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 970647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 985d4f98a2SYan Zheng 995d4f98a2SYan Zheng struct backref_cache { 1005d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 1015d4f98a2SYan Zheng struct rb_root rb_root; 1023fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 1033fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 1043fd0a558SYan, Zheng /* 1053fd0a558SYan, Zheng * list of blocks that have been cowed but some block 1063fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 1073fd0a558SYan, Zheng * new location 1083fd0a558SYan, Zheng */ 1095d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1103fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1113fd0a558SYan, Zheng struct list_head leaves; 1123fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1133fd0a558SYan, Zheng struct list_head changed; 1143fd0a558SYan, Zheng /* list of detached backref node. */ 1153fd0a558SYan, Zheng struct list_head detached; 1163fd0a558SYan, Zheng 1173fd0a558SYan, Zheng u64 last_trans; 1183fd0a558SYan, Zheng 1193fd0a558SYan, Zheng int nr_nodes; 1203fd0a558SYan, Zheng int nr_edges; 1215d4f98a2SYan Zheng }; 1225d4f98a2SYan Zheng 1235d4f98a2SYan Zheng /* 1245d4f98a2SYan Zheng * map address of tree root to tree 1255d4f98a2SYan Zheng */ 1265d4f98a2SYan Zheng struct mapping_node { 1275d4f98a2SYan Zheng struct rb_node rb_node; 1285d4f98a2SYan Zheng u64 bytenr; 1295d4f98a2SYan Zheng void *data; 1305d4f98a2SYan Zheng }; 1315d4f98a2SYan Zheng 1325d4f98a2SYan Zheng struct mapping_tree { 1335d4f98a2SYan Zheng struct rb_root rb_root; 1345d4f98a2SYan Zheng spinlock_t lock; 1355d4f98a2SYan Zheng }; 1365d4f98a2SYan Zheng 1375d4f98a2SYan Zheng /* 1385d4f98a2SYan Zheng * present a tree block to process 1395d4f98a2SYan Zheng */ 1405d4f98a2SYan Zheng struct tree_block { 1415d4f98a2SYan Zheng struct rb_node rb_node; 1425d4f98a2SYan Zheng u64 bytenr; 1435d4f98a2SYan Zheng struct btrfs_key key; 1445d4f98a2SYan Zheng unsigned int level:8; 1455d4f98a2SYan Zheng unsigned int key_ready:1; 1465d4f98a2SYan Zheng }; 1475d4f98a2SYan Zheng 1480257bb82SYan, Zheng #define MAX_EXTENTS 128 1490257bb82SYan, Zheng 1500257bb82SYan, Zheng struct file_extent_cluster { 1510257bb82SYan, Zheng u64 start; 1520257bb82SYan, Zheng u64 end; 1530257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1540257bb82SYan, Zheng unsigned int nr; 1550257bb82SYan, Zheng }; 1560257bb82SYan, Zheng 1575d4f98a2SYan Zheng struct reloc_control { 1585d4f98a2SYan Zheng /* block group to relocate */ 1595d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group; 1605d4f98a2SYan Zheng /* extent tree */ 1615d4f98a2SYan Zheng struct btrfs_root *extent_root; 1625d4f98a2SYan Zheng /* inode for moving data */ 1635d4f98a2SYan Zheng struct inode *data_inode; 1643fd0a558SYan, Zheng 1653fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1663fd0a558SYan, Zheng 1673fd0a558SYan, Zheng struct backref_cache backref_cache; 1683fd0a558SYan, Zheng 1693fd0a558SYan, Zheng struct file_extent_cluster cluster; 1705d4f98a2SYan Zheng /* tree blocks have been processed */ 1715d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1725d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1735d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1745d4f98a2SYan Zheng /* list of reloc trees */ 1755d4f98a2SYan Zheng struct list_head reloc_roots; 1763fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1773fd0a558SYan, Zheng u64 merging_rsv_size; 1783fd0a558SYan, Zheng /* size of relocated tree nodes */ 1793fd0a558SYan, Zheng u64 nodes_relocated; 1800647bf56SWang Shilong /* reserved size for block group relocation*/ 1810647bf56SWang Shilong u64 reserved_bytes; 1823fd0a558SYan, Zheng 1835d4f98a2SYan Zheng u64 search_start; 1845d4f98a2SYan Zheng u64 extents_found; 1853fd0a558SYan, Zheng 1863fd0a558SYan, Zheng unsigned int stage:8; 1873fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1883fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1895d4f98a2SYan Zheng unsigned int found_file_extent:1; 1905d4f98a2SYan Zheng }; 1915d4f98a2SYan Zheng 1925d4f98a2SYan Zheng /* stages of data relocation */ 1935d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1945d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1955d4f98a2SYan Zheng 1963fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 1973fd0a558SYan, Zheng struct backref_node *node); 1983fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 1993fd0a558SYan, Zheng struct backref_node *node); 2005d4f98a2SYan Zheng 2015d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 2025d4f98a2SYan Zheng { 2036bef4d31SEric Paris tree->rb_root = RB_ROOT; 2045d4f98a2SYan Zheng spin_lock_init(&tree->lock); 2055d4f98a2SYan Zheng } 2065d4f98a2SYan Zheng 2075d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 2085d4f98a2SYan Zheng { 2095d4f98a2SYan Zheng int i; 2106bef4d31SEric Paris cache->rb_root = RB_ROOT; 2115d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2125d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2133fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2143fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2153fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2165d4f98a2SYan Zheng } 2175d4f98a2SYan Zheng 2183fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2195d4f98a2SYan Zheng { 2203fd0a558SYan, Zheng struct backref_node *node; 2213fd0a558SYan, Zheng int i; 2223fd0a558SYan, Zheng 2233fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2243fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2253fd0a558SYan, Zheng struct backref_node, list); 2263fd0a558SYan, Zheng remove_backref_node(cache, node); 2273fd0a558SYan, Zheng } 2283fd0a558SYan, Zheng 2293fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2303fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2313fd0a558SYan, Zheng struct backref_node, lower); 2323fd0a558SYan, Zheng remove_backref_node(cache, node); 2333fd0a558SYan, Zheng } 2343fd0a558SYan, Zheng 2353fd0a558SYan, Zheng cache->last_trans = 0; 2363fd0a558SYan, Zheng 2373fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2383fd0a558SYan, Zheng BUG_ON(!list_empty(&cache->pending[i])); 2393fd0a558SYan, Zheng BUG_ON(!list_empty(&cache->changed)); 2403fd0a558SYan, Zheng BUG_ON(!list_empty(&cache->detached)); 2413fd0a558SYan, Zheng BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root)); 2423fd0a558SYan, Zheng BUG_ON(cache->nr_nodes); 2433fd0a558SYan, Zheng BUG_ON(cache->nr_edges); 2443fd0a558SYan, Zheng } 2453fd0a558SYan, Zheng 2463fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2473fd0a558SYan, Zheng { 2483fd0a558SYan, Zheng struct backref_node *node; 2493fd0a558SYan, Zheng 2503fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2513fd0a558SYan, Zheng if (node) { 2523fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2535d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2545d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2555d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2563fd0a558SYan, Zheng cache->nr_nodes++; 2573fd0a558SYan, Zheng } 2583fd0a558SYan, Zheng return node; 2593fd0a558SYan, Zheng } 2603fd0a558SYan, Zheng 2613fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 2623fd0a558SYan, Zheng struct backref_node *node) 2633fd0a558SYan, Zheng { 2643fd0a558SYan, Zheng if (node) { 2653fd0a558SYan, Zheng cache->nr_nodes--; 2663fd0a558SYan, Zheng kfree(node); 2673fd0a558SYan, Zheng } 2683fd0a558SYan, Zheng } 2693fd0a558SYan, Zheng 2703fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 2713fd0a558SYan, Zheng { 2723fd0a558SYan, Zheng struct backref_edge *edge; 2733fd0a558SYan, Zheng 2743fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 2753fd0a558SYan, Zheng if (edge) 2763fd0a558SYan, Zheng cache->nr_edges++; 2773fd0a558SYan, Zheng return edge; 2783fd0a558SYan, Zheng } 2793fd0a558SYan, Zheng 2803fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 2813fd0a558SYan, Zheng struct backref_edge *edge) 2823fd0a558SYan, Zheng { 2833fd0a558SYan, Zheng if (edge) { 2843fd0a558SYan, Zheng cache->nr_edges--; 2853fd0a558SYan, Zheng kfree(edge); 2863fd0a558SYan, Zheng } 2875d4f98a2SYan Zheng } 2885d4f98a2SYan Zheng 2895d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 2905d4f98a2SYan Zheng struct rb_node *node) 2915d4f98a2SYan Zheng { 2925d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 2935d4f98a2SYan Zheng struct rb_node *parent = NULL; 2945d4f98a2SYan Zheng struct tree_entry *entry; 2955d4f98a2SYan Zheng 2965d4f98a2SYan Zheng while (*p) { 2975d4f98a2SYan Zheng parent = *p; 2985d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 2995d4f98a2SYan Zheng 3005d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3015d4f98a2SYan Zheng p = &(*p)->rb_left; 3025d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3035d4f98a2SYan Zheng p = &(*p)->rb_right; 3045d4f98a2SYan Zheng else 3055d4f98a2SYan Zheng return parent; 3065d4f98a2SYan Zheng } 3075d4f98a2SYan Zheng 3085d4f98a2SYan Zheng rb_link_node(node, parent, p); 3095d4f98a2SYan Zheng rb_insert_color(node, root); 3105d4f98a2SYan Zheng return NULL; 3115d4f98a2SYan Zheng } 3125d4f98a2SYan Zheng 3135d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3145d4f98a2SYan Zheng { 3155d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3165d4f98a2SYan Zheng struct tree_entry *entry; 3175d4f98a2SYan Zheng 3185d4f98a2SYan Zheng while (n) { 3195d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3205d4f98a2SYan Zheng 3215d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3225d4f98a2SYan Zheng n = n->rb_left; 3235d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3245d4f98a2SYan Zheng n = n->rb_right; 3255d4f98a2SYan Zheng else 3265d4f98a2SYan Zheng return n; 3275d4f98a2SYan Zheng } 3285d4f98a2SYan Zheng return NULL; 3295d4f98a2SYan Zheng } 3305d4f98a2SYan Zheng 33148a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 33243c04fb1SJeff Mahoney { 33343c04fb1SJeff Mahoney 33443c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 33543c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 33643c04fb1SJeff Mahoney rb_node); 33743c04fb1SJeff Mahoney if (bnode->root) 33843c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 33943c04fb1SJeff Mahoney btrfs_panic(fs_info, errno, "Inconsistency in backref cache " 340351fd353SDavid Sterba "found at offset %llu", bytenr); 34143c04fb1SJeff Mahoney } 34243c04fb1SJeff Mahoney 3435d4f98a2SYan Zheng /* 3445d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3455d4f98a2SYan Zheng */ 3465d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3475d4f98a2SYan Zheng struct backref_edge *edges[], 3485d4f98a2SYan Zheng int *index) 3495d4f98a2SYan Zheng { 3505d4f98a2SYan Zheng struct backref_edge *edge; 3515d4f98a2SYan Zheng int idx = *index; 3525d4f98a2SYan Zheng 3535d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3545d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3555d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3565d4f98a2SYan Zheng edges[idx++] = edge; 3575d4f98a2SYan Zheng node = edge->node[UPPER]; 3585d4f98a2SYan Zheng } 3593fd0a558SYan, Zheng BUG_ON(node->detached); 3605d4f98a2SYan Zheng *index = idx; 3615d4f98a2SYan Zheng return node; 3625d4f98a2SYan Zheng } 3635d4f98a2SYan Zheng 3645d4f98a2SYan Zheng /* 3655d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 3665d4f98a2SYan Zheng */ 3675d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 3685d4f98a2SYan Zheng int *index) 3695d4f98a2SYan Zheng { 3705d4f98a2SYan Zheng struct backref_edge *edge; 3715d4f98a2SYan Zheng struct backref_node *lower; 3725d4f98a2SYan Zheng int idx = *index; 3735d4f98a2SYan Zheng 3745d4f98a2SYan Zheng while (idx > 0) { 3755d4f98a2SYan Zheng edge = edges[idx - 1]; 3765d4f98a2SYan Zheng lower = edge->node[LOWER]; 3775d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 3785d4f98a2SYan Zheng idx--; 3795d4f98a2SYan Zheng continue; 3805d4f98a2SYan Zheng } 3815d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 3825d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3835d4f98a2SYan Zheng edges[idx - 1] = edge; 3845d4f98a2SYan Zheng *index = idx; 3855d4f98a2SYan Zheng return edge->node[UPPER]; 3865d4f98a2SYan Zheng } 3875d4f98a2SYan Zheng *index = 0; 3885d4f98a2SYan Zheng return NULL; 3895d4f98a2SYan Zheng } 3905d4f98a2SYan Zheng 3913fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 3925d4f98a2SYan Zheng { 3935d4f98a2SYan Zheng if (node->locked) { 3945d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 3955d4f98a2SYan Zheng node->locked = 0; 3965d4f98a2SYan Zheng } 3973fd0a558SYan, Zheng } 3983fd0a558SYan, Zheng 3993fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 4003fd0a558SYan, Zheng { 4013fd0a558SYan, Zheng if (node->eb) { 4023fd0a558SYan, Zheng unlock_node_buffer(node); 4035d4f98a2SYan Zheng free_extent_buffer(node->eb); 4045d4f98a2SYan Zheng node->eb = NULL; 4055d4f98a2SYan Zheng } 4065d4f98a2SYan Zheng } 4075d4f98a2SYan Zheng 4085d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4095d4f98a2SYan Zheng struct backref_node *node) 4105d4f98a2SYan Zheng { 4115d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4125d4f98a2SYan Zheng 4135d4f98a2SYan Zheng drop_node_buffer(node); 4143fd0a558SYan, Zheng list_del(&node->list); 4155d4f98a2SYan Zheng list_del(&node->lower); 4163fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4175d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4183fd0a558SYan, Zheng free_backref_node(tree, node); 4195d4f98a2SYan Zheng } 4205d4f98a2SYan Zheng 4215d4f98a2SYan Zheng /* 4225d4f98a2SYan Zheng * remove a backref node from the backref cache 4235d4f98a2SYan Zheng */ 4245d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4255d4f98a2SYan Zheng struct backref_node *node) 4265d4f98a2SYan Zheng { 4275d4f98a2SYan Zheng struct backref_node *upper; 4285d4f98a2SYan Zheng struct backref_edge *edge; 4295d4f98a2SYan Zheng 4305d4f98a2SYan Zheng if (!node) 4315d4f98a2SYan Zheng return; 4325d4f98a2SYan Zheng 4333fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4345d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4355d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4365d4f98a2SYan Zheng list[LOWER]); 4375d4f98a2SYan Zheng upper = edge->node[UPPER]; 4385d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4395d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4403fd0a558SYan, Zheng free_backref_edge(cache, edge); 4413fd0a558SYan, Zheng 4423fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4433fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4443fd0a558SYan, Zheng drop_backref_node(cache, node); 4453fd0a558SYan, Zheng node = upper; 4463fd0a558SYan, Zheng node->lowest = 1; 4473fd0a558SYan, Zheng continue; 4483fd0a558SYan, Zheng } 4495d4f98a2SYan Zheng /* 4503fd0a558SYan, Zheng * add the node to leaf node list if no other 4515d4f98a2SYan Zheng * child block cached. 4525d4f98a2SYan Zheng */ 4535d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4543fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4555d4f98a2SYan Zheng upper->lowest = 1; 4565d4f98a2SYan Zheng } 4575d4f98a2SYan Zheng } 4583fd0a558SYan, Zheng 4595d4f98a2SYan Zheng drop_backref_node(cache, node); 4605d4f98a2SYan Zheng } 4615d4f98a2SYan Zheng 4623fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 4633fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 4643fd0a558SYan, Zheng { 4653fd0a558SYan, Zheng struct rb_node *rb_node; 4663fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 4673fd0a558SYan, Zheng node->bytenr = bytenr; 4683fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 46943c04fb1SJeff Mahoney if (rb_node) 47043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 4713fd0a558SYan, Zheng } 4723fd0a558SYan, Zheng 4733fd0a558SYan, Zheng /* 4743fd0a558SYan, Zheng * update backref cache after a transaction commit 4753fd0a558SYan, Zheng */ 4763fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 4773fd0a558SYan, Zheng struct backref_cache *cache) 4783fd0a558SYan, Zheng { 4793fd0a558SYan, Zheng struct backref_node *node; 4803fd0a558SYan, Zheng int level = 0; 4813fd0a558SYan, Zheng 4823fd0a558SYan, Zheng if (cache->last_trans == 0) { 4833fd0a558SYan, Zheng cache->last_trans = trans->transid; 4843fd0a558SYan, Zheng return 0; 4853fd0a558SYan, Zheng } 4863fd0a558SYan, Zheng 4873fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 4883fd0a558SYan, Zheng return 0; 4893fd0a558SYan, Zheng 4903fd0a558SYan, Zheng /* 4913fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 4923fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 4933fd0a558SYan, Zheng * so the detached nodes are no longer useful. 4943fd0a558SYan, Zheng */ 4953fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 4963fd0a558SYan, Zheng node = list_entry(cache->detached.next, 4973fd0a558SYan, Zheng struct backref_node, list); 4983fd0a558SYan, Zheng remove_backref_node(cache, node); 4993fd0a558SYan, Zheng } 5003fd0a558SYan, Zheng 5013fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 5023fd0a558SYan, Zheng node = list_entry(cache->changed.next, 5033fd0a558SYan, Zheng struct backref_node, list); 5043fd0a558SYan, Zheng list_del_init(&node->list); 5053fd0a558SYan, Zheng BUG_ON(node->pending); 5063fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5073fd0a558SYan, Zheng } 5083fd0a558SYan, Zheng 5093fd0a558SYan, Zheng /* 5103fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5113fd0a558SYan, Zheng * errors during processing the pending nodes. 5123fd0a558SYan, Zheng */ 5133fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5143fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5153fd0a558SYan, Zheng BUG_ON(!node->pending); 5163fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5173fd0a558SYan, Zheng continue; 5183fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5193fd0a558SYan, Zheng } 5203fd0a558SYan, Zheng } 5213fd0a558SYan, Zheng 5223fd0a558SYan, Zheng cache->last_trans = 0; 5233fd0a558SYan, Zheng return 1; 5243fd0a558SYan, Zheng } 5253fd0a558SYan, Zheng 526f2a97a9dSDavid Sterba 5273fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5283fd0a558SYan, Zheng { 5293fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5303fd0a558SYan, Zheng 53127cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 5323fd0a558SYan, Zheng return 0; 5333fd0a558SYan, Zheng 5343fd0a558SYan, Zheng reloc_root = root->reloc_root; 5353fd0a558SYan, Zheng if (!reloc_root) 5363fd0a558SYan, Zheng return 0; 5373fd0a558SYan, Zheng 5383fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5393fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5403fd0a558SYan, Zheng return 0; 5413fd0a558SYan, Zheng /* 5423fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5433fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5443fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5453fd0a558SYan, Zheng * relocation. 5463fd0a558SYan, Zheng */ 5473fd0a558SYan, Zheng return 1; 5483fd0a558SYan, Zheng } 5495d4f98a2SYan Zheng /* 5505d4f98a2SYan Zheng * find reloc tree by address of tree root 5515d4f98a2SYan Zheng */ 5525d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5535d4f98a2SYan Zheng u64 bytenr) 5545d4f98a2SYan Zheng { 5555d4f98a2SYan Zheng struct rb_node *rb_node; 5565d4f98a2SYan Zheng struct mapping_node *node; 5575d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5585d4f98a2SYan Zheng 5595d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5605d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5615d4f98a2SYan Zheng if (rb_node) { 5625d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5635d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5645d4f98a2SYan Zheng } 5655d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5665d4f98a2SYan Zheng return root; 5675d4f98a2SYan Zheng } 5685d4f98a2SYan Zheng 5695d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5705d4f98a2SYan Zheng { 5715d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5725d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5735d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 5745d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 5755d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 57666463748SWang Shilong root_objectid == BTRFS_CSUM_TREE_OBJECTID || 57766463748SWang Shilong root_objectid == BTRFS_UUID_TREE_OBJECTID || 57866463748SWang Shilong root_objectid == BTRFS_QUOTA_TREE_OBJECTID) 5795d4f98a2SYan Zheng return 1; 5805d4f98a2SYan Zheng return 0; 5815d4f98a2SYan Zheng } 5825d4f98a2SYan Zheng 5835d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 5845d4f98a2SYan Zheng u64 root_objectid) 5855d4f98a2SYan Zheng { 5865d4f98a2SYan Zheng struct btrfs_key key; 5875d4f98a2SYan Zheng 5885d4f98a2SYan Zheng key.objectid = root_objectid; 5895d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 5905d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 5915d4f98a2SYan Zheng key.offset = 0; 5925d4f98a2SYan Zheng else 5935d4f98a2SYan Zheng key.offset = (u64)-1; 5945d4f98a2SYan Zheng 595c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 5965d4f98a2SYan Zheng } 5975d4f98a2SYan Zheng 5985d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 5995d4f98a2SYan Zheng static noinline_for_stack 6005d4f98a2SYan Zheng struct btrfs_root *find_tree_root(struct reloc_control *rc, 6015d4f98a2SYan Zheng struct extent_buffer *leaf, 6025d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0) 6035d4f98a2SYan Zheng { 6045d4f98a2SYan Zheng struct btrfs_root *root; 6055d4f98a2SYan Zheng u64 root_objectid = btrfs_ref_root_v0(leaf, ref0); 6065d4f98a2SYan Zheng u64 generation = btrfs_ref_generation_v0(leaf, ref0); 6075d4f98a2SYan Zheng 6085d4f98a2SYan Zheng BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID); 6095d4f98a2SYan Zheng 6105d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, root_objectid); 6115d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 6125d4f98a2SYan Zheng 61327cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && 6145d4f98a2SYan Zheng generation != btrfs_root_generation(&root->root_item)) 6155d4f98a2SYan Zheng return NULL; 6165d4f98a2SYan Zheng 6175d4f98a2SYan Zheng return root; 6185d4f98a2SYan Zheng } 6195d4f98a2SYan Zheng #endif 6205d4f98a2SYan Zheng 6215d4f98a2SYan Zheng static noinline_for_stack 6225d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 6235d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 6245d4f98a2SYan Zheng { 6253173a18fSJosef Bacik struct btrfs_key key; 6265d4f98a2SYan Zheng struct btrfs_extent_item *ei; 6275d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6285d4f98a2SYan Zheng u32 item_size; 6295d4f98a2SYan Zheng 6303173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6313173a18fSJosef Bacik 6325d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 6335d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 6345d4f98a2SYan Zheng if (item_size < sizeof(*ei)) { 6355d4f98a2SYan Zheng WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 6365d4f98a2SYan Zheng return 1; 6375d4f98a2SYan Zheng } 6385d4f98a2SYan Zheng #endif 6395d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6405d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6415d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6425d4f98a2SYan Zheng 6433173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6443173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6455d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6465d4f98a2SYan Zheng return 1; 6475d4f98a2SYan Zheng } 648d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 649d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 650d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 651d062d13cSJosef Bacik return 1; 652d062d13cSJosef Bacik } 6535d4f98a2SYan Zheng 6543173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6555d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6565d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6573173a18fSJosef Bacik } else { 6583173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6593173a18fSJosef Bacik } 6605d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6615d4f98a2SYan Zheng return 0; 6625d4f98a2SYan Zheng } 6635d4f98a2SYan Zheng 6645d4f98a2SYan Zheng /* 6655d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6665d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6675d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6685d4f98a2SYan Zheng * 6695d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 6705d4f98a2SYan Zheng * to find upper level blocks that refernece the block, and then check 6715d4f98a2SYan Zheng * bakcrefs of these upper level blocks recursively. the recursion stop 6725d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6735d4f98a2SYan Zheng * 6745d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6755d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6765d4f98a2SYan Zheng * block are also cached. 6775d4f98a2SYan Zheng */ 6783fd0a558SYan, Zheng static noinline_for_stack 6793fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6805d4f98a2SYan Zheng struct btrfs_key *node_key, 6815d4f98a2SYan Zheng int level, u64 bytenr) 6825d4f98a2SYan Zheng { 6833fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 6845d4f98a2SYan Zheng struct btrfs_path *path1; 6855d4f98a2SYan Zheng struct btrfs_path *path2; 6865d4f98a2SYan Zheng struct extent_buffer *eb; 6875d4f98a2SYan Zheng struct btrfs_root *root; 6885d4f98a2SYan Zheng struct backref_node *cur; 6895d4f98a2SYan Zheng struct backref_node *upper; 6905d4f98a2SYan Zheng struct backref_node *lower; 6915d4f98a2SYan Zheng struct backref_node *node = NULL; 6925d4f98a2SYan Zheng struct backref_node *exist = NULL; 6935d4f98a2SYan Zheng struct backref_edge *edge; 6945d4f98a2SYan Zheng struct rb_node *rb_node; 6955d4f98a2SYan Zheng struct btrfs_key key; 6965d4f98a2SYan Zheng unsigned long end; 6975d4f98a2SYan Zheng unsigned long ptr; 6985d4f98a2SYan Zheng LIST_HEAD(list); 6993fd0a558SYan, Zheng LIST_HEAD(useless); 7003fd0a558SYan, Zheng int cowonly; 7015d4f98a2SYan Zheng int ret; 7025d4f98a2SYan Zheng int err = 0; 703b6c60c80SJosef Bacik bool need_check = true; 7045d4f98a2SYan Zheng 7055d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 7065d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 7075d4f98a2SYan Zheng if (!path1 || !path2) { 7085d4f98a2SYan Zheng err = -ENOMEM; 7095d4f98a2SYan Zheng goto out; 7105d4f98a2SYan Zheng } 711026fd317SJosef Bacik path1->reada = 1; 712026fd317SJosef Bacik path2->reada = 2; 7135d4f98a2SYan Zheng 7143fd0a558SYan, Zheng node = alloc_backref_node(cache); 7155d4f98a2SYan Zheng if (!node) { 7165d4f98a2SYan Zheng err = -ENOMEM; 7175d4f98a2SYan Zheng goto out; 7185d4f98a2SYan Zheng } 7195d4f98a2SYan Zheng 7205d4f98a2SYan Zheng node->bytenr = bytenr; 7215d4f98a2SYan Zheng node->level = level; 7225d4f98a2SYan Zheng node->lowest = 1; 7235d4f98a2SYan Zheng cur = node; 7245d4f98a2SYan Zheng again: 7255d4f98a2SYan Zheng end = 0; 7265d4f98a2SYan Zheng ptr = 0; 7275d4f98a2SYan Zheng key.objectid = cur->bytenr; 7283173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7295d4f98a2SYan Zheng key.offset = (u64)-1; 7305d4f98a2SYan Zheng 7315d4f98a2SYan Zheng path1->search_commit_root = 1; 7325d4f98a2SYan Zheng path1->skip_locking = 1; 7335d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7345d4f98a2SYan Zheng 0, 0); 7355d4f98a2SYan Zheng if (ret < 0) { 7365d4f98a2SYan Zheng err = ret; 7375d4f98a2SYan Zheng goto out; 7385d4f98a2SYan Zheng } 7395d4f98a2SYan Zheng BUG_ON(!ret || !path1->slots[0]); 7405d4f98a2SYan Zheng 7415d4f98a2SYan Zheng path1->slots[0]--; 7425d4f98a2SYan Zheng 7435d4f98a2SYan Zheng WARN_ON(cur->checked); 7445d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7455d4f98a2SYan Zheng /* 74670f23fd6SJustin P. Mattock * the backref was added previously when processing 7475d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7485d4f98a2SYan Zheng */ 7495d4f98a2SYan Zheng BUG_ON(!list_is_singular(&cur->upper)); 7505d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7515d4f98a2SYan Zheng list[LOWER]); 7525d4f98a2SYan Zheng BUG_ON(!list_empty(&edge->list[UPPER])); 7535d4f98a2SYan Zheng exist = edge->node[UPPER]; 7545d4f98a2SYan Zheng /* 7555d4f98a2SYan Zheng * add the upper level block to pending list if we need 7565d4f98a2SYan Zheng * check its backrefs 7575d4f98a2SYan Zheng */ 7585d4f98a2SYan Zheng if (!exist->checked) 7595d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7605d4f98a2SYan Zheng } else { 7615d4f98a2SYan Zheng exist = NULL; 7625d4f98a2SYan Zheng } 7635d4f98a2SYan Zheng 7645d4f98a2SYan Zheng while (1) { 7655d4f98a2SYan Zheng cond_resched(); 7665d4f98a2SYan Zheng eb = path1->nodes[0]; 7675d4f98a2SYan Zheng 7685d4f98a2SYan Zheng if (ptr >= end) { 7695d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7705d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7715d4f98a2SYan Zheng if (ret < 0) { 7725d4f98a2SYan Zheng err = ret; 7735d4f98a2SYan Zheng goto out; 7745d4f98a2SYan Zheng } 7755d4f98a2SYan Zheng if (ret > 0) 7765d4f98a2SYan Zheng break; 7775d4f98a2SYan Zheng eb = path1->nodes[0]; 7785d4f98a2SYan Zheng } 7795d4f98a2SYan Zheng 7805d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7815d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7825d4f98a2SYan Zheng WARN_ON(exist); 7835d4f98a2SYan Zheng break; 7845d4f98a2SYan Zheng } 7855d4f98a2SYan Zheng 7863173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7873173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7885d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7895d4f98a2SYan Zheng &ptr, &end); 7905d4f98a2SYan Zheng if (ret) 7915d4f98a2SYan Zheng goto next; 7925d4f98a2SYan Zheng } 7935d4f98a2SYan Zheng } 7945d4f98a2SYan Zheng 7955d4f98a2SYan Zheng if (ptr < end) { 7965d4f98a2SYan Zheng /* update key for inline back ref */ 7975d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 7985d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 7995d4f98a2SYan Zheng key.type = btrfs_extent_inline_ref_type(eb, iref); 8005d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 8015d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 8025d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 8035d4f98a2SYan Zheng } 8045d4f98a2SYan Zheng 8055d4f98a2SYan Zheng if (exist && 8065d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8075d4f98a2SYan Zheng exist->owner == key.offset) || 8085d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8095d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8105d4f98a2SYan Zheng exist = NULL; 8115d4f98a2SYan Zheng goto next; 8125d4f98a2SYan Zheng } 8135d4f98a2SYan Zheng 8145d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 8155d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY || 8165d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 8173fd0a558SYan, Zheng if (key.type == BTRFS_EXTENT_REF_V0_KEY) { 8185d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 8195d4f98a2SYan Zheng ref0 = btrfs_item_ptr(eb, path1->slots[0], 8205d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 8213fd0a558SYan, Zheng if (key.objectid == key.offset) { 822046f264fSYan, Zheng root = find_tree_root(rc, eb, ref0); 8233fd0a558SYan, Zheng if (root && !should_ignore_root(root)) 8245d4f98a2SYan Zheng cur->root = root; 8255d4f98a2SYan Zheng else 8263fd0a558SYan, Zheng list_add(&cur->list, &useless); 8275d4f98a2SYan Zheng break; 8285d4f98a2SYan Zheng } 829046f264fSYan, Zheng if (is_cowonly_root(btrfs_ref_root_v0(eb, 830046f264fSYan, Zheng ref0))) 831046f264fSYan, Zheng cur->cowonly = 1; 8323fd0a558SYan, Zheng } 8335d4f98a2SYan Zheng #else 8345d4f98a2SYan Zheng BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); 8355d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8365d4f98a2SYan Zheng #endif 8375d4f98a2SYan Zheng if (key.objectid == key.offset) { 8385d4f98a2SYan Zheng /* 8395d4f98a2SYan Zheng * only root blocks of reloc trees use 8405d4f98a2SYan Zheng * backref of this type. 8415d4f98a2SYan Zheng */ 8425d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 8435d4f98a2SYan Zheng BUG_ON(!root); 8445d4f98a2SYan Zheng cur->root = root; 8455d4f98a2SYan Zheng break; 8465d4f98a2SYan Zheng } 8475d4f98a2SYan Zheng 8483fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8495d4f98a2SYan Zheng if (!edge) { 8505d4f98a2SYan Zheng err = -ENOMEM; 8515d4f98a2SYan Zheng goto out; 8525d4f98a2SYan Zheng } 8535d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8545d4f98a2SYan Zheng if (!rb_node) { 8553fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8565d4f98a2SYan Zheng if (!upper) { 8573fd0a558SYan, Zheng free_backref_edge(cache, edge); 8585d4f98a2SYan Zheng err = -ENOMEM; 8595d4f98a2SYan Zheng goto out; 8605d4f98a2SYan Zheng } 8615d4f98a2SYan Zheng upper->bytenr = key.offset; 8625d4f98a2SYan Zheng upper->level = cur->level + 1; 8635d4f98a2SYan Zheng /* 8645d4f98a2SYan Zheng * backrefs for the upper level block isn't 8655d4f98a2SYan Zheng * cached, add the block to pending list 8665d4f98a2SYan Zheng */ 8675d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8685d4f98a2SYan Zheng } else { 8695d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8705d4f98a2SYan Zheng rb_node); 8713fd0a558SYan, Zheng BUG_ON(!upper->checked); 8725d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8735d4f98a2SYan Zheng } 8743fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8755d4f98a2SYan Zheng edge->node[LOWER] = cur; 8763fd0a558SYan, Zheng edge->node[UPPER] = upper; 8775d4f98a2SYan Zheng 8785d4f98a2SYan Zheng goto next; 8795d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8805d4f98a2SYan Zheng goto next; 8815d4f98a2SYan Zheng } 8825d4f98a2SYan Zheng 8835d4f98a2SYan Zheng /* key.type == BTRFS_TREE_BLOCK_REF_KEY */ 8845d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8855d4f98a2SYan Zheng if (IS_ERR(root)) { 8865d4f98a2SYan Zheng err = PTR_ERR(root); 8875d4f98a2SYan Zheng goto out; 8885d4f98a2SYan Zheng } 8895d4f98a2SYan Zheng 89027cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 8913fd0a558SYan, Zheng cur->cowonly = 1; 8923fd0a558SYan, Zheng 8935d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 8945d4f98a2SYan Zheng /* tree root */ 8955d4f98a2SYan Zheng BUG_ON(btrfs_root_bytenr(&root->root_item) != 8965d4f98a2SYan Zheng cur->bytenr); 8973fd0a558SYan, Zheng if (should_ignore_root(root)) 8983fd0a558SYan, Zheng list_add(&cur->list, &useless); 8993fd0a558SYan, Zheng else 9005d4f98a2SYan Zheng cur->root = root; 9015d4f98a2SYan Zheng break; 9025d4f98a2SYan Zheng } 9035d4f98a2SYan Zheng 9045d4f98a2SYan Zheng level = cur->level + 1; 9055d4f98a2SYan Zheng 9065d4f98a2SYan Zheng /* 9075d4f98a2SYan Zheng * searching the tree to find upper level blocks 9085d4f98a2SYan Zheng * reference the block. 9095d4f98a2SYan Zheng */ 9105d4f98a2SYan Zheng path2->search_commit_root = 1; 9115d4f98a2SYan Zheng path2->skip_locking = 1; 9125d4f98a2SYan Zheng path2->lowest_level = level; 9135d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 9145d4f98a2SYan Zheng path2->lowest_level = 0; 9155d4f98a2SYan Zheng if (ret < 0) { 9165d4f98a2SYan Zheng err = ret; 9175d4f98a2SYan Zheng goto out; 9185d4f98a2SYan Zheng } 91933c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 92033c66f43SYan Zheng path2->slots[level]--; 9215d4f98a2SYan Zheng 9225d4f98a2SYan Zheng eb = path2->nodes[level]; 9235d4f98a2SYan Zheng WARN_ON(btrfs_node_blockptr(eb, path2->slots[level]) != 9245d4f98a2SYan Zheng cur->bytenr); 9255d4f98a2SYan Zheng 9265d4f98a2SYan Zheng lower = cur; 927b6c60c80SJosef Bacik need_check = true; 9285d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9295d4f98a2SYan Zheng if (!path2->nodes[level]) { 9305d4f98a2SYan Zheng BUG_ON(btrfs_root_bytenr(&root->root_item) != 9315d4f98a2SYan Zheng lower->bytenr); 9323fd0a558SYan, Zheng if (should_ignore_root(root)) 9333fd0a558SYan, Zheng list_add(&lower->list, &useless); 9343fd0a558SYan, Zheng else 9355d4f98a2SYan Zheng lower->root = root; 9365d4f98a2SYan Zheng break; 9375d4f98a2SYan Zheng } 9385d4f98a2SYan Zheng 9393fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9405d4f98a2SYan Zheng if (!edge) { 9415d4f98a2SYan Zheng err = -ENOMEM; 9425d4f98a2SYan Zheng goto out; 9435d4f98a2SYan Zheng } 9445d4f98a2SYan Zheng 9455d4f98a2SYan Zheng eb = path2->nodes[level]; 9465d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9475d4f98a2SYan Zheng if (!rb_node) { 9483fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9495d4f98a2SYan Zheng if (!upper) { 9503fd0a558SYan, Zheng free_backref_edge(cache, edge); 9515d4f98a2SYan Zheng err = -ENOMEM; 9525d4f98a2SYan Zheng goto out; 9535d4f98a2SYan Zheng } 9545d4f98a2SYan Zheng upper->bytenr = eb->start; 9555d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9565d4f98a2SYan Zheng upper->level = lower->level + 1; 95727cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 95827cdeb70SMiao Xie &root->state)) 9593fd0a558SYan, Zheng upper->cowonly = 1; 9605d4f98a2SYan Zheng 9615d4f98a2SYan Zheng /* 9625d4f98a2SYan Zheng * if we know the block isn't shared 9635d4f98a2SYan Zheng * we can void checking its backrefs. 9645d4f98a2SYan Zheng */ 9655d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9665d4f98a2SYan Zheng upper->checked = 0; 9675d4f98a2SYan Zheng else 9685d4f98a2SYan Zheng upper->checked = 1; 9695d4f98a2SYan Zheng 9705d4f98a2SYan Zheng /* 9715d4f98a2SYan Zheng * add the block to pending list if we 972b6c60c80SJosef Bacik * need check its backrefs, we only do this once 973b6c60c80SJosef Bacik * while walking up a tree as we will catch 974b6c60c80SJosef Bacik * anything else later on. 9755d4f98a2SYan Zheng */ 976b6c60c80SJosef Bacik if (!upper->checked && need_check) { 977b6c60c80SJosef Bacik need_check = false; 9785d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9795d4f98a2SYan Zheng &list); 9805d4f98a2SYan Zheng } else 9815d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9825d4f98a2SYan Zheng } else { 9835d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9845d4f98a2SYan Zheng rb_node); 9855d4f98a2SYan Zheng BUG_ON(!upper->checked); 9865d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9873fd0a558SYan, Zheng if (!upper->owner) 9883fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 9895d4f98a2SYan Zheng } 9905d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 9915d4f98a2SYan Zheng edge->node[LOWER] = lower; 9923fd0a558SYan, Zheng edge->node[UPPER] = upper; 9935d4f98a2SYan Zheng 9945d4f98a2SYan Zheng if (rb_node) 9955d4f98a2SYan Zheng break; 9965d4f98a2SYan Zheng lower = upper; 9975d4f98a2SYan Zheng upper = NULL; 9985d4f98a2SYan Zheng } 999b3b4aa74SDavid Sterba btrfs_release_path(path2); 10005d4f98a2SYan Zheng next: 10015d4f98a2SYan Zheng if (ptr < end) { 10025d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 10035d4f98a2SYan Zheng if (ptr >= end) { 10045d4f98a2SYan Zheng WARN_ON(ptr > end); 10055d4f98a2SYan Zheng ptr = 0; 10065d4f98a2SYan Zheng end = 0; 10075d4f98a2SYan Zheng } 10085d4f98a2SYan Zheng } 10095d4f98a2SYan Zheng if (ptr >= end) 10105d4f98a2SYan Zheng path1->slots[0]++; 10115d4f98a2SYan Zheng } 1012b3b4aa74SDavid Sterba btrfs_release_path(path1); 10135d4f98a2SYan Zheng 10145d4f98a2SYan Zheng cur->checked = 1; 10155d4f98a2SYan Zheng WARN_ON(exist); 10165d4f98a2SYan Zheng 10175d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10185d4f98a2SYan Zheng if (!list_empty(&list)) { 10195d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10205d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10215d4f98a2SYan Zheng cur = edge->node[UPPER]; 10225d4f98a2SYan Zheng goto again; 10235d4f98a2SYan Zheng } 10245d4f98a2SYan Zheng 10255d4f98a2SYan Zheng /* 10265d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10275d4f98a2SYan Zheng * into the cache. 10285d4f98a2SYan Zheng */ 10295d4f98a2SYan Zheng BUG_ON(!node->checked); 10303fd0a558SYan, Zheng cowonly = node->cowonly; 10313fd0a558SYan, Zheng if (!cowonly) { 10323fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10333fd0a558SYan, Zheng &node->rb_node); 103443c04fb1SJeff Mahoney if (rb_node) 103543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10363fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10373fd0a558SYan, Zheng } 10385d4f98a2SYan Zheng 10395d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10405d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10415d4f98a2SYan Zheng 10425d4f98a2SYan Zheng while (!list_empty(&list)) { 10435d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10445d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10455d4f98a2SYan Zheng upper = edge->node[UPPER]; 10463fd0a558SYan, Zheng if (upper->detached) { 10473fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10483fd0a558SYan, Zheng lower = edge->node[LOWER]; 10493fd0a558SYan, Zheng free_backref_edge(cache, edge); 10503fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10513fd0a558SYan, Zheng list_add(&lower->list, &useless); 10523fd0a558SYan, Zheng continue; 10533fd0a558SYan, Zheng } 10545d4f98a2SYan Zheng 10555d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10565d4f98a2SYan Zheng if (upper->lowest) { 10575d4f98a2SYan Zheng list_del_init(&upper->lower); 10585d4f98a2SYan Zheng upper->lowest = 0; 10595d4f98a2SYan Zheng } 10605d4f98a2SYan Zheng 10615d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10625d4f98a2SYan Zheng continue; 10635d4f98a2SYan Zheng } 10645d4f98a2SYan Zheng 10655d4f98a2SYan Zheng BUG_ON(!upper->checked); 10663fd0a558SYan, Zheng BUG_ON(cowonly != upper->cowonly); 10673fd0a558SYan, Zheng if (!cowonly) { 10685d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10695d4f98a2SYan Zheng &upper->rb_node); 107043c04fb1SJeff Mahoney if (rb_node) 107143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 107243c04fb1SJeff Mahoney upper->bytenr); 10733fd0a558SYan, Zheng } 10745d4f98a2SYan Zheng 10755d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10765d4f98a2SYan Zheng 10775d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 10785d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10795d4f98a2SYan Zheng } 10803fd0a558SYan, Zheng /* 10813fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 10823fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 10833fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 10843fd0a558SYan, Zheng * lookup. 10853fd0a558SYan, Zheng */ 10863fd0a558SYan, Zheng while (!list_empty(&useless)) { 10873fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 10883fd0a558SYan, Zheng list_del_init(&upper->list); 10893fd0a558SYan, Zheng BUG_ON(!list_empty(&upper->upper)); 10903fd0a558SYan, Zheng if (upper == node) 10913fd0a558SYan, Zheng node = NULL; 10923fd0a558SYan, Zheng if (upper->lowest) { 10933fd0a558SYan, Zheng list_del_init(&upper->lower); 10943fd0a558SYan, Zheng upper->lowest = 0; 10953fd0a558SYan, Zheng } 10963fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 10973fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 10983fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 10993fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11003fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11013fd0a558SYan, Zheng lower = edge->node[LOWER]; 11023fd0a558SYan, Zheng free_backref_edge(cache, edge); 11033fd0a558SYan, Zheng 11043fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11053fd0a558SYan, Zheng list_add(&lower->list, &useless); 11063fd0a558SYan, Zheng } 11073fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11083fd0a558SYan, Zheng if (upper->level > 0) { 11093fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11103fd0a558SYan, Zheng upper->detached = 1; 11113fd0a558SYan, Zheng } else { 11123fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11133fd0a558SYan, Zheng free_backref_node(cache, upper); 11143fd0a558SYan, Zheng } 11153fd0a558SYan, Zheng } 11165d4f98a2SYan Zheng out: 11175d4f98a2SYan Zheng btrfs_free_path(path1); 11185d4f98a2SYan Zheng btrfs_free_path(path2); 11195d4f98a2SYan Zheng if (err) { 11203fd0a558SYan, Zheng while (!list_empty(&useless)) { 11213fd0a558SYan, Zheng lower = list_entry(useless.next, 11223fd0a558SYan, Zheng struct backref_node, upper); 11233fd0a558SYan, Zheng list_del_init(&lower->upper); 11243fd0a558SYan, Zheng } 11255d4f98a2SYan Zheng upper = node; 11263fd0a558SYan, Zheng INIT_LIST_HEAD(&list); 11275d4f98a2SYan Zheng while (upper) { 11285d4f98a2SYan Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 11295d4f98a2SYan Zheng list_splice_tail(&upper->upper, &list); 11303fd0a558SYan, Zheng free_backref_node(cache, upper); 11315d4f98a2SYan Zheng } 11325d4f98a2SYan Zheng 11335d4f98a2SYan Zheng if (list_empty(&list)) 11345d4f98a2SYan Zheng break; 11355d4f98a2SYan Zheng 11365d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, 11375d4f98a2SYan Zheng list[LOWER]); 11383fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11395d4f98a2SYan Zheng upper = edge->node[UPPER]; 11403fd0a558SYan, Zheng free_backref_edge(cache, edge); 11415d4f98a2SYan Zheng } 11425d4f98a2SYan Zheng return ERR_PTR(err); 11435d4f98a2SYan Zheng } 11443fd0a558SYan, Zheng BUG_ON(node && node->detached); 11455d4f98a2SYan Zheng return node; 11465d4f98a2SYan Zheng } 11475d4f98a2SYan Zheng 11485d4f98a2SYan Zheng /* 11493fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11503fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11513fd0a558SYan, Zheng * corresponds to root of source tree 11523fd0a558SYan, Zheng */ 11533fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 11543fd0a558SYan, Zheng struct reloc_control *rc, 11553fd0a558SYan, Zheng struct btrfs_root *src, 11563fd0a558SYan, Zheng struct btrfs_root *dest) 11573fd0a558SYan, Zheng { 11583fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 11593fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 11603fd0a558SYan, Zheng struct backref_node *node = NULL; 11613fd0a558SYan, Zheng struct backref_node *new_node; 11623fd0a558SYan, Zheng struct backref_edge *edge; 11633fd0a558SYan, Zheng struct backref_edge *new_edge; 11643fd0a558SYan, Zheng struct rb_node *rb_node; 11653fd0a558SYan, Zheng 11663fd0a558SYan, Zheng if (cache->last_trans > 0) 11673fd0a558SYan, Zheng update_backref_cache(trans, cache); 11683fd0a558SYan, Zheng 11693fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 11703fd0a558SYan, Zheng if (rb_node) { 11713fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 11723fd0a558SYan, Zheng if (node->detached) 11733fd0a558SYan, Zheng node = NULL; 11743fd0a558SYan, Zheng else 11753fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 11763fd0a558SYan, Zheng } 11773fd0a558SYan, Zheng 11783fd0a558SYan, Zheng if (!node) { 11793fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 11803fd0a558SYan, Zheng reloc_root->commit_root->start); 11813fd0a558SYan, Zheng if (rb_node) { 11823fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 11833fd0a558SYan, Zheng rb_node); 11843fd0a558SYan, Zheng BUG_ON(node->detached); 11853fd0a558SYan, Zheng } 11863fd0a558SYan, Zheng } 11873fd0a558SYan, Zheng 11883fd0a558SYan, Zheng if (!node) 11893fd0a558SYan, Zheng return 0; 11903fd0a558SYan, Zheng 11913fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 11923fd0a558SYan, Zheng if (!new_node) 11933fd0a558SYan, Zheng return -ENOMEM; 11943fd0a558SYan, Zheng 11953fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 11963fd0a558SYan, Zheng new_node->level = node->level; 11973fd0a558SYan, Zheng new_node->lowest = node->lowest; 11986848ad64SYan, Zheng new_node->checked = 1; 11993fd0a558SYan, Zheng new_node->root = dest; 12003fd0a558SYan, Zheng 12013fd0a558SYan, Zheng if (!node->lowest) { 12023fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12033fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12043fd0a558SYan, Zheng if (!new_edge) 12053fd0a558SYan, Zheng goto fail; 12063fd0a558SYan, Zheng 12073fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12083fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12093fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12103fd0a558SYan, Zheng &new_node->lower); 12113fd0a558SYan, Zheng } 121276b9e23dSMiao Xie } else { 121376b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12143fd0a558SYan, Zheng } 12153fd0a558SYan, Zheng 12163fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12173fd0a558SYan, Zheng &new_node->rb_node); 121843c04fb1SJeff Mahoney if (rb_node) 121943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12203fd0a558SYan, Zheng 12213fd0a558SYan, Zheng if (!new_node->lowest) { 12223fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12233fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12243fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12253fd0a558SYan, Zheng } 12263fd0a558SYan, Zheng } 12273fd0a558SYan, Zheng return 0; 12283fd0a558SYan, Zheng fail: 12293fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12303fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12313fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12323fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12333fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12343fd0a558SYan, Zheng } 12353fd0a558SYan, Zheng free_backref_node(cache, new_node); 12363fd0a558SYan, Zheng return -ENOMEM; 12373fd0a558SYan, Zheng } 12383fd0a558SYan, Zheng 12393fd0a558SYan, Zheng /* 12405d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12415d4f98a2SYan Zheng */ 1242ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12435d4f98a2SYan Zheng { 12445d4f98a2SYan Zheng struct rb_node *rb_node; 12455d4f98a2SYan Zheng struct mapping_node *node; 12465d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 12475d4f98a2SYan Zheng 12485d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1249ffd7b339SJeff Mahoney if (!node) 1250ffd7b339SJeff Mahoney return -ENOMEM; 12515d4f98a2SYan Zheng 12525d4f98a2SYan Zheng node->bytenr = root->node->start; 12535d4f98a2SYan Zheng node->data = root; 12545d4f98a2SYan Zheng 12555d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12565d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 12575d4f98a2SYan Zheng node->bytenr, &node->rb_node); 12585d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1259ffd7b339SJeff Mahoney if (rb_node) { 1260ffd7b339SJeff Mahoney btrfs_panic(root->fs_info, -EEXIST, "Duplicate root found " 1261ffd7b339SJeff Mahoney "for start=%llu while inserting into relocation " 1262351fd353SDavid Sterba "tree", node->bytenr); 126323291a04SDan Carpenter kfree(node); 126423291a04SDan Carpenter return -EEXIST; 1265ffd7b339SJeff Mahoney } 12665d4f98a2SYan Zheng 12675d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 12685d4f98a2SYan Zheng return 0; 12695d4f98a2SYan Zheng } 12705d4f98a2SYan Zheng 12715d4f98a2SYan Zheng /* 1272c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 12735d4f98a2SYan Zheng * mapping 12745d4f98a2SYan Zheng */ 1275c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 12765d4f98a2SYan Zheng { 12775d4f98a2SYan Zheng struct rb_node *rb_node; 12785d4f98a2SYan Zheng struct mapping_node *node = NULL; 12795d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 12805d4f98a2SYan Zheng 12815d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12825d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1283c974c464SWang Shilong root->node->start); 1284c974c464SWang Shilong if (rb_node) { 1285c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1286c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1287c974c464SWang Shilong } 1288c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1289c974c464SWang Shilong 1290c974c464SWang Shilong if (!node) 1291c974c464SWang Shilong return; 1292c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1293c974c464SWang Shilong 1294c974c464SWang Shilong spin_lock(&root->fs_info->trans_lock); 1295c974c464SWang Shilong list_del_init(&root->root_list); 1296c974c464SWang Shilong spin_unlock(&root->fs_info->trans_lock); 1297c974c464SWang Shilong kfree(node); 1298c974c464SWang Shilong } 1299c974c464SWang Shilong 1300c974c464SWang Shilong /* 1301c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1302c974c464SWang Shilong * mapping 1303c974c464SWang Shilong */ 1304c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1305c974c464SWang Shilong { 1306c974c464SWang Shilong struct rb_node *rb_node; 1307c974c464SWang Shilong struct mapping_node *node = NULL; 1308c974c464SWang Shilong struct reloc_control *rc = root->fs_info->reloc_ctl; 1309c974c464SWang Shilong 1310c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1311c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1312c974c464SWang Shilong root->node->start); 13135d4f98a2SYan Zheng if (rb_node) { 13145d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13155d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13165d4f98a2SYan Zheng } 13175d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13185d4f98a2SYan Zheng 13198f71f3e0SLiu Bo if (!node) 13208f71f3e0SLiu Bo return 0; 13215d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13225d4f98a2SYan Zheng 13235d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1324c974c464SWang Shilong node->bytenr = new_bytenr; 13255d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13265d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13275d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 132843c04fb1SJeff Mahoney if (rb_node) 132943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13305d4f98a2SYan Zheng return 0; 13315d4f98a2SYan Zheng } 13325d4f98a2SYan Zheng 13333fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13343fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13355d4f98a2SYan Zheng { 13365d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13375d4f98a2SYan Zheng struct extent_buffer *eb; 13385d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13395d4f98a2SYan Zheng struct btrfs_key root_key; 13405bc7247aSMiao Xie u64 last_snap = 0; 13415d4f98a2SYan Zheng int ret; 13425d4f98a2SYan Zheng 13435d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13445d4f98a2SYan Zheng BUG_ON(!root_item); 13455d4f98a2SYan Zheng 13465d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13475d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13483fd0a558SYan, Zheng root_key.offset = objectid; 13495d4f98a2SYan Zheng 13503fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 13513fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 13525d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 13535d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 13545d4f98a2SYan Zheng BUG_ON(ret); 13555d4f98a2SYan Zheng 13565bc7247aSMiao Xie last_snap = btrfs_root_last_snapshot(&root->root_item); 13573fd0a558SYan, Zheng btrfs_set_root_last_snapshot(&root->root_item, 13583fd0a558SYan, Zheng trans->transid - 1); 13593fd0a558SYan, Zheng } else { 13603fd0a558SYan, Zheng /* 13613fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 13623fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 13633fd0a558SYan, Zheng * modified after it was created have RELOC flag 13643fd0a558SYan, Zheng * set in their headers. so it's OK to not update 13653fd0a558SYan, Zheng * the 'last_snapshot'. 13663fd0a558SYan, Zheng */ 13673fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 13683fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 13693fd0a558SYan, Zheng BUG_ON(ret); 13703fd0a558SYan, Zheng } 13713fd0a558SYan, Zheng 13725d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 13735d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 13745d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 13755d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 13763fd0a558SYan, Zheng 13773fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 13783fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 13793fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 13803fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 13815d4f98a2SYan Zheng root_item->drop_level = 0; 13825bc7247aSMiao Xie /* 13835bc7247aSMiao Xie * abuse rtransid, it is safe because it is impossible to 13845bc7247aSMiao Xie * receive data into a relocation tree. 13855bc7247aSMiao Xie */ 13865bc7247aSMiao Xie btrfs_set_root_rtransid(root_item, last_snap); 13875bc7247aSMiao Xie btrfs_set_root_otransid(root_item, trans->transid); 13883fd0a558SYan, Zheng } 13895d4f98a2SYan Zheng 13905d4f98a2SYan Zheng btrfs_tree_unlock(eb); 13915d4f98a2SYan Zheng free_extent_buffer(eb); 13925d4f98a2SYan Zheng 13935d4f98a2SYan Zheng ret = btrfs_insert_root(trans, root->fs_info->tree_root, 13945d4f98a2SYan Zheng &root_key, root_item); 13955d4f98a2SYan Zheng BUG_ON(ret); 13965d4f98a2SYan Zheng kfree(root_item); 13975d4f98a2SYan Zheng 1398cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root->fs_info->tree_root, &root_key); 13995d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14005d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14013fd0a558SYan, Zheng return reloc_root; 14023fd0a558SYan, Zheng } 14033fd0a558SYan, Zheng 14043fd0a558SYan, Zheng /* 14053fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 14063fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 14073fd0a558SYan, Zheng */ 14083fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 14093fd0a558SYan, Zheng struct btrfs_root *root) 14103fd0a558SYan, Zheng { 14113fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14123fd0a558SYan, Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 141320dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14143fd0a558SYan, Zheng int clear_rsv = 0; 1415ffd7b339SJeff Mahoney int ret; 14163fd0a558SYan, Zheng 14173fd0a558SYan, Zheng if (root->reloc_root) { 14183fd0a558SYan, Zheng reloc_root = root->reloc_root; 14193fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14203fd0a558SYan, Zheng return 0; 14213fd0a558SYan, Zheng } 14223fd0a558SYan, Zheng 14233fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14243fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14253fd0a558SYan, Zheng return 0; 14263fd0a558SYan, Zheng 142720dd2cbfSMiao Xie if (!trans->reloc_reserved) { 142820dd2cbfSMiao Xie rsv = trans->block_rsv; 14293fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14303fd0a558SYan, Zheng clear_rsv = 1; 14313fd0a558SYan, Zheng } 14323fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14333fd0a558SYan, Zheng if (clear_rsv) 143420dd2cbfSMiao Xie trans->block_rsv = rsv; 14355d4f98a2SYan Zheng 1436ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1437ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14385d4f98a2SYan Zheng root->reloc_root = reloc_root; 14395d4f98a2SYan Zheng return 0; 14405d4f98a2SYan Zheng } 14415d4f98a2SYan Zheng 14425d4f98a2SYan Zheng /* 14435d4f98a2SYan Zheng * update root item of reloc tree 14445d4f98a2SYan Zheng */ 14455d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 14465d4f98a2SYan Zheng struct btrfs_root *root) 14475d4f98a2SYan Zheng { 14485d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14495d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14505d4f98a2SYan Zheng int ret; 14515d4f98a2SYan Zheng 14525d4f98a2SYan Zheng if (!root->reloc_root) 14537585717fSChris Mason goto out; 14545d4f98a2SYan Zheng 14555d4f98a2SYan Zheng reloc_root = root->reloc_root; 14565d4f98a2SYan Zheng root_item = &reloc_root->root_item; 14575d4f98a2SYan Zheng 14583fd0a558SYan, Zheng if (root->fs_info->reloc_ctl->merge_reloc_tree && 14593fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 14605d4f98a2SYan Zheng root->reloc_root = NULL; 1461c974c464SWang Shilong __del_reloc_root(reloc_root); 14625d4f98a2SYan Zheng } 14635d4f98a2SYan Zheng 14645d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 14655d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 14665d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 14675d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 14685d4f98a2SYan Zheng } 14695d4f98a2SYan Zheng 14705d4f98a2SYan Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 14715d4f98a2SYan Zheng &reloc_root->root_key, root_item); 14725d4f98a2SYan Zheng BUG_ON(ret); 14737585717fSChris Mason 14747585717fSChris Mason out: 14755d4f98a2SYan Zheng return 0; 14765d4f98a2SYan Zheng } 14775d4f98a2SYan Zheng 14785d4f98a2SYan Zheng /* 14795d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 14805d4f98a2SYan Zheng * in a subvolume 14815d4f98a2SYan Zheng */ 14825d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 14835d4f98a2SYan Zheng { 14845d4f98a2SYan Zheng struct rb_node *node; 14855d4f98a2SYan Zheng struct rb_node *prev; 14865d4f98a2SYan Zheng struct btrfs_inode *entry; 14875d4f98a2SYan Zheng struct inode *inode; 14885d4f98a2SYan Zheng 14895d4f98a2SYan Zheng spin_lock(&root->inode_lock); 14905d4f98a2SYan Zheng again: 14915d4f98a2SYan Zheng node = root->inode_tree.rb_node; 14925d4f98a2SYan Zheng prev = NULL; 14935d4f98a2SYan Zheng while (node) { 14945d4f98a2SYan Zheng prev = node; 14955d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 14965d4f98a2SYan Zheng 149733345d01SLi Zefan if (objectid < btrfs_ino(&entry->vfs_inode)) 14985d4f98a2SYan Zheng node = node->rb_left; 149933345d01SLi Zefan else if (objectid > btrfs_ino(&entry->vfs_inode)) 15005d4f98a2SYan Zheng node = node->rb_right; 15015d4f98a2SYan Zheng else 15025d4f98a2SYan Zheng break; 15035d4f98a2SYan Zheng } 15045d4f98a2SYan Zheng if (!node) { 15055d4f98a2SYan Zheng while (prev) { 15065d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 150733345d01SLi Zefan if (objectid <= btrfs_ino(&entry->vfs_inode)) { 15085d4f98a2SYan Zheng node = prev; 15095d4f98a2SYan Zheng break; 15105d4f98a2SYan Zheng } 15115d4f98a2SYan Zheng prev = rb_next(prev); 15125d4f98a2SYan Zheng } 15135d4f98a2SYan Zheng } 15145d4f98a2SYan Zheng while (node) { 15155d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15165d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15175d4f98a2SYan Zheng if (inode) { 15185d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15195d4f98a2SYan Zheng return inode; 15205d4f98a2SYan Zheng } 15215d4f98a2SYan Zheng 152233345d01SLi Zefan objectid = btrfs_ino(&entry->vfs_inode) + 1; 15235d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15245d4f98a2SYan Zheng goto again; 15255d4f98a2SYan Zheng 15265d4f98a2SYan Zheng node = rb_next(node); 15275d4f98a2SYan Zheng } 15285d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15295d4f98a2SYan Zheng return NULL; 15305d4f98a2SYan Zheng } 15315d4f98a2SYan Zheng 15325d4f98a2SYan Zheng static int in_block_group(u64 bytenr, 15335d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group) 15345d4f98a2SYan Zheng { 15355d4f98a2SYan Zheng if (bytenr >= block_group->key.objectid && 15365d4f98a2SYan Zheng bytenr < block_group->key.objectid + block_group->key.offset) 15375d4f98a2SYan Zheng return 1; 15385d4f98a2SYan Zheng return 0; 15395d4f98a2SYan Zheng } 15405d4f98a2SYan Zheng 15415d4f98a2SYan Zheng /* 15425d4f98a2SYan Zheng * get new location of data 15435d4f98a2SYan Zheng */ 15445d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 15455d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 15465d4f98a2SYan Zheng { 15475d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 15485d4f98a2SYan Zheng struct btrfs_path *path; 15495d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15505d4f98a2SYan Zheng struct extent_buffer *leaf; 15515d4f98a2SYan Zheng int ret; 15525d4f98a2SYan Zheng 15535d4f98a2SYan Zheng path = btrfs_alloc_path(); 15545d4f98a2SYan Zheng if (!path) 15555d4f98a2SYan Zheng return -ENOMEM; 15565d4f98a2SYan Zheng 15575d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 155833345d01SLi Zefan ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode), 15595d4f98a2SYan Zheng bytenr, 0); 15605d4f98a2SYan Zheng if (ret < 0) 15615d4f98a2SYan Zheng goto out; 15625d4f98a2SYan Zheng if (ret > 0) { 15635d4f98a2SYan Zheng ret = -ENOENT; 15645d4f98a2SYan Zheng goto out; 15655d4f98a2SYan Zheng } 15665d4f98a2SYan Zheng 15675d4f98a2SYan Zheng leaf = path->nodes[0]; 15685d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 15695d4f98a2SYan Zheng struct btrfs_file_extent_item); 15705d4f98a2SYan Zheng 15715d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 15725d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 15735d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 15745d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 15755d4f98a2SYan Zheng 15765d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 157783d4cfd4SJosef Bacik ret = -EINVAL; 15785d4f98a2SYan Zheng goto out; 15795d4f98a2SYan Zheng } 15805d4f98a2SYan Zheng 15815d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 15825d4f98a2SYan Zheng ret = 0; 15835d4f98a2SYan Zheng out: 15845d4f98a2SYan Zheng btrfs_free_path(path); 15855d4f98a2SYan Zheng return ret; 15865d4f98a2SYan Zheng } 15875d4f98a2SYan Zheng 15885d4f98a2SYan Zheng /* 15895d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 15905d4f98a2SYan Zheng * the new locations. 15915d4f98a2SYan Zheng */ 15923fd0a558SYan, Zheng static noinline_for_stack 15933fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 15945d4f98a2SYan Zheng struct reloc_control *rc, 15955d4f98a2SYan Zheng struct btrfs_root *root, 15963fd0a558SYan, Zheng struct extent_buffer *leaf) 15975d4f98a2SYan Zheng { 15985d4f98a2SYan Zheng struct btrfs_key key; 15995d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16005d4f98a2SYan Zheng struct inode *inode = NULL; 16015d4f98a2SYan Zheng u64 parent; 16025d4f98a2SYan Zheng u64 bytenr; 16033fd0a558SYan, Zheng u64 new_bytenr = 0; 16045d4f98a2SYan Zheng u64 num_bytes; 16055d4f98a2SYan Zheng u64 end; 16065d4f98a2SYan Zheng u32 nritems; 16075d4f98a2SYan Zheng u32 i; 160883d4cfd4SJosef Bacik int ret = 0; 16095d4f98a2SYan Zheng int first = 1; 16105d4f98a2SYan Zheng int dirty = 0; 16115d4f98a2SYan Zheng 16125d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16135d4f98a2SYan Zheng return 0; 16145d4f98a2SYan Zheng 16155d4f98a2SYan Zheng /* reloc trees always use full backref */ 16165d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16175d4f98a2SYan Zheng parent = leaf->start; 16185d4f98a2SYan Zheng else 16195d4f98a2SYan Zheng parent = 0; 16205d4f98a2SYan Zheng 16215d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16225d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 16235d4f98a2SYan Zheng cond_resched(); 16245d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16255d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16265d4f98a2SYan Zheng continue; 16275d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16285d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16295d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16305d4f98a2SYan Zheng continue; 16315d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16325d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16335d4f98a2SYan Zheng if (bytenr == 0) 16345d4f98a2SYan Zheng continue; 16355d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16365d4f98a2SYan Zheng continue; 16375d4f98a2SYan Zheng 16385d4f98a2SYan Zheng /* 16395d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16405d4f98a2SYan Zheng * to complete and drop the extent cache 16415d4f98a2SYan Zheng */ 16425d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16435d4f98a2SYan Zheng if (first) { 16445d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16455d4f98a2SYan Zheng first = 0; 164633345d01SLi Zefan } else if (inode && btrfs_ino(inode) < key.objectid) { 16473fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16485d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16495d4f98a2SYan Zheng } 165033345d01SLi Zefan if (inode && btrfs_ino(inode) == key.objectid) { 16515d4f98a2SYan Zheng end = key.offset + 16525d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 16535d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 16545d4f98a2SYan Zheng root->sectorsize)); 16555d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 16565d4f98a2SYan Zheng end--; 16575d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1658d0082371SJeff Mahoney key.offset, end); 16595d4f98a2SYan Zheng if (!ret) 16605d4f98a2SYan Zheng continue; 16615d4f98a2SYan Zheng 16625d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, key.offset, end, 16635d4f98a2SYan Zheng 1); 16645d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1665d0082371SJeff Mahoney key.offset, end); 16665d4f98a2SYan Zheng } 16675d4f98a2SYan Zheng } 16685d4f98a2SYan Zheng 16695d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 16705d4f98a2SYan Zheng bytenr, num_bytes); 167183d4cfd4SJosef Bacik if (ret) { 167283d4cfd4SJosef Bacik /* 167383d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 167483d4cfd4SJosef Bacik * in the file extent yet. 167583d4cfd4SJosef Bacik */ 167683d4cfd4SJosef Bacik break; 16773fd0a558SYan, Zheng } 16785d4f98a2SYan Zheng 16795d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 16805d4f98a2SYan Zheng dirty = 1; 16815d4f98a2SYan Zheng 16825d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 16835d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, new_bytenr, 16845d4f98a2SYan Zheng num_bytes, parent, 16855d4f98a2SYan Zheng btrfs_header_owner(leaf), 168666d7e7f0SArne Jansen key.objectid, key.offset, 1); 168783d4cfd4SJosef Bacik if (ret) { 168883d4cfd4SJosef Bacik btrfs_abort_transaction(trans, root, ret); 168983d4cfd4SJosef Bacik break; 169083d4cfd4SJosef Bacik } 16915d4f98a2SYan Zheng 16925d4f98a2SYan Zheng ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 16935d4f98a2SYan Zheng parent, btrfs_header_owner(leaf), 169466d7e7f0SArne Jansen key.objectid, key.offset, 1); 169583d4cfd4SJosef Bacik if (ret) { 169683d4cfd4SJosef Bacik btrfs_abort_transaction(trans, root, ret); 169783d4cfd4SJosef Bacik break; 169883d4cfd4SJosef Bacik } 16995d4f98a2SYan Zheng } 17005d4f98a2SYan Zheng if (dirty) 17015d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17023fd0a558SYan, Zheng if (inode) 17033fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 170483d4cfd4SJosef Bacik return ret; 17055d4f98a2SYan Zheng } 17065d4f98a2SYan Zheng 17075d4f98a2SYan Zheng static noinline_for_stack 17085d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17095d4f98a2SYan Zheng struct btrfs_path *path, int level) 17105d4f98a2SYan Zheng { 17115d4f98a2SYan Zheng struct btrfs_disk_key key1; 17125d4f98a2SYan Zheng struct btrfs_disk_key key2; 17135d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17145d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17155d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17165d4f98a2SYan Zheng } 17175d4f98a2SYan Zheng 17185d4f98a2SYan Zheng /* 17195d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17205d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17215d4f98a2SYan Zheng * reloc tree was create can be replaced. 17225d4f98a2SYan Zheng * 17235d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17245d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17255d4f98a2SYan Zheng * errors, a negative error number is returned. 17265d4f98a2SYan Zheng */ 17273fd0a558SYan, Zheng static noinline_for_stack 17283fd0a558SYan, Zheng int replace_path(struct btrfs_trans_handle *trans, 17295d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17305d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17315d4f98a2SYan Zheng int lowest_level, int max_level) 17325d4f98a2SYan Zheng { 17335d4f98a2SYan Zheng struct extent_buffer *eb; 17345d4f98a2SYan Zheng struct extent_buffer *parent; 17355d4f98a2SYan Zheng struct btrfs_key key; 17365d4f98a2SYan Zheng u64 old_bytenr; 17375d4f98a2SYan Zheng u64 new_bytenr; 17385d4f98a2SYan Zheng u64 old_ptr_gen; 17395d4f98a2SYan Zheng u64 new_ptr_gen; 17405d4f98a2SYan Zheng u64 last_snapshot; 17415d4f98a2SYan Zheng u32 blocksize; 17423fd0a558SYan, Zheng int cow = 0; 17435d4f98a2SYan Zheng int level; 17445d4f98a2SYan Zheng int ret; 17455d4f98a2SYan Zheng int slot; 17465d4f98a2SYan Zheng 17475d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17485d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17495d4f98a2SYan Zheng 17505d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 17513fd0a558SYan, Zheng again: 17525d4f98a2SYan Zheng slot = path->slots[lowest_level]; 17535d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 17545d4f98a2SYan Zheng 17555d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 17565d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 17575d4f98a2SYan Zheng level = btrfs_header_level(eb); 17585d4f98a2SYan Zheng 17595d4f98a2SYan Zheng if (level < lowest_level) { 17605d4f98a2SYan Zheng btrfs_tree_unlock(eb); 17615d4f98a2SYan Zheng free_extent_buffer(eb); 17625d4f98a2SYan Zheng return 0; 17635d4f98a2SYan Zheng } 17645d4f98a2SYan Zheng 17653fd0a558SYan, Zheng if (cow) { 17665d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 17675d4f98a2SYan Zheng BUG_ON(ret); 17683fd0a558SYan, Zheng } 17695d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 17705d4f98a2SYan Zheng 17715d4f98a2SYan Zheng if (next_key) { 17725d4f98a2SYan Zheng next_key->objectid = (u64)-1; 17735d4f98a2SYan Zheng next_key->type = (u8)-1; 17745d4f98a2SYan Zheng next_key->offset = (u64)-1; 17755d4f98a2SYan Zheng } 17765d4f98a2SYan Zheng 17775d4f98a2SYan Zheng parent = eb; 17785d4f98a2SYan Zheng while (1) { 17795d4f98a2SYan Zheng level = btrfs_header_level(parent); 17805d4f98a2SYan Zheng BUG_ON(level < lowest_level); 17815d4f98a2SYan Zheng 17825d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 17835d4f98a2SYan Zheng if (ret && slot > 0) 17845d4f98a2SYan Zheng slot--; 17855d4f98a2SYan Zheng 17865d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 17875d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 17885d4f98a2SYan Zheng 17895d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 1790707e8a07SDavid Sterba blocksize = dest->nodesize; 17915d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 17925d4f98a2SYan Zheng 17935d4f98a2SYan Zheng if (level <= max_level) { 17945d4f98a2SYan Zheng eb = path->nodes[level]; 17955d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 17965d4f98a2SYan Zheng path->slots[level]); 17975d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 17985d4f98a2SYan Zheng path->slots[level]); 17995d4f98a2SYan Zheng } else { 18005d4f98a2SYan Zheng new_bytenr = 0; 18015d4f98a2SYan Zheng new_ptr_gen = 0; 18025d4f98a2SYan Zheng } 18035d4f98a2SYan Zheng 1804fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18055d4f98a2SYan Zheng ret = level; 18065d4f98a2SYan Zheng break; 18075d4f98a2SYan Zheng } 18085d4f98a2SYan Zheng 18095d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18105d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18113fd0a558SYan, Zheng if (level <= lowest_level) { 18125d4f98a2SYan Zheng ret = 0; 18135d4f98a2SYan Zheng break; 18145d4f98a2SYan Zheng } 18155d4f98a2SYan Zheng 18165d4f98a2SYan Zheng eb = read_tree_block(dest, old_bytenr, blocksize, 18175d4f98a2SYan Zheng old_ptr_gen); 1818416bc658SJosef Bacik if (!eb || !extent_buffer_uptodate(eb)) { 1819416bc658SJosef Bacik ret = (!eb) ? -ENOMEM : -EIO; 1820416bc658SJosef Bacik free_extent_buffer(eb); 1821379cde74SStefan Behrens break; 1822416bc658SJosef Bacik } 18235d4f98a2SYan Zheng btrfs_tree_lock(eb); 18243fd0a558SYan, Zheng if (cow) { 18255d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18265d4f98a2SYan Zheng slot, &eb); 18275d4f98a2SYan Zheng BUG_ON(ret); 18285d4f98a2SYan Zheng } 18293fd0a558SYan, Zheng btrfs_set_lock_blocking(eb); 18305d4f98a2SYan Zheng 18315d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18325d4f98a2SYan Zheng free_extent_buffer(parent); 18335d4f98a2SYan Zheng 18345d4f98a2SYan Zheng parent = eb; 18355d4f98a2SYan Zheng continue; 18365d4f98a2SYan Zheng } 18375d4f98a2SYan Zheng 18383fd0a558SYan, Zheng if (!cow) { 18393fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18403fd0a558SYan, Zheng free_extent_buffer(parent); 18413fd0a558SYan, Zheng cow = 1; 18423fd0a558SYan, Zheng goto again; 18433fd0a558SYan, Zheng } 18443fd0a558SYan, Zheng 18455d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18465d4f98a2SYan Zheng path->slots[level]); 1847b3b4aa74SDavid Sterba btrfs_release_path(path); 18485d4f98a2SYan Zheng 18495d4f98a2SYan Zheng path->lowest_level = level; 18505d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 18515d4f98a2SYan Zheng path->lowest_level = 0; 18525d4f98a2SYan Zheng BUG_ON(ret); 18535d4f98a2SYan Zheng 18545d4f98a2SYan Zheng /* 18555d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 18565d4f98a2SYan Zheng */ 18575d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 18585d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 18595d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 18605d4f98a2SYan Zheng 18615d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 18625d4f98a2SYan Zheng path->slots[level], old_bytenr); 18635d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 18645d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 18655d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 18665d4f98a2SYan Zheng 18675d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize, 18685d4f98a2SYan Zheng path->nodes[level]->start, 186966d7e7f0SArne Jansen src->root_key.objectid, level - 1, 0, 187066d7e7f0SArne Jansen 1); 18715d4f98a2SYan Zheng BUG_ON(ret); 18725d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize, 18735d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 187466d7e7f0SArne Jansen 0, 1); 18755d4f98a2SYan Zheng BUG_ON(ret); 18765d4f98a2SYan Zheng 18775d4f98a2SYan Zheng ret = btrfs_free_extent(trans, src, new_bytenr, blocksize, 18785d4f98a2SYan Zheng path->nodes[level]->start, 187966d7e7f0SArne Jansen src->root_key.objectid, level - 1, 0, 188066d7e7f0SArne Jansen 1); 18815d4f98a2SYan Zheng BUG_ON(ret); 18825d4f98a2SYan Zheng 18835d4f98a2SYan Zheng ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize, 18845d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 188566d7e7f0SArne Jansen 0, 1); 18865d4f98a2SYan Zheng BUG_ON(ret); 18875d4f98a2SYan Zheng 18885d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 18895d4f98a2SYan Zheng 18905d4f98a2SYan Zheng ret = level; 18915d4f98a2SYan Zheng break; 18925d4f98a2SYan Zheng } 18935d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18945d4f98a2SYan Zheng free_extent_buffer(parent); 18955d4f98a2SYan Zheng return ret; 18965d4f98a2SYan Zheng } 18975d4f98a2SYan Zheng 18985d4f98a2SYan Zheng /* 18995d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19005d4f98a2SYan Zheng */ 19015d4f98a2SYan Zheng static noinline_for_stack 19025d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19035d4f98a2SYan Zheng int *level) 19045d4f98a2SYan Zheng { 19055d4f98a2SYan Zheng struct extent_buffer *eb; 19065d4f98a2SYan Zheng int i; 19075d4f98a2SYan Zheng u64 last_snapshot; 19085d4f98a2SYan Zheng u32 nritems; 19095d4f98a2SYan Zheng 19105d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19115d4f98a2SYan Zheng 19125d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19135d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19145d4f98a2SYan Zheng path->nodes[i] = NULL; 19155d4f98a2SYan Zheng } 19165d4f98a2SYan Zheng 19175d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 19185d4f98a2SYan Zheng eb = path->nodes[i]; 19195d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19205d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 19215d4f98a2SYan Zheng path->slots[i]++; 19225d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 19235d4f98a2SYan Zheng last_snapshot) 19245d4f98a2SYan Zheng continue; 19255d4f98a2SYan Zheng 19265d4f98a2SYan Zheng *level = i; 19275d4f98a2SYan Zheng return 0; 19285d4f98a2SYan Zheng } 19295d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19305d4f98a2SYan Zheng path->nodes[i] = NULL; 19315d4f98a2SYan Zheng } 19325d4f98a2SYan Zheng return 1; 19335d4f98a2SYan Zheng } 19345d4f98a2SYan Zheng 19355d4f98a2SYan Zheng /* 19365d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 19375d4f98a2SYan Zheng */ 19385d4f98a2SYan Zheng static noinline_for_stack 19395d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19405d4f98a2SYan Zheng int *level) 19415d4f98a2SYan Zheng { 19425d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 19435d4f98a2SYan Zheng int i; 19445d4f98a2SYan Zheng u64 bytenr; 19455d4f98a2SYan Zheng u64 ptr_gen = 0; 19465d4f98a2SYan Zheng u64 last_snapshot; 19475d4f98a2SYan Zheng u32 blocksize; 19485d4f98a2SYan Zheng u32 nritems; 19495d4f98a2SYan Zheng 19505d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19515d4f98a2SYan Zheng 19525d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 19535d4f98a2SYan Zheng eb = path->nodes[i]; 19545d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19555d4f98a2SYan Zheng while (path->slots[i] < nritems) { 19565d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 19575d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 19585d4f98a2SYan Zheng break; 19595d4f98a2SYan Zheng path->slots[i]++; 19605d4f98a2SYan Zheng } 19615d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 19625d4f98a2SYan Zheng if (i == *level) 19635d4f98a2SYan Zheng break; 19645d4f98a2SYan Zheng *level = i + 1; 19655d4f98a2SYan Zheng return 0; 19665d4f98a2SYan Zheng } 19675d4f98a2SYan Zheng if (i == 1) { 19685d4f98a2SYan Zheng *level = i; 19695d4f98a2SYan Zheng return 0; 19705d4f98a2SYan Zheng } 19715d4f98a2SYan Zheng 19725d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 1973707e8a07SDavid Sterba blocksize = root->nodesize; 19745d4f98a2SYan Zheng eb = read_tree_block(root, bytenr, blocksize, ptr_gen); 1975416bc658SJosef Bacik if (!eb || !extent_buffer_uptodate(eb)) { 1976416bc658SJosef Bacik free_extent_buffer(eb); 1977416bc658SJosef Bacik return -EIO; 1978416bc658SJosef Bacik } 19795d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 19805d4f98a2SYan Zheng path->nodes[i - 1] = eb; 19815d4f98a2SYan Zheng path->slots[i - 1] = 0; 19825d4f98a2SYan Zheng } 19835d4f98a2SYan Zheng return 1; 19845d4f98a2SYan Zheng } 19855d4f98a2SYan Zheng 19865d4f98a2SYan Zheng /* 19875d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 19885d4f98a2SYan Zheng * [min_key, max_key) 19895d4f98a2SYan Zheng */ 19905d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 19915d4f98a2SYan Zheng struct btrfs_key *min_key, 19925d4f98a2SYan Zheng struct btrfs_key *max_key) 19935d4f98a2SYan Zheng { 19945d4f98a2SYan Zheng struct inode *inode = NULL; 19955d4f98a2SYan Zheng u64 objectid; 19965d4f98a2SYan Zheng u64 start, end; 199733345d01SLi Zefan u64 ino; 19985d4f98a2SYan Zheng 19995d4f98a2SYan Zheng objectid = min_key->objectid; 20005d4f98a2SYan Zheng while (1) { 20015d4f98a2SYan Zheng cond_resched(); 20025d4f98a2SYan Zheng iput(inode); 20035d4f98a2SYan Zheng 20045d4f98a2SYan Zheng if (objectid > max_key->objectid) 20055d4f98a2SYan Zheng break; 20065d4f98a2SYan Zheng 20075d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20085d4f98a2SYan Zheng if (!inode) 20095d4f98a2SYan Zheng break; 201033345d01SLi Zefan ino = btrfs_ino(inode); 20115d4f98a2SYan Zheng 201233345d01SLi Zefan if (ino > max_key->objectid) { 20135d4f98a2SYan Zheng iput(inode); 20145d4f98a2SYan Zheng break; 20155d4f98a2SYan Zheng } 20165d4f98a2SYan Zheng 201733345d01SLi Zefan objectid = ino + 1; 20185d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 20195d4f98a2SYan Zheng continue; 20205d4f98a2SYan Zheng 202133345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 20225d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 20235d4f98a2SYan Zheng continue; 20245d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 20255d4f98a2SYan Zheng start = 0; 20265d4f98a2SYan Zheng else { 20275d4f98a2SYan Zheng start = min_key->offset; 20285d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(start, root->sectorsize)); 20295d4f98a2SYan Zheng } 20305d4f98a2SYan Zheng } else { 20315d4f98a2SYan Zheng start = 0; 20325d4f98a2SYan Zheng } 20335d4f98a2SYan Zheng 203433345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 20355d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 20365d4f98a2SYan Zheng continue; 20375d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 20385d4f98a2SYan Zheng end = (u64)-1; 20395d4f98a2SYan Zheng } else { 20405d4f98a2SYan Zheng if (max_key->offset == 0) 20415d4f98a2SYan Zheng continue; 20425d4f98a2SYan Zheng end = max_key->offset; 20435d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 20445d4f98a2SYan Zheng end--; 20455d4f98a2SYan Zheng } 20465d4f98a2SYan Zheng } else { 20475d4f98a2SYan Zheng end = (u64)-1; 20485d4f98a2SYan Zheng } 20495d4f98a2SYan Zheng 20505d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2051d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 20525d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 1); 2053d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 20545d4f98a2SYan Zheng } 20555d4f98a2SYan Zheng return 0; 20565d4f98a2SYan Zheng } 20575d4f98a2SYan Zheng 20585d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 20595d4f98a2SYan Zheng struct btrfs_key *key) 20605d4f98a2SYan Zheng 20615d4f98a2SYan Zheng { 20625d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 20635d4f98a2SYan Zheng if (!path->nodes[level]) 20645d4f98a2SYan Zheng break; 20655d4f98a2SYan Zheng if (path->slots[level] + 1 < 20665d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 20675d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 20685d4f98a2SYan Zheng path->slots[level] + 1); 20695d4f98a2SYan Zheng return 0; 20705d4f98a2SYan Zheng } 20715d4f98a2SYan Zheng level++; 20725d4f98a2SYan Zheng } 20735d4f98a2SYan Zheng return 1; 20745d4f98a2SYan Zheng } 20755d4f98a2SYan Zheng 20765d4f98a2SYan Zheng /* 20775d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 20785d4f98a2SYan Zheng * fs tree. 20795d4f98a2SYan Zheng */ 20805d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 20815d4f98a2SYan Zheng struct btrfs_root *root) 20825d4f98a2SYan Zheng { 20835d4f98a2SYan Zheng LIST_HEAD(inode_list); 20845d4f98a2SYan Zheng struct btrfs_key key; 20855d4f98a2SYan Zheng struct btrfs_key next_key; 20869e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 20875d4f98a2SYan Zheng struct btrfs_root *reloc_root; 20885d4f98a2SYan Zheng struct btrfs_root_item *root_item; 20895d4f98a2SYan Zheng struct btrfs_path *path; 20903fd0a558SYan, Zheng struct extent_buffer *leaf; 20915d4f98a2SYan Zheng int level; 20925d4f98a2SYan Zheng int max_level; 20935d4f98a2SYan Zheng int replaced = 0; 20945d4f98a2SYan Zheng int ret; 20955d4f98a2SYan Zheng int err = 0; 20963fd0a558SYan, Zheng u32 min_reserved; 20975d4f98a2SYan Zheng 20985d4f98a2SYan Zheng path = btrfs_alloc_path(); 20995d4f98a2SYan Zheng if (!path) 21005d4f98a2SYan Zheng return -ENOMEM; 2101026fd317SJosef Bacik path->reada = 1; 21025d4f98a2SYan Zheng 21035d4f98a2SYan Zheng reloc_root = root->reloc_root; 21045d4f98a2SYan Zheng root_item = &reloc_root->root_item; 21055d4f98a2SYan Zheng 21065d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 21075d4f98a2SYan Zheng level = btrfs_root_level(root_item); 21085d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 21095d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 21105d4f98a2SYan Zheng path->slots[level] = 0; 21115d4f98a2SYan Zheng } else { 21125d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 21135d4f98a2SYan Zheng 21145d4f98a2SYan Zheng level = root_item->drop_level; 21155d4f98a2SYan Zheng BUG_ON(level == 0); 21165d4f98a2SYan Zheng path->lowest_level = level; 21175d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 211833c66f43SYan Zheng path->lowest_level = 0; 21195d4f98a2SYan Zheng if (ret < 0) { 21205d4f98a2SYan Zheng btrfs_free_path(path); 21215d4f98a2SYan Zheng return ret; 21225d4f98a2SYan Zheng } 21235d4f98a2SYan Zheng 21245d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 21255d4f98a2SYan Zheng path->slots[level]); 21265d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 21275d4f98a2SYan Zheng 21285d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 21295d4f98a2SYan Zheng } 21305d4f98a2SYan Zheng 21313fd0a558SYan, Zheng min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 21325d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 21335d4f98a2SYan Zheng 21345d4f98a2SYan Zheng while (1) { 213508e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 213608e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 21373fd0a558SYan, Zheng if (ret) { 21389e6a0c52SJosef Bacik err = ret; 21399e6a0c52SJosef Bacik goto out; 21403fd0a558SYan, Zheng } 21419e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 21429e6a0c52SJosef Bacik if (IS_ERR(trans)) { 21439e6a0c52SJosef Bacik err = PTR_ERR(trans); 21449e6a0c52SJosef Bacik trans = NULL; 21459e6a0c52SJosef Bacik goto out; 21469e6a0c52SJosef Bacik } 21479e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 21483fd0a558SYan, Zheng 21493fd0a558SYan, Zheng replaced = 0; 21505d4f98a2SYan Zheng max_level = level; 21515d4f98a2SYan Zheng 21525d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 21535d4f98a2SYan Zheng if (ret < 0) { 21545d4f98a2SYan Zheng err = ret; 21555d4f98a2SYan Zheng goto out; 21565d4f98a2SYan Zheng } 21575d4f98a2SYan Zheng if (ret > 0) 21585d4f98a2SYan Zheng break; 21595d4f98a2SYan Zheng 21605d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 21615d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 21625d4f98a2SYan Zheng ret = 0; 21635d4f98a2SYan Zheng } else { 21643fd0a558SYan, Zheng ret = replace_path(trans, root, reloc_root, path, 21653fd0a558SYan, Zheng &next_key, level, max_level); 21665d4f98a2SYan Zheng } 21675d4f98a2SYan Zheng if (ret < 0) { 21685d4f98a2SYan Zheng err = ret; 21695d4f98a2SYan Zheng goto out; 21705d4f98a2SYan Zheng } 21715d4f98a2SYan Zheng 21725d4f98a2SYan Zheng if (ret > 0) { 21735d4f98a2SYan Zheng level = ret; 21745d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 21755d4f98a2SYan Zheng path->slots[level]); 21765d4f98a2SYan Zheng replaced = 1; 21775d4f98a2SYan Zheng } 21785d4f98a2SYan Zheng 21795d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 21805d4f98a2SYan Zheng if (ret > 0) 21815d4f98a2SYan Zheng break; 21825d4f98a2SYan Zheng 21835d4f98a2SYan Zheng BUG_ON(level == 0); 21845d4f98a2SYan Zheng /* 21855d4f98a2SYan Zheng * save the merging progress in the drop_progress. 21865d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 21875d4f98a2SYan Zheng */ 21885d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 21895d4f98a2SYan Zheng path->slots[level]); 21905d4f98a2SYan Zheng root_item->drop_level = level; 21915d4f98a2SYan Zheng 21923fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 21939e6a0c52SJosef Bacik trans = NULL; 21945d4f98a2SYan Zheng 2195b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 21965d4f98a2SYan Zheng 21975d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 21985d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 21995d4f98a2SYan Zheng } 22005d4f98a2SYan Zheng 22015d4f98a2SYan Zheng /* 22025d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 22035d4f98a2SYan Zheng * relocated and the block is tree root. 22045d4f98a2SYan Zheng */ 22055d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 22065d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 22075d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 22085d4f98a2SYan Zheng free_extent_buffer(leaf); 22095d4f98a2SYan Zheng if (ret < 0) 22105d4f98a2SYan Zheng err = ret; 22115d4f98a2SYan Zheng out: 22125d4f98a2SYan Zheng btrfs_free_path(path); 22135d4f98a2SYan Zheng 22145d4f98a2SYan Zheng if (err == 0) { 22155d4f98a2SYan Zheng memset(&root_item->drop_progress, 0, 22165d4f98a2SYan Zheng sizeof(root_item->drop_progress)); 22175d4f98a2SYan Zheng root_item->drop_level = 0; 22185d4f98a2SYan Zheng btrfs_set_root_refs(root_item, 0); 22193fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 22205d4f98a2SYan Zheng } 22215d4f98a2SYan Zheng 22229e6a0c52SJosef Bacik if (trans) 22233fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 22245d4f98a2SYan Zheng 2225b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 22265d4f98a2SYan Zheng 22275d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22285d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22295d4f98a2SYan Zheng 22305d4f98a2SYan Zheng return err; 22315d4f98a2SYan Zheng } 22325d4f98a2SYan Zheng 22333fd0a558SYan, Zheng static noinline_for_stack 22343fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 22355d4f98a2SYan Zheng { 22363fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 22373fd0a558SYan, Zheng struct btrfs_root *reloc_root; 22385d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 22393fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 22403fd0a558SYan, Zheng u64 num_bytes = 0; 22413fd0a558SYan, Zheng int ret; 22423fd0a558SYan, Zheng 22437585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 22443fd0a558SYan, Zheng rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22453fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 22467585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 22477585717fSChris Mason 22483fd0a558SYan, Zheng again: 22493fd0a558SYan, Zheng if (!err) { 22503fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 225108e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 225208e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22533fd0a558SYan, Zheng if (ret) 22543fd0a558SYan, Zheng err = ret; 22553fd0a558SYan, Zheng } 22563fd0a558SYan, Zheng 22577a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 22583612b495STsutomu Itoh if (IS_ERR(trans)) { 22593612b495STsutomu Itoh if (!err) 22603612b495STsutomu Itoh btrfs_block_rsv_release(rc->extent_root, 22613612b495STsutomu Itoh rc->block_rsv, num_bytes); 22623612b495STsutomu Itoh return PTR_ERR(trans); 22633612b495STsutomu Itoh } 22643fd0a558SYan, Zheng 22653fd0a558SYan, Zheng if (!err) { 22663fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 22673fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 22683fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, 22693fd0a558SYan, Zheng rc->block_rsv, num_bytes); 22703fd0a558SYan, Zheng goto again; 22713fd0a558SYan, Zheng } 22723fd0a558SYan, Zheng } 22733fd0a558SYan, Zheng 22743fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 22753fd0a558SYan, Zheng 22763fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 22773fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 22783fd0a558SYan, Zheng struct btrfs_root, root_list); 22793fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 22803fd0a558SYan, Zheng 22813fd0a558SYan, Zheng root = read_fs_root(reloc_root->fs_info, 22823fd0a558SYan, Zheng reloc_root->root_key.offset); 22833fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 22843fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 22853fd0a558SYan, Zheng 22863fd0a558SYan, Zheng /* 22873fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 22883fd0a558SYan, Zheng * knows it should resumes merging 22893fd0a558SYan, Zheng */ 22903fd0a558SYan, Zheng if (!err) 22913fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 22923fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 22933fd0a558SYan, Zheng 22943fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 22953fd0a558SYan, Zheng } 22963fd0a558SYan, Zheng 22973fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 22983fd0a558SYan, Zheng 22993fd0a558SYan, Zheng if (!err) 23003fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 23013fd0a558SYan, Zheng else 23023fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 23033fd0a558SYan, Zheng return err; 23043fd0a558SYan, Zheng } 23053fd0a558SYan, Zheng 23063fd0a558SYan, Zheng static noinline_for_stack 2307aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2308aca1bba6SLiu Bo { 2309aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2310aca1bba6SLiu Bo 2311aca1bba6SLiu Bo while (!list_empty(list)) { 2312aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2313aca1bba6SLiu Bo root_list); 2314c974c464SWang Shilong __del_reloc_root(reloc_root); 2315aca1bba6SLiu Bo } 2316aca1bba6SLiu Bo } 2317aca1bba6SLiu Bo 2318aca1bba6SLiu Bo static noinline_for_stack 23193fd0a558SYan, Zheng int merge_reloc_roots(struct reloc_control *rc) 23203fd0a558SYan, Zheng { 23215d4f98a2SYan Zheng struct btrfs_root *root; 23225d4f98a2SYan Zheng struct btrfs_root *reloc_root; 23235bc7247aSMiao Xie u64 last_snap; 23245bc7247aSMiao Xie u64 otransid; 23255bc7247aSMiao Xie u64 objectid; 23263fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23273fd0a558SYan, Zheng int found = 0; 2328aca1bba6SLiu Bo int ret = 0; 23293fd0a558SYan, Zheng again: 23303fd0a558SYan, Zheng root = rc->extent_root; 23317585717fSChris Mason 23327585717fSChris Mason /* 23337585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 23347585717fSChris Mason * we have to make sure nobody is in the middle of 23357585717fSChris Mason * adding their roots to the list while we are 23367585717fSChris Mason * doing this splice 23377585717fSChris Mason */ 23387585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 23393fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 23407585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 23415d4f98a2SYan Zheng 23423fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 23433fd0a558SYan, Zheng found = 1; 23443fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 23453fd0a558SYan, Zheng struct btrfs_root, root_list); 23465d4f98a2SYan Zheng 23475d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 23485d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, 23495d4f98a2SYan Zheng reloc_root->root_key.offset); 23505d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 23515d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 23525d4f98a2SYan Zheng 23533fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2354b37b39cdSJosef Bacik if (ret) { 235525e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 235625e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 235725e293c2SWang Shilong &reloc_roots); 2358aca1bba6SLiu Bo goto out; 2359b37b39cdSJosef Bacik } 23603fd0a558SYan, Zheng } else { 23613fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 23623fd0a558SYan, Zheng } 23635bc7247aSMiao Xie 23645bc7247aSMiao Xie /* 23655bc7247aSMiao Xie * we keep the old last snapshod transid in rtranid when we 23665bc7247aSMiao Xie * created the relocation tree. 23675bc7247aSMiao Xie */ 23685bc7247aSMiao Xie last_snap = btrfs_root_rtransid(&reloc_root->root_item); 23695bc7247aSMiao Xie otransid = btrfs_root_otransid(&reloc_root->root_item); 23705bc7247aSMiao Xie objectid = reloc_root->root_key.offset; 23715bc7247aSMiao Xie 23722c536799SJeff Mahoney ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1); 2373aca1bba6SLiu Bo if (ret < 0) { 2374aca1bba6SLiu Bo if (list_empty(&reloc_root->root_list)) 2375aca1bba6SLiu Bo list_add_tail(&reloc_root->root_list, 2376aca1bba6SLiu Bo &reloc_roots); 2377aca1bba6SLiu Bo goto out; 2378aca1bba6SLiu Bo } 23795d4f98a2SYan Zheng } 23805d4f98a2SYan Zheng 23813fd0a558SYan, Zheng if (found) { 23823fd0a558SYan, Zheng found = 0; 23833fd0a558SYan, Zheng goto again; 23845d4f98a2SYan Zheng } 2385aca1bba6SLiu Bo out: 2386aca1bba6SLiu Bo if (ret) { 2387aca1bba6SLiu Bo btrfs_std_error(root->fs_info, ret); 2388aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2389aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2390467bb1d2SWang Shilong 2391467bb1d2SWang Shilong /* new reloc root may be added */ 2392467bb1d2SWang Shilong mutex_lock(&root->fs_info->reloc_mutex); 2393467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 2394467bb1d2SWang Shilong mutex_unlock(&root->fs_info->reloc_mutex); 2395467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2396467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2397aca1bba6SLiu Bo } 2398aca1bba6SLiu Bo 23995d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 2400aca1bba6SLiu Bo return ret; 24015d4f98a2SYan Zheng } 24025d4f98a2SYan Zheng 24035d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 24045d4f98a2SYan Zheng { 24055d4f98a2SYan Zheng struct tree_block *block; 24065d4f98a2SYan Zheng struct rb_node *rb_node; 24075d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 24085d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 24095d4f98a2SYan Zheng rb_erase(rb_node, blocks); 24105d4f98a2SYan Zheng kfree(block); 24115d4f98a2SYan Zheng } 24125d4f98a2SYan Zheng } 24135d4f98a2SYan Zheng 24145d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 24155d4f98a2SYan Zheng struct btrfs_root *reloc_root) 24165d4f98a2SYan Zheng { 24175d4f98a2SYan Zheng struct btrfs_root *root; 24185d4f98a2SYan Zheng 24195d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 24205d4f98a2SYan Zheng return 0; 24215d4f98a2SYan Zheng 24225d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset); 24235d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24245d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24255d4f98a2SYan Zheng 24265d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 24275d4f98a2SYan Zheng } 24285d4f98a2SYan Zheng 24293fd0a558SYan, Zheng static noinline_for_stack 24303fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 24313fd0a558SYan, Zheng struct reloc_control *rc, 24325d4f98a2SYan Zheng struct backref_node *node, 2433dc4103f9SWang Shilong struct backref_edge *edges[]) 24345d4f98a2SYan Zheng { 24355d4f98a2SYan Zheng struct backref_node *next; 24365d4f98a2SYan Zheng struct btrfs_root *root; 24373fd0a558SYan, Zheng int index = 0; 24383fd0a558SYan, Zheng 24395d4f98a2SYan Zheng next = node; 24405d4f98a2SYan Zheng while (1) { 24415d4f98a2SYan Zheng cond_resched(); 24425d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 24435d4f98a2SYan Zheng root = next->root; 24443fd0a558SYan, Zheng BUG_ON(!root); 244527cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 24465d4f98a2SYan Zheng 24475d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 24485d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 24495d4f98a2SYan Zheng break; 24505d4f98a2SYan Zheng } 24515d4f98a2SYan Zheng 24525d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 24533fd0a558SYan, Zheng root = root->reloc_root; 24543fd0a558SYan, Zheng 24553fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 24563fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 24573fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 24583fd0a558SYan, Zheng next->new_bytenr = root->node->start; 24593fd0a558SYan, Zheng next->root = root; 24603fd0a558SYan, Zheng list_add_tail(&next->list, 24613fd0a558SYan, Zheng &rc->backref_cache.changed); 24623fd0a558SYan, Zheng __mark_block_processed(rc, next); 24635d4f98a2SYan Zheng break; 24645d4f98a2SYan Zheng } 24655d4f98a2SYan Zheng 24663fd0a558SYan, Zheng WARN_ON(1); 24675d4f98a2SYan Zheng root = NULL; 24685d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 24695d4f98a2SYan Zheng if (!next || next->level <= node->level) 24705d4f98a2SYan Zheng break; 24715d4f98a2SYan Zheng } 24723fd0a558SYan, Zheng if (!root) 24733fd0a558SYan, Zheng return NULL; 24745d4f98a2SYan Zheng 24753fd0a558SYan, Zheng next = node; 24763fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 24773fd0a558SYan, Zheng while (1) { 24783fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 24793fd0a558SYan, Zheng if (--index < 0) 24803fd0a558SYan, Zheng break; 24813fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 24823fd0a558SYan, Zheng } 24835d4f98a2SYan Zheng return root; 24845d4f98a2SYan Zheng } 24855d4f98a2SYan Zheng 24863fd0a558SYan, Zheng /* 24873fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 24883fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 24893fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 24903fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 24913fd0a558SYan, Zheng */ 24925d4f98a2SYan Zheng static noinline_for_stack 24935d4f98a2SYan Zheng struct btrfs_root *select_one_root(struct btrfs_trans_handle *trans, 24945d4f98a2SYan Zheng struct backref_node *node) 24955d4f98a2SYan Zheng { 24963fd0a558SYan, Zheng struct backref_node *next; 24973fd0a558SYan, Zheng struct btrfs_root *root; 24983fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 24995d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25003fd0a558SYan, Zheng int index = 0; 25013fd0a558SYan, Zheng 25023fd0a558SYan, Zheng next = node; 25033fd0a558SYan, Zheng while (1) { 25043fd0a558SYan, Zheng cond_resched(); 25053fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 25063fd0a558SYan, Zheng root = next->root; 25073fd0a558SYan, Zheng BUG_ON(!root); 25083fd0a558SYan, Zheng 250925985edcSLucas De Marchi /* no other choice for non-references counted tree */ 251027cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 25113fd0a558SYan, Zheng return root; 25123fd0a558SYan, Zheng 25133fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 25143fd0a558SYan, Zheng fs_root = root; 25153fd0a558SYan, Zheng 25163fd0a558SYan, Zheng if (next != node) 25173fd0a558SYan, Zheng return NULL; 25183fd0a558SYan, Zheng 25193fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 25203fd0a558SYan, Zheng if (!next || next->level <= node->level) 25213fd0a558SYan, Zheng break; 25223fd0a558SYan, Zheng } 25233fd0a558SYan, Zheng 25243fd0a558SYan, Zheng if (!fs_root) 25253fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 25263fd0a558SYan, Zheng return fs_root; 25275d4f98a2SYan Zheng } 25285d4f98a2SYan Zheng 25295d4f98a2SYan Zheng static noinline_for_stack 25303fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 25313fd0a558SYan, Zheng struct backref_node *node, int reserve) 25325d4f98a2SYan Zheng { 25333fd0a558SYan, Zheng struct backref_node *next = node; 25343fd0a558SYan, Zheng struct backref_edge *edge; 25353fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25363fd0a558SYan, Zheng u64 num_bytes = 0; 25373fd0a558SYan, Zheng int index = 0; 25385d4f98a2SYan Zheng 25393fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 25403fd0a558SYan, Zheng 25413fd0a558SYan, Zheng while (next) { 25423fd0a558SYan, Zheng cond_resched(); 25435d4f98a2SYan Zheng while (1) { 25443fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 25455d4f98a2SYan Zheng break; 25465d4f98a2SYan Zheng 2547707e8a07SDavid Sterba num_bytes += rc->extent_root->nodesize; 25483fd0a558SYan, Zheng 25493fd0a558SYan, Zheng if (list_empty(&next->upper)) 25503fd0a558SYan, Zheng break; 25513fd0a558SYan, Zheng 25523fd0a558SYan, Zheng edge = list_entry(next->upper.next, 25533fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 25543fd0a558SYan, Zheng edges[index++] = edge; 25553fd0a558SYan, Zheng next = edge->node[UPPER]; 25565d4f98a2SYan Zheng } 25573fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 25583fd0a558SYan, Zheng } 25593fd0a558SYan, Zheng return num_bytes; 25603fd0a558SYan, Zheng } 25613fd0a558SYan, Zheng 25623fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 25633fd0a558SYan, Zheng struct reloc_control *rc, 25643fd0a558SYan, Zheng struct backref_node *node) 25653fd0a558SYan, Zheng { 25663fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 25673fd0a558SYan, Zheng u64 num_bytes; 25683fd0a558SYan, Zheng int ret; 25690647bf56SWang Shilong u64 tmp; 25703fd0a558SYan, Zheng 25713fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 25723fd0a558SYan, Zheng 25733fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 25740647bf56SWang Shilong rc->reserved_bytes += num_bytes; 25750647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 257608e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 25773fd0a558SYan, Zheng if (ret) { 25780647bf56SWang Shilong if (ret == -EAGAIN) { 25790647bf56SWang Shilong tmp = rc->extent_root->nodesize * 25800647bf56SWang Shilong RELOCATION_RESERVED_NODES; 25810647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 25820647bf56SWang Shilong tmp <<= 1; 25830647bf56SWang Shilong /* 25840647bf56SWang Shilong * only one thread can access block_rsv at this point, 25850647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 25860647bf56SWang Shilong * we expand more reservation size here to allow enough 25870647bf56SWang Shilong * space for relocation and we will return eailer in 25880647bf56SWang Shilong * enospc case. 25890647bf56SWang Shilong */ 25900647bf56SWang Shilong rc->block_rsv->size = tmp + rc->extent_root->nodesize * 25910647bf56SWang Shilong RELOCATION_RESERVED_NODES; 25920647bf56SWang Shilong } 25933fd0a558SYan, Zheng return ret; 25943fd0a558SYan, Zheng } 25953fd0a558SYan, Zheng 25963fd0a558SYan, Zheng return 0; 25973fd0a558SYan, Zheng } 25983fd0a558SYan, Zheng 25995d4f98a2SYan Zheng /* 26005d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 26015d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 26025d4f98a2SYan Zheng * 26035d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 26045d4f98a2SYan Zheng * in that case this function just updates pointers. 26055d4f98a2SYan Zheng */ 26065d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 26073fd0a558SYan, Zheng struct reloc_control *rc, 26085d4f98a2SYan Zheng struct backref_node *node, 26095d4f98a2SYan Zheng struct btrfs_key *key, 26105d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 26115d4f98a2SYan Zheng { 26125d4f98a2SYan Zheng struct backref_node *upper; 26135d4f98a2SYan Zheng struct backref_edge *edge; 26145d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26155d4f98a2SYan Zheng struct btrfs_root *root; 26165d4f98a2SYan Zheng struct extent_buffer *eb; 26175d4f98a2SYan Zheng u32 blocksize; 26185d4f98a2SYan Zheng u64 bytenr; 26195d4f98a2SYan Zheng u64 generation; 26205d4f98a2SYan Zheng int slot; 26215d4f98a2SYan Zheng int ret; 26225d4f98a2SYan Zheng int err = 0; 26235d4f98a2SYan Zheng 26245d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 26255d4f98a2SYan Zheng 26265d4f98a2SYan Zheng path->lowest_level = node->level + 1; 26273fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 26285d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 26295d4f98a2SYan Zheng cond_resched(); 26305d4f98a2SYan Zheng 26315d4f98a2SYan Zheng upper = edge->node[UPPER]; 2632dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 26333fd0a558SYan, Zheng BUG_ON(!root); 26345d4f98a2SYan Zheng 26353fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 26363fd0a558SYan, Zheng if (!lowest) { 26373fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 26383fd0a558SYan, Zheng upper->level, &slot); 26393fd0a558SYan, Zheng BUG_ON(ret); 26403fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 26413fd0a558SYan, Zheng if (node->eb->start == bytenr) 26423fd0a558SYan, Zheng goto next; 26433fd0a558SYan, Zheng } 26445d4f98a2SYan Zheng drop_node_buffer(upper); 26453fd0a558SYan, Zheng } 26465d4f98a2SYan Zheng 26475d4f98a2SYan Zheng if (!upper->eb) { 26485d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 26495d4f98a2SYan Zheng if (ret < 0) { 26505d4f98a2SYan Zheng err = ret; 26515d4f98a2SYan Zheng break; 26525d4f98a2SYan Zheng } 26535d4f98a2SYan Zheng BUG_ON(ret > 0); 26545d4f98a2SYan Zheng 26553fd0a558SYan, Zheng if (!upper->eb) { 26563fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 26573fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 26583fd0a558SYan, Zheng } else { 26593fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 26603fd0a558SYan, Zheng } 26613fd0a558SYan, Zheng 26623fd0a558SYan, Zheng upper->locked = 1; 26633fd0a558SYan, Zheng path->locks[upper->level] = 0; 26643fd0a558SYan, Zheng 26655d4f98a2SYan Zheng slot = path->slots[upper->level]; 2666b3b4aa74SDavid Sterba btrfs_release_path(path); 26675d4f98a2SYan Zheng } else { 26685d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 26695d4f98a2SYan Zheng &slot); 26705d4f98a2SYan Zheng BUG_ON(ret); 26715d4f98a2SYan Zheng } 26725d4f98a2SYan Zheng 26735d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 26743fd0a558SYan, Zheng if (lowest) { 26753fd0a558SYan, Zheng BUG_ON(bytenr != node->bytenr); 26765d4f98a2SYan Zheng } else { 26773fd0a558SYan, Zheng if (node->eb->start == bytenr) 26783fd0a558SYan, Zheng goto next; 26795d4f98a2SYan Zheng } 26805d4f98a2SYan Zheng 2681707e8a07SDavid Sterba blocksize = root->nodesize; 26825d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 26835d4f98a2SYan Zheng eb = read_tree_block(root, bytenr, blocksize, generation); 2684416bc658SJosef Bacik if (!eb || !extent_buffer_uptodate(eb)) { 2685416bc658SJosef Bacik free_extent_buffer(eb); 268697d9a8a4STsutomu Itoh err = -EIO; 268797d9a8a4STsutomu Itoh goto next; 268897d9a8a4STsutomu Itoh } 26895d4f98a2SYan Zheng btrfs_tree_lock(eb); 26905d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 26915d4f98a2SYan Zheng 26925d4f98a2SYan Zheng if (!node->eb) { 26935d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 26945d4f98a2SYan Zheng slot, &eb); 26953fd0a558SYan, Zheng btrfs_tree_unlock(eb); 26963fd0a558SYan, Zheng free_extent_buffer(eb); 26975d4f98a2SYan Zheng if (ret < 0) { 26985d4f98a2SYan Zheng err = ret; 26993fd0a558SYan, Zheng goto next; 27005d4f98a2SYan Zheng } 27013fd0a558SYan, Zheng BUG_ON(node->eb != eb); 27025d4f98a2SYan Zheng } else { 27035d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 27045d4f98a2SYan Zheng node->eb->start); 27055d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 27065d4f98a2SYan Zheng trans->transid); 27075d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 27085d4f98a2SYan Zheng 27095d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, 27105d4f98a2SYan Zheng node->eb->start, blocksize, 27115d4f98a2SYan Zheng upper->eb->start, 27125d4f98a2SYan Zheng btrfs_header_owner(upper->eb), 271366d7e7f0SArne Jansen node->level, 0, 1); 27145d4f98a2SYan Zheng BUG_ON(ret); 27155d4f98a2SYan Zheng 27165d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 27175d4f98a2SYan Zheng BUG_ON(ret); 27185d4f98a2SYan Zheng } 27193fd0a558SYan, Zheng next: 27203fd0a558SYan, Zheng if (!upper->pending) 27213fd0a558SYan, Zheng drop_node_buffer(upper); 27223fd0a558SYan, Zheng else 27233fd0a558SYan, Zheng unlock_node_buffer(upper); 27243fd0a558SYan, Zheng if (err) 27253fd0a558SYan, Zheng break; 27265d4f98a2SYan Zheng } 27273fd0a558SYan, Zheng 27283fd0a558SYan, Zheng if (!err && node->pending) { 27293fd0a558SYan, Zheng drop_node_buffer(node); 27303fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 27313fd0a558SYan, Zheng node->pending = 0; 27325d4f98a2SYan Zheng } 27333fd0a558SYan, Zheng 27345d4f98a2SYan Zheng path->lowest_level = 0; 27353fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 27365d4f98a2SYan Zheng return err; 27375d4f98a2SYan Zheng } 27385d4f98a2SYan Zheng 27395d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 27403fd0a558SYan, Zheng struct reloc_control *rc, 27415d4f98a2SYan Zheng struct backref_node *node, 27425d4f98a2SYan Zheng struct btrfs_path *path) 27435d4f98a2SYan Zheng { 27445d4f98a2SYan Zheng struct btrfs_key key; 27455d4f98a2SYan Zheng 27465d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 27473fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 27485d4f98a2SYan Zheng } 27495d4f98a2SYan Zheng 27505d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 27513fd0a558SYan, Zheng struct reloc_control *rc, 27523fd0a558SYan, Zheng struct btrfs_path *path, int err) 27535d4f98a2SYan Zheng { 27543fd0a558SYan, Zheng LIST_HEAD(list); 27553fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 27565d4f98a2SYan Zheng struct backref_node *node; 27575d4f98a2SYan Zheng int level; 27585d4f98a2SYan Zheng int ret; 27595d4f98a2SYan Zheng 27605d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 27615d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 27625d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 27633fd0a558SYan, Zheng struct backref_node, list); 27643fd0a558SYan, Zheng list_move_tail(&node->list, &list); 27653fd0a558SYan, Zheng BUG_ON(!node->pending); 27665d4f98a2SYan Zheng 27673fd0a558SYan, Zheng if (!err) { 27683fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 27695d4f98a2SYan Zheng if (ret < 0) 27705d4f98a2SYan Zheng err = ret; 27715d4f98a2SYan Zheng } 27725d4f98a2SYan Zheng } 27733fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 27743fd0a558SYan, Zheng } 27755d4f98a2SYan Zheng return err; 27765d4f98a2SYan Zheng } 27775d4f98a2SYan Zheng 27785d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 27793fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 27803fd0a558SYan, Zheng { 27813fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 27823fd0a558SYan, Zheng EXTENT_DIRTY, GFP_NOFS); 27833fd0a558SYan, Zheng } 27843fd0a558SYan, Zheng 27853fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 27865d4f98a2SYan Zheng struct backref_node *node) 27875d4f98a2SYan Zheng { 27885d4f98a2SYan Zheng u32 blocksize; 27895d4f98a2SYan Zheng if (node->level == 0 || 27905d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2791707e8a07SDavid Sterba blocksize = rc->extent_root->nodesize; 27923fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 27935d4f98a2SYan Zheng } 27945d4f98a2SYan Zheng node->processed = 1; 27955d4f98a2SYan Zheng } 27965d4f98a2SYan Zheng 27975d4f98a2SYan Zheng /* 27985d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 27995d4f98a2SYan Zheng * as processed. 28005d4f98a2SYan Zheng */ 28015d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 28025d4f98a2SYan Zheng struct backref_node *node) 28035d4f98a2SYan Zheng { 28045d4f98a2SYan Zheng struct backref_node *next = node; 28055d4f98a2SYan Zheng struct backref_edge *edge; 28065d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 28075d4f98a2SYan Zheng int index = 0; 28085d4f98a2SYan Zheng 28095d4f98a2SYan Zheng while (next) { 28105d4f98a2SYan Zheng cond_resched(); 28115d4f98a2SYan Zheng while (1) { 28125d4f98a2SYan Zheng if (next->processed) 28135d4f98a2SYan Zheng break; 28145d4f98a2SYan Zheng 28153fd0a558SYan, Zheng __mark_block_processed(rc, next); 28165d4f98a2SYan Zheng 28175d4f98a2SYan Zheng if (list_empty(&next->upper)) 28185d4f98a2SYan Zheng break; 28195d4f98a2SYan Zheng 28205d4f98a2SYan Zheng edge = list_entry(next->upper.next, 28215d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 28225d4f98a2SYan Zheng edges[index++] = edge; 28235d4f98a2SYan Zheng next = edge->node[UPPER]; 28245d4f98a2SYan Zheng } 28255d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 28265d4f98a2SYan Zheng } 28275d4f98a2SYan Zheng } 28285d4f98a2SYan Zheng 28295d4f98a2SYan Zheng static int tree_block_processed(u64 bytenr, u32 blocksize, 28305d4f98a2SYan Zheng struct reloc_control *rc) 28315d4f98a2SYan Zheng { 28325d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 28339655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 28345d4f98a2SYan Zheng return 1; 28355d4f98a2SYan Zheng return 0; 28365d4f98a2SYan Zheng } 28375d4f98a2SYan Zheng 28385d4f98a2SYan Zheng static int get_tree_block_key(struct reloc_control *rc, 28395d4f98a2SYan Zheng struct tree_block *block) 28405d4f98a2SYan Zheng { 28415d4f98a2SYan Zheng struct extent_buffer *eb; 28425d4f98a2SYan Zheng 28435d4f98a2SYan Zheng BUG_ON(block->key_ready); 28445d4f98a2SYan Zheng eb = read_tree_block(rc->extent_root, block->bytenr, 28455d4f98a2SYan Zheng block->key.objectid, block->key.offset); 2846416bc658SJosef Bacik if (!eb || !extent_buffer_uptodate(eb)) { 2847416bc658SJosef Bacik free_extent_buffer(eb); 2848416bc658SJosef Bacik return -EIO; 2849416bc658SJosef Bacik } 28505d4f98a2SYan Zheng WARN_ON(btrfs_header_level(eb) != block->level); 28515d4f98a2SYan Zheng if (block->level == 0) 28525d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 28535d4f98a2SYan Zheng else 28545d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 28555d4f98a2SYan Zheng free_extent_buffer(eb); 28565d4f98a2SYan Zheng block->key_ready = 1; 28575d4f98a2SYan Zheng return 0; 28585d4f98a2SYan Zheng } 28595d4f98a2SYan Zheng 28605d4f98a2SYan Zheng /* 28615d4f98a2SYan Zheng * helper function to relocate a tree block 28625d4f98a2SYan Zheng */ 28635d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 28645d4f98a2SYan Zheng struct reloc_control *rc, 28655d4f98a2SYan Zheng struct backref_node *node, 28665d4f98a2SYan Zheng struct btrfs_key *key, 28675d4f98a2SYan Zheng struct btrfs_path *path) 28685d4f98a2SYan Zheng { 28695d4f98a2SYan Zheng struct btrfs_root *root; 28703fd0a558SYan, Zheng int ret = 0; 28715d4f98a2SYan Zheng 28723fd0a558SYan, Zheng if (!node) 28735d4f98a2SYan Zheng return 0; 28743fd0a558SYan, Zheng 28753fd0a558SYan, Zheng BUG_ON(node->processed); 28763fd0a558SYan, Zheng root = select_one_root(trans, node); 28773fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 28783fd0a558SYan, Zheng update_processed_blocks(rc, node); 28793fd0a558SYan, Zheng goto out; 28805d4f98a2SYan Zheng } 28815d4f98a2SYan Zheng 288227cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 28833fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 28843fd0a558SYan, Zheng if (ret) 28855d4f98a2SYan Zheng goto out; 28865d4f98a2SYan Zheng } 28873fd0a558SYan, Zheng 28883fd0a558SYan, Zheng if (root) { 288927cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 28903fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 28913fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 28923fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 28933fd0a558SYan, Zheng root = root->reloc_root; 28943fd0a558SYan, Zheng node->new_bytenr = root->node->start; 28953fd0a558SYan, Zheng node->root = root; 28963fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 28973fd0a558SYan, Zheng } else { 28985d4f98a2SYan Zheng path->lowest_level = node->level; 28995d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2900b3b4aa74SDavid Sterba btrfs_release_path(path); 29013fd0a558SYan, Zheng if (ret > 0) 29025d4f98a2SYan Zheng ret = 0; 29033fd0a558SYan, Zheng } 29043fd0a558SYan, Zheng if (!ret) 29053fd0a558SYan, Zheng update_processed_blocks(rc, node); 29063fd0a558SYan, Zheng } else { 29073fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 29083fd0a558SYan, Zheng } 29095d4f98a2SYan Zheng out: 29100647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 29113fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 29125d4f98a2SYan Zheng return ret; 29135d4f98a2SYan Zheng } 29145d4f98a2SYan Zheng 29155d4f98a2SYan Zheng /* 29165d4f98a2SYan Zheng * relocate a list of blocks 29175d4f98a2SYan Zheng */ 29185d4f98a2SYan Zheng static noinline_for_stack 29195d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 29205d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 29215d4f98a2SYan Zheng { 29225d4f98a2SYan Zheng struct backref_node *node; 29235d4f98a2SYan Zheng struct btrfs_path *path; 29245d4f98a2SYan Zheng struct tree_block *block; 29255d4f98a2SYan Zheng struct rb_node *rb_node; 29265d4f98a2SYan Zheng int ret; 29275d4f98a2SYan Zheng int err = 0; 29285d4f98a2SYan Zheng 29295d4f98a2SYan Zheng path = btrfs_alloc_path(); 2930e1a12670SLiu Bo if (!path) { 2931e1a12670SLiu Bo err = -ENOMEM; 293234c2b290SDavid Sterba goto out_free_blocks; 2933e1a12670SLiu Bo } 29345d4f98a2SYan Zheng 29355d4f98a2SYan Zheng rb_node = rb_first(blocks); 29365d4f98a2SYan Zheng while (rb_node) { 29375d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 29385d4f98a2SYan Zheng if (!block->key_ready) 2939453848a0SDavid Sterba readahead_tree_block(rc->extent_root, block->bytenr, 2940453848a0SDavid Sterba block->key.objectid); 29415d4f98a2SYan Zheng rb_node = rb_next(rb_node); 29425d4f98a2SYan Zheng } 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); 294734c2b290SDavid Sterba if (!block->key_ready) { 294834c2b290SDavid Sterba err = get_tree_block_key(rc, block); 294934c2b290SDavid Sterba if (err) 295034c2b290SDavid Sterba goto out_free_path; 295134c2b290SDavid Sterba } 29525d4f98a2SYan Zheng rb_node = rb_next(rb_node); 29535d4f98a2SYan Zheng } 29545d4f98a2SYan Zheng 29555d4f98a2SYan Zheng rb_node = rb_first(blocks); 29565d4f98a2SYan Zheng while (rb_node) { 29575d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 29585d4f98a2SYan Zheng 29593fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 29605d4f98a2SYan Zheng block->level, block->bytenr); 29615d4f98a2SYan Zheng if (IS_ERR(node)) { 29625d4f98a2SYan Zheng err = PTR_ERR(node); 29635d4f98a2SYan Zheng goto out; 29645d4f98a2SYan Zheng } 29655d4f98a2SYan Zheng 29665d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 29675d4f98a2SYan Zheng path); 29685d4f98a2SYan Zheng if (ret < 0) { 29693fd0a558SYan, Zheng if (ret != -EAGAIN || rb_node == rb_first(blocks)) 29705d4f98a2SYan Zheng err = ret; 29715d4f98a2SYan Zheng goto out; 29725d4f98a2SYan Zheng } 29735d4f98a2SYan Zheng rb_node = rb_next(rb_node); 29745d4f98a2SYan Zheng } 29755d4f98a2SYan Zheng out: 29763fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 29775d4f98a2SYan Zheng 297834c2b290SDavid Sterba out_free_path: 29795d4f98a2SYan Zheng btrfs_free_path(path); 298034c2b290SDavid Sterba out_free_blocks: 2981e1a12670SLiu Bo free_block_list(blocks); 29825d4f98a2SYan Zheng return err; 29835d4f98a2SYan Zheng } 29845d4f98a2SYan Zheng 29855d4f98a2SYan Zheng static noinline_for_stack 2986efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 2987efa56464SYan, Zheng struct file_extent_cluster *cluster) 2988efa56464SYan, Zheng { 2989efa56464SYan, Zheng u64 alloc_hint = 0; 2990efa56464SYan, Zheng u64 start; 2991efa56464SYan, Zheng u64 end; 2992efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 2993efa56464SYan, Zheng u64 num_bytes; 2994efa56464SYan, Zheng int nr = 0; 2995efa56464SYan, Zheng int ret = 0; 2996efa56464SYan, Zheng 2997efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 2998efa56464SYan, Zheng mutex_lock(&inode->i_mutex); 2999efa56464SYan, Zheng 3000efa56464SYan, Zheng ret = btrfs_check_data_free_space(inode, cluster->end + 3001efa56464SYan, Zheng 1 - cluster->start); 3002efa56464SYan, Zheng if (ret) 3003efa56464SYan, Zheng goto out; 3004efa56464SYan, Zheng 3005efa56464SYan, Zheng while (nr < cluster->nr) { 3006efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3007efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3008efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3009efa56464SYan, Zheng else 3010efa56464SYan, Zheng end = cluster->end - offset; 3011efa56464SYan, Zheng 3012d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3013efa56464SYan, Zheng num_bytes = end + 1 - start; 3014efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3015efa56464SYan, Zheng num_bytes, num_bytes, 3016efa56464SYan, Zheng end + 1, &alloc_hint); 3017d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3018efa56464SYan, Zheng if (ret) 3019efa56464SYan, Zheng break; 3020efa56464SYan, Zheng nr++; 3021efa56464SYan, Zheng } 3022efa56464SYan, Zheng btrfs_free_reserved_data_space(inode, cluster->end + 3023efa56464SYan, Zheng 1 - cluster->start); 3024efa56464SYan, Zheng out: 3025efa56464SYan, Zheng mutex_unlock(&inode->i_mutex); 3026efa56464SYan, Zheng return ret; 3027efa56464SYan, Zheng } 3028efa56464SYan, Zheng 3029efa56464SYan, Zheng static noinline_for_stack 30300257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 30310257bb82SYan, Zheng u64 block_start) 30325d4f98a2SYan Zheng { 30335d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 30345d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 30355d4f98a2SYan Zheng struct extent_map *em; 30360257bb82SYan, Zheng int ret = 0; 30375d4f98a2SYan Zheng 3038172ddd60SDavid Sterba em = alloc_extent_map(); 30390257bb82SYan, Zheng if (!em) 30400257bb82SYan, Zheng return -ENOMEM; 30410257bb82SYan, Zheng 30425d4f98a2SYan Zheng em->start = start; 30430257bb82SYan, Zheng em->len = end + 1 - start; 30440257bb82SYan, Zheng em->block_len = em->len; 30450257bb82SYan, Zheng em->block_start = block_start; 30465d4f98a2SYan Zheng em->bdev = root->fs_info->fs_devices->latest_bdev; 30475d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 30485d4f98a2SYan Zheng 3049d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 30505d4f98a2SYan Zheng while (1) { 3051890871beSChris Mason write_lock(&em_tree->lock); 305209a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3053890871beSChris Mason write_unlock(&em_tree->lock); 30545d4f98a2SYan Zheng if (ret != -EEXIST) { 30555d4f98a2SYan Zheng free_extent_map(em); 30565d4f98a2SYan Zheng break; 30575d4f98a2SYan Zheng } 30585d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 0); 30595d4f98a2SYan Zheng } 3060d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 30610257bb82SYan, Zheng return ret; 30620257bb82SYan, Zheng } 30635d4f98a2SYan Zheng 30640257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 30650257bb82SYan, Zheng struct file_extent_cluster *cluster) 30660257bb82SYan, Zheng { 30670257bb82SYan, Zheng u64 page_start; 30680257bb82SYan, Zheng u64 page_end; 30690257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 30700257bb82SYan, Zheng unsigned long index; 30710257bb82SYan, Zheng unsigned long last_index; 30720257bb82SYan, Zheng struct page *page; 30730257bb82SYan, Zheng struct file_ra_state *ra; 30743b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 30750257bb82SYan, Zheng int nr = 0; 30760257bb82SYan, Zheng int ret = 0; 30770257bb82SYan, Zheng 30780257bb82SYan, Zheng if (!cluster->nr) 30790257bb82SYan, Zheng return 0; 30800257bb82SYan, Zheng 30810257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 30820257bb82SYan, Zheng if (!ra) 30830257bb82SYan, Zheng return -ENOMEM; 30840257bb82SYan, Zheng 3085efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 30860257bb82SYan, Zheng if (ret) 3087efa56464SYan, Zheng goto out; 30880257bb82SYan, Zheng 30890257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 30900257bb82SYan, Zheng 3091efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3092efa56464SYan, Zheng cluster->end - offset, cluster->start); 3093efa56464SYan, Zheng if (ret) 3094efa56464SYan, Zheng goto out; 3095efa56464SYan, Zheng 3096efa56464SYan, Zheng index = (cluster->start - offset) >> PAGE_CACHE_SHIFT; 3097efa56464SYan, Zheng last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT; 30980257bb82SYan, Zheng while (index <= last_index) { 3099efa56464SYan, Zheng ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE); 3100efa56464SYan, Zheng if (ret) 3101efa56464SYan, Zheng goto out; 3102efa56464SYan, Zheng 31030257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 31040257bb82SYan, Zheng if (!page) { 31050257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 31060257bb82SYan, Zheng ra, NULL, index, 31070257bb82SYan, Zheng last_index + 1 - index); 3108a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 31093b16a4e3SJosef Bacik mask); 31100257bb82SYan, Zheng if (!page) { 3111efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 3112efa56464SYan, Zheng PAGE_CACHE_SIZE); 31130257bb82SYan, Zheng ret = -ENOMEM; 3114efa56464SYan, Zheng goto out; 31150257bb82SYan, Zheng } 31160257bb82SYan, Zheng } 31170257bb82SYan, Zheng 31180257bb82SYan, Zheng if (PageReadahead(page)) { 31190257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 31200257bb82SYan, Zheng ra, NULL, page, index, 31210257bb82SYan, Zheng last_index + 1 - index); 31220257bb82SYan, Zheng } 31230257bb82SYan, Zheng 31240257bb82SYan, Zheng if (!PageUptodate(page)) { 31250257bb82SYan, Zheng btrfs_readpage(NULL, page); 31260257bb82SYan, Zheng lock_page(page); 31270257bb82SYan, Zheng if (!PageUptodate(page)) { 31280257bb82SYan, Zheng unlock_page(page); 31290257bb82SYan, Zheng page_cache_release(page); 3130efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 3131efa56464SYan, Zheng PAGE_CACHE_SIZE); 31320257bb82SYan, Zheng ret = -EIO; 3133efa56464SYan, Zheng goto out; 31340257bb82SYan, Zheng } 31350257bb82SYan, Zheng } 31360257bb82SYan, Zheng 31374eee4fa4SMiao Xie page_start = page_offset(page); 31380257bb82SYan, Zheng page_end = page_start + PAGE_CACHE_SIZE - 1; 31390257bb82SYan, Zheng 3140d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 31410257bb82SYan, Zheng 31420257bb82SYan, Zheng set_page_extent_mapped(page); 31430257bb82SYan, Zheng 31440257bb82SYan, Zheng if (nr < cluster->nr && 31450257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 31460257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 31470257bb82SYan, Zheng page_start, page_end, 31480257bb82SYan, Zheng EXTENT_BOUNDARY, GFP_NOFS); 31490257bb82SYan, Zheng nr++; 31500257bb82SYan, Zheng } 31510257bb82SYan, Zheng 3152efa56464SYan, Zheng btrfs_set_extent_delalloc(inode, page_start, page_end, NULL); 31530257bb82SYan, Zheng set_page_dirty(page); 31540257bb82SYan, Zheng 31550257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3156d0082371SJeff Mahoney page_start, page_end); 31570257bb82SYan, Zheng unlock_page(page); 31580257bb82SYan, Zheng page_cache_release(page); 31590257bb82SYan, Zheng 31600257bb82SYan, Zheng index++; 3161efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 3162efa56464SYan, Zheng btrfs_throttle(BTRFS_I(inode)->root); 31630257bb82SYan, Zheng } 31640257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3165efa56464SYan, Zheng out: 31660257bb82SYan, Zheng kfree(ra); 31670257bb82SYan, Zheng return ret; 31680257bb82SYan, Zheng } 31690257bb82SYan, Zheng 31700257bb82SYan, Zheng static noinline_for_stack 31710257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 31720257bb82SYan, Zheng struct file_extent_cluster *cluster) 31730257bb82SYan, Zheng { 31740257bb82SYan, Zheng int ret; 31750257bb82SYan, Zheng 31760257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 31770257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31780257bb82SYan, Zheng if (ret) 31790257bb82SYan, Zheng return ret; 31800257bb82SYan, Zheng cluster->nr = 0; 31810257bb82SYan, Zheng } 31820257bb82SYan, Zheng 31830257bb82SYan, Zheng if (!cluster->nr) 31840257bb82SYan, Zheng cluster->start = extent_key->objectid; 31850257bb82SYan, Zheng else 31860257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 31870257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 31880257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 31890257bb82SYan, Zheng cluster->nr++; 31900257bb82SYan, Zheng 31910257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 31920257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31930257bb82SYan, Zheng if (ret) 31940257bb82SYan, Zheng return ret; 31950257bb82SYan, Zheng cluster->nr = 0; 31960257bb82SYan, Zheng } 31970257bb82SYan, Zheng return 0; 31985d4f98a2SYan Zheng } 31995d4f98a2SYan Zheng 32005d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 32015d4f98a2SYan Zheng static int get_ref_objectid_v0(struct reloc_control *rc, 32025d4f98a2SYan Zheng struct btrfs_path *path, 32035d4f98a2SYan Zheng struct btrfs_key *extent_key, 32045d4f98a2SYan Zheng u64 *ref_objectid, int *path_change) 32055d4f98a2SYan Zheng { 32065d4f98a2SYan Zheng struct btrfs_key key; 32075d4f98a2SYan Zheng struct extent_buffer *leaf; 32085d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 32095d4f98a2SYan Zheng int ret; 32105d4f98a2SYan Zheng int slot; 32115d4f98a2SYan Zheng 32125d4f98a2SYan Zheng leaf = path->nodes[0]; 32135d4f98a2SYan Zheng slot = path->slots[0]; 32145d4f98a2SYan Zheng while (1) { 32155d4f98a2SYan Zheng if (slot >= btrfs_header_nritems(leaf)) { 32165d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 32175d4f98a2SYan Zheng if (ret < 0) 32185d4f98a2SYan Zheng return ret; 32195d4f98a2SYan Zheng BUG_ON(ret > 0); 32205d4f98a2SYan Zheng leaf = path->nodes[0]; 32215d4f98a2SYan Zheng slot = path->slots[0]; 32225d4f98a2SYan Zheng if (path_change) 32235d4f98a2SYan Zheng *path_change = 1; 32245d4f98a2SYan Zheng } 32255d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 32265d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 32275d4f98a2SYan Zheng return -ENOENT; 32285d4f98a2SYan Zheng 32295d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_REF_V0_KEY) { 32305d4f98a2SYan Zheng slot++; 32315d4f98a2SYan Zheng continue; 32325d4f98a2SYan Zheng } 32335d4f98a2SYan Zheng ref0 = btrfs_item_ptr(leaf, slot, 32345d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 32355d4f98a2SYan Zheng *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0); 32365d4f98a2SYan Zheng break; 32375d4f98a2SYan Zheng } 32385d4f98a2SYan Zheng return 0; 32395d4f98a2SYan Zheng } 32405d4f98a2SYan Zheng #endif 32415d4f98a2SYan Zheng 32425d4f98a2SYan Zheng /* 32435d4f98a2SYan Zheng * helper to add a tree block to the list. 32445d4f98a2SYan Zheng * the major work is getting the generation and level of the block 32455d4f98a2SYan Zheng */ 32465d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 32475d4f98a2SYan Zheng struct btrfs_key *extent_key, 32485d4f98a2SYan Zheng struct btrfs_path *path, 32495d4f98a2SYan Zheng struct rb_root *blocks) 32505d4f98a2SYan Zheng { 32515d4f98a2SYan Zheng struct extent_buffer *eb; 32525d4f98a2SYan Zheng struct btrfs_extent_item *ei; 32535d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 32545d4f98a2SYan Zheng struct tree_block *block; 32555d4f98a2SYan Zheng struct rb_node *rb_node; 32565d4f98a2SYan Zheng u32 item_size; 32575d4f98a2SYan Zheng int level = -1; 32587fdf4b60SWang Shilong u64 generation; 32595d4f98a2SYan Zheng 32605d4f98a2SYan Zheng eb = path->nodes[0]; 32615d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 32625d4f98a2SYan Zheng 32633173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 32643173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 32655d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 32665d4f98a2SYan Zheng struct btrfs_extent_item); 32673173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 32685d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 32695d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 32705d4f98a2SYan Zheng } else { 32713173a18fSJosef Bacik level = (int)extent_key->offset; 32723173a18fSJosef Bacik } 32733173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 32743173a18fSJosef Bacik } else { 32755d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 32765d4f98a2SYan Zheng u64 ref_owner; 32775d4f98a2SYan Zheng int ret; 32785d4f98a2SYan Zheng 32795d4f98a2SYan Zheng BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 32805d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, extent_key, 32815d4f98a2SYan Zheng &ref_owner, NULL); 3282411fc6bcSAndi Kleen if (ret < 0) 3283411fc6bcSAndi Kleen return ret; 32845d4f98a2SYan Zheng BUG_ON(ref_owner >= BTRFS_MAX_LEVEL); 32855d4f98a2SYan Zheng level = (int)ref_owner; 32865d4f98a2SYan Zheng /* FIXME: get real generation */ 32875d4f98a2SYan Zheng generation = 0; 32885d4f98a2SYan Zheng #else 32895d4f98a2SYan Zheng BUG(); 32905d4f98a2SYan Zheng #endif 32915d4f98a2SYan Zheng } 32925d4f98a2SYan Zheng 3293b3b4aa74SDavid Sterba btrfs_release_path(path); 32945d4f98a2SYan Zheng 32955d4f98a2SYan Zheng BUG_ON(level == -1); 32965d4f98a2SYan Zheng 32975d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 32985d4f98a2SYan Zheng if (!block) 32995d4f98a2SYan Zheng return -ENOMEM; 33005d4f98a2SYan Zheng 33015d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3302707e8a07SDavid Sterba block->key.objectid = rc->extent_root->nodesize; 33035d4f98a2SYan Zheng block->key.offset = generation; 33045d4f98a2SYan Zheng block->level = level; 33055d4f98a2SYan Zheng block->key_ready = 0; 33065d4f98a2SYan Zheng 33075d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 330843c04fb1SJeff Mahoney if (rb_node) 330943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 33105d4f98a2SYan Zheng 33115d4f98a2SYan Zheng return 0; 33125d4f98a2SYan Zheng } 33135d4f98a2SYan Zheng 33145d4f98a2SYan Zheng /* 33155d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 33165d4f98a2SYan Zheng */ 33175d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 33185d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 33195d4f98a2SYan Zheng struct rb_root *blocks) 33205d4f98a2SYan Zheng { 33215d4f98a2SYan Zheng struct btrfs_path *path; 33225d4f98a2SYan Zheng struct btrfs_key key; 33235d4f98a2SYan Zheng int ret; 3324aee68ee5SJosef Bacik bool skinny = btrfs_fs_incompat(rc->extent_root->fs_info, 3325aee68ee5SJosef Bacik SKINNY_METADATA); 33265d4f98a2SYan Zheng 33275d4f98a2SYan Zheng if (tree_block_processed(bytenr, blocksize, rc)) 33285d4f98a2SYan Zheng return 0; 33295d4f98a2SYan Zheng 33305d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 33315d4f98a2SYan Zheng return 0; 33325d4f98a2SYan Zheng 33335d4f98a2SYan Zheng path = btrfs_alloc_path(); 33345d4f98a2SYan Zheng if (!path) 33355d4f98a2SYan Zheng return -ENOMEM; 3336aee68ee5SJosef Bacik again: 33375d4f98a2SYan Zheng key.objectid = bytenr; 3338aee68ee5SJosef Bacik if (skinny) { 3339aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3340aee68ee5SJosef Bacik key.offset = (u64)-1; 3341aee68ee5SJosef Bacik } else { 33425d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 33435d4f98a2SYan Zheng key.offset = blocksize; 3344aee68ee5SJosef Bacik } 33455d4f98a2SYan Zheng 33465d4f98a2SYan Zheng path->search_commit_root = 1; 33475d4f98a2SYan Zheng path->skip_locking = 1; 33485d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 33495d4f98a2SYan Zheng if (ret < 0) 33505d4f98a2SYan Zheng goto out; 33515d4f98a2SYan Zheng 3352aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3353aee68ee5SJosef Bacik if (path->slots[0]) { 3354aee68ee5SJosef Bacik path->slots[0]--; 3355aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3356aee68ee5SJosef Bacik path->slots[0]); 33573173a18fSJosef Bacik if (key.objectid == bytenr && 3358aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3359aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3360aee68ee5SJosef Bacik key.offset == blocksize))) 33613173a18fSJosef Bacik ret = 0; 33623173a18fSJosef Bacik } 3363aee68ee5SJosef Bacik 3364aee68ee5SJosef Bacik if (ret) { 3365aee68ee5SJosef Bacik skinny = false; 3366aee68ee5SJosef Bacik btrfs_release_path(path); 3367aee68ee5SJosef Bacik goto again; 3368aee68ee5SJosef Bacik } 3369aee68ee5SJosef Bacik } 33703173a18fSJosef Bacik BUG_ON(ret); 33713173a18fSJosef Bacik 33725d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 33735d4f98a2SYan Zheng out: 33745d4f98a2SYan Zheng btrfs_free_path(path); 33755d4f98a2SYan Zheng return ret; 33765d4f98a2SYan Zheng } 33775d4f98a2SYan Zheng 33785d4f98a2SYan Zheng /* 33795d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 33805d4f98a2SYan Zheng */ 33815d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 33825d4f98a2SYan Zheng struct extent_buffer *eb) 33835d4f98a2SYan Zheng { 33845d4f98a2SYan Zheng u64 flags; 33855d4f98a2SYan Zheng int ret; 33865d4f98a2SYan Zheng 33875d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 33885d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 33895d4f98a2SYan Zheng return 1; 33905d4f98a2SYan Zheng 33913fd0a558SYan, Zheng ret = btrfs_lookup_extent_info(NULL, rc->extent_root, 33923173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 33933173a18fSJosef Bacik NULL, &flags); 33945d4f98a2SYan Zheng BUG_ON(ret); 33955d4f98a2SYan Zheng 33965d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 33975d4f98a2SYan Zheng ret = 1; 33985d4f98a2SYan Zheng else 33995d4f98a2SYan Zheng ret = 0; 34005d4f98a2SYan Zheng return ret; 34015d4f98a2SYan Zheng } 34025d4f98a2SYan Zheng 34030af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 34040af3d00bSJosef Bacik struct inode *inode, u64 ino) 34050af3d00bSJosef Bacik { 34060af3d00bSJosef Bacik struct btrfs_key key; 34070af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 34080af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 34090af3d00bSJosef Bacik int ret = 0; 34100af3d00bSJosef Bacik 34110af3d00bSJosef Bacik if (inode) 34120af3d00bSJosef Bacik goto truncate; 34130af3d00bSJosef Bacik 34140af3d00bSJosef Bacik key.objectid = ino; 34150af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 34160af3d00bSJosef Bacik key.offset = 0; 34170af3d00bSJosef Bacik 34180af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 3419f54fb859STsutomu Itoh if (IS_ERR(inode) || is_bad_inode(inode)) { 3420f54fb859STsutomu Itoh if (!IS_ERR(inode)) 34210af3d00bSJosef Bacik iput(inode); 34220af3d00bSJosef Bacik return -ENOENT; 34230af3d00bSJosef Bacik } 34240af3d00bSJosef Bacik 34250af3d00bSJosef Bacik truncate: 34267b61cd92SMiao Xie ret = btrfs_check_trunc_cache_free_space(root, 34277b61cd92SMiao Xie &fs_info->global_block_rsv); 34287b61cd92SMiao Xie if (ret) 34297b61cd92SMiao Xie goto out; 34307b61cd92SMiao Xie 34317a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 34320af3d00bSJosef Bacik if (IS_ERR(trans)) { 34333612b495STsutomu Itoh ret = PTR_ERR(trans); 34340af3d00bSJosef Bacik goto out; 34350af3d00bSJosef Bacik } 34360af3d00bSJosef Bacik 343774514323SFilipe David Borba Manana ret = btrfs_truncate_free_space_cache(root, trans, inode); 34380af3d00bSJosef Bacik 34390af3d00bSJosef Bacik btrfs_end_transaction(trans, root); 3440b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 34410af3d00bSJosef Bacik out: 34420af3d00bSJosef Bacik iput(inode); 34430af3d00bSJosef Bacik return ret; 34440af3d00bSJosef Bacik } 34450af3d00bSJosef Bacik 34465d4f98a2SYan Zheng /* 34475d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 34485d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 34495d4f98a2SYan Zheng */ 34505d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 34515d4f98a2SYan Zheng struct btrfs_key *extent_key, 34525d4f98a2SYan Zheng struct extent_buffer *leaf, 34535d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 34545d4f98a2SYan Zheng struct rb_root *blocks) 34555d4f98a2SYan Zheng { 34565d4f98a2SYan Zheng struct btrfs_path *path; 34575d4f98a2SYan Zheng struct tree_block *block; 34585d4f98a2SYan Zheng struct btrfs_root *root; 34595d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 34605d4f98a2SYan Zheng struct rb_node *rb_node; 34615d4f98a2SYan Zheng struct btrfs_key key; 34625d4f98a2SYan Zheng u64 ref_root; 34635d4f98a2SYan Zheng u64 ref_objectid; 34645d4f98a2SYan Zheng u64 ref_offset; 34655d4f98a2SYan Zheng u32 ref_count; 34665d4f98a2SYan Zheng u32 nritems; 34675d4f98a2SYan Zheng int err = 0; 34685d4f98a2SYan Zheng int added = 0; 34695d4f98a2SYan Zheng int counted; 34705d4f98a2SYan Zheng int ret; 34715d4f98a2SYan Zheng 34725d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 34735d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 34745d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 34755d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 34765d4f98a2SYan Zheng 34770af3d00bSJosef Bacik /* 34780af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 34790af3d00bSJosef Bacik * it and redo the search. 34800af3d00bSJosef Bacik */ 34810af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 34820af3d00bSJosef Bacik ret = delete_block_group_cache(rc->extent_root->fs_info, 34830af3d00bSJosef Bacik NULL, ref_objectid); 34840af3d00bSJosef Bacik if (ret != -ENOENT) 34850af3d00bSJosef Bacik return ret; 34860af3d00bSJosef Bacik ret = 0; 34870af3d00bSJosef Bacik } 34880af3d00bSJosef Bacik 34890af3d00bSJosef Bacik path = btrfs_alloc_path(); 34900af3d00bSJosef Bacik if (!path) 34910af3d00bSJosef Bacik return -ENOMEM; 3492026fd317SJosef Bacik path->reada = 1; 34930af3d00bSJosef Bacik 34945d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, ref_root); 34955d4f98a2SYan Zheng if (IS_ERR(root)) { 34965d4f98a2SYan Zheng err = PTR_ERR(root); 34975d4f98a2SYan Zheng goto out; 34985d4f98a2SYan Zheng } 34995d4f98a2SYan Zheng 35005d4f98a2SYan Zheng key.objectid = ref_objectid; 35015d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 350284850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 350384850e8dSYan, Zheng key.offset = 0; 350484850e8dSYan, Zheng else 350584850e8dSYan, Zheng key.offset = ref_offset; 35065d4f98a2SYan Zheng 35075d4f98a2SYan Zheng path->search_commit_root = 1; 35085d4f98a2SYan Zheng path->skip_locking = 1; 35095d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 35105d4f98a2SYan Zheng if (ret < 0) { 35115d4f98a2SYan Zheng err = ret; 35125d4f98a2SYan Zheng goto out; 35135d4f98a2SYan Zheng } 35145d4f98a2SYan Zheng 35155d4f98a2SYan Zheng leaf = path->nodes[0]; 35165d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 35175d4f98a2SYan Zheng /* 35185d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 35195d4f98a2SYan Zheng * are not counted in 35205d4f98a2SYan Zheng */ 35215d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 35225d4f98a2SYan Zheng counted = 0; 35235d4f98a2SYan Zheng else 35245d4f98a2SYan Zheng counted = 1; 35255d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 35265d4f98a2SYan Zheng if (rb_node) { 35275d4f98a2SYan Zheng if (counted) 35285d4f98a2SYan Zheng added = 1; 35295d4f98a2SYan Zheng else 35305d4f98a2SYan Zheng path->slots[0] = nritems; 35315d4f98a2SYan Zheng } 35325d4f98a2SYan Zheng 35335d4f98a2SYan Zheng while (ref_count > 0) { 35345d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 35355d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 35365d4f98a2SYan Zheng if (ret < 0) { 35375d4f98a2SYan Zheng err = ret; 35385d4f98a2SYan Zheng goto out; 35395d4f98a2SYan Zheng } 3540fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 35415d4f98a2SYan Zheng goto out; 35425d4f98a2SYan Zheng 35435d4f98a2SYan Zheng leaf = path->nodes[0]; 35445d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 35455d4f98a2SYan Zheng added = 0; 35465d4f98a2SYan Zheng 35475d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 35485d4f98a2SYan Zheng counted = 0; 35495d4f98a2SYan Zheng else 35505d4f98a2SYan Zheng counted = 1; 35515d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 35525d4f98a2SYan Zheng if (rb_node) { 35535d4f98a2SYan Zheng if (counted) 35545d4f98a2SYan Zheng added = 1; 35555d4f98a2SYan Zheng else 35565d4f98a2SYan Zheng path->slots[0] = nritems; 35575d4f98a2SYan Zheng } 35585d4f98a2SYan Zheng } 35595d4f98a2SYan Zheng 35605d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3561fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3562fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 35635d4f98a2SYan Zheng break; 35645d4f98a2SYan Zheng 35655d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 35665d4f98a2SYan Zheng struct btrfs_file_extent_item); 35675d4f98a2SYan Zheng 35685d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 35695d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 35705d4f98a2SYan Zheng goto next; 35715d4f98a2SYan Zheng 35725d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 35735d4f98a2SYan Zheng extent_key->objectid) 35745d4f98a2SYan Zheng goto next; 35755d4f98a2SYan Zheng 35765d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 35775d4f98a2SYan Zheng if (key.offset != ref_offset) 35785d4f98a2SYan Zheng goto next; 35795d4f98a2SYan Zheng 35805d4f98a2SYan Zheng if (counted) 35815d4f98a2SYan Zheng ref_count--; 35825d4f98a2SYan Zheng if (added) 35835d4f98a2SYan Zheng goto next; 35845d4f98a2SYan Zheng 35855d4f98a2SYan Zheng if (!tree_block_processed(leaf->start, leaf->len, rc)) { 35865d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 35875d4f98a2SYan Zheng if (!block) { 35885d4f98a2SYan Zheng err = -ENOMEM; 35895d4f98a2SYan Zheng break; 35905d4f98a2SYan Zheng } 35915d4f98a2SYan Zheng block->bytenr = leaf->start; 35925d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 35935d4f98a2SYan Zheng block->level = 0; 35945d4f98a2SYan Zheng block->key_ready = 1; 35955d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 35965d4f98a2SYan Zheng &block->rb_node); 359743c04fb1SJeff Mahoney if (rb_node) 359843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 359943c04fb1SJeff Mahoney block->bytenr); 36005d4f98a2SYan Zheng } 36015d4f98a2SYan Zheng if (counted) 36025d4f98a2SYan Zheng added = 1; 36035d4f98a2SYan Zheng else 36045d4f98a2SYan Zheng path->slots[0] = nritems; 36055d4f98a2SYan Zheng next: 36065d4f98a2SYan Zheng path->slots[0]++; 36075d4f98a2SYan Zheng 36085d4f98a2SYan Zheng } 36095d4f98a2SYan Zheng out: 36105d4f98a2SYan Zheng btrfs_free_path(path); 36115d4f98a2SYan Zheng return err; 36125d4f98a2SYan Zheng } 36135d4f98a2SYan Zheng 36145d4f98a2SYan Zheng /* 36152c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 36165d4f98a2SYan Zheng */ 36175d4f98a2SYan Zheng static noinline_for_stack 36185d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 36195d4f98a2SYan Zheng struct btrfs_key *extent_key, 36205d4f98a2SYan Zheng struct btrfs_path *path, 36215d4f98a2SYan Zheng struct rb_root *blocks) 36225d4f98a2SYan Zheng { 36235d4f98a2SYan Zheng struct btrfs_key key; 36245d4f98a2SYan Zheng struct extent_buffer *eb; 36255d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 36265d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 36275d4f98a2SYan Zheng unsigned long ptr; 36285d4f98a2SYan Zheng unsigned long end; 3629707e8a07SDavid Sterba u32 blocksize = rc->extent_root->nodesize; 3630647f63bdSFilipe David Borba Manana int ret = 0; 36315d4f98a2SYan Zheng int err = 0; 36325d4f98a2SYan Zheng 36335d4f98a2SYan Zheng eb = path->nodes[0]; 36345d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 36355d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 36365d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 36375d4f98a2SYan Zheng if (ptr + sizeof(struct btrfs_extent_item_v0) == end) 36385d4f98a2SYan Zheng ptr = end; 36395d4f98a2SYan Zheng else 36405d4f98a2SYan Zheng #endif 36415d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 36425d4f98a2SYan Zheng 36435d4f98a2SYan Zheng while (ptr < end) { 36445d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 36455d4f98a2SYan Zheng key.type = btrfs_extent_inline_ref_type(eb, iref); 36465d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 36475d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 36485d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 36495d4f98a2SYan Zheng blocks); 36505d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 36515d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 36525d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 36535d4f98a2SYan Zheng eb, dref, blocks); 36545d4f98a2SYan Zheng } else { 36555d4f98a2SYan Zheng BUG(); 36565d4f98a2SYan Zheng } 3657647f63bdSFilipe David Borba Manana if (ret) { 3658647f63bdSFilipe David Borba Manana err = ret; 3659647f63bdSFilipe David Borba Manana goto out; 3660647f63bdSFilipe David Borba Manana } 36615d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 36625d4f98a2SYan Zheng } 36635d4f98a2SYan Zheng WARN_ON(ptr > end); 36645d4f98a2SYan Zheng 36655d4f98a2SYan Zheng while (1) { 36665d4f98a2SYan Zheng cond_resched(); 36675d4f98a2SYan Zheng eb = path->nodes[0]; 36685d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 36695d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 36705d4f98a2SYan Zheng if (ret < 0) { 36715d4f98a2SYan Zheng err = ret; 36725d4f98a2SYan Zheng break; 36735d4f98a2SYan Zheng } 36745d4f98a2SYan Zheng if (ret > 0) 36755d4f98a2SYan Zheng break; 36765d4f98a2SYan Zheng eb = path->nodes[0]; 36775d4f98a2SYan Zheng } 36785d4f98a2SYan Zheng 36795d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 36805d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 36815d4f98a2SYan Zheng break; 36825d4f98a2SYan Zheng 36835d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 36845d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY || 36855d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 36865d4f98a2SYan Zheng #else 36875d4f98a2SYan Zheng BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); 36885d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 36895d4f98a2SYan Zheng #endif 36905d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 36915d4f98a2SYan Zheng blocks); 36925d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 36935d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 36945d4f98a2SYan Zheng struct btrfs_extent_data_ref); 36955d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 36965d4f98a2SYan Zheng eb, dref, blocks); 36975d4f98a2SYan Zheng } else { 36985d4f98a2SYan Zheng ret = 0; 36995d4f98a2SYan Zheng } 37005d4f98a2SYan Zheng if (ret) { 37015d4f98a2SYan Zheng err = ret; 37025d4f98a2SYan Zheng break; 37035d4f98a2SYan Zheng } 37045d4f98a2SYan Zheng path->slots[0]++; 37055d4f98a2SYan Zheng } 3706647f63bdSFilipe David Borba Manana out: 3707b3b4aa74SDavid Sterba btrfs_release_path(path); 37085d4f98a2SYan Zheng if (err) 37095d4f98a2SYan Zheng free_block_list(blocks); 37105d4f98a2SYan Zheng return err; 37115d4f98a2SYan Zheng } 37125d4f98a2SYan Zheng 37135d4f98a2SYan Zheng /* 37142c016dc2SLiu Bo * helper to find next unprocessed extent 37155d4f98a2SYan Zheng */ 37165d4f98a2SYan Zheng static noinline_for_stack 37175d4f98a2SYan Zheng int find_next_extent(struct btrfs_trans_handle *trans, 37183fd0a558SYan, Zheng struct reloc_control *rc, struct btrfs_path *path, 37193fd0a558SYan, Zheng struct btrfs_key *extent_key) 37205d4f98a2SYan Zheng { 37215d4f98a2SYan Zheng struct btrfs_key key; 37225d4f98a2SYan Zheng struct extent_buffer *leaf; 37235d4f98a2SYan Zheng u64 start, end, last; 37245d4f98a2SYan Zheng int ret; 37255d4f98a2SYan Zheng 37265d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 37275d4f98a2SYan Zheng while (1) { 37285d4f98a2SYan Zheng cond_resched(); 37295d4f98a2SYan Zheng if (rc->search_start >= last) { 37305d4f98a2SYan Zheng ret = 1; 37315d4f98a2SYan Zheng break; 37325d4f98a2SYan Zheng } 37335d4f98a2SYan Zheng 37345d4f98a2SYan Zheng key.objectid = rc->search_start; 37355d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 37365d4f98a2SYan Zheng key.offset = 0; 37375d4f98a2SYan Zheng 37385d4f98a2SYan Zheng path->search_commit_root = 1; 37395d4f98a2SYan Zheng path->skip_locking = 1; 37405d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 37415d4f98a2SYan Zheng 0, 0); 37425d4f98a2SYan Zheng if (ret < 0) 37435d4f98a2SYan Zheng break; 37445d4f98a2SYan Zheng next: 37455d4f98a2SYan Zheng leaf = path->nodes[0]; 37465d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 37475d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 37485d4f98a2SYan Zheng if (ret != 0) 37495d4f98a2SYan Zheng break; 37505d4f98a2SYan Zheng leaf = path->nodes[0]; 37515d4f98a2SYan Zheng } 37525d4f98a2SYan Zheng 37535d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 37545d4f98a2SYan Zheng if (key.objectid >= last) { 37555d4f98a2SYan Zheng ret = 1; 37565d4f98a2SYan Zheng break; 37575d4f98a2SYan Zheng } 37585d4f98a2SYan Zheng 37593173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 37603173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 37613173a18fSJosef Bacik path->slots[0]++; 37623173a18fSJosef Bacik goto next; 37633173a18fSJosef Bacik } 37643173a18fSJosef Bacik 37653173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 37665d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 37675d4f98a2SYan Zheng path->slots[0]++; 37685d4f98a2SYan Zheng goto next; 37695d4f98a2SYan Zheng } 37705d4f98a2SYan Zheng 37713173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 3772707e8a07SDavid Sterba key.objectid + rc->extent_root->nodesize <= 37733173a18fSJosef Bacik rc->search_start) { 37743173a18fSJosef Bacik path->slots[0]++; 37753173a18fSJosef Bacik goto next; 37763173a18fSJosef Bacik } 37773173a18fSJosef Bacik 37785d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 37795d4f98a2SYan Zheng key.objectid, &start, &end, 3780e6138876SJosef Bacik EXTENT_DIRTY, NULL); 37815d4f98a2SYan Zheng 37825d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3783b3b4aa74SDavid Sterba btrfs_release_path(path); 37845d4f98a2SYan Zheng rc->search_start = end + 1; 37855d4f98a2SYan Zheng } else { 37863173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 37875d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 37883173a18fSJosef Bacik else 37893173a18fSJosef Bacik rc->search_start = key.objectid + 3790707e8a07SDavid Sterba rc->extent_root->nodesize; 37913fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 37925d4f98a2SYan Zheng return 0; 37935d4f98a2SYan Zheng } 37945d4f98a2SYan Zheng } 3795b3b4aa74SDavid Sterba btrfs_release_path(path); 37965d4f98a2SYan Zheng return ret; 37975d4f98a2SYan Zheng } 37985d4f98a2SYan Zheng 37995d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 38005d4f98a2SYan Zheng { 38015d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38027585717fSChris Mason 38037585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38045d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 38057585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38065d4f98a2SYan Zheng } 38075d4f98a2SYan Zheng 38085d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 38095d4f98a2SYan Zheng { 38105d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38117585717fSChris Mason 38127585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38135d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 38147585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38155d4f98a2SYan Zheng } 38165d4f98a2SYan Zheng 38175d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 38185d4f98a2SYan Zheng { 38195d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38205d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38215d4f98a2SYan Zheng return 1; 38225d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 38235d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38245d4f98a2SYan Zheng return 1; 38255d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38265d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 38275d4f98a2SYan Zheng return 1; 38285d4f98a2SYan Zheng return 0; 38295d4f98a2SYan Zheng } 38305d4f98a2SYan Zheng 38313fd0a558SYan, Zheng static noinline_for_stack 38323fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 38333fd0a558SYan, Zheng { 38343fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 38353fd0a558SYan, Zheng 383666d8f3ddSMiao Xie rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root, 383766d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 38383fd0a558SYan, Zheng if (!rc->block_rsv) 38393fd0a558SYan, Zheng return -ENOMEM; 38403fd0a558SYan, Zheng 38413fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 38423fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 38433fd0a558SYan, Zheng rc->extents_found = 0; 38443fd0a558SYan, Zheng rc->nodes_relocated = 0; 38453fd0a558SYan, Zheng rc->merging_rsv_size = 0; 38460647bf56SWang Shilong rc->reserved_bytes = 0; 38470647bf56SWang Shilong rc->block_rsv->size = rc->extent_root->nodesize * 38480647bf56SWang Shilong RELOCATION_RESERVED_NODES; 38493fd0a558SYan, Zheng 38503fd0a558SYan, Zheng rc->create_reloc_tree = 1; 38513fd0a558SYan, Zheng set_reloc_control(rc); 38523fd0a558SYan, Zheng 38537a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 385428818947SLiu Bo if (IS_ERR(trans)) { 385528818947SLiu Bo unset_reloc_control(rc); 385628818947SLiu Bo /* 385728818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 385828818947SLiu Bo * cleanup. And callers are responsible to free the above 385928818947SLiu Bo * block rsv. 386028818947SLiu Bo */ 386128818947SLiu Bo return PTR_ERR(trans); 386228818947SLiu Bo } 38633fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 38643fd0a558SYan, Zheng return 0; 38653fd0a558SYan, Zheng } 386676dda93cSYan, Zheng 38675d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 38685d4f98a2SYan Zheng { 38695d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 38705d4f98a2SYan Zheng struct btrfs_key key; 38715d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 38725d4f98a2SYan Zheng struct btrfs_path *path; 38735d4f98a2SYan Zheng struct btrfs_extent_item *ei; 38745d4f98a2SYan Zheng u64 flags; 38755d4f98a2SYan Zheng u32 item_size; 38765d4f98a2SYan Zheng int ret; 38775d4f98a2SYan Zheng int err = 0; 3878c87f08caSChris Mason int progress = 0; 38795d4f98a2SYan Zheng 38805d4f98a2SYan Zheng path = btrfs_alloc_path(); 38813fd0a558SYan, Zheng if (!path) 38825d4f98a2SYan Zheng return -ENOMEM; 3883026fd317SJosef Bacik path->reada = 1; 38843fd0a558SYan, Zheng 38853fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 38863fd0a558SYan, Zheng if (ret) { 38873fd0a558SYan, Zheng err = ret; 38883fd0a558SYan, Zheng goto out_free; 38892423fdfbSJiri Slaby } 38905d4f98a2SYan Zheng 38915d4f98a2SYan Zheng while (1) { 38920647bf56SWang Shilong rc->reserved_bytes = 0; 38930647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 38940647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 38950647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 38960647bf56SWang Shilong if (ret) { 38970647bf56SWang Shilong err = ret; 38980647bf56SWang Shilong break; 38990647bf56SWang Shilong } 3900c87f08caSChris Mason progress++; 3901a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 39020f788c58SLiu Bo if (IS_ERR(trans)) { 39030f788c58SLiu Bo err = PTR_ERR(trans); 39040f788c58SLiu Bo trans = NULL; 39050f788c58SLiu Bo break; 39060f788c58SLiu Bo } 3907c87f08caSChris Mason restart: 39083fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 39093fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 39103fd0a558SYan, Zheng continue; 39113fd0a558SYan, Zheng } 39123fd0a558SYan, Zheng 39133fd0a558SYan, Zheng ret = find_next_extent(trans, rc, path, &key); 39145d4f98a2SYan Zheng if (ret < 0) 39155d4f98a2SYan Zheng err = ret; 39165d4f98a2SYan Zheng if (ret != 0) 39175d4f98a2SYan Zheng break; 39185d4f98a2SYan Zheng 39195d4f98a2SYan Zheng rc->extents_found++; 39205d4f98a2SYan Zheng 39215d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 39225d4f98a2SYan Zheng struct btrfs_extent_item); 39233fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 39245d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 39255d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 39265d4f98a2SYan Zheng ret = check_extent_flags(flags); 39275d4f98a2SYan Zheng BUG_ON(ret); 39285d4f98a2SYan Zheng 39295d4f98a2SYan Zheng } else { 39305d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 39315d4f98a2SYan Zheng u64 ref_owner; 39325d4f98a2SYan Zheng int path_change = 0; 39335d4f98a2SYan Zheng 39345d4f98a2SYan Zheng BUG_ON(item_size != 39355d4f98a2SYan Zheng sizeof(struct btrfs_extent_item_v0)); 39365d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, &key, &ref_owner, 39375d4f98a2SYan Zheng &path_change); 39385d4f98a2SYan Zheng if (ref_owner < BTRFS_FIRST_FREE_OBJECTID) 39395d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_TREE_BLOCK; 39405d4f98a2SYan Zheng else 39415d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_DATA; 39425d4f98a2SYan Zheng 39435d4f98a2SYan Zheng if (path_change) { 3944b3b4aa74SDavid Sterba btrfs_release_path(path); 39455d4f98a2SYan Zheng 39465d4f98a2SYan Zheng path->search_commit_root = 1; 39475d4f98a2SYan Zheng path->skip_locking = 1; 39485d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, 39495d4f98a2SYan Zheng &key, path, 0, 0); 39505d4f98a2SYan Zheng if (ret < 0) { 39515d4f98a2SYan Zheng err = ret; 39525d4f98a2SYan Zheng break; 39535d4f98a2SYan Zheng } 39545d4f98a2SYan Zheng BUG_ON(ret > 0); 39555d4f98a2SYan Zheng } 39565d4f98a2SYan Zheng #else 39575d4f98a2SYan Zheng BUG(); 39585d4f98a2SYan Zheng #endif 39595d4f98a2SYan Zheng } 39605d4f98a2SYan Zheng 39615d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 39625d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 39635d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 39645d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 39655d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 39665d4f98a2SYan Zheng } else { 3967b3b4aa74SDavid Sterba btrfs_release_path(path); 39685d4f98a2SYan Zheng ret = 0; 39695d4f98a2SYan Zheng } 39705d4f98a2SYan Zheng if (ret < 0) { 39713fd0a558SYan, Zheng err = ret; 39725d4f98a2SYan Zheng break; 39735d4f98a2SYan Zheng } 39745d4f98a2SYan Zheng 39755d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 39765d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 39775d4f98a2SYan Zheng if (ret < 0) { 39781708cc57SWang Shilong /* 39791708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 39801708cc57SWang Shilong * backref cache when committing transaction. 39811708cc57SWang Shilong */ 39821708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 39831708cc57SWang Shilong 39843fd0a558SYan, Zheng if (ret != -EAGAIN) { 39855d4f98a2SYan Zheng err = ret; 39865d4f98a2SYan Zheng break; 39875d4f98a2SYan Zheng } 39883fd0a558SYan, Zheng rc->extents_found--; 39893fd0a558SYan, Zheng rc->search_start = key.objectid; 39903fd0a558SYan, Zheng } 39915d4f98a2SYan Zheng } 39925d4f98a2SYan Zheng 39933fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 3994b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 39953fd0a558SYan, Zheng trans = NULL; 39965d4f98a2SYan Zheng 39975d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 39985d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 39995d4f98a2SYan Zheng rc->found_file_extent = 1; 40000257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 40013fd0a558SYan, Zheng &key, &rc->cluster); 40025d4f98a2SYan Zheng if (ret < 0) { 40035d4f98a2SYan Zheng err = ret; 40045d4f98a2SYan Zheng break; 40055d4f98a2SYan Zheng } 40065d4f98a2SYan Zheng } 40075d4f98a2SYan Zheng } 4008c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 4009c87f08caSChris Mason ret = btrfs_force_chunk_alloc(trans, rc->extent_root, 4010c87f08caSChris Mason rc->block_group->flags); 4011c87f08caSChris Mason if (ret == 0) { 4012c87f08caSChris Mason err = 0; 4013c87f08caSChris Mason progress = 0; 4014c87f08caSChris Mason goto restart; 4015c87f08caSChris Mason } 4016c87f08caSChris Mason } 40173fd0a558SYan, Zheng 4018b3b4aa74SDavid Sterba btrfs_release_path(path); 40193fd0a558SYan, Zheng clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY, 40203fd0a558SYan, Zheng GFP_NOFS); 40215d4f98a2SYan Zheng 40225d4f98a2SYan Zheng if (trans) { 40233fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 4024b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 40255d4f98a2SYan Zheng } 40265d4f98a2SYan Zheng 40270257bb82SYan, Zheng if (!err) { 40283fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 40293fd0a558SYan, Zheng &rc->cluster); 40300257bb82SYan, Zheng if (ret < 0) 40310257bb82SYan, Zheng err = ret; 40320257bb82SYan, Zheng } 40330257bb82SYan, Zheng 40343fd0a558SYan, Zheng rc->create_reloc_tree = 0; 40353fd0a558SYan, Zheng set_reloc_control(rc); 40360257bb82SYan, Zheng 40373fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 40383fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 40395d4f98a2SYan Zheng 40403fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 40415d4f98a2SYan Zheng 40425d4f98a2SYan Zheng merge_reloc_roots(rc); 40435d4f98a2SYan Zheng 40443fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 40455d4f98a2SYan Zheng unset_reloc_control(rc); 40463fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 40475d4f98a2SYan Zheng 40485d4f98a2SYan Zheng /* get rid of pinned extents */ 40497a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 40503612b495STsutomu Itoh if (IS_ERR(trans)) 40513612b495STsutomu Itoh err = PTR_ERR(trans); 40523612b495STsutomu Itoh else 40535d4f98a2SYan Zheng btrfs_commit_transaction(trans, rc->extent_root); 40543fd0a558SYan, Zheng out_free: 40553fd0a558SYan, Zheng btrfs_free_block_rsv(rc->extent_root, rc->block_rsv); 40563fd0a558SYan, Zheng btrfs_free_path(path); 40575d4f98a2SYan Zheng return err; 40585d4f98a2SYan Zheng } 40595d4f98a2SYan Zheng 40605d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 40610257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 40625d4f98a2SYan Zheng { 40635d4f98a2SYan Zheng struct btrfs_path *path; 40645d4f98a2SYan Zheng struct btrfs_inode_item *item; 40655d4f98a2SYan Zheng struct extent_buffer *leaf; 40665d4f98a2SYan Zheng int ret; 40675d4f98a2SYan Zheng 40685d4f98a2SYan Zheng path = btrfs_alloc_path(); 40695d4f98a2SYan Zheng if (!path) 40705d4f98a2SYan Zheng return -ENOMEM; 40715d4f98a2SYan Zheng 40725d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 40735d4f98a2SYan Zheng if (ret) 40745d4f98a2SYan Zheng goto out; 40755d4f98a2SYan Zheng 40765d4f98a2SYan Zheng leaf = path->nodes[0]; 40775d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 40785d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); 40795d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 40800257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 40815d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 40823fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 40833fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 40845d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 40855d4f98a2SYan Zheng out: 40865d4f98a2SYan Zheng btrfs_free_path(path); 40875d4f98a2SYan Zheng return ret; 40885d4f98a2SYan Zheng } 40895d4f98a2SYan Zheng 40905d4f98a2SYan Zheng /* 40915d4f98a2SYan Zheng * helper to create inode for data relocation. 40925d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 40935d4f98a2SYan Zheng */ 40943fd0a558SYan, Zheng static noinline_for_stack 40953fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 40965d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 40975d4f98a2SYan Zheng { 40985d4f98a2SYan Zheng struct inode *inode = NULL; 40995d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41005d4f98a2SYan Zheng struct btrfs_root *root; 41015d4f98a2SYan Zheng struct btrfs_key key; 41025d4f98a2SYan Zheng u64 objectid = BTRFS_FIRST_FREE_OBJECTID; 41035d4f98a2SYan Zheng int err = 0; 41045d4f98a2SYan Zheng 41055d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 41065d4f98a2SYan Zheng if (IS_ERR(root)) 41075d4f98a2SYan Zheng return ERR_CAST(root); 41085d4f98a2SYan Zheng 4109a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 41103fd0a558SYan, Zheng if (IS_ERR(trans)) 41113fd0a558SYan, Zheng return ERR_CAST(trans); 41125d4f98a2SYan Zheng 4113581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 41145d4f98a2SYan Zheng if (err) 41155d4f98a2SYan Zheng goto out; 41165d4f98a2SYan Zheng 41170257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 41185d4f98a2SYan Zheng BUG_ON(err); 41195d4f98a2SYan Zheng 41205d4f98a2SYan Zheng key.objectid = objectid; 41215d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 41225d4f98a2SYan Zheng key.offset = 0; 412373f73415SJosef Bacik inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); 41245d4f98a2SYan Zheng BUG_ON(IS_ERR(inode) || is_bad_inode(inode)); 41255d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 41265d4f98a2SYan Zheng 41275d4f98a2SYan Zheng err = btrfs_orphan_add(trans, inode); 41285d4f98a2SYan Zheng out: 41295d4f98a2SYan Zheng btrfs_end_transaction(trans, root); 4130b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 41315d4f98a2SYan Zheng if (err) { 41325d4f98a2SYan Zheng if (inode) 41335d4f98a2SYan Zheng iput(inode); 41345d4f98a2SYan Zheng inode = ERR_PTR(err); 41355d4f98a2SYan Zheng } 41365d4f98a2SYan Zheng return inode; 41375d4f98a2SYan Zheng } 41385d4f98a2SYan Zheng 4139a9995eecSJosef Bacik static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 41403fd0a558SYan, Zheng { 41413fd0a558SYan, Zheng struct reloc_control *rc; 41423fd0a558SYan, Zheng 41433fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 41443fd0a558SYan, Zheng if (!rc) 41453fd0a558SYan, Zheng return NULL; 41463fd0a558SYan, Zheng 41473fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 41483fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 41493fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 4150a9995eecSJosef Bacik extent_io_tree_init(&rc->processed_blocks, 4151a9995eecSJosef Bacik fs_info->btree_inode->i_mapping); 41523fd0a558SYan, Zheng return rc; 41533fd0a558SYan, Zheng } 41543fd0a558SYan, Zheng 41555d4f98a2SYan Zheng /* 41565d4f98a2SYan Zheng * function to relocate all extents in a block group. 41575d4f98a2SYan Zheng */ 41585d4f98a2SYan Zheng int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start) 41595d4f98a2SYan Zheng { 41605d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = extent_root->fs_info; 41615d4f98a2SYan Zheng struct reloc_control *rc; 41620af3d00bSJosef Bacik struct inode *inode; 41630af3d00bSJosef Bacik struct btrfs_path *path; 41645d4f98a2SYan Zheng int ret; 4165f0486c68SYan, Zheng int rw = 0; 41665d4f98a2SYan Zheng int err = 0; 41675d4f98a2SYan Zheng 4168a9995eecSJosef Bacik rc = alloc_reloc_control(fs_info); 41695d4f98a2SYan Zheng if (!rc) 41705d4f98a2SYan Zheng return -ENOMEM; 41715d4f98a2SYan Zheng 4172f0486c68SYan, Zheng rc->extent_root = extent_root; 41733fd0a558SYan, Zheng 41745d4f98a2SYan Zheng rc->block_group = btrfs_lookup_block_group(fs_info, group_start); 41755d4f98a2SYan Zheng BUG_ON(!rc->block_group); 41765d4f98a2SYan Zheng 4177f0486c68SYan, Zheng if (!rc->block_group->ro) { 4178f0486c68SYan, Zheng ret = btrfs_set_block_group_ro(extent_root, rc->block_group); 4179f0486c68SYan, Zheng if (ret) { 4180f0486c68SYan, Zheng err = ret; 4181f0486c68SYan, Zheng goto out; 4182f0486c68SYan, Zheng } 4183f0486c68SYan, Zheng rw = 1; 4184f0486c68SYan, Zheng } 4185f0486c68SYan, Zheng 41860af3d00bSJosef Bacik path = btrfs_alloc_path(); 41870af3d00bSJosef Bacik if (!path) { 41880af3d00bSJosef Bacik err = -ENOMEM; 41890af3d00bSJosef Bacik goto out; 41900af3d00bSJosef Bacik } 41910af3d00bSJosef Bacik 41920af3d00bSJosef Bacik inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group, 41930af3d00bSJosef Bacik path); 41940af3d00bSJosef Bacik btrfs_free_path(path); 41950af3d00bSJosef Bacik 41960af3d00bSJosef Bacik if (!IS_ERR(inode)) 41970af3d00bSJosef Bacik ret = delete_block_group_cache(fs_info, inode, 0); 41980af3d00bSJosef Bacik else 41990af3d00bSJosef Bacik ret = PTR_ERR(inode); 42000af3d00bSJosef Bacik 42010af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 42020af3d00bSJosef Bacik err = ret; 42030af3d00bSJosef Bacik goto out; 42040af3d00bSJosef Bacik } 42050af3d00bSJosef Bacik 42065d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 42075d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 42085d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 42095d4f98a2SYan Zheng rc->data_inode = NULL; 42105d4f98a2SYan Zheng goto out; 42115d4f98a2SYan Zheng } 42125d4f98a2SYan Zheng 4213efe120a0SFrank Holton btrfs_info(extent_root->fs_info, "relocating block group %llu flags %llu", 4214c1c9ff7cSGeert Uytterhoeven rc->block_group->key.objectid, rc->block_group->flags); 42155d4f98a2SYan Zheng 42166c255e67SMiao Xie ret = btrfs_start_delalloc_roots(fs_info, 0, -1); 42178ccf6f19SMiao Xie if (ret < 0) { 42188ccf6f19SMiao Xie err = ret; 42198ccf6f19SMiao Xie goto out; 42208ccf6f19SMiao Xie } 4221b0244199SMiao Xie btrfs_wait_ordered_roots(fs_info, -1); 42225d4f98a2SYan Zheng 42235d4f98a2SYan Zheng while (1) { 422476dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 42255d4f98a2SYan Zheng ret = relocate_block_group(rc); 422676dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 42275d4f98a2SYan Zheng if (ret < 0) { 42285d4f98a2SYan Zheng err = ret; 42293fd0a558SYan, Zheng goto out; 42305d4f98a2SYan Zheng } 42315d4f98a2SYan Zheng 42325d4f98a2SYan Zheng if (rc->extents_found == 0) 42335d4f98a2SYan Zheng break; 42345d4f98a2SYan Zheng 4235efe120a0SFrank Holton btrfs_info(extent_root->fs_info, "found %llu extents", 4236c1c9ff7cSGeert Uytterhoeven rc->extents_found); 42375d4f98a2SYan Zheng 42385d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 42390ef8b726SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 42400ef8b726SJosef Bacik (u64)-1); 42410ef8b726SJosef Bacik if (ret) { 42420ef8b726SJosef Bacik err = ret; 42430ef8b726SJosef Bacik goto out; 42440ef8b726SJosef Bacik } 42455d4f98a2SYan Zheng invalidate_mapping_pages(rc->data_inode->i_mapping, 42465d4f98a2SYan Zheng 0, -1); 42475d4f98a2SYan Zheng rc->stage = UPDATE_DATA_PTRS; 42485d4f98a2SYan Zheng } 42495d4f98a2SYan Zheng } 42505d4f98a2SYan Zheng 42515d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 42525d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 42535d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 42545d4f98a2SYan Zheng out: 4255f0486c68SYan, Zheng if (err && rw) 4256f0486c68SYan, Zheng btrfs_set_block_group_rw(extent_root, rc->block_group); 42575d4f98a2SYan Zheng iput(rc->data_inode); 42585d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 42595d4f98a2SYan Zheng kfree(rc); 42605d4f98a2SYan Zheng return err; 42615d4f98a2SYan Zheng } 42625d4f98a2SYan Zheng 426376dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 426476dda93cSYan, Zheng { 426576dda93cSYan, Zheng struct btrfs_trans_handle *trans; 426679787eaaSJeff Mahoney int ret, err; 426776dda93cSYan, Zheng 4268a22285a6SYan, Zheng trans = btrfs_start_transaction(root->fs_info->tree_root, 0); 426979787eaaSJeff Mahoney if (IS_ERR(trans)) 427079787eaaSJeff Mahoney return PTR_ERR(trans); 427176dda93cSYan, Zheng 427276dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 427376dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 427476dda93cSYan, Zheng root->root_item.drop_level = 0; 427576dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 427676dda93cSYan, Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 427776dda93cSYan, Zheng &root->root_key, &root->root_item); 427876dda93cSYan, Zheng 427979787eaaSJeff Mahoney err = btrfs_end_transaction(trans, root->fs_info->tree_root); 428079787eaaSJeff Mahoney if (err) 428179787eaaSJeff Mahoney return err; 428279787eaaSJeff Mahoney return ret; 428376dda93cSYan, Zheng } 428476dda93cSYan, Zheng 42855d4f98a2SYan Zheng /* 42865d4f98a2SYan Zheng * recover relocation interrupted by system crash. 42875d4f98a2SYan Zheng * 42885d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 42895d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 42905d4f98a2SYan Zheng */ 42915d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 42925d4f98a2SYan Zheng { 42935d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 42945d4f98a2SYan Zheng struct btrfs_key key; 42955d4f98a2SYan Zheng struct btrfs_root *fs_root; 42965d4f98a2SYan Zheng struct btrfs_root *reloc_root; 42975d4f98a2SYan Zheng struct btrfs_path *path; 42985d4f98a2SYan Zheng struct extent_buffer *leaf; 42995d4f98a2SYan Zheng struct reloc_control *rc = NULL; 43005d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 43015d4f98a2SYan Zheng int ret; 43025d4f98a2SYan Zheng int err = 0; 43035d4f98a2SYan Zheng 43045d4f98a2SYan Zheng path = btrfs_alloc_path(); 43055d4f98a2SYan Zheng if (!path) 43065d4f98a2SYan Zheng return -ENOMEM; 4307026fd317SJosef Bacik path->reada = -1; 43085d4f98a2SYan Zheng 43095d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 43105d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 43115d4f98a2SYan Zheng key.offset = (u64)-1; 43125d4f98a2SYan Zheng 43135d4f98a2SYan Zheng while (1) { 43145d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, 43155d4f98a2SYan Zheng path, 0, 0); 43165d4f98a2SYan Zheng if (ret < 0) { 43175d4f98a2SYan Zheng err = ret; 43185d4f98a2SYan Zheng goto out; 43195d4f98a2SYan Zheng } 43205d4f98a2SYan Zheng if (ret > 0) { 43215d4f98a2SYan Zheng if (path->slots[0] == 0) 43225d4f98a2SYan Zheng break; 43235d4f98a2SYan Zheng path->slots[0]--; 43245d4f98a2SYan Zheng } 43255d4f98a2SYan Zheng leaf = path->nodes[0]; 43265d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4327b3b4aa74SDavid Sterba btrfs_release_path(path); 43285d4f98a2SYan Zheng 43295d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 43305d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 43315d4f98a2SYan Zheng break; 43325d4f98a2SYan Zheng 4333cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 43345d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 43355d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 43365d4f98a2SYan Zheng goto out; 43375d4f98a2SYan Zheng } 43385d4f98a2SYan Zheng 43395d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 43405d4f98a2SYan Zheng 43415d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 43425d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 43435d4f98a2SYan Zheng reloc_root->root_key.offset); 43445d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 434576dda93cSYan, Zheng ret = PTR_ERR(fs_root); 434676dda93cSYan, Zheng if (ret != -ENOENT) { 434776dda93cSYan, Zheng err = ret; 43485d4f98a2SYan Zheng goto out; 43495d4f98a2SYan Zheng } 435079787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 435179787eaaSJeff Mahoney if (ret < 0) { 435279787eaaSJeff Mahoney err = ret; 435379787eaaSJeff Mahoney goto out; 435479787eaaSJeff Mahoney } 435576dda93cSYan, Zheng } 43565d4f98a2SYan Zheng } 43575d4f98a2SYan Zheng 43585d4f98a2SYan Zheng if (key.offset == 0) 43595d4f98a2SYan Zheng break; 43605d4f98a2SYan Zheng 43615d4f98a2SYan Zheng key.offset--; 43625d4f98a2SYan Zheng } 4363b3b4aa74SDavid Sterba btrfs_release_path(path); 43645d4f98a2SYan Zheng 43655d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 43665d4f98a2SYan Zheng goto out; 43675d4f98a2SYan Zheng 4368a9995eecSJosef Bacik rc = alloc_reloc_control(root->fs_info); 43695d4f98a2SYan Zheng if (!rc) { 43705d4f98a2SYan Zheng err = -ENOMEM; 43715d4f98a2SYan Zheng goto out; 43725d4f98a2SYan Zheng } 43735d4f98a2SYan Zheng 43745d4f98a2SYan Zheng rc->extent_root = root->fs_info->extent_root; 43755d4f98a2SYan Zheng 43765d4f98a2SYan Zheng set_reloc_control(rc); 43775d4f98a2SYan Zheng 43787a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 43793612b495STsutomu Itoh if (IS_ERR(trans)) { 43803612b495STsutomu Itoh unset_reloc_control(rc); 43813612b495STsutomu Itoh err = PTR_ERR(trans); 43823612b495STsutomu Itoh goto out_free; 43833612b495STsutomu Itoh } 43843fd0a558SYan, Zheng 43853fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 43863fd0a558SYan, Zheng 43875d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 43885d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 43895d4f98a2SYan Zheng struct btrfs_root, root_list); 43905d4f98a2SYan Zheng list_del(&reloc_root->root_list); 43915d4f98a2SYan Zheng 43925d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 43935d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 43945d4f98a2SYan Zheng &rc->reloc_roots); 43955d4f98a2SYan Zheng continue; 43965d4f98a2SYan Zheng } 43975d4f98a2SYan Zheng 43985d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 43995d4f98a2SYan Zheng reloc_root->root_key.offset); 440079787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 440179787eaaSJeff Mahoney err = PTR_ERR(fs_root); 440279787eaaSJeff Mahoney goto out_free; 440379787eaaSJeff Mahoney } 44045d4f98a2SYan Zheng 4405ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 440679787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 44075d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 44085d4f98a2SYan Zheng } 44095d4f98a2SYan Zheng 441079787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 441179787eaaSJeff Mahoney if (err) 441279787eaaSJeff Mahoney goto out_free; 44135d4f98a2SYan Zheng 44145d4f98a2SYan Zheng merge_reloc_roots(rc); 44155d4f98a2SYan Zheng 44165d4f98a2SYan Zheng unset_reloc_control(rc); 44175d4f98a2SYan Zheng 44187a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 44193612b495STsutomu Itoh if (IS_ERR(trans)) 44203612b495STsutomu Itoh err = PTR_ERR(trans); 44213612b495STsutomu Itoh else 442279787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 44233612b495STsutomu Itoh out_free: 44245d4f98a2SYan Zheng kfree(rc); 44253612b495STsutomu Itoh out: 4426aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4427aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4428aca1bba6SLiu Bo 44295d4f98a2SYan Zheng btrfs_free_path(path); 44305d4f98a2SYan Zheng 44315d4f98a2SYan Zheng if (err == 0) { 44325d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 44335d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 44345d4f98a2SYan Zheng BTRFS_DATA_RELOC_TREE_OBJECTID); 44355d4f98a2SYan Zheng if (IS_ERR(fs_root)) 44365d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4437d7ce5843SMiao Xie else 443866b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 44395d4f98a2SYan Zheng } 44405d4f98a2SYan Zheng return err; 44415d4f98a2SYan Zheng } 44425d4f98a2SYan Zheng 44435d4f98a2SYan Zheng /* 44445d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 44455d4f98a2SYan Zheng * 44465d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 44475d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 44485d4f98a2SYan Zheng */ 44495d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 44505d4f98a2SYan Zheng { 44515d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 44525d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 44535d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 44545d4f98a2SYan Zheng int ret; 44555d4f98a2SYan Zheng u64 disk_bytenr; 44564577b014SJosef Bacik u64 new_bytenr; 44575d4f98a2SYan Zheng LIST_HEAD(list); 44585d4f98a2SYan Zheng 44595d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 44605d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 44615d4f98a2SYan Zheng 44625d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 44635d4f98a2SYan Zheng ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr, 4464a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 446579787eaaSJeff Mahoney if (ret) 446679787eaaSJeff Mahoney goto out; 44675d4f98a2SYan Zheng 44685d4f98a2SYan Zheng while (!list_empty(&list)) { 44695d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 44705d4f98a2SYan Zheng list_del_init(&sums->list); 44715d4f98a2SYan Zheng 44724577b014SJosef Bacik /* 44734577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 44744577b014SJosef Bacik * We need to do this because we will read in entire prealloc 44754577b014SJosef Bacik * extents but we may have written to say the middle of the 44764577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 44774577b014SJosef Bacik * the right disk offset. 44784577b014SJosef Bacik * 44794577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 44804577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 44814577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 44824577b014SJosef Bacik * disk length. 44834577b014SJosef Bacik */ 44844577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 44854577b014SJosef Bacik sums->bytenr = new_bytenr; 44865d4f98a2SYan Zheng 44875d4f98a2SYan Zheng btrfs_add_ordered_sum(inode, ordered, sums); 44885d4f98a2SYan Zheng } 448979787eaaSJeff Mahoney out: 44905d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4491411fc6bcSAndi Kleen return ret; 44925d4f98a2SYan Zheng } 44933fd0a558SYan, Zheng 449483d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 44953fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 44963fd0a558SYan, Zheng struct extent_buffer *cow) 44973fd0a558SYan, Zheng { 44983fd0a558SYan, Zheng struct reloc_control *rc; 44993fd0a558SYan, Zheng struct backref_node *node; 45003fd0a558SYan, Zheng int first_cow = 0; 45013fd0a558SYan, Zheng int level; 450283d4cfd4SJosef Bacik int ret = 0; 45033fd0a558SYan, Zheng 45043fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 45053fd0a558SYan, Zheng if (!rc) 450683d4cfd4SJosef Bacik return 0; 45073fd0a558SYan, Zheng 45083fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 45093fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 45103fd0a558SYan, Zheng 4511c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4512c974c464SWang Shilong if (buf == root->node) 4513c974c464SWang Shilong __update_reloc_root(root, cow->start); 4514c974c464SWang Shilong } 4515c974c464SWang Shilong 45163fd0a558SYan, Zheng level = btrfs_header_level(buf); 45173fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 45183fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 45193fd0a558SYan, Zheng first_cow = 1; 45203fd0a558SYan, Zheng 45213fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 45223fd0a558SYan, Zheng rc->create_reloc_tree) { 45233fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 45243fd0a558SYan, Zheng 45253fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 45263fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 45273fd0a558SYan, Zheng node->new_bytenr != buf->start); 45283fd0a558SYan, Zheng 45293fd0a558SYan, Zheng drop_node_buffer(node); 45303fd0a558SYan, Zheng extent_buffer_get(cow); 45313fd0a558SYan, Zheng node->eb = cow; 45323fd0a558SYan, Zheng node->new_bytenr = cow->start; 45333fd0a558SYan, Zheng 45343fd0a558SYan, Zheng if (!node->pending) { 45353fd0a558SYan, Zheng list_move_tail(&node->list, 45363fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 45373fd0a558SYan, Zheng node->pending = 1; 45383fd0a558SYan, Zheng } 45393fd0a558SYan, Zheng 45403fd0a558SYan, Zheng if (first_cow) 45413fd0a558SYan, Zheng __mark_block_processed(rc, node); 45423fd0a558SYan, Zheng 45433fd0a558SYan, Zheng if (first_cow && level > 0) 45443fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 45453fd0a558SYan, Zheng } 45463fd0a558SYan, Zheng 454783d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 45483fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 454983d4cfd4SJosef Bacik return ret; 45503fd0a558SYan, Zheng } 45513fd0a558SYan, Zheng 45523fd0a558SYan, Zheng /* 45533fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 45543fd0a558SYan, Zheng * requried for relocating tree blocks in the snapshot 45553fd0a558SYan, Zheng */ 45563fd0a558SYan, Zheng void btrfs_reloc_pre_snapshot(struct btrfs_trans_handle *trans, 45573fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending, 45583fd0a558SYan, Zheng u64 *bytes_to_reserve) 45593fd0a558SYan, Zheng { 45603fd0a558SYan, Zheng struct btrfs_root *root; 45613fd0a558SYan, Zheng struct reloc_control *rc; 45623fd0a558SYan, Zheng 45633fd0a558SYan, Zheng root = pending->root; 45643fd0a558SYan, Zheng if (!root->reloc_root) 45653fd0a558SYan, Zheng return; 45663fd0a558SYan, Zheng 45673fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 45683fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 45693fd0a558SYan, Zheng return; 45703fd0a558SYan, Zheng 45713fd0a558SYan, Zheng root = root->reloc_root; 45723fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 45733fd0a558SYan, Zheng /* 45743fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 45753fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 45763fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 45773fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 45783fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 45793fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 45803fd0a558SYan, Zheng * tree may use more space than it frees. so we need 45813fd0a558SYan, Zheng * reserve extra space. 45823fd0a558SYan, Zheng */ 45833fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 45843fd0a558SYan, Zheng } 45853fd0a558SYan, Zheng 45863fd0a558SYan, Zheng /* 45873fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 45883fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 45893fd0a558SYan, Zheng */ 459049b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 45913fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 45923fd0a558SYan, Zheng { 45933fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 45943fd0a558SYan, Zheng struct btrfs_root *reloc_root; 45953fd0a558SYan, Zheng struct btrfs_root *new_root; 45963fd0a558SYan, Zheng struct reloc_control *rc; 45973fd0a558SYan, Zheng int ret; 45983fd0a558SYan, Zheng 45993fd0a558SYan, Zheng if (!root->reloc_root) 460049b25e05SJeff Mahoney return 0; 46013fd0a558SYan, Zheng 46023fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 46033fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 46043fd0a558SYan, Zheng 46053fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 46063fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 46073fd0a558SYan, Zheng rc->block_rsv, 46083fd0a558SYan, Zheng rc->nodes_relocated); 460949b25e05SJeff Mahoney if (ret) 461049b25e05SJeff Mahoney return ret; 46113fd0a558SYan, Zheng } 46123fd0a558SYan, Zheng 46133fd0a558SYan, Zheng new_root = pending->snap; 46143fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 46153fd0a558SYan, Zheng new_root->root_key.objectid); 461649b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 461749b25e05SJeff Mahoney return PTR_ERR(reloc_root); 46183fd0a558SYan, Zheng 4619ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4620ffd7b339SJeff Mahoney BUG_ON(ret < 0); 46213fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 46223fd0a558SYan, Zheng 462349b25e05SJeff Mahoney if (rc->create_reloc_tree) 46243fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 462549b25e05SJeff Mahoney return ret; 46263fd0a558SYan, Zheng } 4627