15d4f98a2SYan Zheng /* 25d4f98a2SYan Zheng * Copyright (C) 2009 Oracle. All rights reserved. 35d4f98a2SYan Zheng * 45d4f98a2SYan Zheng * This program is free software; you can redistribute it and/or 55d4f98a2SYan Zheng * modify it under the terms of the GNU General Public 65d4f98a2SYan Zheng * License v2 as published by the Free Software Foundation. 75d4f98a2SYan Zheng * 85d4f98a2SYan Zheng * This program is distributed in the hope that it will be useful, 95d4f98a2SYan Zheng * but WITHOUT ANY WARRANTY; without even the implied warranty of 105d4f98a2SYan Zheng * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 115d4f98a2SYan Zheng * General Public License for more details. 125d4f98a2SYan Zheng * 135d4f98a2SYan Zheng * You should have received a copy of the GNU General Public 145d4f98a2SYan Zheng * License along with this program; if not, write to the 155d4f98a2SYan Zheng * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 165d4f98a2SYan Zheng * Boston, MA 021110-1307, USA. 175d4f98a2SYan Zheng */ 185d4f98a2SYan Zheng 195d4f98a2SYan Zheng #include <linux/sched.h> 205d4f98a2SYan Zheng #include <linux/pagemap.h> 215d4f98a2SYan Zheng #include <linux/writeback.h> 225d4f98a2SYan Zheng #include <linux/blkdev.h> 235d4f98a2SYan Zheng #include <linux/rbtree.h> 245a0e3ad6STejun Heo #include <linux/slab.h> 255d4f98a2SYan Zheng #include "ctree.h" 265d4f98a2SYan Zheng #include "disk-io.h" 275d4f98a2SYan Zheng #include "transaction.h" 285d4f98a2SYan Zheng #include "volumes.h" 295d4f98a2SYan Zheng #include "locking.h" 305d4f98a2SYan Zheng #include "btrfs_inode.h" 315d4f98a2SYan Zheng #include "async-thread.h" 320af3d00bSJosef Bacik #include "free-space-cache.h" 33581bb050SLi Zefan #include "inode-map.h" 345d4f98a2SYan Zheng 355d4f98a2SYan Zheng /* 365d4f98a2SYan Zheng * backref_node, mapping_node and tree_block start with this 375d4f98a2SYan Zheng */ 385d4f98a2SYan Zheng struct tree_entry { 395d4f98a2SYan Zheng struct rb_node rb_node; 405d4f98a2SYan Zheng u64 bytenr; 415d4f98a2SYan Zheng }; 425d4f98a2SYan Zheng 435d4f98a2SYan Zheng /* 445d4f98a2SYan Zheng * present a tree block in the backref cache 455d4f98a2SYan Zheng */ 465d4f98a2SYan Zheng struct backref_node { 475d4f98a2SYan Zheng struct rb_node rb_node; 485d4f98a2SYan Zheng u64 bytenr; 493fd0a558SYan, Zheng 503fd0a558SYan, Zheng u64 new_bytenr; 513fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 525d4f98a2SYan Zheng u64 owner; 533fd0a558SYan, Zheng /* link to pending, changed or detached list */ 543fd0a558SYan, Zheng struct list_head list; 555d4f98a2SYan Zheng /* list of upper level blocks reference this block */ 565d4f98a2SYan Zheng struct list_head upper; 575d4f98a2SYan Zheng /* list of child blocks in the cache */ 585d4f98a2SYan Zheng struct list_head lower; 595d4f98a2SYan Zheng /* NULL if this node is not tree root */ 605d4f98a2SYan Zheng struct btrfs_root *root; 615d4f98a2SYan Zheng /* extent buffer got by COW the block */ 625d4f98a2SYan Zheng struct extent_buffer *eb; 635d4f98a2SYan Zheng /* level of tree block */ 645d4f98a2SYan Zheng unsigned int level:8; 653fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 663fd0a558SYan, Zheng unsigned int cowonly:1; 673fd0a558SYan, Zheng /* 1 if no child node in the cache */ 685d4f98a2SYan Zheng unsigned int lowest:1; 695d4f98a2SYan Zheng /* is the extent buffer locked */ 705d4f98a2SYan Zheng unsigned int locked:1; 715d4f98a2SYan Zheng /* has the block been processed */ 725d4f98a2SYan Zheng unsigned int processed:1; 735d4f98a2SYan Zheng /* have backrefs of this block been checked */ 745d4f98a2SYan Zheng unsigned int checked:1; 753fd0a558SYan, Zheng /* 763fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 773fd0a558SYan, Zheng * level block pointers may not point to the new location 783fd0a558SYan, Zheng */ 793fd0a558SYan, Zheng unsigned int pending:1; 803fd0a558SYan, Zheng /* 813fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 823fd0a558SYan, Zheng * backref node. 833fd0a558SYan, Zheng */ 843fd0a558SYan, Zheng unsigned int detached:1; 855d4f98a2SYan Zheng }; 865d4f98a2SYan Zheng 875d4f98a2SYan Zheng /* 885d4f98a2SYan Zheng * present a block pointer in the backref cache 895d4f98a2SYan Zheng */ 905d4f98a2SYan Zheng struct backref_edge { 915d4f98a2SYan Zheng struct list_head list[2]; 925d4f98a2SYan Zheng struct backref_node *node[2]; 935d4f98a2SYan Zheng }; 945d4f98a2SYan Zheng 955d4f98a2SYan Zheng #define LOWER 0 965d4f98a2SYan Zheng #define UPPER 1 975d4f98a2SYan Zheng 985d4f98a2SYan Zheng struct backref_cache { 995d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 1005d4f98a2SYan Zheng struct rb_root rb_root; 1013fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 1023fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 1033fd0a558SYan, Zheng /* 1043fd0a558SYan, Zheng * list of blocks that have been cowed but some block 1053fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 1063fd0a558SYan, Zheng * new location 1073fd0a558SYan, Zheng */ 1085d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1093fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1103fd0a558SYan, Zheng struct list_head leaves; 1113fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1123fd0a558SYan, Zheng struct list_head changed; 1133fd0a558SYan, Zheng /* list of detached backref node. */ 1143fd0a558SYan, Zheng struct list_head detached; 1153fd0a558SYan, Zheng 1163fd0a558SYan, Zheng u64 last_trans; 1173fd0a558SYan, Zheng 1183fd0a558SYan, Zheng int nr_nodes; 1193fd0a558SYan, Zheng int nr_edges; 1205d4f98a2SYan Zheng }; 1215d4f98a2SYan Zheng 1225d4f98a2SYan Zheng /* 1235d4f98a2SYan Zheng * map address of tree root to tree 1245d4f98a2SYan Zheng */ 1255d4f98a2SYan Zheng struct mapping_node { 1265d4f98a2SYan Zheng struct rb_node rb_node; 1275d4f98a2SYan Zheng u64 bytenr; 1285d4f98a2SYan Zheng void *data; 1295d4f98a2SYan Zheng }; 1305d4f98a2SYan Zheng 1315d4f98a2SYan Zheng struct mapping_tree { 1325d4f98a2SYan Zheng struct rb_root rb_root; 1335d4f98a2SYan Zheng spinlock_t lock; 1345d4f98a2SYan Zheng }; 1355d4f98a2SYan Zheng 1365d4f98a2SYan Zheng /* 1375d4f98a2SYan Zheng * present a tree block to process 1385d4f98a2SYan Zheng */ 1395d4f98a2SYan Zheng struct tree_block { 1405d4f98a2SYan Zheng struct rb_node rb_node; 1415d4f98a2SYan Zheng u64 bytenr; 1425d4f98a2SYan Zheng struct btrfs_key key; 1435d4f98a2SYan Zheng unsigned int level:8; 1445d4f98a2SYan Zheng unsigned int key_ready:1; 1455d4f98a2SYan Zheng }; 1465d4f98a2SYan Zheng 1470257bb82SYan, Zheng #define MAX_EXTENTS 128 1480257bb82SYan, Zheng 1490257bb82SYan, Zheng struct file_extent_cluster { 1500257bb82SYan, Zheng u64 start; 1510257bb82SYan, Zheng u64 end; 1520257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1530257bb82SYan, Zheng unsigned int nr; 1540257bb82SYan, Zheng }; 1550257bb82SYan, Zheng 1565d4f98a2SYan Zheng struct reloc_control { 1575d4f98a2SYan Zheng /* block group to relocate */ 1585d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group; 1595d4f98a2SYan Zheng /* extent tree */ 1605d4f98a2SYan Zheng struct btrfs_root *extent_root; 1615d4f98a2SYan Zheng /* inode for moving data */ 1625d4f98a2SYan Zheng struct inode *data_inode; 1633fd0a558SYan, Zheng 1643fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1653fd0a558SYan, Zheng 1663fd0a558SYan, Zheng struct backref_cache backref_cache; 1673fd0a558SYan, Zheng 1683fd0a558SYan, Zheng struct file_extent_cluster cluster; 1695d4f98a2SYan Zheng /* tree blocks have been processed */ 1705d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1715d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1725d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1735d4f98a2SYan Zheng /* list of reloc trees */ 1745d4f98a2SYan Zheng struct list_head reloc_roots; 1753fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1763fd0a558SYan, Zheng u64 merging_rsv_size; 1773fd0a558SYan, Zheng /* size of relocated tree nodes */ 1783fd0a558SYan, Zheng u64 nodes_relocated; 1793fd0a558SYan, Zheng 1805d4f98a2SYan Zheng u64 search_start; 1815d4f98a2SYan Zheng u64 extents_found; 1823fd0a558SYan, Zheng 1833fd0a558SYan, Zheng unsigned int stage:8; 1843fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1853fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1865d4f98a2SYan Zheng unsigned int found_file_extent:1; 1873fd0a558SYan, Zheng unsigned int commit_transaction:1; 1885d4f98a2SYan Zheng }; 1895d4f98a2SYan Zheng 1905d4f98a2SYan Zheng /* stages of data relocation */ 1915d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1925d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1935d4f98a2SYan Zheng 1943fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 1953fd0a558SYan, Zheng struct backref_node *node); 1963fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 1973fd0a558SYan, Zheng struct backref_node *node); 1985d4f98a2SYan Zheng 1995d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 2005d4f98a2SYan Zheng { 2016bef4d31SEric Paris tree->rb_root = RB_ROOT; 2025d4f98a2SYan Zheng spin_lock_init(&tree->lock); 2035d4f98a2SYan Zheng } 2045d4f98a2SYan Zheng 2055d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 2065d4f98a2SYan Zheng { 2075d4f98a2SYan Zheng int i; 2086bef4d31SEric Paris cache->rb_root = RB_ROOT; 2095d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2105d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2113fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2123fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2133fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2145d4f98a2SYan Zheng } 2155d4f98a2SYan Zheng 2163fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2175d4f98a2SYan Zheng { 2183fd0a558SYan, Zheng struct backref_node *node; 2193fd0a558SYan, Zheng int i; 2203fd0a558SYan, Zheng 2213fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2223fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2233fd0a558SYan, Zheng struct backref_node, list); 2243fd0a558SYan, Zheng remove_backref_node(cache, node); 2253fd0a558SYan, Zheng } 2263fd0a558SYan, Zheng 2273fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2283fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2293fd0a558SYan, Zheng struct backref_node, lower); 2303fd0a558SYan, Zheng remove_backref_node(cache, node); 2313fd0a558SYan, Zheng } 2323fd0a558SYan, Zheng 2333fd0a558SYan, Zheng cache->last_trans = 0; 2343fd0a558SYan, Zheng 2353fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2363fd0a558SYan, Zheng BUG_ON(!list_empty(&cache->pending[i])); 2373fd0a558SYan, Zheng BUG_ON(!list_empty(&cache->changed)); 2383fd0a558SYan, Zheng BUG_ON(!list_empty(&cache->detached)); 2393fd0a558SYan, Zheng BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root)); 2403fd0a558SYan, Zheng BUG_ON(cache->nr_nodes); 2413fd0a558SYan, Zheng BUG_ON(cache->nr_edges); 2423fd0a558SYan, Zheng } 2433fd0a558SYan, Zheng 2443fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2453fd0a558SYan, Zheng { 2463fd0a558SYan, Zheng struct backref_node *node; 2473fd0a558SYan, Zheng 2483fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2493fd0a558SYan, Zheng if (node) { 2503fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2515d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2525d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2535d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2543fd0a558SYan, Zheng cache->nr_nodes++; 2553fd0a558SYan, Zheng } 2563fd0a558SYan, Zheng return node; 2573fd0a558SYan, Zheng } 2583fd0a558SYan, Zheng 2593fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 2603fd0a558SYan, Zheng struct backref_node *node) 2613fd0a558SYan, Zheng { 2623fd0a558SYan, Zheng if (node) { 2633fd0a558SYan, Zheng cache->nr_nodes--; 2643fd0a558SYan, Zheng kfree(node); 2653fd0a558SYan, Zheng } 2663fd0a558SYan, Zheng } 2673fd0a558SYan, Zheng 2683fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 2693fd0a558SYan, Zheng { 2703fd0a558SYan, Zheng struct backref_edge *edge; 2713fd0a558SYan, Zheng 2723fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 2733fd0a558SYan, Zheng if (edge) 2743fd0a558SYan, Zheng cache->nr_edges++; 2753fd0a558SYan, Zheng return edge; 2763fd0a558SYan, Zheng } 2773fd0a558SYan, Zheng 2783fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 2793fd0a558SYan, Zheng struct backref_edge *edge) 2803fd0a558SYan, Zheng { 2813fd0a558SYan, Zheng if (edge) { 2823fd0a558SYan, Zheng cache->nr_edges--; 2833fd0a558SYan, Zheng kfree(edge); 2843fd0a558SYan, Zheng } 2855d4f98a2SYan Zheng } 2865d4f98a2SYan Zheng 2875d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 2885d4f98a2SYan Zheng struct rb_node *node) 2895d4f98a2SYan Zheng { 2905d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 2915d4f98a2SYan Zheng struct rb_node *parent = NULL; 2925d4f98a2SYan Zheng struct tree_entry *entry; 2935d4f98a2SYan Zheng 2945d4f98a2SYan Zheng while (*p) { 2955d4f98a2SYan Zheng parent = *p; 2965d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 2975d4f98a2SYan Zheng 2985d4f98a2SYan Zheng if (bytenr < entry->bytenr) 2995d4f98a2SYan Zheng p = &(*p)->rb_left; 3005d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3015d4f98a2SYan Zheng p = &(*p)->rb_right; 3025d4f98a2SYan Zheng else 3035d4f98a2SYan Zheng return parent; 3045d4f98a2SYan Zheng } 3055d4f98a2SYan Zheng 3065d4f98a2SYan Zheng rb_link_node(node, parent, p); 3075d4f98a2SYan Zheng rb_insert_color(node, root); 3085d4f98a2SYan Zheng return NULL; 3095d4f98a2SYan Zheng } 3105d4f98a2SYan Zheng 3115d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3125d4f98a2SYan Zheng { 3135d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3145d4f98a2SYan Zheng struct tree_entry *entry; 3155d4f98a2SYan Zheng 3165d4f98a2SYan Zheng while (n) { 3175d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3185d4f98a2SYan Zheng 3195d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3205d4f98a2SYan Zheng n = n->rb_left; 3215d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3225d4f98a2SYan Zheng n = n->rb_right; 3235d4f98a2SYan Zheng else 3245d4f98a2SYan Zheng return n; 3255d4f98a2SYan Zheng } 3265d4f98a2SYan Zheng return NULL; 3275d4f98a2SYan Zheng } 3285d4f98a2SYan Zheng 32948a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 33043c04fb1SJeff Mahoney { 33143c04fb1SJeff Mahoney 33243c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 33343c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 33443c04fb1SJeff Mahoney rb_node); 33543c04fb1SJeff Mahoney if (bnode->root) 33643c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 33743c04fb1SJeff Mahoney btrfs_panic(fs_info, errno, "Inconsistency in backref cache " 338c1c9ff7cSGeert Uytterhoeven "found at offset %llu\n", bytenr); 33943c04fb1SJeff Mahoney } 34043c04fb1SJeff Mahoney 3415d4f98a2SYan Zheng /* 3425d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3435d4f98a2SYan Zheng */ 3445d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3455d4f98a2SYan Zheng struct backref_edge *edges[], 3465d4f98a2SYan Zheng int *index) 3475d4f98a2SYan Zheng { 3485d4f98a2SYan Zheng struct backref_edge *edge; 3495d4f98a2SYan Zheng int idx = *index; 3505d4f98a2SYan Zheng 3515d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3525d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3535d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3545d4f98a2SYan Zheng edges[idx++] = edge; 3555d4f98a2SYan Zheng node = edge->node[UPPER]; 3565d4f98a2SYan Zheng } 3573fd0a558SYan, Zheng BUG_ON(node->detached); 3585d4f98a2SYan Zheng *index = idx; 3595d4f98a2SYan Zheng return node; 3605d4f98a2SYan Zheng } 3615d4f98a2SYan Zheng 3625d4f98a2SYan Zheng /* 3635d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 3645d4f98a2SYan Zheng */ 3655d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 3665d4f98a2SYan Zheng int *index) 3675d4f98a2SYan Zheng { 3685d4f98a2SYan Zheng struct backref_edge *edge; 3695d4f98a2SYan Zheng struct backref_node *lower; 3705d4f98a2SYan Zheng int idx = *index; 3715d4f98a2SYan Zheng 3725d4f98a2SYan Zheng while (idx > 0) { 3735d4f98a2SYan Zheng edge = edges[idx - 1]; 3745d4f98a2SYan Zheng lower = edge->node[LOWER]; 3755d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 3765d4f98a2SYan Zheng idx--; 3775d4f98a2SYan Zheng continue; 3785d4f98a2SYan Zheng } 3795d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 3805d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3815d4f98a2SYan Zheng edges[idx - 1] = edge; 3825d4f98a2SYan Zheng *index = idx; 3835d4f98a2SYan Zheng return edge->node[UPPER]; 3845d4f98a2SYan Zheng } 3855d4f98a2SYan Zheng *index = 0; 3865d4f98a2SYan Zheng return NULL; 3875d4f98a2SYan Zheng } 3885d4f98a2SYan Zheng 3893fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 3905d4f98a2SYan Zheng { 3915d4f98a2SYan Zheng if (node->locked) { 3925d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 3935d4f98a2SYan Zheng node->locked = 0; 3945d4f98a2SYan Zheng } 3953fd0a558SYan, Zheng } 3963fd0a558SYan, Zheng 3973fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 3983fd0a558SYan, Zheng { 3993fd0a558SYan, Zheng if (node->eb) { 4003fd0a558SYan, Zheng unlock_node_buffer(node); 4015d4f98a2SYan Zheng free_extent_buffer(node->eb); 4025d4f98a2SYan Zheng node->eb = NULL; 4035d4f98a2SYan Zheng } 4045d4f98a2SYan Zheng } 4055d4f98a2SYan Zheng 4065d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4075d4f98a2SYan Zheng struct backref_node *node) 4085d4f98a2SYan Zheng { 4095d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4105d4f98a2SYan Zheng 4115d4f98a2SYan Zheng drop_node_buffer(node); 4123fd0a558SYan, Zheng list_del(&node->list); 4135d4f98a2SYan Zheng list_del(&node->lower); 4143fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4155d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4163fd0a558SYan, Zheng free_backref_node(tree, node); 4175d4f98a2SYan Zheng } 4185d4f98a2SYan Zheng 4195d4f98a2SYan Zheng /* 4205d4f98a2SYan Zheng * remove a backref node from the backref cache 4215d4f98a2SYan Zheng */ 4225d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4235d4f98a2SYan Zheng struct backref_node *node) 4245d4f98a2SYan Zheng { 4255d4f98a2SYan Zheng struct backref_node *upper; 4265d4f98a2SYan Zheng struct backref_edge *edge; 4275d4f98a2SYan Zheng 4285d4f98a2SYan Zheng if (!node) 4295d4f98a2SYan Zheng return; 4305d4f98a2SYan Zheng 4313fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4325d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4335d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4345d4f98a2SYan Zheng list[LOWER]); 4355d4f98a2SYan Zheng upper = edge->node[UPPER]; 4365d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4375d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4383fd0a558SYan, Zheng free_backref_edge(cache, edge); 4393fd0a558SYan, Zheng 4403fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4413fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4423fd0a558SYan, Zheng drop_backref_node(cache, node); 4433fd0a558SYan, Zheng node = upper; 4443fd0a558SYan, Zheng node->lowest = 1; 4453fd0a558SYan, Zheng continue; 4463fd0a558SYan, Zheng } 4475d4f98a2SYan Zheng /* 4483fd0a558SYan, Zheng * add the node to leaf node list if no other 4495d4f98a2SYan Zheng * child block cached. 4505d4f98a2SYan Zheng */ 4515d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4523fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4535d4f98a2SYan Zheng upper->lowest = 1; 4545d4f98a2SYan Zheng } 4555d4f98a2SYan Zheng } 4563fd0a558SYan, Zheng 4575d4f98a2SYan Zheng drop_backref_node(cache, node); 4585d4f98a2SYan Zheng } 4595d4f98a2SYan Zheng 4603fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 4613fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 4623fd0a558SYan, Zheng { 4633fd0a558SYan, Zheng struct rb_node *rb_node; 4643fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 4653fd0a558SYan, Zheng node->bytenr = bytenr; 4663fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 46743c04fb1SJeff Mahoney if (rb_node) 46843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 4693fd0a558SYan, Zheng } 4703fd0a558SYan, Zheng 4713fd0a558SYan, Zheng /* 4723fd0a558SYan, Zheng * update backref cache after a transaction commit 4733fd0a558SYan, Zheng */ 4743fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 4753fd0a558SYan, Zheng struct backref_cache *cache) 4763fd0a558SYan, Zheng { 4773fd0a558SYan, Zheng struct backref_node *node; 4783fd0a558SYan, Zheng int level = 0; 4793fd0a558SYan, Zheng 4803fd0a558SYan, Zheng if (cache->last_trans == 0) { 4813fd0a558SYan, Zheng cache->last_trans = trans->transid; 4823fd0a558SYan, Zheng return 0; 4833fd0a558SYan, Zheng } 4843fd0a558SYan, Zheng 4853fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 4863fd0a558SYan, Zheng return 0; 4873fd0a558SYan, Zheng 4883fd0a558SYan, Zheng /* 4893fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 4903fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 4913fd0a558SYan, Zheng * so the detached nodes are no longer useful. 4923fd0a558SYan, Zheng */ 4933fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 4943fd0a558SYan, Zheng node = list_entry(cache->detached.next, 4953fd0a558SYan, Zheng struct backref_node, list); 4963fd0a558SYan, Zheng remove_backref_node(cache, node); 4973fd0a558SYan, Zheng } 4983fd0a558SYan, Zheng 4993fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 5003fd0a558SYan, Zheng node = list_entry(cache->changed.next, 5013fd0a558SYan, Zheng struct backref_node, list); 5023fd0a558SYan, Zheng list_del_init(&node->list); 5033fd0a558SYan, Zheng BUG_ON(node->pending); 5043fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5053fd0a558SYan, Zheng } 5063fd0a558SYan, Zheng 5073fd0a558SYan, Zheng /* 5083fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5093fd0a558SYan, Zheng * errors during processing the pending nodes. 5103fd0a558SYan, Zheng */ 5113fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5123fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5133fd0a558SYan, Zheng BUG_ON(!node->pending); 5143fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5153fd0a558SYan, Zheng continue; 5163fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5173fd0a558SYan, Zheng } 5183fd0a558SYan, Zheng } 5193fd0a558SYan, Zheng 5203fd0a558SYan, Zheng cache->last_trans = 0; 5213fd0a558SYan, Zheng return 1; 5223fd0a558SYan, Zheng } 5233fd0a558SYan, Zheng 524f2a97a9dSDavid Sterba 5253fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5263fd0a558SYan, Zheng { 5273fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5283fd0a558SYan, Zheng 5293fd0a558SYan, Zheng if (!root->ref_cows) 5303fd0a558SYan, Zheng return 0; 5313fd0a558SYan, Zheng 5323fd0a558SYan, Zheng reloc_root = root->reloc_root; 5333fd0a558SYan, Zheng if (!reloc_root) 5343fd0a558SYan, Zheng return 0; 5353fd0a558SYan, Zheng 5363fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5373fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5383fd0a558SYan, Zheng return 0; 5393fd0a558SYan, Zheng /* 5403fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5413fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5423fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5433fd0a558SYan, Zheng * relocation. 5443fd0a558SYan, Zheng */ 5453fd0a558SYan, Zheng return 1; 5463fd0a558SYan, Zheng } 5475d4f98a2SYan Zheng /* 5485d4f98a2SYan Zheng * find reloc tree by address of tree root 5495d4f98a2SYan Zheng */ 5505d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5515d4f98a2SYan Zheng u64 bytenr) 5525d4f98a2SYan Zheng { 5535d4f98a2SYan Zheng struct rb_node *rb_node; 5545d4f98a2SYan Zheng struct mapping_node *node; 5555d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5565d4f98a2SYan Zheng 5575d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5585d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5595d4f98a2SYan Zheng if (rb_node) { 5605d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5615d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5625d4f98a2SYan Zheng } 5635d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5645d4f98a2SYan Zheng return root; 5655d4f98a2SYan Zheng } 5665d4f98a2SYan Zheng 5675d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5685d4f98a2SYan Zheng { 5695d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5705d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5715d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 5725d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 5735d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 5745d4f98a2SYan Zheng root_objectid == BTRFS_CSUM_TREE_OBJECTID) 5755d4f98a2SYan Zheng return 1; 5765d4f98a2SYan Zheng return 0; 5775d4f98a2SYan Zheng } 5785d4f98a2SYan Zheng 5795d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 5805d4f98a2SYan Zheng u64 root_objectid) 5815d4f98a2SYan Zheng { 5825d4f98a2SYan Zheng struct btrfs_key key; 5835d4f98a2SYan Zheng 5845d4f98a2SYan Zheng key.objectid = root_objectid; 5855d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 5865d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 5875d4f98a2SYan Zheng key.offset = 0; 5885d4f98a2SYan Zheng else 5895d4f98a2SYan Zheng key.offset = (u64)-1; 5905d4f98a2SYan Zheng 591c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 5925d4f98a2SYan Zheng } 5935d4f98a2SYan Zheng 5945d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 5955d4f98a2SYan Zheng static noinline_for_stack 5965d4f98a2SYan Zheng struct btrfs_root *find_tree_root(struct reloc_control *rc, 5975d4f98a2SYan Zheng struct extent_buffer *leaf, 5985d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0) 5995d4f98a2SYan Zheng { 6005d4f98a2SYan Zheng struct btrfs_root *root; 6015d4f98a2SYan Zheng u64 root_objectid = btrfs_ref_root_v0(leaf, ref0); 6025d4f98a2SYan Zheng u64 generation = btrfs_ref_generation_v0(leaf, ref0); 6035d4f98a2SYan Zheng 6045d4f98a2SYan Zheng BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID); 6055d4f98a2SYan Zheng 6065d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, root_objectid); 6075d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 6085d4f98a2SYan Zheng 6095d4f98a2SYan Zheng if (root->ref_cows && 6105d4f98a2SYan Zheng generation != btrfs_root_generation(&root->root_item)) 6115d4f98a2SYan Zheng return NULL; 6125d4f98a2SYan Zheng 6135d4f98a2SYan Zheng return root; 6145d4f98a2SYan Zheng } 6155d4f98a2SYan Zheng #endif 6165d4f98a2SYan Zheng 6175d4f98a2SYan Zheng static noinline_for_stack 6185d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 6195d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 6205d4f98a2SYan Zheng { 6213173a18fSJosef Bacik struct btrfs_key key; 6225d4f98a2SYan Zheng struct btrfs_extent_item *ei; 6235d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6245d4f98a2SYan Zheng u32 item_size; 6255d4f98a2SYan Zheng 6263173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6273173a18fSJosef Bacik 6285d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 6295d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 6305d4f98a2SYan Zheng if (item_size < sizeof(*ei)) { 6315d4f98a2SYan Zheng WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 6325d4f98a2SYan Zheng return 1; 6335d4f98a2SYan Zheng } 6345d4f98a2SYan Zheng #endif 6355d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6365d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6375d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6385d4f98a2SYan Zheng 6393173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6403173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6415d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6425d4f98a2SYan Zheng return 1; 6435d4f98a2SYan Zheng } 644d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 645d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 646d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 647d062d13cSJosef Bacik return 1; 648d062d13cSJosef Bacik } 6495d4f98a2SYan Zheng 6503173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6515d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6525d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6533173a18fSJosef Bacik } else { 6543173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6553173a18fSJosef Bacik } 6565d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6575d4f98a2SYan Zheng return 0; 6585d4f98a2SYan Zheng } 6595d4f98a2SYan Zheng 6605d4f98a2SYan Zheng /* 6615d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6625d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6635d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6645d4f98a2SYan Zheng * 6655d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 6665d4f98a2SYan Zheng * to find upper level blocks that refernece the block, and then check 6675d4f98a2SYan Zheng * bakcrefs of these upper level blocks recursively. the recursion stop 6685d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6695d4f98a2SYan Zheng * 6705d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6715d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6725d4f98a2SYan Zheng * block are also cached. 6735d4f98a2SYan Zheng */ 6743fd0a558SYan, Zheng static noinline_for_stack 6753fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6765d4f98a2SYan Zheng struct btrfs_key *node_key, 6775d4f98a2SYan Zheng int level, u64 bytenr) 6785d4f98a2SYan Zheng { 6793fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 6805d4f98a2SYan Zheng struct btrfs_path *path1; 6815d4f98a2SYan Zheng struct btrfs_path *path2; 6825d4f98a2SYan Zheng struct extent_buffer *eb; 6835d4f98a2SYan Zheng struct btrfs_root *root; 6845d4f98a2SYan Zheng struct backref_node *cur; 6855d4f98a2SYan Zheng struct backref_node *upper; 6865d4f98a2SYan Zheng struct backref_node *lower; 6875d4f98a2SYan Zheng struct backref_node *node = NULL; 6885d4f98a2SYan Zheng struct backref_node *exist = NULL; 6895d4f98a2SYan Zheng struct backref_edge *edge; 6905d4f98a2SYan Zheng struct rb_node *rb_node; 6915d4f98a2SYan Zheng struct btrfs_key key; 6925d4f98a2SYan Zheng unsigned long end; 6935d4f98a2SYan Zheng unsigned long ptr; 6945d4f98a2SYan Zheng LIST_HEAD(list); 6953fd0a558SYan, Zheng LIST_HEAD(useless); 6963fd0a558SYan, Zheng int cowonly; 6975d4f98a2SYan Zheng int ret; 6985d4f98a2SYan Zheng int err = 0; 699b6c60c80SJosef Bacik bool need_check = true; 7005d4f98a2SYan Zheng 7015d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 7025d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 7035d4f98a2SYan Zheng if (!path1 || !path2) { 7045d4f98a2SYan Zheng err = -ENOMEM; 7055d4f98a2SYan Zheng goto out; 7065d4f98a2SYan Zheng } 707026fd317SJosef Bacik path1->reada = 1; 708026fd317SJosef Bacik path2->reada = 2; 7095d4f98a2SYan Zheng 7103fd0a558SYan, Zheng node = alloc_backref_node(cache); 7115d4f98a2SYan Zheng if (!node) { 7125d4f98a2SYan Zheng err = -ENOMEM; 7135d4f98a2SYan Zheng goto out; 7145d4f98a2SYan Zheng } 7155d4f98a2SYan Zheng 7165d4f98a2SYan Zheng node->bytenr = bytenr; 7175d4f98a2SYan Zheng node->level = level; 7185d4f98a2SYan Zheng node->lowest = 1; 7195d4f98a2SYan Zheng cur = node; 7205d4f98a2SYan Zheng again: 7215d4f98a2SYan Zheng end = 0; 7225d4f98a2SYan Zheng ptr = 0; 7235d4f98a2SYan Zheng key.objectid = cur->bytenr; 7243173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7255d4f98a2SYan Zheng key.offset = (u64)-1; 7265d4f98a2SYan Zheng 7275d4f98a2SYan Zheng path1->search_commit_root = 1; 7285d4f98a2SYan Zheng path1->skip_locking = 1; 7295d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7305d4f98a2SYan Zheng 0, 0); 7315d4f98a2SYan Zheng if (ret < 0) { 7325d4f98a2SYan Zheng err = ret; 7335d4f98a2SYan Zheng goto out; 7345d4f98a2SYan Zheng } 7355d4f98a2SYan Zheng BUG_ON(!ret || !path1->slots[0]); 7365d4f98a2SYan Zheng 7375d4f98a2SYan Zheng path1->slots[0]--; 7385d4f98a2SYan Zheng 7395d4f98a2SYan Zheng WARN_ON(cur->checked); 7405d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7415d4f98a2SYan Zheng /* 74270f23fd6SJustin P. Mattock * the backref was added previously when processing 7435d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7445d4f98a2SYan Zheng */ 7455d4f98a2SYan Zheng BUG_ON(!list_is_singular(&cur->upper)); 7465d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7475d4f98a2SYan Zheng list[LOWER]); 7485d4f98a2SYan Zheng BUG_ON(!list_empty(&edge->list[UPPER])); 7495d4f98a2SYan Zheng exist = edge->node[UPPER]; 7505d4f98a2SYan Zheng /* 7515d4f98a2SYan Zheng * add the upper level block to pending list if we need 7525d4f98a2SYan Zheng * check its backrefs 7535d4f98a2SYan Zheng */ 7545d4f98a2SYan Zheng if (!exist->checked) 7555d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7565d4f98a2SYan Zheng } else { 7575d4f98a2SYan Zheng exist = NULL; 7585d4f98a2SYan Zheng } 7595d4f98a2SYan Zheng 7605d4f98a2SYan Zheng while (1) { 7615d4f98a2SYan Zheng cond_resched(); 7625d4f98a2SYan Zheng eb = path1->nodes[0]; 7635d4f98a2SYan Zheng 7645d4f98a2SYan Zheng if (ptr >= end) { 7655d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7665d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7675d4f98a2SYan Zheng if (ret < 0) { 7685d4f98a2SYan Zheng err = ret; 7695d4f98a2SYan Zheng goto out; 7705d4f98a2SYan Zheng } 7715d4f98a2SYan Zheng if (ret > 0) 7725d4f98a2SYan Zheng break; 7735d4f98a2SYan Zheng eb = path1->nodes[0]; 7745d4f98a2SYan Zheng } 7755d4f98a2SYan Zheng 7765d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7775d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7785d4f98a2SYan Zheng WARN_ON(exist); 7795d4f98a2SYan Zheng break; 7805d4f98a2SYan Zheng } 7815d4f98a2SYan Zheng 7823173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7833173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7845d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7855d4f98a2SYan Zheng &ptr, &end); 7865d4f98a2SYan Zheng if (ret) 7875d4f98a2SYan Zheng goto next; 7885d4f98a2SYan Zheng } 7895d4f98a2SYan Zheng } 7905d4f98a2SYan Zheng 7915d4f98a2SYan Zheng if (ptr < end) { 7925d4f98a2SYan Zheng /* update key for inline back ref */ 7935d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 7945d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 7955d4f98a2SYan Zheng key.type = btrfs_extent_inline_ref_type(eb, iref); 7965d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 7975d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 7985d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 7995d4f98a2SYan Zheng } 8005d4f98a2SYan Zheng 8015d4f98a2SYan Zheng if (exist && 8025d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8035d4f98a2SYan Zheng exist->owner == key.offset) || 8045d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8055d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8065d4f98a2SYan Zheng exist = NULL; 8075d4f98a2SYan Zheng goto next; 8085d4f98a2SYan Zheng } 8095d4f98a2SYan Zheng 8105d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 8115d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY || 8125d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 8133fd0a558SYan, Zheng if (key.type == BTRFS_EXTENT_REF_V0_KEY) { 8145d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 8155d4f98a2SYan Zheng ref0 = btrfs_item_ptr(eb, path1->slots[0], 8165d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 8173fd0a558SYan, Zheng if (key.objectid == key.offset) { 818046f264fSYan, Zheng root = find_tree_root(rc, eb, ref0); 8193fd0a558SYan, Zheng if (root && !should_ignore_root(root)) 8205d4f98a2SYan Zheng cur->root = root; 8215d4f98a2SYan Zheng else 8223fd0a558SYan, Zheng list_add(&cur->list, &useless); 8235d4f98a2SYan Zheng break; 8245d4f98a2SYan Zheng } 825046f264fSYan, Zheng if (is_cowonly_root(btrfs_ref_root_v0(eb, 826046f264fSYan, Zheng ref0))) 827046f264fSYan, Zheng cur->cowonly = 1; 8283fd0a558SYan, Zheng } 8295d4f98a2SYan Zheng #else 8305d4f98a2SYan Zheng BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); 8315d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8325d4f98a2SYan Zheng #endif 8335d4f98a2SYan Zheng if (key.objectid == key.offset) { 8345d4f98a2SYan Zheng /* 8355d4f98a2SYan Zheng * only root blocks of reloc trees use 8365d4f98a2SYan Zheng * backref of this type. 8375d4f98a2SYan Zheng */ 8385d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 8395d4f98a2SYan Zheng BUG_ON(!root); 8405d4f98a2SYan Zheng cur->root = root; 8415d4f98a2SYan Zheng break; 8425d4f98a2SYan Zheng } 8435d4f98a2SYan Zheng 8443fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8455d4f98a2SYan Zheng if (!edge) { 8465d4f98a2SYan Zheng err = -ENOMEM; 8475d4f98a2SYan Zheng goto out; 8485d4f98a2SYan Zheng } 8495d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8505d4f98a2SYan Zheng if (!rb_node) { 8513fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8525d4f98a2SYan Zheng if (!upper) { 8533fd0a558SYan, Zheng free_backref_edge(cache, edge); 8545d4f98a2SYan Zheng err = -ENOMEM; 8555d4f98a2SYan Zheng goto out; 8565d4f98a2SYan Zheng } 8575d4f98a2SYan Zheng upper->bytenr = key.offset; 8585d4f98a2SYan Zheng upper->level = cur->level + 1; 8595d4f98a2SYan Zheng /* 8605d4f98a2SYan Zheng * backrefs for the upper level block isn't 8615d4f98a2SYan Zheng * cached, add the block to pending list 8625d4f98a2SYan Zheng */ 8635d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8645d4f98a2SYan Zheng } else { 8655d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8665d4f98a2SYan Zheng rb_node); 8673fd0a558SYan, Zheng BUG_ON(!upper->checked); 8685d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8695d4f98a2SYan Zheng } 8703fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8715d4f98a2SYan Zheng edge->node[LOWER] = cur; 8723fd0a558SYan, Zheng edge->node[UPPER] = upper; 8735d4f98a2SYan Zheng 8745d4f98a2SYan Zheng goto next; 8755d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8765d4f98a2SYan Zheng goto next; 8775d4f98a2SYan Zheng } 8785d4f98a2SYan Zheng 8795d4f98a2SYan Zheng /* key.type == BTRFS_TREE_BLOCK_REF_KEY */ 8805d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8815d4f98a2SYan Zheng if (IS_ERR(root)) { 8825d4f98a2SYan Zheng err = PTR_ERR(root); 8835d4f98a2SYan Zheng goto out; 8845d4f98a2SYan Zheng } 8855d4f98a2SYan Zheng 8863fd0a558SYan, Zheng if (!root->ref_cows) 8873fd0a558SYan, Zheng cur->cowonly = 1; 8883fd0a558SYan, Zheng 8895d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 8905d4f98a2SYan Zheng /* tree root */ 8915d4f98a2SYan Zheng BUG_ON(btrfs_root_bytenr(&root->root_item) != 8925d4f98a2SYan Zheng cur->bytenr); 8933fd0a558SYan, Zheng if (should_ignore_root(root)) 8943fd0a558SYan, Zheng list_add(&cur->list, &useless); 8953fd0a558SYan, Zheng else 8965d4f98a2SYan Zheng cur->root = root; 8975d4f98a2SYan Zheng break; 8985d4f98a2SYan Zheng } 8995d4f98a2SYan Zheng 9005d4f98a2SYan Zheng level = cur->level + 1; 9015d4f98a2SYan Zheng 9025d4f98a2SYan Zheng /* 9035d4f98a2SYan Zheng * searching the tree to find upper level blocks 9045d4f98a2SYan Zheng * reference the block. 9055d4f98a2SYan Zheng */ 9065d4f98a2SYan Zheng path2->search_commit_root = 1; 9075d4f98a2SYan Zheng path2->skip_locking = 1; 9085d4f98a2SYan Zheng path2->lowest_level = level; 9095d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 9105d4f98a2SYan Zheng path2->lowest_level = 0; 9115d4f98a2SYan Zheng if (ret < 0) { 9125d4f98a2SYan Zheng err = ret; 9135d4f98a2SYan Zheng goto out; 9145d4f98a2SYan Zheng } 91533c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 91633c66f43SYan Zheng path2->slots[level]--; 9175d4f98a2SYan Zheng 9185d4f98a2SYan Zheng eb = path2->nodes[level]; 9195d4f98a2SYan Zheng WARN_ON(btrfs_node_blockptr(eb, path2->slots[level]) != 9205d4f98a2SYan Zheng cur->bytenr); 9215d4f98a2SYan Zheng 9225d4f98a2SYan Zheng lower = cur; 923b6c60c80SJosef Bacik need_check = true; 9245d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9255d4f98a2SYan Zheng if (!path2->nodes[level]) { 9265d4f98a2SYan Zheng BUG_ON(btrfs_root_bytenr(&root->root_item) != 9275d4f98a2SYan Zheng lower->bytenr); 9283fd0a558SYan, Zheng if (should_ignore_root(root)) 9293fd0a558SYan, Zheng list_add(&lower->list, &useless); 9303fd0a558SYan, Zheng else 9315d4f98a2SYan Zheng lower->root = root; 9325d4f98a2SYan Zheng break; 9335d4f98a2SYan Zheng } 9345d4f98a2SYan Zheng 9353fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9365d4f98a2SYan Zheng if (!edge) { 9375d4f98a2SYan Zheng err = -ENOMEM; 9385d4f98a2SYan Zheng goto out; 9395d4f98a2SYan Zheng } 9405d4f98a2SYan Zheng 9415d4f98a2SYan Zheng eb = path2->nodes[level]; 9425d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9435d4f98a2SYan Zheng if (!rb_node) { 9443fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9455d4f98a2SYan Zheng if (!upper) { 9463fd0a558SYan, Zheng free_backref_edge(cache, edge); 9475d4f98a2SYan Zheng err = -ENOMEM; 9485d4f98a2SYan Zheng goto out; 9495d4f98a2SYan Zheng } 9505d4f98a2SYan Zheng upper->bytenr = eb->start; 9515d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9525d4f98a2SYan Zheng upper->level = lower->level + 1; 9533fd0a558SYan, Zheng if (!root->ref_cows) 9543fd0a558SYan, Zheng upper->cowonly = 1; 9555d4f98a2SYan Zheng 9565d4f98a2SYan Zheng /* 9575d4f98a2SYan Zheng * if we know the block isn't shared 9585d4f98a2SYan Zheng * we can void checking its backrefs. 9595d4f98a2SYan Zheng */ 9605d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9615d4f98a2SYan Zheng upper->checked = 0; 9625d4f98a2SYan Zheng else 9635d4f98a2SYan Zheng upper->checked = 1; 9645d4f98a2SYan Zheng 9655d4f98a2SYan Zheng /* 9665d4f98a2SYan Zheng * add the block to pending list if we 967b6c60c80SJosef Bacik * need check its backrefs, we only do this once 968b6c60c80SJosef Bacik * while walking up a tree as we will catch 969b6c60c80SJosef Bacik * anything else later on. 9705d4f98a2SYan Zheng */ 971b6c60c80SJosef Bacik if (!upper->checked && need_check) { 972b6c60c80SJosef Bacik need_check = false; 9735d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9745d4f98a2SYan Zheng &list); 9755d4f98a2SYan Zheng } else 9765d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9775d4f98a2SYan Zheng } else { 9785d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9795d4f98a2SYan Zheng rb_node); 9805d4f98a2SYan Zheng BUG_ON(!upper->checked); 9815d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9823fd0a558SYan, Zheng if (!upper->owner) 9833fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 9845d4f98a2SYan Zheng } 9855d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 9865d4f98a2SYan Zheng edge->node[LOWER] = lower; 9873fd0a558SYan, Zheng edge->node[UPPER] = upper; 9885d4f98a2SYan Zheng 9895d4f98a2SYan Zheng if (rb_node) 9905d4f98a2SYan Zheng break; 9915d4f98a2SYan Zheng lower = upper; 9925d4f98a2SYan Zheng upper = NULL; 9935d4f98a2SYan Zheng } 994b3b4aa74SDavid Sterba btrfs_release_path(path2); 9955d4f98a2SYan Zheng next: 9965d4f98a2SYan Zheng if (ptr < end) { 9975d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 9985d4f98a2SYan Zheng if (ptr >= end) { 9995d4f98a2SYan Zheng WARN_ON(ptr > end); 10005d4f98a2SYan Zheng ptr = 0; 10015d4f98a2SYan Zheng end = 0; 10025d4f98a2SYan Zheng } 10035d4f98a2SYan Zheng } 10045d4f98a2SYan Zheng if (ptr >= end) 10055d4f98a2SYan Zheng path1->slots[0]++; 10065d4f98a2SYan Zheng } 1007b3b4aa74SDavid Sterba btrfs_release_path(path1); 10085d4f98a2SYan Zheng 10095d4f98a2SYan Zheng cur->checked = 1; 10105d4f98a2SYan Zheng WARN_ON(exist); 10115d4f98a2SYan Zheng 10125d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10135d4f98a2SYan Zheng if (!list_empty(&list)) { 10145d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10155d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10165d4f98a2SYan Zheng cur = edge->node[UPPER]; 10175d4f98a2SYan Zheng goto again; 10185d4f98a2SYan Zheng } 10195d4f98a2SYan Zheng 10205d4f98a2SYan Zheng /* 10215d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10225d4f98a2SYan Zheng * into the cache. 10235d4f98a2SYan Zheng */ 10245d4f98a2SYan Zheng BUG_ON(!node->checked); 10253fd0a558SYan, Zheng cowonly = node->cowonly; 10263fd0a558SYan, Zheng if (!cowonly) { 10273fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10283fd0a558SYan, Zheng &node->rb_node); 102943c04fb1SJeff Mahoney if (rb_node) 103043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10313fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10323fd0a558SYan, Zheng } 10335d4f98a2SYan Zheng 10345d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10355d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10365d4f98a2SYan Zheng 10375d4f98a2SYan Zheng while (!list_empty(&list)) { 10385d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10395d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10405d4f98a2SYan Zheng upper = edge->node[UPPER]; 10413fd0a558SYan, Zheng if (upper->detached) { 10423fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10433fd0a558SYan, Zheng lower = edge->node[LOWER]; 10443fd0a558SYan, Zheng free_backref_edge(cache, edge); 10453fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10463fd0a558SYan, Zheng list_add(&lower->list, &useless); 10473fd0a558SYan, Zheng continue; 10483fd0a558SYan, Zheng } 10495d4f98a2SYan Zheng 10505d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10515d4f98a2SYan Zheng if (upper->lowest) { 10525d4f98a2SYan Zheng list_del_init(&upper->lower); 10535d4f98a2SYan Zheng upper->lowest = 0; 10545d4f98a2SYan Zheng } 10555d4f98a2SYan Zheng 10565d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10575d4f98a2SYan Zheng continue; 10585d4f98a2SYan Zheng } 10595d4f98a2SYan Zheng 10605d4f98a2SYan Zheng BUG_ON(!upper->checked); 10613fd0a558SYan, Zheng BUG_ON(cowonly != upper->cowonly); 10623fd0a558SYan, Zheng if (!cowonly) { 10635d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10645d4f98a2SYan Zheng &upper->rb_node); 106543c04fb1SJeff Mahoney if (rb_node) 106643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 106743c04fb1SJeff Mahoney upper->bytenr); 10683fd0a558SYan, Zheng } 10695d4f98a2SYan Zheng 10705d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10715d4f98a2SYan Zheng 10725d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 10735d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10745d4f98a2SYan Zheng } 10753fd0a558SYan, Zheng /* 10763fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 10773fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 10783fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 10793fd0a558SYan, Zheng * lookup. 10803fd0a558SYan, Zheng */ 10813fd0a558SYan, Zheng while (!list_empty(&useless)) { 10823fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 10833fd0a558SYan, Zheng list_del_init(&upper->list); 10843fd0a558SYan, Zheng BUG_ON(!list_empty(&upper->upper)); 10853fd0a558SYan, Zheng if (upper == node) 10863fd0a558SYan, Zheng node = NULL; 10873fd0a558SYan, Zheng if (upper->lowest) { 10883fd0a558SYan, Zheng list_del_init(&upper->lower); 10893fd0a558SYan, Zheng upper->lowest = 0; 10903fd0a558SYan, Zheng } 10913fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 10923fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 10933fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 10943fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 10953fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10963fd0a558SYan, Zheng lower = edge->node[LOWER]; 10973fd0a558SYan, Zheng free_backref_edge(cache, edge); 10983fd0a558SYan, Zheng 10993fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11003fd0a558SYan, Zheng list_add(&lower->list, &useless); 11013fd0a558SYan, Zheng } 11023fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11033fd0a558SYan, Zheng if (upper->level > 0) { 11043fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11053fd0a558SYan, Zheng upper->detached = 1; 11063fd0a558SYan, Zheng } else { 11073fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11083fd0a558SYan, Zheng free_backref_node(cache, upper); 11093fd0a558SYan, Zheng } 11103fd0a558SYan, Zheng } 11115d4f98a2SYan Zheng out: 11125d4f98a2SYan Zheng btrfs_free_path(path1); 11135d4f98a2SYan Zheng btrfs_free_path(path2); 11145d4f98a2SYan Zheng if (err) { 11153fd0a558SYan, Zheng while (!list_empty(&useless)) { 11163fd0a558SYan, Zheng lower = list_entry(useless.next, 11173fd0a558SYan, Zheng struct backref_node, upper); 11183fd0a558SYan, Zheng list_del_init(&lower->upper); 11193fd0a558SYan, Zheng } 11205d4f98a2SYan Zheng upper = node; 11213fd0a558SYan, Zheng INIT_LIST_HEAD(&list); 11225d4f98a2SYan Zheng while (upper) { 11235d4f98a2SYan Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 11245d4f98a2SYan Zheng list_splice_tail(&upper->upper, &list); 11253fd0a558SYan, Zheng free_backref_node(cache, upper); 11265d4f98a2SYan Zheng } 11275d4f98a2SYan Zheng 11285d4f98a2SYan Zheng if (list_empty(&list)) 11295d4f98a2SYan Zheng break; 11305d4f98a2SYan Zheng 11315d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, 11325d4f98a2SYan Zheng list[LOWER]); 11333fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11345d4f98a2SYan Zheng upper = edge->node[UPPER]; 11353fd0a558SYan, Zheng free_backref_edge(cache, edge); 11365d4f98a2SYan Zheng } 11375d4f98a2SYan Zheng return ERR_PTR(err); 11385d4f98a2SYan Zheng } 11393fd0a558SYan, Zheng BUG_ON(node && node->detached); 11405d4f98a2SYan Zheng return node; 11415d4f98a2SYan Zheng } 11425d4f98a2SYan Zheng 11435d4f98a2SYan Zheng /* 11443fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11453fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11463fd0a558SYan, Zheng * corresponds to root of source tree 11473fd0a558SYan, Zheng */ 11483fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 11493fd0a558SYan, Zheng struct reloc_control *rc, 11503fd0a558SYan, Zheng struct btrfs_root *src, 11513fd0a558SYan, Zheng struct btrfs_root *dest) 11523fd0a558SYan, Zheng { 11533fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 11543fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 11553fd0a558SYan, Zheng struct backref_node *node = NULL; 11563fd0a558SYan, Zheng struct backref_node *new_node; 11573fd0a558SYan, Zheng struct backref_edge *edge; 11583fd0a558SYan, Zheng struct backref_edge *new_edge; 11593fd0a558SYan, Zheng struct rb_node *rb_node; 11603fd0a558SYan, Zheng 11613fd0a558SYan, Zheng if (cache->last_trans > 0) 11623fd0a558SYan, Zheng update_backref_cache(trans, cache); 11633fd0a558SYan, Zheng 11643fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 11653fd0a558SYan, Zheng if (rb_node) { 11663fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 11673fd0a558SYan, Zheng if (node->detached) 11683fd0a558SYan, Zheng node = NULL; 11693fd0a558SYan, Zheng else 11703fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 11713fd0a558SYan, Zheng } 11723fd0a558SYan, Zheng 11733fd0a558SYan, Zheng if (!node) { 11743fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 11753fd0a558SYan, Zheng reloc_root->commit_root->start); 11763fd0a558SYan, Zheng if (rb_node) { 11773fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 11783fd0a558SYan, Zheng rb_node); 11793fd0a558SYan, Zheng BUG_ON(node->detached); 11803fd0a558SYan, Zheng } 11813fd0a558SYan, Zheng } 11823fd0a558SYan, Zheng 11833fd0a558SYan, Zheng if (!node) 11843fd0a558SYan, Zheng return 0; 11853fd0a558SYan, Zheng 11863fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 11873fd0a558SYan, Zheng if (!new_node) 11883fd0a558SYan, Zheng return -ENOMEM; 11893fd0a558SYan, Zheng 11903fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 11913fd0a558SYan, Zheng new_node->level = node->level; 11923fd0a558SYan, Zheng new_node->lowest = node->lowest; 11936848ad64SYan, Zheng new_node->checked = 1; 11943fd0a558SYan, Zheng new_node->root = dest; 11953fd0a558SYan, Zheng 11963fd0a558SYan, Zheng if (!node->lowest) { 11973fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 11983fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 11993fd0a558SYan, Zheng if (!new_edge) 12003fd0a558SYan, Zheng goto fail; 12013fd0a558SYan, Zheng 12023fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12033fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12043fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12053fd0a558SYan, Zheng &new_node->lower); 12063fd0a558SYan, Zheng } 120776b9e23dSMiao Xie } else { 120876b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12093fd0a558SYan, Zheng } 12103fd0a558SYan, Zheng 12113fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12123fd0a558SYan, Zheng &new_node->rb_node); 121343c04fb1SJeff Mahoney if (rb_node) 121443c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12153fd0a558SYan, Zheng 12163fd0a558SYan, Zheng if (!new_node->lowest) { 12173fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12183fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12193fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12203fd0a558SYan, Zheng } 12213fd0a558SYan, Zheng } 12223fd0a558SYan, Zheng return 0; 12233fd0a558SYan, Zheng fail: 12243fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12253fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12263fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12273fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12283fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12293fd0a558SYan, Zheng } 12303fd0a558SYan, Zheng free_backref_node(cache, new_node); 12313fd0a558SYan, Zheng return -ENOMEM; 12323fd0a558SYan, Zheng } 12333fd0a558SYan, Zheng 12343fd0a558SYan, Zheng /* 12355d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12365d4f98a2SYan Zheng */ 1237ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12385d4f98a2SYan Zheng { 12395d4f98a2SYan Zheng struct rb_node *rb_node; 12405d4f98a2SYan Zheng struct mapping_node *node; 12415d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 12425d4f98a2SYan Zheng 12435d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1244ffd7b339SJeff Mahoney if (!node) 1245ffd7b339SJeff Mahoney return -ENOMEM; 12465d4f98a2SYan Zheng 12475d4f98a2SYan Zheng node->bytenr = root->node->start; 12485d4f98a2SYan Zheng node->data = root; 12495d4f98a2SYan Zheng 12505d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12515d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 12525d4f98a2SYan Zheng node->bytenr, &node->rb_node); 12535d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1254ffd7b339SJeff Mahoney if (rb_node) { 1255ffd7b339SJeff Mahoney btrfs_panic(root->fs_info, -EEXIST, "Duplicate root found " 1256ffd7b339SJeff Mahoney "for start=%llu while inserting into relocation " 1257533574c6SJoe Perches "tree\n", node->bytenr); 125823291a04SDan Carpenter kfree(node); 125923291a04SDan Carpenter return -EEXIST; 1260ffd7b339SJeff Mahoney } 12615d4f98a2SYan Zheng 12625d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 12635d4f98a2SYan Zheng return 0; 12645d4f98a2SYan Zheng } 12655d4f98a2SYan Zheng 12665d4f98a2SYan Zheng /* 12675d4f98a2SYan Zheng * helper to update/delete the 'address of tree root -> reloc tree' 12685d4f98a2SYan Zheng * mapping 12695d4f98a2SYan Zheng */ 12705d4f98a2SYan Zheng static int __update_reloc_root(struct btrfs_root *root, int del) 12715d4f98a2SYan Zheng { 12725d4f98a2SYan Zheng struct rb_node *rb_node; 12735d4f98a2SYan Zheng struct mapping_node *node = NULL; 12745d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 12755d4f98a2SYan Zheng 12765d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12775d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 12785d4f98a2SYan Zheng root->commit_root->start); 12795d4f98a2SYan Zheng if (rb_node) { 12805d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 12815d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 12825d4f98a2SYan Zheng } 12835d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 12845d4f98a2SYan Zheng 12858f71f3e0SLiu Bo if (!node) 12868f71f3e0SLiu Bo return 0; 12875d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 12885d4f98a2SYan Zheng 12895d4f98a2SYan Zheng if (!del) { 12905d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12915d4f98a2SYan Zheng node->bytenr = root->node->start; 12925d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 12935d4f98a2SYan Zheng node->bytenr, &node->rb_node); 12945d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 129543c04fb1SJeff Mahoney if (rb_node) 129643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 12975d4f98a2SYan Zheng } else { 12981daf3540SDaniel J Blueman spin_lock(&root->fs_info->trans_lock); 12995d4f98a2SYan Zheng list_del_init(&root->root_list); 13001daf3540SDaniel J Blueman spin_unlock(&root->fs_info->trans_lock); 13015d4f98a2SYan Zheng kfree(node); 13025d4f98a2SYan Zheng } 13035d4f98a2SYan Zheng return 0; 13045d4f98a2SYan Zheng } 13055d4f98a2SYan Zheng 13063fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13073fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13085d4f98a2SYan Zheng { 13095d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13105d4f98a2SYan Zheng struct extent_buffer *eb; 13115d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13125d4f98a2SYan Zheng struct btrfs_key root_key; 13135bc7247aSMiao Xie u64 last_snap = 0; 13145d4f98a2SYan Zheng int ret; 13155d4f98a2SYan Zheng 13165d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13175d4f98a2SYan Zheng BUG_ON(!root_item); 13185d4f98a2SYan Zheng 13195d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13205d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13213fd0a558SYan, Zheng root_key.offset = objectid; 13225d4f98a2SYan Zheng 13233fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 13243fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 13255d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 13265d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 13275d4f98a2SYan Zheng BUG_ON(ret); 13285d4f98a2SYan Zheng 13295bc7247aSMiao Xie last_snap = btrfs_root_last_snapshot(&root->root_item); 13303fd0a558SYan, Zheng btrfs_set_root_last_snapshot(&root->root_item, 13313fd0a558SYan, Zheng trans->transid - 1); 13323fd0a558SYan, Zheng } else { 13333fd0a558SYan, Zheng /* 13343fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 13353fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 13363fd0a558SYan, Zheng * modified after it was created have RELOC flag 13373fd0a558SYan, Zheng * set in their headers. so it's OK to not update 13383fd0a558SYan, Zheng * the 'last_snapshot'. 13393fd0a558SYan, Zheng */ 13403fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 13413fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 13423fd0a558SYan, Zheng BUG_ON(ret); 13433fd0a558SYan, Zheng } 13443fd0a558SYan, Zheng 13455d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 13465d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 13475d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 13485d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 13493fd0a558SYan, Zheng 13503fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 13513fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 13523fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 13533fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 13545d4f98a2SYan Zheng root_item->drop_level = 0; 13555bc7247aSMiao Xie /* 13565bc7247aSMiao Xie * abuse rtransid, it is safe because it is impossible to 13575bc7247aSMiao Xie * receive data into a relocation tree. 13585bc7247aSMiao Xie */ 13595bc7247aSMiao Xie btrfs_set_root_rtransid(root_item, last_snap); 13605bc7247aSMiao Xie btrfs_set_root_otransid(root_item, trans->transid); 13613fd0a558SYan, Zheng } 13625d4f98a2SYan Zheng 13635d4f98a2SYan Zheng btrfs_tree_unlock(eb); 13645d4f98a2SYan Zheng free_extent_buffer(eb); 13655d4f98a2SYan Zheng 13665d4f98a2SYan Zheng ret = btrfs_insert_root(trans, root->fs_info->tree_root, 13675d4f98a2SYan Zheng &root_key, root_item); 13685d4f98a2SYan Zheng BUG_ON(ret); 13695d4f98a2SYan Zheng kfree(root_item); 13705d4f98a2SYan Zheng 1371cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root->fs_info->tree_root, &root_key); 13725d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 13735d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 13743fd0a558SYan, Zheng return reloc_root; 13753fd0a558SYan, Zheng } 13763fd0a558SYan, Zheng 13773fd0a558SYan, Zheng /* 13783fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 13793fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 13803fd0a558SYan, Zheng */ 13813fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 13823fd0a558SYan, Zheng struct btrfs_root *root) 13833fd0a558SYan, Zheng { 13843fd0a558SYan, Zheng struct btrfs_root *reloc_root; 13853fd0a558SYan, Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 138620dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 13873fd0a558SYan, Zheng int clear_rsv = 0; 1388ffd7b339SJeff Mahoney int ret; 13893fd0a558SYan, Zheng 13903fd0a558SYan, Zheng if (root->reloc_root) { 13913fd0a558SYan, Zheng reloc_root = root->reloc_root; 13923fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 13933fd0a558SYan, Zheng return 0; 13943fd0a558SYan, Zheng } 13953fd0a558SYan, Zheng 13963fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 13973fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 13983fd0a558SYan, Zheng return 0; 13993fd0a558SYan, Zheng 140020dd2cbfSMiao Xie if (!trans->reloc_reserved) { 140120dd2cbfSMiao Xie rsv = trans->block_rsv; 14023fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14033fd0a558SYan, Zheng clear_rsv = 1; 14043fd0a558SYan, Zheng } 14053fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14063fd0a558SYan, Zheng if (clear_rsv) 140720dd2cbfSMiao Xie trans->block_rsv = rsv; 14085d4f98a2SYan Zheng 1409ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1410ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14115d4f98a2SYan Zheng root->reloc_root = reloc_root; 14125d4f98a2SYan Zheng return 0; 14135d4f98a2SYan Zheng } 14145d4f98a2SYan Zheng 14155d4f98a2SYan Zheng /* 14165d4f98a2SYan Zheng * update root item of reloc tree 14175d4f98a2SYan Zheng */ 14185d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 14195d4f98a2SYan Zheng struct btrfs_root *root) 14205d4f98a2SYan Zheng { 14215d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14225d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14235d4f98a2SYan Zheng int del = 0; 14245d4f98a2SYan Zheng int ret; 14255d4f98a2SYan Zheng 14265d4f98a2SYan Zheng if (!root->reloc_root) 14277585717fSChris Mason goto out; 14285d4f98a2SYan Zheng 14295d4f98a2SYan Zheng reloc_root = root->reloc_root; 14305d4f98a2SYan Zheng root_item = &reloc_root->root_item; 14315d4f98a2SYan Zheng 14323fd0a558SYan, Zheng if (root->fs_info->reloc_ctl->merge_reloc_tree && 14333fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 14345d4f98a2SYan Zheng root->reloc_root = NULL; 14355d4f98a2SYan Zheng del = 1; 14365d4f98a2SYan Zheng } 14375d4f98a2SYan Zheng 14385d4f98a2SYan Zheng __update_reloc_root(reloc_root, del); 14395d4f98a2SYan Zheng 14405d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 14415d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 14425d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 14435d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 14445d4f98a2SYan Zheng } 14455d4f98a2SYan Zheng 14465d4f98a2SYan Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 14475d4f98a2SYan Zheng &reloc_root->root_key, root_item); 14485d4f98a2SYan Zheng BUG_ON(ret); 14497585717fSChris Mason 14507585717fSChris Mason out: 14515d4f98a2SYan Zheng return 0; 14525d4f98a2SYan Zheng } 14535d4f98a2SYan Zheng 14545d4f98a2SYan Zheng /* 14555d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 14565d4f98a2SYan Zheng * in a subvolume 14575d4f98a2SYan Zheng */ 14585d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 14595d4f98a2SYan Zheng { 14605d4f98a2SYan Zheng struct rb_node *node; 14615d4f98a2SYan Zheng struct rb_node *prev; 14625d4f98a2SYan Zheng struct btrfs_inode *entry; 14635d4f98a2SYan Zheng struct inode *inode; 14645d4f98a2SYan Zheng 14655d4f98a2SYan Zheng spin_lock(&root->inode_lock); 14665d4f98a2SYan Zheng again: 14675d4f98a2SYan Zheng node = root->inode_tree.rb_node; 14685d4f98a2SYan Zheng prev = NULL; 14695d4f98a2SYan Zheng while (node) { 14705d4f98a2SYan Zheng prev = node; 14715d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 14725d4f98a2SYan Zheng 147333345d01SLi Zefan if (objectid < btrfs_ino(&entry->vfs_inode)) 14745d4f98a2SYan Zheng node = node->rb_left; 147533345d01SLi Zefan else if (objectid > btrfs_ino(&entry->vfs_inode)) 14765d4f98a2SYan Zheng node = node->rb_right; 14775d4f98a2SYan Zheng else 14785d4f98a2SYan Zheng break; 14795d4f98a2SYan Zheng } 14805d4f98a2SYan Zheng if (!node) { 14815d4f98a2SYan Zheng while (prev) { 14825d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 148333345d01SLi Zefan if (objectid <= btrfs_ino(&entry->vfs_inode)) { 14845d4f98a2SYan Zheng node = prev; 14855d4f98a2SYan Zheng break; 14865d4f98a2SYan Zheng } 14875d4f98a2SYan Zheng prev = rb_next(prev); 14885d4f98a2SYan Zheng } 14895d4f98a2SYan Zheng } 14905d4f98a2SYan Zheng while (node) { 14915d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 14925d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 14935d4f98a2SYan Zheng if (inode) { 14945d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 14955d4f98a2SYan Zheng return inode; 14965d4f98a2SYan Zheng } 14975d4f98a2SYan Zheng 149833345d01SLi Zefan objectid = btrfs_ino(&entry->vfs_inode) + 1; 14995d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15005d4f98a2SYan Zheng goto again; 15015d4f98a2SYan Zheng 15025d4f98a2SYan Zheng node = rb_next(node); 15035d4f98a2SYan Zheng } 15045d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15055d4f98a2SYan Zheng return NULL; 15065d4f98a2SYan Zheng } 15075d4f98a2SYan Zheng 15085d4f98a2SYan Zheng static int in_block_group(u64 bytenr, 15095d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group) 15105d4f98a2SYan Zheng { 15115d4f98a2SYan Zheng if (bytenr >= block_group->key.objectid && 15125d4f98a2SYan Zheng bytenr < block_group->key.objectid + block_group->key.offset) 15135d4f98a2SYan Zheng return 1; 15145d4f98a2SYan Zheng return 0; 15155d4f98a2SYan Zheng } 15165d4f98a2SYan Zheng 15175d4f98a2SYan Zheng /* 15185d4f98a2SYan Zheng * get new location of data 15195d4f98a2SYan Zheng */ 15205d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 15215d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 15225d4f98a2SYan Zheng { 15235d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 15245d4f98a2SYan Zheng struct btrfs_path *path; 15255d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15265d4f98a2SYan Zheng struct extent_buffer *leaf; 15275d4f98a2SYan Zheng int ret; 15285d4f98a2SYan Zheng 15295d4f98a2SYan Zheng path = btrfs_alloc_path(); 15305d4f98a2SYan Zheng if (!path) 15315d4f98a2SYan Zheng return -ENOMEM; 15325d4f98a2SYan Zheng 15335d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 153433345d01SLi Zefan ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode), 15355d4f98a2SYan Zheng bytenr, 0); 15365d4f98a2SYan Zheng if (ret < 0) 15375d4f98a2SYan Zheng goto out; 15385d4f98a2SYan Zheng if (ret > 0) { 15395d4f98a2SYan Zheng ret = -ENOENT; 15405d4f98a2SYan Zheng goto out; 15415d4f98a2SYan Zheng } 15425d4f98a2SYan Zheng 15435d4f98a2SYan Zheng leaf = path->nodes[0]; 15445d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 15455d4f98a2SYan Zheng struct btrfs_file_extent_item); 15465d4f98a2SYan Zheng 15475d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 15485d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 15495d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 15505d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 15515d4f98a2SYan Zheng 15525d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 155383d4cfd4SJosef Bacik ret = -EINVAL; 15545d4f98a2SYan Zheng goto out; 15555d4f98a2SYan Zheng } 15565d4f98a2SYan Zheng 15575d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 15585d4f98a2SYan Zheng ret = 0; 15595d4f98a2SYan Zheng out: 15605d4f98a2SYan Zheng btrfs_free_path(path); 15615d4f98a2SYan Zheng return ret; 15625d4f98a2SYan Zheng } 15635d4f98a2SYan Zheng 15645d4f98a2SYan Zheng /* 15655d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 15665d4f98a2SYan Zheng * the new locations. 15675d4f98a2SYan Zheng */ 15683fd0a558SYan, Zheng static noinline_for_stack 15693fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 15705d4f98a2SYan Zheng struct reloc_control *rc, 15715d4f98a2SYan Zheng struct btrfs_root *root, 15723fd0a558SYan, Zheng struct extent_buffer *leaf) 15735d4f98a2SYan Zheng { 15745d4f98a2SYan Zheng struct btrfs_key key; 15755d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15765d4f98a2SYan Zheng struct inode *inode = NULL; 15775d4f98a2SYan Zheng u64 parent; 15785d4f98a2SYan Zheng u64 bytenr; 15793fd0a558SYan, Zheng u64 new_bytenr = 0; 15805d4f98a2SYan Zheng u64 num_bytes; 15815d4f98a2SYan Zheng u64 end; 15825d4f98a2SYan Zheng u32 nritems; 15835d4f98a2SYan Zheng u32 i; 158483d4cfd4SJosef Bacik int ret = 0; 15855d4f98a2SYan Zheng int first = 1; 15865d4f98a2SYan Zheng int dirty = 0; 15875d4f98a2SYan Zheng 15885d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 15895d4f98a2SYan Zheng return 0; 15905d4f98a2SYan Zheng 15915d4f98a2SYan Zheng /* reloc trees always use full backref */ 15925d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 15935d4f98a2SYan Zheng parent = leaf->start; 15945d4f98a2SYan Zheng else 15955d4f98a2SYan Zheng parent = 0; 15965d4f98a2SYan Zheng 15975d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 15985d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 15995d4f98a2SYan Zheng cond_resched(); 16005d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16015d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16025d4f98a2SYan Zheng continue; 16035d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16045d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16055d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16065d4f98a2SYan Zheng continue; 16075d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16085d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16095d4f98a2SYan Zheng if (bytenr == 0) 16105d4f98a2SYan Zheng continue; 16115d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16125d4f98a2SYan Zheng continue; 16135d4f98a2SYan Zheng 16145d4f98a2SYan Zheng /* 16155d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16165d4f98a2SYan Zheng * to complete and drop the extent cache 16175d4f98a2SYan Zheng */ 16185d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16195d4f98a2SYan Zheng if (first) { 16205d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16215d4f98a2SYan Zheng first = 0; 162233345d01SLi Zefan } else if (inode && btrfs_ino(inode) < key.objectid) { 16233fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16245d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16255d4f98a2SYan Zheng } 162633345d01SLi Zefan if (inode && btrfs_ino(inode) == key.objectid) { 16275d4f98a2SYan Zheng end = key.offset + 16285d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 16295d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 16305d4f98a2SYan Zheng root->sectorsize)); 16315d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 16325d4f98a2SYan Zheng end--; 16335d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1634d0082371SJeff Mahoney key.offset, end); 16355d4f98a2SYan Zheng if (!ret) 16365d4f98a2SYan Zheng continue; 16375d4f98a2SYan Zheng 16385d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, key.offset, end, 16395d4f98a2SYan Zheng 1); 16405d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1641d0082371SJeff Mahoney key.offset, end); 16425d4f98a2SYan Zheng } 16435d4f98a2SYan Zheng } 16445d4f98a2SYan Zheng 16455d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 16465d4f98a2SYan Zheng bytenr, num_bytes); 164783d4cfd4SJosef Bacik if (ret) { 164883d4cfd4SJosef Bacik /* 164983d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 165083d4cfd4SJosef Bacik * in the file extent yet. 165183d4cfd4SJosef Bacik */ 165283d4cfd4SJosef Bacik break; 16533fd0a558SYan, Zheng } 16545d4f98a2SYan Zheng 16555d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 16565d4f98a2SYan Zheng dirty = 1; 16575d4f98a2SYan Zheng 16585d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 16595d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, new_bytenr, 16605d4f98a2SYan Zheng num_bytes, parent, 16615d4f98a2SYan Zheng btrfs_header_owner(leaf), 166266d7e7f0SArne Jansen key.objectid, key.offset, 1); 166383d4cfd4SJosef Bacik if (ret) { 166483d4cfd4SJosef Bacik btrfs_abort_transaction(trans, root, ret); 166583d4cfd4SJosef Bacik break; 166683d4cfd4SJosef Bacik } 16675d4f98a2SYan Zheng 16685d4f98a2SYan Zheng ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 16695d4f98a2SYan Zheng parent, btrfs_header_owner(leaf), 167066d7e7f0SArne Jansen key.objectid, key.offset, 1); 167183d4cfd4SJosef Bacik if (ret) { 167283d4cfd4SJosef Bacik btrfs_abort_transaction(trans, root, ret); 167383d4cfd4SJosef Bacik break; 167483d4cfd4SJosef Bacik } 16755d4f98a2SYan Zheng } 16765d4f98a2SYan Zheng if (dirty) 16775d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 16783fd0a558SYan, Zheng if (inode) 16793fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 168083d4cfd4SJosef Bacik return ret; 16815d4f98a2SYan Zheng } 16825d4f98a2SYan Zheng 16835d4f98a2SYan Zheng static noinline_for_stack 16845d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 16855d4f98a2SYan Zheng struct btrfs_path *path, int level) 16865d4f98a2SYan Zheng { 16875d4f98a2SYan Zheng struct btrfs_disk_key key1; 16885d4f98a2SYan Zheng struct btrfs_disk_key key2; 16895d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 16905d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 16915d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 16925d4f98a2SYan Zheng } 16935d4f98a2SYan Zheng 16945d4f98a2SYan Zheng /* 16955d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 16965d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 16975d4f98a2SYan Zheng * reloc tree was create can be replaced. 16985d4f98a2SYan Zheng * 16995d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17005d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17015d4f98a2SYan Zheng * errors, a negative error number is returned. 17025d4f98a2SYan Zheng */ 17033fd0a558SYan, Zheng static noinline_for_stack 17043fd0a558SYan, Zheng int replace_path(struct btrfs_trans_handle *trans, 17055d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17065d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17075d4f98a2SYan Zheng int lowest_level, int max_level) 17085d4f98a2SYan Zheng { 17095d4f98a2SYan Zheng struct extent_buffer *eb; 17105d4f98a2SYan Zheng struct extent_buffer *parent; 17115d4f98a2SYan Zheng struct btrfs_key key; 17125d4f98a2SYan Zheng u64 old_bytenr; 17135d4f98a2SYan Zheng u64 new_bytenr; 17145d4f98a2SYan Zheng u64 old_ptr_gen; 17155d4f98a2SYan Zheng u64 new_ptr_gen; 17165d4f98a2SYan Zheng u64 last_snapshot; 17175d4f98a2SYan Zheng u32 blocksize; 17183fd0a558SYan, Zheng int cow = 0; 17195d4f98a2SYan Zheng int level; 17205d4f98a2SYan Zheng int ret; 17215d4f98a2SYan Zheng int slot; 17225d4f98a2SYan Zheng 17235d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17245d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17255d4f98a2SYan Zheng 17265d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 17273fd0a558SYan, Zheng again: 17285d4f98a2SYan Zheng slot = path->slots[lowest_level]; 17295d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 17305d4f98a2SYan Zheng 17315d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 17325d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 17335d4f98a2SYan Zheng level = btrfs_header_level(eb); 17345d4f98a2SYan Zheng 17355d4f98a2SYan Zheng if (level < lowest_level) { 17365d4f98a2SYan Zheng btrfs_tree_unlock(eb); 17375d4f98a2SYan Zheng free_extent_buffer(eb); 17385d4f98a2SYan Zheng return 0; 17395d4f98a2SYan Zheng } 17405d4f98a2SYan Zheng 17413fd0a558SYan, Zheng if (cow) { 17425d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 17435d4f98a2SYan Zheng BUG_ON(ret); 17443fd0a558SYan, Zheng } 17455d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 17465d4f98a2SYan Zheng 17475d4f98a2SYan Zheng if (next_key) { 17485d4f98a2SYan Zheng next_key->objectid = (u64)-1; 17495d4f98a2SYan Zheng next_key->type = (u8)-1; 17505d4f98a2SYan Zheng next_key->offset = (u64)-1; 17515d4f98a2SYan Zheng } 17525d4f98a2SYan Zheng 17535d4f98a2SYan Zheng parent = eb; 17545d4f98a2SYan Zheng while (1) { 17555d4f98a2SYan Zheng level = btrfs_header_level(parent); 17565d4f98a2SYan Zheng BUG_ON(level < lowest_level); 17575d4f98a2SYan Zheng 17585d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 17595d4f98a2SYan Zheng if (ret && slot > 0) 17605d4f98a2SYan Zheng slot--; 17615d4f98a2SYan Zheng 17625d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 17635d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 17645d4f98a2SYan Zheng 17655d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 17665d4f98a2SYan Zheng blocksize = btrfs_level_size(dest, level - 1); 17675d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 17685d4f98a2SYan Zheng 17695d4f98a2SYan Zheng if (level <= max_level) { 17705d4f98a2SYan Zheng eb = path->nodes[level]; 17715d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 17725d4f98a2SYan Zheng path->slots[level]); 17735d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 17745d4f98a2SYan Zheng path->slots[level]); 17755d4f98a2SYan Zheng } else { 17765d4f98a2SYan Zheng new_bytenr = 0; 17775d4f98a2SYan Zheng new_ptr_gen = 0; 17785d4f98a2SYan Zheng } 17795d4f98a2SYan Zheng 17805d4f98a2SYan Zheng if (new_bytenr > 0 && new_bytenr == old_bytenr) { 17815d4f98a2SYan Zheng WARN_ON(1); 17825d4f98a2SYan Zheng ret = level; 17835d4f98a2SYan Zheng break; 17845d4f98a2SYan Zheng } 17855d4f98a2SYan Zheng 17865d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 17875d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 17883fd0a558SYan, Zheng if (level <= lowest_level) { 17895d4f98a2SYan Zheng ret = 0; 17905d4f98a2SYan Zheng break; 17915d4f98a2SYan Zheng } 17925d4f98a2SYan Zheng 17935d4f98a2SYan Zheng eb = read_tree_block(dest, old_bytenr, blocksize, 17945d4f98a2SYan Zheng old_ptr_gen); 1795416bc658SJosef Bacik if (!eb || !extent_buffer_uptodate(eb)) { 1796416bc658SJosef Bacik ret = (!eb) ? -ENOMEM : -EIO; 1797416bc658SJosef Bacik free_extent_buffer(eb); 1798379cde74SStefan Behrens break; 1799416bc658SJosef Bacik } 18005d4f98a2SYan Zheng btrfs_tree_lock(eb); 18013fd0a558SYan, Zheng if (cow) { 18025d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18035d4f98a2SYan Zheng slot, &eb); 18045d4f98a2SYan Zheng BUG_ON(ret); 18055d4f98a2SYan Zheng } 18063fd0a558SYan, Zheng btrfs_set_lock_blocking(eb); 18075d4f98a2SYan Zheng 18085d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18095d4f98a2SYan Zheng free_extent_buffer(parent); 18105d4f98a2SYan Zheng 18115d4f98a2SYan Zheng parent = eb; 18125d4f98a2SYan Zheng continue; 18135d4f98a2SYan Zheng } 18145d4f98a2SYan Zheng 18153fd0a558SYan, Zheng if (!cow) { 18163fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18173fd0a558SYan, Zheng free_extent_buffer(parent); 18183fd0a558SYan, Zheng cow = 1; 18193fd0a558SYan, Zheng goto again; 18203fd0a558SYan, Zheng } 18213fd0a558SYan, Zheng 18225d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18235d4f98a2SYan Zheng path->slots[level]); 1824b3b4aa74SDavid Sterba btrfs_release_path(path); 18255d4f98a2SYan Zheng 18265d4f98a2SYan Zheng path->lowest_level = level; 18275d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 18285d4f98a2SYan Zheng path->lowest_level = 0; 18295d4f98a2SYan Zheng BUG_ON(ret); 18305d4f98a2SYan Zheng 18315d4f98a2SYan Zheng /* 18325d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 18335d4f98a2SYan Zheng */ 18345d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 18355d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 18365d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 18375d4f98a2SYan Zheng 18385d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 18395d4f98a2SYan Zheng path->slots[level], old_bytenr); 18405d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 18415d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 18425d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 18435d4f98a2SYan Zheng 18445d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize, 18455d4f98a2SYan Zheng path->nodes[level]->start, 184666d7e7f0SArne Jansen src->root_key.objectid, level - 1, 0, 184766d7e7f0SArne Jansen 1); 18485d4f98a2SYan Zheng BUG_ON(ret); 18495d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize, 18505d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 185166d7e7f0SArne Jansen 0, 1); 18525d4f98a2SYan Zheng BUG_ON(ret); 18535d4f98a2SYan Zheng 18545d4f98a2SYan Zheng ret = btrfs_free_extent(trans, src, new_bytenr, blocksize, 18555d4f98a2SYan Zheng path->nodes[level]->start, 185666d7e7f0SArne Jansen src->root_key.objectid, level - 1, 0, 185766d7e7f0SArne Jansen 1); 18585d4f98a2SYan Zheng BUG_ON(ret); 18595d4f98a2SYan Zheng 18605d4f98a2SYan Zheng ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize, 18615d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 186266d7e7f0SArne Jansen 0, 1); 18635d4f98a2SYan Zheng BUG_ON(ret); 18645d4f98a2SYan Zheng 18655d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 18665d4f98a2SYan Zheng 18675d4f98a2SYan Zheng ret = level; 18685d4f98a2SYan Zheng break; 18695d4f98a2SYan Zheng } 18705d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18715d4f98a2SYan Zheng free_extent_buffer(parent); 18725d4f98a2SYan Zheng return ret; 18735d4f98a2SYan Zheng } 18745d4f98a2SYan Zheng 18755d4f98a2SYan Zheng /* 18765d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 18775d4f98a2SYan Zheng */ 18785d4f98a2SYan Zheng static noinline_for_stack 18795d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 18805d4f98a2SYan Zheng int *level) 18815d4f98a2SYan Zheng { 18825d4f98a2SYan Zheng struct extent_buffer *eb; 18835d4f98a2SYan Zheng int i; 18845d4f98a2SYan Zheng u64 last_snapshot; 18855d4f98a2SYan Zheng u32 nritems; 18865d4f98a2SYan Zheng 18875d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 18885d4f98a2SYan Zheng 18895d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 18905d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 18915d4f98a2SYan Zheng path->nodes[i] = NULL; 18925d4f98a2SYan Zheng } 18935d4f98a2SYan Zheng 18945d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 18955d4f98a2SYan Zheng eb = path->nodes[i]; 18965d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 18975d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 18985d4f98a2SYan Zheng path->slots[i]++; 18995d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 19005d4f98a2SYan Zheng last_snapshot) 19015d4f98a2SYan Zheng continue; 19025d4f98a2SYan Zheng 19035d4f98a2SYan Zheng *level = i; 19045d4f98a2SYan Zheng return 0; 19055d4f98a2SYan Zheng } 19065d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19075d4f98a2SYan Zheng path->nodes[i] = NULL; 19085d4f98a2SYan Zheng } 19095d4f98a2SYan Zheng return 1; 19105d4f98a2SYan Zheng } 19115d4f98a2SYan Zheng 19125d4f98a2SYan Zheng /* 19135d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 19145d4f98a2SYan Zheng */ 19155d4f98a2SYan Zheng static noinline_for_stack 19165d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19175d4f98a2SYan Zheng int *level) 19185d4f98a2SYan Zheng { 19195d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 19205d4f98a2SYan Zheng int i; 19215d4f98a2SYan Zheng u64 bytenr; 19225d4f98a2SYan Zheng u64 ptr_gen = 0; 19235d4f98a2SYan Zheng u64 last_snapshot; 19245d4f98a2SYan Zheng u32 blocksize; 19255d4f98a2SYan Zheng u32 nritems; 19265d4f98a2SYan Zheng 19275d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19285d4f98a2SYan Zheng 19295d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 19305d4f98a2SYan Zheng eb = path->nodes[i]; 19315d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19325d4f98a2SYan Zheng while (path->slots[i] < nritems) { 19335d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 19345d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 19355d4f98a2SYan Zheng break; 19365d4f98a2SYan Zheng path->slots[i]++; 19375d4f98a2SYan Zheng } 19385d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 19395d4f98a2SYan Zheng if (i == *level) 19405d4f98a2SYan Zheng break; 19415d4f98a2SYan Zheng *level = i + 1; 19425d4f98a2SYan Zheng return 0; 19435d4f98a2SYan Zheng } 19445d4f98a2SYan Zheng if (i == 1) { 19455d4f98a2SYan Zheng *level = i; 19465d4f98a2SYan Zheng return 0; 19475d4f98a2SYan Zheng } 19485d4f98a2SYan Zheng 19495d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 19505d4f98a2SYan Zheng blocksize = btrfs_level_size(root, i - 1); 19515d4f98a2SYan Zheng eb = read_tree_block(root, bytenr, blocksize, ptr_gen); 1952416bc658SJosef Bacik if (!eb || !extent_buffer_uptodate(eb)) { 1953416bc658SJosef Bacik free_extent_buffer(eb); 1954416bc658SJosef Bacik return -EIO; 1955416bc658SJosef Bacik } 19565d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 19575d4f98a2SYan Zheng path->nodes[i - 1] = eb; 19585d4f98a2SYan Zheng path->slots[i - 1] = 0; 19595d4f98a2SYan Zheng } 19605d4f98a2SYan Zheng return 1; 19615d4f98a2SYan Zheng } 19625d4f98a2SYan Zheng 19635d4f98a2SYan Zheng /* 19645d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 19655d4f98a2SYan Zheng * [min_key, max_key) 19665d4f98a2SYan Zheng */ 19675d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 19685d4f98a2SYan Zheng struct btrfs_key *min_key, 19695d4f98a2SYan Zheng struct btrfs_key *max_key) 19705d4f98a2SYan Zheng { 19715d4f98a2SYan Zheng struct inode *inode = NULL; 19725d4f98a2SYan Zheng u64 objectid; 19735d4f98a2SYan Zheng u64 start, end; 197433345d01SLi Zefan u64 ino; 19755d4f98a2SYan Zheng 19765d4f98a2SYan Zheng objectid = min_key->objectid; 19775d4f98a2SYan Zheng while (1) { 19785d4f98a2SYan Zheng cond_resched(); 19795d4f98a2SYan Zheng iput(inode); 19805d4f98a2SYan Zheng 19815d4f98a2SYan Zheng if (objectid > max_key->objectid) 19825d4f98a2SYan Zheng break; 19835d4f98a2SYan Zheng 19845d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 19855d4f98a2SYan Zheng if (!inode) 19865d4f98a2SYan Zheng break; 198733345d01SLi Zefan ino = btrfs_ino(inode); 19885d4f98a2SYan Zheng 198933345d01SLi Zefan if (ino > max_key->objectid) { 19905d4f98a2SYan Zheng iput(inode); 19915d4f98a2SYan Zheng break; 19925d4f98a2SYan Zheng } 19935d4f98a2SYan Zheng 199433345d01SLi Zefan objectid = ino + 1; 19955d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 19965d4f98a2SYan Zheng continue; 19975d4f98a2SYan Zheng 199833345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 19995d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 20005d4f98a2SYan Zheng continue; 20015d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 20025d4f98a2SYan Zheng start = 0; 20035d4f98a2SYan Zheng else { 20045d4f98a2SYan Zheng start = min_key->offset; 20055d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(start, root->sectorsize)); 20065d4f98a2SYan Zheng } 20075d4f98a2SYan Zheng } else { 20085d4f98a2SYan Zheng start = 0; 20095d4f98a2SYan Zheng } 20105d4f98a2SYan Zheng 201133345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 20125d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 20135d4f98a2SYan Zheng continue; 20145d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 20155d4f98a2SYan Zheng end = (u64)-1; 20165d4f98a2SYan Zheng } else { 20175d4f98a2SYan Zheng if (max_key->offset == 0) 20185d4f98a2SYan Zheng continue; 20195d4f98a2SYan Zheng end = max_key->offset; 20205d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 20215d4f98a2SYan Zheng end--; 20225d4f98a2SYan Zheng } 20235d4f98a2SYan Zheng } else { 20245d4f98a2SYan Zheng end = (u64)-1; 20255d4f98a2SYan Zheng } 20265d4f98a2SYan Zheng 20275d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2028d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 20295d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 1); 2030d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 20315d4f98a2SYan Zheng } 20325d4f98a2SYan Zheng return 0; 20335d4f98a2SYan Zheng } 20345d4f98a2SYan Zheng 20355d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 20365d4f98a2SYan Zheng struct btrfs_key *key) 20375d4f98a2SYan Zheng 20385d4f98a2SYan Zheng { 20395d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 20405d4f98a2SYan Zheng if (!path->nodes[level]) 20415d4f98a2SYan Zheng break; 20425d4f98a2SYan Zheng if (path->slots[level] + 1 < 20435d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 20445d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 20455d4f98a2SYan Zheng path->slots[level] + 1); 20465d4f98a2SYan Zheng return 0; 20475d4f98a2SYan Zheng } 20485d4f98a2SYan Zheng level++; 20495d4f98a2SYan Zheng } 20505d4f98a2SYan Zheng return 1; 20515d4f98a2SYan Zheng } 20525d4f98a2SYan Zheng 20535d4f98a2SYan Zheng /* 20545d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 20555d4f98a2SYan Zheng * fs tree. 20565d4f98a2SYan Zheng */ 20575d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 20585d4f98a2SYan Zheng struct btrfs_root *root) 20595d4f98a2SYan Zheng { 20605d4f98a2SYan Zheng LIST_HEAD(inode_list); 20615d4f98a2SYan Zheng struct btrfs_key key; 20625d4f98a2SYan Zheng struct btrfs_key next_key; 20639e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 20645d4f98a2SYan Zheng struct btrfs_root *reloc_root; 20655d4f98a2SYan Zheng struct btrfs_root_item *root_item; 20665d4f98a2SYan Zheng struct btrfs_path *path; 20673fd0a558SYan, Zheng struct extent_buffer *leaf; 20685d4f98a2SYan Zheng int level; 20695d4f98a2SYan Zheng int max_level; 20705d4f98a2SYan Zheng int replaced = 0; 20715d4f98a2SYan Zheng int ret; 20725d4f98a2SYan Zheng int err = 0; 20733fd0a558SYan, Zheng u32 min_reserved; 20745d4f98a2SYan Zheng 20755d4f98a2SYan Zheng path = btrfs_alloc_path(); 20765d4f98a2SYan Zheng if (!path) 20775d4f98a2SYan Zheng return -ENOMEM; 2078026fd317SJosef Bacik path->reada = 1; 20795d4f98a2SYan Zheng 20805d4f98a2SYan Zheng reloc_root = root->reloc_root; 20815d4f98a2SYan Zheng root_item = &reloc_root->root_item; 20825d4f98a2SYan Zheng 20835d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 20845d4f98a2SYan Zheng level = btrfs_root_level(root_item); 20855d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 20865d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 20875d4f98a2SYan Zheng path->slots[level] = 0; 20885d4f98a2SYan Zheng } else { 20895d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 20905d4f98a2SYan Zheng 20915d4f98a2SYan Zheng level = root_item->drop_level; 20925d4f98a2SYan Zheng BUG_ON(level == 0); 20935d4f98a2SYan Zheng path->lowest_level = level; 20945d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 209533c66f43SYan Zheng path->lowest_level = 0; 20965d4f98a2SYan Zheng if (ret < 0) { 20975d4f98a2SYan Zheng btrfs_free_path(path); 20985d4f98a2SYan Zheng return ret; 20995d4f98a2SYan Zheng } 21005d4f98a2SYan Zheng 21015d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 21025d4f98a2SYan Zheng path->slots[level]); 21035d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 21045d4f98a2SYan Zheng 21055d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 21065d4f98a2SYan Zheng } 21075d4f98a2SYan Zheng 21083fd0a558SYan, Zheng min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 21095d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 21105d4f98a2SYan Zheng 21115d4f98a2SYan Zheng while (1) { 211208e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 211308e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 21143fd0a558SYan, Zheng if (ret) { 21159e6a0c52SJosef Bacik err = ret; 21169e6a0c52SJosef Bacik goto out; 21173fd0a558SYan, Zheng } 21189e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 21199e6a0c52SJosef Bacik if (IS_ERR(trans)) { 21209e6a0c52SJosef Bacik err = PTR_ERR(trans); 21219e6a0c52SJosef Bacik trans = NULL; 21229e6a0c52SJosef Bacik goto out; 21239e6a0c52SJosef Bacik } 21249e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 21253fd0a558SYan, Zheng 21263fd0a558SYan, Zheng replaced = 0; 21275d4f98a2SYan Zheng max_level = level; 21285d4f98a2SYan Zheng 21295d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 21305d4f98a2SYan Zheng if (ret < 0) { 21315d4f98a2SYan Zheng err = ret; 21325d4f98a2SYan Zheng goto out; 21335d4f98a2SYan Zheng } 21345d4f98a2SYan Zheng if (ret > 0) 21355d4f98a2SYan Zheng break; 21365d4f98a2SYan Zheng 21375d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 21385d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 21395d4f98a2SYan Zheng ret = 0; 21405d4f98a2SYan Zheng } else { 21413fd0a558SYan, Zheng ret = replace_path(trans, root, reloc_root, path, 21423fd0a558SYan, Zheng &next_key, level, max_level); 21435d4f98a2SYan Zheng } 21445d4f98a2SYan Zheng if (ret < 0) { 21455d4f98a2SYan Zheng err = ret; 21465d4f98a2SYan Zheng goto out; 21475d4f98a2SYan Zheng } 21485d4f98a2SYan Zheng 21495d4f98a2SYan Zheng if (ret > 0) { 21505d4f98a2SYan Zheng level = ret; 21515d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 21525d4f98a2SYan Zheng path->slots[level]); 21535d4f98a2SYan Zheng replaced = 1; 21545d4f98a2SYan Zheng } 21555d4f98a2SYan Zheng 21565d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 21575d4f98a2SYan Zheng if (ret > 0) 21585d4f98a2SYan Zheng break; 21595d4f98a2SYan Zheng 21605d4f98a2SYan Zheng BUG_ON(level == 0); 21615d4f98a2SYan Zheng /* 21625d4f98a2SYan Zheng * save the merging progress in the drop_progress. 21635d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 21645d4f98a2SYan Zheng */ 21655d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 21665d4f98a2SYan Zheng path->slots[level]); 21675d4f98a2SYan Zheng root_item->drop_level = level; 21685d4f98a2SYan Zheng 21693fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 21709e6a0c52SJosef Bacik trans = NULL; 21715d4f98a2SYan Zheng 2172b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 21735d4f98a2SYan Zheng 21745d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 21755d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 21765d4f98a2SYan Zheng } 21775d4f98a2SYan Zheng 21785d4f98a2SYan Zheng /* 21795d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 21805d4f98a2SYan Zheng * relocated and the block is tree root. 21815d4f98a2SYan Zheng */ 21825d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 21835d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 21845d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 21855d4f98a2SYan Zheng free_extent_buffer(leaf); 21865d4f98a2SYan Zheng if (ret < 0) 21875d4f98a2SYan Zheng err = ret; 21885d4f98a2SYan Zheng out: 21895d4f98a2SYan Zheng btrfs_free_path(path); 21905d4f98a2SYan Zheng 21915d4f98a2SYan Zheng if (err == 0) { 21925d4f98a2SYan Zheng memset(&root_item->drop_progress, 0, 21935d4f98a2SYan Zheng sizeof(root_item->drop_progress)); 21945d4f98a2SYan Zheng root_item->drop_level = 0; 21955d4f98a2SYan Zheng btrfs_set_root_refs(root_item, 0); 21963fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 21975d4f98a2SYan Zheng } 21985d4f98a2SYan Zheng 21999e6a0c52SJosef Bacik if (trans) 22003fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 22015d4f98a2SYan Zheng 2202b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 22035d4f98a2SYan Zheng 22045d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22055d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22065d4f98a2SYan Zheng 22075d4f98a2SYan Zheng return err; 22085d4f98a2SYan Zheng } 22095d4f98a2SYan Zheng 22103fd0a558SYan, Zheng static noinline_for_stack 22113fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 22125d4f98a2SYan Zheng { 22133fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 22143fd0a558SYan, Zheng struct btrfs_root *reloc_root; 22155d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 22163fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 22173fd0a558SYan, Zheng u64 num_bytes = 0; 22183fd0a558SYan, Zheng int ret; 22193fd0a558SYan, Zheng 22207585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 22213fd0a558SYan, Zheng rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22223fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 22237585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 22247585717fSChris Mason 22253fd0a558SYan, Zheng again: 22263fd0a558SYan, Zheng if (!err) { 22273fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 222808e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 222908e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22303fd0a558SYan, Zheng if (ret) 22313fd0a558SYan, Zheng err = ret; 22323fd0a558SYan, Zheng } 22333fd0a558SYan, Zheng 22347a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 22353612b495STsutomu Itoh if (IS_ERR(trans)) { 22363612b495STsutomu Itoh if (!err) 22373612b495STsutomu Itoh btrfs_block_rsv_release(rc->extent_root, 22383612b495STsutomu Itoh rc->block_rsv, num_bytes); 22393612b495STsutomu Itoh return PTR_ERR(trans); 22403612b495STsutomu Itoh } 22413fd0a558SYan, Zheng 22423fd0a558SYan, Zheng if (!err) { 22433fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 22443fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 22453fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, 22463fd0a558SYan, Zheng rc->block_rsv, num_bytes); 22473fd0a558SYan, Zheng goto again; 22483fd0a558SYan, Zheng } 22493fd0a558SYan, Zheng } 22503fd0a558SYan, Zheng 22513fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 22523fd0a558SYan, Zheng 22533fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 22543fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 22553fd0a558SYan, Zheng struct btrfs_root, root_list); 22563fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 22573fd0a558SYan, Zheng 22583fd0a558SYan, Zheng root = read_fs_root(reloc_root->fs_info, 22593fd0a558SYan, Zheng reloc_root->root_key.offset); 22603fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 22613fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 22623fd0a558SYan, Zheng 22633fd0a558SYan, Zheng /* 22643fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 22653fd0a558SYan, Zheng * knows it should resumes merging 22663fd0a558SYan, Zheng */ 22673fd0a558SYan, Zheng if (!err) 22683fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 22693fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 22703fd0a558SYan, Zheng 22713fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 22723fd0a558SYan, Zheng } 22733fd0a558SYan, Zheng 22743fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 22753fd0a558SYan, Zheng 22763fd0a558SYan, Zheng if (!err) 22773fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 22783fd0a558SYan, Zheng else 22793fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 22803fd0a558SYan, Zheng return err; 22813fd0a558SYan, Zheng } 22823fd0a558SYan, Zheng 22833fd0a558SYan, Zheng static noinline_for_stack 2284aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2285aca1bba6SLiu Bo { 2286aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2287aca1bba6SLiu Bo 2288aca1bba6SLiu Bo while (!list_empty(list)) { 2289aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2290aca1bba6SLiu Bo root_list); 2291aca1bba6SLiu Bo __update_reloc_root(reloc_root, 1); 2292aca1bba6SLiu Bo free_extent_buffer(reloc_root->node); 2293aca1bba6SLiu Bo free_extent_buffer(reloc_root->commit_root); 2294aca1bba6SLiu Bo kfree(reloc_root); 2295aca1bba6SLiu Bo } 2296aca1bba6SLiu Bo } 2297aca1bba6SLiu Bo 2298aca1bba6SLiu Bo static noinline_for_stack 22993fd0a558SYan, Zheng int merge_reloc_roots(struct reloc_control *rc) 23003fd0a558SYan, Zheng { 23015bc7247aSMiao Xie struct btrfs_trans_handle *trans; 23025d4f98a2SYan Zheng struct btrfs_root *root; 23035d4f98a2SYan Zheng struct btrfs_root *reloc_root; 23045bc7247aSMiao Xie u64 last_snap; 23055bc7247aSMiao Xie u64 otransid; 23065bc7247aSMiao Xie u64 objectid; 23073fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23083fd0a558SYan, Zheng int found = 0; 2309aca1bba6SLiu Bo int ret = 0; 23103fd0a558SYan, Zheng again: 23113fd0a558SYan, Zheng root = rc->extent_root; 23127585717fSChris Mason 23137585717fSChris Mason /* 23147585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 23157585717fSChris Mason * we have to make sure nobody is in the middle of 23167585717fSChris Mason * adding their roots to the list while we are 23177585717fSChris Mason * doing this splice 23187585717fSChris Mason */ 23197585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 23203fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 23217585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 23225d4f98a2SYan Zheng 23233fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 23243fd0a558SYan, Zheng found = 1; 23253fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 23263fd0a558SYan, Zheng struct btrfs_root, root_list); 23275d4f98a2SYan Zheng 23285d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 23295d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, 23305d4f98a2SYan Zheng reloc_root->root_key.offset); 23315d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 23325d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 23335d4f98a2SYan Zheng 23343fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2335b37b39cdSJosef Bacik if (ret) { 2336b37b39cdSJosef Bacik __update_reloc_root(reloc_root, 1); 2337b37b39cdSJosef Bacik free_extent_buffer(reloc_root->node); 2338b37b39cdSJosef Bacik free_extent_buffer(reloc_root->commit_root); 2339b37b39cdSJosef Bacik kfree(reloc_root); 2340aca1bba6SLiu Bo goto out; 2341b37b39cdSJosef Bacik } 23423fd0a558SYan, Zheng } else { 23433fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 23443fd0a558SYan, Zheng } 23455bc7247aSMiao Xie 23465bc7247aSMiao Xie /* 23475bc7247aSMiao Xie * we keep the old last snapshod transid in rtranid when we 23485bc7247aSMiao Xie * created the relocation tree. 23495bc7247aSMiao Xie */ 23505bc7247aSMiao Xie last_snap = btrfs_root_rtransid(&reloc_root->root_item); 23515bc7247aSMiao Xie otransid = btrfs_root_otransid(&reloc_root->root_item); 23525bc7247aSMiao Xie objectid = reloc_root->root_key.offset; 23535bc7247aSMiao Xie 23542c536799SJeff Mahoney ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1); 2355aca1bba6SLiu Bo if (ret < 0) { 2356aca1bba6SLiu Bo if (list_empty(&reloc_root->root_list)) 2357aca1bba6SLiu Bo list_add_tail(&reloc_root->root_list, 2358aca1bba6SLiu Bo &reloc_roots); 2359aca1bba6SLiu Bo goto out; 23605bc7247aSMiao Xie } else if (!ret) { 23615bc7247aSMiao Xie /* 23625bc7247aSMiao Xie * recover the last snapshot tranid to avoid 23635bc7247aSMiao Xie * the space balance break NOCOW. 23645bc7247aSMiao Xie */ 23655bc7247aSMiao Xie root = read_fs_root(rc->extent_root->fs_info, 23665bc7247aSMiao Xie objectid); 23675bc7247aSMiao Xie if (IS_ERR(root)) 23685bc7247aSMiao Xie continue; 23695bc7247aSMiao Xie 23705bc7247aSMiao Xie trans = btrfs_join_transaction(root); 23715bc7247aSMiao Xie BUG_ON(IS_ERR(trans)); 23725bc7247aSMiao Xie 23735bc7247aSMiao Xie /* Check if the fs/file tree was snapshoted or not. */ 23745bc7247aSMiao Xie if (btrfs_root_last_snapshot(&root->root_item) == 23755bc7247aSMiao Xie otransid - 1) 23765bc7247aSMiao Xie btrfs_set_root_last_snapshot(&root->root_item, 23775bc7247aSMiao Xie last_snap); 23785bc7247aSMiao Xie 23795bc7247aSMiao Xie btrfs_end_transaction(trans, root); 2380aca1bba6SLiu Bo } 23815d4f98a2SYan Zheng } 23825d4f98a2SYan Zheng 23833fd0a558SYan, Zheng if (found) { 23843fd0a558SYan, Zheng found = 0; 23853fd0a558SYan, Zheng goto again; 23865d4f98a2SYan Zheng } 2387aca1bba6SLiu Bo out: 2388aca1bba6SLiu Bo if (ret) { 2389aca1bba6SLiu Bo btrfs_std_error(root->fs_info, ret); 2390aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2391aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2392aca1bba6SLiu Bo } 2393aca1bba6SLiu Bo 23945d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 2395aca1bba6SLiu Bo return ret; 23965d4f98a2SYan Zheng } 23975d4f98a2SYan Zheng 23985d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 23995d4f98a2SYan Zheng { 24005d4f98a2SYan Zheng struct tree_block *block; 24015d4f98a2SYan Zheng struct rb_node *rb_node; 24025d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 24035d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 24045d4f98a2SYan Zheng rb_erase(rb_node, blocks); 24055d4f98a2SYan Zheng kfree(block); 24065d4f98a2SYan Zheng } 24075d4f98a2SYan Zheng } 24085d4f98a2SYan Zheng 24095d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 24105d4f98a2SYan Zheng struct btrfs_root *reloc_root) 24115d4f98a2SYan Zheng { 24125d4f98a2SYan Zheng struct btrfs_root *root; 24135d4f98a2SYan Zheng 24145d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 24155d4f98a2SYan Zheng return 0; 24165d4f98a2SYan Zheng 24175d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset); 24185d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24195d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24205d4f98a2SYan Zheng 24215d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 24225d4f98a2SYan Zheng } 24235d4f98a2SYan Zheng 24243fd0a558SYan, Zheng static noinline_for_stack 24253fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 24263fd0a558SYan, Zheng struct reloc_control *rc, 24275d4f98a2SYan Zheng struct backref_node *node, 24283fd0a558SYan, Zheng struct backref_edge *edges[], int *nr) 24295d4f98a2SYan Zheng { 24305d4f98a2SYan Zheng struct backref_node *next; 24315d4f98a2SYan Zheng struct btrfs_root *root; 24323fd0a558SYan, Zheng int index = 0; 24333fd0a558SYan, Zheng 24345d4f98a2SYan Zheng next = node; 24355d4f98a2SYan Zheng while (1) { 24365d4f98a2SYan Zheng cond_resched(); 24375d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 24385d4f98a2SYan Zheng root = next->root; 24393fd0a558SYan, Zheng BUG_ON(!root); 24403fd0a558SYan, Zheng BUG_ON(!root->ref_cows); 24415d4f98a2SYan Zheng 24425d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 24435d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 24445d4f98a2SYan Zheng break; 24455d4f98a2SYan Zheng } 24465d4f98a2SYan Zheng 24475d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 24483fd0a558SYan, Zheng root = root->reloc_root; 24493fd0a558SYan, Zheng 24503fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 24513fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 24523fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 24533fd0a558SYan, Zheng next->new_bytenr = root->node->start; 24543fd0a558SYan, Zheng next->root = root; 24553fd0a558SYan, Zheng list_add_tail(&next->list, 24563fd0a558SYan, Zheng &rc->backref_cache.changed); 24573fd0a558SYan, Zheng __mark_block_processed(rc, next); 24585d4f98a2SYan Zheng break; 24595d4f98a2SYan Zheng } 24605d4f98a2SYan Zheng 24613fd0a558SYan, Zheng WARN_ON(1); 24625d4f98a2SYan Zheng root = NULL; 24635d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 24645d4f98a2SYan Zheng if (!next || next->level <= node->level) 24655d4f98a2SYan Zheng break; 24665d4f98a2SYan Zheng } 24673fd0a558SYan, Zheng if (!root) 24683fd0a558SYan, Zheng return NULL; 24695d4f98a2SYan Zheng 24705d4f98a2SYan Zheng *nr = index; 24713fd0a558SYan, Zheng next = node; 24723fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 24733fd0a558SYan, Zheng while (1) { 24743fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 24753fd0a558SYan, Zheng if (--index < 0) 24763fd0a558SYan, Zheng break; 24773fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 24783fd0a558SYan, Zheng } 24795d4f98a2SYan Zheng return root; 24805d4f98a2SYan Zheng } 24815d4f98a2SYan Zheng 24823fd0a558SYan, Zheng /* 24833fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 24843fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 24853fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 24863fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 24873fd0a558SYan, Zheng */ 24885d4f98a2SYan Zheng static noinline_for_stack 24895d4f98a2SYan Zheng struct btrfs_root *select_one_root(struct btrfs_trans_handle *trans, 24905d4f98a2SYan Zheng struct backref_node *node) 24915d4f98a2SYan Zheng { 24923fd0a558SYan, Zheng struct backref_node *next; 24933fd0a558SYan, Zheng struct btrfs_root *root; 24943fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 24955d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 24963fd0a558SYan, Zheng int index = 0; 24973fd0a558SYan, Zheng 24983fd0a558SYan, Zheng next = node; 24993fd0a558SYan, Zheng while (1) { 25003fd0a558SYan, Zheng cond_resched(); 25013fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 25023fd0a558SYan, Zheng root = next->root; 25033fd0a558SYan, Zheng BUG_ON(!root); 25043fd0a558SYan, Zheng 250525985edcSLucas De Marchi /* no other choice for non-references counted tree */ 25063fd0a558SYan, Zheng if (!root->ref_cows) 25073fd0a558SYan, Zheng return root; 25083fd0a558SYan, Zheng 25093fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 25103fd0a558SYan, Zheng fs_root = root; 25113fd0a558SYan, Zheng 25123fd0a558SYan, Zheng if (next != node) 25133fd0a558SYan, Zheng return NULL; 25143fd0a558SYan, Zheng 25153fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 25163fd0a558SYan, Zheng if (!next || next->level <= node->level) 25173fd0a558SYan, Zheng break; 25183fd0a558SYan, Zheng } 25193fd0a558SYan, Zheng 25203fd0a558SYan, Zheng if (!fs_root) 25213fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 25223fd0a558SYan, Zheng return fs_root; 25235d4f98a2SYan Zheng } 25245d4f98a2SYan Zheng 25255d4f98a2SYan Zheng static noinline_for_stack 25263fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 25273fd0a558SYan, Zheng struct backref_node *node, int reserve) 25285d4f98a2SYan Zheng { 25293fd0a558SYan, Zheng struct backref_node *next = node; 25303fd0a558SYan, Zheng struct backref_edge *edge; 25313fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25323fd0a558SYan, Zheng u64 num_bytes = 0; 25333fd0a558SYan, Zheng int index = 0; 25345d4f98a2SYan Zheng 25353fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 25363fd0a558SYan, Zheng 25373fd0a558SYan, Zheng while (next) { 25383fd0a558SYan, Zheng cond_resched(); 25395d4f98a2SYan Zheng while (1) { 25403fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 25415d4f98a2SYan Zheng break; 25425d4f98a2SYan Zheng 25433fd0a558SYan, Zheng num_bytes += btrfs_level_size(rc->extent_root, 25443fd0a558SYan, Zheng next->level); 25453fd0a558SYan, Zheng 25463fd0a558SYan, Zheng if (list_empty(&next->upper)) 25473fd0a558SYan, Zheng break; 25483fd0a558SYan, Zheng 25493fd0a558SYan, Zheng edge = list_entry(next->upper.next, 25503fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 25513fd0a558SYan, Zheng edges[index++] = edge; 25523fd0a558SYan, Zheng next = edge->node[UPPER]; 25535d4f98a2SYan Zheng } 25543fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 25553fd0a558SYan, Zheng } 25563fd0a558SYan, Zheng return num_bytes; 25573fd0a558SYan, Zheng } 25583fd0a558SYan, Zheng 25593fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 25603fd0a558SYan, Zheng struct reloc_control *rc, 25613fd0a558SYan, Zheng struct backref_node *node) 25623fd0a558SYan, Zheng { 25633fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 25643fd0a558SYan, Zheng u64 num_bytes; 25653fd0a558SYan, Zheng int ret; 25663fd0a558SYan, Zheng 25673fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 25683fd0a558SYan, Zheng 25693fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 257008e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 257108e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 25723fd0a558SYan, Zheng if (ret) { 25733fd0a558SYan, Zheng if (ret == -EAGAIN) 25743fd0a558SYan, Zheng rc->commit_transaction = 1; 25753fd0a558SYan, Zheng return ret; 25763fd0a558SYan, Zheng } 25773fd0a558SYan, Zheng 25783fd0a558SYan, Zheng return 0; 25793fd0a558SYan, Zheng } 25803fd0a558SYan, Zheng 25813fd0a558SYan, Zheng static void release_metadata_space(struct reloc_control *rc, 25823fd0a558SYan, Zheng struct backref_node *node) 25833fd0a558SYan, Zheng { 25843fd0a558SYan, Zheng u64 num_bytes = calcu_metadata_size(rc, node, 0) * 2; 25853fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, num_bytes); 25865d4f98a2SYan Zheng } 25875d4f98a2SYan Zheng 25885d4f98a2SYan Zheng /* 25895d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 25905d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 25915d4f98a2SYan Zheng * 25925d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 25935d4f98a2SYan Zheng * in that case this function just updates pointers. 25945d4f98a2SYan Zheng */ 25955d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 25963fd0a558SYan, Zheng struct reloc_control *rc, 25975d4f98a2SYan Zheng struct backref_node *node, 25985d4f98a2SYan Zheng struct btrfs_key *key, 25995d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 26005d4f98a2SYan Zheng { 26015d4f98a2SYan Zheng struct backref_node *upper; 26025d4f98a2SYan Zheng struct backref_edge *edge; 26035d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26045d4f98a2SYan Zheng struct btrfs_root *root; 26055d4f98a2SYan Zheng struct extent_buffer *eb; 26065d4f98a2SYan Zheng u32 blocksize; 26075d4f98a2SYan Zheng u64 bytenr; 26085d4f98a2SYan Zheng u64 generation; 26095d4f98a2SYan Zheng int nr; 26105d4f98a2SYan Zheng int slot; 26115d4f98a2SYan Zheng int ret; 26125d4f98a2SYan Zheng int err = 0; 26135d4f98a2SYan Zheng 26145d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 26155d4f98a2SYan Zheng 26165d4f98a2SYan Zheng path->lowest_level = node->level + 1; 26173fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 26185d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 26195d4f98a2SYan Zheng cond_resched(); 26205d4f98a2SYan Zheng 26215d4f98a2SYan Zheng upper = edge->node[UPPER]; 26223fd0a558SYan, Zheng root = select_reloc_root(trans, rc, upper, edges, &nr); 26233fd0a558SYan, Zheng BUG_ON(!root); 26245d4f98a2SYan Zheng 26253fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 26263fd0a558SYan, Zheng if (!lowest) { 26273fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 26283fd0a558SYan, Zheng upper->level, &slot); 26293fd0a558SYan, Zheng BUG_ON(ret); 26303fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 26313fd0a558SYan, Zheng if (node->eb->start == bytenr) 26323fd0a558SYan, Zheng goto next; 26333fd0a558SYan, Zheng } 26345d4f98a2SYan Zheng drop_node_buffer(upper); 26353fd0a558SYan, Zheng } 26365d4f98a2SYan Zheng 26375d4f98a2SYan Zheng if (!upper->eb) { 26385d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 26395d4f98a2SYan Zheng if (ret < 0) { 26405d4f98a2SYan Zheng err = ret; 26415d4f98a2SYan Zheng break; 26425d4f98a2SYan Zheng } 26435d4f98a2SYan Zheng BUG_ON(ret > 0); 26445d4f98a2SYan Zheng 26453fd0a558SYan, Zheng if (!upper->eb) { 26463fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 26473fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 26483fd0a558SYan, Zheng } else { 26493fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 26503fd0a558SYan, Zheng } 26513fd0a558SYan, Zheng 26523fd0a558SYan, Zheng upper->locked = 1; 26533fd0a558SYan, Zheng path->locks[upper->level] = 0; 26543fd0a558SYan, Zheng 26555d4f98a2SYan Zheng slot = path->slots[upper->level]; 2656b3b4aa74SDavid Sterba btrfs_release_path(path); 26575d4f98a2SYan Zheng } else { 26585d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 26595d4f98a2SYan Zheng &slot); 26605d4f98a2SYan Zheng BUG_ON(ret); 26615d4f98a2SYan Zheng } 26625d4f98a2SYan Zheng 26635d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 26643fd0a558SYan, Zheng if (lowest) { 26653fd0a558SYan, Zheng BUG_ON(bytenr != node->bytenr); 26665d4f98a2SYan Zheng } else { 26673fd0a558SYan, Zheng if (node->eb->start == bytenr) 26683fd0a558SYan, Zheng goto next; 26695d4f98a2SYan Zheng } 26705d4f98a2SYan Zheng 26715d4f98a2SYan Zheng blocksize = btrfs_level_size(root, node->level); 26725d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 26735d4f98a2SYan Zheng eb = read_tree_block(root, bytenr, blocksize, generation); 2674416bc658SJosef Bacik if (!eb || !extent_buffer_uptodate(eb)) { 2675416bc658SJosef Bacik free_extent_buffer(eb); 267697d9a8a4STsutomu Itoh err = -EIO; 267797d9a8a4STsutomu Itoh goto next; 267897d9a8a4STsutomu Itoh } 26795d4f98a2SYan Zheng btrfs_tree_lock(eb); 26805d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 26815d4f98a2SYan Zheng 26825d4f98a2SYan Zheng if (!node->eb) { 26835d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 26845d4f98a2SYan Zheng slot, &eb); 26853fd0a558SYan, Zheng btrfs_tree_unlock(eb); 26863fd0a558SYan, Zheng free_extent_buffer(eb); 26875d4f98a2SYan Zheng if (ret < 0) { 26885d4f98a2SYan Zheng err = ret; 26893fd0a558SYan, Zheng goto next; 26905d4f98a2SYan Zheng } 26913fd0a558SYan, Zheng BUG_ON(node->eb != eb); 26925d4f98a2SYan Zheng } else { 26935d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 26945d4f98a2SYan Zheng node->eb->start); 26955d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 26965d4f98a2SYan Zheng trans->transid); 26975d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 26985d4f98a2SYan Zheng 26995d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, 27005d4f98a2SYan Zheng node->eb->start, blocksize, 27015d4f98a2SYan Zheng upper->eb->start, 27025d4f98a2SYan Zheng btrfs_header_owner(upper->eb), 270366d7e7f0SArne Jansen node->level, 0, 1); 27045d4f98a2SYan Zheng BUG_ON(ret); 27055d4f98a2SYan Zheng 27065d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 27075d4f98a2SYan Zheng BUG_ON(ret); 27085d4f98a2SYan Zheng } 27093fd0a558SYan, Zheng next: 27103fd0a558SYan, Zheng if (!upper->pending) 27113fd0a558SYan, Zheng drop_node_buffer(upper); 27123fd0a558SYan, Zheng else 27133fd0a558SYan, Zheng unlock_node_buffer(upper); 27143fd0a558SYan, Zheng if (err) 27153fd0a558SYan, Zheng break; 27165d4f98a2SYan Zheng } 27173fd0a558SYan, Zheng 27183fd0a558SYan, Zheng if (!err && node->pending) { 27193fd0a558SYan, Zheng drop_node_buffer(node); 27203fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 27213fd0a558SYan, Zheng node->pending = 0; 27225d4f98a2SYan Zheng } 27233fd0a558SYan, Zheng 27245d4f98a2SYan Zheng path->lowest_level = 0; 27253fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 27265d4f98a2SYan Zheng return err; 27275d4f98a2SYan Zheng } 27285d4f98a2SYan Zheng 27295d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 27303fd0a558SYan, Zheng struct reloc_control *rc, 27315d4f98a2SYan Zheng struct backref_node *node, 27325d4f98a2SYan Zheng struct btrfs_path *path) 27335d4f98a2SYan Zheng { 27345d4f98a2SYan Zheng struct btrfs_key key; 27355d4f98a2SYan Zheng 27365d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 27373fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 27385d4f98a2SYan Zheng } 27395d4f98a2SYan Zheng 27405d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 27413fd0a558SYan, Zheng struct reloc_control *rc, 27423fd0a558SYan, Zheng struct btrfs_path *path, int err) 27435d4f98a2SYan Zheng { 27443fd0a558SYan, Zheng LIST_HEAD(list); 27453fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 27465d4f98a2SYan Zheng struct backref_node *node; 27475d4f98a2SYan Zheng int level; 27485d4f98a2SYan Zheng int ret; 27495d4f98a2SYan Zheng 27505d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 27515d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 27525d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 27533fd0a558SYan, Zheng struct backref_node, list); 27543fd0a558SYan, Zheng list_move_tail(&node->list, &list); 27553fd0a558SYan, Zheng BUG_ON(!node->pending); 27565d4f98a2SYan Zheng 27573fd0a558SYan, Zheng if (!err) { 27583fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 27595d4f98a2SYan Zheng if (ret < 0) 27605d4f98a2SYan Zheng err = ret; 27615d4f98a2SYan Zheng } 27625d4f98a2SYan Zheng } 27633fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 27643fd0a558SYan, Zheng } 27655d4f98a2SYan Zheng return err; 27665d4f98a2SYan Zheng } 27675d4f98a2SYan Zheng 27685d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 27693fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 27703fd0a558SYan, Zheng { 27713fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 27723fd0a558SYan, Zheng EXTENT_DIRTY, GFP_NOFS); 27733fd0a558SYan, Zheng } 27743fd0a558SYan, Zheng 27753fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 27765d4f98a2SYan Zheng struct backref_node *node) 27775d4f98a2SYan Zheng { 27785d4f98a2SYan Zheng u32 blocksize; 27795d4f98a2SYan Zheng if (node->level == 0 || 27805d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 27815d4f98a2SYan Zheng blocksize = btrfs_level_size(rc->extent_root, node->level); 27823fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 27835d4f98a2SYan Zheng } 27845d4f98a2SYan Zheng node->processed = 1; 27855d4f98a2SYan Zheng } 27865d4f98a2SYan Zheng 27875d4f98a2SYan Zheng /* 27885d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 27895d4f98a2SYan Zheng * as processed. 27905d4f98a2SYan Zheng */ 27915d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 27925d4f98a2SYan Zheng struct backref_node *node) 27935d4f98a2SYan Zheng { 27945d4f98a2SYan Zheng struct backref_node *next = node; 27955d4f98a2SYan Zheng struct backref_edge *edge; 27965d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27975d4f98a2SYan Zheng int index = 0; 27985d4f98a2SYan Zheng 27995d4f98a2SYan Zheng while (next) { 28005d4f98a2SYan Zheng cond_resched(); 28015d4f98a2SYan Zheng while (1) { 28025d4f98a2SYan Zheng if (next->processed) 28035d4f98a2SYan Zheng break; 28045d4f98a2SYan Zheng 28053fd0a558SYan, Zheng __mark_block_processed(rc, next); 28065d4f98a2SYan Zheng 28075d4f98a2SYan Zheng if (list_empty(&next->upper)) 28085d4f98a2SYan Zheng break; 28095d4f98a2SYan Zheng 28105d4f98a2SYan Zheng edge = list_entry(next->upper.next, 28115d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 28125d4f98a2SYan Zheng edges[index++] = edge; 28135d4f98a2SYan Zheng next = edge->node[UPPER]; 28145d4f98a2SYan Zheng } 28155d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 28165d4f98a2SYan Zheng } 28175d4f98a2SYan Zheng } 28185d4f98a2SYan Zheng 28195d4f98a2SYan Zheng static int tree_block_processed(u64 bytenr, u32 blocksize, 28205d4f98a2SYan Zheng struct reloc_control *rc) 28215d4f98a2SYan Zheng { 28225d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 28239655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 28245d4f98a2SYan Zheng return 1; 28255d4f98a2SYan Zheng return 0; 28265d4f98a2SYan Zheng } 28275d4f98a2SYan Zheng 28285d4f98a2SYan Zheng static int get_tree_block_key(struct reloc_control *rc, 28295d4f98a2SYan Zheng struct tree_block *block) 28305d4f98a2SYan Zheng { 28315d4f98a2SYan Zheng struct extent_buffer *eb; 28325d4f98a2SYan Zheng 28335d4f98a2SYan Zheng BUG_ON(block->key_ready); 28345d4f98a2SYan Zheng eb = read_tree_block(rc->extent_root, block->bytenr, 28355d4f98a2SYan Zheng block->key.objectid, block->key.offset); 2836416bc658SJosef Bacik if (!eb || !extent_buffer_uptodate(eb)) { 2837416bc658SJosef Bacik free_extent_buffer(eb); 2838416bc658SJosef Bacik return -EIO; 2839416bc658SJosef Bacik } 28405d4f98a2SYan Zheng WARN_ON(btrfs_header_level(eb) != block->level); 28415d4f98a2SYan Zheng if (block->level == 0) 28425d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 28435d4f98a2SYan Zheng else 28445d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 28455d4f98a2SYan Zheng free_extent_buffer(eb); 28465d4f98a2SYan Zheng block->key_ready = 1; 28475d4f98a2SYan Zheng return 0; 28485d4f98a2SYan Zheng } 28495d4f98a2SYan Zheng 28505d4f98a2SYan Zheng static int reada_tree_block(struct reloc_control *rc, 28515d4f98a2SYan Zheng struct tree_block *block) 28525d4f98a2SYan Zheng { 28535d4f98a2SYan Zheng BUG_ON(block->key_ready); 28543173a18fSJosef Bacik if (block->key.type == BTRFS_METADATA_ITEM_KEY) 28553173a18fSJosef Bacik readahead_tree_block(rc->extent_root, block->bytenr, 28563173a18fSJosef Bacik block->key.objectid, 28573173a18fSJosef Bacik rc->extent_root->leafsize); 28583173a18fSJosef Bacik else 28595d4f98a2SYan Zheng readahead_tree_block(rc->extent_root, block->bytenr, 28605d4f98a2SYan Zheng block->key.objectid, block->key.offset); 28615d4f98a2SYan Zheng return 0; 28625d4f98a2SYan Zheng } 28635d4f98a2SYan Zheng 28645d4f98a2SYan Zheng /* 28655d4f98a2SYan Zheng * helper function to relocate a tree block 28665d4f98a2SYan Zheng */ 28675d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 28685d4f98a2SYan Zheng struct reloc_control *rc, 28695d4f98a2SYan Zheng struct backref_node *node, 28705d4f98a2SYan Zheng struct btrfs_key *key, 28715d4f98a2SYan Zheng struct btrfs_path *path) 28725d4f98a2SYan Zheng { 28735d4f98a2SYan Zheng struct btrfs_root *root; 28743fd0a558SYan, Zheng int release = 0; 28753fd0a558SYan, Zheng int ret = 0; 28765d4f98a2SYan Zheng 28773fd0a558SYan, Zheng if (!node) 28785d4f98a2SYan Zheng return 0; 28793fd0a558SYan, Zheng 28803fd0a558SYan, Zheng BUG_ON(node->processed); 28813fd0a558SYan, Zheng root = select_one_root(trans, node); 28823fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 28833fd0a558SYan, Zheng update_processed_blocks(rc, node); 28843fd0a558SYan, Zheng goto out; 28855d4f98a2SYan Zheng } 28865d4f98a2SYan Zheng 28873fd0a558SYan, Zheng if (!root || root->ref_cows) { 28883fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 28893fd0a558SYan, Zheng if (ret) 28905d4f98a2SYan Zheng goto out; 28913fd0a558SYan, Zheng release = 1; 28925d4f98a2SYan Zheng } 28933fd0a558SYan, Zheng 28943fd0a558SYan, Zheng if (root) { 28953fd0a558SYan, Zheng if (root->ref_cows) { 28963fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 28973fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 28983fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 28993fd0a558SYan, Zheng root = root->reloc_root; 29003fd0a558SYan, Zheng node->new_bytenr = root->node->start; 29013fd0a558SYan, Zheng node->root = root; 29023fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 29033fd0a558SYan, Zheng } else { 29045d4f98a2SYan Zheng path->lowest_level = node->level; 29055d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2906b3b4aa74SDavid Sterba btrfs_release_path(path); 29073fd0a558SYan, Zheng if (ret > 0) 29085d4f98a2SYan Zheng ret = 0; 29093fd0a558SYan, Zheng } 29103fd0a558SYan, Zheng if (!ret) 29113fd0a558SYan, Zheng update_processed_blocks(rc, node); 29123fd0a558SYan, Zheng } else { 29133fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 29143fd0a558SYan, Zheng } 29155d4f98a2SYan Zheng out: 29163fd0a558SYan, Zheng if (ret || node->level == 0 || node->cowonly) { 29173fd0a558SYan, Zheng if (release) 29183fd0a558SYan, Zheng release_metadata_space(rc, node); 29193fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 29203fd0a558SYan, Zheng } 29215d4f98a2SYan Zheng return ret; 29225d4f98a2SYan Zheng } 29235d4f98a2SYan Zheng 29245d4f98a2SYan Zheng /* 29255d4f98a2SYan Zheng * relocate a list of blocks 29265d4f98a2SYan Zheng */ 29275d4f98a2SYan Zheng static noinline_for_stack 29285d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 29295d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 29305d4f98a2SYan Zheng { 29315d4f98a2SYan Zheng struct backref_node *node; 29325d4f98a2SYan Zheng struct btrfs_path *path; 29335d4f98a2SYan Zheng struct tree_block *block; 29345d4f98a2SYan Zheng struct rb_node *rb_node; 29355d4f98a2SYan Zheng int ret; 29365d4f98a2SYan Zheng int err = 0; 29375d4f98a2SYan Zheng 29385d4f98a2SYan Zheng path = btrfs_alloc_path(); 2939e1a12670SLiu Bo if (!path) { 2940e1a12670SLiu Bo err = -ENOMEM; 294134c2b290SDavid Sterba goto out_free_blocks; 2942e1a12670SLiu Bo } 29435d4f98a2SYan Zheng 29445d4f98a2SYan Zheng rb_node = rb_first(blocks); 29455d4f98a2SYan Zheng while (rb_node) { 29465d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 29475d4f98a2SYan Zheng if (!block->key_ready) 29485d4f98a2SYan Zheng reada_tree_block(rc, block); 29495d4f98a2SYan Zheng rb_node = rb_next(rb_node); 29505d4f98a2SYan Zheng } 29515d4f98a2SYan Zheng 29525d4f98a2SYan Zheng rb_node = rb_first(blocks); 29535d4f98a2SYan Zheng while (rb_node) { 29545d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 295534c2b290SDavid Sterba if (!block->key_ready) { 295634c2b290SDavid Sterba err = get_tree_block_key(rc, block); 295734c2b290SDavid Sterba if (err) 295834c2b290SDavid Sterba goto out_free_path; 295934c2b290SDavid Sterba } 29605d4f98a2SYan Zheng rb_node = rb_next(rb_node); 29615d4f98a2SYan Zheng } 29625d4f98a2SYan Zheng 29635d4f98a2SYan Zheng rb_node = rb_first(blocks); 29645d4f98a2SYan Zheng while (rb_node) { 29655d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 29665d4f98a2SYan Zheng 29673fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 29685d4f98a2SYan Zheng block->level, block->bytenr); 29695d4f98a2SYan Zheng if (IS_ERR(node)) { 29705d4f98a2SYan Zheng err = PTR_ERR(node); 29715d4f98a2SYan Zheng goto out; 29725d4f98a2SYan Zheng } 29735d4f98a2SYan Zheng 29745d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 29755d4f98a2SYan Zheng path); 29765d4f98a2SYan Zheng if (ret < 0) { 29773fd0a558SYan, Zheng if (ret != -EAGAIN || rb_node == rb_first(blocks)) 29785d4f98a2SYan Zheng err = ret; 29795d4f98a2SYan Zheng goto out; 29805d4f98a2SYan Zheng } 29815d4f98a2SYan Zheng rb_node = rb_next(rb_node); 29825d4f98a2SYan Zheng } 29835d4f98a2SYan Zheng out: 29843fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 29855d4f98a2SYan Zheng 298634c2b290SDavid Sterba out_free_path: 29875d4f98a2SYan Zheng btrfs_free_path(path); 298834c2b290SDavid Sterba out_free_blocks: 2989e1a12670SLiu Bo free_block_list(blocks); 29905d4f98a2SYan Zheng return err; 29915d4f98a2SYan Zheng } 29925d4f98a2SYan Zheng 29935d4f98a2SYan Zheng static noinline_for_stack 2994efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 2995efa56464SYan, Zheng struct file_extent_cluster *cluster) 2996efa56464SYan, Zheng { 2997efa56464SYan, Zheng u64 alloc_hint = 0; 2998efa56464SYan, Zheng u64 start; 2999efa56464SYan, Zheng u64 end; 3000efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3001efa56464SYan, Zheng u64 num_bytes; 3002efa56464SYan, Zheng int nr = 0; 3003efa56464SYan, Zheng int ret = 0; 3004efa56464SYan, Zheng 3005efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 3006efa56464SYan, Zheng mutex_lock(&inode->i_mutex); 3007efa56464SYan, Zheng 3008efa56464SYan, Zheng ret = btrfs_check_data_free_space(inode, cluster->end + 3009efa56464SYan, Zheng 1 - cluster->start); 3010efa56464SYan, Zheng if (ret) 3011efa56464SYan, Zheng goto out; 3012efa56464SYan, Zheng 3013efa56464SYan, Zheng while (nr < cluster->nr) { 3014efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3015efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3016efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3017efa56464SYan, Zheng else 3018efa56464SYan, Zheng end = cluster->end - offset; 3019efa56464SYan, Zheng 3020d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3021efa56464SYan, Zheng num_bytes = end + 1 - start; 3022efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3023efa56464SYan, Zheng num_bytes, num_bytes, 3024efa56464SYan, Zheng end + 1, &alloc_hint); 3025d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3026efa56464SYan, Zheng if (ret) 3027efa56464SYan, Zheng break; 3028efa56464SYan, Zheng nr++; 3029efa56464SYan, Zheng } 3030efa56464SYan, Zheng btrfs_free_reserved_data_space(inode, cluster->end + 3031efa56464SYan, Zheng 1 - cluster->start); 3032efa56464SYan, Zheng out: 3033efa56464SYan, Zheng mutex_unlock(&inode->i_mutex); 3034efa56464SYan, Zheng return ret; 3035efa56464SYan, Zheng } 3036efa56464SYan, Zheng 3037efa56464SYan, Zheng static noinline_for_stack 30380257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 30390257bb82SYan, Zheng u64 block_start) 30405d4f98a2SYan Zheng { 30415d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 30425d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 30435d4f98a2SYan Zheng struct extent_map *em; 30440257bb82SYan, Zheng int ret = 0; 30455d4f98a2SYan Zheng 3046172ddd60SDavid Sterba em = alloc_extent_map(); 30470257bb82SYan, Zheng if (!em) 30480257bb82SYan, Zheng return -ENOMEM; 30490257bb82SYan, Zheng 30505d4f98a2SYan Zheng em->start = start; 30510257bb82SYan, Zheng em->len = end + 1 - start; 30520257bb82SYan, Zheng em->block_len = em->len; 30530257bb82SYan, Zheng em->block_start = block_start; 30545d4f98a2SYan Zheng em->bdev = root->fs_info->fs_devices->latest_bdev; 30555d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 30565d4f98a2SYan Zheng 3057d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 30585d4f98a2SYan Zheng while (1) { 3059890871beSChris Mason write_lock(&em_tree->lock); 306009a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3061890871beSChris Mason write_unlock(&em_tree->lock); 30625d4f98a2SYan Zheng if (ret != -EEXIST) { 30635d4f98a2SYan Zheng free_extent_map(em); 30645d4f98a2SYan Zheng break; 30655d4f98a2SYan Zheng } 30665d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 0); 30675d4f98a2SYan Zheng } 3068d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 30690257bb82SYan, Zheng return ret; 30700257bb82SYan, Zheng } 30715d4f98a2SYan Zheng 30720257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 30730257bb82SYan, Zheng struct file_extent_cluster *cluster) 30740257bb82SYan, Zheng { 30750257bb82SYan, Zheng u64 page_start; 30760257bb82SYan, Zheng u64 page_end; 30770257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 30780257bb82SYan, Zheng unsigned long index; 30790257bb82SYan, Zheng unsigned long last_index; 30800257bb82SYan, Zheng struct page *page; 30810257bb82SYan, Zheng struct file_ra_state *ra; 30823b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 30830257bb82SYan, Zheng int nr = 0; 30840257bb82SYan, Zheng int ret = 0; 30850257bb82SYan, Zheng 30860257bb82SYan, Zheng if (!cluster->nr) 30870257bb82SYan, Zheng return 0; 30880257bb82SYan, Zheng 30890257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 30900257bb82SYan, Zheng if (!ra) 30910257bb82SYan, Zheng return -ENOMEM; 30920257bb82SYan, Zheng 3093efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 30940257bb82SYan, Zheng if (ret) 3095efa56464SYan, Zheng goto out; 30960257bb82SYan, Zheng 30970257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 30980257bb82SYan, Zheng 3099efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3100efa56464SYan, Zheng cluster->end - offset, cluster->start); 3101efa56464SYan, Zheng if (ret) 3102efa56464SYan, Zheng goto out; 3103efa56464SYan, Zheng 3104efa56464SYan, Zheng index = (cluster->start - offset) >> PAGE_CACHE_SHIFT; 3105efa56464SYan, Zheng last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT; 31060257bb82SYan, Zheng while (index <= last_index) { 3107efa56464SYan, Zheng ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE); 3108efa56464SYan, Zheng if (ret) 3109efa56464SYan, Zheng goto out; 3110efa56464SYan, Zheng 31110257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 31120257bb82SYan, Zheng if (!page) { 31130257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 31140257bb82SYan, Zheng ra, NULL, index, 31150257bb82SYan, Zheng last_index + 1 - index); 3116a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 31173b16a4e3SJosef Bacik mask); 31180257bb82SYan, Zheng if (!page) { 3119efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 3120efa56464SYan, Zheng PAGE_CACHE_SIZE); 31210257bb82SYan, Zheng ret = -ENOMEM; 3122efa56464SYan, Zheng goto out; 31230257bb82SYan, Zheng } 31240257bb82SYan, Zheng } 31250257bb82SYan, Zheng 31260257bb82SYan, Zheng if (PageReadahead(page)) { 31270257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 31280257bb82SYan, Zheng ra, NULL, page, index, 31290257bb82SYan, Zheng last_index + 1 - index); 31300257bb82SYan, Zheng } 31310257bb82SYan, Zheng 31320257bb82SYan, Zheng if (!PageUptodate(page)) { 31330257bb82SYan, Zheng btrfs_readpage(NULL, page); 31340257bb82SYan, Zheng lock_page(page); 31350257bb82SYan, Zheng if (!PageUptodate(page)) { 31360257bb82SYan, Zheng unlock_page(page); 31370257bb82SYan, Zheng page_cache_release(page); 3138efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 3139efa56464SYan, Zheng PAGE_CACHE_SIZE); 31400257bb82SYan, Zheng ret = -EIO; 3141efa56464SYan, Zheng goto out; 31420257bb82SYan, Zheng } 31430257bb82SYan, Zheng } 31440257bb82SYan, Zheng 31454eee4fa4SMiao Xie page_start = page_offset(page); 31460257bb82SYan, Zheng page_end = page_start + PAGE_CACHE_SIZE - 1; 31470257bb82SYan, Zheng 3148d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 31490257bb82SYan, Zheng 31500257bb82SYan, Zheng set_page_extent_mapped(page); 31510257bb82SYan, Zheng 31520257bb82SYan, Zheng if (nr < cluster->nr && 31530257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 31540257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 31550257bb82SYan, Zheng page_start, page_end, 31560257bb82SYan, Zheng EXTENT_BOUNDARY, GFP_NOFS); 31570257bb82SYan, Zheng nr++; 31580257bb82SYan, Zheng } 31590257bb82SYan, Zheng 3160efa56464SYan, Zheng btrfs_set_extent_delalloc(inode, page_start, page_end, NULL); 31610257bb82SYan, Zheng set_page_dirty(page); 31620257bb82SYan, Zheng 31630257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3164d0082371SJeff Mahoney page_start, page_end); 31650257bb82SYan, Zheng unlock_page(page); 31660257bb82SYan, Zheng page_cache_release(page); 31670257bb82SYan, Zheng 31680257bb82SYan, Zheng index++; 3169efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 3170efa56464SYan, Zheng btrfs_throttle(BTRFS_I(inode)->root); 31710257bb82SYan, Zheng } 31720257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3173efa56464SYan, Zheng out: 31740257bb82SYan, Zheng kfree(ra); 31750257bb82SYan, Zheng return ret; 31760257bb82SYan, Zheng } 31770257bb82SYan, Zheng 31780257bb82SYan, Zheng static noinline_for_stack 31790257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 31800257bb82SYan, Zheng struct file_extent_cluster *cluster) 31810257bb82SYan, Zheng { 31820257bb82SYan, Zheng int ret; 31830257bb82SYan, Zheng 31840257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 31850257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31860257bb82SYan, Zheng if (ret) 31870257bb82SYan, Zheng return ret; 31880257bb82SYan, Zheng cluster->nr = 0; 31890257bb82SYan, Zheng } 31900257bb82SYan, Zheng 31910257bb82SYan, Zheng if (!cluster->nr) 31920257bb82SYan, Zheng cluster->start = extent_key->objectid; 31930257bb82SYan, Zheng else 31940257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 31950257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 31960257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 31970257bb82SYan, Zheng cluster->nr++; 31980257bb82SYan, Zheng 31990257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 32000257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 32010257bb82SYan, Zheng if (ret) 32020257bb82SYan, Zheng return ret; 32030257bb82SYan, Zheng cluster->nr = 0; 32040257bb82SYan, Zheng } 32050257bb82SYan, Zheng return 0; 32065d4f98a2SYan Zheng } 32075d4f98a2SYan Zheng 32085d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 32095d4f98a2SYan Zheng static int get_ref_objectid_v0(struct reloc_control *rc, 32105d4f98a2SYan Zheng struct btrfs_path *path, 32115d4f98a2SYan Zheng struct btrfs_key *extent_key, 32125d4f98a2SYan Zheng u64 *ref_objectid, int *path_change) 32135d4f98a2SYan Zheng { 32145d4f98a2SYan Zheng struct btrfs_key key; 32155d4f98a2SYan Zheng struct extent_buffer *leaf; 32165d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 32175d4f98a2SYan Zheng int ret; 32185d4f98a2SYan Zheng int slot; 32195d4f98a2SYan Zheng 32205d4f98a2SYan Zheng leaf = path->nodes[0]; 32215d4f98a2SYan Zheng slot = path->slots[0]; 32225d4f98a2SYan Zheng while (1) { 32235d4f98a2SYan Zheng if (slot >= btrfs_header_nritems(leaf)) { 32245d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 32255d4f98a2SYan Zheng if (ret < 0) 32265d4f98a2SYan Zheng return ret; 32275d4f98a2SYan Zheng BUG_ON(ret > 0); 32285d4f98a2SYan Zheng leaf = path->nodes[0]; 32295d4f98a2SYan Zheng slot = path->slots[0]; 32305d4f98a2SYan Zheng if (path_change) 32315d4f98a2SYan Zheng *path_change = 1; 32325d4f98a2SYan Zheng } 32335d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 32345d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 32355d4f98a2SYan Zheng return -ENOENT; 32365d4f98a2SYan Zheng 32375d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_REF_V0_KEY) { 32385d4f98a2SYan Zheng slot++; 32395d4f98a2SYan Zheng continue; 32405d4f98a2SYan Zheng } 32415d4f98a2SYan Zheng ref0 = btrfs_item_ptr(leaf, slot, 32425d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 32435d4f98a2SYan Zheng *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0); 32445d4f98a2SYan Zheng break; 32455d4f98a2SYan Zheng } 32465d4f98a2SYan Zheng return 0; 32475d4f98a2SYan Zheng } 32485d4f98a2SYan Zheng #endif 32495d4f98a2SYan Zheng 32505d4f98a2SYan Zheng /* 32515d4f98a2SYan Zheng * helper to add a tree block to the list. 32525d4f98a2SYan Zheng * the major work is getting the generation and level of the block 32535d4f98a2SYan Zheng */ 32545d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 32555d4f98a2SYan Zheng struct btrfs_key *extent_key, 32565d4f98a2SYan Zheng struct btrfs_path *path, 32575d4f98a2SYan Zheng struct rb_root *blocks) 32585d4f98a2SYan Zheng { 32595d4f98a2SYan Zheng struct extent_buffer *eb; 32605d4f98a2SYan Zheng struct btrfs_extent_item *ei; 32615d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 32625d4f98a2SYan Zheng struct tree_block *block; 32635d4f98a2SYan Zheng struct rb_node *rb_node; 32645d4f98a2SYan Zheng u32 item_size; 32655d4f98a2SYan Zheng int level = -1; 32667fdf4b60SWang Shilong u64 generation; 32675d4f98a2SYan Zheng 32685d4f98a2SYan Zheng eb = path->nodes[0]; 32695d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 32705d4f98a2SYan Zheng 32713173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 32723173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 32735d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 32745d4f98a2SYan Zheng struct btrfs_extent_item); 32753173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 32765d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 32775d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 32785d4f98a2SYan Zheng } else { 32793173a18fSJosef Bacik level = (int)extent_key->offset; 32803173a18fSJosef Bacik } 32813173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 32823173a18fSJosef Bacik } else { 32835d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 32845d4f98a2SYan Zheng u64 ref_owner; 32855d4f98a2SYan Zheng int ret; 32865d4f98a2SYan Zheng 32875d4f98a2SYan Zheng BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 32885d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, extent_key, 32895d4f98a2SYan Zheng &ref_owner, NULL); 3290411fc6bcSAndi Kleen if (ret < 0) 3291411fc6bcSAndi Kleen return ret; 32925d4f98a2SYan Zheng BUG_ON(ref_owner >= BTRFS_MAX_LEVEL); 32935d4f98a2SYan Zheng level = (int)ref_owner; 32945d4f98a2SYan Zheng /* FIXME: get real generation */ 32955d4f98a2SYan Zheng generation = 0; 32965d4f98a2SYan Zheng #else 32975d4f98a2SYan Zheng BUG(); 32985d4f98a2SYan Zheng #endif 32995d4f98a2SYan Zheng } 33005d4f98a2SYan Zheng 3301b3b4aa74SDavid Sterba btrfs_release_path(path); 33025d4f98a2SYan Zheng 33035d4f98a2SYan Zheng BUG_ON(level == -1); 33045d4f98a2SYan Zheng 33055d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 33065d4f98a2SYan Zheng if (!block) 33075d4f98a2SYan Zheng return -ENOMEM; 33085d4f98a2SYan Zheng 33095d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 33103173a18fSJosef Bacik block->key.objectid = rc->extent_root->leafsize; 33115d4f98a2SYan Zheng block->key.offset = generation; 33125d4f98a2SYan Zheng block->level = level; 33135d4f98a2SYan Zheng block->key_ready = 0; 33145d4f98a2SYan Zheng 33155d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 331643c04fb1SJeff Mahoney if (rb_node) 331743c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 33185d4f98a2SYan Zheng 33195d4f98a2SYan Zheng return 0; 33205d4f98a2SYan Zheng } 33215d4f98a2SYan Zheng 33225d4f98a2SYan Zheng /* 33235d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 33245d4f98a2SYan Zheng */ 33255d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 33265d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 33275d4f98a2SYan Zheng struct rb_root *blocks) 33285d4f98a2SYan Zheng { 33295d4f98a2SYan Zheng struct btrfs_path *path; 33305d4f98a2SYan Zheng struct btrfs_key key; 33315d4f98a2SYan Zheng int ret; 3332aee68ee5SJosef Bacik bool skinny = btrfs_fs_incompat(rc->extent_root->fs_info, 3333aee68ee5SJosef Bacik SKINNY_METADATA); 33345d4f98a2SYan Zheng 33355d4f98a2SYan Zheng if (tree_block_processed(bytenr, blocksize, rc)) 33365d4f98a2SYan Zheng return 0; 33375d4f98a2SYan Zheng 33385d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 33395d4f98a2SYan Zheng return 0; 33405d4f98a2SYan Zheng 33415d4f98a2SYan Zheng path = btrfs_alloc_path(); 33425d4f98a2SYan Zheng if (!path) 33435d4f98a2SYan Zheng return -ENOMEM; 3344aee68ee5SJosef Bacik again: 33455d4f98a2SYan Zheng key.objectid = bytenr; 3346aee68ee5SJosef Bacik if (skinny) { 3347aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3348aee68ee5SJosef Bacik key.offset = (u64)-1; 3349aee68ee5SJosef Bacik } else { 33505d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 33515d4f98a2SYan Zheng key.offset = blocksize; 3352aee68ee5SJosef Bacik } 33535d4f98a2SYan Zheng 33545d4f98a2SYan Zheng path->search_commit_root = 1; 33555d4f98a2SYan Zheng path->skip_locking = 1; 33565d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 33575d4f98a2SYan Zheng if (ret < 0) 33585d4f98a2SYan Zheng goto out; 33595d4f98a2SYan Zheng 3360aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3361aee68ee5SJosef Bacik if (path->slots[0]) { 3362aee68ee5SJosef Bacik path->slots[0]--; 3363aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3364aee68ee5SJosef Bacik path->slots[0]); 33653173a18fSJosef Bacik if (key.objectid == bytenr && 3366aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3367aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3368aee68ee5SJosef Bacik key.offset == blocksize))) 33693173a18fSJosef Bacik ret = 0; 33703173a18fSJosef Bacik } 3371aee68ee5SJosef Bacik 3372aee68ee5SJosef Bacik if (ret) { 3373aee68ee5SJosef Bacik skinny = false; 3374aee68ee5SJosef Bacik btrfs_release_path(path); 3375aee68ee5SJosef Bacik goto again; 3376aee68ee5SJosef Bacik } 3377aee68ee5SJosef Bacik } 33783173a18fSJosef Bacik BUG_ON(ret); 33793173a18fSJosef Bacik 33805d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 33815d4f98a2SYan Zheng out: 33825d4f98a2SYan Zheng btrfs_free_path(path); 33835d4f98a2SYan Zheng return ret; 33845d4f98a2SYan Zheng } 33855d4f98a2SYan Zheng 33865d4f98a2SYan Zheng /* 33875d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 33885d4f98a2SYan Zheng */ 33895d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 33905d4f98a2SYan Zheng struct extent_buffer *eb) 33915d4f98a2SYan Zheng { 33925d4f98a2SYan Zheng u64 flags; 33935d4f98a2SYan Zheng int ret; 33945d4f98a2SYan Zheng 33955d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 33965d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 33975d4f98a2SYan Zheng return 1; 33985d4f98a2SYan Zheng 33993fd0a558SYan, Zheng ret = btrfs_lookup_extent_info(NULL, rc->extent_root, 34003173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 34013173a18fSJosef Bacik NULL, &flags); 34025d4f98a2SYan Zheng BUG_ON(ret); 34035d4f98a2SYan Zheng 34045d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 34055d4f98a2SYan Zheng ret = 1; 34065d4f98a2SYan Zheng else 34075d4f98a2SYan Zheng ret = 0; 34085d4f98a2SYan Zheng return ret; 34095d4f98a2SYan Zheng } 34105d4f98a2SYan Zheng 34110af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 34120af3d00bSJosef Bacik struct inode *inode, u64 ino) 34130af3d00bSJosef Bacik { 34140af3d00bSJosef Bacik struct btrfs_key key; 34150af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 34160af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 34170af3d00bSJosef Bacik int ret = 0; 34180af3d00bSJosef Bacik 34190af3d00bSJosef Bacik if (inode) 34200af3d00bSJosef Bacik goto truncate; 34210af3d00bSJosef Bacik 34220af3d00bSJosef Bacik key.objectid = ino; 34230af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 34240af3d00bSJosef Bacik key.offset = 0; 34250af3d00bSJosef Bacik 34260af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 3427f54fb859STsutomu Itoh if (IS_ERR(inode) || is_bad_inode(inode)) { 3428f54fb859STsutomu Itoh if (!IS_ERR(inode)) 34290af3d00bSJosef Bacik iput(inode); 34300af3d00bSJosef Bacik return -ENOENT; 34310af3d00bSJosef Bacik } 34320af3d00bSJosef Bacik 34330af3d00bSJosef Bacik truncate: 34347b61cd92SMiao Xie ret = btrfs_check_trunc_cache_free_space(root, 34357b61cd92SMiao Xie &fs_info->global_block_rsv); 34367b61cd92SMiao Xie if (ret) 34377b61cd92SMiao Xie goto out; 34387b61cd92SMiao Xie 34397a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 34400af3d00bSJosef Bacik if (IS_ERR(trans)) { 34413612b495STsutomu Itoh ret = PTR_ERR(trans); 34420af3d00bSJosef Bacik goto out; 34430af3d00bSJosef Bacik } 34440af3d00bSJosef Bacik 344574514323SFilipe David Borba Manana ret = btrfs_truncate_free_space_cache(root, trans, inode); 34460af3d00bSJosef Bacik 34470af3d00bSJosef Bacik btrfs_end_transaction(trans, root); 3448b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 34490af3d00bSJosef Bacik out: 34500af3d00bSJosef Bacik iput(inode); 34510af3d00bSJosef Bacik return ret; 34520af3d00bSJosef Bacik } 34530af3d00bSJosef Bacik 34545d4f98a2SYan Zheng /* 34555d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 34565d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 34575d4f98a2SYan Zheng */ 34585d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 34595d4f98a2SYan Zheng struct btrfs_key *extent_key, 34605d4f98a2SYan Zheng struct extent_buffer *leaf, 34615d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 34625d4f98a2SYan Zheng struct rb_root *blocks) 34635d4f98a2SYan Zheng { 34645d4f98a2SYan Zheng struct btrfs_path *path; 34655d4f98a2SYan Zheng struct tree_block *block; 34665d4f98a2SYan Zheng struct btrfs_root *root; 34675d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 34685d4f98a2SYan Zheng struct rb_node *rb_node; 34695d4f98a2SYan Zheng struct btrfs_key key; 34705d4f98a2SYan Zheng u64 ref_root; 34715d4f98a2SYan Zheng u64 ref_objectid; 34725d4f98a2SYan Zheng u64 ref_offset; 34735d4f98a2SYan Zheng u32 ref_count; 34745d4f98a2SYan Zheng u32 nritems; 34755d4f98a2SYan Zheng int err = 0; 34765d4f98a2SYan Zheng int added = 0; 34775d4f98a2SYan Zheng int counted; 34785d4f98a2SYan Zheng int ret; 34795d4f98a2SYan Zheng 34805d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 34815d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 34825d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 34835d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 34845d4f98a2SYan Zheng 34850af3d00bSJosef Bacik /* 34860af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 34870af3d00bSJosef Bacik * it and redo the search. 34880af3d00bSJosef Bacik */ 34890af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 34900af3d00bSJosef Bacik ret = delete_block_group_cache(rc->extent_root->fs_info, 34910af3d00bSJosef Bacik NULL, ref_objectid); 34920af3d00bSJosef Bacik if (ret != -ENOENT) 34930af3d00bSJosef Bacik return ret; 34940af3d00bSJosef Bacik ret = 0; 34950af3d00bSJosef Bacik } 34960af3d00bSJosef Bacik 34970af3d00bSJosef Bacik path = btrfs_alloc_path(); 34980af3d00bSJosef Bacik if (!path) 34990af3d00bSJosef Bacik return -ENOMEM; 3500026fd317SJosef Bacik path->reada = 1; 35010af3d00bSJosef Bacik 35025d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, ref_root); 35035d4f98a2SYan Zheng if (IS_ERR(root)) { 35045d4f98a2SYan Zheng err = PTR_ERR(root); 35055d4f98a2SYan Zheng goto out; 35065d4f98a2SYan Zheng } 35075d4f98a2SYan Zheng 35085d4f98a2SYan Zheng key.objectid = ref_objectid; 35095d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 351084850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 351184850e8dSYan, Zheng key.offset = 0; 351284850e8dSYan, Zheng else 351384850e8dSYan, Zheng key.offset = ref_offset; 35145d4f98a2SYan Zheng 35155d4f98a2SYan Zheng path->search_commit_root = 1; 35165d4f98a2SYan Zheng path->skip_locking = 1; 35175d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 35185d4f98a2SYan Zheng if (ret < 0) { 35195d4f98a2SYan Zheng err = ret; 35205d4f98a2SYan Zheng goto out; 35215d4f98a2SYan Zheng } 35225d4f98a2SYan Zheng 35235d4f98a2SYan Zheng leaf = path->nodes[0]; 35245d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 35255d4f98a2SYan Zheng /* 35265d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 35275d4f98a2SYan Zheng * are not counted in 35285d4f98a2SYan Zheng */ 35295d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 35305d4f98a2SYan Zheng counted = 0; 35315d4f98a2SYan Zheng else 35325d4f98a2SYan Zheng counted = 1; 35335d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 35345d4f98a2SYan Zheng if (rb_node) { 35355d4f98a2SYan Zheng if (counted) 35365d4f98a2SYan Zheng added = 1; 35375d4f98a2SYan Zheng else 35385d4f98a2SYan Zheng path->slots[0] = nritems; 35395d4f98a2SYan Zheng } 35405d4f98a2SYan Zheng 35415d4f98a2SYan Zheng while (ref_count > 0) { 35425d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 35435d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 35445d4f98a2SYan Zheng if (ret < 0) { 35455d4f98a2SYan Zheng err = ret; 35465d4f98a2SYan Zheng goto out; 35475d4f98a2SYan Zheng } 35485d4f98a2SYan Zheng if (ret > 0) { 35495d4f98a2SYan Zheng WARN_ON(1); 35505d4f98a2SYan Zheng goto out; 35515d4f98a2SYan Zheng } 35525d4f98a2SYan Zheng 35535d4f98a2SYan Zheng leaf = path->nodes[0]; 35545d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 35555d4f98a2SYan Zheng added = 0; 35565d4f98a2SYan Zheng 35575d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 35585d4f98a2SYan Zheng counted = 0; 35595d4f98a2SYan Zheng else 35605d4f98a2SYan Zheng counted = 1; 35615d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 35625d4f98a2SYan Zheng if (rb_node) { 35635d4f98a2SYan Zheng if (counted) 35645d4f98a2SYan Zheng added = 1; 35655d4f98a2SYan Zheng else 35665d4f98a2SYan Zheng path->slots[0] = nritems; 35675d4f98a2SYan Zheng } 35685d4f98a2SYan Zheng } 35695d4f98a2SYan Zheng 35705d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 35715d4f98a2SYan Zheng if (key.objectid != ref_objectid || 35725d4f98a2SYan Zheng key.type != BTRFS_EXTENT_DATA_KEY) { 35735d4f98a2SYan Zheng WARN_ON(1); 35745d4f98a2SYan Zheng break; 35755d4f98a2SYan Zheng } 35765d4f98a2SYan Zheng 35775d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 35785d4f98a2SYan Zheng struct btrfs_file_extent_item); 35795d4f98a2SYan Zheng 35805d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 35815d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 35825d4f98a2SYan Zheng goto next; 35835d4f98a2SYan Zheng 35845d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 35855d4f98a2SYan Zheng extent_key->objectid) 35865d4f98a2SYan Zheng goto next; 35875d4f98a2SYan Zheng 35885d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 35895d4f98a2SYan Zheng if (key.offset != ref_offset) 35905d4f98a2SYan Zheng goto next; 35915d4f98a2SYan Zheng 35925d4f98a2SYan Zheng if (counted) 35935d4f98a2SYan Zheng ref_count--; 35945d4f98a2SYan Zheng if (added) 35955d4f98a2SYan Zheng goto next; 35965d4f98a2SYan Zheng 35975d4f98a2SYan Zheng if (!tree_block_processed(leaf->start, leaf->len, rc)) { 35985d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 35995d4f98a2SYan Zheng if (!block) { 36005d4f98a2SYan Zheng err = -ENOMEM; 36015d4f98a2SYan Zheng break; 36025d4f98a2SYan Zheng } 36035d4f98a2SYan Zheng block->bytenr = leaf->start; 36045d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 36055d4f98a2SYan Zheng block->level = 0; 36065d4f98a2SYan Zheng block->key_ready = 1; 36075d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 36085d4f98a2SYan Zheng &block->rb_node); 360943c04fb1SJeff Mahoney if (rb_node) 361043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 361143c04fb1SJeff Mahoney block->bytenr); 36125d4f98a2SYan Zheng } 36135d4f98a2SYan Zheng if (counted) 36145d4f98a2SYan Zheng added = 1; 36155d4f98a2SYan Zheng else 36165d4f98a2SYan Zheng path->slots[0] = nritems; 36175d4f98a2SYan Zheng next: 36185d4f98a2SYan Zheng path->slots[0]++; 36195d4f98a2SYan Zheng 36205d4f98a2SYan Zheng } 36215d4f98a2SYan Zheng out: 36225d4f98a2SYan Zheng btrfs_free_path(path); 36235d4f98a2SYan Zheng return err; 36245d4f98a2SYan Zheng } 36255d4f98a2SYan Zheng 36265d4f98a2SYan Zheng /* 36272c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 36285d4f98a2SYan Zheng */ 36295d4f98a2SYan Zheng static noinline_for_stack 36305d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 36315d4f98a2SYan Zheng struct btrfs_key *extent_key, 36325d4f98a2SYan Zheng struct btrfs_path *path, 36335d4f98a2SYan Zheng struct rb_root *blocks) 36345d4f98a2SYan Zheng { 36355d4f98a2SYan Zheng struct btrfs_key key; 36365d4f98a2SYan Zheng struct extent_buffer *eb; 36375d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 36385d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 36395d4f98a2SYan Zheng unsigned long ptr; 36405d4f98a2SYan Zheng unsigned long end; 36413fd0a558SYan, Zheng u32 blocksize = btrfs_level_size(rc->extent_root, 0); 3642647f63bdSFilipe David Borba Manana int ret = 0; 36435d4f98a2SYan Zheng int err = 0; 36445d4f98a2SYan Zheng 36455d4f98a2SYan Zheng eb = path->nodes[0]; 36465d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 36475d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 36485d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 36495d4f98a2SYan Zheng if (ptr + sizeof(struct btrfs_extent_item_v0) == end) 36505d4f98a2SYan Zheng ptr = end; 36515d4f98a2SYan Zheng else 36525d4f98a2SYan Zheng #endif 36535d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 36545d4f98a2SYan Zheng 36555d4f98a2SYan Zheng while (ptr < end) { 36565d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 36575d4f98a2SYan Zheng key.type = btrfs_extent_inline_ref_type(eb, iref); 36585d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 36595d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 36605d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 36615d4f98a2SYan Zheng blocks); 36625d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 36635d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 36645d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 36655d4f98a2SYan Zheng eb, dref, blocks); 36665d4f98a2SYan Zheng } else { 36675d4f98a2SYan Zheng BUG(); 36685d4f98a2SYan Zheng } 3669647f63bdSFilipe David Borba Manana if (ret) { 3670647f63bdSFilipe David Borba Manana err = ret; 3671647f63bdSFilipe David Borba Manana goto out; 3672647f63bdSFilipe David Borba Manana } 36735d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 36745d4f98a2SYan Zheng } 36755d4f98a2SYan Zheng WARN_ON(ptr > end); 36765d4f98a2SYan Zheng 36775d4f98a2SYan Zheng while (1) { 36785d4f98a2SYan Zheng cond_resched(); 36795d4f98a2SYan Zheng eb = path->nodes[0]; 36805d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 36815d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 36825d4f98a2SYan Zheng if (ret < 0) { 36835d4f98a2SYan Zheng err = ret; 36845d4f98a2SYan Zheng break; 36855d4f98a2SYan Zheng } 36865d4f98a2SYan Zheng if (ret > 0) 36875d4f98a2SYan Zheng break; 36885d4f98a2SYan Zheng eb = path->nodes[0]; 36895d4f98a2SYan Zheng } 36905d4f98a2SYan Zheng 36915d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 36925d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 36935d4f98a2SYan Zheng break; 36945d4f98a2SYan Zheng 36955d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 36965d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY || 36975d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 36985d4f98a2SYan Zheng #else 36995d4f98a2SYan Zheng BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); 37005d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37015d4f98a2SYan Zheng #endif 37025d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37035d4f98a2SYan Zheng blocks); 37045d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37055d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 37065d4f98a2SYan Zheng struct btrfs_extent_data_ref); 37075d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37085d4f98a2SYan Zheng eb, dref, blocks); 37095d4f98a2SYan Zheng } else { 37105d4f98a2SYan Zheng ret = 0; 37115d4f98a2SYan Zheng } 37125d4f98a2SYan Zheng if (ret) { 37135d4f98a2SYan Zheng err = ret; 37145d4f98a2SYan Zheng break; 37155d4f98a2SYan Zheng } 37165d4f98a2SYan Zheng path->slots[0]++; 37175d4f98a2SYan Zheng } 3718647f63bdSFilipe David Borba Manana out: 3719b3b4aa74SDavid Sterba btrfs_release_path(path); 37205d4f98a2SYan Zheng if (err) 37215d4f98a2SYan Zheng free_block_list(blocks); 37225d4f98a2SYan Zheng return err; 37235d4f98a2SYan Zheng } 37245d4f98a2SYan Zheng 37255d4f98a2SYan Zheng /* 37262c016dc2SLiu Bo * helper to find next unprocessed extent 37275d4f98a2SYan Zheng */ 37285d4f98a2SYan Zheng static noinline_for_stack 37295d4f98a2SYan Zheng int find_next_extent(struct btrfs_trans_handle *trans, 37303fd0a558SYan, Zheng struct reloc_control *rc, struct btrfs_path *path, 37313fd0a558SYan, Zheng struct btrfs_key *extent_key) 37325d4f98a2SYan Zheng { 37335d4f98a2SYan Zheng struct btrfs_key key; 37345d4f98a2SYan Zheng struct extent_buffer *leaf; 37355d4f98a2SYan Zheng u64 start, end, last; 37365d4f98a2SYan Zheng int ret; 37375d4f98a2SYan Zheng 37385d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 37395d4f98a2SYan Zheng while (1) { 37405d4f98a2SYan Zheng cond_resched(); 37415d4f98a2SYan Zheng if (rc->search_start >= last) { 37425d4f98a2SYan Zheng ret = 1; 37435d4f98a2SYan Zheng break; 37445d4f98a2SYan Zheng } 37455d4f98a2SYan Zheng 37465d4f98a2SYan Zheng key.objectid = rc->search_start; 37475d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 37485d4f98a2SYan Zheng key.offset = 0; 37495d4f98a2SYan Zheng 37505d4f98a2SYan Zheng path->search_commit_root = 1; 37515d4f98a2SYan Zheng path->skip_locking = 1; 37525d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 37535d4f98a2SYan Zheng 0, 0); 37545d4f98a2SYan Zheng if (ret < 0) 37555d4f98a2SYan Zheng break; 37565d4f98a2SYan Zheng next: 37575d4f98a2SYan Zheng leaf = path->nodes[0]; 37585d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 37595d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 37605d4f98a2SYan Zheng if (ret != 0) 37615d4f98a2SYan Zheng break; 37625d4f98a2SYan Zheng leaf = path->nodes[0]; 37635d4f98a2SYan Zheng } 37645d4f98a2SYan Zheng 37655d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 37665d4f98a2SYan Zheng if (key.objectid >= last) { 37675d4f98a2SYan Zheng ret = 1; 37685d4f98a2SYan Zheng break; 37695d4f98a2SYan Zheng } 37705d4f98a2SYan Zheng 37713173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 37723173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 37733173a18fSJosef Bacik path->slots[0]++; 37743173a18fSJosef Bacik goto next; 37753173a18fSJosef Bacik } 37763173a18fSJosef Bacik 37773173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 37785d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 37795d4f98a2SYan Zheng path->slots[0]++; 37805d4f98a2SYan Zheng goto next; 37815d4f98a2SYan Zheng } 37825d4f98a2SYan Zheng 37833173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 37843173a18fSJosef Bacik key.objectid + rc->extent_root->leafsize <= 37853173a18fSJosef Bacik rc->search_start) { 37863173a18fSJosef Bacik path->slots[0]++; 37873173a18fSJosef Bacik goto next; 37883173a18fSJosef Bacik } 37893173a18fSJosef Bacik 37905d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 37915d4f98a2SYan Zheng key.objectid, &start, &end, 3792e6138876SJosef Bacik EXTENT_DIRTY, NULL); 37935d4f98a2SYan Zheng 37945d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3795b3b4aa74SDavid Sterba btrfs_release_path(path); 37965d4f98a2SYan Zheng rc->search_start = end + 1; 37975d4f98a2SYan Zheng } else { 37983173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 37995d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 38003173a18fSJosef Bacik else 38013173a18fSJosef Bacik rc->search_start = key.objectid + 38023173a18fSJosef Bacik rc->extent_root->leafsize; 38033fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 38045d4f98a2SYan Zheng return 0; 38055d4f98a2SYan Zheng } 38065d4f98a2SYan Zheng } 3807b3b4aa74SDavid Sterba btrfs_release_path(path); 38085d4f98a2SYan Zheng return ret; 38095d4f98a2SYan Zheng } 38105d4f98a2SYan Zheng 38115d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 38125d4f98a2SYan Zheng { 38135d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38147585717fSChris Mason 38157585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38165d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 38177585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38185d4f98a2SYan Zheng } 38195d4f98a2SYan Zheng 38205d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 38215d4f98a2SYan Zheng { 38225d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38237585717fSChris Mason 38247585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38255d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 38267585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38275d4f98a2SYan Zheng } 38285d4f98a2SYan Zheng 38295d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 38305d4f98a2SYan Zheng { 38315d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38325d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38335d4f98a2SYan Zheng return 1; 38345d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 38355d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38365d4f98a2SYan Zheng return 1; 38375d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38385d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 38395d4f98a2SYan Zheng return 1; 38405d4f98a2SYan Zheng return 0; 38415d4f98a2SYan Zheng } 38425d4f98a2SYan Zheng 38433fd0a558SYan, Zheng static noinline_for_stack 38443fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 38453fd0a558SYan, Zheng { 38463fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 38473fd0a558SYan, Zheng int ret; 38483fd0a558SYan, Zheng 384966d8f3ddSMiao Xie rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root, 385066d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 38513fd0a558SYan, Zheng if (!rc->block_rsv) 38523fd0a558SYan, Zheng return -ENOMEM; 38533fd0a558SYan, Zheng 38543fd0a558SYan, Zheng /* 38553fd0a558SYan, Zheng * reserve some space for creating reloc trees. 38563fd0a558SYan, Zheng * btrfs_init_reloc_root will use them when there 38573fd0a558SYan, Zheng * is no reservation in transaction handle. 38583fd0a558SYan, Zheng */ 38594a92b1b8SJosef Bacik ret = btrfs_block_rsv_add(rc->extent_root, rc->block_rsv, 386008e007d2SMiao Xie rc->extent_root->nodesize * 256, 386108e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 38623fd0a558SYan, Zheng if (ret) 38633fd0a558SYan, Zheng return ret; 38643fd0a558SYan, Zheng 38653fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 38663fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 38673fd0a558SYan, Zheng rc->extents_found = 0; 38683fd0a558SYan, Zheng rc->nodes_relocated = 0; 38693fd0a558SYan, Zheng rc->merging_rsv_size = 0; 38703fd0a558SYan, Zheng 38713fd0a558SYan, Zheng rc->create_reloc_tree = 1; 38723fd0a558SYan, Zheng set_reloc_control(rc); 38733fd0a558SYan, Zheng 38747a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 387528818947SLiu Bo if (IS_ERR(trans)) { 387628818947SLiu Bo unset_reloc_control(rc); 387728818947SLiu Bo /* 387828818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 387928818947SLiu Bo * cleanup. And callers are responsible to free the above 388028818947SLiu Bo * block rsv. 388128818947SLiu Bo */ 388228818947SLiu Bo return PTR_ERR(trans); 388328818947SLiu Bo } 38843fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 38853fd0a558SYan, Zheng return 0; 38863fd0a558SYan, Zheng } 388776dda93cSYan, Zheng 38885d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 38895d4f98a2SYan Zheng { 38905d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 38915d4f98a2SYan Zheng struct btrfs_key key; 38925d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 38935d4f98a2SYan Zheng struct btrfs_path *path; 38945d4f98a2SYan Zheng struct btrfs_extent_item *ei; 38955d4f98a2SYan Zheng u64 flags; 38965d4f98a2SYan Zheng u32 item_size; 38975d4f98a2SYan Zheng int ret; 38985d4f98a2SYan Zheng int err = 0; 3899c87f08caSChris Mason int progress = 0; 39005d4f98a2SYan Zheng 39015d4f98a2SYan Zheng path = btrfs_alloc_path(); 39023fd0a558SYan, Zheng if (!path) 39035d4f98a2SYan Zheng return -ENOMEM; 3904026fd317SJosef Bacik path->reada = 1; 39053fd0a558SYan, Zheng 39063fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 39073fd0a558SYan, Zheng if (ret) { 39083fd0a558SYan, Zheng err = ret; 39093fd0a558SYan, Zheng goto out_free; 39102423fdfbSJiri Slaby } 39115d4f98a2SYan Zheng 39125d4f98a2SYan Zheng while (1) { 3913c87f08caSChris Mason progress++; 3914a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 39150f788c58SLiu Bo if (IS_ERR(trans)) { 39160f788c58SLiu Bo err = PTR_ERR(trans); 39170f788c58SLiu Bo trans = NULL; 39180f788c58SLiu Bo break; 39190f788c58SLiu Bo } 3920c87f08caSChris Mason restart: 39213fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 39223fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 39233fd0a558SYan, Zheng continue; 39243fd0a558SYan, Zheng } 39253fd0a558SYan, Zheng 39263fd0a558SYan, Zheng ret = find_next_extent(trans, rc, path, &key); 39275d4f98a2SYan Zheng if (ret < 0) 39285d4f98a2SYan Zheng err = ret; 39295d4f98a2SYan Zheng if (ret != 0) 39305d4f98a2SYan Zheng break; 39315d4f98a2SYan Zheng 39325d4f98a2SYan Zheng rc->extents_found++; 39335d4f98a2SYan Zheng 39345d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 39355d4f98a2SYan Zheng struct btrfs_extent_item); 39363fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 39375d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 39385d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 39395d4f98a2SYan Zheng ret = check_extent_flags(flags); 39405d4f98a2SYan Zheng BUG_ON(ret); 39415d4f98a2SYan Zheng 39425d4f98a2SYan Zheng } else { 39435d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 39445d4f98a2SYan Zheng u64 ref_owner; 39455d4f98a2SYan Zheng int path_change = 0; 39465d4f98a2SYan Zheng 39475d4f98a2SYan Zheng BUG_ON(item_size != 39485d4f98a2SYan Zheng sizeof(struct btrfs_extent_item_v0)); 39495d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, &key, &ref_owner, 39505d4f98a2SYan Zheng &path_change); 39515d4f98a2SYan Zheng if (ref_owner < BTRFS_FIRST_FREE_OBJECTID) 39525d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_TREE_BLOCK; 39535d4f98a2SYan Zheng else 39545d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_DATA; 39555d4f98a2SYan Zheng 39565d4f98a2SYan Zheng if (path_change) { 3957b3b4aa74SDavid Sterba btrfs_release_path(path); 39585d4f98a2SYan Zheng 39595d4f98a2SYan Zheng path->search_commit_root = 1; 39605d4f98a2SYan Zheng path->skip_locking = 1; 39615d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, 39625d4f98a2SYan Zheng &key, path, 0, 0); 39635d4f98a2SYan Zheng if (ret < 0) { 39645d4f98a2SYan Zheng err = ret; 39655d4f98a2SYan Zheng break; 39665d4f98a2SYan Zheng } 39675d4f98a2SYan Zheng BUG_ON(ret > 0); 39685d4f98a2SYan Zheng } 39695d4f98a2SYan Zheng #else 39705d4f98a2SYan Zheng BUG(); 39715d4f98a2SYan Zheng #endif 39725d4f98a2SYan Zheng } 39735d4f98a2SYan Zheng 39745d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 39755d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 39765d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 39775d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 39785d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 39795d4f98a2SYan Zheng } else { 3980b3b4aa74SDavid Sterba btrfs_release_path(path); 39815d4f98a2SYan Zheng ret = 0; 39825d4f98a2SYan Zheng } 39835d4f98a2SYan Zheng if (ret < 0) { 39843fd0a558SYan, Zheng err = ret; 39855d4f98a2SYan Zheng break; 39865d4f98a2SYan Zheng } 39875d4f98a2SYan Zheng 39885d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 39895d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 39905d4f98a2SYan Zheng if (ret < 0) { 39913fd0a558SYan, Zheng if (ret != -EAGAIN) { 39925d4f98a2SYan Zheng err = ret; 39935d4f98a2SYan Zheng break; 39945d4f98a2SYan Zheng } 39953fd0a558SYan, Zheng rc->extents_found--; 39963fd0a558SYan, Zheng rc->search_start = key.objectid; 39973fd0a558SYan, Zheng } 39985d4f98a2SYan Zheng } 39995d4f98a2SYan Zheng 40003fd0a558SYan, Zheng if (rc->commit_transaction) { 40013fd0a558SYan, Zheng rc->commit_transaction = 0; 40023fd0a558SYan, Zheng ret = btrfs_commit_transaction(trans, rc->extent_root); 40033fd0a558SYan, Zheng BUG_ON(ret); 40043fd0a558SYan, Zheng } else { 40053fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 4006b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 40073fd0a558SYan, Zheng } 40083fd0a558SYan, Zheng trans = NULL; 40095d4f98a2SYan Zheng 40105d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 40115d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40125d4f98a2SYan Zheng rc->found_file_extent = 1; 40130257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 40143fd0a558SYan, Zheng &key, &rc->cluster); 40155d4f98a2SYan Zheng if (ret < 0) { 40165d4f98a2SYan Zheng err = ret; 40175d4f98a2SYan Zheng break; 40185d4f98a2SYan Zheng } 40195d4f98a2SYan Zheng } 40205d4f98a2SYan Zheng } 4021c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 4022c87f08caSChris Mason ret = btrfs_force_chunk_alloc(trans, rc->extent_root, 4023c87f08caSChris Mason rc->block_group->flags); 4024c87f08caSChris Mason if (ret == 0) { 4025c87f08caSChris Mason err = 0; 4026c87f08caSChris Mason progress = 0; 4027c87f08caSChris Mason goto restart; 4028c87f08caSChris Mason } 4029c87f08caSChris Mason } 40303fd0a558SYan, Zheng 4031b3b4aa74SDavid Sterba btrfs_release_path(path); 40323fd0a558SYan, Zheng clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY, 40333fd0a558SYan, Zheng GFP_NOFS); 40345d4f98a2SYan Zheng 40355d4f98a2SYan Zheng if (trans) { 40363fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 4037b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 40385d4f98a2SYan Zheng } 40395d4f98a2SYan Zheng 40400257bb82SYan, Zheng if (!err) { 40413fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 40423fd0a558SYan, Zheng &rc->cluster); 40430257bb82SYan, Zheng if (ret < 0) 40440257bb82SYan, Zheng err = ret; 40450257bb82SYan, Zheng } 40460257bb82SYan, Zheng 40473fd0a558SYan, Zheng rc->create_reloc_tree = 0; 40483fd0a558SYan, Zheng set_reloc_control(rc); 40490257bb82SYan, Zheng 40503fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 40513fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 40525d4f98a2SYan Zheng 40533fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 40545d4f98a2SYan Zheng 40555d4f98a2SYan Zheng merge_reloc_roots(rc); 40565d4f98a2SYan Zheng 40573fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 40585d4f98a2SYan Zheng unset_reloc_control(rc); 40593fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 40605d4f98a2SYan Zheng 40615d4f98a2SYan Zheng /* get rid of pinned extents */ 40627a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 40633612b495STsutomu Itoh if (IS_ERR(trans)) 40643612b495STsutomu Itoh err = PTR_ERR(trans); 40653612b495STsutomu Itoh else 40665d4f98a2SYan Zheng btrfs_commit_transaction(trans, rc->extent_root); 40673fd0a558SYan, Zheng out_free: 40683fd0a558SYan, Zheng btrfs_free_block_rsv(rc->extent_root, rc->block_rsv); 40693fd0a558SYan, Zheng btrfs_free_path(path); 40705d4f98a2SYan Zheng return err; 40715d4f98a2SYan Zheng } 40725d4f98a2SYan Zheng 40735d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 40740257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 40755d4f98a2SYan Zheng { 40765d4f98a2SYan Zheng struct btrfs_path *path; 40775d4f98a2SYan Zheng struct btrfs_inode_item *item; 40785d4f98a2SYan Zheng struct extent_buffer *leaf; 40795d4f98a2SYan Zheng int ret; 40805d4f98a2SYan Zheng 40815d4f98a2SYan Zheng path = btrfs_alloc_path(); 40825d4f98a2SYan Zheng if (!path) 40835d4f98a2SYan Zheng return -ENOMEM; 40845d4f98a2SYan Zheng 40855d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 40865d4f98a2SYan Zheng if (ret) 40875d4f98a2SYan Zheng goto out; 40885d4f98a2SYan Zheng 40895d4f98a2SYan Zheng leaf = path->nodes[0]; 40905d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 40915d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); 40925d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 40930257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 40945d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 40953fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 40963fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 40975d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 4098b3b4aa74SDavid Sterba btrfs_release_path(path); 40995d4f98a2SYan Zheng out: 41005d4f98a2SYan Zheng btrfs_free_path(path); 41015d4f98a2SYan Zheng return ret; 41025d4f98a2SYan Zheng } 41035d4f98a2SYan Zheng 41045d4f98a2SYan Zheng /* 41055d4f98a2SYan Zheng * helper to create inode for data relocation. 41065d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 41075d4f98a2SYan Zheng */ 41083fd0a558SYan, Zheng static noinline_for_stack 41093fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 41105d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 41115d4f98a2SYan Zheng { 41125d4f98a2SYan Zheng struct inode *inode = NULL; 41135d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41145d4f98a2SYan Zheng struct btrfs_root *root; 41155d4f98a2SYan Zheng struct btrfs_key key; 41165d4f98a2SYan Zheng u64 objectid = BTRFS_FIRST_FREE_OBJECTID; 41175d4f98a2SYan Zheng int err = 0; 41185d4f98a2SYan Zheng 41195d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 41205d4f98a2SYan Zheng if (IS_ERR(root)) 41215d4f98a2SYan Zheng return ERR_CAST(root); 41225d4f98a2SYan Zheng 4123a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 41243fd0a558SYan, Zheng if (IS_ERR(trans)) 41253fd0a558SYan, Zheng return ERR_CAST(trans); 41265d4f98a2SYan Zheng 4127581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 41285d4f98a2SYan Zheng if (err) 41295d4f98a2SYan Zheng goto out; 41305d4f98a2SYan Zheng 41310257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 41325d4f98a2SYan Zheng BUG_ON(err); 41335d4f98a2SYan Zheng 41345d4f98a2SYan Zheng key.objectid = objectid; 41355d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 41365d4f98a2SYan Zheng key.offset = 0; 413773f73415SJosef Bacik inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); 41385d4f98a2SYan Zheng BUG_ON(IS_ERR(inode) || is_bad_inode(inode)); 41395d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 41405d4f98a2SYan Zheng 41415d4f98a2SYan Zheng err = btrfs_orphan_add(trans, inode); 41425d4f98a2SYan Zheng out: 41435d4f98a2SYan Zheng btrfs_end_transaction(trans, root); 4144b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 41455d4f98a2SYan Zheng if (err) { 41465d4f98a2SYan Zheng if (inode) 41475d4f98a2SYan Zheng iput(inode); 41485d4f98a2SYan Zheng inode = ERR_PTR(err); 41495d4f98a2SYan Zheng } 41505d4f98a2SYan Zheng return inode; 41515d4f98a2SYan Zheng } 41525d4f98a2SYan Zheng 4153a9995eecSJosef Bacik static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 41543fd0a558SYan, Zheng { 41553fd0a558SYan, Zheng struct reloc_control *rc; 41563fd0a558SYan, Zheng 41573fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 41583fd0a558SYan, Zheng if (!rc) 41593fd0a558SYan, Zheng return NULL; 41603fd0a558SYan, Zheng 41613fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 41623fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 41633fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 4164a9995eecSJosef Bacik extent_io_tree_init(&rc->processed_blocks, 4165a9995eecSJosef Bacik fs_info->btree_inode->i_mapping); 41663fd0a558SYan, Zheng return rc; 41673fd0a558SYan, Zheng } 41683fd0a558SYan, Zheng 41695d4f98a2SYan Zheng /* 41705d4f98a2SYan Zheng * function to relocate all extents in a block group. 41715d4f98a2SYan Zheng */ 41725d4f98a2SYan Zheng int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start) 41735d4f98a2SYan Zheng { 41745d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = extent_root->fs_info; 41755d4f98a2SYan Zheng struct reloc_control *rc; 41760af3d00bSJosef Bacik struct inode *inode; 41770af3d00bSJosef Bacik struct btrfs_path *path; 41785d4f98a2SYan Zheng int ret; 4179f0486c68SYan, Zheng int rw = 0; 41805d4f98a2SYan Zheng int err = 0; 41815d4f98a2SYan Zheng 4182a9995eecSJosef Bacik rc = alloc_reloc_control(fs_info); 41835d4f98a2SYan Zheng if (!rc) 41845d4f98a2SYan Zheng return -ENOMEM; 41855d4f98a2SYan Zheng 4186f0486c68SYan, Zheng rc->extent_root = extent_root; 41873fd0a558SYan, Zheng 41885d4f98a2SYan Zheng rc->block_group = btrfs_lookup_block_group(fs_info, group_start); 41895d4f98a2SYan Zheng BUG_ON(!rc->block_group); 41905d4f98a2SYan Zheng 4191f0486c68SYan, Zheng if (!rc->block_group->ro) { 4192f0486c68SYan, Zheng ret = btrfs_set_block_group_ro(extent_root, rc->block_group); 4193f0486c68SYan, Zheng if (ret) { 4194f0486c68SYan, Zheng err = ret; 4195f0486c68SYan, Zheng goto out; 4196f0486c68SYan, Zheng } 4197f0486c68SYan, Zheng rw = 1; 4198f0486c68SYan, Zheng } 4199f0486c68SYan, Zheng 42000af3d00bSJosef Bacik path = btrfs_alloc_path(); 42010af3d00bSJosef Bacik if (!path) { 42020af3d00bSJosef Bacik err = -ENOMEM; 42030af3d00bSJosef Bacik goto out; 42040af3d00bSJosef Bacik } 42050af3d00bSJosef Bacik 42060af3d00bSJosef Bacik inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group, 42070af3d00bSJosef Bacik path); 42080af3d00bSJosef Bacik btrfs_free_path(path); 42090af3d00bSJosef Bacik 42100af3d00bSJosef Bacik if (!IS_ERR(inode)) 42110af3d00bSJosef Bacik ret = delete_block_group_cache(fs_info, inode, 0); 42120af3d00bSJosef Bacik else 42130af3d00bSJosef Bacik ret = PTR_ERR(inode); 42140af3d00bSJosef Bacik 42150af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 42160af3d00bSJosef Bacik err = ret; 42170af3d00bSJosef Bacik goto out; 42180af3d00bSJosef Bacik } 42190af3d00bSJosef Bacik 42205d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 42215d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 42225d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 42235d4f98a2SYan Zheng rc->data_inode = NULL; 42245d4f98a2SYan Zheng goto out; 42255d4f98a2SYan Zheng } 42265d4f98a2SYan Zheng 42275d4f98a2SYan Zheng printk(KERN_INFO "btrfs: relocating block group %llu flags %llu\n", 4228c1c9ff7cSGeert Uytterhoeven rc->block_group->key.objectid, rc->block_group->flags); 42295d4f98a2SYan Zheng 4230eb73c1b7SMiao Xie ret = btrfs_start_all_delalloc_inodes(fs_info, 0); 42318ccf6f19SMiao Xie if (ret < 0) { 42328ccf6f19SMiao Xie err = ret; 42338ccf6f19SMiao Xie goto out; 42348ccf6f19SMiao Xie } 4235f0de181cSJosef Bacik btrfs_wait_all_ordered_extents(fs_info); 42365d4f98a2SYan Zheng 42375d4f98a2SYan Zheng while (1) { 423876dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 42395d4f98a2SYan Zheng ret = relocate_block_group(rc); 424076dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 42415d4f98a2SYan Zheng if (ret < 0) { 42425d4f98a2SYan Zheng err = ret; 42433fd0a558SYan, Zheng goto out; 42445d4f98a2SYan Zheng } 42455d4f98a2SYan Zheng 42465d4f98a2SYan Zheng if (rc->extents_found == 0) 42475d4f98a2SYan Zheng break; 42485d4f98a2SYan Zheng 42495d4f98a2SYan Zheng printk(KERN_INFO "btrfs: found %llu extents\n", 4250c1c9ff7cSGeert Uytterhoeven rc->extents_found); 42515d4f98a2SYan Zheng 42525d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 42530ef8b726SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 42540ef8b726SJosef Bacik (u64)-1); 42550ef8b726SJosef Bacik if (ret) { 42560ef8b726SJosef Bacik err = ret; 42570ef8b726SJosef Bacik goto out; 42580ef8b726SJosef Bacik } 42595d4f98a2SYan Zheng invalidate_mapping_pages(rc->data_inode->i_mapping, 42605d4f98a2SYan Zheng 0, -1); 42615d4f98a2SYan Zheng rc->stage = UPDATE_DATA_PTRS; 42625d4f98a2SYan Zheng } 42635d4f98a2SYan Zheng } 42645d4f98a2SYan Zheng 42650257bb82SYan, Zheng filemap_write_and_wait_range(fs_info->btree_inode->i_mapping, 42665d4f98a2SYan Zheng rc->block_group->key.objectid, 42675d4f98a2SYan Zheng rc->block_group->key.objectid + 42685d4f98a2SYan Zheng rc->block_group->key.offset - 1); 42695d4f98a2SYan Zheng 42705d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 42715d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 42725d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 42735d4f98a2SYan Zheng out: 4274f0486c68SYan, Zheng if (err && rw) 4275f0486c68SYan, Zheng btrfs_set_block_group_rw(extent_root, rc->block_group); 42765d4f98a2SYan Zheng iput(rc->data_inode); 42775d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 42785d4f98a2SYan Zheng kfree(rc); 42795d4f98a2SYan Zheng return err; 42805d4f98a2SYan Zheng } 42815d4f98a2SYan Zheng 428276dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 428376dda93cSYan, Zheng { 428476dda93cSYan, Zheng struct btrfs_trans_handle *trans; 428579787eaaSJeff Mahoney int ret, err; 428676dda93cSYan, Zheng 4287a22285a6SYan, Zheng trans = btrfs_start_transaction(root->fs_info->tree_root, 0); 428879787eaaSJeff Mahoney if (IS_ERR(trans)) 428979787eaaSJeff Mahoney return PTR_ERR(trans); 429076dda93cSYan, Zheng 429176dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 429276dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 429376dda93cSYan, Zheng root->root_item.drop_level = 0; 429476dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 429576dda93cSYan, Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 429676dda93cSYan, Zheng &root->root_key, &root->root_item); 429776dda93cSYan, Zheng 429879787eaaSJeff Mahoney err = btrfs_end_transaction(trans, root->fs_info->tree_root); 429979787eaaSJeff Mahoney if (err) 430079787eaaSJeff Mahoney return err; 430179787eaaSJeff Mahoney return ret; 430276dda93cSYan, Zheng } 430376dda93cSYan, Zheng 43045d4f98a2SYan Zheng /* 43055d4f98a2SYan Zheng * recover relocation interrupted by system crash. 43065d4f98a2SYan Zheng * 43075d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 43085d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 43095d4f98a2SYan Zheng */ 43105d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 43115d4f98a2SYan Zheng { 43125d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 43135d4f98a2SYan Zheng struct btrfs_key key; 43145d4f98a2SYan Zheng struct btrfs_root *fs_root; 43155d4f98a2SYan Zheng struct btrfs_root *reloc_root; 43165d4f98a2SYan Zheng struct btrfs_path *path; 43175d4f98a2SYan Zheng struct extent_buffer *leaf; 43185d4f98a2SYan Zheng struct reloc_control *rc = NULL; 43195d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 43205d4f98a2SYan Zheng int ret; 43215d4f98a2SYan Zheng int err = 0; 43225d4f98a2SYan Zheng 43235d4f98a2SYan Zheng path = btrfs_alloc_path(); 43245d4f98a2SYan Zheng if (!path) 43255d4f98a2SYan Zheng return -ENOMEM; 4326026fd317SJosef Bacik path->reada = -1; 43275d4f98a2SYan Zheng 43285d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 43295d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 43305d4f98a2SYan Zheng key.offset = (u64)-1; 43315d4f98a2SYan Zheng 43325d4f98a2SYan Zheng while (1) { 43335d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, 43345d4f98a2SYan Zheng path, 0, 0); 43355d4f98a2SYan Zheng if (ret < 0) { 43365d4f98a2SYan Zheng err = ret; 43375d4f98a2SYan Zheng goto out; 43385d4f98a2SYan Zheng } 43395d4f98a2SYan Zheng if (ret > 0) { 43405d4f98a2SYan Zheng if (path->slots[0] == 0) 43415d4f98a2SYan Zheng break; 43425d4f98a2SYan Zheng path->slots[0]--; 43435d4f98a2SYan Zheng } 43445d4f98a2SYan Zheng leaf = path->nodes[0]; 43455d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4346b3b4aa74SDavid Sterba btrfs_release_path(path); 43475d4f98a2SYan Zheng 43485d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 43495d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 43505d4f98a2SYan Zheng break; 43515d4f98a2SYan Zheng 4352cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 43535d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 43545d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 43555d4f98a2SYan Zheng goto out; 43565d4f98a2SYan Zheng } 43575d4f98a2SYan Zheng 43585d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 43595d4f98a2SYan Zheng 43605d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 43615d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 43625d4f98a2SYan Zheng reloc_root->root_key.offset); 43635d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 436476dda93cSYan, Zheng ret = PTR_ERR(fs_root); 436576dda93cSYan, Zheng if (ret != -ENOENT) { 436676dda93cSYan, Zheng err = ret; 43675d4f98a2SYan Zheng goto out; 43685d4f98a2SYan Zheng } 436979787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 437079787eaaSJeff Mahoney if (ret < 0) { 437179787eaaSJeff Mahoney err = ret; 437279787eaaSJeff Mahoney goto out; 437379787eaaSJeff Mahoney } 437476dda93cSYan, Zheng } 43755d4f98a2SYan Zheng } 43765d4f98a2SYan Zheng 43775d4f98a2SYan Zheng if (key.offset == 0) 43785d4f98a2SYan Zheng break; 43795d4f98a2SYan Zheng 43805d4f98a2SYan Zheng key.offset--; 43815d4f98a2SYan Zheng } 4382b3b4aa74SDavid Sterba btrfs_release_path(path); 43835d4f98a2SYan Zheng 43845d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 43855d4f98a2SYan Zheng goto out; 43865d4f98a2SYan Zheng 4387a9995eecSJosef Bacik rc = alloc_reloc_control(root->fs_info); 43885d4f98a2SYan Zheng if (!rc) { 43895d4f98a2SYan Zheng err = -ENOMEM; 43905d4f98a2SYan Zheng goto out; 43915d4f98a2SYan Zheng } 43925d4f98a2SYan Zheng 43935d4f98a2SYan Zheng rc->extent_root = root->fs_info->extent_root; 43945d4f98a2SYan Zheng 43955d4f98a2SYan Zheng set_reloc_control(rc); 43965d4f98a2SYan Zheng 43977a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 43983612b495STsutomu Itoh if (IS_ERR(trans)) { 43993612b495STsutomu Itoh unset_reloc_control(rc); 44003612b495STsutomu Itoh err = PTR_ERR(trans); 44013612b495STsutomu Itoh goto out_free; 44023612b495STsutomu Itoh } 44033fd0a558SYan, Zheng 44043fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 44053fd0a558SYan, Zheng 44065d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 44075d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 44085d4f98a2SYan Zheng struct btrfs_root, root_list); 44095d4f98a2SYan Zheng list_del(&reloc_root->root_list); 44105d4f98a2SYan Zheng 44115d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 44125d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 44135d4f98a2SYan Zheng &rc->reloc_roots); 44145d4f98a2SYan Zheng continue; 44155d4f98a2SYan Zheng } 44165d4f98a2SYan Zheng 44175d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 44185d4f98a2SYan Zheng reloc_root->root_key.offset); 441979787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 442079787eaaSJeff Mahoney err = PTR_ERR(fs_root); 442179787eaaSJeff Mahoney goto out_free; 442279787eaaSJeff Mahoney } 44235d4f98a2SYan Zheng 4424ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 442579787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 44265d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 44275d4f98a2SYan Zheng } 44285d4f98a2SYan Zheng 442979787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 443079787eaaSJeff Mahoney if (err) 443179787eaaSJeff Mahoney goto out_free; 44325d4f98a2SYan Zheng 44335d4f98a2SYan Zheng merge_reloc_roots(rc); 44345d4f98a2SYan Zheng 44355d4f98a2SYan Zheng unset_reloc_control(rc); 44365d4f98a2SYan Zheng 44377a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 44383612b495STsutomu Itoh if (IS_ERR(trans)) 44393612b495STsutomu Itoh err = PTR_ERR(trans); 44403612b495STsutomu Itoh else 444179787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 44423612b495STsutomu Itoh out_free: 44435d4f98a2SYan Zheng kfree(rc); 44443612b495STsutomu Itoh out: 4445aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4446aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4447aca1bba6SLiu Bo 44485d4f98a2SYan Zheng btrfs_free_path(path); 44495d4f98a2SYan Zheng 44505d4f98a2SYan Zheng if (err == 0) { 44515d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 44525d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 44535d4f98a2SYan Zheng BTRFS_DATA_RELOC_TREE_OBJECTID); 44545d4f98a2SYan Zheng if (IS_ERR(fs_root)) 44555d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4456d7ce5843SMiao Xie else 445766b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 44585d4f98a2SYan Zheng } 44595d4f98a2SYan Zheng return err; 44605d4f98a2SYan Zheng } 44615d4f98a2SYan Zheng 44625d4f98a2SYan Zheng /* 44635d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 44645d4f98a2SYan Zheng * 44655d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 44665d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 44675d4f98a2SYan Zheng */ 44685d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 44695d4f98a2SYan Zheng { 44705d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 44715d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 44725d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 44735d4f98a2SYan Zheng int ret; 44745d4f98a2SYan Zheng u64 disk_bytenr; 44754577b014SJosef Bacik u64 new_bytenr; 44765d4f98a2SYan Zheng LIST_HEAD(list); 44775d4f98a2SYan Zheng 44785d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 44795d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 44805d4f98a2SYan Zheng 44815d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 44825d4f98a2SYan Zheng ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr, 4483a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 448479787eaaSJeff Mahoney if (ret) 448579787eaaSJeff Mahoney goto out; 44865d4f98a2SYan Zheng 44875d4f98a2SYan Zheng while (!list_empty(&list)) { 44885d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 44895d4f98a2SYan Zheng list_del_init(&sums->list); 44905d4f98a2SYan Zheng 44914577b014SJosef Bacik /* 44924577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 44934577b014SJosef Bacik * We need to do this because we will read in entire prealloc 44944577b014SJosef Bacik * extents but we may have written to say the middle of the 44954577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 44964577b014SJosef Bacik * the right disk offset. 44974577b014SJosef Bacik * 44984577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 44994577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 45004577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 45014577b014SJosef Bacik * disk length. 45024577b014SJosef Bacik */ 45034577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 45044577b014SJosef Bacik sums->bytenr = new_bytenr; 45055d4f98a2SYan Zheng 45065d4f98a2SYan Zheng btrfs_add_ordered_sum(inode, ordered, sums); 45075d4f98a2SYan Zheng } 450879787eaaSJeff Mahoney out: 45095d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4510411fc6bcSAndi Kleen return ret; 45115d4f98a2SYan Zheng } 45123fd0a558SYan, Zheng 451383d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 45143fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 45153fd0a558SYan, Zheng struct extent_buffer *cow) 45163fd0a558SYan, Zheng { 45173fd0a558SYan, Zheng struct reloc_control *rc; 45183fd0a558SYan, Zheng struct backref_node *node; 45193fd0a558SYan, Zheng int first_cow = 0; 45203fd0a558SYan, Zheng int level; 452183d4cfd4SJosef Bacik int ret = 0; 45223fd0a558SYan, Zheng 45233fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 45243fd0a558SYan, Zheng if (!rc) 452583d4cfd4SJosef Bacik return 0; 45263fd0a558SYan, Zheng 45273fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 45283fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 45293fd0a558SYan, Zheng 45303fd0a558SYan, Zheng level = btrfs_header_level(buf); 45313fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 45323fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 45333fd0a558SYan, Zheng first_cow = 1; 45343fd0a558SYan, Zheng 45353fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 45363fd0a558SYan, Zheng rc->create_reloc_tree) { 45373fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 45383fd0a558SYan, Zheng 45393fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 45403fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 45413fd0a558SYan, Zheng node->new_bytenr != buf->start); 45423fd0a558SYan, Zheng 45433fd0a558SYan, Zheng drop_node_buffer(node); 45443fd0a558SYan, Zheng extent_buffer_get(cow); 45453fd0a558SYan, Zheng node->eb = cow; 45463fd0a558SYan, Zheng node->new_bytenr = cow->start; 45473fd0a558SYan, Zheng 45483fd0a558SYan, Zheng if (!node->pending) { 45493fd0a558SYan, Zheng list_move_tail(&node->list, 45503fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 45513fd0a558SYan, Zheng node->pending = 1; 45523fd0a558SYan, Zheng } 45533fd0a558SYan, Zheng 45543fd0a558SYan, Zheng if (first_cow) 45553fd0a558SYan, Zheng __mark_block_processed(rc, node); 45563fd0a558SYan, Zheng 45573fd0a558SYan, Zheng if (first_cow && level > 0) 45583fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 45593fd0a558SYan, Zheng } 45603fd0a558SYan, Zheng 456183d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 45623fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 456383d4cfd4SJosef Bacik return ret; 45643fd0a558SYan, Zheng } 45653fd0a558SYan, Zheng 45663fd0a558SYan, Zheng /* 45673fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 45683fd0a558SYan, Zheng * requried for relocating tree blocks in the snapshot 45693fd0a558SYan, Zheng */ 45703fd0a558SYan, Zheng void btrfs_reloc_pre_snapshot(struct btrfs_trans_handle *trans, 45713fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending, 45723fd0a558SYan, Zheng u64 *bytes_to_reserve) 45733fd0a558SYan, Zheng { 45743fd0a558SYan, Zheng struct btrfs_root *root; 45753fd0a558SYan, Zheng struct reloc_control *rc; 45763fd0a558SYan, Zheng 45773fd0a558SYan, Zheng root = pending->root; 45783fd0a558SYan, Zheng if (!root->reloc_root) 45793fd0a558SYan, Zheng return; 45803fd0a558SYan, Zheng 45813fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 45823fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 45833fd0a558SYan, Zheng return; 45843fd0a558SYan, Zheng 45853fd0a558SYan, Zheng root = root->reloc_root; 45863fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 45873fd0a558SYan, Zheng /* 45883fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 45893fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 45903fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 45913fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 45923fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 45933fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 45943fd0a558SYan, Zheng * tree may use more space than it frees. so we need 45953fd0a558SYan, Zheng * reserve extra space. 45963fd0a558SYan, Zheng */ 45973fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 45983fd0a558SYan, Zheng } 45993fd0a558SYan, Zheng 46003fd0a558SYan, Zheng /* 46013fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 46023fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 46033fd0a558SYan, Zheng */ 460449b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 46053fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 46063fd0a558SYan, Zheng { 46073fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 46083fd0a558SYan, Zheng struct btrfs_root *reloc_root; 46093fd0a558SYan, Zheng struct btrfs_root *new_root; 46103fd0a558SYan, Zheng struct reloc_control *rc; 46113fd0a558SYan, Zheng int ret; 46123fd0a558SYan, Zheng 46133fd0a558SYan, Zheng if (!root->reloc_root) 461449b25e05SJeff Mahoney return 0; 46153fd0a558SYan, Zheng 46163fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 46173fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 46183fd0a558SYan, Zheng 46193fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 46203fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 46213fd0a558SYan, Zheng rc->block_rsv, 46223fd0a558SYan, Zheng rc->nodes_relocated); 462349b25e05SJeff Mahoney if (ret) 462449b25e05SJeff Mahoney return ret; 46253fd0a558SYan, Zheng } 46263fd0a558SYan, Zheng 46273fd0a558SYan, Zheng new_root = pending->snap; 46283fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 46293fd0a558SYan, Zheng new_root->root_key.objectid); 463049b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 463149b25e05SJeff Mahoney return PTR_ERR(reloc_root); 46323fd0a558SYan, Zheng 4633ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4634ffd7b339SJeff Mahoney BUG_ON(ret < 0); 46353fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 46363fd0a558SYan, Zheng 463749b25e05SJeff Mahoney if (rc->create_reloc_tree) 46383fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 463949b25e05SJeff Mahoney return ret; 46403fd0a558SYan, Zheng } 4641