28913fd0a558SYan, Zheng 28925d4f98a2SYan Zheng path->lowest_level = 0; 28933fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 28945d4f98a2SYan Zheng return err; 28955d4f98a2SYan Zheng } 28965d4f98a2SYan Zheng 28975d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 28983fd0a558SYan, Zheng struct reloc_control *rc, 28995d4f98a2SYan Zheng struct backref_node *node, 29005d4f98a2SYan Zheng struct btrfs_path *path) 29015d4f98a2SYan Zheng { 29025d4f98a2SYan Zheng struct btrfs_key key; 29035d4f98a2SYan Zheng 29045d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 29053fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 29065d4f98a2SYan Zheng } 29075d4f98a2SYan Zheng 29085d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 29093fd0a558SYan, Zheng struct reloc_control *rc, 29103fd0a558SYan, Zheng struct btrfs_path *path, int err) 29115d4f98a2SYan Zheng { 29123fd0a558SYan, Zheng LIST_HEAD(list); 29133fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 29145d4f98a2SYan Zheng struct backref_node *node; 29155d4f98a2SYan Zheng int level; 29165d4f98a2SYan Zheng int ret; 29175d4f98a2SYan Zheng 29185d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 29195d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 29205d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 29213fd0a558SYan, Zheng struct backref_node, list); 29223fd0a558SYan, Zheng list_move_tail(&node->list, &list); 29233fd0a558SYan, Zheng BUG_ON(!node->pending); 29245d4f98a2SYan Zheng 29253fd0a558SYan, Zheng if (!err) { 29263fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 29275d4f98a2SYan Zheng if (ret < 0) 29285d4f98a2SYan Zheng err = ret; 29295d4f98a2SYan Zheng } 29305d4f98a2SYan Zheng } 29313fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 29323fd0a558SYan, Zheng } 29335d4f98a2SYan Zheng return err; 29345d4f98a2SYan Zheng } 29355d4f98a2SYan Zheng 29365d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 29373fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 29383fd0a558SYan, Zheng { 29393fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2940ceeb0ae7SDavid Sterba EXTENT_DIRTY); 29413fd0a558SYan, Zheng } 29423fd0a558SYan, Zheng 29433fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 29445d4f98a2SYan Zheng struct backref_node *node) 29455d4f98a2SYan Zheng { 29465d4f98a2SYan Zheng u32 blocksize; 29475d4f98a2SYan Zheng if (node->level == 0 || 29485d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2949da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 29503fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 29515d4f98a2SYan Zheng } 29525d4f98a2SYan Zheng node->processed = 1; 29535d4f98a2SYan Zheng } 29545d4f98a2SYan Zheng 29555d4f98a2SYan Zheng /* 29565d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 29575d4f98a2SYan Zheng * as processed. 29585d4f98a2SYan Zheng */ 29595d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 29605d4f98a2SYan Zheng struct backref_node *node) 29615d4f98a2SYan Zheng { 29625d4f98a2SYan Zheng struct backref_node *next = node; 29635d4f98a2SYan Zheng struct backref_edge *edge; 29645d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 29655d4f98a2SYan Zheng int index = 0; 29665d4f98a2SYan Zheng 29675d4f98a2SYan Zheng while (next) { 29685d4f98a2SYan Zheng cond_resched(); 29695d4f98a2SYan Zheng while (1) { 29705d4f98a2SYan Zheng if (next->processed) 29715d4f98a2SYan Zheng break; 29725d4f98a2SYan Zheng 29733fd0a558SYan, Zheng __mark_block_processed(rc, next); 29745d4f98a2SYan Zheng 29755d4f98a2SYan Zheng if (list_empty(&next->upper)) 29765d4f98a2SYan Zheng break; 29775d4f98a2SYan Zheng 29785d4f98a2SYan Zheng edge = list_entry(next->upper.next, 29795d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 29805d4f98a2SYan Zheng edges[index++] = edge; 29815d4f98a2SYan Zheng next = edge->node[UPPER]; 29825d4f98a2SYan Zheng } 29835d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 29845d4f98a2SYan Zheng } 29855d4f98a2SYan Zheng } 29865d4f98a2SYan Zheng 29877476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 29885d4f98a2SYan Zheng { 2989da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 29907476dfdaSDavid Sterba 29915d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 29929655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 29935d4f98a2SYan Zheng return 1; 29945d4f98a2SYan Zheng return 0; 29955d4f98a2SYan Zheng } 29965d4f98a2SYan Zheng 29972ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 29985d4f98a2SYan Zheng struct tree_block *block) 29995d4f98a2SYan Zheng { 30005d4f98a2SYan Zheng struct extent_buffer *eb; 30015d4f98a2SYan Zheng 30025d4f98a2SYan Zheng BUG_ON(block->key_ready); 3003581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 3004581c1760SQu Wenruo block->level, NULL); 300564c043deSLiu Bo if (IS_ERR(eb)) { 300664c043deSLiu Bo return PTR_ERR(eb); 300764c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3008416bc658SJosef Bacik free_extent_buffer(eb); 3009416bc658SJosef Bacik return -EIO; 3010416bc658SJosef Bacik } 30115d4f98a2SYan Zheng if (block->level == 0) 30125d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 30135d4f98a2SYan Zheng else 30145d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 30155d4f98a2SYan Zheng free_extent_buffer(eb); 30165d4f98a2SYan Zheng block->key_ready = 1; 30175d4f98a2SYan Zheng return 0; 30185d4f98a2SYan Zheng } 30195d4f98a2SYan Zheng 30205d4f98a2SYan Zheng /* 30215d4f98a2SYan Zheng * helper function to relocate a tree block 30225d4f98a2SYan Zheng */ 30235d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 30245d4f98a2SYan Zheng struct reloc_control *rc, 30255d4f98a2SYan Zheng struct backref_node *node, 30265d4f98a2SYan Zheng struct btrfs_key *key, 30275d4f98a2SYan Zheng struct btrfs_path *path) 30285d4f98a2SYan Zheng { 30295d4f98a2SYan Zheng struct btrfs_root *root; 30303fd0a558SYan, Zheng int ret = 0; 30315d4f98a2SYan Zheng 30323fd0a558SYan, Zheng if (!node) 30335d4f98a2SYan Zheng return 0; 30343fd0a558SYan, Zheng 30353fd0a558SYan, Zheng BUG_ON(node->processed); 3036147d256eSZhaolei root = select_one_root(node); 30373fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 30383fd0a558SYan, Zheng update_processed_blocks(rc, node); 30393fd0a558SYan, Zheng goto out; 30405d4f98a2SYan Zheng } 30415d4f98a2SYan Zheng 304227cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30433fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 30443fd0a558SYan, Zheng if (ret) 30455d4f98a2SYan Zheng goto out; 30465d4f98a2SYan Zheng } 30473fd0a558SYan, Zheng 30483fd0a558SYan, Zheng if (root) { 304927cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30503fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 30513fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 30523fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 30533fd0a558SYan, Zheng root = root->reloc_root; 30543fd0a558SYan, Zheng node->new_bytenr = root->node->start; 30553fd0a558SYan, Zheng node->root = root; 30563fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 30573fd0a558SYan, Zheng } else { 30585d4f98a2SYan Zheng path->lowest_level = node->level; 30595d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3060b3b4aa74SDavid Sterba btrfs_release_path(path); 30613fd0a558SYan, Zheng if (ret > 0) 30625d4f98a2SYan Zheng ret = 0; 30633fd0a558SYan, Zheng } 30643fd0a558SYan, Zheng if (!ret) 30653fd0a558SYan, Zheng update_processed_blocks(rc, node); 30663fd0a558SYan, Zheng } else { 30673fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 30683fd0a558SYan, Zheng } 30695d4f98a2SYan Zheng out: 30700647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 30713fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 30725d4f98a2SYan Zheng return ret; 30735d4f98a2SYan Zheng } 30745d4f98a2SYan Zheng 30755d4f98a2SYan Zheng /* 30765d4f98a2SYan Zheng * relocate a list of blocks 30775d4f98a2SYan Zheng */ 30785d4f98a2SYan Zheng static noinline_for_stack 30795d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 30805d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 30815d4f98a2SYan Zheng { 30822ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 30835d4f98a2SYan Zheng struct backref_node *node; 30845d4f98a2SYan Zheng struct btrfs_path *path; 30855d4f98a2SYan Zheng struct tree_block *block; 308698ff7b94SQu Wenruo struct tree_block *next; 30875d4f98a2SYan Zheng int ret; 30885d4f98a2SYan Zheng int err = 0; 30895d4f98a2SYan Zheng 30905d4f98a2SYan Zheng path = btrfs_alloc_path(); 3091e1a12670SLiu Bo if (!path) { 3092e1a12670SLiu Bo err = -ENOMEM; 309334c2b290SDavid Sterba goto out_free_blocks; 3094e1a12670SLiu Bo } 30955d4f98a2SYan Zheng 309698ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 309798ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 30985d4f98a2SYan Zheng if (!block->key_ready) 30992ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 31005d4f98a2SYan Zheng } 31015d4f98a2SYan Zheng 310298ff7b94SQu Wenruo /* Get first keys */ 310398ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 310434c2b290SDavid Sterba if (!block->key_ready) { 31052ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 310634c2b290SDavid Sterba if (err) 310734c2b290SDavid Sterba goto out_free_path; 310834c2b290SDavid Sterba } 31095d4f98a2SYan Zheng } 31105d4f98a2SYan Zheng 311198ff7b94SQu Wenruo /* Do tree relocation */ 311298ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 31133fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 31145d4f98a2SYan Zheng block->level, block->bytenr); 31155d4f98a2SYan Zheng if (IS_ERR(node)) { 31165d4f98a2SYan Zheng err = PTR_ERR(node); 31175d4f98a2SYan Zheng goto out; 31185d4f98a2SYan Zheng } 31195d4f98a2SYan Zheng 31205d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 31215d4f98a2SYan Zheng path); 31225d4f98a2SYan Zheng if (ret < 0) { 312398ff7b94SQu Wenruo if (ret != -EAGAIN || &block->rb_node == rb_first(blocks)) 31245d4f98a2SYan Zheng err = ret; 31255d4f98a2SYan Zheng goto out; 31265d4f98a2SYan Zheng } 31275d4f98a2SYan Zheng } 31285d4f98a2SYan Zheng out: 31293fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 31305d4f98a2SYan Zheng 313134c2b290SDavid Sterba out_free_path: 31325d4f98a2SYan Zheng btrfs_free_path(path); 313334c2b290SDavid Sterba out_free_blocks: 3134e1a12670SLiu Bo free_block_list(blocks); 31355d4f98a2SYan Zheng return err; 31365d4f98a2SYan Zheng } 31375d4f98a2SYan Zheng 31385d4f98a2SYan Zheng static noinline_for_stack 3139efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3140efa56464SYan, Zheng struct file_extent_cluster *cluster) 3141efa56464SYan, Zheng { 3142efa56464SYan, Zheng u64 alloc_hint = 0; 3143efa56464SYan, Zheng u64 start; 3144efa56464SYan, Zheng u64 end; 3145efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3146efa56464SYan, Zheng u64 num_bytes; 3147efa56464SYan, Zheng int nr = 0; 3148efa56464SYan, Zheng int ret = 0; 3149dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3150dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 315118513091SWang Xiaoguang u64 cur_offset; 3152364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3153efa56464SYan, Zheng 3154efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 31555955102cSAl Viro inode_lock(inode); 3156efa56464SYan, Zheng 3157364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3158dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3159efa56464SYan, Zheng if (ret) 3160efa56464SYan, Zheng goto out; 3161efa56464SYan, Zheng 316218513091SWang Xiaoguang cur_offset = prealloc_start; 3163efa56464SYan, Zheng while (nr < cluster->nr) { 3164efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3165efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3166efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3167efa56464SYan, Zheng else 3168efa56464SYan, Zheng end = cluster->end - offset; 3169efa56464SYan, Zheng 3170d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3171efa56464SYan, Zheng num_bytes = end + 1 - start; 317218513091SWang Xiaoguang if (cur_offset < start) 3173bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3174bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3175efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3176efa56464SYan, Zheng num_bytes, num_bytes, 3177efa56464SYan, Zheng end + 1, &alloc_hint); 317818513091SWang Xiaoguang cur_offset = end + 1; 3179d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3180efa56464SYan, Zheng if (ret) 3181efa56464SYan, Zheng break; 3182efa56464SYan, Zheng nr++; 3183efa56464SYan, Zheng } 318418513091SWang Xiaoguang if (cur_offset < prealloc_end) 3185bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3186bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3187efa56464SYan, Zheng out: 31885955102cSAl Viro inode_unlock(inode); 3189364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3190efa56464SYan, Zheng return ret; 3191efa56464SYan, Zheng } 3192efa56464SYan, Zheng 3193efa56464SYan, Zheng static noinline_for_stack 31940257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 31950257bb82SYan, Zheng u64 block_start) 31965d4f98a2SYan Zheng { 31975d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 31985d4f98a2SYan Zheng struct extent_map *em; 31990257bb82SYan, Zheng int ret = 0; 32005d4f98a2SYan Zheng 3201172ddd60SDavid Sterba em = alloc_extent_map(); 32020257bb82SYan, Zheng if (!em) 32030257bb82SYan, Zheng return -ENOMEM; 32040257bb82SYan, Zheng 32055d4f98a2SYan Zheng em->start = start; 32060257bb82SYan, Zheng em->len = end + 1 - start; 32070257bb82SYan, Zheng em->block_len = em->len; 32080257bb82SYan, Zheng em->block_start = block_start; 32095d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 32105d4f98a2SYan Zheng 3211d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 32125d4f98a2SYan Zheng while (1) { 3213890871beSChris Mason write_lock(&em_tree->lock); 321409a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3215890871beSChris Mason write_unlock(&em_tree->lock); 32165d4f98a2SYan Zheng if (ret != -EEXIST) { 32175d4f98a2SYan Zheng free_extent_map(em); 32185d4f98a2SYan Zheng break; 32195d4f98a2SYan Zheng } 3220dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 32215d4f98a2SYan Zheng } 3222d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 32230257bb82SYan, Zheng return ret; 32240257bb82SYan, Zheng } 32255d4f98a2SYan Zheng 32260257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 32270257bb82SYan, Zheng struct file_extent_cluster *cluster) 32280257bb82SYan, Zheng { 32292ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 32300257bb82SYan, Zheng u64 page_start; 32310257bb82SYan, Zheng u64 page_end; 32320257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 32330257bb82SYan, Zheng unsigned long index; 32340257bb82SYan, Zheng unsigned long last_index; 32350257bb82SYan, Zheng struct page *page; 32360257bb82SYan, Zheng struct file_ra_state *ra; 32373b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 32380257bb82SYan, Zheng int nr = 0; 32390257bb82SYan, Zheng int ret = 0; 32400257bb82SYan, Zheng 32410257bb82SYan, Zheng if (!cluster->nr) 32420257bb82SYan, Zheng return 0; 32430257bb82SYan, Zheng 32440257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 32450257bb82SYan, Zheng if (!ra) 32460257bb82SYan, Zheng return -ENOMEM; 32470257bb82SYan, Zheng 3248efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 32490257bb82SYan, Zheng if (ret) 3250efa56464SYan, Zheng goto out; 32510257bb82SYan, Zheng 32520257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 32530257bb82SYan, Zheng 3254efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3255efa56464SYan, Zheng cluster->end - offset, cluster->start); 3256efa56464SYan, Zheng if (ret) 3257efa56464SYan, Zheng goto out; 3258efa56464SYan, Zheng 325909cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 326009cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 32610257bb82SYan, Zheng while (index <= last_index) { 32629f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 32639f3db423SNikolay Borisov PAGE_SIZE); 3264efa56464SYan, Zheng if (ret) 3265efa56464SYan, Zheng goto out; 3266efa56464SYan, Zheng 32670257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 32680257bb82SYan, Zheng if (!page) { 32690257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 32700257bb82SYan, Zheng ra, NULL, index, 32710257bb82SYan, Zheng last_index + 1 - index); 3272a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 32733b16a4e3SJosef Bacik mask); 32740257bb82SYan, Zheng if (!page) { 3275691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 327643b18595SQu Wenruo PAGE_SIZE, true); 327744db1216SFilipe Manana btrfs_delalloc_release_extents(BTRFS_I(inode), 32788702ba93SQu Wenruo PAGE_SIZE); 32790257bb82SYan, Zheng ret = -ENOMEM; 3280efa56464SYan, Zheng goto out; 32810257bb82SYan, Zheng } 32820257bb82SYan, Zheng } 32830257bb82SYan, Zheng 32840257bb82SYan, Zheng if (PageReadahead(page)) { 32850257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 32860257bb82SYan, Zheng ra, NULL, page, index, 32870257bb82SYan, Zheng last_index + 1 - index); 32880257bb82SYan, Zheng } 32890257bb82SYan, Zheng 32900257bb82SYan, Zheng if (!PageUptodate(page)) { 32910257bb82SYan, Zheng btrfs_readpage(NULL, page); 32920257bb82SYan, Zheng lock_page(page); 32930257bb82SYan, Zheng if (!PageUptodate(page)) { 32940257bb82SYan, Zheng unlock_page(page); 329509cbfeafSKirill A. Shutemov put_page(page); 3296691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 329743b18595SQu Wenruo PAGE_SIZE, true); 32988b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 32998702ba93SQu Wenruo PAGE_SIZE); 33000257bb82SYan, Zheng ret = -EIO; 3301efa56464SYan, Zheng goto out; 33020257bb82SYan, Zheng } 33030257bb82SYan, Zheng } 33040257bb82SYan, Zheng 33054eee4fa4SMiao Xie page_start = page_offset(page); 330609cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 33070257bb82SYan, Zheng 3308d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 33090257bb82SYan, Zheng 33100257bb82SYan, Zheng set_page_extent_mapped(page); 33110257bb82SYan, Zheng 33120257bb82SYan, Zheng if (nr < cluster->nr && 33130257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 33140257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 33150257bb82SYan, Zheng page_start, page_end, 3316ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 33170257bb82SYan, Zheng nr++; 33180257bb82SYan, Zheng } 33190257bb82SYan, Zheng 3320765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3321330a5827SNikolay Borisov NULL); 3322765f3cebSNikolay Borisov if (ret) { 3323765f3cebSNikolay Borisov unlock_page(page); 3324765f3cebSNikolay Borisov put_page(page); 3325765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 332643b18595SQu Wenruo PAGE_SIZE, true); 3327765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 33288702ba93SQu Wenruo PAGE_SIZE); 3329765f3cebSNikolay Borisov 3330765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3331765f3cebSNikolay Borisov page_start, page_end, 3332765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3333765f3cebSNikolay Borisov goto out; 3334765f3cebSNikolay Borisov 3335765f3cebSNikolay Borisov } 33360257bb82SYan, Zheng set_page_dirty(page); 33370257bb82SYan, Zheng 33380257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3339d0082371SJeff Mahoney page_start, page_end); 33400257bb82SYan, Zheng unlock_page(page); 334109cbfeafSKirill A. Shutemov put_page(page); 33420257bb82SYan, Zheng 33430257bb82SYan, Zheng index++; 33448702ba93SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); 3345efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 33462ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 33470257bb82SYan, Zheng } 33480257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3349efa56464SYan, Zheng out: 33500257bb82SYan, Zheng kfree(ra); 33510257bb82SYan, Zheng return ret; 33520257bb82SYan, Zheng } 33530257bb82SYan, Zheng 33540257bb82SYan, Zheng static noinline_for_stack 33550257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 33560257bb82SYan, Zheng struct file_extent_cluster *cluster) 33570257bb82SYan, Zheng { 33580257bb82SYan, Zheng int ret; 33590257bb82SYan, Zheng 33600257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 33610257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33620257bb82SYan, Zheng if (ret) 33630257bb82SYan, Zheng return ret; 33640257bb82SYan, Zheng cluster->nr = 0; 33650257bb82SYan, Zheng } 33660257bb82SYan, Zheng 33670257bb82SYan, Zheng if (!cluster->nr) 33680257bb82SYan, Zheng cluster->start = extent_key->objectid; 33690257bb82SYan, Zheng else 33700257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 33710257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 33720257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 33730257bb82SYan, Zheng cluster->nr++; 33740257bb82SYan, Zheng 33750257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 33760257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33770257bb82SYan, Zheng if (ret) 33780257bb82SYan, Zheng return ret; 33790257bb82SYan, Zheng cluster->nr = 0; 33800257bb82SYan, Zheng } 33810257bb82SYan, Zheng return 0; 33825d4f98a2SYan Zheng } 33835d4f98a2SYan Zheng 33845d4f98a2SYan Zheng /* 33855d4f98a2SYan Zheng * helper to add a tree block to the list. 33865d4f98a2SYan Zheng * the major work is getting the generation and level of the block 33875d4f98a2SYan Zheng */ 33885d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 33895d4f98a2SYan Zheng struct btrfs_key *extent_key, 33905d4f98a2SYan Zheng struct btrfs_path *path, 33915d4f98a2SYan Zheng struct rb_root *blocks) 33925d4f98a2SYan Zheng { 33935d4f98a2SYan Zheng struct extent_buffer *eb; 33945d4f98a2SYan Zheng struct btrfs_extent_item *ei; 33955d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 33965d4f98a2SYan Zheng struct tree_block *block; 33975d4f98a2SYan Zheng struct rb_node *rb_node; 33985d4f98a2SYan Zheng u32 item_size; 33995d4f98a2SYan Zheng int level = -1; 34007fdf4b60SWang Shilong u64 generation; 34015d4f98a2SYan Zheng 34025d4f98a2SYan Zheng eb = path->nodes[0]; 34035d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 34045d4f98a2SYan Zheng 34053173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 34063173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 34075d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 34085d4f98a2SYan Zheng struct btrfs_extent_item); 34093173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 34105d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 34115d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 34125d4f98a2SYan Zheng } else { 34133173a18fSJosef Bacik level = (int)extent_key->offset; 34143173a18fSJosef Bacik } 34153173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 34166d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3417ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3418ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3419ba3c2b19SNikolay Borisov return -EINVAL; 34203173a18fSJosef Bacik } else { 34215d4f98a2SYan Zheng BUG(); 34225d4f98a2SYan Zheng } 34235d4f98a2SYan Zheng 3424b3b4aa74SDavid Sterba btrfs_release_path(path); 34255d4f98a2SYan Zheng 34265d4f98a2SYan Zheng BUG_ON(level == -1); 34275d4f98a2SYan Zheng 34285d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 34295d4f98a2SYan Zheng if (!block) 34305d4f98a2SYan Zheng return -ENOMEM; 34315d4f98a2SYan Zheng 34325d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3433da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 34345d4f98a2SYan Zheng block->key.offset = generation; 34355d4f98a2SYan Zheng block->level = level; 34365d4f98a2SYan Zheng block->key_ready = 0; 34375d4f98a2SYan Zheng 34385d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 343943c04fb1SJeff Mahoney if (rb_node) 344043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 34415d4f98a2SYan Zheng 34425d4f98a2SYan Zheng return 0; 34435d4f98a2SYan Zheng } 34445d4f98a2SYan Zheng 34455d4f98a2SYan Zheng /* 34465d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 34475d4f98a2SYan Zheng */ 34485d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 34495d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 34505d4f98a2SYan Zheng struct rb_root *blocks) 34515d4f98a2SYan Zheng { 34520b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34535d4f98a2SYan Zheng struct btrfs_path *path; 34545d4f98a2SYan Zheng struct btrfs_key key; 34555d4f98a2SYan Zheng int ret; 34560b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 34575d4f98a2SYan Zheng 34587476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 34595d4f98a2SYan Zheng return 0; 34605d4f98a2SYan Zheng 34615d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 34625d4f98a2SYan Zheng return 0; 34635d4f98a2SYan Zheng 34645d4f98a2SYan Zheng path = btrfs_alloc_path(); 34655d4f98a2SYan Zheng if (!path) 34665d4f98a2SYan Zheng return -ENOMEM; 3467aee68ee5SJosef Bacik again: 34685d4f98a2SYan Zheng key.objectid = bytenr; 3469aee68ee5SJosef Bacik if (skinny) { 3470aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3471aee68ee5SJosef Bacik key.offset = (u64)-1; 3472aee68ee5SJosef Bacik } else { 34735d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34745d4f98a2SYan Zheng key.offset = blocksize; 3475aee68ee5SJosef Bacik } 34765d4f98a2SYan Zheng 34775d4f98a2SYan Zheng path->search_commit_root = 1; 34785d4f98a2SYan Zheng path->skip_locking = 1; 34795d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 34805d4f98a2SYan Zheng if (ret < 0) 34815d4f98a2SYan Zheng goto out; 34825d4f98a2SYan Zheng 3483aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3484aee68ee5SJosef Bacik if (path->slots[0]) { 3485aee68ee5SJosef Bacik path->slots[0]--; 3486aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3487aee68ee5SJosef Bacik path->slots[0]); 34883173a18fSJosef Bacik if (key.objectid == bytenr && 3489aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3490aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3491aee68ee5SJosef Bacik key.offset == blocksize))) 34923173a18fSJosef Bacik ret = 0; 34933173a18fSJosef Bacik } 3494aee68ee5SJosef Bacik 3495aee68ee5SJosef Bacik if (ret) { 3496aee68ee5SJosef Bacik skinny = false; 3497aee68ee5SJosef Bacik btrfs_release_path(path); 3498aee68ee5SJosef Bacik goto again; 3499aee68ee5SJosef Bacik } 3500aee68ee5SJosef Bacik } 3501cdccee99SLiu Bo if (ret) { 3502cdccee99SLiu Bo ASSERT(ret == 1); 3503cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3504cdccee99SLiu Bo btrfs_err(fs_info, 3505cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3506cdccee99SLiu Bo bytenr); 3507cdccee99SLiu Bo WARN_ON(1); 3508cdccee99SLiu Bo ret = -EINVAL; 3509cdccee99SLiu Bo goto out; 3510cdccee99SLiu Bo } 35113173a18fSJosef Bacik 35125d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 35135d4f98a2SYan Zheng out: 35145d4f98a2SYan Zheng btrfs_free_path(path); 35155d4f98a2SYan Zheng return ret; 35165d4f98a2SYan Zheng } 35175d4f98a2SYan Zheng 35185d4f98a2SYan Zheng /* 35195d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 35205d4f98a2SYan Zheng */ 35215d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 35225d4f98a2SYan Zheng struct extent_buffer *eb) 35235d4f98a2SYan Zheng { 35245d4f98a2SYan Zheng u64 flags; 35255d4f98a2SYan Zheng int ret; 35265d4f98a2SYan Zheng 35275d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 35285d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 35295d4f98a2SYan Zheng return 1; 35305d4f98a2SYan Zheng 35312ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info, 35323173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 35333173a18fSJosef Bacik NULL, &flags); 35345d4f98a2SYan Zheng BUG_ON(ret); 35355d4f98a2SYan Zheng 35365d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 35375d4f98a2SYan Zheng ret = 1; 35385d4f98a2SYan Zheng else 35395d4f98a2SYan Zheng ret = 0; 35405d4f98a2SYan Zheng return ret; 35415d4f98a2SYan Zheng } 35425d4f98a2SYan Zheng 35430af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 354432da5386SDavid Sterba struct btrfs_block_group *block_group, 35451bbc621eSChris Mason struct inode *inode, 35461bbc621eSChris Mason u64 ino) 35470af3d00bSJosef Bacik { 35480af3d00bSJosef Bacik struct btrfs_key key; 35490af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 35500af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 35510af3d00bSJosef Bacik int ret = 0; 35520af3d00bSJosef Bacik 35530af3d00bSJosef Bacik if (inode) 35540af3d00bSJosef Bacik goto truncate; 35550af3d00bSJosef Bacik 35560af3d00bSJosef Bacik key.objectid = ino; 35570af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 35580af3d00bSJosef Bacik key.offset = 0; 35590af3d00bSJosef Bacik 35604c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 35612e19f1f9SAl Viro if (IS_ERR(inode)) 35620af3d00bSJosef Bacik return -ENOENT; 35630af3d00bSJosef Bacik 35640af3d00bSJosef Bacik truncate: 35652ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 35667b61cd92SMiao Xie &fs_info->global_block_rsv); 35677b61cd92SMiao Xie if (ret) 35687b61cd92SMiao Xie goto out; 35697b61cd92SMiao Xie 35707a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 35710af3d00bSJosef Bacik if (IS_ERR(trans)) { 35723612b495STsutomu Itoh ret = PTR_ERR(trans); 35730af3d00bSJosef Bacik goto out; 35740af3d00bSJosef Bacik } 35750af3d00bSJosef Bacik 357677ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 35770af3d00bSJosef Bacik 35783a45bb20SJeff Mahoney btrfs_end_transaction(trans); 35792ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 35800af3d00bSJosef Bacik out: 35810af3d00bSJosef Bacik iput(inode); 35820af3d00bSJosef Bacik return ret; 35830af3d00bSJosef Bacik } 35840af3d00bSJosef Bacik 35855d4f98a2SYan Zheng /* 35865d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 35875d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 35885d4f98a2SYan Zheng */ 35895d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 35905d4f98a2SYan Zheng struct btrfs_key *extent_key, 35915d4f98a2SYan Zheng struct extent_buffer *leaf, 35925d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 35935d4f98a2SYan Zheng struct rb_root *blocks) 35945d4f98a2SYan Zheng { 35950b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35965d4f98a2SYan Zheng struct btrfs_path *path; 35975d4f98a2SYan Zheng struct tree_block *block; 35985d4f98a2SYan Zheng struct btrfs_root *root; 35995d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 36005d4f98a2SYan Zheng struct rb_node *rb_node; 36015d4f98a2SYan Zheng struct btrfs_key key; 36025d4f98a2SYan Zheng u64 ref_root; 36035d4f98a2SYan Zheng u64 ref_objectid; 36045d4f98a2SYan Zheng u64 ref_offset; 36055d4f98a2SYan Zheng u32 ref_count; 36065d4f98a2SYan Zheng u32 nritems; 36075d4f98a2SYan Zheng int err = 0; 36085d4f98a2SYan Zheng int added = 0; 36095d4f98a2SYan Zheng int counted; 36105d4f98a2SYan Zheng int ret; 36115d4f98a2SYan Zheng 36125d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 36135d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 36145d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 36155d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 36165d4f98a2SYan Zheng 36170af3d00bSJosef Bacik /* 36180af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 36190af3d00bSJosef Bacik * it and redo the search. 36200af3d00bSJosef Bacik */ 36210af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 36220b246afaSJeff Mahoney ret = delete_block_group_cache(fs_info, rc->block_group, 36230af3d00bSJosef Bacik NULL, ref_objectid); 36240af3d00bSJosef Bacik if (ret != -ENOENT) 36250af3d00bSJosef Bacik return ret; 36260af3d00bSJosef Bacik ret = 0; 36270af3d00bSJosef Bacik } 36280af3d00bSJosef Bacik 36290af3d00bSJosef Bacik path = btrfs_alloc_path(); 36300af3d00bSJosef Bacik if (!path) 36310af3d00bSJosef Bacik return -ENOMEM; 3632e4058b54SDavid Sterba path->reada = READA_FORWARD; 36330af3d00bSJosef Bacik 36340b246afaSJeff Mahoney root = read_fs_root(fs_info, ref_root); 36355d4f98a2SYan Zheng if (IS_ERR(root)) { 36365d4f98a2SYan Zheng err = PTR_ERR(root); 36375d4f98a2SYan Zheng goto out; 36385d4f98a2SYan Zheng } 36395d4f98a2SYan Zheng 36405d4f98a2SYan Zheng key.objectid = ref_objectid; 36415d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 364284850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 364384850e8dSYan, Zheng key.offset = 0; 364484850e8dSYan, Zheng else 364584850e8dSYan, Zheng key.offset = ref_offset; 36465d4f98a2SYan Zheng 36475d4f98a2SYan Zheng path->search_commit_root = 1; 36485d4f98a2SYan Zheng path->skip_locking = 1; 36495d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 36505d4f98a2SYan Zheng if (ret < 0) { 36515d4f98a2SYan Zheng err = ret; 36525d4f98a2SYan Zheng goto out; 36535d4f98a2SYan Zheng } 36545d4f98a2SYan Zheng 36555d4f98a2SYan Zheng leaf = path->nodes[0]; 36565d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36575d4f98a2SYan Zheng /* 36585d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 36595d4f98a2SYan Zheng * are not counted in 36605d4f98a2SYan Zheng */ 36615d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36625d4f98a2SYan Zheng counted = 0; 36635d4f98a2SYan Zheng else 36645d4f98a2SYan Zheng counted = 1; 36655d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36665d4f98a2SYan Zheng if (rb_node) { 36675d4f98a2SYan Zheng if (counted) 36685d4f98a2SYan Zheng added = 1; 36695d4f98a2SYan Zheng else 36705d4f98a2SYan Zheng path->slots[0] = nritems; 36715d4f98a2SYan Zheng } 36725d4f98a2SYan Zheng 36735d4f98a2SYan Zheng while (ref_count > 0) { 36745d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 36755d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 36765d4f98a2SYan Zheng if (ret < 0) { 36775d4f98a2SYan Zheng err = ret; 36785d4f98a2SYan Zheng goto out; 36795d4f98a2SYan Zheng } 3680fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 36815d4f98a2SYan Zheng goto out; 36825d4f98a2SYan Zheng 36835d4f98a2SYan Zheng leaf = path->nodes[0]; 36845d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36855d4f98a2SYan Zheng added = 0; 36865d4f98a2SYan Zheng 36875d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36885d4f98a2SYan Zheng counted = 0; 36895d4f98a2SYan Zheng else 36905d4f98a2SYan Zheng counted = 1; 36915d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36925d4f98a2SYan Zheng if (rb_node) { 36935d4f98a2SYan Zheng if (counted) 36945d4f98a2SYan Zheng added = 1; 36955d4f98a2SYan Zheng else 36965d4f98a2SYan Zheng path->slots[0] = nritems; 36975d4f98a2SYan Zheng } 36985d4f98a2SYan Zheng } 36995d4f98a2SYan Zheng 37005d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3701fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3702fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 37035d4f98a2SYan Zheng break; 37045d4f98a2SYan Zheng 37055d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 37065d4f98a2SYan Zheng struct btrfs_file_extent_item); 37075d4f98a2SYan Zheng 37085d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 37095d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 37105d4f98a2SYan Zheng goto next; 37115d4f98a2SYan Zheng 37125d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 37135d4f98a2SYan Zheng extent_key->objectid) 37145d4f98a2SYan Zheng goto next; 37155d4f98a2SYan Zheng 37165d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 37175d4f98a2SYan Zheng if (key.offset != ref_offset) 37185d4f98a2SYan Zheng goto next; 37195d4f98a2SYan Zheng 37205d4f98a2SYan Zheng if (counted) 37215d4f98a2SYan Zheng ref_count--; 37225d4f98a2SYan Zheng if (added) 37235d4f98a2SYan Zheng goto next; 37245d4f98a2SYan Zheng 37257476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 37265d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 37275d4f98a2SYan Zheng if (!block) { 37285d4f98a2SYan Zheng err = -ENOMEM; 37295d4f98a2SYan Zheng break; 37305d4f98a2SYan Zheng } 37315d4f98a2SYan Zheng block->bytenr = leaf->start; 37325d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 37335d4f98a2SYan Zheng block->level = 0; 37345d4f98a2SYan Zheng block->key_ready = 1; 37355d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 37365d4f98a2SYan Zheng &block->rb_node); 373743c04fb1SJeff Mahoney if (rb_node) 373843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 373943c04fb1SJeff Mahoney block->bytenr); 37405d4f98a2SYan Zheng } 37415d4f98a2SYan Zheng if (counted) 37425d4f98a2SYan Zheng added = 1; 37435d4f98a2SYan Zheng else 37445d4f98a2SYan Zheng path->slots[0] = nritems; 37455d4f98a2SYan Zheng next: 37465d4f98a2SYan Zheng path->slots[0]++; 37475d4f98a2SYan Zheng 37485d4f98a2SYan Zheng } 37495d4f98a2SYan Zheng out: 37505d4f98a2SYan Zheng btrfs_free_path(path); 37515d4f98a2SYan Zheng return err; 37525d4f98a2SYan Zheng } 37535d4f98a2SYan Zheng 37545d4f98a2SYan Zheng /* 37552c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37565d4f98a2SYan Zheng */ 37575d4f98a2SYan Zheng static noinline_for_stack 37585d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 37595d4f98a2SYan Zheng struct btrfs_key *extent_key, 37605d4f98a2SYan Zheng struct btrfs_path *path, 37615d4f98a2SYan Zheng struct rb_root *blocks) 37625d4f98a2SYan Zheng { 37635d4f98a2SYan Zheng struct btrfs_key key; 37645d4f98a2SYan Zheng struct extent_buffer *eb; 37655d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 37665d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 37675d4f98a2SYan Zheng unsigned long ptr; 37685d4f98a2SYan Zheng unsigned long end; 3769da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 3770647f63bdSFilipe David Borba Manana int ret = 0; 37715d4f98a2SYan Zheng int err = 0; 37725d4f98a2SYan Zheng 37735d4f98a2SYan Zheng eb = path->nodes[0]; 37745d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 37755d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 37765d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 37775d4f98a2SYan Zheng 37785d4f98a2SYan Zheng while (ptr < end) { 37795d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 37803de28d57SLiu Bo key.type = btrfs_get_extent_inline_ref_type(eb, iref, 37813de28d57SLiu Bo BTRFS_REF_TYPE_DATA); 37825d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37835d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 37845d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37855d4f98a2SYan Zheng blocks); 37865d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37875d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 37885d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37895d4f98a2SYan Zheng eb, dref, blocks); 37905d4f98a2SYan Zheng } else { 3791af431dcbSSu Yue ret = -EUCLEAN; 3792b14c55a1SLiu Bo btrfs_err(rc->extent_root->fs_info, 3793b14c55a1SLiu Bo "extent %llu slot %d has an invalid inline ref type", 3794b14c55a1SLiu Bo eb->start, path->slots[0]); 37955d4f98a2SYan Zheng } 3796647f63bdSFilipe David Borba Manana if (ret) { 3797647f63bdSFilipe David Borba Manana err = ret; 3798647f63bdSFilipe David Borba Manana goto out; 3799647f63bdSFilipe David Borba Manana } 38005d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 38015d4f98a2SYan Zheng } 38025d4f98a2SYan Zheng WARN_ON(ptr > end); 38035d4f98a2SYan Zheng 38045d4f98a2SYan Zheng while (1) { 38055d4f98a2SYan Zheng cond_resched(); 38065d4f98a2SYan Zheng eb = path->nodes[0]; 38075d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 38085d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38095d4f98a2SYan Zheng if (ret < 0) { 38105d4f98a2SYan Zheng err = ret; 38115d4f98a2SYan Zheng break; 38125d4f98a2SYan Zheng } 38135d4f98a2SYan Zheng if (ret > 0) 38145d4f98a2SYan Zheng break; 38155d4f98a2SYan Zheng eb = path->nodes[0]; 38165d4f98a2SYan Zheng } 38175d4f98a2SYan Zheng 38185d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 38195d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 38205d4f98a2SYan Zheng break; 38215d4f98a2SYan Zheng 38225d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 38235d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 38245d4f98a2SYan Zheng blocks); 38255d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 38265d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 38275d4f98a2SYan Zheng struct btrfs_extent_data_ref); 38285d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 38295d4f98a2SYan Zheng eb, dref, blocks); 38306d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 3831ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3832ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3833ba3c2b19SNikolay Borisov ret = -EINVAL; 38345d4f98a2SYan Zheng } else { 38355d4f98a2SYan Zheng ret = 0; 38365d4f98a2SYan Zheng } 38375d4f98a2SYan Zheng if (ret) { 38385d4f98a2SYan Zheng err = ret; 38395d4f98a2SYan Zheng break; 38405d4f98a2SYan Zheng } 38415d4f98a2SYan Zheng path->slots[0]++; 38425d4f98a2SYan Zheng } 3843647f63bdSFilipe David Borba Manana out: 3844b3b4aa74SDavid Sterba btrfs_release_path(path); 38455d4f98a2SYan Zheng if (err) 38465d4f98a2SYan Zheng free_block_list(blocks); 38475d4f98a2SYan Zheng return err; 38485d4f98a2SYan Zheng } 38495d4f98a2SYan Zheng 38505d4f98a2SYan Zheng /* 38512c016dc2SLiu Bo * helper to find next unprocessed extent 38525d4f98a2SYan Zheng */ 38535d4f98a2SYan Zheng static noinline_for_stack 3854147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 38553fd0a558SYan, Zheng struct btrfs_key *extent_key) 38565d4f98a2SYan Zheng { 38570b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38585d4f98a2SYan Zheng struct btrfs_key key; 38595d4f98a2SYan Zheng struct extent_buffer *leaf; 38605d4f98a2SYan Zheng u64 start, end, last; 38615d4f98a2SYan Zheng int ret; 38625d4f98a2SYan Zheng 3863b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 38645d4f98a2SYan Zheng while (1) { 38655d4f98a2SYan Zheng cond_resched(); 38665d4f98a2SYan Zheng if (rc->search_start >= last) { 38675d4f98a2SYan Zheng ret = 1; 38685d4f98a2SYan Zheng break; 38695d4f98a2SYan Zheng } 38705d4f98a2SYan Zheng 38715d4f98a2SYan Zheng key.objectid = rc->search_start; 38725d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 38735d4f98a2SYan Zheng key.offset = 0; 38745d4f98a2SYan Zheng 38755d4f98a2SYan Zheng path->search_commit_root = 1; 38765d4f98a2SYan Zheng path->skip_locking = 1; 38775d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 38785d4f98a2SYan Zheng 0, 0); 38795d4f98a2SYan Zheng if (ret < 0) 38805d4f98a2SYan Zheng break; 38815d4f98a2SYan Zheng next: 38825d4f98a2SYan Zheng leaf = path->nodes[0]; 38835d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38845d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38855d4f98a2SYan Zheng if (ret != 0) 38865d4f98a2SYan Zheng break; 38875d4f98a2SYan Zheng leaf = path->nodes[0]; 38885d4f98a2SYan Zheng } 38895d4f98a2SYan Zheng 38905d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38915d4f98a2SYan Zheng if (key.objectid >= last) { 38925d4f98a2SYan Zheng ret = 1; 38935d4f98a2SYan Zheng break; 38945d4f98a2SYan Zheng } 38955d4f98a2SYan Zheng 38963173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38973173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38983173a18fSJosef Bacik path->slots[0]++; 38993173a18fSJosef Bacik goto next; 39003173a18fSJosef Bacik } 39013173a18fSJosef Bacik 39023173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 39035d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 39045d4f98a2SYan Zheng path->slots[0]++; 39055d4f98a2SYan Zheng goto next; 39065d4f98a2SYan Zheng } 39075d4f98a2SYan Zheng 39083173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 39090b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 39103173a18fSJosef Bacik rc->search_start) { 39113173a18fSJosef Bacik path->slots[0]++; 39123173a18fSJosef Bacik goto next; 39133173a18fSJosef Bacik } 39143173a18fSJosef Bacik 39155d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 39165d4f98a2SYan Zheng key.objectid, &start, &end, 3917e6138876SJosef Bacik EXTENT_DIRTY, NULL); 39185d4f98a2SYan Zheng 39195d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3920b3b4aa74SDavid Sterba btrfs_release_path(path); 39215d4f98a2SYan Zheng rc->search_start = end + 1; 39225d4f98a2SYan Zheng } else { 39233173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 39245d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 39253173a18fSJosef Bacik else 39263173a18fSJosef Bacik rc->search_start = key.objectid + 39270b246afaSJeff Mahoney fs_info->nodesize; 39283fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 39295d4f98a2SYan Zheng return 0; 39305d4f98a2SYan Zheng } 39315d4f98a2SYan Zheng } 3932b3b4aa74SDavid Sterba btrfs_release_path(path); 39335d4f98a2SYan Zheng return ret; 39345d4f98a2SYan Zheng } 39355d4f98a2SYan Zheng 39365d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 39375d4f98a2SYan Zheng { 39385d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39397585717fSChris Mason 39407585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39415d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 39427585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39435d4f98a2SYan Zheng } 39445d4f98a2SYan Zheng 39455d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 39465d4f98a2SYan Zheng { 39475d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39487585717fSChris Mason 39497585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39505d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 39517585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39525d4f98a2SYan Zheng } 39535d4f98a2SYan Zheng 39545d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 39555d4f98a2SYan Zheng { 39565d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39575d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39585d4f98a2SYan Zheng return 1; 39595d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 39605d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39615d4f98a2SYan Zheng return 1; 39625d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39635d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 39645d4f98a2SYan Zheng return 1; 39655d4f98a2SYan Zheng return 0; 39665d4f98a2SYan Zheng } 39675d4f98a2SYan Zheng 39683fd0a558SYan, Zheng static noinline_for_stack 39693fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 39703fd0a558SYan, Zheng { 39713fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3972ac2fabacSJosef Bacik int ret; 39733fd0a558SYan, Zheng 39742ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 397566d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 39763fd0a558SYan, Zheng if (!rc->block_rsv) 39773fd0a558SYan, Zheng return -ENOMEM; 39783fd0a558SYan, Zheng 39793fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3980b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 39813fd0a558SYan, Zheng rc->extents_found = 0; 39823fd0a558SYan, Zheng rc->nodes_relocated = 0; 39833fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39840647bf56SWang Shilong rc->reserved_bytes = 0; 3985da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 39860647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3987ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3988ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3989ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3990ac2fabacSJosef Bacik if (ret) 3991ac2fabacSJosef Bacik return ret; 39923fd0a558SYan, Zheng 39933fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39943fd0a558SYan, Zheng set_reloc_control(rc); 39953fd0a558SYan, Zheng 39967a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 399728818947SLiu Bo if (IS_ERR(trans)) { 399828818947SLiu Bo unset_reloc_control(rc); 399928818947SLiu Bo /* 400028818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 400128818947SLiu Bo * cleanup. And callers are responsible to free the above 400228818947SLiu Bo * block rsv. 400328818947SLiu Bo */ 400428818947SLiu Bo return PTR_ERR(trans); 400528818947SLiu Bo } 40063a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 40073fd0a558SYan, Zheng return 0; 40083fd0a558SYan, Zheng } 400976dda93cSYan, Zheng 40105d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 40115d4f98a2SYan Zheng { 40122ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40135d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 40145d4f98a2SYan Zheng struct btrfs_key key; 40155d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 40165d4f98a2SYan Zheng struct btrfs_path *path; 40175d4f98a2SYan Zheng struct btrfs_extent_item *ei; 40185d4f98a2SYan Zheng u64 flags; 40195d4f98a2SYan Zheng u32 item_size; 40205d4f98a2SYan Zheng int ret; 40215d4f98a2SYan Zheng int err = 0; 4022c87f08caSChris Mason int progress = 0; 40235d4f98a2SYan Zheng 40245d4f98a2SYan Zheng path = btrfs_alloc_path(); 40253fd0a558SYan, Zheng if (!path) 40265d4f98a2SYan Zheng return -ENOMEM; 4027e4058b54SDavid Sterba path->reada = READA_FORWARD; 40283fd0a558SYan, Zheng 40293fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 40303fd0a558SYan, Zheng if (ret) { 40313fd0a558SYan, Zheng err = ret; 40323fd0a558SYan, Zheng goto out_free; 40332423fdfbSJiri Slaby } 40345d4f98a2SYan Zheng 40355d4f98a2SYan Zheng while (1) { 40360647bf56SWang Shilong rc->reserved_bytes = 0; 40370647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40380647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40390647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40400647bf56SWang Shilong if (ret) { 40410647bf56SWang Shilong err = ret; 40420647bf56SWang Shilong break; 40430647bf56SWang Shilong } 4044c87f08caSChris Mason progress++; 4045a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40460f788c58SLiu Bo if (IS_ERR(trans)) { 40470f788c58SLiu Bo err = PTR_ERR(trans); 40480f788c58SLiu Bo trans = NULL; 40490f788c58SLiu Bo break; 40500f788c58SLiu Bo } 4051c87f08caSChris Mason restart: 40523fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40533a45bb20SJeff Mahoney btrfs_end_transaction(trans); 405442a657f5SPan Bian trans = NULL; 40553fd0a558SYan, Zheng continue; 40563fd0a558SYan, Zheng } 40573fd0a558SYan, Zheng 4058147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40595d4f98a2SYan Zheng if (ret < 0) 40605d4f98a2SYan Zheng err = ret; 40615d4f98a2SYan Zheng if (ret != 0) 40625d4f98a2SYan Zheng break; 40635d4f98a2SYan Zheng 40645d4f98a2SYan Zheng rc->extents_found++; 40655d4f98a2SYan Zheng 40665d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40675d4f98a2SYan Zheng struct btrfs_extent_item); 40683fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40695d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40705d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40715d4f98a2SYan Zheng ret = check_extent_flags(flags); 40725d4f98a2SYan Zheng BUG_ON(ret); 40736d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 4074ba3c2b19SNikolay Borisov err = -EINVAL; 4075ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 4076ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 4077ba3c2b19SNikolay Borisov break; 40785d4f98a2SYan Zheng } else { 40795d4f98a2SYan Zheng BUG(); 40805d4f98a2SYan Zheng } 40815d4f98a2SYan Zheng 40825d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 40835d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 40845d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 40855d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40865d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 40875d4f98a2SYan Zheng } else { 4088b3b4aa74SDavid Sterba btrfs_release_path(path); 40895d4f98a2SYan Zheng ret = 0; 40905d4f98a2SYan Zheng } 40915d4f98a2SYan Zheng if (ret < 0) { 40923fd0a558SYan, Zheng err = ret; 40935d4f98a2SYan Zheng break; 40945d4f98a2SYan Zheng } 40955d4f98a2SYan Zheng 40965d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 40975d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 40985d4f98a2SYan Zheng if (ret < 0) { 40991708cc57SWang Shilong /* 41001708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 41011708cc57SWang Shilong * backref cache when committing transaction. 41021708cc57SWang Shilong */ 41031708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 41041708cc57SWang Shilong 41053fd0a558SYan, Zheng if (ret != -EAGAIN) { 41065d4f98a2SYan Zheng err = ret; 41075d4f98a2SYan Zheng break; 41085d4f98a2SYan Zheng } 41093fd0a558SYan, Zheng rc->extents_found--; 41103fd0a558SYan, Zheng rc->search_start = key.objectid; 41113fd0a558SYan, Zheng } 41125d4f98a2SYan Zheng } 41135d4f98a2SYan Zheng 41143a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41152ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41163fd0a558SYan, Zheng trans = NULL; 41175d4f98a2SYan Zheng 41185d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 41195d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41205d4f98a2SYan Zheng rc->found_file_extent = 1; 41210257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 41223fd0a558SYan, Zheng &key, &rc->cluster); 41235d4f98a2SYan Zheng if (ret < 0) { 41245d4f98a2SYan Zheng err = ret; 41255d4f98a2SYan Zheng break; 41265d4f98a2SYan Zheng } 41275d4f98a2SYan Zheng } 41285d4f98a2SYan Zheng } 4129c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 413043a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 41319689457bSShilong Wang if (ret == 1) { 4132c87f08caSChris Mason err = 0; 4133c87f08caSChris Mason progress = 0; 4134c87f08caSChris Mason goto restart; 4135c87f08caSChris Mason } 4136c87f08caSChris Mason } 41373fd0a558SYan, Zheng 4138b3b4aa74SDavid Sterba btrfs_release_path(path); 413991166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41405d4f98a2SYan Zheng 41415d4f98a2SYan Zheng if (trans) { 41423a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41432ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41445d4f98a2SYan Zheng } 41455d4f98a2SYan Zheng 41460257bb82SYan, Zheng if (!err) { 41473fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 41483fd0a558SYan, Zheng &rc->cluster); 41490257bb82SYan, Zheng if (ret < 0) 41500257bb82SYan, Zheng err = ret; 41510257bb82SYan, Zheng } 41520257bb82SYan, Zheng 41533fd0a558SYan, Zheng rc->create_reloc_tree = 0; 41543fd0a558SYan, Zheng set_reloc_control(rc); 41550257bb82SYan, Zheng 41563fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 41572ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41585d4f98a2SYan Zheng 41593fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 41605d4f98a2SYan Zheng 41615d4f98a2SYan Zheng merge_reloc_roots(rc); 41625d4f98a2SYan Zheng 41633fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 41645d4f98a2SYan Zheng unset_reloc_control(rc); 41652ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41665d4f98a2SYan Zheng 41675d4f98a2SYan Zheng /* get rid of pinned extents */ 41687a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 416962b99540SQu Wenruo if (IS_ERR(trans)) { 41703612b495STsutomu Itoh err = PTR_ERR(trans); 417162b99540SQu Wenruo goto out_free; 417262b99540SQu Wenruo } 41733a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 4174d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4175d2311e69SQu Wenruo if (ret < 0 && !err) 4176d2311e69SQu Wenruo err = ret; 41773fd0a558SYan, Zheng out_free: 41782ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 41793fd0a558SYan, Zheng btrfs_free_path(path); 41805d4f98a2SYan Zheng return err; 41815d4f98a2SYan Zheng } 41825d4f98a2SYan Zheng 41835d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 41840257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 41855d4f98a2SYan Zheng { 41865d4f98a2SYan Zheng struct btrfs_path *path; 41875d4f98a2SYan Zheng struct btrfs_inode_item *item; 41885d4f98a2SYan Zheng struct extent_buffer *leaf; 41895d4f98a2SYan Zheng int ret; 41905d4f98a2SYan Zheng 41915d4f98a2SYan Zheng path = btrfs_alloc_path(); 41925d4f98a2SYan Zheng if (!path) 41935d4f98a2SYan Zheng return -ENOMEM; 41945d4f98a2SYan Zheng 41955d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 41965d4f98a2SYan Zheng if (ret) 41975d4f98a2SYan Zheng goto out; 41985d4f98a2SYan Zheng 41995d4f98a2SYan Zheng leaf = path->nodes[0]; 42005d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4201b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 42025d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 42030257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 42045d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 42053fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 42063fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 42075d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 42085d4f98a2SYan Zheng out: 42095d4f98a2SYan Zheng btrfs_free_path(path); 42105d4f98a2SYan Zheng return ret; 42115d4f98a2SYan Zheng } 42125d4f98a2SYan Zheng 42135d4f98a2SYan Zheng /* 42145d4f98a2SYan Zheng * helper to create inode for data relocation. 42155d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 42165d4f98a2SYan Zheng */ 42173fd0a558SYan, Zheng static noinline_for_stack 42183fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 421932da5386SDavid Sterba struct btrfs_block_group *group) 42205d4f98a2SYan Zheng { 42215d4f98a2SYan Zheng struct inode *inode = NULL; 42225d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42235d4f98a2SYan Zheng struct btrfs_root *root; 42245d4f98a2SYan Zheng struct btrfs_key key; 42254624900dSZhaolei u64 objectid; 42265d4f98a2SYan Zheng int err = 0; 42275d4f98a2SYan Zheng 42285d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 42295d4f98a2SYan Zheng if (IS_ERR(root)) 42305d4f98a2SYan Zheng return ERR_CAST(root); 42315d4f98a2SYan Zheng 4232a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 42333fd0a558SYan, Zheng if (IS_ERR(trans)) 42343fd0a558SYan, Zheng return ERR_CAST(trans); 42355d4f98a2SYan Zheng 4236581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42375d4f98a2SYan Zheng if (err) 42385d4f98a2SYan Zheng goto out; 42395d4f98a2SYan Zheng 42400257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42415d4f98a2SYan Zheng BUG_ON(err); 42425d4f98a2SYan Zheng 42435d4f98a2SYan Zheng key.objectid = objectid; 42445d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 42455d4f98a2SYan Zheng key.offset = 0; 42464c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 42472e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 4248b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 42495d4f98a2SYan Zheng 425073f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 42515d4f98a2SYan Zheng out: 42523a45bb20SJeff Mahoney btrfs_end_transaction(trans); 42532ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42545d4f98a2SYan Zheng if (err) { 42555d4f98a2SYan Zheng if (inode) 42565d4f98a2SYan Zheng iput(inode); 42575d4f98a2SYan Zheng inode = ERR_PTR(err); 42585d4f98a2SYan Zheng } 42595d4f98a2SYan Zheng return inode; 42605d4f98a2SYan Zheng } 42615d4f98a2SYan Zheng 4262c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 42633fd0a558SYan, Zheng { 42643fd0a558SYan, Zheng struct reloc_control *rc; 42653fd0a558SYan, Zheng 42663fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 42673fd0a558SYan, Zheng if (!rc) 42683fd0a558SYan, Zheng return NULL; 42693fd0a558SYan, Zheng 42703fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4271d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 42723fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 42733fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 427443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 427543eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 42763fd0a558SYan, Zheng return rc; 42773fd0a558SYan, Zheng } 42783fd0a558SYan, Zheng 42795d4f98a2SYan Zheng /* 4280ebce0e01SAdam Borowski * Print the block group being relocated 4281ebce0e01SAdam Borowski */ 4282ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 428332da5386SDavid Sterba struct btrfs_block_group *block_group) 4284ebce0e01SAdam Borowski { 4285f89e09cfSAnand Jain char buf[128] = {'\0'}; 4286ebce0e01SAdam Borowski 4287f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4288ebce0e01SAdam Borowski 4289ebce0e01SAdam Borowski btrfs_info(fs_info, 4290ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4291b3470b5dSDavid Sterba block_group->start, buf); 4292ebce0e01SAdam Borowski } 4293ebce0e01SAdam Borowski 4294ebce0e01SAdam Borowski /* 42955d4f98a2SYan Zheng * function to relocate all extents in a block group. 42965d4f98a2SYan Zheng */ 42976bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 42985d4f98a2SYan Zheng { 429932da5386SDavid Sterba struct btrfs_block_group *bg; 43006bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 43015d4f98a2SYan Zheng struct reloc_control *rc; 43020af3d00bSJosef Bacik struct inode *inode; 43030af3d00bSJosef Bacik struct btrfs_path *path; 43045d4f98a2SYan Zheng int ret; 4305f0486c68SYan, Zheng int rw = 0; 43065d4f98a2SYan Zheng int err = 0; 43075d4f98a2SYan Zheng 4308eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4309eede2bf3SOmar Sandoval if (!bg) 4310eede2bf3SOmar Sandoval return -ENOENT; 4311eede2bf3SOmar Sandoval 4312eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4313eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4314eede2bf3SOmar Sandoval return -ETXTBSY; 4315eede2bf3SOmar Sandoval } 4316eede2bf3SOmar Sandoval 4317c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4318eede2bf3SOmar Sandoval if (!rc) { 4319eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 43205d4f98a2SYan Zheng return -ENOMEM; 4321eede2bf3SOmar Sandoval } 43225d4f98a2SYan Zheng 4323f0486c68SYan, Zheng rc->extent_root = extent_root; 4324eede2bf3SOmar Sandoval rc->block_group = bg; 43255d4f98a2SYan Zheng 4326b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4327f0486c68SYan, Zheng if (ret) { 4328f0486c68SYan, Zheng err = ret; 4329f0486c68SYan, Zheng goto out; 4330f0486c68SYan, Zheng } 4331f0486c68SYan, Zheng rw = 1; 4332f0486c68SYan, Zheng 43330af3d00bSJosef Bacik path = btrfs_alloc_path(); 43340af3d00bSJosef Bacik if (!path) { 43350af3d00bSJosef Bacik err = -ENOMEM; 43360af3d00bSJosef Bacik goto out; 43370af3d00bSJosef Bacik } 43380af3d00bSJosef Bacik 43397949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 43400af3d00bSJosef Bacik btrfs_free_path(path); 43410af3d00bSJosef Bacik 43420af3d00bSJosef Bacik if (!IS_ERR(inode)) 43431bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43440af3d00bSJosef Bacik else 43450af3d00bSJosef Bacik ret = PTR_ERR(inode); 43460af3d00bSJosef Bacik 43470af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 43480af3d00bSJosef Bacik err = ret; 43490af3d00bSJosef Bacik goto out; 43500af3d00bSJosef Bacik } 43510af3d00bSJosef Bacik 43525d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 43535d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 43545d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 43555d4f98a2SYan Zheng rc->data_inode = NULL; 43565d4f98a2SYan Zheng goto out; 43575d4f98a2SYan Zheng } 43585d4f98a2SYan Zheng 43590b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 43605d4f98a2SYan Zheng 43619cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4362f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 43636374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4364b3470b5dSDavid Sterba rc->block_group->start, 4365b3470b5dSDavid Sterba rc->block_group->length); 43665d4f98a2SYan Zheng 43675d4f98a2SYan Zheng while (1) { 436876dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 43695d4f98a2SYan Zheng ret = relocate_block_group(rc); 437076dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4371ff612ba7SJosef Bacik if (ret < 0) 43725d4f98a2SYan Zheng err = ret; 4373ff612ba7SJosef Bacik 4374ff612ba7SJosef Bacik /* 4375ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4376ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4377ff612ba7SJosef Bacik * different block group we could end up hitting the 4378ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4379ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4380ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4381ff612ba7SJosef Bacik * out of the loop if we hit an error. 4382ff612ba7SJosef Bacik */ 4383ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4384ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4385ff612ba7SJosef Bacik (u64)-1); 4386ff612ba7SJosef Bacik if (ret) 4387ff612ba7SJosef Bacik err = ret; 4388ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4389ff612ba7SJosef Bacik 0, -1); 4390ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 43915d4f98a2SYan Zheng } 43925d4f98a2SYan Zheng 4393ff612ba7SJosef Bacik if (err < 0) 4394ff612ba7SJosef Bacik goto out; 4395ff612ba7SJosef Bacik 43965d4f98a2SYan Zheng if (rc->extents_found == 0) 43975d4f98a2SYan Zheng break; 43985d4f98a2SYan Zheng 43990b246afaSJeff Mahoney btrfs_info(fs_info, "found %llu extents", rc->extents_found); 44005d4f98a2SYan Zheng 44015d4f98a2SYan Zheng } 44025d4f98a2SYan Zheng 44035d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 44045d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4405bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 44065d4f98a2SYan Zheng out: 4407f0486c68SYan, Zheng if (err && rw) 44082ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 44095d4f98a2SYan Zheng iput(rc->data_inode); 44105d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 44115d4f98a2SYan Zheng kfree(rc); 44125d4f98a2SYan Zheng return err; 44135d4f98a2SYan Zheng } 44145d4f98a2SYan Zheng 441576dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 441676dda93cSYan, Zheng { 44170b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 441876dda93cSYan, Zheng struct btrfs_trans_handle *trans; 441979787eaaSJeff Mahoney int ret, err; 442076dda93cSYan, Zheng 44210b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 442279787eaaSJeff Mahoney if (IS_ERR(trans)) 442379787eaaSJeff Mahoney return PTR_ERR(trans); 442476dda93cSYan, Zheng 442576dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 442676dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 442776dda93cSYan, Zheng root->root_item.drop_level = 0; 442876dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 44290b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 443076dda93cSYan, Zheng &root->root_key, &root->root_item); 443176dda93cSYan, Zheng 44323a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 443379787eaaSJeff Mahoney if (err) 443479787eaaSJeff Mahoney return err; 443579787eaaSJeff Mahoney return ret; 443676dda93cSYan, Zheng } 443776dda93cSYan, Zheng 44385d4f98a2SYan Zheng /* 44395d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44405d4f98a2SYan Zheng * 44415d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44425d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44435d4f98a2SYan Zheng */ 44445d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44455d4f98a2SYan Zheng { 44460b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 44475d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 44485d4f98a2SYan Zheng struct btrfs_key key; 44495d4f98a2SYan Zheng struct btrfs_root *fs_root; 44505d4f98a2SYan Zheng struct btrfs_root *reloc_root; 44515d4f98a2SYan Zheng struct btrfs_path *path; 44525d4f98a2SYan Zheng struct extent_buffer *leaf; 44535d4f98a2SYan Zheng struct reloc_control *rc = NULL; 44545d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 44555d4f98a2SYan Zheng int ret; 44565d4f98a2SYan Zheng int err = 0; 44575d4f98a2SYan Zheng 44585d4f98a2SYan Zheng path = btrfs_alloc_path(); 44595d4f98a2SYan Zheng if (!path) 44605d4f98a2SYan Zheng return -ENOMEM; 4461e4058b54SDavid Sterba path->reada = READA_BACK; 44625d4f98a2SYan Zheng 44635d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 44645d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 44655d4f98a2SYan Zheng key.offset = (u64)-1; 44665d4f98a2SYan Zheng 44675d4f98a2SYan Zheng while (1) { 44680b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 44695d4f98a2SYan Zheng path, 0, 0); 44705d4f98a2SYan Zheng if (ret < 0) { 44715d4f98a2SYan Zheng err = ret; 44725d4f98a2SYan Zheng goto out; 44735d4f98a2SYan Zheng } 44745d4f98a2SYan Zheng if (ret > 0) { 44755d4f98a2SYan Zheng if (path->slots[0] == 0) 44765d4f98a2SYan Zheng break; 44775d4f98a2SYan Zheng path->slots[0]--; 44785d4f98a2SYan Zheng } 44795d4f98a2SYan Zheng leaf = path->nodes[0]; 44805d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4481b3b4aa74SDavid Sterba btrfs_release_path(path); 44825d4f98a2SYan Zheng 44835d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 44845d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 44855d4f98a2SYan Zheng break; 44865d4f98a2SYan Zheng 4487cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 44885d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 44895d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 44905d4f98a2SYan Zheng goto out; 44915d4f98a2SYan Zheng } 44925d4f98a2SYan Zheng 44935d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 44945d4f98a2SYan Zheng 44955d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 44960b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 44975d4f98a2SYan Zheng reloc_root->root_key.offset); 44985d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 449976dda93cSYan, Zheng ret = PTR_ERR(fs_root); 450076dda93cSYan, Zheng if (ret != -ENOENT) { 450176dda93cSYan, Zheng err = ret; 45025d4f98a2SYan Zheng goto out; 45035d4f98a2SYan Zheng } 450479787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 450579787eaaSJeff Mahoney if (ret < 0) { 450679787eaaSJeff Mahoney err = ret; 450779787eaaSJeff Mahoney goto out; 450879787eaaSJeff Mahoney } 450976dda93cSYan, Zheng } 45105d4f98a2SYan Zheng } 45115d4f98a2SYan Zheng 45125d4f98a2SYan Zheng if (key.offset == 0) 45135d4f98a2SYan Zheng break; 45145d4f98a2SYan Zheng 45155d4f98a2SYan Zheng key.offset--; 45165d4f98a2SYan Zheng } 4517b3b4aa74SDavid Sterba btrfs_release_path(path); 45185d4f98a2SYan Zheng 45195d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 45205d4f98a2SYan Zheng goto out; 45215d4f98a2SYan Zheng 4522c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 45235d4f98a2SYan Zheng if (!rc) { 45245d4f98a2SYan Zheng err = -ENOMEM; 45255d4f98a2SYan Zheng goto out; 45265d4f98a2SYan Zheng } 45275d4f98a2SYan Zheng 45280b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 45295d4f98a2SYan Zheng 45305d4f98a2SYan Zheng set_reloc_control(rc); 45315d4f98a2SYan Zheng 45327a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45333612b495STsutomu Itoh if (IS_ERR(trans)) { 45343612b495STsutomu Itoh unset_reloc_control(rc); 45353612b495STsutomu Itoh err = PTR_ERR(trans); 45363612b495STsutomu Itoh goto out_free; 45373612b495STsutomu Itoh } 45383fd0a558SYan, Zheng 45393fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45403fd0a558SYan, Zheng 45415d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45425d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45435d4f98a2SYan Zheng struct btrfs_root, root_list); 45445d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45455d4f98a2SYan Zheng 45465d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 45475d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 45485d4f98a2SYan Zheng &rc->reloc_roots); 45495d4f98a2SYan Zheng continue; 45505d4f98a2SYan Zheng } 45515d4f98a2SYan Zheng 45520b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 455379787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 455479787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4555ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 455679787eaaSJeff Mahoney goto out_free; 455779787eaaSJeff Mahoney } 45585d4f98a2SYan Zheng 4559ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 456079787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 45615d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 45625d4f98a2SYan Zheng } 45635d4f98a2SYan Zheng 45643a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 456579787eaaSJeff Mahoney if (err) 456679787eaaSJeff Mahoney goto out_free; 45675d4f98a2SYan Zheng 45685d4f98a2SYan Zheng merge_reloc_roots(rc); 45695d4f98a2SYan Zheng 45705d4f98a2SYan Zheng unset_reloc_control(rc); 45715d4f98a2SYan Zheng 45727a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 457362b99540SQu Wenruo if (IS_ERR(trans)) { 45743612b495STsutomu Itoh err = PTR_ERR(trans); 457562b99540SQu Wenruo goto out_free; 457662b99540SQu Wenruo } 45773a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 4578d2311e69SQu Wenruo 4579d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4580d2311e69SQu Wenruo if (ret < 0 && !err) 4581d2311e69SQu Wenruo err = ret; 45823612b495STsutomu Itoh out_free: 45835d4f98a2SYan Zheng kfree(rc); 45843612b495STsutomu Itoh out: 4585aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4586aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4587aca1bba6SLiu Bo 45885d4f98a2SYan Zheng btrfs_free_path(path); 45895d4f98a2SYan Zheng 45905d4f98a2SYan Zheng if (err == 0) { 45915d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 45920b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 45935d4f98a2SYan Zheng if (IS_ERR(fs_root)) 45945d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4595d7ce5843SMiao Xie else 459666b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 45975d4f98a2SYan Zheng } 45985d4f98a2SYan Zheng return err; 45995d4f98a2SYan Zheng } 46005d4f98a2SYan Zheng 46015d4f98a2SYan Zheng /* 46025d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 46035d4f98a2SYan Zheng * 46045d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 46055d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 46065d4f98a2SYan Zheng */ 46075d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 46085d4f98a2SYan Zheng { 46090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 46105d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 46115d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 46125d4f98a2SYan Zheng int ret; 46135d4f98a2SYan Zheng u64 disk_bytenr; 46144577b014SJosef Bacik u64 new_bytenr; 46155d4f98a2SYan Zheng LIST_HEAD(list); 46165d4f98a2SYan Zheng 46175d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 46185d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 46195d4f98a2SYan Zheng 46205d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46210b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4622a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 462379787eaaSJeff Mahoney if (ret) 462479787eaaSJeff Mahoney goto out; 46255d4f98a2SYan Zheng 46265d4f98a2SYan Zheng while (!list_empty(&list)) { 46275d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46285d4f98a2SYan Zheng list_del_init(&sums->list); 46295d4f98a2SYan Zheng 46304577b014SJosef Bacik /* 46314577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46324577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46334577b014SJosef Bacik * extents but we may have written to say the middle of the 46344577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46354577b014SJosef Bacik * the right disk offset. 46364577b014SJosef Bacik * 46374577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46384577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46394577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46404577b014SJosef Bacik * disk length. 46414577b014SJosef Bacik */ 46424577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 46434577b014SJosef Bacik sums->bytenr = new_bytenr; 46445d4f98a2SYan Zheng 4645f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 46465d4f98a2SYan Zheng } 464779787eaaSJeff Mahoney out: 46485d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4649411fc6bcSAndi Kleen return ret; 46505d4f98a2SYan Zheng } 46513fd0a558SYan, Zheng 465283d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 46533fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 46543fd0a558SYan, Zheng struct extent_buffer *cow) 46553fd0a558SYan, Zheng { 46560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46573fd0a558SYan, Zheng struct reloc_control *rc; 46583fd0a558SYan, Zheng struct backref_node *node; 46593fd0a558SYan, Zheng int first_cow = 0; 46603fd0a558SYan, Zheng int level; 466183d4cfd4SJosef Bacik int ret = 0; 46623fd0a558SYan, Zheng 46630b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 46643fd0a558SYan, Zheng if (!rc) 466583d4cfd4SJosef Bacik return 0; 46663fd0a558SYan, Zheng 46673fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46683fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46693fd0a558SYan, Zheng 4670c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4671c974c464SWang Shilong if (buf == root->node) 4672c974c464SWang Shilong __update_reloc_root(root, cow->start); 4673c974c464SWang Shilong } 4674c974c464SWang Shilong 46753fd0a558SYan, Zheng level = btrfs_header_level(buf); 46763fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46773fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 46783fd0a558SYan, Zheng first_cow = 1; 46793fd0a558SYan, Zheng 46803fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 46813fd0a558SYan, Zheng rc->create_reloc_tree) { 46823fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 46833fd0a558SYan, Zheng 46843fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 46853fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 46863fd0a558SYan, Zheng node->new_bytenr != buf->start); 46873fd0a558SYan, Zheng 46883fd0a558SYan, Zheng drop_node_buffer(node); 468967439dadSDavid Sterba atomic_inc(&cow->refs); 46903fd0a558SYan, Zheng node->eb = cow; 46913fd0a558SYan, Zheng node->new_bytenr = cow->start; 46923fd0a558SYan, Zheng 46933fd0a558SYan, Zheng if (!node->pending) { 46943fd0a558SYan, Zheng list_move_tail(&node->list, 46953fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 46963fd0a558SYan, Zheng node->pending = 1; 46973fd0a558SYan, Zheng } 46983fd0a558SYan, Zheng 46993fd0a558SYan, Zheng if (first_cow) 47003fd0a558SYan, Zheng __mark_block_processed(rc, node); 47013fd0a558SYan, Zheng 47023fd0a558SYan, Zheng if (first_cow && level > 0) 47033fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 47043fd0a558SYan, Zheng } 47053fd0a558SYan, Zheng 470683d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 47073fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 470883d4cfd4SJosef Bacik return ret; 47093fd0a558SYan, Zheng } 47103fd0a558SYan, Zheng 47113fd0a558SYan, Zheng /* 47123fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 471301327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 47143fd0a558SYan, Zheng */ 4715147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 47163fd0a558SYan, Zheng u64 *bytes_to_reserve) 47173fd0a558SYan, Zheng { 471810995c04SQu Wenruo struct btrfs_root *root = pending->root; 471910995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 47203fd0a558SYan, Zheng 472110995c04SQu Wenruo if (!root->reloc_root || !rc) 47223fd0a558SYan, Zheng return; 47233fd0a558SYan, Zheng 47243fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47253fd0a558SYan, Zheng return; 47263fd0a558SYan, Zheng 47273fd0a558SYan, Zheng root = root->reloc_root; 47283fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47293fd0a558SYan, Zheng /* 47303fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47313fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47323fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47333fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47343fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47353fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 47363fd0a558SYan, Zheng * tree may use more space than it frees. so we need 47373fd0a558SYan, Zheng * reserve extra space. 47383fd0a558SYan, Zheng */ 47393fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47403fd0a558SYan, Zheng } 47413fd0a558SYan, Zheng 47423fd0a558SYan, Zheng /* 47433fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47443fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 47453fd0a558SYan, Zheng */ 474649b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 47473fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 47483fd0a558SYan, Zheng { 47493fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 47503fd0a558SYan, Zheng struct btrfs_root *reloc_root; 47513fd0a558SYan, Zheng struct btrfs_root *new_root; 475210995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 47533fd0a558SYan, Zheng int ret; 47543fd0a558SYan, Zheng 475510995c04SQu Wenruo if (!root->reloc_root || !rc) 475649b25e05SJeff Mahoney return 0; 47573fd0a558SYan, Zheng 47583fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47593fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 47603fd0a558SYan, Zheng 47613fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 47623fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 47633fd0a558SYan, Zheng rc->block_rsv, 47643a584174SLu Fengqi rc->nodes_relocated, true); 476549b25e05SJeff Mahoney if (ret) 476649b25e05SJeff Mahoney return ret; 47673fd0a558SYan, Zheng } 47683fd0a558SYan, Zheng 47693fd0a558SYan, Zheng new_root = pending->snap; 47703fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47713fd0a558SYan, Zheng new_root->root_key.objectid); 477249b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 477349b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47743fd0a558SYan, Zheng 4775ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4776ffd7b339SJeff Mahoney BUG_ON(ret < 0); 47773fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 47783fd0a558SYan, Zheng 477949b25e05SJeff Mahoney if (rc->create_reloc_tree) 47803fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 478149b25e05SJeff Mahoney return ret; 47823fd0a558SYan, Zheng } 4783