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 5206282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 5216282675eSQu Wenruo { 5226282675eSQu Wenruo /* 5236282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 5246282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 5256282675eSQu Wenruo * trying to access reloc_root 5266282675eSQu Wenruo */ 5276282675eSQu Wenruo smp_rmb(); 5286282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 5296282675eSQu Wenruo return true; 5306282675eSQu Wenruo return false; 5316282675eSQu Wenruo } 5326282675eSQu Wenruo 5336282675eSQu Wenruo /* 5346282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 5356282675eSQu Wenruo * 5366282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 5376282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 5386282675eSQu Wenruo * from no reloc root. But should_ignore_root() below is a special case. 5396282675eSQu Wenruo */ 5406282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 5416282675eSQu Wenruo { 5426282675eSQu Wenruo if (reloc_root_is_dead(root)) 5436282675eSQu Wenruo return false; 5446282675eSQu Wenruo if (!root->reloc_root) 5456282675eSQu Wenruo return false; 5466282675eSQu Wenruo return true; 5476282675eSQu Wenruo } 548f2a97a9dSDavid Sterba 5493fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5503fd0a558SYan, Zheng { 5513fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5523fd0a558SYan, Zheng 55327cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 5543fd0a558SYan, Zheng return 0; 5553fd0a558SYan, Zheng 5566282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 5576282675eSQu Wenruo if (reloc_root_is_dead(root)) 5586282675eSQu Wenruo return 1; 5596282675eSQu Wenruo 5603fd0a558SYan, Zheng reloc_root = root->reloc_root; 5613fd0a558SYan, Zheng if (!reloc_root) 5623fd0a558SYan, Zheng return 0; 5633fd0a558SYan, Zheng 5643fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5653fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5663fd0a558SYan, Zheng return 0; 5673fd0a558SYan, Zheng /* 5683fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5693fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5703fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5713fd0a558SYan, Zheng * relocation. 5723fd0a558SYan, Zheng */ 5733fd0a558SYan, Zheng return 1; 5743fd0a558SYan, Zheng } 5755d4f98a2SYan Zheng /* 5765d4f98a2SYan Zheng * find reloc tree by address of tree root 5775d4f98a2SYan Zheng */ 5785d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5795d4f98a2SYan Zheng u64 bytenr) 5805d4f98a2SYan Zheng { 5815d4f98a2SYan Zheng struct rb_node *rb_node; 5825d4f98a2SYan Zheng struct mapping_node *node; 5835d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5845d4f98a2SYan Zheng 5855d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5865d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5875d4f98a2SYan Zheng if (rb_node) { 5885d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5895d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5905d4f98a2SYan Zheng } 5915d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5925d4f98a2SYan Zheng return root; 5935d4f98a2SYan Zheng } 5945d4f98a2SYan Zheng 5955d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5965d4f98a2SYan Zheng { 5975d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5985d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5995d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 6005d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 6015d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 60266463748SWang Shilong root_objectid == BTRFS_CSUM_TREE_OBJECTID || 60366463748SWang Shilong root_objectid == BTRFS_UUID_TREE_OBJECTID || 6043e4c5efbSDavid Sterba root_objectid == BTRFS_QUOTA_TREE_OBJECTID || 6053e4c5efbSDavid Sterba root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 6065d4f98a2SYan Zheng return 1; 6075d4f98a2SYan Zheng return 0; 6085d4f98a2SYan Zheng } 6095d4f98a2SYan Zheng 6105d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 6115d4f98a2SYan Zheng u64 root_objectid) 6125d4f98a2SYan Zheng { 6135d4f98a2SYan Zheng struct btrfs_key key; 6145d4f98a2SYan Zheng 6155d4f98a2SYan Zheng key.objectid = root_objectid; 6165d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 6175d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 6185d4f98a2SYan Zheng key.offset = 0; 6195d4f98a2SYan Zheng else 6205d4f98a2SYan Zheng key.offset = (u64)-1; 6215d4f98a2SYan Zheng 622c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 6235d4f98a2SYan Zheng } 6245d4f98a2SYan Zheng 6255d4f98a2SYan Zheng static noinline_for_stack 6265d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 6275d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 6285d4f98a2SYan Zheng { 6293173a18fSJosef Bacik struct btrfs_key key; 6305d4f98a2SYan Zheng struct btrfs_extent_item *ei; 6315d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6325d4f98a2SYan Zheng u32 item_size; 6335d4f98a2SYan Zheng 6343173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6353173a18fSJosef Bacik 6365d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 637ba3c2b19SNikolay Borisov if (item_size < sizeof(*ei)) { 638ba3c2b19SNikolay Borisov btrfs_print_v0_err(leaf->fs_info); 639ba3c2b19SNikolay Borisov btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL); 640ba3c2b19SNikolay Borisov return 1; 641ba3c2b19SNikolay Borisov } 6425d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6435d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6445d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6455d4f98a2SYan Zheng 6463173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6473173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6485d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6495d4f98a2SYan Zheng return 1; 6505d4f98a2SYan Zheng } 651d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 652d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 653d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 654d062d13cSJosef Bacik return 1; 655d062d13cSJosef Bacik } 6565d4f98a2SYan Zheng 6573173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6585d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6595d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6603173a18fSJosef Bacik } else { 6613173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6623173a18fSJosef Bacik } 6635d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6645d4f98a2SYan Zheng return 0; 6655d4f98a2SYan Zheng } 6665d4f98a2SYan Zheng 6675d4f98a2SYan Zheng /* 6685d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6695d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6705d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6715d4f98a2SYan Zheng * 6725d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 67301327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 67401327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 6755d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6765d4f98a2SYan Zheng * 6775d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6785d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6795d4f98a2SYan Zheng * block are also cached. 6805d4f98a2SYan Zheng */ 6813fd0a558SYan, Zheng static noinline_for_stack 6823fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6835d4f98a2SYan Zheng struct btrfs_key *node_key, 6845d4f98a2SYan Zheng int level, u64 bytenr) 6855d4f98a2SYan Zheng { 6863fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 687fa6ac715SQu Wenruo struct btrfs_path *path1; /* For searching extent root */ 688fa6ac715SQu Wenruo struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */ 6895d4f98a2SYan Zheng struct extent_buffer *eb; 6905d4f98a2SYan Zheng struct btrfs_root *root; 6915d4f98a2SYan Zheng struct backref_node *cur; 6925d4f98a2SYan Zheng struct backref_node *upper; 6935d4f98a2SYan Zheng struct backref_node *lower; 6945d4f98a2SYan Zheng struct backref_node *node = NULL; 6955d4f98a2SYan Zheng struct backref_node *exist = NULL; 6965d4f98a2SYan Zheng struct backref_edge *edge; 6975d4f98a2SYan Zheng struct rb_node *rb_node; 6985d4f98a2SYan Zheng struct btrfs_key key; 6995d4f98a2SYan Zheng unsigned long end; 7005d4f98a2SYan Zheng unsigned long ptr; 701fa6ac715SQu Wenruo LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */ 7023fd0a558SYan, Zheng LIST_HEAD(useless); 7033fd0a558SYan, Zheng int cowonly; 7045d4f98a2SYan Zheng int ret; 7055d4f98a2SYan Zheng int err = 0; 706b6c60c80SJosef Bacik bool need_check = true; 7075d4f98a2SYan Zheng 7085d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 7095d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 7105d4f98a2SYan Zheng if (!path1 || !path2) { 7115d4f98a2SYan Zheng err = -ENOMEM; 7125d4f98a2SYan Zheng goto out; 7135d4f98a2SYan Zheng } 714e4058b54SDavid Sterba path1->reada = READA_FORWARD; 715e4058b54SDavid Sterba path2->reada = READA_FORWARD; 7165d4f98a2SYan Zheng 7173fd0a558SYan, Zheng node = alloc_backref_node(cache); 7185d4f98a2SYan Zheng if (!node) { 7195d4f98a2SYan Zheng err = -ENOMEM; 7205d4f98a2SYan Zheng goto out; 7215d4f98a2SYan Zheng } 7225d4f98a2SYan Zheng 7235d4f98a2SYan Zheng node->bytenr = bytenr; 7245d4f98a2SYan Zheng node->level = level; 7255d4f98a2SYan Zheng node->lowest = 1; 7265d4f98a2SYan Zheng cur = node; 7275d4f98a2SYan Zheng again: 7285d4f98a2SYan Zheng end = 0; 7295d4f98a2SYan Zheng ptr = 0; 7305d4f98a2SYan Zheng key.objectid = cur->bytenr; 7313173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7325d4f98a2SYan Zheng key.offset = (u64)-1; 7335d4f98a2SYan Zheng 7345d4f98a2SYan Zheng path1->search_commit_root = 1; 7355d4f98a2SYan Zheng path1->skip_locking = 1; 7365d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7375d4f98a2SYan Zheng 0, 0); 7385d4f98a2SYan Zheng if (ret < 0) { 7395d4f98a2SYan Zheng err = ret; 7405d4f98a2SYan Zheng goto out; 7415d4f98a2SYan Zheng } 74275bfb9afSJosef Bacik ASSERT(ret); 74375bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7445d4f98a2SYan Zheng 7455d4f98a2SYan Zheng path1->slots[0]--; 7465d4f98a2SYan Zheng 7475d4f98a2SYan Zheng WARN_ON(cur->checked); 7485d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7495d4f98a2SYan Zheng /* 75070f23fd6SJustin P. Mattock * the backref was added previously when processing 7515d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7525d4f98a2SYan Zheng */ 75375bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7545d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7555d4f98a2SYan Zheng list[LOWER]); 75675bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7575d4f98a2SYan Zheng exist = edge->node[UPPER]; 7585d4f98a2SYan Zheng /* 7595d4f98a2SYan Zheng * add the upper level block to pending list if we need 7605d4f98a2SYan Zheng * check its backrefs 7615d4f98a2SYan Zheng */ 7625d4f98a2SYan Zheng if (!exist->checked) 7635d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7645d4f98a2SYan Zheng } else { 7655d4f98a2SYan Zheng exist = NULL; 7665d4f98a2SYan Zheng } 7675d4f98a2SYan Zheng 7685d4f98a2SYan Zheng while (1) { 7695d4f98a2SYan Zheng cond_resched(); 7705d4f98a2SYan Zheng eb = path1->nodes[0]; 7715d4f98a2SYan Zheng 7725d4f98a2SYan Zheng if (ptr >= end) { 7735d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7745d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7755d4f98a2SYan Zheng if (ret < 0) { 7765d4f98a2SYan Zheng err = ret; 7775d4f98a2SYan Zheng goto out; 7785d4f98a2SYan Zheng } 7795d4f98a2SYan Zheng if (ret > 0) 7805d4f98a2SYan Zheng break; 7815d4f98a2SYan Zheng eb = path1->nodes[0]; 7825d4f98a2SYan Zheng } 7835d4f98a2SYan Zheng 7845d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7855d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7865d4f98a2SYan Zheng WARN_ON(exist); 7875d4f98a2SYan Zheng break; 7885d4f98a2SYan Zheng } 7895d4f98a2SYan Zheng 7903173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7913173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7925d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7935d4f98a2SYan Zheng &ptr, &end); 7945d4f98a2SYan Zheng if (ret) 7955d4f98a2SYan Zheng goto next; 7965d4f98a2SYan Zheng } 7975d4f98a2SYan Zheng } 7985d4f98a2SYan Zheng 7995d4f98a2SYan Zheng if (ptr < end) { 8005d4f98a2SYan Zheng /* update key for inline back ref */ 8015d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 8023de28d57SLiu Bo int type; 8035d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 8043de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 8053de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 8063de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 807af431dcbSSu Yue err = -EUCLEAN; 8083de28d57SLiu Bo goto out; 8093de28d57SLiu Bo } 8103de28d57SLiu Bo key.type = type; 8115d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 8123de28d57SLiu Bo 8135d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 8145d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 8155d4f98a2SYan Zheng } 8165d4f98a2SYan Zheng 817fa6ac715SQu Wenruo /* 818fa6ac715SQu Wenruo * Parent node found and matches current inline ref, no need to 819fa6ac715SQu Wenruo * rebuild this node for this inline ref. 820fa6ac715SQu Wenruo */ 8215d4f98a2SYan Zheng if (exist && 8225d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8235d4f98a2SYan Zheng exist->owner == key.offset) || 8245d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8255d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8265d4f98a2SYan Zheng exist = NULL; 8275d4f98a2SYan Zheng goto next; 8285d4f98a2SYan Zheng } 8295d4f98a2SYan Zheng 830fa6ac715SQu Wenruo /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ 8315d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8325d4f98a2SYan Zheng if (key.objectid == key.offset) { 8335d4f98a2SYan Zheng /* 834fa6ac715SQu Wenruo * Only root blocks of reloc trees use backref 835fa6ac715SQu Wenruo * pointing to itself. 8365d4f98a2SYan Zheng */ 8375d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 83875bfb9afSJosef Bacik ASSERT(root); 8395d4f98a2SYan Zheng cur->root = root; 8405d4f98a2SYan Zheng break; 8415d4f98a2SYan Zheng } 8425d4f98a2SYan Zheng 8433fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8445d4f98a2SYan Zheng if (!edge) { 8455d4f98a2SYan Zheng err = -ENOMEM; 8465d4f98a2SYan Zheng goto out; 8475d4f98a2SYan Zheng } 8485d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8495d4f98a2SYan Zheng if (!rb_node) { 8503fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8515d4f98a2SYan Zheng if (!upper) { 8523fd0a558SYan, Zheng free_backref_edge(cache, edge); 8535d4f98a2SYan Zheng err = -ENOMEM; 8545d4f98a2SYan Zheng goto out; 8555d4f98a2SYan Zheng } 8565d4f98a2SYan Zheng upper->bytenr = key.offset; 8575d4f98a2SYan Zheng upper->level = cur->level + 1; 8585d4f98a2SYan Zheng /* 8595d4f98a2SYan Zheng * backrefs for the upper level block isn't 8605d4f98a2SYan Zheng * cached, add the block to pending list 8615d4f98a2SYan Zheng */ 8625d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8635d4f98a2SYan Zheng } else { 8645d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8655d4f98a2SYan Zheng rb_node); 86675bfb9afSJosef Bacik ASSERT(upper->checked); 8675d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8685d4f98a2SYan Zheng } 8693fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8705d4f98a2SYan Zheng edge->node[LOWER] = cur; 8713fd0a558SYan, Zheng edge->node[UPPER] = upper; 8725d4f98a2SYan Zheng 8735d4f98a2SYan Zheng goto next; 8746d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 875ba3c2b19SNikolay Borisov err = -EINVAL; 876ba3c2b19SNikolay Borisov btrfs_print_v0_err(rc->extent_root->fs_info); 877ba3c2b19SNikolay Borisov btrfs_handle_fs_error(rc->extent_root->fs_info, err, 878ba3c2b19SNikolay Borisov NULL); 879ba3c2b19SNikolay Borisov goto out; 8805d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8815d4f98a2SYan Zheng goto next; 8825d4f98a2SYan Zheng } 8835d4f98a2SYan Zheng 884fa6ac715SQu Wenruo /* 885fa6ac715SQu Wenruo * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset 886fa6ac715SQu Wenruo * means the root objectid. We need to search the tree to get 887fa6ac715SQu Wenruo * its parent bytenr. 888fa6ac715SQu Wenruo */ 8895d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8905d4f98a2SYan Zheng if (IS_ERR(root)) { 8915d4f98a2SYan Zheng err = PTR_ERR(root); 8925d4f98a2SYan Zheng goto out; 8935d4f98a2SYan Zheng } 8945d4f98a2SYan Zheng 89527cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 8963fd0a558SYan, Zheng cur->cowonly = 1; 8973fd0a558SYan, Zheng 8985d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 8995d4f98a2SYan Zheng /* tree root */ 90075bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9015d4f98a2SYan Zheng cur->bytenr); 9023fd0a558SYan, Zheng if (should_ignore_root(root)) 9033fd0a558SYan, Zheng list_add(&cur->list, &useless); 9043fd0a558SYan, Zheng else 9055d4f98a2SYan Zheng cur->root = root; 9065d4f98a2SYan Zheng break; 9075d4f98a2SYan Zheng } 9085d4f98a2SYan Zheng 9095d4f98a2SYan Zheng level = cur->level + 1; 9105d4f98a2SYan Zheng 911fa6ac715SQu Wenruo /* Search the tree to find parent blocks referring the block. */ 9125d4f98a2SYan Zheng path2->search_commit_root = 1; 9135d4f98a2SYan Zheng path2->skip_locking = 1; 9145d4f98a2SYan Zheng path2->lowest_level = level; 9155d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 9165d4f98a2SYan Zheng path2->lowest_level = 0; 9175d4f98a2SYan Zheng if (ret < 0) { 9185d4f98a2SYan Zheng err = ret; 9195d4f98a2SYan Zheng goto out; 9205d4f98a2SYan Zheng } 92133c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 92233c66f43SYan Zheng path2->slots[level]--; 9235d4f98a2SYan Zheng 9245d4f98a2SYan Zheng eb = path2->nodes[level]; 9253561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 9263561b9dbSLiu Bo cur->bytenr) { 9273561b9dbSLiu Bo btrfs_err(root->fs_info, 9283561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 9294fd786e6SMisono Tomohiro cur->bytenr, level - 1, 9304fd786e6SMisono Tomohiro root->root_key.objectid, 9313561b9dbSLiu Bo node_key->objectid, node_key->type, 9323561b9dbSLiu Bo node_key->offset); 9333561b9dbSLiu Bo err = -ENOENT; 9343561b9dbSLiu Bo goto out; 9353561b9dbSLiu Bo } 9365d4f98a2SYan Zheng lower = cur; 937b6c60c80SJosef Bacik need_check = true; 938fa6ac715SQu Wenruo 939fa6ac715SQu Wenruo /* Add all nodes and edges in the path */ 9405d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9415d4f98a2SYan Zheng if (!path2->nodes[level]) { 94275bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9435d4f98a2SYan Zheng lower->bytenr); 9443fd0a558SYan, Zheng if (should_ignore_root(root)) 9453fd0a558SYan, Zheng list_add(&lower->list, &useless); 9463fd0a558SYan, Zheng else 9475d4f98a2SYan Zheng lower->root = root; 9485d4f98a2SYan Zheng break; 9495d4f98a2SYan Zheng } 9505d4f98a2SYan Zheng 9513fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9525d4f98a2SYan Zheng if (!edge) { 9535d4f98a2SYan Zheng err = -ENOMEM; 9545d4f98a2SYan Zheng goto out; 9555d4f98a2SYan Zheng } 9565d4f98a2SYan Zheng 9575d4f98a2SYan Zheng eb = path2->nodes[level]; 9585d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9595d4f98a2SYan Zheng if (!rb_node) { 9603fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9615d4f98a2SYan Zheng if (!upper) { 9623fd0a558SYan, Zheng free_backref_edge(cache, edge); 9635d4f98a2SYan Zheng err = -ENOMEM; 9645d4f98a2SYan Zheng goto out; 9655d4f98a2SYan Zheng } 9665d4f98a2SYan Zheng upper->bytenr = eb->start; 9675d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9685d4f98a2SYan Zheng upper->level = lower->level + 1; 96927cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 97027cdeb70SMiao Xie &root->state)) 9713fd0a558SYan, Zheng upper->cowonly = 1; 9725d4f98a2SYan Zheng 9735d4f98a2SYan Zheng /* 9745d4f98a2SYan Zheng * if we know the block isn't shared 9755d4f98a2SYan Zheng * we can void checking its backrefs. 9765d4f98a2SYan Zheng */ 9775d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9785d4f98a2SYan Zheng upper->checked = 0; 9795d4f98a2SYan Zheng else 9805d4f98a2SYan Zheng upper->checked = 1; 9815d4f98a2SYan Zheng 9825d4f98a2SYan Zheng /* 9835d4f98a2SYan Zheng * add the block to pending list if we 984b6c60c80SJosef Bacik * need check its backrefs, we only do this once 985b6c60c80SJosef Bacik * while walking up a tree as we will catch 986b6c60c80SJosef Bacik * anything else later on. 9875d4f98a2SYan Zheng */ 988b6c60c80SJosef Bacik if (!upper->checked && need_check) { 989b6c60c80SJosef Bacik need_check = false; 9905d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9915d4f98a2SYan Zheng &list); 992bbe90514SJosef Bacik } else { 993bbe90514SJosef Bacik if (upper->checked) 994bbe90514SJosef Bacik need_check = true; 9955d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 996bbe90514SJosef Bacik } 9975d4f98a2SYan Zheng } else { 9985d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9995d4f98a2SYan Zheng rb_node); 100075bfb9afSJosef Bacik ASSERT(upper->checked); 10015d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 10023fd0a558SYan, Zheng if (!upper->owner) 10033fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 10045d4f98a2SYan Zheng } 10055d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 10065d4f98a2SYan Zheng edge->node[LOWER] = lower; 10073fd0a558SYan, Zheng edge->node[UPPER] = upper; 10085d4f98a2SYan Zheng 10095d4f98a2SYan Zheng if (rb_node) 10105d4f98a2SYan Zheng break; 10115d4f98a2SYan Zheng lower = upper; 10125d4f98a2SYan Zheng upper = NULL; 10135d4f98a2SYan Zheng } 1014b3b4aa74SDavid Sterba btrfs_release_path(path2); 10155d4f98a2SYan Zheng next: 10165d4f98a2SYan Zheng if (ptr < end) { 10175d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 10185d4f98a2SYan Zheng if (ptr >= end) { 10195d4f98a2SYan Zheng WARN_ON(ptr > end); 10205d4f98a2SYan Zheng ptr = 0; 10215d4f98a2SYan Zheng end = 0; 10225d4f98a2SYan Zheng } 10235d4f98a2SYan Zheng } 10245d4f98a2SYan Zheng if (ptr >= end) 10255d4f98a2SYan Zheng path1->slots[0]++; 10265d4f98a2SYan Zheng } 1027b3b4aa74SDavid Sterba btrfs_release_path(path1); 10285d4f98a2SYan Zheng 10295d4f98a2SYan Zheng cur->checked = 1; 10305d4f98a2SYan Zheng WARN_ON(exist); 10315d4f98a2SYan Zheng 10325d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10335d4f98a2SYan Zheng if (!list_empty(&list)) { 10345d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10355d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10365d4f98a2SYan Zheng cur = edge->node[UPPER]; 10375d4f98a2SYan Zheng goto again; 10385d4f98a2SYan Zheng } 10395d4f98a2SYan Zheng 10405d4f98a2SYan Zheng /* 10415d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10425d4f98a2SYan Zheng * into the cache. 10435d4f98a2SYan Zheng */ 104475bfb9afSJosef Bacik ASSERT(node->checked); 10453fd0a558SYan, Zheng cowonly = node->cowonly; 10463fd0a558SYan, Zheng if (!cowonly) { 10473fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10483fd0a558SYan, Zheng &node->rb_node); 104943c04fb1SJeff Mahoney if (rb_node) 105043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10513fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10523fd0a558SYan, Zheng } 10535d4f98a2SYan Zheng 10545d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10555d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10565d4f98a2SYan Zheng 10575d4f98a2SYan Zheng while (!list_empty(&list)) { 10585d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10595d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10605d4f98a2SYan Zheng upper = edge->node[UPPER]; 10613fd0a558SYan, Zheng if (upper->detached) { 10623fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10633fd0a558SYan, Zheng lower = edge->node[LOWER]; 10643fd0a558SYan, Zheng free_backref_edge(cache, edge); 10653fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10663fd0a558SYan, Zheng list_add(&lower->list, &useless); 10673fd0a558SYan, Zheng continue; 10683fd0a558SYan, Zheng } 10695d4f98a2SYan Zheng 10705d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10715d4f98a2SYan Zheng if (upper->lowest) { 10725d4f98a2SYan Zheng list_del_init(&upper->lower); 10735d4f98a2SYan Zheng upper->lowest = 0; 10745d4f98a2SYan Zheng } 10755d4f98a2SYan Zheng 10765d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10775d4f98a2SYan Zheng continue; 10785d4f98a2SYan Zheng } 10795d4f98a2SYan Zheng 108075bfb9afSJosef Bacik if (!upper->checked) { 108175bfb9afSJosef Bacik /* 108275bfb9afSJosef Bacik * Still want to blow up for developers since this is a 108375bfb9afSJosef Bacik * logic bug. 108475bfb9afSJosef Bacik */ 108575bfb9afSJosef Bacik ASSERT(0); 108675bfb9afSJosef Bacik err = -EINVAL; 108775bfb9afSJosef Bacik goto out; 108875bfb9afSJosef Bacik } 108975bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 109075bfb9afSJosef Bacik ASSERT(0); 109175bfb9afSJosef Bacik err = -EINVAL; 109275bfb9afSJosef Bacik goto out; 109375bfb9afSJosef Bacik } 109475bfb9afSJosef Bacik 10953fd0a558SYan, Zheng if (!cowonly) { 10965d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10975d4f98a2SYan Zheng &upper->rb_node); 109843c04fb1SJeff Mahoney if (rb_node) 109943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 110043c04fb1SJeff Mahoney upper->bytenr); 11013fd0a558SYan, Zheng } 11025d4f98a2SYan Zheng 11035d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11045d4f98a2SYan Zheng 11055d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 11065d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 11075d4f98a2SYan Zheng } 11083fd0a558SYan, Zheng /* 11093fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 11103fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 11113fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 11123fd0a558SYan, Zheng * lookup. 11133fd0a558SYan, Zheng */ 11143fd0a558SYan, Zheng while (!list_empty(&useless)) { 11153fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 11163fd0a558SYan, Zheng list_del_init(&upper->list); 111775bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 11183fd0a558SYan, Zheng if (upper == node) 11193fd0a558SYan, Zheng node = NULL; 11203fd0a558SYan, Zheng if (upper->lowest) { 11213fd0a558SYan, Zheng list_del_init(&upper->lower); 11223fd0a558SYan, Zheng upper->lowest = 0; 11233fd0a558SYan, Zheng } 11243fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 11253fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 11263fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 11273fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11283fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11293fd0a558SYan, Zheng lower = edge->node[LOWER]; 11303fd0a558SYan, Zheng free_backref_edge(cache, edge); 11313fd0a558SYan, Zheng 11323fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11333fd0a558SYan, Zheng list_add(&lower->list, &useless); 11343fd0a558SYan, Zheng } 11353fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11363fd0a558SYan, Zheng if (upper->level > 0) { 11373fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11383fd0a558SYan, Zheng upper->detached = 1; 11393fd0a558SYan, Zheng } else { 11403fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11413fd0a558SYan, Zheng free_backref_node(cache, upper); 11423fd0a558SYan, Zheng } 11433fd0a558SYan, Zheng } 11445d4f98a2SYan Zheng out: 11455d4f98a2SYan Zheng btrfs_free_path(path1); 11465d4f98a2SYan Zheng btrfs_free_path(path2); 11475d4f98a2SYan Zheng if (err) { 11483fd0a558SYan, Zheng while (!list_empty(&useless)) { 11493fd0a558SYan, Zheng lower = list_entry(useless.next, 115075bfb9afSJosef Bacik struct backref_node, list); 115175bfb9afSJosef Bacik list_del_init(&lower->list); 11523fd0a558SYan, Zheng } 115375bfb9afSJosef Bacik while (!list_empty(&list)) { 115475bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 115575bfb9afSJosef Bacik list[UPPER]); 115675bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11573fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 115875bfb9afSJosef Bacik lower = edge->node[LOWER]; 11595d4f98a2SYan Zheng upper = edge->node[UPPER]; 11603fd0a558SYan, Zheng free_backref_edge(cache, edge); 116175bfb9afSJosef Bacik 116275bfb9afSJosef Bacik /* 116375bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 116475bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 116575bfb9afSJosef Bacik */ 116675bfb9afSJosef Bacik if (list_empty(&lower->upper) && 116775bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 116875bfb9afSJosef Bacik list_add(&lower->list, &useless); 116975bfb9afSJosef Bacik 117075bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 117175bfb9afSJosef Bacik continue; 117275bfb9afSJosef Bacik 117301327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 117475bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 117575bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 117675bfb9afSJosef Bacik if (list_empty(&upper->upper)) 117775bfb9afSJosef Bacik list_add(&upper->list, &useless); 117875bfb9afSJosef Bacik } 117975bfb9afSJosef Bacik 118075bfb9afSJosef Bacik while (!list_empty(&useless)) { 118175bfb9afSJosef Bacik lower = list_entry(useless.next, 118275bfb9afSJosef Bacik struct backref_node, list); 118375bfb9afSJosef Bacik list_del_init(&lower->list); 11840fd8c3daSLiu Bo if (lower == node) 11850fd8c3daSLiu Bo node = NULL; 118675bfb9afSJosef Bacik free_backref_node(cache, lower); 11875d4f98a2SYan Zheng } 11880fd8c3daSLiu Bo 11890fd8c3daSLiu Bo free_backref_node(cache, node); 11905d4f98a2SYan Zheng return ERR_PTR(err); 11915d4f98a2SYan Zheng } 119275bfb9afSJosef Bacik ASSERT(!node || !node->detached); 11935d4f98a2SYan Zheng return node; 11945d4f98a2SYan Zheng } 11955d4f98a2SYan Zheng 11965d4f98a2SYan Zheng /* 11973fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11983fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11993fd0a558SYan, Zheng * corresponds to root of source tree 12003fd0a558SYan, Zheng */ 12013fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 12023fd0a558SYan, Zheng struct reloc_control *rc, 12033fd0a558SYan, Zheng struct btrfs_root *src, 12043fd0a558SYan, Zheng struct btrfs_root *dest) 12053fd0a558SYan, Zheng { 12063fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 12073fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 12083fd0a558SYan, Zheng struct backref_node *node = NULL; 12093fd0a558SYan, Zheng struct backref_node *new_node; 12103fd0a558SYan, Zheng struct backref_edge *edge; 12113fd0a558SYan, Zheng struct backref_edge *new_edge; 12123fd0a558SYan, Zheng struct rb_node *rb_node; 12133fd0a558SYan, Zheng 12143fd0a558SYan, Zheng if (cache->last_trans > 0) 12153fd0a558SYan, Zheng update_backref_cache(trans, cache); 12163fd0a558SYan, Zheng 12173fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 12183fd0a558SYan, Zheng if (rb_node) { 12193fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 12203fd0a558SYan, Zheng if (node->detached) 12213fd0a558SYan, Zheng node = NULL; 12223fd0a558SYan, Zheng else 12233fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 12243fd0a558SYan, Zheng } 12253fd0a558SYan, Zheng 12263fd0a558SYan, Zheng if (!node) { 12273fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 12283fd0a558SYan, Zheng reloc_root->commit_root->start); 12293fd0a558SYan, Zheng if (rb_node) { 12303fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 12313fd0a558SYan, Zheng rb_node); 12323fd0a558SYan, Zheng BUG_ON(node->detached); 12333fd0a558SYan, Zheng } 12343fd0a558SYan, Zheng } 12353fd0a558SYan, Zheng 12363fd0a558SYan, Zheng if (!node) 12373fd0a558SYan, Zheng return 0; 12383fd0a558SYan, Zheng 12393fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12403fd0a558SYan, Zheng if (!new_node) 12413fd0a558SYan, Zheng return -ENOMEM; 12423fd0a558SYan, Zheng 12433fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12443fd0a558SYan, Zheng new_node->level = node->level; 12453fd0a558SYan, Zheng new_node->lowest = node->lowest; 12466848ad64SYan, Zheng new_node->checked = 1; 12473fd0a558SYan, Zheng new_node->root = dest; 12483fd0a558SYan, Zheng 12493fd0a558SYan, Zheng if (!node->lowest) { 12503fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12513fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12523fd0a558SYan, Zheng if (!new_edge) 12533fd0a558SYan, Zheng goto fail; 12543fd0a558SYan, Zheng 12553fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12563fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12573fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12583fd0a558SYan, Zheng &new_node->lower); 12593fd0a558SYan, Zheng } 126076b9e23dSMiao Xie } else { 126176b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12623fd0a558SYan, Zheng } 12633fd0a558SYan, Zheng 12643fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12653fd0a558SYan, Zheng &new_node->rb_node); 126643c04fb1SJeff Mahoney if (rb_node) 126743c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12683fd0a558SYan, Zheng 12693fd0a558SYan, Zheng if (!new_node->lowest) { 12703fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12713fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12723fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12733fd0a558SYan, Zheng } 12743fd0a558SYan, Zheng } 12753fd0a558SYan, Zheng return 0; 12763fd0a558SYan, Zheng fail: 12773fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12783fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12793fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12803fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12813fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12823fd0a558SYan, Zheng } 12833fd0a558SYan, Zheng free_backref_node(cache, new_node); 12843fd0a558SYan, Zheng return -ENOMEM; 12853fd0a558SYan, Zheng } 12863fd0a558SYan, Zheng 12873fd0a558SYan, Zheng /* 12885d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12895d4f98a2SYan Zheng */ 1290ffd7b339SJeff Mahoney static int __must_check __add_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; 12950b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 12965d4f98a2SYan Zheng 12975d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1298ffd7b339SJeff Mahoney if (!node) 1299ffd7b339SJeff Mahoney return -ENOMEM; 13005d4f98a2SYan Zheng 13015d4f98a2SYan Zheng node->bytenr = root->node->start; 13025d4f98a2SYan Zheng node->data = root; 13035d4f98a2SYan Zheng 13045d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13055d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13065d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13075d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1308ffd7b339SJeff Mahoney if (rb_node) { 13090b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 13105d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 13115d163e0eSJeff Mahoney node->bytenr); 1312ffd7b339SJeff Mahoney } 13135d4f98a2SYan Zheng 13145d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 13155d4f98a2SYan Zheng return 0; 13165d4f98a2SYan Zheng } 13175d4f98a2SYan Zheng 13185d4f98a2SYan Zheng /* 1319c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 13205d4f98a2SYan Zheng * mapping 13215d4f98a2SYan Zheng */ 1322c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 13235d4f98a2SYan Zheng { 13240b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13255d4f98a2SYan Zheng struct rb_node *rb_node; 13265d4f98a2SYan Zheng struct mapping_node *node = NULL; 13270b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 13285d4f98a2SYan Zheng 132965c6e82bSQu Wenruo if (rc && root->node) { 13305d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13315d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1332c974c464SWang Shilong root->node->start); 1333c974c464SWang Shilong if (rb_node) { 1334c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1335c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1336c974c464SWang Shilong } 1337c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1338c974c464SWang Shilong if (!node) 1339c974c464SWang Shilong return; 1340c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1341389305b2SQu Wenruo } 1342c974c464SWang Shilong 13430b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1344c974c464SWang Shilong list_del_init(&root->root_list); 13450b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1346c974c464SWang Shilong kfree(node); 1347c974c464SWang Shilong } 1348c974c464SWang Shilong 1349c974c464SWang Shilong /* 1350c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1351c974c464SWang Shilong * mapping 1352c974c464SWang Shilong */ 1353c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1354c974c464SWang Shilong { 13550b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1356c974c464SWang Shilong struct rb_node *rb_node; 1357c974c464SWang Shilong struct mapping_node *node = NULL; 13580b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1359c974c464SWang Shilong 1360c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1361c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1362c974c464SWang Shilong root->node->start); 13635d4f98a2SYan Zheng if (rb_node) { 13645d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13655d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13665d4f98a2SYan Zheng } 13675d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13685d4f98a2SYan Zheng 13698f71f3e0SLiu Bo if (!node) 13708f71f3e0SLiu Bo return 0; 13715d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13725d4f98a2SYan Zheng 13735d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1374c974c464SWang Shilong node->bytenr = new_bytenr; 13755d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13765d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13775d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 137843c04fb1SJeff Mahoney if (rb_node) 137943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13805d4f98a2SYan Zheng return 0; 13815d4f98a2SYan Zheng } 13825d4f98a2SYan Zheng 13833fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13843fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13855d4f98a2SYan Zheng { 13860b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13875d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13885d4f98a2SYan Zheng struct extent_buffer *eb; 13895d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13905d4f98a2SYan Zheng struct btrfs_key root_key; 13915d4f98a2SYan Zheng int ret; 13925d4f98a2SYan Zheng 13935d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13945d4f98a2SYan Zheng BUG_ON(!root_item); 13955d4f98a2SYan Zheng 13965d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13975d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13983fd0a558SYan, Zheng root_key.offset = objectid; 13995d4f98a2SYan Zheng 14003fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1401054570a1SFilipe Manana u64 commit_root_gen; 1402054570a1SFilipe Manana 14033fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 14045d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 14055d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 14065d4f98a2SYan Zheng BUG_ON(ret); 1407054570a1SFilipe Manana /* 1408054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1409054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1410054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1411054570a1SFilipe Manana * either inside the critical section of a transaction commit 1412054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1413054570a1SFilipe Manana * it's created before the transaction commit is started. 1414054570a1SFilipe Manana */ 1415054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1416054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 14173fd0a558SYan, Zheng } else { 14183fd0a558SYan, Zheng /* 14193fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 14203fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 14213fd0a558SYan, Zheng * modified after it was created have RELOC flag 14223fd0a558SYan, Zheng * set in their headers. so it's OK to not update 14233fd0a558SYan, Zheng * the 'last_snapshot'. 14243fd0a558SYan, Zheng */ 14253fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 14263fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 14273fd0a558SYan, Zheng BUG_ON(ret); 14283fd0a558SYan, Zheng } 14293fd0a558SYan, Zheng 14305d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 14315d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14325d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14335d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14343fd0a558SYan, Zheng 14353fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14363fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14373fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14383fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14395d4f98a2SYan Zheng root_item->drop_level = 0; 14403fd0a558SYan, Zheng } 14415d4f98a2SYan Zheng 14425d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14435d4f98a2SYan Zheng free_extent_buffer(eb); 14445d4f98a2SYan Zheng 14450b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 14465d4f98a2SYan Zheng &root_key, root_item); 14475d4f98a2SYan Zheng BUG_ON(ret); 14485d4f98a2SYan Zheng kfree(root_item); 14495d4f98a2SYan Zheng 14500b246afaSJeff Mahoney reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key); 14515d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14525d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14533fd0a558SYan, Zheng return reloc_root; 14543fd0a558SYan, Zheng } 14553fd0a558SYan, Zheng 14563fd0a558SYan, Zheng /* 14573fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 14583fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 14593fd0a558SYan, Zheng */ 14603fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 14613fd0a558SYan, Zheng struct btrfs_root *root) 14623fd0a558SYan, Zheng { 14630b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14643fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14650b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 146620dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14673fd0a558SYan, Zheng int clear_rsv = 0; 1468ffd7b339SJeff Mahoney int ret; 14693fd0a558SYan, Zheng 14701fac4a54SQu Wenruo /* 14711fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 14721fac4a54SQu Wenruo * create/update the dead reloc tree 14731fac4a54SQu Wenruo */ 14746282675eSQu Wenruo if (reloc_root_is_dead(root)) 14751fac4a54SQu Wenruo return 0; 14761fac4a54SQu Wenruo 14773fd0a558SYan, Zheng if (root->reloc_root) { 14783fd0a558SYan, Zheng reloc_root = root->reloc_root; 14793fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14803fd0a558SYan, Zheng return 0; 14813fd0a558SYan, Zheng } 14823fd0a558SYan, Zheng 14833fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14843fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14853fd0a558SYan, Zheng return 0; 14863fd0a558SYan, Zheng 148720dd2cbfSMiao Xie if (!trans->reloc_reserved) { 148820dd2cbfSMiao Xie rsv = trans->block_rsv; 14893fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14903fd0a558SYan, Zheng clear_rsv = 1; 14913fd0a558SYan, Zheng } 14923fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14933fd0a558SYan, Zheng if (clear_rsv) 149420dd2cbfSMiao Xie trans->block_rsv = rsv; 14955d4f98a2SYan Zheng 1496ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1497ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14985d4f98a2SYan Zheng root->reloc_root = reloc_root; 14995d4f98a2SYan Zheng return 0; 15005d4f98a2SYan Zheng } 15015d4f98a2SYan Zheng 15025d4f98a2SYan Zheng /* 15035d4f98a2SYan Zheng * update root item of reloc tree 15045d4f98a2SYan Zheng */ 15055d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 15065d4f98a2SYan Zheng struct btrfs_root *root) 15075d4f98a2SYan Zheng { 15080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15095d4f98a2SYan Zheng struct btrfs_root *reloc_root; 15105d4f98a2SYan Zheng struct btrfs_root_item *root_item; 15115d4f98a2SYan Zheng int ret; 15125d4f98a2SYan Zheng 15136282675eSQu Wenruo if (!have_reloc_root(root)) 15147585717fSChris Mason goto out; 15155d4f98a2SYan Zheng 15165d4f98a2SYan Zheng reloc_root = root->reloc_root; 15175d4f98a2SYan Zheng root_item = &reloc_root->root_item; 15185d4f98a2SYan Zheng 1519d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 15200b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 15213fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 1522d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 15236282675eSQu Wenruo /* 15246282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 15256282675eSQu Wenruo * have_reloc_root will not touch it from now on. 15266282675eSQu Wenruo */ 15276282675eSQu Wenruo smp_wmb(); 1528c974c464SWang Shilong __del_reloc_root(reloc_root); 15295d4f98a2SYan Zheng } 15305d4f98a2SYan Zheng 15315d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 15325d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 15335d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 15345d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 15355d4f98a2SYan Zheng } 15365d4f98a2SYan Zheng 15370b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 15385d4f98a2SYan Zheng &reloc_root->root_key, root_item); 15395d4f98a2SYan Zheng BUG_ON(ret); 15407585717fSChris Mason 15417585717fSChris Mason out: 15425d4f98a2SYan Zheng return 0; 15435d4f98a2SYan Zheng } 15445d4f98a2SYan Zheng 15455d4f98a2SYan Zheng /* 15465d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15475d4f98a2SYan Zheng * in a subvolume 15485d4f98a2SYan Zheng */ 15495d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15505d4f98a2SYan Zheng { 15515d4f98a2SYan Zheng struct rb_node *node; 15525d4f98a2SYan Zheng struct rb_node *prev; 15535d4f98a2SYan Zheng struct btrfs_inode *entry; 15545d4f98a2SYan Zheng struct inode *inode; 15555d4f98a2SYan Zheng 15565d4f98a2SYan Zheng spin_lock(&root->inode_lock); 15575d4f98a2SYan Zheng again: 15585d4f98a2SYan Zheng node = root->inode_tree.rb_node; 15595d4f98a2SYan Zheng prev = NULL; 15605d4f98a2SYan Zheng while (node) { 15615d4f98a2SYan Zheng prev = node; 15625d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15635d4f98a2SYan Zheng 15644a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 15655d4f98a2SYan Zheng node = node->rb_left; 15664a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 15675d4f98a2SYan Zheng node = node->rb_right; 15685d4f98a2SYan Zheng else 15695d4f98a2SYan Zheng break; 15705d4f98a2SYan Zheng } 15715d4f98a2SYan Zheng if (!node) { 15725d4f98a2SYan Zheng while (prev) { 15735d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 15744a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 15755d4f98a2SYan Zheng node = prev; 15765d4f98a2SYan Zheng break; 15775d4f98a2SYan Zheng } 15785d4f98a2SYan Zheng prev = rb_next(prev); 15795d4f98a2SYan Zheng } 15805d4f98a2SYan Zheng } 15815d4f98a2SYan Zheng while (node) { 15825d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15835d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15845d4f98a2SYan Zheng if (inode) { 15855d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15865d4f98a2SYan Zheng return inode; 15875d4f98a2SYan Zheng } 15885d4f98a2SYan Zheng 15894a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 15905d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15915d4f98a2SYan Zheng goto again; 15925d4f98a2SYan Zheng 15935d4f98a2SYan Zheng node = rb_next(node); 15945d4f98a2SYan Zheng } 15955d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15965d4f98a2SYan Zheng return NULL; 15975d4f98a2SYan Zheng } 15985d4f98a2SYan Zheng 159932da5386SDavid Sterba static int in_block_group(u64 bytenr, struct btrfs_block_group *block_group) 16005d4f98a2SYan Zheng { 1601b3470b5dSDavid Sterba if (bytenr >= block_group->start && 1602b3470b5dSDavid Sterba bytenr < block_group->start + block_group->length) 16035d4f98a2SYan Zheng return 1; 16045d4f98a2SYan Zheng return 0; 16055d4f98a2SYan Zheng } 16065d4f98a2SYan Zheng 16075d4f98a2SYan Zheng /* 16085d4f98a2SYan Zheng * get new location of data 16095d4f98a2SYan Zheng */ 16105d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 16115d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 16125d4f98a2SYan Zheng { 16135d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 16145d4f98a2SYan Zheng struct btrfs_path *path; 16155d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16165d4f98a2SYan Zheng struct extent_buffer *leaf; 16175d4f98a2SYan Zheng int ret; 16185d4f98a2SYan Zheng 16195d4f98a2SYan Zheng path = btrfs_alloc_path(); 16205d4f98a2SYan Zheng if (!path) 16215d4f98a2SYan Zheng return -ENOMEM; 16225d4f98a2SYan Zheng 16235d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1624f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1625f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 16265d4f98a2SYan Zheng if (ret < 0) 16275d4f98a2SYan Zheng goto out; 16285d4f98a2SYan Zheng if (ret > 0) { 16295d4f98a2SYan Zheng ret = -ENOENT; 16305d4f98a2SYan Zheng goto out; 16315d4f98a2SYan Zheng } 16325d4f98a2SYan Zheng 16335d4f98a2SYan Zheng leaf = path->nodes[0]; 16345d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 16355d4f98a2SYan Zheng struct btrfs_file_extent_item); 16365d4f98a2SYan Zheng 16375d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 16385d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 16395d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 16405d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 16415d4f98a2SYan Zheng 16425d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 164383d4cfd4SJosef Bacik ret = -EINVAL; 16445d4f98a2SYan Zheng goto out; 16455d4f98a2SYan Zheng } 16465d4f98a2SYan Zheng 16475d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16485d4f98a2SYan Zheng ret = 0; 16495d4f98a2SYan Zheng out: 16505d4f98a2SYan Zheng btrfs_free_path(path); 16515d4f98a2SYan Zheng return ret; 16525d4f98a2SYan Zheng } 16535d4f98a2SYan Zheng 16545d4f98a2SYan Zheng /* 16555d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16565d4f98a2SYan Zheng * the new locations. 16575d4f98a2SYan Zheng */ 16583fd0a558SYan, Zheng static noinline_for_stack 16593fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 16605d4f98a2SYan Zheng struct reloc_control *rc, 16615d4f98a2SYan Zheng struct btrfs_root *root, 16623fd0a558SYan, Zheng struct extent_buffer *leaf) 16635d4f98a2SYan Zheng { 16640b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 16655d4f98a2SYan Zheng struct btrfs_key key; 16665d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16675d4f98a2SYan Zheng struct inode *inode = NULL; 16685d4f98a2SYan Zheng u64 parent; 16695d4f98a2SYan Zheng u64 bytenr; 16703fd0a558SYan, Zheng u64 new_bytenr = 0; 16715d4f98a2SYan Zheng u64 num_bytes; 16725d4f98a2SYan Zheng u64 end; 16735d4f98a2SYan Zheng u32 nritems; 16745d4f98a2SYan Zheng u32 i; 167583d4cfd4SJosef Bacik int ret = 0; 16765d4f98a2SYan Zheng int first = 1; 16775d4f98a2SYan Zheng int dirty = 0; 16785d4f98a2SYan Zheng 16795d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16805d4f98a2SYan Zheng return 0; 16815d4f98a2SYan Zheng 16825d4f98a2SYan Zheng /* reloc trees always use full backref */ 16835d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16845d4f98a2SYan Zheng parent = leaf->start; 16855d4f98a2SYan Zheng else 16865d4f98a2SYan Zheng parent = 0; 16875d4f98a2SYan Zheng 16885d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16895d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 169082fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 169182fa113fSQu Wenruo 16925d4f98a2SYan Zheng cond_resched(); 16935d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16945d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16955d4f98a2SYan Zheng continue; 16965d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16975d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16985d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16995d4f98a2SYan Zheng continue; 17005d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 17015d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 17025d4f98a2SYan Zheng if (bytenr == 0) 17035d4f98a2SYan Zheng continue; 17045d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 17055d4f98a2SYan Zheng continue; 17065d4f98a2SYan Zheng 17075d4f98a2SYan Zheng /* 17085d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 17095d4f98a2SYan Zheng * to complete and drop the extent cache 17105d4f98a2SYan Zheng */ 17115d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 17125d4f98a2SYan Zheng if (first) { 17135d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17145d4f98a2SYan Zheng first = 0; 17154a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 17163fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 17175d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17185d4f98a2SYan Zheng } 17194a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 17205d4f98a2SYan Zheng end = key.offset + 17215d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 17225d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 17230b246afaSJeff Mahoney fs_info->sectorsize)); 17240b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 17255d4f98a2SYan Zheng end--; 17265d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1727d0082371SJeff Mahoney key.offset, end); 17285d4f98a2SYan Zheng if (!ret) 17295d4f98a2SYan Zheng continue; 17305d4f98a2SYan Zheng 1731dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1732dcdbc059SNikolay Borisov key.offset, end, 1); 17335d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1734d0082371SJeff Mahoney key.offset, end); 17355d4f98a2SYan Zheng } 17365d4f98a2SYan Zheng } 17375d4f98a2SYan Zheng 17385d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 17395d4f98a2SYan Zheng bytenr, num_bytes); 174083d4cfd4SJosef Bacik if (ret) { 174183d4cfd4SJosef Bacik /* 174283d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 174383d4cfd4SJosef Bacik * in the file extent yet. 174483d4cfd4SJosef Bacik */ 174583d4cfd4SJosef Bacik break; 17463fd0a558SYan, Zheng } 17475d4f98a2SYan Zheng 17485d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17495d4f98a2SYan Zheng dirty = 1; 17505d4f98a2SYan Zheng 17515d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 175282fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 175382fa113fSQu Wenruo num_bytes, parent); 175482fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 175582fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1756b06c4bf5SFilipe Manana key.objectid, key.offset); 175782fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 175883d4cfd4SJosef Bacik if (ret) { 175966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 176083d4cfd4SJosef Bacik break; 176183d4cfd4SJosef Bacik } 17625d4f98a2SYan Zheng 1763ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1764ffd4bb2aSQu Wenruo num_bytes, parent); 1765ffd4bb2aSQu Wenruo ref.real_root = root->root_key.objectid; 1766ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1767b06c4bf5SFilipe Manana key.objectid, key.offset); 1768ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 176983d4cfd4SJosef Bacik if (ret) { 177066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 177183d4cfd4SJosef Bacik break; 177283d4cfd4SJosef Bacik } 17735d4f98a2SYan Zheng } 17745d4f98a2SYan Zheng if (dirty) 17755d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17763fd0a558SYan, Zheng if (inode) 17773fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 177883d4cfd4SJosef Bacik return ret; 17795d4f98a2SYan Zheng } 17805d4f98a2SYan Zheng 17815d4f98a2SYan Zheng static noinline_for_stack 17825d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17835d4f98a2SYan Zheng struct btrfs_path *path, int level) 17845d4f98a2SYan Zheng { 17855d4f98a2SYan Zheng struct btrfs_disk_key key1; 17865d4f98a2SYan Zheng struct btrfs_disk_key key2; 17875d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17885d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17895d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17905d4f98a2SYan Zheng } 17915d4f98a2SYan Zheng 17925d4f98a2SYan Zheng /* 17935d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17945d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17955d4f98a2SYan Zheng * reloc tree was create can be replaced. 17965d4f98a2SYan Zheng * 17975d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17985d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17995d4f98a2SYan Zheng * errors, a negative error number is returned. 18005d4f98a2SYan Zheng */ 18013fd0a558SYan, Zheng static noinline_for_stack 18023d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 18035d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 18045d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 18055d4f98a2SYan Zheng int lowest_level, int max_level) 18065d4f98a2SYan Zheng { 18070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 18085d4f98a2SYan Zheng struct extent_buffer *eb; 18095d4f98a2SYan Zheng struct extent_buffer *parent; 181082fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 18115d4f98a2SYan Zheng struct btrfs_key key; 18125d4f98a2SYan Zheng u64 old_bytenr; 18135d4f98a2SYan Zheng u64 new_bytenr; 18145d4f98a2SYan Zheng u64 old_ptr_gen; 18155d4f98a2SYan Zheng u64 new_ptr_gen; 18165d4f98a2SYan Zheng u64 last_snapshot; 18175d4f98a2SYan Zheng u32 blocksize; 18183fd0a558SYan, Zheng int cow = 0; 18195d4f98a2SYan Zheng int level; 18205d4f98a2SYan Zheng int ret; 18215d4f98a2SYan Zheng int slot; 18225d4f98a2SYan Zheng 18235d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 18245d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 18255d4f98a2SYan Zheng 18265d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 18273fd0a558SYan, Zheng again: 18285d4f98a2SYan Zheng slot = path->slots[lowest_level]; 18295d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 18305d4f98a2SYan Zheng 18315d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 18328bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18335d4f98a2SYan Zheng level = btrfs_header_level(eb); 18345d4f98a2SYan Zheng 18355d4f98a2SYan Zheng if (level < lowest_level) { 18365d4f98a2SYan Zheng btrfs_tree_unlock(eb); 18375d4f98a2SYan Zheng free_extent_buffer(eb); 18385d4f98a2SYan Zheng return 0; 18395d4f98a2SYan Zheng } 18405d4f98a2SYan Zheng 18413fd0a558SYan, Zheng if (cow) { 18425d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18435d4f98a2SYan Zheng BUG_ON(ret); 18443fd0a558SYan, Zheng } 18458bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18465d4f98a2SYan Zheng 18475d4f98a2SYan Zheng if (next_key) { 18485d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18495d4f98a2SYan Zheng next_key->type = (u8)-1; 18505d4f98a2SYan Zheng next_key->offset = (u64)-1; 18515d4f98a2SYan Zheng } 18525d4f98a2SYan Zheng 18535d4f98a2SYan Zheng parent = eb; 18545d4f98a2SYan Zheng while (1) { 1855581c1760SQu Wenruo struct btrfs_key first_key; 1856581c1760SQu Wenruo 18575d4f98a2SYan Zheng level = btrfs_header_level(parent); 18585d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18595d4f98a2SYan Zheng 18605d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 1861cbca7d59SFilipe Manana if (ret < 0) 1862cbca7d59SFilipe Manana break; 18635d4f98a2SYan Zheng if (ret && slot > 0) 18645d4f98a2SYan Zheng slot--; 18655d4f98a2SYan Zheng 18665d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18675d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18685d4f98a2SYan Zheng 18695d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 18700b246afaSJeff Mahoney blocksize = fs_info->nodesize; 18715d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 187217515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 18735d4f98a2SYan Zheng 18745d4f98a2SYan Zheng if (level <= max_level) { 18755d4f98a2SYan Zheng eb = path->nodes[level]; 18765d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18775d4f98a2SYan Zheng path->slots[level]); 18785d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18795d4f98a2SYan Zheng path->slots[level]); 18805d4f98a2SYan Zheng } else { 18815d4f98a2SYan Zheng new_bytenr = 0; 18825d4f98a2SYan Zheng new_ptr_gen = 0; 18835d4f98a2SYan Zheng } 18845d4f98a2SYan Zheng 1885fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18865d4f98a2SYan Zheng ret = level; 18875d4f98a2SYan Zheng break; 18885d4f98a2SYan Zheng } 18895d4f98a2SYan Zheng 18905d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18915d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18923fd0a558SYan, Zheng if (level <= lowest_level) { 18935d4f98a2SYan Zheng ret = 0; 18945d4f98a2SYan Zheng break; 18955d4f98a2SYan Zheng } 18965d4f98a2SYan Zheng 1897581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 1898581c1760SQu Wenruo level - 1, &first_key); 189964c043deSLiu Bo if (IS_ERR(eb)) { 190064c043deSLiu Bo ret = PTR_ERR(eb); 1901264813acSLiu Bo break; 190264c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 190364c043deSLiu Bo ret = -EIO; 1904416bc658SJosef Bacik free_extent_buffer(eb); 1905379cde74SStefan Behrens break; 1906416bc658SJosef Bacik } 19075d4f98a2SYan Zheng btrfs_tree_lock(eb); 19083fd0a558SYan, Zheng if (cow) { 19095d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 19105d4f98a2SYan Zheng slot, &eb); 19115d4f98a2SYan Zheng BUG_ON(ret); 19125d4f98a2SYan Zheng } 19138bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 19145d4f98a2SYan Zheng 19155d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19165d4f98a2SYan Zheng free_extent_buffer(parent); 19175d4f98a2SYan Zheng 19185d4f98a2SYan Zheng parent = eb; 19195d4f98a2SYan Zheng continue; 19205d4f98a2SYan Zheng } 19215d4f98a2SYan Zheng 19223fd0a558SYan, Zheng if (!cow) { 19233fd0a558SYan, Zheng btrfs_tree_unlock(parent); 19243fd0a558SYan, Zheng free_extent_buffer(parent); 19253fd0a558SYan, Zheng cow = 1; 19263fd0a558SYan, Zheng goto again; 19273fd0a558SYan, Zheng } 19283fd0a558SYan, Zheng 19295d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 19305d4f98a2SYan Zheng path->slots[level]); 1931b3b4aa74SDavid Sterba btrfs_release_path(path); 19325d4f98a2SYan Zheng 19335d4f98a2SYan Zheng path->lowest_level = level; 19345d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 19355d4f98a2SYan Zheng path->lowest_level = 0; 19365d4f98a2SYan Zheng BUG_ON(ret); 19375d4f98a2SYan Zheng 19385d4f98a2SYan Zheng /* 1939824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1940824d8dffSQu Wenruo * 1941824d8dffSQu Wenruo * We must trace both trees. 1942824d8dffSQu Wenruo * 1) Tree reloc subtree 1943824d8dffSQu Wenruo * If not traced, we will leak data numbers 1944824d8dffSQu Wenruo * 2) Fs subtree 1945824d8dffSQu Wenruo * If not traced, we will double count old data 1946f616f5cdSQu Wenruo * 1947f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1948f616f5cdSQu Wenruo * the swapped tree blocks. 1949f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1950f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1951824d8dffSQu Wenruo */ 1952370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1953370a11b8SQu Wenruo rc->block_group, parent, slot, 1954370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1955370a11b8SQu Wenruo last_snapshot); 1956370a11b8SQu Wenruo if (ret < 0) 1957370a11b8SQu Wenruo break; 1958824d8dffSQu Wenruo /* 19595d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 19605d4f98a2SYan Zheng */ 19615d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 19625d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19635d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19645d4f98a2SYan Zheng 19655d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19665d4f98a2SYan Zheng path->slots[level], old_bytenr); 19675d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19685d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19695d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19705d4f98a2SYan Zheng 197182fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 197282fa113fSQu Wenruo blocksize, path->nodes[level]->start); 197382fa113fSQu Wenruo ref.skip_qgroup = true; 197482fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 197582fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 19765d4f98a2SYan Zheng BUG_ON(ret); 197782fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 197882fa113fSQu Wenruo blocksize, 0); 197982fa113fSQu Wenruo ref.skip_qgroup = true; 198082fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 198182fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 19825d4f98a2SYan Zheng BUG_ON(ret); 19835d4f98a2SYan Zheng 1984ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1985ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1986ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 1987ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1988ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 19895d4f98a2SYan Zheng BUG_ON(ret); 19905d4f98a2SYan Zheng 1991ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1992ffd4bb2aSQu Wenruo blocksize, 0); 1993ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 1994ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1995ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 19965d4f98a2SYan Zheng BUG_ON(ret); 19975d4f98a2SYan Zheng 19985d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19995d4f98a2SYan Zheng 20005d4f98a2SYan Zheng ret = level; 20015d4f98a2SYan Zheng break; 20025d4f98a2SYan Zheng } 20035d4f98a2SYan Zheng btrfs_tree_unlock(parent); 20045d4f98a2SYan Zheng free_extent_buffer(parent); 20055d4f98a2SYan Zheng return ret; 20065d4f98a2SYan Zheng } 20075d4f98a2SYan Zheng 20085d4f98a2SYan Zheng /* 20095d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 20105d4f98a2SYan Zheng */ 20115d4f98a2SYan Zheng static noinline_for_stack 20125d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20135d4f98a2SYan Zheng int *level) 20145d4f98a2SYan Zheng { 20155d4f98a2SYan Zheng struct extent_buffer *eb; 20165d4f98a2SYan Zheng int i; 20175d4f98a2SYan Zheng u64 last_snapshot; 20185d4f98a2SYan Zheng u32 nritems; 20195d4f98a2SYan Zheng 20205d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20215d4f98a2SYan Zheng 20225d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 20235d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20245d4f98a2SYan Zheng path->nodes[i] = NULL; 20255d4f98a2SYan Zheng } 20265d4f98a2SYan Zheng 20275d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 20285d4f98a2SYan Zheng eb = path->nodes[i]; 20295d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20305d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 20315d4f98a2SYan Zheng path->slots[i]++; 20325d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 20335d4f98a2SYan Zheng last_snapshot) 20345d4f98a2SYan Zheng continue; 20355d4f98a2SYan Zheng 20365d4f98a2SYan Zheng *level = i; 20375d4f98a2SYan Zheng return 0; 20385d4f98a2SYan Zheng } 20395d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20405d4f98a2SYan Zheng path->nodes[i] = NULL; 20415d4f98a2SYan Zheng } 20425d4f98a2SYan Zheng return 1; 20435d4f98a2SYan Zheng } 20445d4f98a2SYan Zheng 20455d4f98a2SYan Zheng /* 20465d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 20475d4f98a2SYan Zheng */ 20485d4f98a2SYan Zheng static noinline_for_stack 20495d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20505d4f98a2SYan Zheng int *level) 20515d4f98a2SYan Zheng { 20522ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20535d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 20545d4f98a2SYan Zheng int i; 20555d4f98a2SYan Zheng u64 bytenr; 20565d4f98a2SYan Zheng u64 ptr_gen = 0; 20575d4f98a2SYan Zheng u64 last_snapshot; 20585d4f98a2SYan Zheng u32 nritems; 20595d4f98a2SYan Zheng 20605d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20615d4f98a2SYan Zheng 20625d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2063581c1760SQu Wenruo struct btrfs_key first_key; 2064581c1760SQu Wenruo 20655d4f98a2SYan Zheng eb = path->nodes[i]; 20665d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20675d4f98a2SYan Zheng while (path->slots[i] < nritems) { 20685d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 20695d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 20705d4f98a2SYan Zheng break; 20715d4f98a2SYan Zheng path->slots[i]++; 20725d4f98a2SYan Zheng } 20735d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20745d4f98a2SYan Zheng if (i == *level) 20755d4f98a2SYan Zheng break; 20765d4f98a2SYan Zheng *level = i + 1; 20775d4f98a2SYan Zheng return 0; 20785d4f98a2SYan Zheng } 20795d4f98a2SYan Zheng if (i == 1) { 20805d4f98a2SYan Zheng *level = i; 20815d4f98a2SYan Zheng return 0; 20825d4f98a2SYan Zheng } 20835d4f98a2SYan Zheng 20845d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2085581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2086581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2087581c1760SQu Wenruo &first_key); 208864c043deSLiu Bo if (IS_ERR(eb)) { 208964c043deSLiu Bo return PTR_ERR(eb); 209064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2091416bc658SJosef Bacik free_extent_buffer(eb); 2092416bc658SJosef Bacik return -EIO; 2093416bc658SJosef Bacik } 20945d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20955d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20965d4f98a2SYan Zheng path->slots[i - 1] = 0; 20975d4f98a2SYan Zheng } 20985d4f98a2SYan Zheng return 1; 20995d4f98a2SYan Zheng } 21005d4f98a2SYan Zheng 21015d4f98a2SYan Zheng /* 21025d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 21035d4f98a2SYan Zheng * [min_key, max_key) 21045d4f98a2SYan Zheng */ 21055d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 21065d4f98a2SYan Zheng struct btrfs_key *min_key, 21075d4f98a2SYan Zheng struct btrfs_key *max_key) 21085d4f98a2SYan Zheng { 21090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 21105d4f98a2SYan Zheng struct inode *inode = NULL; 21115d4f98a2SYan Zheng u64 objectid; 21125d4f98a2SYan Zheng u64 start, end; 211333345d01SLi Zefan u64 ino; 21145d4f98a2SYan Zheng 21155d4f98a2SYan Zheng objectid = min_key->objectid; 21165d4f98a2SYan Zheng while (1) { 21175d4f98a2SYan Zheng cond_resched(); 21185d4f98a2SYan Zheng iput(inode); 21195d4f98a2SYan Zheng 21205d4f98a2SYan Zheng if (objectid > max_key->objectid) 21215d4f98a2SYan Zheng break; 21225d4f98a2SYan Zheng 21235d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 21245d4f98a2SYan Zheng if (!inode) 21255d4f98a2SYan Zheng break; 21264a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 21275d4f98a2SYan Zheng 212833345d01SLi Zefan if (ino > max_key->objectid) { 21295d4f98a2SYan Zheng iput(inode); 21305d4f98a2SYan Zheng break; 21315d4f98a2SYan Zheng } 21325d4f98a2SYan Zheng 213333345d01SLi Zefan objectid = ino + 1; 21345d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 21355d4f98a2SYan Zheng continue; 21365d4f98a2SYan Zheng 213733345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 21385d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 21395d4f98a2SYan Zheng continue; 21405d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 21415d4f98a2SYan Zheng start = 0; 21425d4f98a2SYan Zheng else { 21435d4f98a2SYan Zheng start = min_key->offset; 21440b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 21455d4f98a2SYan Zheng } 21465d4f98a2SYan Zheng } else { 21475d4f98a2SYan Zheng start = 0; 21485d4f98a2SYan Zheng } 21495d4f98a2SYan Zheng 215033345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 21515d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 21525d4f98a2SYan Zheng continue; 21535d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 21545d4f98a2SYan Zheng end = (u64)-1; 21555d4f98a2SYan Zheng } else { 21565d4f98a2SYan Zheng if (max_key->offset == 0) 21575d4f98a2SYan Zheng continue; 21585d4f98a2SYan Zheng end = max_key->offset; 21590b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 21605d4f98a2SYan Zheng end--; 21615d4f98a2SYan Zheng } 21625d4f98a2SYan Zheng } else { 21635d4f98a2SYan Zheng end = (u64)-1; 21645d4f98a2SYan Zheng } 21655d4f98a2SYan Zheng 21665d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2167d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2168dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2169d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 21705d4f98a2SYan Zheng } 21715d4f98a2SYan Zheng return 0; 21725d4f98a2SYan Zheng } 21735d4f98a2SYan Zheng 21745d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 21755d4f98a2SYan Zheng struct btrfs_key *key) 21765d4f98a2SYan Zheng 21775d4f98a2SYan Zheng { 21785d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21795d4f98a2SYan Zheng if (!path->nodes[level]) 21805d4f98a2SYan Zheng break; 21815d4f98a2SYan Zheng if (path->slots[level] + 1 < 21825d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21835d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21845d4f98a2SYan Zheng path->slots[level] + 1); 21855d4f98a2SYan Zheng return 0; 21865d4f98a2SYan Zheng } 21875d4f98a2SYan Zheng level++; 21885d4f98a2SYan Zheng } 21895d4f98a2SYan Zheng return 1; 21905d4f98a2SYan Zheng } 21915d4f98a2SYan Zheng 21925d4f98a2SYan Zheng /* 2193d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 2194d2311e69SQu Wenruo */ 2195d2311e69SQu Wenruo static void insert_dirty_subvol(struct btrfs_trans_handle *trans, 2196d2311e69SQu Wenruo struct reloc_control *rc, 2197d2311e69SQu Wenruo struct btrfs_root *root) 2198d2311e69SQu Wenruo { 2199d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2200d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 2201d2311e69SQu Wenruo 2202d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 2203d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 2204d2311e69SQu Wenruo ASSERT(reloc_root); 2205d2311e69SQu Wenruo 2206d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 2207d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 2208d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 2209d2311e69SQu Wenruo reloc_root_item->drop_level = 0; 2210d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 2211d2311e69SQu Wenruo btrfs_update_reloc_root(trans, root); 2212d2311e69SQu Wenruo 2213d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 2214d2311e69SQu Wenruo btrfs_grab_fs_root(root); 2215d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 2216d2311e69SQu Wenruo } 2217d2311e69SQu Wenruo } 2218d2311e69SQu Wenruo 2219d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 2220d2311e69SQu Wenruo { 2221d2311e69SQu Wenruo struct btrfs_root *root; 2222d2311e69SQu Wenruo struct btrfs_root *next; 2223d2311e69SQu Wenruo int ret = 0; 222430d40577SQu Wenruo int ret2; 2225d2311e69SQu Wenruo 2226d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 2227d2311e69SQu Wenruo reloc_dirty_list) { 222830d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 222930d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 2230d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2231d2311e69SQu Wenruo 2232d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 2233d2311e69SQu Wenruo root->reloc_root = NULL; 2234d2311e69SQu Wenruo if (reloc_root) { 2235d2311e69SQu Wenruo 2236d2311e69SQu Wenruo ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1); 2237d2311e69SQu Wenruo if (ret2 < 0 && !ret) 2238d2311e69SQu Wenruo ret = ret2; 2239d2311e69SQu Wenruo } 22406282675eSQu Wenruo /* 22416282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 22426282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 22436282675eSQu Wenruo */ 22446282675eSQu Wenruo smp_wmb(); 22451fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 2246d2311e69SQu Wenruo btrfs_put_fs_root(root); 224730d40577SQu Wenruo } else { 224830d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 224930d40577SQu Wenruo ret2 = btrfs_drop_snapshot(root, NULL, 0, 1); 225030d40577SQu Wenruo if (ret2 < 0 && !ret) 225130d40577SQu Wenruo ret = ret2; 225230d40577SQu Wenruo } 2253d2311e69SQu Wenruo } 2254d2311e69SQu Wenruo return ret; 2255d2311e69SQu Wenruo } 2256d2311e69SQu Wenruo 2257d2311e69SQu Wenruo /* 22585d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 22595d4f98a2SYan Zheng * fs tree. 22605d4f98a2SYan Zheng */ 22615d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 22625d4f98a2SYan Zheng struct btrfs_root *root) 22635d4f98a2SYan Zheng { 22640b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 22655d4f98a2SYan Zheng struct btrfs_key key; 22665d4f98a2SYan Zheng struct btrfs_key next_key; 22679e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 22685d4f98a2SYan Zheng struct btrfs_root *reloc_root; 22695d4f98a2SYan Zheng struct btrfs_root_item *root_item; 22705d4f98a2SYan Zheng struct btrfs_path *path; 22713fd0a558SYan, Zheng struct extent_buffer *leaf; 22725d4f98a2SYan Zheng int level; 22735d4f98a2SYan Zheng int max_level; 22745d4f98a2SYan Zheng int replaced = 0; 22755d4f98a2SYan Zheng int ret; 22765d4f98a2SYan Zheng int err = 0; 22773fd0a558SYan, Zheng u32 min_reserved; 22785d4f98a2SYan Zheng 22795d4f98a2SYan Zheng path = btrfs_alloc_path(); 22805d4f98a2SYan Zheng if (!path) 22815d4f98a2SYan Zheng return -ENOMEM; 2282e4058b54SDavid Sterba path->reada = READA_FORWARD; 22835d4f98a2SYan Zheng 22845d4f98a2SYan Zheng reloc_root = root->reloc_root; 22855d4f98a2SYan Zheng root_item = &reloc_root->root_item; 22865d4f98a2SYan Zheng 22875d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 22885d4f98a2SYan Zheng level = btrfs_root_level(root_item); 228967439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 22905d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 22915d4f98a2SYan Zheng path->slots[level] = 0; 22925d4f98a2SYan Zheng } else { 22935d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 22945d4f98a2SYan Zheng 22955d4f98a2SYan Zheng level = root_item->drop_level; 22965d4f98a2SYan Zheng BUG_ON(level == 0); 22975d4f98a2SYan Zheng path->lowest_level = level; 22985d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 229933c66f43SYan Zheng path->lowest_level = 0; 23005d4f98a2SYan Zheng if (ret < 0) { 23015d4f98a2SYan Zheng btrfs_free_path(path); 23025d4f98a2SYan Zheng return ret; 23035d4f98a2SYan Zheng } 23045d4f98a2SYan Zheng 23055d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 23065d4f98a2SYan Zheng path->slots[level]); 23075d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 23085d4f98a2SYan Zheng 23095d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 23105d4f98a2SYan Zheng } 23115d4f98a2SYan Zheng 23120b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23135d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 23145d4f98a2SYan Zheng 23155d4f98a2SYan Zheng while (1) { 231608e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 231708e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23183fd0a558SYan, Zheng if (ret) { 23199e6a0c52SJosef Bacik err = ret; 23209e6a0c52SJosef Bacik goto out; 23213fd0a558SYan, Zheng } 23229e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 23239e6a0c52SJosef Bacik if (IS_ERR(trans)) { 23249e6a0c52SJosef Bacik err = PTR_ERR(trans); 23259e6a0c52SJosef Bacik trans = NULL; 23269e6a0c52SJosef Bacik goto out; 23279e6a0c52SJosef Bacik } 23289e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 23293fd0a558SYan, Zheng 23303fd0a558SYan, Zheng replaced = 0; 23315d4f98a2SYan Zheng max_level = level; 23325d4f98a2SYan Zheng 23335d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 23345d4f98a2SYan Zheng if (ret < 0) { 23355d4f98a2SYan Zheng err = ret; 23365d4f98a2SYan Zheng goto out; 23375d4f98a2SYan Zheng } 23385d4f98a2SYan Zheng if (ret > 0) 23395d4f98a2SYan Zheng break; 23405d4f98a2SYan Zheng 23415d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 23425d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 23435d4f98a2SYan Zheng ret = 0; 23445d4f98a2SYan Zheng } else { 23453d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 23463fd0a558SYan, Zheng &next_key, level, max_level); 23475d4f98a2SYan Zheng } 23485d4f98a2SYan Zheng if (ret < 0) { 23495d4f98a2SYan Zheng err = ret; 23505d4f98a2SYan Zheng goto out; 23515d4f98a2SYan Zheng } 23525d4f98a2SYan Zheng 23535d4f98a2SYan Zheng if (ret > 0) { 23545d4f98a2SYan Zheng level = ret; 23555d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 23565d4f98a2SYan Zheng path->slots[level]); 23575d4f98a2SYan Zheng replaced = 1; 23585d4f98a2SYan Zheng } 23595d4f98a2SYan Zheng 23605d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 23615d4f98a2SYan Zheng if (ret > 0) 23625d4f98a2SYan Zheng break; 23635d4f98a2SYan Zheng 23645d4f98a2SYan Zheng BUG_ON(level == 0); 23655d4f98a2SYan Zheng /* 23665d4f98a2SYan Zheng * save the merging progress in the drop_progress. 23675d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 23685d4f98a2SYan Zheng */ 23695d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 23705d4f98a2SYan Zheng path->slots[level]); 23715d4f98a2SYan Zheng root_item->drop_level = level; 23725d4f98a2SYan Zheng 23733a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23749e6a0c52SJosef Bacik trans = NULL; 23755d4f98a2SYan Zheng 23762ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23775d4f98a2SYan Zheng 23785d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23795d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23805d4f98a2SYan Zheng } 23815d4f98a2SYan Zheng 23825d4f98a2SYan Zheng /* 23835d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 23845d4f98a2SYan Zheng * relocated and the block is tree root. 23855d4f98a2SYan Zheng */ 23865d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 23875d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 23885d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 23895d4f98a2SYan Zheng free_extent_buffer(leaf); 23905d4f98a2SYan Zheng if (ret < 0) 23915d4f98a2SYan Zheng err = ret; 23925d4f98a2SYan Zheng out: 23935d4f98a2SYan Zheng btrfs_free_path(path); 23945d4f98a2SYan Zheng 2395d2311e69SQu Wenruo if (err == 0) 2396d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 23975d4f98a2SYan Zheng 23989e6a0c52SJosef Bacik if (trans) 23993a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 24005d4f98a2SYan Zheng 24012ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 24025d4f98a2SYan Zheng 24035d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 24045d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 24055d4f98a2SYan Zheng 24065d4f98a2SYan Zheng return err; 24075d4f98a2SYan Zheng } 24085d4f98a2SYan Zheng 24093fd0a558SYan, Zheng static noinline_for_stack 24103fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 24115d4f98a2SYan Zheng { 24123fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 24130b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 24143fd0a558SYan, Zheng struct btrfs_root *reloc_root; 24155d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 24163fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24173fd0a558SYan, Zheng u64 num_bytes = 0; 24183fd0a558SYan, Zheng int ret; 24193fd0a558SYan, Zheng 24200b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24210b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 24223fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 24230b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24247585717fSChris Mason 24253fd0a558SYan, Zheng again: 24263fd0a558SYan, Zheng if (!err) { 24273fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 242808e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 242908e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 24303fd0a558SYan, Zheng if (ret) 24313fd0a558SYan, Zheng err = ret; 24323fd0a558SYan, Zheng } 24333fd0a558SYan, Zheng 24347a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 24353612b495STsutomu Itoh if (IS_ERR(trans)) { 24363612b495STsutomu Itoh if (!err) 24372ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 24382ff7e61eSJeff Mahoney num_bytes); 24393612b495STsutomu Itoh return PTR_ERR(trans); 24403612b495STsutomu Itoh } 24413fd0a558SYan, Zheng 24423fd0a558SYan, Zheng if (!err) { 24433fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 24443a45bb20SJeff Mahoney btrfs_end_transaction(trans); 24452ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 24462ff7e61eSJeff Mahoney num_bytes); 24473fd0a558SYan, Zheng goto again; 24483fd0a558SYan, Zheng } 24493fd0a558SYan, Zheng } 24503fd0a558SYan, Zheng 24513fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 24523fd0a558SYan, Zheng 24533fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 24543fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 24553fd0a558SYan, Zheng struct btrfs_root, root_list); 24563fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24573fd0a558SYan, Zheng 24580b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 24593fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 24603fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 24613fd0a558SYan, Zheng 24623fd0a558SYan, Zheng /* 24633fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 24643fd0a558SYan, Zheng * knows it should resumes merging 24653fd0a558SYan, Zheng */ 24663fd0a558SYan, Zheng if (!err) 24673fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 24683fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 24693fd0a558SYan, Zheng 24703fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 24713fd0a558SYan, Zheng } 24723fd0a558SYan, Zheng 24733fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 24743fd0a558SYan, Zheng 24753fd0a558SYan, Zheng if (!err) 24763a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 24773fd0a558SYan, Zheng else 24783a45bb20SJeff Mahoney btrfs_end_transaction(trans); 24793fd0a558SYan, Zheng return err; 24803fd0a558SYan, Zheng } 24813fd0a558SYan, Zheng 24823fd0a558SYan, Zheng static noinline_for_stack 2483aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2484aca1bba6SLiu Bo { 2485aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2486aca1bba6SLiu Bo 2487aca1bba6SLiu Bo while (!list_empty(list)) { 2488aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2489aca1bba6SLiu Bo root_list); 2490bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 24916bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 24926bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 24936bdf131fSJosef Bacik reloc_root->node = NULL; 24946bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2495aca1bba6SLiu Bo } 2496aca1bba6SLiu Bo } 2497aca1bba6SLiu Bo 2498aca1bba6SLiu Bo static noinline_for_stack 249994404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 25003fd0a558SYan, Zheng { 25010b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 25025d4f98a2SYan Zheng struct btrfs_root *root; 25035d4f98a2SYan Zheng struct btrfs_root *reloc_root; 25043fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 25053fd0a558SYan, Zheng int found = 0; 2506aca1bba6SLiu Bo int ret = 0; 25073fd0a558SYan, Zheng again: 25083fd0a558SYan, Zheng root = rc->extent_root; 25097585717fSChris Mason 25107585717fSChris Mason /* 25117585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 25127585717fSChris Mason * we have to make sure nobody is in the middle of 25137585717fSChris Mason * adding their roots to the list while we are 25147585717fSChris Mason * doing this splice 25157585717fSChris Mason */ 25160b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 25173fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 25180b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 25195d4f98a2SYan Zheng 25203fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 25213fd0a558SYan, Zheng found = 1; 25223fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 25233fd0a558SYan, Zheng struct btrfs_root, root_list); 25245d4f98a2SYan Zheng 25255d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 25260b246afaSJeff Mahoney root = read_fs_root(fs_info, 25275d4f98a2SYan Zheng reloc_root->root_key.offset); 25285d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25295d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25305d4f98a2SYan Zheng 25313fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2532b37b39cdSJosef Bacik if (ret) { 253325e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 253425e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 253525e293c2SWang Shilong &reloc_roots); 2536aca1bba6SLiu Bo goto out; 2537b37b39cdSJosef Bacik } 25383fd0a558SYan, Zheng } else { 25393fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 254030d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 254130d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 254230d40577SQu Wenruo &rc->dirty_subvol_roots); 25433fd0a558SYan, Zheng } 25445d4f98a2SYan Zheng } 25455d4f98a2SYan Zheng 25463fd0a558SYan, Zheng if (found) { 25473fd0a558SYan, Zheng found = 0; 25483fd0a558SYan, Zheng goto again; 25495d4f98a2SYan Zheng } 2550aca1bba6SLiu Bo out: 2551aca1bba6SLiu Bo if (ret) { 25520b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2553aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2554aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2555467bb1d2SWang Shilong 2556467bb1d2SWang Shilong /* new reloc root may be added */ 25570b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2558467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 25590b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2560467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2561467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2562aca1bba6SLiu Bo } 2563aca1bba6SLiu Bo 25645d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 25655d4f98a2SYan Zheng } 25665d4f98a2SYan Zheng 25675d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 25685d4f98a2SYan Zheng { 25695d4f98a2SYan Zheng struct tree_block *block; 25705d4f98a2SYan Zheng struct rb_node *rb_node; 25715d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 25725d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 25735d4f98a2SYan Zheng rb_erase(rb_node, blocks); 25745d4f98a2SYan Zheng kfree(block); 25755d4f98a2SYan Zheng } 25765d4f98a2SYan Zheng } 25775d4f98a2SYan Zheng 25785d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 25795d4f98a2SYan Zheng struct btrfs_root *reloc_root) 25805d4f98a2SYan Zheng { 25810b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 25825d4f98a2SYan Zheng struct btrfs_root *root; 25835d4f98a2SYan Zheng 25845d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 25855d4f98a2SYan Zheng return 0; 25865d4f98a2SYan Zheng 25870b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25885d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25895d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25905d4f98a2SYan Zheng 25915d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 25925d4f98a2SYan Zheng } 25935d4f98a2SYan Zheng 25943fd0a558SYan, Zheng static noinline_for_stack 25953fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 25963fd0a558SYan, Zheng struct reloc_control *rc, 25975d4f98a2SYan Zheng struct backref_node *node, 2598dc4103f9SWang Shilong struct backref_edge *edges[]) 25995d4f98a2SYan Zheng { 26005d4f98a2SYan Zheng struct backref_node *next; 26015d4f98a2SYan Zheng struct btrfs_root *root; 26023fd0a558SYan, Zheng int index = 0; 26033fd0a558SYan, Zheng 26045d4f98a2SYan Zheng next = node; 26055d4f98a2SYan Zheng while (1) { 26065d4f98a2SYan Zheng cond_resched(); 26075d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 26085d4f98a2SYan Zheng root = next->root; 26093fd0a558SYan, Zheng BUG_ON(!root); 261027cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 26115d4f98a2SYan Zheng 26125d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 26135d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 26145d4f98a2SYan Zheng break; 26155d4f98a2SYan Zheng } 26165d4f98a2SYan Zheng 26175d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 26183fd0a558SYan, Zheng root = root->reloc_root; 26193fd0a558SYan, Zheng 26203fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 26213fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 26223fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 26233fd0a558SYan, Zheng next->new_bytenr = root->node->start; 26243fd0a558SYan, Zheng next->root = root; 26253fd0a558SYan, Zheng list_add_tail(&next->list, 26263fd0a558SYan, Zheng &rc->backref_cache.changed); 26273fd0a558SYan, Zheng __mark_block_processed(rc, next); 26285d4f98a2SYan Zheng break; 26295d4f98a2SYan Zheng } 26305d4f98a2SYan Zheng 26313fd0a558SYan, Zheng WARN_ON(1); 26325d4f98a2SYan Zheng root = NULL; 26335d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 26345d4f98a2SYan Zheng if (!next || next->level <= node->level) 26355d4f98a2SYan Zheng break; 26365d4f98a2SYan Zheng } 26373fd0a558SYan, Zheng if (!root) 26383fd0a558SYan, Zheng return NULL; 26395d4f98a2SYan Zheng 26403fd0a558SYan, Zheng next = node; 26413fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 26423fd0a558SYan, Zheng while (1) { 26433fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 26443fd0a558SYan, Zheng if (--index < 0) 26453fd0a558SYan, Zheng break; 26463fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 26473fd0a558SYan, Zheng } 26485d4f98a2SYan Zheng return root; 26495d4f98a2SYan Zheng } 26505d4f98a2SYan Zheng 26513fd0a558SYan, Zheng /* 26523fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 26533fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 26543fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 26553fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 26563fd0a558SYan, Zheng */ 26575d4f98a2SYan Zheng static noinline_for_stack 2658147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 26595d4f98a2SYan Zheng { 26603fd0a558SYan, Zheng struct backref_node *next; 26613fd0a558SYan, Zheng struct btrfs_root *root; 26623fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 26635d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26643fd0a558SYan, Zheng int index = 0; 26653fd0a558SYan, Zheng 26663fd0a558SYan, Zheng next = node; 26673fd0a558SYan, Zheng while (1) { 26683fd0a558SYan, Zheng cond_resched(); 26693fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 26703fd0a558SYan, Zheng root = next->root; 26713fd0a558SYan, Zheng BUG_ON(!root); 26723fd0a558SYan, Zheng 267325985edcSLucas De Marchi /* no other choice for non-references counted tree */ 267427cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 26753fd0a558SYan, Zheng return root; 26763fd0a558SYan, Zheng 26773fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 26783fd0a558SYan, Zheng fs_root = root; 26793fd0a558SYan, Zheng 26803fd0a558SYan, Zheng if (next != node) 26813fd0a558SYan, Zheng return NULL; 26823fd0a558SYan, Zheng 26833fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26843fd0a558SYan, Zheng if (!next || next->level <= node->level) 26853fd0a558SYan, Zheng break; 26863fd0a558SYan, Zheng } 26873fd0a558SYan, Zheng 26883fd0a558SYan, Zheng if (!fs_root) 26893fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 26903fd0a558SYan, Zheng return fs_root; 26915d4f98a2SYan Zheng } 26925d4f98a2SYan Zheng 26935d4f98a2SYan Zheng static noinline_for_stack 26943fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 26953fd0a558SYan, Zheng struct backref_node *node, int reserve) 26965d4f98a2SYan Zheng { 26970b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 26983fd0a558SYan, Zheng struct backref_node *next = node; 26993fd0a558SYan, Zheng struct backref_edge *edge; 27003fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27013fd0a558SYan, Zheng u64 num_bytes = 0; 27023fd0a558SYan, Zheng int index = 0; 27035d4f98a2SYan Zheng 27043fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 27053fd0a558SYan, Zheng 27063fd0a558SYan, Zheng while (next) { 27073fd0a558SYan, Zheng cond_resched(); 27085d4f98a2SYan Zheng while (1) { 27093fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 27105d4f98a2SYan Zheng break; 27115d4f98a2SYan Zheng 27120b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 27133fd0a558SYan, Zheng 27143fd0a558SYan, Zheng if (list_empty(&next->upper)) 27153fd0a558SYan, Zheng break; 27163fd0a558SYan, Zheng 27173fd0a558SYan, Zheng edge = list_entry(next->upper.next, 27183fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 27193fd0a558SYan, Zheng edges[index++] = edge; 27203fd0a558SYan, Zheng next = edge->node[UPPER]; 27215d4f98a2SYan Zheng } 27223fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 27233fd0a558SYan, Zheng } 27243fd0a558SYan, Zheng return num_bytes; 27253fd0a558SYan, Zheng } 27263fd0a558SYan, Zheng 27273fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 27283fd0a558SYan, Zheng struct reloc_control *rc, 27293fd0a558SYan, Zheng struct backref_node *node) 27303fd0a558SYan, Zheng { 27313fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2732da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 27333fd0a558SYan, Zheng u64 num_bytes; 27343fd0a558SYan, Zheng int ret; 27350647bf56SWang Shilong u64 tmp; 27363fd0a558SYan, Zheng 27373fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 27383fd0a558SYan, Zheng 27393fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 27400647bf56SWang Shilong rc->reserved_bytes += num_bytes; 27418ca17f0fSJosef Bacik 27428ca17f0fSJosef Bacik /* 27438ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 27448ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 27458ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 27468ca17f0fSJosef Bacik */ 27470647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 27488ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 27493fd0a558SYan, Zheng if (ret) { 2750da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 27510647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 27520647bf56SWang Shilong tmp <<= 1; 27530647bf56SWang Shilong /* 27540647bf56SWang Shilong * only one thread can access block_rsv at this point, 27550647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 27560647bf56SWang Shilong * we expand more reservation size here to allow enough 275752042d8eSAndrea Gelmini * space for relocation and we will return earlier in 27580647bf56SWang Shilong * enospc case. 27590647bf56SWang Shilong */ 2760da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 27610647bf56SWang Shilong RELOCATION_RESERVED_NODES; 27628ca17f0fSJosef Bacik return -EAGAIN; 27633fd0a558SYan, Zheng } 27643fd0a558SYan, Zheng 27653fd0a558SYan, Zheng return 0; 27663fd0a558SYan, Zheng } 27673fd0a558SYan, Zheng 27685d4f98a2SYan Zheng /* 27695d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 27705d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 27715d4f98a2SYan Zheng * 27725d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 27735d4f98a2SYan Zheng * in that case this function just updates pointers. 27745d4f98a2SYan Zheng */ 27755d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 27763fd0a558SYan, Zheng struct reloc_control *rc, 27775d4f98a2SYan Zheng struct backref_node *node, 27785d4f98a2SYan Zheng struct btrfs_key *key, 27795d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 27805d4f98a2SYan Zheng { 27812ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27825d4f98a2SYan Zheng struct backref_node *upper; 27835d4f98a2SYan Zheng struct backref_edge *edge; 27845d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27855d4f98a2SYan Zheng struct btrfs_root *root; 27865d4f98a2SYan Zheng struct extent_buffer *eb; 27875d4f98a2SYan Zheng u32 blocksize; 27885d4f98a2SYan Zheng u64 bytenr; 27895d4f98a2SYan Zheng u64 generation; 27905d4f98a2SYan Zheng int slot; 27915d4f98a2SYan Zheng int ret; 27925d4f98a2SYan Zheng int err = 0; 27935d4f98a2SYan Zheng 27945d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 27955d4f98a2SYan Zheng 27965d4f98a2SYan Zheng path->lowest_level = node->level + 1; 27973fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 27985d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2799581c1760SQu Wenruo struct btrfs_key first_key; 280082fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2801581c1760SQu Wenruo 28025d4f98a2SYan Zheng cond_resched(); 28035d4f98a2SYan Zheng 28045d4f98a2SYan Zheng upper = edge->node[UPPER]; 2805dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 28063fd0a558SYan, Zheng BUG_ON(!root); 28075d4f98a2SYan Zheng 28083fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 28093fd0a558SYan, Zheng if (!lowest) { 28103fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 28113fd0a558SYan, Zheng upper->level, &slot); 2812cbca7d59SFilipe Manana if (ret < 0) { 2813cbca7d59SFilipe Manana err = ret; 2814cbca7d59SFilipe Manana goto next; 2815cbca7d59SFilipe Manana } 28163fd0a558SYan, Zheng BUG_ON(ret); 28173fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 28183fd0a558SYan, Zheng if (node->eb->start == bytenr) 28193fd0a558SYan, Zheng goto next; 28203fd0a558SYan, Zheng } 28215d4f98a2SYan Zheng drop_node_buffer(upper); 28223fd0a558SYan, Zheng } 28235d4f98a2SYan Zheng 28245d4f98a2SYan Zheng if (!upper->eb) { 28255d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 28263561b9dbSLiu Bo if (ret) { 28273561b9dbSLiu Bo if (ret < 0) 28285d4f98a2SYan Zheng err = ret; 28293561b9dbSLiu Bo else 28303561b9dbSLiu Bo err = -ENOENT; 28313561b9dbSLiu Bo 28323561b9dbSLiu Bo btrfs_release_path(path); 28335d4f98a2SYan Zheng break; 28345d4f98a2SYan Zheng } 28355d4f98a2SYan Zheng 28363fd0a558SYan, Zheng if (!upper->eb) { 28373fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 28383fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 28393fd0a558SYan, Zheng } else { 28403fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 28413fd0a558SYan, Zheng } 28423fd0a558SYan, Zheng 28433fd0a558SYan, Zheng upper->locked = 1; 28443fd0a558SYan, Zheng path->locks[upper->level] = 0; 28453fd0a558SYan, Zheng 28465d4f98a2SYan Zheng slot = path->slots[upper->level]; 2847b3b4aa74SDavid Sterba btrfs_release_path(path); 28485d4f98a2SYan Zheng } else { 28495d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 28505d4f98a2SYan Zheng &slot); 2851cbca7d59SFilipe Manana if (ret < 0) { 2852cbca7d59SFilipe Manana err = ret; 2853cbca7d59SFilipe Manana goto next; 2854cbca7d59SFilipe Manana } 28555d4f98a2SYan Zheng BUG_ON(ret); 28565d4f98a2SYan Zheng } 28575d4f98a2SYan Zheng 28585d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 28593fd0a558SYan, Zheng if (lowest) { 28604547f4d8SLiu Bo if (bytenr != node->bytenr) { 28614547f4d8SLiu Bo btrfs_err(root->fs_info, 28624547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 28634547f4d8SLiu Bo bytenr, node->bytenr, slot, 28644547f4d8SLiu Bo upper->eb->start); 28654547f4d8SLiu Bo err = -EIO; 28664547f4d8SLiu Bo goto next; 28674547f4d8SLiu Bo } 28685d4f98a2SYan Zheng } else { 28693fd0a558SYan, Zheng if (node->eb->start == bytenr) 28703fd0a558SYan, Zheng goto next; 28715d4f98a2SYan Zheng } 28725d4f98a2SYan Zheng 2873da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 28745d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2875581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2876581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2877581c1760SQu Wenruo upper->level - 1, &first_key); 287864c043deSLiu Bo if (IS_ERR(eb)) { 287964c043deSLiu Bo err = PTR_ERR(eb); 288064c043deSLiu Bo goto next; 288164c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2882416bc658SJosef Bacik free_extent_buffer(eb); 288397d9a8a4STsutomu Itoh err = -EIO; 288497d9a8a4STsutomu Itoh goto next; 288597d9a8a4STsutomu Itoh } 28865d4f98a2SYan Zheng btrfs_tree_lock(eb); 28878bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 28885d4f98a2SYan Zheng 28895d4f98a2SYan Zheng if (!node->eb) { 28905d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 28915d4f98a2SYan Zheng slot, &eb); 28923fd0a558SYan, Zheng btrfs_tree_unlock(eb); 28933fd0a558SYan, Zheng free_extent_buffer(eb); 28945d4f98a2SYan Zheng if (ret < 0) { 28955d4f98a2SYan Zheng err = ret; 28963fd0a558SYan, Zheng goto next; 28975d4f98a2SYan Zheng } 28983fd0a558SYan, Zheng BUG_ON(node->eb != eb); 28995d4f98a2SYan Zheng } else { 29005d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 29015d4f98a2SYan Zheng node->eb->start); 29025d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 29035d4f98a2SYan Zheng trans->transid); 29045d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 29055d4f98a2SYan Zheng 290682fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 29075d4f98a2SYan Zheng node->eb->start, blocksize, 290882fa113fSQu Wenruo upper->eb->start); 290982fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 291082fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 291182fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 291282fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 29135d4f98a2SYan Zheng BUG_ON(ret); 29145d4f98a2SYan Zheng 29155d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 29165d4f98a2SYan Zheng BUG_ON(ret); 29175d4f98a2SYan Zheng } 29183fd0a558SYan, Zheng next: 29193fd0a558SYan, Zheng if (!upper->pending) 29203fd0a558SYan, Zheng drop_node_buffer(upper); 29213fd0a558SYan, Zheng else 29223fd0a558SYan, Zheng unlock_node_buffer(upper); 29233fd0a558SYan, Zheng if (err) 29243fd0a558SYan, Zheng break; 29255d4f98a2SYan Zheng } 29263fd0a558SYan, Zheng 29273fd0a558SYan, Zheng if (!err && node->pending) { 29283fd0a558SYan, Zheng drop_node_buffer(node); 29293fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 29303fd0a558SYan, Zheng node->pending = 0; 29315d4f98a2SYan Zheng } 29323fd0a558SYan, Zheng 29335d4f98a2SYan Zheng path->lowest_level = 0; 29343fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 29355d4f98a2SYan Zheng return err; 29365d4f98a2SYan Zheng } 29375d4f98a2SYan Zheng 29385d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 29393fd0a558SYan, Zheng struct reloc_control *rc, 29405d4f98a2SYan Zheng struct backref_node *node, 29415d4f98a2SYan Zheng struct btrfs_path *path) 29425d4f98a2SYan Zheng { 29435d4f98a2SYan Zheng struct btrfs_key key; 29445d4f98a2SYan Zheng 29455d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 29463fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 29475d4f98a2SYan Zheng } 29485d4f98a2SYan Zheng 29495d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 29503fd0a558SYan, Zheng struct reloc_control *rc, 29513fd0a558SYan, Zheng struct btrfs_path *path, int err) 29525d4f98a2SYan Zheng { 29533fd0a558SYan, Zheng LIST_HEAD(list); 29543fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 29555d4f98a2SYan Zheng struct backref_node *node; 29565d4f98a2SYan Zheng int level; 29575d4f98a2SYan Zheng int ret; 29585d4f98a2SYan Zheng 29595d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 29605d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 29615d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 29623fd0a558SYan, Zheng struct backref_node, list); 29633fd0a558SYan, Zheng list_move_tail(&node->list, &list); 29643fd0a558SYan, Zheng BUG_ON(!node->pending); 29655d4f98a2SYan Zheng 29663fd0a558SYan, Zheng if (!err) { 29673fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 29685d4f98a2SYan Zheng if (ret < 0) 29695d4f98a2SYan Zheng err = ret; 29705d4f98a2SYan Zheng } 29715d4f98a2SYan Zheng } 29723fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 29733fd0a558SYan, Zheng } 29745d4f98a2SYan Zheng return err; 29755d4f98a2SYan Zheng } 29765d4f98a2SYan Zheng 29775d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 29783fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 29793fd0a558SYan, Zheng { 29803fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2981ceeb0ae7SDavid Sterba EXTENT_DIRTY); 29823fd0a558SYan, Zheng } 29833fd0a558SYan, Zheng 29843fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 29855d4f98a2SYan Zheng struct backref_node *node) 29865d4f98a2SYan Zheng { 29875d4f98a2SYan Zheng u32 blocksize; 29885d4f98a2SYan Zheng if (node->level == 0 || 29895d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2990da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 29913fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 29925d4f98a2SYan Zheng } 29935d4f98a2SYan Zheng node->processed = 1; 29945d4f98a2SYan Zheng } 29955d4f98a2SYan Zheng 29965d4f98a2SYan Zheng /* 29975d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 29985d4f98a2SYan Zheng * as processed. 29995d4f98a2SYan Zheng */ 30005d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 30015d4f98a2SYan Zheng struct backref_node *node) 30025d4f98a2SYan Zheng { 30035d4f98a2SYan Zheng struct backref_node *next = node; 30045d4f98a2SYan Zheng struct backref_edge *edge; 30055d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 30065d4f98a2SYan Zheng int index = 0; 30075d4f98a2SYan Zheng 30085d4f98a2SYan Zheng while (next) { 30095d4f98a2SYan Zheng cond_resched(); 30105d4f98a2SYan Zheng while (1) { 30115d4f98a2SYan Zheng if (next->processed) 30125d4f98a2SYan Zheng break; 30135d4f98a2SYan Zheng 30143fd0a558SYan, Zheng __mark_block_processed(rc, next); 30155d4f98a2SYan Zheng 30165d4f98a2SYan Zheng if (list_empty(&next->upper)) 30175d4f98a2SYan Zheng break; 30185d4f98a2SYan Zheng 30195d4f98a2SYan Zheng edge = list_entry(next->upper.next, 30205d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 30215d4f98a2SYan Zheng edges[index++] = edge; 30225d4f98a2SYan Zheng next = edge->node[UPPER]; 30235d4f98a2SYan Zheng } 30245d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 30255d4f98a2SYan Zheng } 30265d4f98a2SYan Zheng } 30275d4f98a2SYan Zheng 30287476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 30295d4f98a2SYan Zheng { 3030da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 30317476dfdaSDavid Sterba 30325d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 30339655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 30345d4f98a2SYan Zheng return 1; 30355d4f98a2SYan Zheng return 0; 30365d4f98a2SYan Zheng } 30375d4f98a2SYan Zheng 30382ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 30395d4f98a2SYan Zheng struct tree_block *block) 30405d4f98a2SYan Zheng { 30415d4f98a2SYan Zheng struct extent_buffer *eb; 30425d4f98a2SYan Zheng 30435d4f98a2SYan Zheng BUG_ON(block->key_ready); 3044581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 3045581c1760SQu Wenruo block->level, NULL); 304664c043deSLiu Bo if (IS_ERR(eb)) { 304764c043deSLiu Bo return PTR_ERR(eb); 304864c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3049416bc658SJosef Bacik free_extent_buffer(eb); 3050416bc658SJosef Bacik return -EIO; 3051416bc658SJosef Bacik } 30525d4f98a2SYan Zheng if (block->level == 0) 30535d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 30545d4f98a2SYan Zheng else 30555d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 30565d4f98a2SYan Zheng free_extent_buffer(eb); 30575d4f98a2SYan Zheng block->key_ready = 1; 30585d4f98a2SYan Zheng return 0; 30595d4f98a2SYan Zheng } 30605d4f98a2SYan Zheng 30615d4f98a2SYan Zheng /* 30625d4f98a2SYan Zheng * helper function to relocate a tree block 30635d4f98a2SYan Zheng */ 30645d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 30655d4f98a2SYan Zheng struct reloc_control *rc, 30665d4f98a2SYan Zheng struct backref_node *node, 30675d4f98a2SYan Zheng struct btrfs_key *key, 30685d4f98a2SYan Zheng struct btrfs_path *path) 30695d4f98a2SYan Zheng { 30705d4f98a2SYan Zheng struct btrfs_root *root; 30713fd0a558SYan, Zheng int ret = 0; 30725d4f98a2SYan Zheng 30733fd0a558SYan, Zheng if (!node) 30745d4f98a2SYan Zheng return 0; 30753fd0a558SYan, Zheng 30763fd0a558SYan, Zheng BUG_ON(node->processed); 3077147d256eSZhaolei root = select_one_root(node); 30783fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 30793fd0a558SYan, Zheng update_processed_blocks(rc, node); 30803fd0a558SYan, Zheng goto out; 30815d4f98a2SYan Zheng } 30825d4f98a2SYan Zheng 308327cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30843fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 30853fd0a558SYan, Zheng if (ret) 30865d4f98a2SYan Zheng goto out; 30875d4f98a2SYan Zheng } 30883fd0a558SYan, Zheng 30893fd0a558SYan, Zheng if (root) { 309027cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30913fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 30923fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 30933fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 30943fd0a558SYan, Zheng root = root->reloc_root; 30953fd0a558SYan, Zheng node->new_bytenr = root->node->start; 30963fd0a558SYan, Zheng node->root = root; 30973fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 30983fd0a558SYan, Zheng } else { 30995d4f98a2SYan Zheng path->lowest_level = node->level; 31005d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3101b3b4aa74SDavid Sterba btrfs_release_path(path); 31023fd0a558SYan, Zheng if (ret > 0) 31035d4f98a2SYan Zheng ret = 0; 31043fd0a558SYan, Zheng } 31053fd0a558SYan, Zheng if (!ret) 31063fd0a558SYan, Zheng update_processed_blocks(rc, node); 31073fd0a558SYan, Zheng } else { 31083fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 31093fd0a558SYan, Zheng } 31105d4f98a2SYan Zheng out: 31110647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 31123fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 31135d4f98a2SYan Zheng return ret; 31145d4f98a2SYan Zheng } 31155d4f98a2SYan Zheng 31165d4f98a2SYan Zheng /* 31175d4f98a2SYan Zheng * relocate a list of blocks 31185d4f98a2SYan Zheng */ 31195d4f98a2SYan Zheng static noinline_for_stack 31205d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 31215d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 31225d4f98a2SYan Zheng { 31232ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 31245d4f98a2SYan Zheng struct backref_node *node; 31255d4f98a2SYan Zheng struct btrfs_path *path; 31265d4f98a2SYan Zheng struct tree_block *block; 312798ff7b94SQu Wenruo struct tree_block *next; 31285d4f98a2SYan Zheng int ret; 31295d4f98a2SYan Zheng int err = 0; 31305d4f98a2SYan Zheng 31315d4f98a2SYan Zheng path = btrfs_alloc_path(); 3132e1a12670SLiu Bo if (!path) { 3133e1a12670SLiu Bo err = -ENOMEM; 313434c2b290SDavid Sterba goto out_free_blocks; 3135e1a12670SLiu Bo } 31365d4f98a2SYan Zheng 313798ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 313898ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 31395d4f98a2SYan Zheng if (!block->key_ready) 31402ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 31415d4f98a2SYan Zheng } 31425d4f98a2SYan Zheng 314398ff7b94SQu Wenruo /* Get first keys */ 314498ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 314534c2b290SDavid Sterba if (!block->key_ready) { 31462ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 314734c2b290SDavid Sterba if (err) 314834c2b290SDavid Sterba goto out_free_path; 314934c2b290SDavid Sterba } 31505d4f98a2SYan Zheng } 31515d4f98a2SYan Zheng 315298ff7b94SQu Wenruo /* Do tree relocation */ 315398ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 31543fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 31555d4f98a2SYan Zheng block->level, block->bytenr); 31565d4f98a2SYan Zheng if (IS_ERR(node)) { 31575d4f98a2SYan Zheng err = PTR_ERR(node); 31585d4f98a2SYan Zheng goto out; 31595d4f98a2SYan Zheng } 31605d4f98a2SYan Zheng 31615d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 31625d4f98a2SYan Zheng path); 31635d4f98a2SYan Zheng if (ret < 0) { 316498ff7b94SQu Wenruo if (ret != -EAGAIN || &block->rb_node == rb_first(blocks)) 31655d4f98a2SYan Zheng err = ret; 31665d4f98a2SYan Zheng goto out; 31675d4f98a2SYan Zheng } 31685d4f98a2SYan Zheng } 31695d4f98a2SYan Zheng out: 31703fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 31715d4f98a2SYan Zheng 317234c2b290SDavid Sterba out_free_path: 31735d4f98a2SYan Zheng btrfs_free_path(path); 317434c2b290SDavid Sterba out_free_blocks: 3175e1a12670SLiu Bo free_block_list(blocks); 31765d4f98a2SYan Zheng return err; 31775d4f98a2SYan Zheng } 31785d4f98a2SYan Zheng 31795d4f98a2SYan Zheng static noinline_for_stack 3180efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3181efa56464SYan, Zheng struct file_extent_cluster *cluster) 3182efa56464SYan, Zheng { 3183efa56464SYan, Zheng u64 alloc_hint = 0; 3184efa56464SYan, Zheng u64 start; 3185efa56464SYan, Zheng u64 end; 3186efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3187efa56464SYan, Zheng u64 num_bytes; 3188efa56464SYan, Zheng int nr = 0; 3189efa56464SYan, Zheng int ret = 0; 3190dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3191dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 319218513091SWang Xiaoguang u64 cur_offset; 3193364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3194efa56464SYan, Zheng 3195efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 31965955102cSAl Viro inode_lock(inode); 3197efa56464SYan, Zheng 3198364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3199dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3200efa56464SYan, Zheng if (ret) 3201efa56464SYan, Zheng goto out; 3202efa56464SYan, Zheng 320318513091SWang Xiaoguang cur_offset = prealloc_start; 3204efa56464SYan, Zheng while (nr < cluster->nr) { 3205efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3206efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3207efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3208efa56464SYan, Zheng else 3209efa56464SYan, Zheng end = cluster->end - offset; 3210efa56464SYan, Zheng 3211d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3212efa56464SYan, Zheng num_bytes = end + 1 - start; 321318513091SWang Xiaoguang if (cur_offset < start) 3214bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3215bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3216efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3217efa56464SYan, Zheng num_bytes, num_bytes, 3218efa56464SYan, Zheng end + 1, &alloc_hint); 321918513091SWang Xiaoguang cur_offset = end + 1; 3220d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3221efa56464SYan, Zheng if (ret) 3222efa56464SYan, Zheng break; 3223efa56464SYan, Zheng nr++; 3224efa56464SYan, Zheng } 322518513091SWang Xiaoguang if (cur_offset < prealloc_end) 3226bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3227bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3228efa56464SYan, Zheng out: 32295955102cSAl Viro inode_unlock(inode); 3230364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3231efa56464SYan, Zheng return ret; 3232efa56464SYan, Zheng } 3233efa56464SYan, Zheng 3234efa56464SYan, Zheng static noinline_for_stack 32350257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 32360257bb82SYan, Zheng u64 block_start) 32375d4f98a2SYan Zheng { 32385d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 32395d4f98a2SYan Zheng struct extent_map *em; 32400257bb82SYan, Zheng int ret = 0; 32415d4f98a2SYan Zheng 3242172ddd60SDavid Sterba em = alloc_extent_map(); 32430257bb82SYan, Zheng if (!em) 32440257bb82SYan, Zheng return -ENOMEM; 32450257bb82SYan, Zheng 32465d4f98a2SYan Zheng em->start = start; 32470257bb82SYan, Zheng em->len = end + 1 - start; 32480257bb82SYan, Zheng em->block_len = em->len; 32490257bb82SYan, Zheng em->block_start = block_start; 32505d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 32515d4f98a2SYan Zheng 3252d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 32535d4f98a2SYan Zheng while (1) { 3254890871beSChris Mason write_lock(&em_tree->lock); 325509a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3256890871beSChris Mason write_unlock(&em_tree->lock); 32575d4f98a2SYan Zheng if (ret != -EEXIST) { 32585d4f98a2SYan Zheng free_extent_map(em); 32595d4f98a2SYan Zheng break; 32605d4f98a2SYan Zheng } 3261dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 32625d4f98a2SYan Zheng } 3263d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 32640257bb82SYan, Zheng return ret; 32650257bb82SYan, Zheng } 32665d4f98a2SYan Zheng 32670257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 32680257bb82SYan, Zheng struct file_extent_cluster *cluster) 32690257bb82SYan, Zheng { 32702ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 32710257bb82SYan, Zheng u64 page_start; 32720257bb82SYan, Zheng u64 page_end; 32730257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 32740257bb82SYan, Zheng unsigned long index; 32750257bb82SYan, Zheng unsigned long last_index; 32760257bb82SYan, Zheng struct page *page; 32770257bb82SYan, Zheng struct file_ra_state *ra; 32783b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 32790257bb82SYan, Zheng int nr = 0; 32800257bb82SYan, Zheng int ret = 0; 32810257bb82SYan, Zheng 32820257bb82SYan, Zheng if (!cluster->nr) 32830257bb82SYan, Zheng return 0; 32840257bb82SYan, Zheng 32850257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 32860257bb82SYan, Zheng if (!ra) 32870257bb82SYan, Zheng return -ENOMEM; 32880257bb82SYan, Zheng 3289efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 32900257bb82SYan, Zheng if (ret) 3291efa56464SYan, Zheng goto out; 32920257bb82SYan, Zheng 32930257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 32940257bb82SYan, Zheng 3295efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3296efa56464SYan, Zheng cluster->end - offset, cluster->start); 3297efa56464SYan, Zheng if (ret) 3298efa56464SYan, Zheng goto out; 3299efa56464SYan, Zheng 330009cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 330109cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 33020257bb82SYan, Zheng while (index <= last_index) { 33039f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 33049f3db423SNikolay Borisov PAGE_SIZE); 3305efa56464SYan, Zheng if (ret) 3306efa56464SYan, Zheng goto out; 3307efa56464SYan, Zheng 33080257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 33090257bb82SYan, Zheng if (!page) { 33100257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 33110257bb82SYan, Zheng ra, NULL, index, 33120257bb82SYan, Zheng last_index + 1 - index); 3313a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 33143b16a4e3SJosef Bacik mask); 33150257bb82SYan, Zheng if (!page) { 3316691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 331743b18595SQu Wenruo PAGE_SIZE, true); 331844db1216SFilipe Manana btrfs_delalloc_release_extents(BTRFS_I(inode), 33198702ba93SQu Wenruo PAGE_SIZE); 33200257bb82SYan, Zheng ret = -ENOMEM; 3321efa56464SYan, Zheng goto out; 33220257bb82SYan, Zheng } 33230257bb82SYan, Zheng } 33240257bb82SYan, Zheng 33250257bb82SYan, Zheng if (PageReadahead(page)) { 33260257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 33270257bb82SYan, Zheng ra, NULL, page, index, 33280257bb82SYan, Zheng last_index + 1 - index); 33290257bb82SYan, Zheng } 33300257bb82SYan, Zheng 33310257bb82SYan, Zheng if (!PageUptodate(page)) { 33320257bb82SYan, Zheng btrfs_readpage(NULL, page); 33330257bb82SYan, Zheng lock_page(page); 33340257bb82SYan, Zheng if (!PageUptodate(page)) { 33350257bb82SYan, Zheng unlock_page(page); 333609cbfeafSKirill A. Shutemov put_page(page); 3337691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 333843b18595SQu Wenruo PAGE_SIZE, true); 33398b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 33408702ba93SQu Wenruo PAGE_SIZE); 33410257bb82SYan, Zheng ret = -EIO; 3342efa56464SYan, Zheng goto out; 33430257bb82SYan, Zheng } 33440257bb82SYan, Zheng } 33450257bb82SYan, Zheng 33464eee4fa4SMiao Xie page_start = page_offset(page); 334709cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 33480257bb82SYan, Zheng 3349d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 33500257bb82SYan, Zheng 33510257bb82SYan, Zheng set_page_extent_mapped(page); 33520257bb82SYan, Zheng 33530257bb82SYan, Zheng if (nr < cluster->nr && 33540257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 33550257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 33560257bb82SYan, Zheng page_start, page_end, 3357ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 33580257bb82SYan, Zheng nr++; 33590257bb82SYan, Zheng } 33600257bb82SYan, Zheng 3361765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3362330a5827SNikolay Borisov NULL); 3363765f3cebSNikolay Borisov if (ret) { 3364765f3cebSNikolay Borisov unlock_page(page); 3365765f3cebSNikolay Borisov put_page(page); 3366765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 336743b18595SQu Wenruo PAGE_SIZE, true); 3368765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 33698702ba93SQu Wenruo PAGE_SIZE); 3370765f3cebSNikolay Borisov 3371765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3372765f3cebSNikolay Borisov page_start, page_end, 3373765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3374765f3cebSNikolay Borisov goto out; 3375765f3cebSNikolay Borisov 3376765f3cebSNikolay Borisov } 33770257bb82SYan, Zheng set_page_dirty(page); 33780257bb82SYan, Zheng 33790257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3380d0082371SJeff Mahoney page_start, page_end); 33810257bb82SYan, Zheng unlock_page(page); 338209cbfeafSKirill A. Shutemov put_page(page); 33830257bb82SYan, Zheng 33840257bb82SYan, Zheng index++; 33858702ba93SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); 3386efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 33872ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 33880257bb82SYan, Zheng } 33890257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3390efa56464SYan, Zheng out: 33910257bb82SYan, Zheng kfree(ra); 33920257bb82SYan, Zheng return ret; 33930257bb82SYan, Zheng } 33940257bb82SYan, Zheng 33950257bb82SYan, Zheng static noinline_for_stack 33960257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 33970257bb82SYan, Zheng struct file_extent_cluster *cluster) 33980257bb82SYan, Zheng { 33990257bb82SYan, Zheng int ret; 34000257bb82SYan, Zheng 34010257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 34020257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 34030257bb82SYan, Zheng if (ret) 34040257bb82SYan, Zheng return ret; 34050257bb82SYan, Zheng cluster->nr = 0; 34060257bb82SYan, Zheng } 34070257bb82SYan, Zheng 34080257bb82SYan, Zheng if (!cluster->nr) 34090257bb82SYan, Zheng cluster->start = extent_key->objectid; 34100257bb82SYan, Zheng else 34110257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 34120257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 34130257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 34140257bb82SYan, Zheng cluster->nr++; 34150257bb82SYan, Zheng 34160257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 34170257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 34180257bb82SYan, Zheng if (ret) 34190257bb82SYan, Zheng return ret; 34200257bb82SYan, Zheng cluster->nr = 0; 34210257bb82SYan, Zheng } 34220257bb82SYan, Zheng return 0; 34235d4f98a2SYan Zheng } 34245d4f98a2SYan Zheng 34255d4f98a2SYan Zheng /* 34265d4f98a2SYan Zheng * helper to add a tree block to the list. 34275d4f98a2SYan Zheng * the major work is getting the generation and level of the block 34285d4f98a2SYan Zheng */ 34295d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 34305d4f98a2SYan Zheng struct btrfs_key *extent_key, 34315d4f98a2SYan Zheng struct btrfs_path *path, 34325d4f98a2SYan Zheng struct rb_root *blocks) 34335d4f98a2SYan Zheng { 34345d4f98a2SYan Zheng struct extent_buffer *eb; 34355d4f98a2SYan Zheng struct btrfs_extent_item *ei; 34365d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 34375d4f98a2SYan Zheng struct tree_block *block; 34385d4f98a2SYan Zheng struct rb_node *rb_node; 34395d4f98a2SYan Zheng u32 item_size; 34405d4f98a2SYan Zheng int level = -1; 34417fdf4b60SWang Shilong u64 generation; 34425d4f98a2SYan Zheng 34435d4f98a2SYan Zheng eb = path->nodes[0]; 34445d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 34455d4f98a2SYan Zheng 34463173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 34473173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 34485d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 34495d4f98a2SYan Zheng struct btrfs_extent_item); 34503173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 34515d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 34525d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 34535d4f98a2SYan Zheng } else { 34543173a18fSJosef Bacik level = (int)extent_key->offset; 34553173a18fSJosef Bacik } 34563173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 34576d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3458ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3459ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3460ba3c2b19SNikolay Borisov return -EINVAL; 34613173a18fSJosef Bacik } else { 34625d4f98a2SYan Zheng BUG(); 34635d4f98a2SYan Zheng } 34645d4f98a2SYan Zheng 3465b3b4aa74SDavid Sterba btrfs_release_path(path); 34665d4f98a2SYan Zheng 34675d4f98a2SYan Zheng BUG_ON(level == -1); 34685d4f98a2SYan Zheng 34695d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 34705d4f98a2SYan Zheng if (!block) 34715d4f98a2SYan Zheng return -ENOMEM; 34725d4f98a2SYan Zheng 34735d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3474da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 34755d4f98a2SYan Zheng block->key.offset = generation; 34765d4f98a2SYan Zheng block->level = level; 34775d4f98a2SYan Zheng block->key_ready = 0; 34785d4f98a2SYan Zheng 34795d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 348043c04fb1SJeff Mahoney if (rb_node) 348143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 34825d4f98a2SYan Zheng 34835d4f98a2SYan Zheng return 0; 34845d4f98a2SYan Zheng } 34855d4f98a2SYan Zheng 34865d4f98a2SYan Zheng /* 34875d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 34885d4f98a2SYan Zheng */ 34895d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 34905d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 34915d4f98a2SYan Zheng struct rb_root *blocks) 34925d4f98a2SYan Zheng { 34930b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34945d4f98a2SYan Zheng struct btrfs_path *path; 34955d4f98a2SYan Zheng struct btrfs_key key; 34965d4f98a2SYan Zheng int ret; 34970b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 34985d4f98a2SYan Zheng 34997476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 35005d4f98a2SYan Zheng return 0; 35015d4f98a2SYan Zheng 35025d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 35035d4f98a2SYan Zheng return 0; 35045d4f98a2SYan Zheng 35055d4f98a2SYan Zheng path = btrfs_alloc_path(); 35065d4f98a2SYan Zheng if (!path) 35075d4f98a2SYan Zheng return -ENOMEM; 3508aee68ee5SJosef Bacik again: 35095d4f98a2SYan Zheng key.objectid = bytenr; 3510aee68ee5SJosef Bacik if (skinny) { 3511aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3512aee68ee5SJosef Bacik key.offset = (u64)-1; 3513aee68ee5SJosef Bacik } else { 35145d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 35155d4f98a2SYan Zheng key.offset = blocksize; 3516aee68ee5SJosef Bacik } 35175d4f98a2SYan Zheng 35185d4f98a2SYan Zheng path->search_commit_root = 1; 35195d4f98a2SYan Zheng path->skip_locking = 1; 35205d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 35215d4f98a2SYan Zheng if (ret < 0) 35225d4f98a2SYan Zheng goto out; 35235d4f98a2SYan Zheng 3524aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3525aee68ee5SJosef Bacik if (path->slots[0]) { 3526aee68ee5SJosef Bacik path->slots[0]--; 3527aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3528aee68ee5SJosef Bacik path->slots[0]); 35293173a18fSJosef Bacik if (key.objectid == bytenr && 3530aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3531aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3532aee68ee5SJosef Bacik key.offset == blocksize))) 35333173a18fSJosef Bacik ret = 0; 35343173a18fSJosef Bacik } 3535aee68ee5SJosef Bacik 3536aee68ee5SJosef Bacik if (ret) { 3537aee68ee5SJosef Bacik skinny = false; 3538aee68ee5SJosef Bacik btrfs_release_path(path); 3539aee68ee5SJosef Bacik goto again; 3540aee68ee5SJosef Bacik } 3541aee68ee5SJosef Bacik } 3542cdccee99SLiu Bo if (ret) { 3543cdccee99SLiu Bo ASSERT(ret == 1); 3544cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3545cdccee99SLiu Bo btrfs_err(fs_info, 3546cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3547cdccee99SLiu Bo bytenr); 3548cdccee99SLiu Bo WARN_ON(1); 3549cdccee99SLiu Bo ret = -EINVAL; 3550cdccee99SLiu Bo goto out; 3551cdccee99SLiu Bo } 35523173a18fSJosef Bacik 35535d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 35545d4f98a2SYan Zheng out: 35555d4f98a2SYan Zheng btrfs_free_path(path); 35565d4f98a2SYan Zheng return ret; 35575d4f98a2SYan Zheng } 35585d4f98a2SYan Zheng 35595d4f98a2SYan Zheng /* 35605d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 35615d4f98a2SYan Zheng */ 35625d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 35635d4f98a2SYan Zheng struct extent_buffer *eb) 35645d4f98a2SYan Zheng { 35655d4f98a2SYan Zheng u64 flags; 35665d4f98a2SYan Zheng int ret; 35675d4f98a2SYan Zheng 35685d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 35695d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 35705d4f98a2SYan Zheng return 1; 35715d4f98a2SYan Zheng 35722ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info, 35733173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 35743173a18fSJosef Bacik NULL, &flags); 35755d4f98a2SYan Zheng BUG_ON(ret); 35765d4f98a2SYan Zheng 35775d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 35785d4f98a2SYan Zheng ret = 1; 35795d4f98a2SYan Zheng else 35805d4f98a2SYan Zheng ret = 0; 35815d4f98a2SYan Zheng return ret; 35825d4f98a2SYan Zheng } 35835d4f98a2SYan Zheng 35840af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 358532da5386SDavid Sterba struct btrfs_block_group *block_group, 35861bbc621eSChris Mason struct inode *inode, 35871bbc621eSChris Mason u64 ino) 35880af3d00bSJosef Bacik { 35890af3d00bSJosef Bacik struct btrfs_key key; 35900af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 35910af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 35920af3d00bSJosef Bacik int ret = 0; 35930af3d00bSJosef Bacik 35940af3d00bSJosef Bacik if (inode) 35950af3d00bSJosef Bacik goto truncate; 35960af3d00bSJosef Bacik 35970af3d00bSJosef Bacik key.objectid = ino; 35980af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 35990af3d00bSJosef Bacik key.offset = 0; 36000af3d00bSJosef Bacik 36014c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 36022e19f1f9SAl Viro if (IS_ERR(inode)) 36030af3d00bSJosef Bacik return -ENOENT; 36040af3d00bSJosef Bacik 36050af3d00bSJosef Bacik truncate: 36062ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 36077b61cd92SMiao Xie &fs_info->global_block_rsv); 36087b61cd92SMiao Xie if (ret) 36097b61cd92SMiao Xie goto out; 36107b61cd92SMiao Xie 36117a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 36120af3d00bSJosef Bacik if (IS_ERR(trans)) { 36133612b495STsutomu Itoh ret = PTR_ERR(trans); 36140af3d00bSJosef Bacik goto out; 36150af3d00bSJosef Bacik } 36160af3d00bSJosef Bacik 361777ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 36180af3d00bSJosef Bacik 36193a45bb20SJeff Mahoney btrfs_end_transaction(trans); 36202ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36210af3d00bSJosef Bacik out: 36220af3d00bSJosef Bacik iput(inode); 36230af3d00bSJosef Bacik return ret; 36240af3d00bSJosef Bacik } 36250af3d00bSJosef Bacik 36265d4f98a2SYan Zheng /* 36275d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 36285d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 36295d4f98a2SYan Zheng */ 36305d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 36315d4f98a2SYan Zheng struct btrfs_key *extent_key, 36325d4f98a2SYan Zheng struct extent_buffer *leaf, 36335d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 36345d4f98a2SYan Zheng struct rb_root *blocks) 36355d4f98a2SYan Zheng { 36360b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 36375d4f98a2SYan Zheng struct btrfs_path *path; 36385d4f98a2SYan Zheng struct tree_block *block; 36395d4f98a2SYan Zheng struct btrfs_root *root; 36405d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 36415d4f98a2SYan Zheng struct rb_node *rb_node; 36425d4f98a2SYan Zheng struct btrfs_key key; 36435d4f98a2SYan Zheng u64 ref_root; 36445d4f98a2SYan Zheng u64 ref_objectid; 36455d4f98a2SYan Zheng u64 ref_offset; 36465d4f98a2SYan Zheng u32 ref_count; 36475d4f98a2SYan Zheng u32 nritems; 36485d4f98a2SYan Zheng int err = 0; 36495d4f98a2SYan Zheng int added = 0; 36505d4f98a2SYan Zheng int counted; 36515d4f98a2SYan Zheng int ret; 36525d4f98a2SYan Zheng 36535d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 36545d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 36555d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 36565d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 36575d4f98a2SYan Zheng 36580af3d00bSJosef Bacik /* 36590af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 36600af3d00bSJosef Bacik * it and redo the search. 36610af3d00bSJosef Bacik */ 36620af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 36630b246afaSJeff Mahoney ret = delete_block_group_cache(fs_info, rc->block_group, 36640af3d00bSJosef Bacik NULL, ref_objectid); 36650af3d00bSJosef Bacik if (ret != -ENOENT) 36660af3d00bSJosef Bacik return ret; 36670af3d00bSJosef Bacik ret = 0; 36680af3d00bSJosef Bacik } 36690af3d00bSJosef Bacik 36700af3d00bSJosef Bacik path = btrfs_alloc_path(); 36710af3d00bSJosef Bacik if (!path) 36720af3d00bSJosef Bacik return -ENOMEM; 3673e4058b54SDavid Sterba path->reada = READA_FORWARD; 36740af3d00bSJosef Bacik 36750b246afaSJeff Mahoney root = read_fs_root(fs_info, ref_root); 36765d4f98a2SYan Zheng if (IS_ERR(root)) { 36775d4f98a2SYan Zheng err = PTR_ERR(root); 36785d4f98a2SYan Zheng goto out; 36795d4f98a2SYan Zheng } 36805d4f98a2SYan Zheng 36815d4f98a2SYan Zheng key.objectid = ref_objectid; 36825d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 368384850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 368484850e8dSYan, Zheng key.offset = 0; 368584850e8dSYan, Zheng else 368684850e8dSYan, Zheng key.offset = ref_offset; 36875d4f98a2SYan Zheng 36885d4f98a2SYan Zheng path->search_commit_root = 1; 36895d4f98a2SYan Zheng path->skip_locking = 1; 36905d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 36915d4f98a2SYan Zheng if (ret < 0) { 36925d4f98a2SYan Zheng err = ret; 36935d4f98a2SYan Zheng goto out; 36945d4f98a2SYan Zheng } 36955d4f98a2SYan Zheng 36965d4f98a2SYan Zheng leaf = path->nodes[0]; 36975d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36985d4f98a2SYan Zheng /* 36995d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 37005d4f98a2SYan Zheng * are not counted in 37015d4f98a2SYan Zheng */ 37025d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 37035d4f98a2SYan Zheng counted = 0; 37045d4f98a2SYan Zheng else 37055d4f98a2SYan Zheng counted = 1; 37065d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 37075d4f98a2SYan Zheng if (rb_node) { 37085d4f98a2SYan Zheng if (counted) 37095d4f98a2SYan Zheng added = 1; 37105d4f98a2SYan Zheng else 37115d4f98a2SYan Zheng path->slots[0] = nritems; 37125d4f98a2SYan Zheng } 37135d4f98a2SYan Zheng 37145d4f98a2SYan Zheng while (ref_count > 0) { 37155d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 37165d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 37175d4f98a2SYan Zheng if (ret < 0) { 37185d4f98a2SYan Zheng err = ret; 37195d4f98a2SYan Zheng goto out; 37205d4f98a2SYan Zheng } 3721fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 37225d4f98a2SYan Zheng goto out; 37235d4f98a2SYan Zheng 37245d4f98a2SYan Zheng leaf = path->nodes[0]; 37255d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 37265d4f98a2SYan Zheng added = 0; 37275d4f98a2SYan Zheng 37285d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 37295d4f98a2SYan Zheng counted = 0; 37305d4f98a2SYan Zheng else 37315d4f98a2SYan Zheng counted = 1; 37325d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 37335d4f98a2SYan Zheng if (rb_node) { 37345d4f98a2SYan Zheng if (counted) 37355d4f98a2SYan Zheng added = 1; 37365d4f98a2SYan Zheng else 37375d4f98a2SYan Zheng path->slots[0] = nritems; 37385d4f98a2SYan Zheng } 37395d4f98a2SYan Zheng } 37405d4f98a2SYan Zheng 37415d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3742fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3743fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 37445d4f98a2SYan Zheng break; 37455d4f98a2SYan Zheng 37465d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 37475d4f98a2SYan Zheng struct btrfs_file_extent_item); 37485d4f98a2SYan Zheng 37495d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 37505d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 37515d4f98a2SYan Zheng goto next; 37525d4f98a2SYan Zheng 37535d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 37545d4f98a2SYan Zheng extent_key->objectid) 37555d4f98a2SYan Zheng goto next; 37565d4f98a2SYan Zheng 37575d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 37585d4f98a2SYan Zheng if (key.offset != ref_offset) 37595d4f98a2SYan Zheng goto next; 37605d4f98a2SYan Zheng 37615d4f98a2SYan Zheng if (counted) 37625d4f98a2SYan Zheng ref_count--; 37635d4f98a2SYan Zheng if (added) 37645d4f98a2SYan Zheng goto next; 37655d4f98a2SYan Zheng 37667476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 37675d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 37685d4f98a2SYan Zheng if (!block) { 37695d4f98a2SYan Zheng err = -ENOMEM; 37705d4f98a2SYan Zheng break; 37715d4f98a2SYan Zheng } 37725d4f98a2SYan Zheng block->bytenr = leaf->start; 37735d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 37745d4f98a2SYan Zheng block->level = 0; 37755d4f98a2SYan Zheng block->key_ready = 1; 37765d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 37775d4f98a2SYan Zheng &block->rb_node); 377843c04fb1SJeff Mahoney if (rb_node) 377943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 378043c04fb1SJeff Mahoney block->bytenr); 37815d4f98a2SYan Zheng } 37825d4f98a2SYan Zheng if (counted) 37835d4f98a2SYan Zheng added = 1; 37845d4f98a2SYan Zheng else 37855d4f98a2SYan Zheng path->slots[0] = nritems; 37865d4f98a2SYan Zheng next: 37875d4f98a2SYan Zheng path->slots[0]++; 37885d4f98a2SYan Zheng 37895d4f98a2SYan Zheng } 37905d4f98a2SYan Zheng out: 37915d4f98a2SYan Zheng btrfs_free_path(path); 37925d4f98a2SYan Zheng return err; 37935d4f98a2SYan Zheng } 37945d4f98a2SYan Zheng 37955d4f98a2SYan Zheng /* 37962c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37975d4f98a2SYan Zheng */ 37985d4f98a2SYan Zheng static noinline_for_stack 37995d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 38005d4f98a2SYan Zheng struct btrfs_key *extent_key, 38015d4f98a2SYan Zheng struct btrfs_path *path, 38025d4f98a2SYan Zheng struct rb_root *blocks) 38035d4f98a2SYan Zheng { 38045d4f98a2SYan Zheng struct btrfs_key key; 38055d4f98a2SYan Zheng struct extent_buffer *eb; 38065d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 38075d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 38085d4f98a2SYan Zheng unsigned long ptr; 38095d4f98a2SYan Zheng unsigned long end; 3810da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 3811647f63bdSFilipe David Borba Manana int ret = 0; 38125d4f98a2SYan Zheng int err = 0; 38135d4f98a2SYan Zheng 38145d4f98a2SYan Zheng eb = path->nodes[0]; 38155d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 38165d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 38175d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 38185d4f98a2SYan Zheng 38195d4f98a2SYan Zheng while (ptr < end) { 38205d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 38213de28d57SLiu Bo key.type = btrfs_get_extent_inline_ref_type(eb, iref, 38223de28d57SLiu Bo BTRFS_REF_TYPE_DATA); 38235d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 38245d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 38255d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 38265d4f98a2SYan Zheng blocks); 38275d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 38285d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 38295d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 38305d4f98a2SYan Zheng eb, dref, blocks); 38315d4f98a2SYan Zheng } else { 3832af431dcbSSu Yue ret = -EUCLEAN; 3833b14c55a1SLiu Bo btrfs_err(rc->extent_root->fs_info, 3834b14c55a1SLiu Bo "extent %llu slot %d has an invalid inline ref type", 3835b14c55a1SLiu Bo eb->start, path->slots[0]); 38365d4f98a2SYan Zheng } 3837647f63bdSFilipe David Borba Manana if (ret) { 3838647f63bdSFilipe David Borba Manana err = ret; 3839647f63bdSFilipe David Borba Manana goto out; 3840647f63bdSFilipe David Borba Manana } 38415d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 38425d4f98a2SYan Zheng } 38435d4f98a2SYan Zheng WARN_ON(ptr > end); 38445d4f98a2SYan Zheng 38455d4f98a2SYan Zheng while (1) { 38465d4f98a2SYan Zheng cond_resched(); 38475d4f98a2SYan Zheng eb = path->nodes[0]; 38485d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 38495d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38505d4f98a2SYan Zheng if (ret < 0) { 38515d4f98a2SYan Zheng err = ret; 38525d4f98a2SYan Zheng break; 38535d4f98a2SYan Zheng } 38545d4f98a2SYan Zheng if (ret > 0) 38555d4f98a2SYan Zheng break; 38565d4f98a2SYan Zheng eb = path->nodes[0]; 38575d4f98a2SYan Zheng } 38585d4f98a2SYan Zheng 38595d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 38605d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 38615d4f98a2SYan Zheng break; 38625d4f98a2SYan Zheng 38635d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 38645d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 38655d4f98a2SYan Zheng blocks); 38665d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 38675d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 38685d4f98a2SYan Zheng struct btrfs_extent_data_ref); 38695d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 38705d4f98a2SYan Zheng eb, dref, blocks); 38716d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 3872ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3873ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3874ba3c2b19SNikolay Borisov ret = -EINVAL; 38755d4f98a2SYan Zheng } else { 38765d4f98a2SYan Zheng ret = 0; 38775d4f98a2SYan Zheng } 38785d4f98a2SYan Zheng if (ret) { 38795d4f98a2SYan Zheng err = ret; 38805d4f98a2SYan Zheng break; 38815d4f98a2SYan Zheng } 38825d4f98a2SYan Zheng path->slots[0]++; 38835d4f98a2SYan Zheng } 3884647f63bdSFilipe David Borba Manana out: 3885b3b4aa74SDavid Sterba btrfs_release_path(path); 38865d4f98a2SYan Zheng if (err) 38875d4f98a2SYan Zheng free_block_list(blocks); 38885d4f98a2SYan Zheng return err; 38895d4f98a2SYan Zheng } 38905d4f98a2SYan Zheng 38915d4f98a2SYan Zheng /* 38922c016dc2SLiu Bo * helper to find next unprocessed extent 38935d4f98a2SYan Zheng */ 38945d4f98a2SYan Zheng static noinline_for_stack 3895147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 38963fd0a558SYan, Zheng struct btrfs_key *extent_key) 38975d4f98a2SYan Zheng { 38980b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38995d4f98a2SYan Zheng struct btrfs_key key; 39005d4f98a2SYan Zheng struct extent_buffer *leaf; 39015d4f98a2SYan Zheng u64 start, end, last; 39025d4f98a2SYan Zheng int ret; 39035d4f98a2SYan Zheng 3904b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 39055d4f98a2SYan Zheng while (1) { 39065d4f98a2SYan Zheng cond_resched(); 39075d4f98a2SYan Zheng if (rc->search_start >= last) { 39085d4f98a2SYan Zheng ret = 1; 39095d4f98a2SYan Zheng break; 39105d4f98a2SYan Zheng } 39115d4f98a2SYan Zheng 39125d4f98a2SYan Zheng key.objectid = rc->search_start; 39135d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 39145d4f98a2SYan Zheng key.offset = 0; 39155d4f98a2SYan Zheng 39165d4f98a2SYan Zheng path->search_commit_root = 1; 39175d4f98a2SYan Zheng path->skip_locking = 1; 39185d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 39195d4f98a2SYan Zheng 0, 0); 39205d4f98a2SYan Zheng if (ret < 0) 39215d4f98a2SYan Zheng break; 39225d4f98a2SYan Zheng next: 39235d4f98a2SYan Zheng leaf = path->nodes[0]; 39245d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 39255d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 39265d4f98a2SYan Zheng if (ret != 0) 39275d4f98a2SYan Zheng break; 39285d4f98a2SYan Zheng leaf = path->nodes[0]; 39295d4f98a2SYan Zheng } 39305d4f98a2SYan Zheng 39315d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 39325d4f98a2SYan Zheng if (key.objectid >= last) { 39335d4f98a2SYan Zheng ret = 1; 39345d4f98a2SYan Zheng break; 39355d4f98a2SYan Zheng } 39365d4f98a2SYan Zheng 39373173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 39383173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 39393173a18fSJosef Bacik path->slots[0]++; 39403173a18fSJosef Bacik goto next; 39413173a18fSJosef Bacik } 39423173a18fSJosef Bacik 39433173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 39445d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 39455d4f98a2SYan Zheng path->slots[0]++; 39465d4f98a2SYan Zheng goto next; 39475d4f98a2SYan Zheng } 39485d4f98a2SYan Zheng 39493173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 39500b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 39513173a18fSJosef Bacik rc->search_start) { 39523173a18fSJosef Bacik path->slots[0]++; 39533173a18fSJosef Bacik goto next; 39543173a18fSJosef Bacik } 39553173a18fSJosef Bacik 39565d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 39575d4f98a2SYan Zheng key.objectid, &start, &end, 3958e6138876SJosef Bacik EXTENT_DIRTY, NULL); 39595d4f98a2SYan Zheng 39605d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3961b3b4aa74SDavid Sterba btrfs_release_path(path); 39625d4f98a2SYan Zheng rc->search_start = end + 1; 39635d4f98a2SYan Zheng } else { 39643173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 39655d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 39663173a18fSJosef Bacik else 39673173a18fSJosef Bacik rc->search_start = key.objectid + 39680b246afaSJeff Mahoney fs_info->nodesize; 39693fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 39705d4f98a2SYan Zheng return 0; 39715d4f98a2SYan Zheng } 39725d4f98a2SYan Zheng } 3973b3b4aa74SDavid Sterba btrfs_release_path(path); 39745d4f98a2SYan Zheng return ret; 39755d4f98a2SYan Zheng } 39765d4f98a2SYan Zheng 39775d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 39785d4f98a2SYan Zheng { 39795d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39807585717fSChris Mason 39817585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39825d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 39837585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39845d4f98a2SYan Zheng } 39855d4f98a2SYan Zheng 39865d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 39875d4f98a2SYan Zheng { 39885d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39897585717fSChris Mason 39907585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39915d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 39927585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39935d4f98a2SYan Zheng } 39945d4f98a2SYan Zheng 39955d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 39965d4f98a2SYan Zheng { 39975d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39985d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39995d4f98a2SYan Zheng return 1; 40005d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 40015d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 40025d4f98a2SYan Zheng return 1; 40035d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 40045d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 40055d4f98a2SYan Zheng return 1; 40065d4f98a2SYan Zheng return 0; 40075d4f98a2SYan Zheng } 40085d4f98a2SYan Zheng 40093fd0a558SYan, Zheng static noinline_for_stack 40103fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 40113fd0a558SYan, Zheng { 40123fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 4013ac2fabacSJosef Bacik int ret; 40143fd0a558SYan, Zheng 40152ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 401666d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 40173fd0a558SYan, Zheng if (!rc->block_rsv) 40183fd0a558SYan, Zheng return -ENOMEM; 40193fd0a558SYan, Zheng 40203fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 4021b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 40223fd0a558SYan, Zheng rc->extents_found = 0; 40233fd0a558SYan, Zheng rc->nodes_relocated = 0; 40243fd0a558SYan, Zheng rc->merging_rsv_size = 0; 40250647bf56SWang Shilong rc->reserved_bytes = 0; 4026da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 40270647bf56SWang Shilong RELOCATION_RESERVED_NODES; 4028ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 4029ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 4030ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 4031ac2fabacSJosef Bacik if (ret) 4032ac2fabacSJosef Bacik return ret; 40333fd0a558SYan, Zheng 40343fd0a558SYan, Zheng rc->create_reloc_tree = 1; 40353fd0a558SYan, Zheng set_reloc_control(rc); 40363fd0a558SYan, Zheng 40377a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 403828818947SLiu Bo if (IS_ERR(trans)) { 403928818947SLiu Bo unset_reloc_control(rc); 404028818947SLiu Bo /* 404128818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 404228818947SLiu Bo * cleanup. And callers are responsible to free the above 404328818947SLiu Bo * block rsv. 404428818947SLiu Bo */ 404528818947SLiu Bo return PTR_ERR(trans); 404628818947SLiu Bo } 40473a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 40483fd0a558SYan, Zheng return 0; 40493fd0a558SYan, Zheng } 405076dda93cSYan, Zheng 40515d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 40525d4f98a2SYan Zheng { 40532ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40545d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 40555d4f98a2SYan Zheng struct btrfs_key key; 40565d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 40575d4f98a2SYan Zheng struct btrfs_path *path; 40585d4f98a2SYan Zheng struct btrfs_extent_item *ei; 40595d4f98a2SYan Zheng u64 flags; 40605d4f98a2SYan Zheng u32 item_size; 40615d4f98a2SYan Zheng int ret; 40625d4f98a2SYan Zheng int err = 0; 4063c87f08caSChris Mason int progress = 0; 40645d4f98a2SYan Zheng 40655d4f98a2SYan Zheng path = btrfs_alloc_path(); 40663fd0a558SYan, Zheng if (!path) 40675d4f98a2SYan Zheng return -ENOMEM; 4068e4058b54SDavid Sterba path->reada = READA_FORWARD; 40693fd0a558SYan, Zheng 40703fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 40713fd0a558SYan, Zheng if (ret) { 40723fd0a558SYan, Zheng err = ret; 40733fd0a558SYan, Zheng goto out_free; 40742423fdfbSJiri Slaby } 40755d4f98a2SYan Zheng 40765d4f98a2SYan Zheng while (1) { 40770647bf56SWang Shilong rc->reserved_bytes = 0; 40780647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40790647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40800647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40810647bf56SWang Shilong if (ret) { 40820647bf56SWang Shilong err = ret; 40830647bf56SWang Shilong break; 40840647bf56SWang Shilong } 4085c87f08caSChris Mason progress++; 4086a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40870f788c58SLiu Bo if (IS_ERR(trans)) { 40880f788c58SLiu Bo err = PTR_ERR(trans); 40890f788c58SLiu Bo trans = NULL; 40900f788c58SLiu Bo break; 40910f788c58SLiu Bo } 4092c87f08caSChris Mason restart: 40933fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40943a45bb20SJeff Mahoney btrfs_end_transaction(trans); 409542a657f5SPan Bian trans = NULL; 40963fd0a558SYan, Zheng continue; 40973fd0a558SYan, Zheng } 40983fd0a558SYan, Zheng 4099147d256eSZhaolei ret = find_next_extent(rc, path, &key); 41005d4f98a2SYan Zheng if (ret < 0) 41015d4f98a2SYan Zheng err = ret; 41025d4f98a2SYan Zheng if (ret != 0) 41035d4f98a2SYan Zheng break; 41045d4f98a2SYan Zheng 41055d4f98a2SYan Zheng rc->extents_found++; 41065d4f98a2SYan Zheng 41075d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 41085d4f98a2SYan Zheng struct btrfs_extent_item); 41093fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 41105d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 41115d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 41125d4f98a2SYan Zheng ret = check_extent_flags(flags); 41135d4f98a2SYan Zheng BUG_ON(ret); 41146d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 4115ba3c2b19SNikolay Borisov err = -EINVAL; 4116ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 4117ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 4118ba3c2b19SNikolay Borisov break; 41195d4f98a2SYan Zheng } else { 41205d4f98a2SYan Zheng BUG(); 41215d4f98a2SYan Zheng } 41225d4f98a2SYan Zheng 41235d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 41245d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 41255d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 41265d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41275d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 41285d4f98a2SYan Zheng } else { 4129b3b4aa74SDavid Sterba btrfs_release_path(path); 41305d4f98a2SYan Zheng ret = 0; 41315d4f98a2SYan Zheng } 41325d4f98a2SYan Zheng if (ret < 0) { 41333fd0a558SYan, Zheng err = ret; 41345d4f98a2SYan Zheng break; 41355d4f98a2SYan Zheng } 41365d4f98a2SYan Zheng 41375d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 41385d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 41395d4f98a2SYan Zheng if (ret < 0) { 41401708cc57SWang Shilong /* 41411708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 41421708cc57SWang Shilong * backref cache when committing transaction. 41431708cc57SWang Shilong */ 41441708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 41451708cc57SWang Shilong 41463fd0a558SYan, Zheng if (ret != -EAGAIN) { 41475d4f98a2SYan Zheng err = ret; 41485d4f98a2SYan Zheng break; 41495d4f98a2SYan Zheng } 41503fd0a558SYan, Zheng rc->extents_found--; 41513fd0a558SYan, Zheng rc->search_start = key.objectid; 41523fd0a558SYan, Zheng } 41535d4f98a2SYan Zheng } 41545d4f98a2SYan Zheng 41553a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41562ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41573fd0a558SYan, Zheng trans = NULL; 41585d4f98a2SYan Zheng 41595d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 41605d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41615d4f98a2SYan Zheng rc->found_file_extent = 1; 41620257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 41633fd0a558SYan, Zheng &key, &rc->cluster); 41645d4f98a2SYan Zheng if (ret < 0) { 41655d4f98a2SYan Zheng err = ret; 41665d4f98a2SYan Zheng break; 41675d4f98a2SYan Zheng } 41685d4f98a2SYan Zheng } 41695d4f98a2SYan Zheng } 4170c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 417143a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 41729689457bSShilong Wang if (ret == 1) { 4173c87f08caSChris Mason err = 0; 4174c87f08caSChris Mason progress = 0; 4175c87f08caSChris Mason goto restart; 4176c87f08caSChris Mason } 4177c87f08caSChris Mason } 41783fd0a558SYan, Zheng 4179b3b4aa74SDavid Sterba btrfs_release_path(path); 418091166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41815d4f98a2SYan Zheng 41825d4f98a2SYan Zheng if (trans) { 41833a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41842ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41855d4f98a2SYan Zheng } 41865d4f98a2SYan Zheng 41870257bb82SYan, Zheng if (!err) { 41883fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 41893fd0a558SYan, Zheng &rc->cluster); 41900257bb82SYan, Zheng if (ret < 0) 41910257bb82SYan, Zheng err = ret; 41920257bb82SYan, Zheng } 41930257bb82SYan, Zheng 41943fd0a558SYan, Zheng rc->create_reloc_tree = 0; 41953fd0a558SYan, Zheng set_reloc_control(rc); 41960257bb82SYan, Zheng 41973fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 41982ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41995d4f98a2SYan Zheng 42003fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 42015d4f98a2SYan Zheng 42025d4f98a2SYan Zheng merge_reloc_roots(rc); 42035d4f98a2SYan Zheng 42043fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 42055d4f98a2SYan Zheng unset_reloc_control(rc); 42062ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 42075d4f98a2SYan Zheng 42085d4f98a2SYan Zheng /* get rid of pinned extents */ 42097a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 421062b99540SQu Wenruo if (IS_ERR(trans)) { 42113612b495STsutomu Itoh err = PTR_ERR(trans); 421262b99540SQu Wenruo goto out_free; 421362b99540SQu Wenruo } 42143a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 4215d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4216d2311e69SQu Wenruo if (ret < 0 && !err) 4217d2311e69SQu Wenruo err = ret; 42183fd0a558SYan, Zheng out_free: 42192ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 42203fd0a558SYan, Zheng btrfs_free_path(path); 42215d4f98a2SYan Zheng return err; 42225d4f98a2SYan Zheng } 42235d4f98a2SYan Zheng 42245d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 42250257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 42265d4f98a2SYan Zheng { 42275d4f98a2SYan Zheng struct btrfs_path *path; 42285d4f98a2SYan Zheng struct btrfs_inode_item *item; 42295d4f98a2SYan Zheng struct extent_buffer *leaf; 42305d4f98a2SYan Zheng int ret; 42315d4f98a2SYan Zheng 42325d4f98a2SYan Zheng path = btrfs_alloc_path(); 42335d4f98a2SYan Zheng if (!path) 42345d4f98a2SYan Zheng return -ENOMEM; 42355d4f98a2SYan Zheng 42365d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 42375d4f98a2SYan Zheng if (ret) 42385d4f98a2SYan Zheng goto out; 42395d4f98a2SYan Zheng 42405d4f98a2SYan Zheng leaf = path->nodes[0]; 42415d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4242b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 42435d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 42440257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 42455d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 42463fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 42473fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 42485d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 42495d4f98a2SYan Zheng out: 42505d4f98a2SYan Zheng btrfs_free_path(path); 42515d4f98a2SYan Zheng return ret; 42525d4f98a2SYan Zheng } 42535d4f98a2SYan Zheng 42545d4f98a2SYan Zheng /* 42555d4f98a2SYan Zheng * helper to create inode for data relocation. 42565d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 42575d4f98a2SYan Zheng */ 42583fd0a558SYan, Zheng static noinline_for_stack 42593fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 426032da5386SDavid Sterba struct btrfs_block_group *group) 42615d4f98a2SYan Zheng { 42625d4f98a2SYan Zheng struct inode *inode = NULL; 42635d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42645d4f98a2SYan Zheng struct btrfs_root *root; 42655d4f98a2SYan Zheng struct btrfs_key key; 42664624900dSZhaolei u64 objectid; 42675d4f98a2SYan Zheng int err = 0; 42685d4f98a2SYan Zheng 42695d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 42705d4f98a2SYan Zheng if (IS_ERR(root)) 42715d4f98a2SYan Zheng return ERR_CAST(root); 42725d4f98a2SYan Zheng 4273a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 42743fd0a558SYan, Zheng if (IS_ERR(trans)) 42753fd0a558SYan, Zheng return ERR_CAST(trans); 42765d4f98a2SYan Zheng 4277581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42785d4f98a2SYan Zheng if (err) 42795d4f98a2SYan Zheng goto out; 42805d4f98a2SYan Zheng 42810257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42825d4f98a2SYan Zheng BUG_ON(err); 42835d4f98a2SYan Zheng 42845d4f98a2SYan Zheng key.objectid = objectid; 42855d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 42865d4f98a2SYan Zheng key.offset = 0; 42874c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 42882e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 4289b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 42905d4f98a2SYan Zheng 429173f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 42925d4f98a2SYan Zheng out: 42933a45bb20SJeff Mahoney btrfs_end_transaction(trans); 42942ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42955d4f98a2SYan Zheng if (err) { 42965d4f98a2SYan Zheng if (inode) 42975d4f98a2SYan Zheng iput(inode); 42985d4f98a2SYan Zheng inode = ERR_PTR(err); 42995d4f98a2SYan Zheng } 43005d4f98a2SYan Zheng return inode; 43015d4f98a2SYan Zheng } 43025d4f98a2SYan Zheng 4303c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 43043fd0a558SYan, Zheng { 43053fd0a558SYan, Zheng struct reloc_control *rc; 43063fd0a558SYan, Zheng 43073fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 43083fd0a558SYan, Zheng if (!rc) 43093fd0a558SYan, Zheng return NULL; 43103fd0a558SYan, Zheng 43113fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4312d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 43133fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 43143fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 431543eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 431643eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 43173fd0a558SYan, Zheng return rc; 43183fd0a558SYan, Zheng } 43193fd0a558SYan, Zheng 43205d4f98a2SYan Zheng /* 4321ebce0e01SAdam Borowski * Print the block group being relocated 4322ebce0e01SAdam Borowski */ 4323ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 432432da5386SDavid Sterba struct btrfs_block_group *block_group) 4325ebce0e01SAdam Borowski { 4326f89e09cfSAnand Jain char buf[128] = {'\0'}; 4327ebce0e01SAdam Borowski 4328f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4329ebce0e01SAdam Borowski 4330ebce0e01SAdam Borowski btrfs_info(fs_info, 4331ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4332b3470b5dSDavid Sterba block_group->start, buf); 4333ebce0e01SAdam Borowski } 4334ebce0e01SAdam Borowski 4335430640e3SQu Wenruo static const char *stage_to_string(int stage) 4336430640e3SQu Wenruo { 4337430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 4338430640e3SQu Wenruo return "move data extents"; 4339430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 4340430640e3SQu Wenruo return "update data pointers"; 4341430640e3SQu Wenruo return "unknown"; 4342430640e3SQu Wenruo } 4343430640e3SQu Wenruo 4344ebce0e01SAdam Borowski /* 43455d4f98a2SYan Zheng * function to relocate all extents in a block group. 43465d4f98a2SYan Zheng */ 43476bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 43485d4f98a2SYan Zheng { 434932da5386SDavid Sterba struct btrfs_block_group *bg; 43506bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 43515d4f98a2SYan Zheng struct reloc_control *rc; 43520af3d00bSJosef Bacik struct inode *inode; 43530af3d00bSJosef Bacik struct btrfs_path *path; 43545d4f98a2SYan Zheng int ret; 4355f0486c68SYan, Zheng int rw = 0; 43565d4f98a2SYan Zheng int err = 0; 43575d4f98a2SYan Zheng 4358eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4359eede2bf3SOmar Sandoval if (!bg) 4360eede2bf3SOmar Sandoval return -ENOENT; 4361eede2bf3SOmar Sandoval 4362eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4363eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4364eede2bf3SOmar Sandoval return -ETXTBSY; 4365eede2bf3SOmar Sandoval } 4366eede2bf3SOmar Sandoval 4367c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4368eede2bf3SOmar Sandoval if (!rc) { 4369eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 43705d4f98a2SYan Zheng return -ENOMEM; 4371eede2bf3SOmar Sandoval } 43725d4f98a2SYan Zheng 4373f0486c68SYan, Zheng rc->extent_root = extent_root; 4374eede2bf3SOmar Sandoval rc->block_group = bg; 43755d4f98a2SYan Zheng 4376b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4377f0486c68SYan, Zheng if (ret) { 4378f0486c68SYan, Zheng err = ret; 4379f0486c68SYan, Zheng goto out; 4380f0486c68SYan, Zheng } 4381f0486c68SYan, Zheng rw = 1; 4382f0486c68SYan, Zheng 43830af3d00bSJosef Bacik path = btrfs_alloc_path(); 43840af3d00bSJosef Bacik if (!path) { 43850af3d00bSJosef Bacik err = -ENOMEM; 43860af3d00bSJosef Bacik goto out; 43870af3d00bSJosef Bacik } 43880af3d00bSJosef Bacik 43897949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 43900af3d00bSJosef Bacik btrfs_free_path(path); 43910af3d00bSJosef Bacik 43920af3d00bSJosef Bacik if (!IS_ERR(inode)) 43931bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43940af3d00bSJosef Bacik else 43950af3d00bSJosef Bacik ret = PTR_ERR(inode); 43960af3d00bSJosef Bacik 43970af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 43980af3d00bSJosef Bacik err = ret; 43990af3d00bSJosef Bacik goto out; 44000af3d00bSJosef Bacik } 44010af3d00bSJosef Bacik 44025d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 44035d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 44045d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 44055d4f98a2SYan Zheng rc->data_inode = NULL; 44065d4f98a2SYan Zheng goto out; 44075d4f98a2SYan Zheng } 44085d4f98a2SYan Zheng 44090b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 44105d4f98a2SYan Zheng 44119cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4412f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 44136374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4414b3470b5dSDavid Sterba rc->block_group->start, 4415b3470b5dSDavid Sterba rc->block_group->length); 44165d4f98a2SYan Zheng 44175d4f98a2SYan Zheng while (1) { 4418430640e3SQu Wenruo int finishes_stage; 4419430640e3SQu Wenruo 442076dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 44215d4f98a2SYan Zheng ret = relocate_block_group(rc); 442276dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4423ff612ba7SJosef Bacik if (ret < 0) 44245d4f98a2SYan Zheng err = ret; 4425ff612ba7SJosef Bacik 4426430640e3SQu Wenruo finishes_stage = rc->stage; 4427ff612ba7SJosef Bacik /* 4428ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4429ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4430ff612ba7SJosef Bacik * different block group we could end up hitting the 4431ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4432ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4433ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4434ff612ba7SJosef Bacik * out of the loop if we hit an error. 4435ff612ba7SJosef Bacik */ 4436ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4437ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4438ff612ba7SJosef Bacik (u64)-1); 4439ff612ba7SJosef Bacik if (ret) 4440ff612ba7SJosef Bacik err = ret; 4441ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4442ff612ba7SJosef Bacik 0, -1); 4443ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 44445d4f98a2SYan Zheng } 44455d4f98a2SYan Zheng 4446ff612ba7SJosef Bacik if (err < 0) 4447ff612ba7SJosef Bacik goto out; 4448ff612ba7SJosef Bacik 44495d4f98a2SYan Zheng if (rc->extents_found == 0) 44505d4f98a2SYan Zheng break; 44515d4f98a2SYan Zheng 4452430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4453430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 44545d4f98a2SYan Zheng } 44555d4f98a2SYan Zheng 44565d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 44575d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4458bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 44595d4f98a2SYan Zheng out: 4460f0486c68SYan, Zheng if (err && rw) 44612ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 44625d4f98a2SYan Zheng iput(rc->data_inode); 44635d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 44645d4f98a2SYan Zheng kfree(rc); 44655d4f98a2SYan Zheng return err; 44665d4f98a2SYan Zheng } 44675d4f98a2SYan Zheng 446876dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 446976dda93cSYan, Zheng { 44700b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 447176dda93cSYan, Zheng struct btrfs_trans_handle *trans; 447279787eaaSJeff Mahoney int ret, err; 447376dda93cSYan, Zheng 44740b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 447579787eaaSJeff Mahoney if (IS_ERR(trans)) 447679787eaaSJeff Mahoney return PTR_ERR(trans); 447776dda93cSYan, Zheng 447876dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 447976dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 448076dda93cSYan, Zheng root->root_item.drop_level = 0; 448176dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 44820b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 448376dda93cSYan, Zheng &root->root_key, &root->root_item); 448476dda93cSYan, Zheng 44853a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 448679787eaaSJeff Mahoney if (err) 448779787eaaSJeff Mahoney return err; 448879787eaaSJeff Mahoney return ret; 448976dda93cSYan, Zheng } 449076dda93cSYan, Zheng 44915d4f98a2SYan Zheng /* 44925d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44935d4f98a2SYan Zheng * 44945d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44955d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44965d4f98a2SYan Zheng */ 44975d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44985d4f98a2SYan Zheng { 44990b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 45005d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 45015d4f98a2SYan Zheng struct btrfs_key key; 45025d4f98a2SYan Zheng struct btrfs_root *fs_root; 45035d4f98a2SYan Zheng struct btrfs_root *reloc_root; 45045d4f98a2SYan Zheng struct btrfs_path *path; 45055d4f98a2SYan Zheng struct extent_buffer *leaf; 45065d4f98a2SYan Zheng struct reloc_control *rc = NULL; 45075d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 45085d4f98a2SYan Zheng int ret; 45095d4f98a2SYan Zheng int err = 0; 45105d4f98a2SYan Zheng 45115d4f98a2SYan Zheng path = btrfs_alloc_path(); 45125d4f98a2SYan Zheng if (!path) 45135d4f98a2SYan Zheng return -ENOMEM; 4514e4058b54SDavid Sterba path->reada = READA_BACK; 45155d4f98a2SYan Zheng 45165d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 45175d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 45185d4f98a2SYan Zheng key.offset = (u64)-1; 45195d4f98a2SYan Zheng 45205d4f98a2SYan Zheng while (1) { 45210b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 45225d4f98a2SYan Zheng path, 0, 0); 45235d4f98a2SYan Zheng if (ret < 0) { 45245d4f98a2SYan Zheng err = ret; 45255d4f98a2SYan Zheng goto out; 45265d4f98a2SYan Zheng } 45275d4f98a2SYan Zheng if (ret > 0) { 45285d4f98a2SYan Zheng if (path->slots[0] == 0) 45295d4f98a2SYan Zheng break; 45305d4f98a2SYan Zheng path->slots[0]--; 45315d4f98a2SYan Zheng } 45325d4f98a2SYan Zheng leaf = path->nodes[0]; 45335d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4534b3b4aa74SDavid Sterba btrfs_release_path(path); 45355d4f98a2SYan Zheng 45365d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 45375d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 45385d4f98a2SYan Zheng break; 45395d4f98a2SYan Zheng 4540cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 45415d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 45425d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 45435d4f98a2SYan Zheng goto out; 45445d4f98a2SYan Zheng } 45455d4f98a2SYan Zheng 45465d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 45475d4f98a2SYan Zheng 45485d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 45490b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 45505d4f98a2SYan Zheng reloc_root->root_key.offset); 45515d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 455276dda93cSYan, Zheng ret = PTR_ERR(fs_root); 455376dda93cSYan, Zheng if (ret != -ENOENT) { 455476dda93cSYan, Zheng err = ret; 45555d4f98a2SYan Zheng goto out; 45565d4f98a2SYan Zheng } 455779787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 455879787eaaSJeff Mahoney if (ret < 0) { 455979787eaaSJeff Mahoney err = ret; 456079787eaaSJeff Mahoney goto out; 456179787eaaSJeff Mahoney } 456276dda93cSYan, Zheng } 45635d4f98a2SYan Zheng } 45645d4f98a2SYan Zheng 45655d4f98a2SYan Zheng if (key.offset == 0) 45665d4f98a2SYan Zheng break; 45675d4f98a2SYan Zheng 45685d4f98a2SYan Zheng key.offset--; 45695d4f98a2SYan Zheng } 4570b3b4aa74SDavid Sterba btrfs_release_path(path); 45715d4f98a2SYan Zheng 45725d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 45735d4f98a2SYan Zheng goto out; 45745d4f98a2SYan Zheng 4575c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 45765d4f98a2SYan Zheng if (!rc) { 45775d4f98a2SYan Zheng err = -ENOMEM; 45785d4f98a2SYan Zheng goto out; 45795d4f98a2SYan Zheng } 45805d4f98a2SYan Zheng 45810b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 45825d4f98a2SYan Zheng 45835d4f98a2SYan Zheng set_reloc_control(rc); 45845d4f98a2SYan Zheng 45857a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45863612b495STsutomu Itoh if (IS_ERR(trans)) { 45873612b495STsutomu Itoh unset_reloc_control(rc); 45883612b495STsutomu Itoh err = PTR_ERR(trans); 45893612b495STsutomu Itoh goto out_free; 45903612b495STsutomu Itoh } 45913fd0a558SYan, Zheng 45923fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45933fd0a558SYan, Zheng 45945d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45955d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45965d4f98a2SYan Zheng struct btrfs_root, root_list); 45975d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45985d4f98a2SYan Zheng 45995d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 46005d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 46015d4f98a2SYan Zheng &rc->reloc_roots); 46025d4f98a2SYan Zheng continue; 46035d4f98a2SYan Zheng } 46045d4f98a2SYan Zheng 46050b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 460679787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 460779787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4608ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 460979787eaaSJeff Mahoney goto out_free; 461079787eaaSJeff Mahoney } 46115d4f98a2SYan Zheng 4612ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 461379787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 46145d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 46155d4f98a2SYan Zheng } 46165d4f98a2SYan Zheng 46173a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 461879787eaaSJeff Mahoney if (err) 461979787eaaSJeff Mahoney goto out_free; 46205d4f98a2SYan Zheng 46215d4f98a2SYan Zheng merge_reloc_roots(rc); 46225d4f98a2SYan Zheng 46235d4f98a2SYan Zheng unset_reloc_control(rc); 46245d4f98a2SYan Zheng 46257a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 462662b99540SQu Wenruo if (IS_ERR(trans)) { 46273612b495STsutomu Itoh err = PTR_ERR(trans); 462862b99540SQu Wenruo goto out_free; 462962b99540SQu Wenruo } 46303a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 4631d2311e69SQu Wenruo 4632d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4633d2311e69SQu Wenruo if (ret < 0 && !err) 4634d2311e69SQu Wenruo err = ret; 46353612b495STsutomu Itoh out_free: 46365d4f98a2SYan Zheng kfree(rc); 46373612b495STsutomu Itoh out: 4638aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4639aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4640aca1bba6SLiu Bo 46415d4f98a2SYan Zheng btrfs_free_path(path); 46425d4f98a2SYan Zheng 46435d4f98a2SYan Zheng if (err == 0) { 46445d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 46450b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 46465d4f98a2SYan Zheng if (IS_ERR(fs_root)) 46475d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4648d7ce5843SMiao Xie else 464966b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 46505d4f98a2SYan Zheng } 46515d4f98a2SYan Zheng return err; 46525d4f98a2SYan Zheng } 46535d4f98a2SYan Zheng 46545d4f98a2SYan Zheng /* 46555d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 46565d4f98a2SYan Zheng * 46575d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 46585d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 46595d4f98a2SYan Zheng */ 46605d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 46615d4f98a2SYan Zheng { 46620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 46635d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 46645d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 46655d4f98a2SYan Zheng int ret; 46665d4f98a2SYan Zheng u64 disk_bytenr; 46674577b014SJosef Bacik u64 new_bytenr; 46685d4f98a2SYan Zheng LIST_HEAD(list); 46695d4f98a2SYan Zheng 46705d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4671*bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 46725d4f98a2SYan Zheng 46735d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46740b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4675a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 467679787eaaSJeff Mahoney if (ret) 467779787eaaSJeff Mahoney goto out; 46785d4f98a2SYan Zheng 46795d4f98a2SYan Zheng while (!list_empty(&list)) { 46805d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46815d4f98a2SYan Zheng list_del_init(&sums->list); 46825d4f98a2SYan Zheng 46834577b014SJosef Bacik /* 46844577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46854577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46864577b014SJosef Bacik * extents but we may have written to say the middle of the 46874577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46884577b014SJosef Bacik * the right disk offset. 46894577b014SJosef Bacik * 46904577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46914577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46924577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46934577b014SJosef Bacik * disk length. 46944577b014SJosef Bacik */ 4695*bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 46964577b014SJosef Bacik sums->bytenr = new_bytenr; 46975d4f98a2SYan Zheng 4698f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 46995d4f98a2SYan Zheng } 470079787eaaSJeff Mahoney out: 47015d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4702411fc6bcSAndi Kleen return ret; 47035d4f98a2SYan Zheng } 47043fd0a558SYan, Zheng 470583d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 47063fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 47073fd0a558SYan, Zheng struct extent_buffer *cow) 47083fd0a558SYan, Zheng { 47090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 47103fd0a558SYan, Zheng struct reloc_control *rc; 47113fd0a558SYan, Zheng struct backref_node *node; 47123fd0a558SYan, Zheng int first_cow = 0; 47133fd0a558SYan, Zheng int level; 471483d4cfd4SJosef Bacik int ret = 0; 47153fd0a558SYan, Zheng 47160b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 47173fd0a558SYan, Zheng if (!rc) 471883d4cfd4SJosef Bacik return 0; 47193fd0a558SYan, Zheng 47203fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 47213fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 47223fd0a558SYan, Zheng 4723c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4724c974c464SWang Shilong if (buf == root->node) 4725c974c464SWang Shilong __update_reloc_root(root, cow->start); 4726c974c464SWang Shilong } 4727c974c464SWang Shilong 47283fd0a558SYan, Zheng level = btrfs_header_level(buf); 47293fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 47303fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 47313fd0a558SYan, Zheng first_cow = 1; 47323fd0a558SYan, Zheng 47333fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 47343fd0a558SYan, Zheng rc->create_reloc_tree) { 47353fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 47363fd0a558SYan, Zheng 47373fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 47383fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 47393fd0a558SYan, Zheng node->new_bytenr != buf->start); 47403fd0a558SYan, Zheng 47413fd0a558SYan, Zheng drop_node_buffer(node); 474267439dadSDavid Sterba atomic_inc(&cow->refs); 47433fd0a558SYan, Zheng node->eb = cow; 47443fd0a558SYan, Zheng node->new_bytenr = cow->start; 47453fd0a558SYan, Zheng 47463fd0a558SYan, Zheng if (!node->pending) { 47473fd0a558SYan, Zheng list_move_tail(&node->list, 47483fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 47493fd0a558SYan, Zheng node->pending = 1; 47503fd0a558SYan, Zheng } 47513fd0a558SYan, Zheng 47523fd0a558SYan, Zheng if (first_cow) 47533fd0a558SYan, Zheng __mark_block_processed(rc, node); 47543fd0a558SYan, Zheng 47553fd0a558SYan, Zheng if (first_cow && level > 0) 47563fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 47573fd0a558SYan, Zheng } 47583fd0a558SYan, Zheng 475983d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 47603fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 476183d4cfd4SJosef Bacik return ret; 47623fd0a558SYan, Zheng } 47633fd0a558SYan, Zheng 47643fd0a558SYan, Zheng /* 47653fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 476601327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 47673fd0a558SYan, Zheng */ 4768147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 47693fd0a558SYan, Zheng u64 *bytes_to_reserve) 47703fd0a558SYan, Zheng { 477110995c04SQu Wenruo struct btrfs_root *root = pending->root; 477210995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 47733fd0a558SYan, Zheng 47746282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 47753fd0a558SYan, Zheng return; 47763fd0a558SYan, Zheng 47773fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47783fd0a558SYan, Zheng return; 47793fd0a558SYan, Zheng 47803fd0a558SYan, Zheng root = root->reloc_root; 47813fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47823fd0a558SYan, Zheng /* 47833fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47843fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47853fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47863fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47873fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47883fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 47893fd0a558SYan, Zheng * tree may use more space than it frees. so we need 47903fd0a558SYan, Zheng * reserve extra space. 47913fd0a558SYan, Zheng */ 47923fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47933fd0a558SYan, Zheng } 47943fd0a558SYan, Zheng 47953fd0a558SYan, Zheng /* 47963fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47973fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 47983fd0a558SYan, Zheng */ 479949b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 48003fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 48013fd0a558SYan, Zheng { 48023fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 48033fd0a558SYan, Zheng struct btrfs_root *reloc_root; 48043fd0a558SYan, Zheng struct btrfs_root *new_root; 480510995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 48063fd0a558SYan, Zheng int ret; 48073fd0a558SYan, Zheng 48086282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 480949b25e05SJeff Mahoney return 0; 48103fd0a558SYan, Zheng 48113fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 48123fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 48133fd0a558SYan, Zheng 48143fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 48153fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 48163fd0a558SYan, Zheng rc->block_rsv, 48173a584174SLu Fengqi rc->nodes_relocated, true); 481849b25e05SJeff Mahoney if (ret) 481949b25e05SJeff Mahoney return ret; 48203fd0a558SYan, Zheng } 48213fd0a558SYan, Zheng 48223fd0a558SYan, Zheng new_root = pending->snap; 48233fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 48243fd0a558SYan, Zheng new_root->root_key.objectid); 482549b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 482649b25e05SJeff Mahoney return PTR_ERR(reloc_root); 48273fd0a558SYan, Zheng 4828ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4829ffd7b339SJeff Mahoney BUG_ON(ret < 0); 48303fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 48313fd0a558SYan, Zheng 483249b25e05SJeff Mahoney if (rc->create_reloc_tree) 48333fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 483449b25e05SJeff Mahoney return ret; 48353fd0a558SYan, Zheng } 4836