15d4f98a2SYan Zheng /* 25d4f98a2SYan Zheng * Copyright (C) 2009 Oracle. All rights reserved. 35d4f98a2SYan Zheng * 45d4f98a2SYan Zheng * This program is free software; you can redistribute it and/or 55d4f98a2SYan Zheng * modify it under the terms of the GNU General Public 65d4f98a2SYan Zheng * License v2 as published by the Free Software Foundation. 75d4f98a2SYan Zheng * 85d4f98a2SYan Zheng * This program is distributed in the hope that it will be useful, 95d4f98a2SYan Zheng * but WITHOUT ANY WARRANTY; without even the implied warranty of 105d4f98a2SYan Zheng * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 115d4f98a2SYan Zheng * General Public License for more details. 125d4f98a2SYan Zheng * 135d4f98a2SYan Zheng * You should have received a copy of the GNU General Public 145d4f98a2SYan Zheng * License along with this program; if not, write to the 155d4f98a2SYan Zheng * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 165d4f98a2SYan Zheng * Boston, MA 021110-1307, USA. 175d4f98a2SYan Zheng */ 185d4f98a2SYan Zheng 195d4f98a2SYan Zheng #include <linux/sched.h> 205d4f98a2SYan Zheng #include <linux/pagemap.h> 215d4f98a2SYan Zheng #include <linux/writeback.h> 225d4f98a2SYan Zheng #include <linux/blkdev.h> 235d4f98a2SYan Zheng #include <linux/rbtree.h> 245a0e3ad6STejun Heo #include <linux/slab.h> 255d4f98a2SYan Zheng #include "ctree.h" 265d4f98a2SYan Zheng #include "disk-io.h" 275d4f98a2SYan Zheng #include "transaction.h" 285d4f98a2SYan Zheng #include "volumes.h" 295d4f98a2SYan Zheng #include "locking.h" 305d4f98a2SYan Zheng #include "btrfs_inode.h" 315d4f98a2SYan Zheng #include "async-thread.h" 320af3d00bSJosef Bacik #include "free-space-cache.h" 33581bb050SLi Zefan #include "inode-map.h" 3462b99540SQu Wenruo #include "qgroup.h" 355d4f98a2SYan Zheng 365d4f98a2SYan Zheng /* 375d4f98a2SYan Zheng * backref_node, mapping_node and tree_block start with this 385d4f98a2SYan Zheng */ 395d4f98a2SYan Zheng struct tree_entry { 405d4f98a2SYan Zheng struct rb_node rb_node; 415d4f98a2SYan Zheng u64 bytenr; 425d4f98a2SYan Zheng }; 435d4f98a2SYan Zheng 445d4f98a2SYan Zheng /* 455d4f98a2SYan Zheng * present a tree block in the backref cache 465d4f98a2SYan Zheng */ 475d4f98a2SYan Zheng struct backref_node { 485d4f98a2SYan Zheng struct rb_node rb_node; 495d4f98a2SYan Zheng u64 bytenr; 503fd0a558SYan, Zheng 513fd0a558SYan, Zheng u64 new_bytenr; 523fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 535d4f98a2SYan Zheng u64 owner; 543fd0a558SYan, Zheng /* link to pending, changed or detached list */ 553fd0a558SYan, Zheng struct list_head list; 565d4f98a2SYan Zheng /* list of upper level blocks reference this block */ 575d4f98a2SYan Zheng struct list_head upper; 585d4f98a2SYan Zheng /* list of child blocks in the cache */ 595d4f98a2SYan Zheng struct list_head lower; 605d4f98a2SYan Zheng /* NULL if this node is not tree root */ 615d4f98a2SYan Zheng struct btrfs_root *root; 625d4f98a2SYan Zheng /* extent buffer got by COW the block */ 635d4f98a2SYan Zheng struct extent_buffer *eb; 645d4f98a2SYan Zheng /* level of tree block */ 655d4f98a2SYan Zheng unsigned int level:8; 663fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 673fd0a558SYan, Zheng unsigned int cowonly:1; 683fd0a558SYan, Zheng /* 1 if no child node in the cache */ 695d4f98a2SYan Zheng unsigned int lowest:1; 705d4f98a2SYan Zheng /* is the extent buffer locked */ 715d4f98a2SYan Zheng unsigned int locked:1; 725d4f98a2SYan Zheng /* has the block been processed */ 735d4f98a2SYan Zheng unsigned int processed:1; 745d4f98a2SYan Zheng /* have backrefs of this block been checked */ 755d4f98a2SYan Zheng unsigned int checked:1; 763fd0a558SYan, Zheng /* 773fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 783fd0a558SYan, Zheng * level block pointers may not point to the new location 793fd0a558SYan, Zheng */ 803fd0a558SYan, Zheng unsigned int pending:1; 813fd0a558SYan, Zheng /* 823fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 833fd0a558SYan, Zheng * backref node. 843fd0a558SYan, Zheng */ 853fd0a558SYan, Zheng unsigned int detached:1; 865d4f98a2SYan Zheng }; 875d4f98a2SYan Zheng 885d4f98a2SYan Zheng /* 895d4f98a2SYan Zheng * present a block pointer in the backref cache 905d4f98a2SYan Zheng */ 915d4f98a2SYan Zheng struct backref_edge { 925d4f98a2SYan Zheng struct list_head list[2]; 935d4f98a2SYan Zheng struct backref_node *node[2]; 945d4f98a2SYan Zheng }; 955d4f98a2SYan Zheng 965d4f98a2SYan Zheng #define LOWER 0 975d4f98a2SYan Zheng #define UPPER 1 980647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 995d4f98a2SYan Zheng 1005d4f98a2SYan Zheng struct backref_cache { 1015d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 1025d4f98a2SYan Zheng struct rb_root rb_root; 1033fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 1043fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 1053fd0a558SYan, Zheng /* 1063fd0a558SYan, Zheng * list of blocks that have been cowed but some block 1073fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 1083fd0a558SYan, Zheng * new location 1093fd0a558SYan, Zheng */ 1105d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1113fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1123fd0a558SYan, Zheng struct list_head leaves; 1133fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1143fd0a558SYan, Zheng struct list_head changed; 1153fd0a558SYan, Zheng /* list of detached backref node. */ 1163fd0a558SYan, Zheng struct list_head detached; 1173fd0a558SYan, Zheng 1183fd0a558SYan, Zheng u64 last_trans; 1193fd0a558SYan, Zheng 1203fd0a558SYan, Zheng int nr_nodes; 1213fd0a558SYan, Zheng int nr_edges; 1225d4f98a2SYan Zheng }; 1235d4f98a2SYan Zheng 1245d4f98a2SYan Zheng /* 1255d4f98a2SYan Zheng * map address of tree root to tree 1265d4f98a2SYan Zheng */ 1275d4f98a2SYan Zheng struct mapping_node { 1285d4f98a2SYan Zheng struct rb_node rb_node; 1295d4f98a2SYan Zheng u64 bytenr; 1305d4f98a2SYan Zheng void *data; 1315d4f98a2SYan Zheng }; 1325d4f98a2SYan Zheng 1335d4f98a2SYan Zheng struct mapping_tree { 1345d4f98a2SYan Zheng struct rb_root rb_root; 1355d4f98a2SYan Zheng spinlock_t lock; 1365d4f98a2SYan Zheng }; 1375d4f98a2SYan Zheng 1385d4f98a2SYan Zheng /* 1395d4f98a2SYan Zheng * present a tree block to process 1405d4f98a2SYan Zheng */ 1415d4f98a2SYan Zheng struct tree_block { 1425d4f98a2SYan Zheng struct rb_node rb_node; 1435d4f98a2SYan Zheng u64 bytenr; 1445d4f98a2SYan Zheng struct btrfs_key key; 1455d4f98a2SYan Zheng unsigned int level:8; 1465d4f98a2SYan Zheng unsigned int key_ready:1; 1475d4f98a2SYan Zheng }; 1485d4f98a2SYan Zheng 1490257bb82SYan, Zheng #define MAX_EXTENTS 128 1500257bb82SYan, Zheng 1510257bb82SYan, Zheng struct file_extent_cluster { 1520257bb82SYan, Zheng u64 start; 1530257bb82SYan, Zheng u64 end; 1540257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1550257bb82SYan, Zheng unsigned int nr; 1560257bb82SYan, Zheng }; 1570257bb82SYan, Zheng 1585d4f98a2SYan Zheng struct reloc_control { 1595d4f98a2SYan Zheng /* block group to relocate */ 1605d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group; 1615d4f98a2SYan Zheng /* extent tree */ 1625d4f98a2SYan Zheng struct btrfs_root *extent_root; 1635d4f98a2SYan Zheng /* inode for moving data */ 1645d4f98a2SYan Zheng struct inode *data_inode; 1653fd0a558SYan, Zheng 1663fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1673fd0a558SYan, Zheng 1683fd0a558SYan, Zheng struct backref_cache backref_cache; 1693fd0a558SYan, Zheng 1703fd0a558SYan, Zheng struct file_extent_cluster cluster; 1715d4f98a2SYan Zheng /* tree blocks have been processed */ 1725d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1735d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1745d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1755d4f98a2SYan Zheng /* list of reloc trees */ 1765d4f98a2SYan Zheng struct list_head reloc_roots; 1773fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1783fd0a558SYan, Zheng u64 merging_rsv_size; 1793fd0a558SYan, Zheng /* size of relocated tree nodes */ 1803fd0a558SYan, Zheng u64 nodes_relocated; 1810647bf56SWang Shilong /* reserved size for block group relocation*/ 1820647bf56SWang Shilong u64 reserved_bytes; 1833fd0a558SYan, Zheng 1845d4f98a2SYan Zheng u64 search_start; 1855d4f98a2SYan Zheng u64 extents_found; 1863fd0a558SYan, Zheng 1873fd0a558SYan, Zheng unsigned int stage:8; 1883fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1893fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1905d4f98a2SYan Zheng unsigned int found_file_extent:1; 1915d4f98a2SYan Zheng }; 1925d4f98a2SYan Zheng 1935d4f98a2SYan Zheng /* stages of data relocation */ 1945d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1955d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1965d4f98a2SYan Zheng 1973fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 1983fd0a558SYan, Zheng struct backref_node *node); 1993fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 2003fd0a558SYan, Zheng struct backref_node *node); 2015d4f98a2SYan Zheng 2025d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 2035d4f98a2SYan Zheng { 2046bef4d31SEric Paris tree->rb_root = RB_ROOT; 2055d4f98a2SYan Zheng spin_lock_init(&tree->lock); 2065d4f98a2SYan Zheng } 2075d4f98a2SYan Zheng 2085d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 2095d4f98a2SYan Zheng { 2105d4f98a2SYan Zheng int i; 2116bef4d31SEric Paris cache->rb_root = RB_ROOT; 2125d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2135d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2143fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2153fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2163fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2175d4f98a2SYan Zheng } 2185d4f98a2SYan Zheng 2193fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2205d4f98a2SYan Zheng { 2213fd0a558SYan, Zheng struct backref_node *node; 2223fd0a558SYan, Zheng int i; 2233fd0a558SYan, Zheng 2243fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2253fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2263fd0a558SYan, Zheng struct backref_node, list); 2273fd0a558SYan, Zheng remove_backref_node(cache, node); 2283fd0a558SYan, Zheng } 2293fd0a558SYan, Zheng 2303fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2313fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2323fd0a558SYan, Zheng struct backref_node, lower); 2333fd0a558SYan, Zheng remove_backref_node(cache, node); 2343fd0a558SYan, Zheng } 2353fd0a558SYan, Zheng 2363fd0a558SYan, Zheng cache->last_trans = 0; 2373fd0a558SYan, Zheng 2383fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 239f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 240f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 241f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 242f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 243f4907095SLiu Bo ASSERT(!cache->nr_nodes); 244f4907095SLiu Bo ASSERT(!cache->nr_edges); 2453fd0a558SYan, Zheng } 2463fd0a558SYan, Zheng 2473fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2483fd0a558SYan, Zheng { 2493fd0a558SYan, Zheng struct backref_node *node; 2503fd0a558SYan, Zheng 2513fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2523fd0a558SYan, Zheng if (node) { 2533fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2545d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2555d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2565d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2573fd0a558SYan, Zheng cache->nr_nodes++; 2583fd0a558SYan, Zheng } 2593fd0a558SYan, Zheng return node; 2603fd0a558SYan, Zheng } 2613fd0a558SYan, Zheng 2623fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 2633fd0a558SYan, Zheng struct backref_node *node) 2643fd0a558SYan, Zheng { 2653fd0a558SYan, Zheng if (node) { 2663fd0a558SYan, Zheng cache->nr_nodes--; 2673fd0a558SYan, Zheng kfree(node); 2683fd0a558SYan, Zheng } 2693fd0a558SYan, Zheng } 2703fd0a558SYan, Zheng 2713fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 2723fd0a558SYan, Zheng { 2733fd0a558SYan, Zheng struct backref_edge *edge; 2743fd0a558SYan, Zheng 2753fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 2763fd0a558SYan, Zheng if (edge) 2773fd0a558SYan, Zheng cache->nr_edges++; 2783fd0a558SYan, Zheng return edge; 2793fd0a558SYan, Zheng } 2803fd0a558SYan, Zheng 2813fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 2823fd0a558SYan, Zheng struct backref_edge *edge) 2833fd0a558SYan, Zheng { 2843fd0a558SYan, Zheng if (edge) { 2853fd0a558SYan, Zheng cache->nr_edges--; 2863fd0a558SYan, Zheng kfree(edge); 2873fd0a558SYan, Zheng } 2885d4f98a2SYan Zheng } 2895d4f98a2SYan Zheng 2905d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 2915d4f98a2SYan Zheng struct rb_node *node) 2925d4f98a2SYan Zheng { 2935d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 2945d4f98a2SYan Zheng struct rb_node *parent = NULL; 2955d4f98a2SYan Zheng struct tree_entry *entry; 2965d4f98a2SYan Zheng 2975d4f98a2SYan Zheng while (*p) { 2985d4f98a2SYan Zheng parent = *p; 2995d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 3005d4f98a2SYan Zheng 3015d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3025d4f98a2SYan Zheng p = &(*p)->rb_left; 3035d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3045d4f98a2SYan Zheng p = &(*p)->rb_right; 3055d4f98a2SYan Zheng else 3065d4f98a2SYan Zheng return parent; 3075d4f98a2SYan Zheng } 3085d4f98a2SYan Zheng 3095d4f98a2SYan Zheng rb_link_node(node, parent, p); 3105d4f98a2SYan Zheng rb_insert_color(node, root); 3115d4f98a2SYan Zheng return NULL; 3125d4f98a2SYan Zheng } 3135d4f98a2SYan Zheng 3145d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3155d4f98a2SYan Zheng { 3165d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3175d4f98a2SYan Zheng struct tree_entry *entry; 3185d4f98a2SYan Zheng 3195d4f98a2SYan Zheng while (n) { 3205d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3215d4f98a2SYan Zheng 3225d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3235d4f98a2SYan Zheng n = n->rb_left; 3245d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3255d4f98a2SYan Zheng n = n->rb_right; 3265d4f98a2SYan Zheng else 3275d4f98a2SYan Zheng return n; 3285d4f98a2SYan Zheng } 3295d4f98a2SYan Zheng return NULL; 3305d4f98a2SYan Zheng } 3315d4f98a2SYan Zheng 33248a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 33343c04fb1SJeff Mahoney { 33443c04fb1SJeff Mahoney 33543c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 33643c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 33743c04fb1SJeff Mahoney rb_node); 33843c04fb1SJeff Mahoney if (bnode->root) 33943c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 3405d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 3415d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 3425d163e0eSJeff Mahoney bytenr); 34343c04fb1SJeff Mahoney } 34443c04fb1SJeff Mahoney 3455d4f98a2SYan Zheng /* 3465d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3475d4f98a2SYan Zheng */ 3485d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3495d4f98a2SYan Zheng struct backref_edge *edges[], 3505d4f98a2SYan Zheng int *index) 3515d4f98a2SYan Zheng { 3525d4f98a2SYan Zheng struct backref_edge *edge; 3535d4f98a2SYan Zheng int idx = *index; 3545d4f98a2SYan Zheng 3555d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3565d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3575d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3585d4f98a2SYan Zheng edges[idx++] = edge; 3595d4f98a2SYan Zheng node = edge->node[UPPER]; 3605d4f98a2SYan Zheng } 3613fd0a558SYan, Zheng BUG_ON(node->detached); 3625d4f98a2SYan Zheng *index = idx; 3635d4f98a2SYan Zheng return node; 3645d4f98a2SYan Zheng } 3655d4f98a2SYan Zheng 3665d4f98a2SYan Zheng /* 3675d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 3685d4f98a2SYan Zheng */ 3695d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 3705d4f98a2SYan Zheng int *index) 3715d4f98a2SYan Zheng { 3725d4f98a2SYan Zheng struct backref_edge *edge; 3735d4f98a2SYan Zheng struct backref_node *lower; 3745d4f98a2SYan Zheng int idx = *index; 3755d4f98a2SYan Zheng 3765d4f98a2SYan Zheng while (idx > 0) { 3775d4f98a2SYan Zheng edge = edges[idx - 1]; 3785d4f98a2SYan Zheng lower = edge->node[LOWER]; 3795d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 3805d4f98a2SYan Zheng idx--; 3815d4f98a2SYan Zheng continue; 3825d4f98a2SYan Zheng } 3835d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 3845d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3855d4f98a2SYan Zheng edges[idx - 1] = edge; 3865d4f98a2SYan Zheng *index = idx; 3875d4f98a2SYan Zheng return edge->node[UPPER]; 3885d4f98a2SYan Zheng } 3895d4f98a2SYan Zheng *index = 0; 3905d4f98a2SYan Zheng return NULL; 3915d4f98a2SYan Zheng } 3925d4f98a2SYan Zheng 3933fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 3945d4f98a2SYan Zheng { 3955d4f98a2SYan Zheng if (node->locked) { 3965d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 3975d4f98a2SYan Zheng node->locked = 0; 3985d4f98a2SYan Zheng } 3993fd0a558SYan, Zheng } 4003fd0a558SYan, Zheng 4013fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 4023fd0a558SYan, Zheng { 4033fd0a558SYan, Zheng if (node->eb) { 4043fd0a558SYan, Zheng unlock_node_buffer(node); 4055d4f98a2SYan Zheng free_extent_buffer(node->eb); 4065d4f98a2SYan Zheng node->eb = NULL; 4075d4f98a2SYan Zheng } 4085d4f98a2SYan Zheng } 4095d4f98a2SYan Zheng 4105d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4115d4f98a2SYan Zheng struct backref_node *node) 4125d4f98a2SYan Zheng { 4135d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4145d4f98a2SYan Zheng 4155d4f98a2SYan Zheng drop_node_buffer(node); 4163fd0a558SYan, Zheng list_del(&node->list); 4175d4f98a2SYan Zheng list_del(&node->lower); 4183fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4195d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4203fd0a558SYan, Zheng free_backref_node(tree, node); 4215d4f98a2SYan Zheng } 4225d4f98a2SYan Zheng 4235d4f98a2SYan Zheng /* 4245d4f98a2SYan Zheng * remove a backref node from the backref cache 4255d4f98a2SYan Zheng */ 4265d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4275d4f98a2SYan Zheng struct backref_node *node) 4285d4f98a2SYan Zheng { 4295d4f98a2SYan Zheng struct backref_node *upper; 4305d4f98a2SYan Zheng struct backref_edge *edge; 4315d4f98a2SYan Zheng 4325d4f98a2SYan Zheng if (!node) 4335d4f98a2SYan Zheng return; 4345d4f98a2SYan Zheng 4353fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4365d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4375d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4385d4f98a2SYan Zheng list[LOWER]); 4395d4f98a2SYan Zheng upper = edge->node[UPPER]; 4405d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4415d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4423fd0a558SYan, Zheng free_backref_edge(cache, edge); 4433fd0a558SYan, Zheng 4443fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4453fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4463fd0a558SYan, Zheng drop_backref_node(cache, node); 4473fd0a558SYan, Zheng node = upper; 4483fd0a558SYan, Zheng node->lowest = 1; 4493fd0a558SYan, Zheng continue; 4503fd0a558SYan, Zheng } 4515d4f98a2SYan Zheng /* 4523fd0a558SYan, Zheng * add the node to leaf node list if no other 4535d4f98a2SYan Zheng * child block cached. 4545d4f98a2SYan Zheng */ 4555d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4563fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4575d4f98a2SYan Zheng upper->lowest = 1; 4585d4f98a2SYan Zheng } 4595d4f98a2SYan Zheng } 4603fd0a558SYan, Zheng 4615d4f98a2SYan Zheng drop_backref_node(cache, node); 4625d4f98a2SYan Zheng } 4635d4f98a2SYan Zheng 4643fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 4653fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 4663fd0a558SYan, Zheng { 4673fd0a558SYan, Zheng struct rb_node *rb_node; 4683fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 4693fd0a558SYan, Zheng node->bytenr = bytenr; 4703fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 47143c04fb1SJeff Mahoney if (rb_node) 47243c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 4733fd0a558SYan, Zheng } 4743fd0a558SYan, Zheng 4753fd0a558SYan, Zheng /* 4763fd0a558SYan, Zheng * update backref cache after a transaction commit 4773fd0a558SYan, Zheng */ 4783fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 4793fd0a558SYan, Zheng struct backref_cache *cache) 4803fd0a558SYan, Zheng { 4813fd0a558SYan, Zheng struct backref_node *node; 4823fd0a558SYan, Zheng int level = 0; 4833fd0a558SYan, Zheng 4843fd0a558SYan, Zheng if (cache->last_trans == 0) { 4853fd0a558SYan, Zheng cache->last_trans = trans->transid; 4863fd0a558SYan, Zheng return 0; 4873fd0a558SYan, Zheng } 4883fd0a558SYan, Zheng 4893fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 4903fd0a558SYan, Zheng return 0; 4913fd0a558SYan, Zheng 4923fd0a558SYan, Zheng /* 4933fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 4943fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 4953fd0a558SYan, Zheng * so the detached nodes are no longer useful. 4963fd0a558SYan, Zheng */ 4973fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 4983fd0a558SYan, Zheng node = list_entry(cache->detached.next, 4993fd0a558SYan, Zheng struct backref_node, list); 5003fd0a558SYan, Zheng remove_backref_node(cache, node); 5013fd0a558SYan, Zheng } 5023fd0a558SYan, Zheng 5033fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 5043fd0a558SYan, Zheng node = list_entry(cache->changed.next, 5053fd0a558SYan, Zheng struct backref_node, list); 5063fd0a558SYan, Zheng list_del_init(&node->list); 5073fd0a558SYan, Zheng BUG_ON(node->pending); 5083fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5093fd0a558SYan, Zheng } 5103fd0a558SYan, Zheng 5113fd0a558SYan, Zheng /* 5123fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5133fd0a558SYan, Zheng * errors during processing the pending nodes. 5143fd0a558SYan, Zheng */ 5153fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5163fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5173fd0a558SYan, Zheng BUG_ON(!node->pending); 5183fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5193fd0a558SYan, Zheng continue; 5203fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5213fd0a558SYan, Zheng } 5223fd0a558SYan, Zheng } 5233fd0a558SYan, Zheng 5243fd0a558SYan, Zheng cache->last_trans = 0; 5253fd0a558SYan, Zheng return 1; 5263fd0a558SYan, Zheng } 5273fd0a558SYan, Zheng 528f2a97a9dSDavid Sterba 5293fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5303fd0a558SYan, Zheng { 5313fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5323fd0a558SYan, Zheng 53327cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 5343fd0a558SYan, Zheng return 0; 5353fd0a558SYan, Zheng 5363fd0a558SYan, Zheng reloc_root = root->reloc_root; 5373fd0a558SYan, Zheng if (!reloc_root) 5383fd0a558SYan, Zheng return 0; 5393fd0a558SYan, Zheng 5403fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5413fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5423fd0a558SYan, Zheng return 0; 5433fd0a558SYan, Zheng /* 5443fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5453fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5463fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5473fd0a558SYan, Zheng * relocation. 5483fd0a558SYan, Zheng */ 5493fd0a558SYan, Zheng return 1; 5503fd0a558SYan, Zheng } 5515d4f98a2SYan Zheng /* 5525d4f98a2SYan Zheng * find reloc tree by address of tree root 5535d4f98a2SYan Zheng */ 5545d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5555d4f98a2SYan Zheng u64 bytenr) 5565d4f98a2SYan Zheng { 5575d4f98a2SYan Zheng struct rb_node *rb_node; 5585d4f98a2SYan Zheng struct mapping_node *node; 5595d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5605d4f98a2SYan Zheng 5615d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5625d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5635d4f98a2SYan Zheng if (rb_node) { 5645d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5655d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5665d4f98a2SYan Zheng } 5675d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5685d4f98a2SYan Zheng return root; 5695d4f98a2SYan Zheng } 5705d4f98a2SYan Zheng 5715d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5725d4f98a2SYan Zheng { 5735d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5745d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5755d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 5765d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 5775d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 57866463748SWang Shilong root_objectid == BTRFS_CSUM_TREE_OBJECTID || 57966463748SWang Shilong root_objectid == BTRFS_UUID_TREE_OBJECTID || 5803e4c5efbSDavid Sterba root_objectid == BTRFS_QUOTA_TREE_OBJECTID || 5813e4c5efbSDavid Sterba root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 5825d4f98a2SYan Zheng return 1; 5835d4f98a2SYan Zheng return 0; 5845d4f98a2SYan Zheng } 5855d4f98a2SYan Zheng 5865d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 5875d4f98a2SYan Zheng u64 root_objectid) 5885d4f98a2SYan Zheng { 5895d4f98a2SYan Zheng struct btrfs_key key; 5905d4f98a2SYan Zheng 5915d4f98a2SYan Zheng key.objectid = root_objectid; 5925d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 5935d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 5945d4f98a2SYan Zheng key.offset = 0; 5955d4f98a2SYan Zheng else 5965d4f98a2SYan Zheng key.offset = (u64)-1; 5975d4f98a2SYan Zheng 598c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 5995d4f98a2SYan Zheng } 6005d4f98a2SYan Zheng 6015d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 6025d4f98a2SYan Zheng static noinline_for_stack 6035d4f98a2SYan Zheng struct btrfs_root *find_tree_root(struct reloc_control *rc, 6045d4f98a2SYan Zheng struct extent_buffer *leaf, 6055d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0) 6065d4f98a2SYan Zheng { 6075d4f98a2SYan Zheng struct btrfs_root *root; 6085d4f98a2SYan Zheng u64 root_objectid = btrfs_ref_root_v0(leaf, ref0); 6095d4f98a2SYan Zheng u64 generation = btrfs_ref_generation_v0(leaf, ref0); 6105d4f98a2SYan Zheng 6115d4f98a2SYan Zheng BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID); 6125d4f98a2SYan Zheng 6135d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, root_objectid); 6145d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 6155d4f98a2SYan Zheng 61627cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && 6175d4f98a2SYan Zheng generation != btrfs_root_generation(&root->root_item)) 6185d4f98a2SYan Zheng return NULL; 6195d4f98a2SYan Zheng 6205d4f98a2SYan Zheng return root; 6215d4f98a2SYan Zheng } 6225d4f98a2SYan Zheng #endif 6235d4f98a2SYan Zheng 6245d4f98a2SYan Zheng static noinline_for_stack 6255d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 6265d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 6275d4f98a2SYan Zheng { 6283173a18fSJosef Bacik struct btrfs_key key; 6295d4f98a2SYan Zheng struct btrfs_extent_item *ei; 6305d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6315d4f98a2SYan Zheng u32 item_size; 6325d4f98a2SYan Zheng 6333173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6343173a18fSJosef Bacik 6355d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 6365d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 6375d4f98a2SYan Zheng if (item_size < sizeof(*ei)) { 6385d4f98a2SYan Zheng WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 6395d4f98a2SYan Zheng return 1; 6405d4f98a2SYan Zheng } 6415d4f98a2SYan Zheng #endif 6425d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6435d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6445d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6455d4f98a2SYan Zheng 6463173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6473173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6485d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6495d4f98a2SYan Zheng return 1; 6505d4f98a2SYan Zheng } 651d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 652d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 653d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 654d062d13cSJosef Bacik return 1; 655d062d13cSJosef Bacik } 6565d4f98a2SYan Zheng 6573173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6585d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6595d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6603173a18fSJosef Bacik } else { 6613173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6623173a18fSJosef Bacik } 6635d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6645d4f98a2SYan Zheng return 0; 6655d4f98a2SYan Zheng } 6665d4f98a2SYan Zheng 6675d4f98a2SYan Zheng /* 6685d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6695d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6705d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6715d4f98a2SYan Zheng * 6725d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 67301327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 67401327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 6755d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6765d4f98a2SYan Zheng * 6775d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6785d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6795d4f98a2SYan Zheng * block are also cached. 6805d4f98a2SYan Zheng */ 6813fd0a558SYan, Zheng static noinline_for_stack 6823fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6835d4f98a2SYan Zheng struct btrfs_key *node_key, 6845d4f98a2SYan Zheng int level, u64 bytenr) 6855d4f98a2SYan Zheng { 6863fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 6875d4f98a2SYan Zheng struct btrfs_path *path1; 6885d4f98a2SYan Zheng struct btrfs_path *path2; 6895d4f98a2SYan Zheng struct extent_buffer *eb; 6905d4f98a2SYan Zheng struct btrfs_root *root; 6915d4f98a2SYan Zheng struct backref_node *cur; 6925d4f98a2SYan Zheng struct backref_node *upper; 6935d4f98a2SYan Zheng struct backref_node *lower; 6945d4f98a2SYan Zheng struct backref_node *node = NULL; 6955d4f98a2SYan Zheng struct backref_node *exist = NULL; 6965d4f98a2SYan Zheng struct backref_edge *edge; 6975d4f98a2SYan Zheng struct rb_node *rb_node; 6985d4f98a2SYan Zheng struct btrfs_key key; 6995d4f98a2SYan Zheng unsigned long end; 7005d4f98a2SYan Zheng unsigned long ptr; 7015d4f98a2SYan Zheng LIST_HEAD(list); 7023fd0a558SYan, Zheng LIST_HEAD(useless); 7033fd0a558SYan, Zheng int cowonly; 7045d4f98a2SYan Zheng int ret; 7055d4f98a2SYan Zheng int err = 0; 706b6c60c80SJosef Bacik bool need_check = true; 7075d4f98a2SYan Zheng 7085d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 7095d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 7105d4f98a2SYan Zheng if (!path1 || !path2) { 7115d4f98a2SYan Zheng err = -ENOMEM; 7125d4f98a2SYan Zheng goto out; 7135d4f98a2SYan Zheng } 714e4058b54SDavid Sterba path1->reada = READA_FORWARD; 715e4058b54SDavid Sterba path2->reada = READA_FORWARD; 7165d4f98a2SYan Zheng 7173fd0a558SYan, Zheng node = alloc_backref_node(cache); 7185d4f98a2SYan Zheng if (!node) { 7195d4f98a2SYan Zheng err = -ENOMEM; 7205d4f98a2SYan Zheng goto out; 7215d4f98a2SYan Zheng } 7225d4f98a2SYan Zheng 7235d4f98a2SYan Zheng node->bytenr = bytenr; 7245d4f98a2SYan Zheng node->level = level; 7255d4f98a2SYan Zheng node->lowest = 1; 7265d4f98a2SYan Zheng cur = node; 7275d4f98a2SYan Zheng again: 7285d4f98a2SYan Zheng end = 0; 7295d4f98a2SYan Zheng ptr = 0; 7305d4f98a2SYan Zheng key.objectid = cur->bytenr; 7313173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7325d4f98a2SYan Zheng key.offset = (u64)-1; 7335d4f98a2SYan Zheng 7345d4f98a2SYan Zheng path1->search_commit_root = 1; 7355d4f98a2SYan Zheng path1->skip_locking = 1; 7365d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7375d4f98a2SYan Zheng 0, 0); 7385d4f98a2SYan Zheng if (ret < 0) { 7395d4f98a2SYan Zheng err = ret; 7405d4f98a2SYan Zheng goto out; 7415d4f98a2SYan Zheng } 74275bfb9afSJosef Bacik ASSERT(ret); 74375bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7445d4f98a2SYan Zheng 7455d4f98a2SYan Zheng path1->slots[0]--; 7465d4f98a2SYan Zheng 7475d4f98a2SYan Zheng WARN_ON(cur->checked); 7485d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7495d4f98a2SYan Zheng /* 75070f23fd6SJustin P. Mattock * the backref was added previously when processing 7515d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7525d4f98a2SYan Zheng */ 75375bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7545d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7555d4f98a2SYan Zheng list[LOWER]); 75675bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7575d4f98a2SYan Zheng exist = edge->node[UPPER]; 7585d4f98a2SYan Zheng /* 7595d4f98a2SYan Zheng * add the upper level block to pending list if we need 7605d4f98a2SYan Zheng * check its backrefs 7615d4f98a2SYan Zheng */ 7625d4f98a2SYan Zheng if (!exist->checked) 7635d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7645d4f98a2SYan Zheng } else { 7655d4f98a2SYan Zheng exist = NULL; 7665d4f98a2SYan Zheng } 7675d4f98a2SYan Zheng 7685d4f98a2SYan Zheng while (1) { 7695d4f98a2SYan Zheng cond_resched(); 7705d4f98a2SYan Zheng eb = path1->nodes[0]; 7715d4f98a2SYan Zheng 7725d4f98a2SYan Zheng if (ptr >= end) { 7735d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7745d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7755d4f98a2SYan Zheng if (ret < 0) { 7765d4f98a2SYan Zheng err = ret; 7775d4f98a2SYan Zheng goto out; 7785d4f98a2SYan Zheng } 7795d4f98a2SYan Zheng if (ret > 0) 7805d4f98a2SYan Zheng break; 7815d4f98a2SYan Zheng eb = path1->nodes[0]; 7825d4f98a2SYan Zheng } 7835d4f98a2SYan Zheng 7845d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7855d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7865d4f98a2SYan Zheng WARN_ON(exist); 7875d4f98a2SYan Zheng break; 7885d4f98a2SYan Zheng } 7895d4f98a2SYan Zheng 7903173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7913173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7925d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7935d4f98a2SYan Zheng &ptr, &end); 7945d4f98a2SYan Zheng if (ret) 7955d4f98a2SYan Zheng goto next; 7965d4f98a2SYan Zheng } 7975d4f98a2SYan Zheng } 7985d4f98a2SYan Zheng 7995d4f98a2SYan Zheng if (ptr < end) { 8005d4f98a2SYan Zheng /* update key for inline back ref */ 8015d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 8025d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 8035d4f98a2SYan Zheng key.type = btrfs_extent_inline_ref_type(eb, iref); 8045d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 8055d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 8065d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 8075d4f98a2SYan Zheng } 8085d4f98a2SYan Zheng 8095d4f98a2SYan Zheng if (exist && 8105d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8115d4f98a2SYan Zheng exist->owner == key.offset) || 8125d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8135d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8145d4f98a2SYan Zheng exist = NULL; 8155d4f98a2SYan Zheng goto next; 8165d4f98a2SYan Zheng } 8175d4f98a2SYan Zheng 8185d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 8195d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY || 8205d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 8213fd0a558SYan, Zheng if (key.type == BTRFS_EXTENT_REF_V0_KEY) { 8225d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 8235d4f98a2SYan Zheng ref0 = btrfs_item_ptr(eb, path1->slots[0], 8245d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 8253fd0a558SYan, Zheng if (key.objectid == key.offset) { 826046f264fSYan, Zheng root = find_tree_root(rc, eb, ref0); 8273fd0a558SYan, Zheng if (root && !should_ignore_root(root)) 8285d4f98a2SYan Zheng cur->root = root; 8295d4f98a2SYan Zheng else 8303fd0a558SYan, Zheng list_add(&cur->list, &useless); 8315d4f98a2SYan Zheng break; 8325d4f98a2SYan Zheng } 833046f264fSYan, Zheng if (is_cowonly_root(btrfs_ref_root_v0(eb, 834046f264fSYan, Zheng ref0))) 835046f264fSYan, Zheng cur->cowonly = 1; 8363fd0a558SYan, Zheng } 8375d4f98a2SYan Zheng #else 83875bfb9afSJosef Bacik ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY); 8395d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8405d4f98a2SYan Zheng #endif 8415d4f98a2SYan Zheng if (key.objectid == key.offset) { 8425d4f98a2SYan Zheng /* 8435d4f98a2SYan Zheng * only root blocks of reloc trees use 8445d4f98a2SYan Zheng * backref of this type. 8455d4f98a2SYan Zheng */ 8465d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 84775bfb9afSJosef Bacik ASSERT(root); 8485d4f98a2SYan Zheng cur->root = root; 8495d4f98a2SYan Zheng break; 8505d4f98a2SYan Zheng } 8515d4f98a2SYan Zheng 8523fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8535d4f98a2SYan Zheng if (!edge) { 8545d4f98a2SYan Zheng err = -ENOMEM; 8555d4f98a2SYan Zheng goto out; 8565d4f98a2SYan Zheng } 8575d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8585d4f98a2SYan Zheng if (!rb_node) { 8593fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8605d4f98a2SYan Zheng if (!upper) { 8613fd0a558SYan, Zheng free_backref_edge(cache, edge); 8625d4f98a2SYan Zheng err = -ENOMEM; 8635d4f98a2SYan Zheng goto out; 8645d4f98a2SYan Zheng } 8655d4f98a2SYan Zheng upper->bytenr = key.offset; 8665d4f98a2SYan Zheng upper->level = cur->level + 1; 8675d4f98a2SYan Zheng /* 8685d4f98a2SYan Zheng * backrefs for the upper level block isn't 8695d4f98a2SYan Zheng * cached, add the block to pending list 8705d4f98a2SYan Zheng */ 8715d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8725d4f98a2SYan Zheng } else { 8735d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8745d4f98a2SYan Zheng rb_node); 87575bfb9afSJosef Bacik ASSERT(upper->checked); 8765d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8775d4f98a2SYan Zheng } 8783fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8795d4f98a2SYan Zheng edge->node[LOWER] = cur; 8803fd0a558SYan, Zheng edge->node[UPPER] = upper; 8815d4f98a2SYan Zheng 8825d4f98a2SYan Zheng goto next; 8835d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8845d4f98a2SYan Zheng goto next; 8855d4f98a2SYan Zheng } 8865d4f98a2SYan Zheng 8875d4f98a2SYan Zheng /* key.type == BTRFS_TREE_BLOCK_REF_KEY */ 8885d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8895d4f98a2SYan Zheng if (IS_ERR(root)) { 8905d4f98a2SYan Zheng err = PTR_ERR(root); 8915d4f98a2SYan Zheng goto out; 8925d4f98a2SYan Zheng } 8935d4f98a2SYan Zheng 89427cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 8953fd0a558SYan, Zheng cur->cowonly = 1; 8963fd0a558SYan, Zheng 8975d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 8985d4f98a2SYan Zheng /* tree root */ 89975bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9005d4f98a2SYan Zheng cur->bytenr); 9013fd0a558SYan, Zheng if (should_ignore_root(root)) 9023fd0a558SYan, Zheng list_add(&cur->list, &useless); 9033fd0a558SYan, Zheng else 9045d4f98a2SYan Zheng cur->root = root; 9055d4f98a2SYan Zheng break; 9065d4f98a2SYan Zheng } 9075d4f98a2SYan Zheng 9085d4f98a2SYan Zheng level = cur->level + 1; 9095d4f98a2SYan Zheng 9105d4f98a2SYan Zheng /* 9115d4f98a2SYan Zheng * searching the tree to find upper level blocks 9125d4f98a2SYan Zheng * reference the block. 9135d4f98a2SYan Zheng */ 9145d4f98a2SYan Zheng path2->search_commit_root = 1; 9155d4f98a2SYan Zheng path2->skip_locking = 1; 9165d4f98a2SYan Zheng path2->lowest_level = level; 9175d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 9185d4f98a2SYan Zheng path2->lowest_level = 0; 9195d4f98a2SYan Zheng if (ret < 0) { 9205d4f98a2SYan Zheng err = ret; 9215d4f98a2SYan Zheng goto out; 9225d4f98a2SYan Zheng } 92333c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 92433c66f43SYan Zheng path2->slots[level]--; 9255d4f98a2SYan Zheng 9265d4f98a2SYan Zheng eb = path2->nodes[level]; 9273561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 9283561b9dbSLiu Bo cur->bytenr) { 9293561b9dbSLiu Bo btrfs_err(root->fs_info, 9303561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 9313561b9dbSLiu Bo cur->bytenr, level - 1, root->objectid, 9323561b9dbSLiu Bo node_key->objectid, node_key->type, 9333561b9dbSLiu Bo node_key->offset); 9343561b9dbSLiu Bo err = -ENOENT; 9353561b9dbSLiu Bo goto out; 9363561b9dbSLiu Bo } 9375d4f98a2SYan Zheng lower = cur; 938b6c60c80SJosef Bacik need_check = true; 9395d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9405d4f98a2SYan Zheng if (!path2->nodes[level]) { 94175bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9425d4f98a2SYan Zheng lower->bytenr); 9433fd0a558SYan, Zheng if (should_ignore_root(root)) 9443fd0a558SYan, Zheng list_add(&lower->list, &useless); 9453fd0a558SYan, Zheng else 9465d4f98a2SYan Zheng lower->root = root; 9475d4f98a2SYan Zheng break; 9485d4f98a2SYan Zheng } 9495d4f98a2SYan Zheng 9503fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9515d4f98a2SYan Zheng if (!edge) { 9525d4f98a2SYan Zheng err = -ENOMEM; 9535d4f98a2SYan Zheng goto out; 9545d4f98a2SYan Zheng } 9555d4f98a2SYan Zheng 9565d4f98a2SYan Zheng eb = path2->nodes[level]; 9575d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9585d4f98a2SYan Zheng if (!rb_node) { 9593fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9605d4f98a2SYan Zheng if (!upper) { 9613fd0a558SYan, Zheng free_backref_edge(cache, edge); 9625d4f98a2SYan Zheng err = -ENOMEM; 9635d4f98a2SYan Zheng goto out; 9645d4f98a2SYan Zheng } 9655d4f98a2SYan Zheng upper->bytenr = eb->start; 9665d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9675d4f98a2SYan Zheng upper->level = lower->level + 1; 96827cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 96927cdeb70SMiao Xie &root->state)) 9703fd0a558SYan, Zheng upper->cowonly = 1; 9715d4f98a2SYan Zheng 9725d4f98a2SYan Zheng /* 9735d4f98a2SYan Zheng * if we know the block isn't shared 9745d4f98a2SYan Zheng * we can void checking its backrefs. 9755d4f98a2SYan Zheng */ 9765d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9775d4f98a2SYan Zheng upper->checked = 0; 9785d4f98a2SYan Zheng else 9795d4f98a2SYan Zheng upper->checked = 1; 9805d4f98a2SYan Zheng 9815d4f98a2SYan Zheng /* 9825d4f98a2SYan Zheng * add the block to pending list if we 983b6c60c80SJosef Bacik * need check its backrefs, we only do this once 984b6c60c80SJosef Bacik * while walking up a tree as we will catch 985b6c60c80SJosef Bacik * anything else later on. 9865d4f98a2SYan Zheng */ 987b6c60c80SJosef Bacik if (!upper->checked && need_check) { 988b6c60c80SJosef Bacik need_check = false; 9895d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9905d4f98a2SYan Zheng &list); 991bbe90514SJosef Bacik } else { 992bbe90514SJosef Bacik if (upper->checked) 993bbe90514SJosef Bacik need_check = true; 9945d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 995bbe90514SJosef Bacik } 9965d4f98a2SYan Zheng } else { 9975d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9985d4f98a2SYan Zheng rb_node); 99975bfb9afSJosef Bacik ASSERT(upper->checked); 10005d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 10013fd0a558SYan, Zheng if (!upper->owner) 10023fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 10035d4f98a2SYan Zheng } 10045d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 10055d4f98a2SYan Zheng edge->node[LOWER] = lower; 10063fd0a558SYan, Zheng edge->node[UPPER] = upper; 10075d4f98a2SYan Zheng 10085d4f98a2SYan Zheng if (rb_node) 10095d4f98a2SYan Zheng break; 10105d4f98a2SYan Zheng lower = upper; 10115d4f98a2SYan Zheng upper = NULL; 10125d4f98a2SYan Zheng } 1013b3b4aa74SDavid Sterba btrfs_release_path(path2); 10145d4f98a2SYan Zheng next: 10155d4f98a2SYan Zheng if (ptr < end) { 10165d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 10175d4f98a2SYan Zheng if (ptr >= end) { 10185d4f98a2SYan Zheng WARN_ON(ptr > end); 10195d4f98a2SYan Zheng ptr = 0; 10205d4f98a2SYan Zheng end = 0; 10215d4f98a2SYan Zheng } 10225d4f98a2SYan Zheng } 10235d4f98a2SYan Zheng if (ptr >= end) 10245d4f98a2SYan Zheng path1->slots[0]++; 10255d4f98a2SYan Zheng } 1026b3b4aa74SDavid Sterba btrfs_release_path(path1); 10275d4f98a2SYan Zheng 10285d4f98a2SYan Zheng cur->checked = 1; 10295d4f98a2SYan Zheng WARN_ON(exist); 10305d4f98a2SYan Zheng 10315d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10325d4f98a2SYan Zheng if (!list_empty(&list)) { 10335d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10345d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10355d4f98a2SYan Zheng cur = edge->node[UPPER]; 10365d4f98a2SYan Zheng goto again; 10375d4f98a2SYan Zheng } 10385d4f98a2SYan Zheng 10395d4f98a2SYan Zheng /* 10405d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10415d4f98a2SYan Zheng * into the cache. 10425d4f98a2SYan Zheng */ 104375bfb9afSJosef Bacik ASSERT(node->checked); 10443fd0a558SYan, Zheng cowonly = node->cowonly; 10453fd0a558SYan, Zheng if (!cowonly) { 10463fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10473fd0a558SYan, Zheng &node->rb_node); 104843c04fb1SJeff Mahoney if (rb_node) 104943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10503fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10513fd0a558SYan, Zheng } 10525d4f98a2SYan Zheng 10535d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10545d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10555d4f98a2SYan Zheng 10565d4f98a2SYan Zheng while (!list_empty(&list)) { 10575d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10585d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10595d4f98a2SYan Zheng upper = edge->node[UPPER]; 10603fd0a558SYan, Zheng if (upper->detached) { 10613fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10623fd0a558SYan, Zheng lower = edge->node[LOWER]; 10633fd0a558SYan, Zheng free_backref_edge(cache, edge); 10643fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10653fd0a558SYan, Zheng list_add(&lower->list, &useless); 10663fd0a558SYan, Zheng continue; 10673fd0a558SYan, Zheng } 10685d4f98a2SYan Zheng 10695d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10705d4f98a2SYan Zheng if (upper->lowest) { 10715d4f98a2SYan Zheng list_del_init(&upper->lower); 10725d4f98a2SYan Zheng upper->lowest = 0; 10735d4f98a2SYan Zheng } 10745d4f98a2SYan Zheng 10755d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10765d4f98a2SYan Zheng continue; 10775d4f98a2SYan Zheng } 10785d4f98a2SYan Zheng 107975bfb9afSJosef Bacik if (!upper->checked) { 108075bfb9afSJosef Bacik /* 108175bfb9afSJosef Bacik * Still want to blow up for developers since this is a 108275bfb9afSJosef Bacik * logic bug. 108375bfb9afSJosef Bacik */ 108475bfb9afSJosef Bacik ASSERT(0); 108575bfb9afSJosef Bacik err = -EINVAL; 108675bfb9afSJosef Bacik goto out; 108775bfb9afSJosef Bacik } 108875bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 108975bfb9afSJosef Bacik ASSERT(0); 109075bfb9afSJosef Bacik err = -EINVAL; 109175bfb9afSJosef Bacik goto out; 109275bfb9afSJosef Bacik } 109375bfb9afSJosef Bacik 10943fd0a558SYan, Zheng if (!cowonly) { 10955d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10965d4f98a2SYan Zheng &upper->rb_node); 109743c04fb1SJeff Mahoney if (rb_node) 109843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 109943c04fb1SJeff Mahoney upper->bytenr); 11003fd0a558SYan, Zheng } 11015d4f98a2SYan Zheng 11025d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11035d4f98a2SYan Zheng 11045d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 11055d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 11065d4f98a2SYan Zheng } 11073fd0a558SYan, Zheng /* 11083fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 11093fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 11103fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 11113fd0a558SYan, Zheng * lookup. 11123fd0a558SYan, Zheng */ 11133fd0a558SYan, Zheng while (!list_empty(&useless)) { 11143fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 11153fd0a558SYan, Zheng list_del_init(&upper->list); 111675bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 11173fd0a558SYan, Zheng if (upper == node) 11183fd0a558SYan, Zheng node = NULL; 11193fd0a558SYan, Zheng if (upper->lowest) { 11203fd0a558SYan, Zheng list_del_init(&upper->lower); 11213fd0a558SYan, Zheng upper->lowest = 0; 11223fd0a558SYan, Zheng } 11233fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 11243fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 11253fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 11263fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11273fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11283fd0a558SYan, Zheng lower = edge->node[LOWER]; 11293fd0a558SYan, Zheng free_backref_edge(cache, edge); 11303fd0a558SYan, Zheng 11313fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11323fd0a558SYan, Zheng list_add(&lower->list, &useless); 11333fd0a558SYan, Zheng } 11343fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11353fd0a558SYan, Zheng if (upper->level > 0) { 11363fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11373fd0a558SYan, Zheng upper->detached = 1; 11383fd0a558SYan, Zheng } else { 11393fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11403fd0a558SYan, Zheng free_backref_node(cache, upper); 11413fd0a558SYan, Zheng } 11423fd0a558SYan, Zheng } 11435d4f98a2SYan Zheng out: 11445d4f98a2SYan Zheng btrfs_free_path(path1); 11455d4f98a2SYan Zheng btrfs_free_path(path2); 11465d4f98a2SYan Zheng if (err) { 11473fd0a558SYan, Zheng while (!list_empty(&useless)) { 11483fd0a558SYan, Zheng lower = list_entry(useless.next, 114975bfb9afSJosef Bacik struct backref_node, list); 115075bfb9afSJosef Bacik list_del_init(&lower->list); 11513fd0a558SYan, Zheng } 115275bfb9afSJosef Bacik while (!list_empty(&list)) { 115375bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 115475bfb9afSJosef Bacik list[UPPER]); 115575bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11563fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 115775bfb9afSJosef Bacik lower = edge->node[LOWER]; 11585d4f98a2SYan Zheng upper = edge->node[UPPER]; 11593fd0a558SYan, Zheng free_backref_edge(cache, edge); 116075bfb9afSJosef Bacik 116175bfb9afSJosef Bacik /* 116275bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 116375bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 116475bfb9afSJosef Bacik */ 116575bfb9afSJosef Bacik if (list_empty(&lower->upper) && 116675bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 116775bfb9afSJosef Bacik list_add(&lower->list, &useless); 116875bfb9afSJosef Bacik 116975bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 117075bfb9afSJosef Bacik continue; 117175bfb9afSJosef Bacik 117201327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 117375bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 117475bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 117575bfb9afSJosef Bacik if (list_empty(&upper->upper)) 117675bfb9afSJosef Bacik list_add(&upper->list, &useless); 117775bfb9afSJosef Bacik } 117875bfb9afSJosef Bacik 117975bfb9afSJosef Bacik while (!list_empty(&useless)) { 118075bfb9afSJosef Bacik lower = list_entry(useless.next, 118175bfb9afSJosef Bacik struct backref_node, list); 118275bfb9afSJosef Bacik list_del_init(&lower->list); 11830fd8c3daSLiu Bo if (lower == node) 11840fd8c3daSLiu Bo node = NULL; 118575bfb9afSJosef Bacik free_backref_node(cache, lower); 11865d4f98a2SYan Zheng } 11870fd8c3daSLiu Bo 11880fd8c3daSLiu Bo free_backref_node(cache, node); 11895d4f98a2SYan Zheng return ERR_PTR(err); 11905d4f98a2SYan Zheng } 119175bfb9afSJosef Bacik ASSERT(!node || !node->detached); 11925d4f98a2SYan Zheng return node; 11935d4f98a2SYan Zheng } 11945d4f98a2SYan Zheng 11955d4f98a2SYan Zheng /* 11963fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11973fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11983fd0a558SYan, Zheng * corresponds to root of source tree 11993fd0a558SYan, Zheng */ 12003fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 12013fd0a558SYan, Zheng struct reloc_control *rc, 12023fd0a558SYan, Zheng struct btrfs_root *src, 12033fd0a558SYan, Zheng struct btrfs_root *dest) 12043fd0a558SYan, Zheng { 12053fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 12063fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 12073fd0a558SYan, Zheng struct backref_node *node = NULL; 12083fd0a558SYan, Zheng struct backref_node *new_node; 12093fd0a558SYan, Zheng struct backref_edge *edge; 12103fd0a558SYan, Zheng struct backref_edge *new_edge; 12113fd0a558SYan, Zheng struct rb_node *rb_node; 12123fd0a558SYan, Zheng 12133fd0a558SYan, Zheng if (cache->last_trans > 0) 12143fd0a558SYan, Zheng update_backref_cache(trans, cache); 12153fd0a558SYan, Zheng 12163fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 12173fd0a558SYan, Zheng if (rb_node) { 12183fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 12193fd0a558SYan, Zheng if (node->detached) 12203fd0a558SYan, Zheng node = NULL; 12213fd0a558SYan, Zheng else 12223fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 12233fd0a558SYan, Zheng } 12243fd0a558SYan, Zheng 12253fd0a558SYan, Zheng if (!node) { 12263fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 12273fd0a558SYan, Zheng reloc_root->commit_root->start); 12283fd0a558SYan, Zheng if (rb_node) { 12293fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 12303fd0a558SYan, Zheng rb_node); 12313fd0a558SYan, Zheng BUG_ON(node->detached); 12323fd0a558SYan, Zheng } 12333fd0a558SYan, Zheng } 12343fd0a558SYan, Zheng 12353fd0a558SYan, Zheng if (!node) 12363fd0a558SYan, Zheng return 0; 12373fd0a558SYan, Zheng 12383fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12393fd0a558SYan, Zheng if (!new_node) 12403fd0a558SYan, Zheng return -ENOMEM; 12413fd0a558SYan, Zheng 12423fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12433fd0a558SYan, Zheng new_node->level = node->level; 12443fd0a558SYan, Zheng new_node->lowest = node->lowest; 12456848ad64SYan, Zheng new_node->checked = 1; 12463fd0a558SYan, Zheng new_node->root = dest; 12473fd0a558SYan, Zheng 12483fd0a558SYan, Zheng if (!node->lowest) { 12493fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12503fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12513fd0a558SYan, Zheng if (!new_edge) 12523fd0a558SYan, Zheng goto fail; 12533fd0a558SYan, Zheng 12543fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12553fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12563fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12573fd0a558SYan, Zheng &new_node->lower); 12583fd0a558SYan, Zheng } 125976b9e23dSMiao Xie } else { 126076b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12613fd0a558SYan, Zheng } 12623fd0a558SYan, Zheng 12633fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12643fd0a558SYan, Zheng &new_node->rb_node); 126543c04fb1SJeff Mahoney if (rb_node) 126643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12673fd0a558SYan, Zheng 12683fd0a558SYan, Zheng if (!new_node->lowest) { 12693fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12703fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12713fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12723fd0a558SYan, Zheng } 12733fd0a558SYan, Zheng } 12743fd0a558SYan, Zheng return 0; 12753fd0a558SYan, Zheng fail: 12763fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12773fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12783fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12793fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12803fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12813fd0a558SYan, Zheng } 12823fd0a558SYan, Zheng free_backref_node(cache, new_node); 12833fd0a558SYan, Zheng return -ENOMEM; 12843fd0a558SYan, Zheng } 12853fd0a558SYan, Zheng 12863fd0a558SYan, Zheng /* 12875d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12885d4f98a2SYan Zheng */ 1289ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12905d4f98a2SYan Zheng { 12915d4f98a2SYan Zheng struct rb_node *rb_node; 12925d4f98a2SYan Zheng struct mapping_node *node; 12935d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 12945d4f98a2SYan Zheng 12955d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1296ffd7b339SJeff Mahoney if (!node) 1297ffd7b339SJeff Mahoney return -ENOMEM; 12985d4f98a2SYan Zheng 12995d4f98a2SYan Zheng node->bytenr = root->node->start; 13005d4f98a2SYan Zheng node->data = root; 13015d4f98a2SYan Zheng 13025d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13035d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13045d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13055d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1306ffd7b339SJeff Mahoney if (rb_node) { 13075d163e0eSJeff Mahoney btrfs_panic(root->fs_info, -EEXIST, 13085d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 13095d163e0eSJeff Mahoney node->bytenr); 131023291a04SDan Carpenter kfree(node); 131123291a04SDan Carpenter return -EEXIST; 1312ffd7b339SJeff Mahoney } 13135d4f98a2SYan Zheng 13145d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 13155d4f98a2SYan Zheng return 0; 13165d4f98a2SYan Zheng } 13175d4f98a2SYan Zheng 13185d4f98a2SYan Zheng /* 1319c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 13205d4f98a2SYan Zheng * mapping 13215d4f98a2SYan Zheng */ 1322c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 13235d4f98a2SYan Zheng { 13245d4f98a2SYan Zheng struct rb_node *rb_node; 13255d4f98a2SYan Zheng struct mapping_node *node = NULL; 13265d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 13275d4f98a2SYan Zheng 13285d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13295d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1330c974c464SWang Shilong root->node->start); 1331c974c464SWang Shilong if (rb_node) { 1332c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1333c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1334c974c464SWang Shilong } 1335c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1336c974c464SWang Shilong 1337c974c464SWang Shilong if (!node) 1338c974c464SWang Shilong return; 1339c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1340c974c464SWang Shilong 1341c974c464SWang Shilong spin_lock(&root->fs_info->trans_lock); 1342c974c464SWang Shilong list_del_init(&root->root_list); 1343c974c464SWang Shilong spin_unlock(&root->fs_info->trans_lock); 1344c974c464SWang Shilong kfree(node); 1345c974c464SWang Shilong } 1346c974c464SWang Shilong 1347c974c464SWang Shilong /* 1348c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1349c974c464SWang Shilong * mapping 1350c974c464SWang Shilong */ 1351c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1352c974c464SWang Shilong { 1353c974c464SWang Shilong struct rb_node *rb_node; 1354c974c464SWang Shilong struct mapping_node *node = NULL; 1355c974c464SWang Shilong struct reloc_control *rc = root->fs_info->reloc_ctl; 1356c974c464SWang Shilong 1357c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1358c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1359c974c464SWang Shilong root->node->start); 13605d4f98a2SYan Zheng if (rb_node) { 13615d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13625d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13635d4f98a2SYan Zheng } 13645d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13655d4f98a2SYan Zheng 13668f71f3e0SLiu Bo if (!node) 13678f71f3e0SLiu Bo return 0; 13685d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13695d4f98a2SYan Zheng 13705d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1371c974c464SWang Shilong node->bytenr = new_bytenr; 13725d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13735d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13745d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 137543c04fb1SJeff Mahoney if (rb_node) 137643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13775d4f98a2SYan Zheng return 0; 13785d4f98a2SYan Zheng } 13795d4f98a2SYan Zheng 13803fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13813fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13825d4f98a2SYan Zheng { 13835d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13845d4f98a2SYan Zheng struct extent_buffer *eb; 13855d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13865d4f98a2SYan Zheng struct btrfs_key root_key; 13875bc7247aSMiao Xie u64 last_snap = 0; 13885d4f98a2SYan Zheng int ret; 13895d4f98a2SYan Zheng 13905d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13915d4f98a2SYan Zheng BUG_ON(!root_item); 13925d4f98a2SYan Zheng 13935d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13945d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13953fd0a558SYan, Zheng root_key.offset = objectid; 13965d4f98a2SYan Zheng 13973fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 13983fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 13995d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 14005d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 14015d4f98a2SYan Zheng BUG_ON(ret); 14025d4f98a2SYan Zheng 14035bc7247aSMiao Xie last_snap = btrfs_root_last_snapshot(&root->root_item); 14043fd0a558SYan, Zheng btrfs_set_root_last_snapshot(&root->root_item, 14053fd0a558SYan, Zheng trans->transid - 1); 14063fd0a558SYan, Zheng } else { 14073fd0a558SYan, Zheng /* 14083fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 14093fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 14103fd0a558SYan, Zheng * modified after it was created have RELOC flag 14113fd0a558SYan, Zheng * set in their headers. so it's OK to not update 14123fd0a558SYan, Zheng * the 'last_snapshot'. 14133fd0a558SYan, Zheng */ 14143fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 14153fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 14163fd0a558SYan, Zheng BUG_ON(ret); 14173fd0a558SYan, Zheng } 14183fd0a558SYan, Zheng 14195d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 14205d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14215d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14225d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14233fd0a558SYan, Zheng 14243fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14253fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14263fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14273fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14285d4f98a2SYan Zheng root_item->drop_level = 0; 14295bc7247aSMiao Xie /* 14305bc7247aSMiao Xie * abuse rtransid, it is safe because it is impossible to 14315bc7247aSMiao Xie * receive data into a relocation tree. 14325bc7247aSMiao Xie */ 14335bc7247aSMiao Xie btrfs_set_root_rtransid(root_item, last_snap); 14345bc7247aSMiao Xie btrfs_set_root_otransid(root_item, trans->transid); 14353fd0a558SYan, Zheng } 14365d4f98a2SYan Zheng 14375d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14385d4f98a2SYan Zheng free_extent_buffer(eb); 14395d4f98a2SYan Zheng 14405d4f98a2SYan Zheng ret = btrfs_insert_root(trans, root->fs_info->tree_root, 14415d4f98a2SYan Zheng &root_key, root_item); 14425d4f98a2SYan Zheng BUG_ON(ret); 14435d4f98a2SYan Zheng kfree(root_item); 14445d4f98a2SYan Zheng 1445cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root->fs_info->tree_root, &root_key); 14465d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14475d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14483fd0a558SYan, Zheng return reloc_root; 14493fd0a558SYan, Zheng } 14503fd0a558SYan, Zheng 14513fd0a558SYan, Zheng /* 14523fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 14533fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 14543fd0a558SYan, Zheng */ 14553fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 14563fd0a558SYan, Zheng struct btrfs_root *root) 14573fd0a558SYan, Zheng { 14583fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14593fd0a558SYan, Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 146020dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14613fd0a558SYan, Zheng int clear_rsv = 0; 1462ffd7b339SJeff Mahoney int ret; 14633fd0a558SYan, Zheng 14643fd0a558SYan, Zheng if (root->reloc_root) { 14653fd0a558SYan, Zheng reloc_root = root->reloc_root; 14663fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14673fd0a558SYan, Zheng return 0; 14683fd0a558SYan, Zheng } 14693fd0a558SYan, Zheng 14703fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14713fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14723fd0a558SYan, Zheng return 0; 14733fd0a558SYan, Zheng 147420dd2cbfSMiao Xie if (!trans->reloc_reserved) { 147520dd2cbfSMiao Xie rsv = trans->block_rsv; 14763fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14773fd0a558SYan, Zheng clear_rsv = 1; 14783fd0a558SYan, Zheng } 14793fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14803fd0a558SYan, Zheng if (clear_rsv) 148120dd2cbfSMiao Xie trans->block_rsv = rsv; 14825d4f98a2SYan Zheng 1483ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1484ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14855d4f98a2SYan Zheng root->reloc_root = reloc_root; 14865d4f98a2SYan Zheng return 0; 14875d4f98a2SYan Zheng } 14885d4f98a2SYan Zheng 14895d4f98a2SYan Zheng /* 14905d4f98a2SYan Zheng * update root item of reloc tree 14915d4f98a2SYan Zheng */ 14925d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 14935d4f98a2SYan Zheng struct btrfs_root *root) 14945d4f98a2SYan Zheng { 14955d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14965d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14975d4f98a2SYan Zheng int ret; 14985d4f98a2SYan Zheng 14995d4f98a2SYan Zheng if (!root->reloc_root) 15007585717fSChris Mason goto out; 15015d4f98a2SYan Zheng 15025d4f98a2SYan Zheng reloc_root = root->reloc_root; 15035d4f98a2SYan Zheng root_item = &reloc_root->root_item; 15045d4f98a2SYan Zheng 15053fd0a558SYan, Zheng if (root->fs_info->reloc_ctl->merge_reloc_tree && 15063fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 15075d4f98a2SYan Zheng root->reloc_root = NULL; 1508c974c464SWang Shilong __del_reloc_root(reloc_root); 15095d4f98a2SYan Zheng } 15105d4f98a2SYan Zheng 15115d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 15125d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 15135d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 15145d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 15155d4f98a2SYan Zheng } 15165d4f98a2SYan Zheng 15175d4f98a2SYan Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 15185d4f98a2SYan Zheng &reloc_root->root_key, root_item); 15195d4f98a2SYan Zheng BUG_ON(ret); 15207585717fSChris Mason 15217585717fSChris Mason out: 15225d4f98a2SYan Zheng return 0; 15235d4f98a2SYan Zheng } 15245d4f98a2SYan Zheng 15255d4f98a2SYan Zheng /* 15265d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15275d4f98a2SYan Zheng * in a subvolume 15285d4f98a2SYan Zheng */ 15295d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15305d4f98a2SYan Zheng { 15315d4f98a2SYan Zheng struct rb_node *node; 15325d4f98a2SYan Zheng struct rb_node *prev; 15335d4f98a2SYan Zheng struct btrfs_inode *entry; 15345d4f98a2SYan Zheng struct inode *inode; 15355d4f98a2SYan Zheng 15365d4f98a2SYan Zheng spin_lock(&root->inode_lock); 15375d4f98a2SYan Zheng again: 15385d4f98a2SYan Zheng node = root->inode_tree.rb_node; 15395d4f98a2SYan Zheng prev = NULL; 15405d4f98a2SYan Zheng while (node) { 15415d4f98a2SYan Zheng prev = node; 15425d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15435d4f98a2SYan Zheng 154433345d01SLi Zefan if (objectid < btrfs_ino(&entry->vfs_inode)) 15455d4f98a2SYan Zheng node = node->rb_left; 154633345d01SLi Zefan else if (objectid > btrfs_ino(&entry->vfs_inode)) 15475d4f98a2SYan Zheng node = node->rb_right; 15485d4f98a2SYan Zheng else 15495d4f98a2SYan Zheng break; 15505d4f98a2SYan Zheng } 15515d4f98a2SYan Zheng if (!node) { 15525d4f98a2SYan Zheng while (prev) { 15535d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 155433345d01SLi Zefan if (objectid <= btrfs_ino(&entry->vfs_inode)) { 15555d4f98a2SYan Zheng node = prev; 15565d4f98a2SYan Zheng break; 15575d4f98a2SYan Zheng } 15585d4f98a2SYan Zheng prev = rb_next(prev); 15595d4f98a2SYan Zheng } 15605d4f98a2SYan Zheng } 15615d4f98a2SYan Zheng while (node) { 15625d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15635d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15645d4f98a2SYan Zheng if (inode) { 15655d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15665d4f98a2SYan Zheng return inode; 15675d4f98a2SYan Zheng } 15685d4f98a2SYan Zheng 156933345d01SLi Zefan objectid = btrfs_ino(&entry->vfs_inode) + 1; 15705d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15715d4f98a2SYan Zheng goto again; 15725d4f98a2SYan Zheng 15735d4f98a2SYan Zheng node = rb_next(node); 15745d4f98a2SYan Zheng } 15755d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15765d4f98a2SYan Zheng return NULL; 15775d4f98a2SYan Zheng } 15785d4f98a2SYan Zheng 15795d4f98a2SYan Zheng static int in_block_group(u64 bytenr, 15805d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group) 15815d4f98a2SYan Zheng { 15825d4f98a2SYan Zheng if (bytenr >= block_group->key.objectid && 15835d4f98a2SYan Zheng bytenr < block_group->key.objectid + block_group->key.offset) 15845d4f98a2SYan Zheng return 1; 15855d4f98a2SYan Zheng return 0; 15865d4f98a2SYan Zheng } 15875d4f98a2SYan Zheng 15885d4f98a2SYan Zheng /* 15895d4f98a2SYan Zheng * get new location of data 15905d4f98a2SYan Zheng */ 15915d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 15925d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 15935d4f98a2SYan Zheng { 15945d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 15955d4f98a2SYan Zheng struct btrfs_path *path; 15965d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15975d4f98a2SYan Zheng struct extent_buffer *leaf; 15985d4f98a2SYan Zheng int ret; 15995d4f98a2SYan Zheng 16005d4f98a2SYan Zheng path = btrfs_alloc_path(); 16015d4f98a2SYan Zheng if (!path) 16025d4f98a2SYan Zheng return -ENOMEM; 16035d4f98a2SYan Zheng 16045d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 160533345d01SLi Zefan ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode), 16065d4f98a2SYan Zheng bytenr, 0); 16075d4f98a2SYan Zheng if (ret < 0) 16085d4f98a2SYan Zheng goto out; 16095d4f98a2SYan Zheng if (ret > 0) { 16105d4f98a2SYan Zheng ret = -ENOENT; 16115d4f98a2SYan Zheng goto out; 16125d4f98a2SYan Zheng } 16135d4f98a2SYan Zheng 16145d4f98a2SYan Zheng leaf = path->nodes[0]; 16155d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 16165d4f98a2SYan Zheng struct btrfs_file_extent_item); 16175d4f98a2SYan Zheng 16185d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 16195d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 16205d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 16215d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 16225d4f98a2SYan Zheng 16235d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 162483d4cfd4SJosef Bacik ret = -EINVAL; 16255d4f98a2SYan Zheng goto out; 16265d4f98a2SYan Zheng } 16275d4f98a2SYan Zheng 16285d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16295d4f98a2SYan Zheng ret = 0; 16305d4f98a2SYan Zheng out: 16315d4f98a2SYan Zheng btrfs_free_path(path); 16325d4f98a2SYan Zheng return ret; 16335d4f98a2SYan Zheng } 16345d4f98a2SYan Zheng 16355d4f98a2SYan Zheng /* 16365d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16375d4f98a2SYan Zheng * the new locations. 16385d4f98a2SYan Zheng */ 16393fd0a558SYan, Zheng static noinline_for_stack 16403fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 16415d4f98a2SYan Zheng struct reloc_control *rc, 16425d4f98a2SYan Zheng struct btrfs_root *root, 16433fd0a558SYan, Zheng struct extent_buffer *leaf) 16445d4f98a2SYan Zheng { 16455d4f98a2SYan Zheng struct btrfs_key key; 16465d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16475d4f98a2SYan Zheng struct inode *inode = NULL; 16485d4f98a2SYan Zheng u64 parent; 16495d4f98a2SYan Zheng u64 bytenr; 16503fd0a558SYan, Zheng u64 new_bytenr = 0; 16515d4f98a2SYan Zheng u64 num_bytes; 16525d4f98a2SYan Zheng u64 end; 16535d4f98a2SYan Zheng u32 nritems; 16545d4f98a2SYan Zheng u32 i; 165583d4cfd4SJosef Bacik int ret = 0; 16565d4f98a2SYan Zheng int first = 1; 16575d4f98a2SYan Zheng int dirty = 0; 16585d4f98a2SYan Zheng 16595d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16605d4f98a2SYan Zheng return 0; 16615d4f98a2SYan Zheng 16625d4f98a2SYan Zheng /* reloc trees always use full backref */ 16635d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16645d4f98a2SYan Zheng parent = leaf->start; 16655d4f98a2SYan Zheng else 16665d4f98a2SYan Zheng parent = 0; 16675d4f98a2SYan Zheng 16685d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16695d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 16705d4f98a2SYan Zheng cond_resched(); 16715d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16725d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16735d4f98a2SYan Zheng continue; 16745d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16755d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16765d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16775d4f98a2SYan Zheng continue; 16785d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16795d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16805d4f98a2SYan Zheng if (bytenr == 0) 16815d4f98a2SYan Zheng continue; 16825d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16835d4f98a2SYan Zheng continue; 16845d4f98a2SYan Zheng 16855d4f98a2SYan Zheng /* 16865d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16875d4f98a2SYan Zheng * to complete and drop the extent cache 16885d4f98a2SYan Zheng */ 16895d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16905d4f98a2SYan Zheng if (first) { 16915d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16925d4f98a2SYan Zheng first = 0; 169333345d01SLi Zefan } else if (inode && btrfs_ino(inode) < key.objectid) { 16943fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16955d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16965d4f98a2SYan Zheng } 169733345d01SLi Zefan if (inode && btrfs_ino(inode) == key.objectid) { 16985d4f98a2SYan Zheng end = key.offset + 16995d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 17005d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 17015d4f98a2SYan Zheng root->sectorsize)); 17025d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 17035d4f98a2SYan Zheng end--; 17045d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1705d0082371SJeff Mahoney key.offset, end); 17065d4f98a2SYan Zheng if (!ret) 17075d4f98a2SYan Zheng continue; 17085d4f98a2SYan Zheng 17095d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, key.offset, end, 17105d4f98a2SYan Zheng 1); 17115d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1712d0082371SJeff Mahoney key.offset, end); 17135d4f98a2SYan Zheng } 17145d4f98a2SYan Zheng } 17155d4f98a2SYan Zheng 17165d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 17175d4f98a2SYan Zheng bytenr, num_bytes); 171883d4cfd4SJosef Bacik if (ret) { 171983d4cfd4SJosef Bacik /* 172083d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 172183d4cfd4SJosef Bacik * in the file extent yet. 172283d4cfd4SJosef Bacik */ 172383d4cfd4SJosef Bacik break; 17243fd0a558SYan, Zheng } 17255d4f98a2SYan Zheng 17265d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17275d4f98a2SYan Zheng dirty = 1; 17285d4f98a2SYan Zheng 17295d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 17305d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, new_bytenr, 17315d4f98a2SYan Zheng num_bytes, parent, 17325d4f98a2SYan Zheng btrfs_header_owner(leaf), 1733b06c4bf5SFilipe Manana key.objectid, key.offset); 173483d4cfd4SJosef Bacik if (ret) { 173566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 173683d4cfd4SJosef Bacik break; 173783d4cfd4SJosef Bacik } 17385d4f98a2SYan Zheng 17395d4f98a2SYan Zheng ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 17405d4f98a2SYan Zheng parent, btrfs_header_owner(leaf), 1741b06c4bf5SFilipe Manana key.objectid, key.offset); 174283d4cfd4SJosef Bacik if (ret) { 174366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 174483d4cfd4SJosef Bacik break; 174583d4cfd4SJosef Bacik } 17465d4f98a2SYan Zheng } 17475d4f98a2SYan Zheng if (dirty) 17485d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17493fd0a558SYan, Zheng if (inode) 17503fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 175183d4cfd4SJosef Bacik return ret; 17525d4f98a2SYan Zheng } 17535d4f98a2SYan Zheng 17545d4f98a2SYan Zheng static noinline_for_stack 17555d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17565d4f98a2SYan Zheng struct btrfs_path *path, int level) 17575d4f98a2SYan Zheng { 17585d4f98a2SYan Zheng struct btrfs_disk_key key1; 17595d4f98a2SYan Zheng struct btrfs_disk_key key2; 17605d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17615d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17625d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17635d4f98a2SYan Zheng } 17645d4f98a2SYan Zheng 17655d4f98a2SYan Zheng /* 17665d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17675d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17685d4f98a2SYan Zheng * reloc tree was create can be replaced. 17695d4f98a2SYan Zheng * 17705d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17715d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17725d4f98a2SYan Zheng * errors, a negative error number is returned. 17735d4f98a2SYan Zheng */ 17743fd0a558SYan, Zheng static noinline_for_stack 17753fd0a558SYan, Zheng int replace_path(struct btrfs_trans_handle *trans, 17765d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17775d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17785d4f98a2SYan Zheng int lowest_level, int max_level) 17795d4f98a2SYan Zheng { 17805d4f98a2SYan Zheng struct extent_buffer *eb; 17815d4f98a2SYan Zheng struct extent_buffer *parent; 17825d4f98a2SYan Zheng struct btrfs_key key; 17835d4f98a2SYan Zheng u64 old_bytenr; 17845d4f98a2SYan Zheng u64 new_bytenr; 17855d4f98a2SYan Zheng u64 old_ptr_gen; 17865d4f98a2SYan Zheng u64 new_ptr_gen; 17875d4f98a2SYan Zheng u64 last_snapshot; 17885d4f98a2SYan Zheng u32 blocksize; 17893fd0a558SYan, Zheng int cow = 0; 17905d4f98a2SYan Zheng int level; 17915d4f98a2SYan Zheng int ret; 17925d4f98a2SYan Zheng int slot; 17935d4f98a2SYan Zheng 17945d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17955d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17965d4f98a2SYan Zheng 17975d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 17983fd0a558SYan, Zheng again: 17995d4f98a2SYan Zheng slot = path->slots[lowest_level]; 18005d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 18015d4f98a2SYan Zheng 18025d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 18035d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 18045d4f98a2SYan Zheng level = btrfs_header_level(eb); 18055d4f98a2SYan Zheng 18065d4f98a2SYan Zheng if (level < lowest_level) { 18075d4f98a2SYan Zheng btrfs_tree_unlock(eb); 18085d4f98a2SYan Zheng free_extent_buffer(eb); 18095d4f98a2SYan Zheng return 0; 18105d4f98a2SYan Zheng } 18115d4f98a2SYan Zheng 18123fd0a558SYan, Zheng if (cow) { 18135d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18145d4f98a2SYan Zheng BUG_ON(ret); 18153fd0a558SYan, Zheng } 18165d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 18175d4f98a2SYan Zheng 18185d4f98a2SYan Zheng if (next_key) { 18195d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18205d4f98a2SYan Zheng next_key->type = (u8)-1; 18215d4f98a2SYan Zheng next_key->offset = (u64)-1; 18225d4f98a2SYan Zheng } 18235d4f98a2SYan Zheng 18245d4f98a2SYan Zheng parent = eb; 18255d4f98a2SYan Zheng while (1) { 18265d4f98a2SYan Zheng level = btrfs_header_level(parent); 18275d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18285d4f98a2SYan Zheng 18295d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 18305d4f98a2SYan Zheng if (ret && slot > 0) 18315d4f98a2SYan Zheng slot--; 18325d4f98a2SYan Zheng 18335d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18345d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18355d4f98a2SYan Zheng 18365d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 1837707e8a07SDavid Sterba blocksize = dest->nodesize; 18385d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 18395d4f98a2SYan Zheng 18405d4f98a2SYan Zheng if (level <= max_level) { 18415d4f98a2SYan Zheng eb = path->nodes[level]; 18425d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18435d4f98a2SYan Zheng path->slots[level]); 18445d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18455d4f98a2SYan Zheng path->slots[level]); 18465d4f98a2SYan Zheng } else { 18475d4f98a2SYan Zheng new_bytenr = 0; 18485d4f98a2SYan Zheng new_ptr_gen = 0; 18495d4f98a2SYan Zheng } 18505d4f98a2SYan Zheng 1851fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18525d4f98a2SYan Zheng ret = level; 18535d4f98a2SYan Zheng break; 18545d4f98a2SYan Zheng } 18555d4f98a2SYan Zheng 18565d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18575d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18583fd0a558SYan, Zheng if (level <= lowest_level) { 18595d4f98a2SYan Zheng ret = 0; 18605d4f98a2SYan Zheng break; 18615d4f98a2SYan Zheng } 18625d4f98a2SYan Zheng 1863ce86cd59SDavid Sterba eb = read_tree_block(dest, old_bytenr, old_ptr_gen); 186464c043deSLiu Bo if (IS_ERR(eb)) { 186564c043deSLiu Bo ret = PTR_ERR(eb); 1866264813acSLiu Bo break; 186764c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 186864c043deSLiu Bo ret = -EIO; 1869416bc658SJosef Bacik free_extent_buffer(eb); 1870379cde74SStefan Behrens break; 1871416bc658SJosef Bacik } 18725d4f98a2SYan Zheng btrfs_tree_lock(eb); 18733fd0a558SYan, Zheng if (cow) { 18745d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18755d4f98a2SYan Zheng slot, &eb); 18765d4f98a2SYan Zheng BUG_ON(ret); 18775d4f98a2SYan Zheng } 18783fd0a558SYan, Zheng btrfs_set_lock_blocking(eb); 18795d4f98a2SYan Zheng 18805d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18815d4f98a2SYan Zheng free_extent_buffer(parent); 18825d4f98a2SYan Zheng 18835d4f98a2SYan Zheng parent = eb; 18845d4f98a2SYan Zheng continue; 18855d4f98a2SYan Zheng } 18865d4f98a2SYan Zheng 18873fd0a558SYan, Zheng if (!cow) { 18883fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18893fd0a558SYan, Zheng free_extent_buffer(parent); 18903fd0a558SYan, Zheng cow = 1; 18913fd0a558SYan, Zheng goto again; 18923fd0a558SYan, Zheng } 18933fd0a558SYan, Zheng 18945d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18955d4f98a2SYan Zheng path->slots[level]); 1896b3b4aa74SDavid Sterba btrfs_release_path(path); 18975d4f98a2SYan Zheng 18985d4f98a2SYan Zheng path->lowest_level = level; 18995d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 19005d4f98a2SYan Zheng path->lowest_level = 0; 19015d4f98a2SYan Zheng BUG_ON(ret); 19025d4f98a2SYan Zheng 19035d4f98a2SYan Zheng /* 19045d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 19055d4f98a2SYan Zheng */ 19065d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 19075d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19085d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19095d4f98a2SYan Zheng 19105d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19115d4f98a2SYan Zheng path->slots[level], old_bytenr); 19125d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19135d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19145d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19155d4f98a2SYan Zheng 19165d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize, 19175d4f98a2SYan Zheng path->nodes[level]->start, 1918b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19195d4f98a2SYan Zheng BUG_ON(ret); 19205d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize, 19215d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 1922b06c4bf5SFilipe Manana 0); 19235d4f98a2SYan Zheng BUG_ON(ret); 19245d4f98a2SYan Zheng 19255d4f98a2SYan Zheng ret = btrfs_free_extent(trans, src, new_bytenr, blocksize, 19265d4f98a2SYan Zheng path->nodes[level]->start, 1927b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19285d4f98a2SYan Zheng BUG_ON(ret); 19295d4f98a2SYan Zheng 19305d4f98a2SYan Zheng ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize, 19315d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 1932b06c4bf5SFilipe Manana 0); 19335d4f98a2SYan Zheng BUG_ON(ret); 19345d4f98a2SYan Zheng 19355d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19365d4f98a2SYan Zheng 19375d4f98a2SYan Zheng ret = level; 19385d4f98a2SYan Zheng break; 19395d4f98a2SYan Zheng } 19405d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19415d4f98a2SYan Zheng free_extent_buffer(parent); 19425d4f98a2SYan Zheng return ret; 19435d4f98a2SYan Zheng } 19445d4f98a2SYan Zheng 19455d4f98a2SYan Zheng /* 19465d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19475d4f98a2SYan Zheng */ 19485d4f98a2SYan Zheng static noinline_for_stack 19495d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19505d4f98a2SYan Zheng int *level) 19515d4f98a2SYan Zheng { 19525d4f98a2SYan Zheng struct extent_buffer *eb; 19535d4f98a2SYan Zheng int i; 19545d4f98a2SYan Zheng u64 last_snapshot; 19555d4f98a2SYan Zheng u32 nritems; 19565d4f98a2SYan Zheng 19575d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19585d4f98a2SYan Zheng 19595d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19605d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19615d4f98a2SYan Zheng path->nodes[i] = NULL; 19625d4f98a2SYan Zheng } 19635d4f98a2SYan Zheng 19645d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 19655d4f98a2SYan Zheng eb = path->nodes[i]; 19665d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19675d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 19685d4f98a2SYan Zheng path->slots[i]++; 19695d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 19705d4f98a2SYan Zheng last_snapshot) 19715d4f98a2SYan Zheng continue; 19725d4f98a2SYan Zheng 19735d4f98a2SYan Zheng *level = i; 19745d4f98a2SYan Zheng return 0; 19755d4f98a2SYan Zheng } 19765d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19775d4f98a2SYan Zheng path->nodes[i] = NULL; 19785d4f98a2SYan Zheng } 19795d4f98a2SYan Zheng return 1; 19805d4f98a2SYan Zheng } 19815d4f98a2SYan Zheng 19825d4f98a2SYan Zheng /* 19835d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 19845d4f98a2SYan Zheng */ 19855d4f98a2SYan Zheng static noinline_for_stack 19865d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19875d4f98a2SYan Zheng int *level) 19885d4f98a2SYan Zheng { 19895d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 19905d4f98a2SYan Zheng int i; 19915d4f98a2SYan Zheng u64 bytenr; 19925d4f98a2SYan Zheng u64 ptr_gen = 0; 19935d4f98a2SYan Zheng u64 last_snapshot; 19945d4f98a2SYan Zheng u32 nritems; 19955d4f98a2SYan Zheng 19965d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19975d4f98a2SYan Zheng 19985d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 19995d4f98a2SYan Zheng eb = path->nodes[i]; 20005d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20015d4f98a2SYan Zheng while (path->slots[i] < nritems) { 20025d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 20035d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 20045d4f98a2SYan Zheng break; 20055d4f98a2SYan Zheng path->slots[i]++; 20065d4f98a2SYan Zheng } 20075d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20085d4f98a2SYan Zheng if (i == *level) 20095d4f98a2SYan Zheng break; 20105d4f98a2SYan Zheng *level = i + 1; 20115d4f98a2SYan Zheng return 0; 20125d4f98a2SYan Zheng } 20135d4f98a2SYan Zheng if (i == 1) { 20145d4f98a2SYan Zheng *level = i; 20155d4f98a2SYan Zheng return 0; 20165d4f98a2SYan Zheng } 20175d4f98a2SYan Zheng 20185d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2019ce86cd59SDavid Sterba eb = read_tree_block(root, bytenr, ptr_gen); 202064c043deSLiu Bo if (IS_ERR(eb)) { 202164c043deSLiu Bo return PTR_ERR(eb); 202264c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2023416bc658SJosef Bacik free_extent_buffer(eb); 2024416bc658SJosef Bacik return -EIO; 2025416bc658SJosef Bacik } 20265d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20275d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20285d4f98a2SYan Zheng path->slots[i - 1] = 0; 20295d4f98a2SYan Zheng } 20305d4f98a2SYan Zheng return 1; 20315d4f98a2SYan Zheng } 20325d4f98a2SYan Zheng 20335d4f98a2SYan Zheng /* 20345d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 20355d4f98a2SYan Zheng * [min_key, max_key) 20365d4f98a2SYan Zheng */ 20375d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 20385d4f98a2SYan Zheng struct btrfs_key *min_key, 20395d4f98a2SYan Zheng struct btrfs_key *max_key) 20405d4f98a2SYan Zheng { 20415d4f98a2SYan Zheng struct inode *inode = NULL; 20425d4f98a2SYan Zheng u64 objectid; 20435d4f98a2SYan Zheng u64 start, end; 204433345d01SLi Zefan u64 ino; 20455d4f98a2SYan Zheng 20465d4f98a2SYan Zheng objectid = min_key->objectid; 20475d4f98a2SYan Zheng while (1) { 20485d4f98a2SYan Zheng cond_resched(); 20495d4f98a2SYan Zheng iput(inode); 20505d4f98a2SYan Zheng 20515d4f98a2SYan Zheng if (objectid > max_key->objectid) 20525d4f98a2SYan Zheng break; 20535d4f98a2SYan Zheng 20545d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20555d4f98a2SYan Zheng if (!inode) 20565d4f98a2SYan Zheng break; 205733345d01SLi Zefan ino = btrfs_ino(inode); 20585d4f98a2SYan Zheng 205933345d01SLi Zefan if (ino > max_key->objectid) { 20605d4f98a2SYan Zheng iput(inode); 20615d4f98a2SYan Zheng break; 20625d4f98a2SYan Zheng } 20635d4f98a2SYan Zheng 206433345d01SLi Zefan objectid = ino + 1; 20655d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 20665d4f98a2SYan Zheng continue; 20675d4f98a2SYan Zheng 206833345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 20695d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 20705d4f98a2SYan Zheng continue; 20715d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 20725d4f98a2SYan Zheng start = 0; 20735d4f98a2SYan Zheng else { 20745d4f98a2SYan Zheng start = min_key->offset; 20755d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(start, root->sectorsize)); 20765d4f98a2SYan Zheng } 20775d4f98a2SYan Zheng } else { 20785d4f98a2SYan Zheng start = 0; 20795d4f98a2SYan Zheng } 20805d4f98a2SYan Zheng 208133345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 20825d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 20835d4f98a2SYan Zheng continue; 20845d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 20855d4f98a2SYan Zheng end = (u64)-1; 20865d4f98a2SYan Zheng } else { 20875d4f98a2SYan Zheng if (max_key->offset == 0) 20885d4f98a2SYan Zheng continue; 20895d4f98a2SYan Zheng end = max_key->offset; 20905d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 20915d4f98a2SYan Zheng end--; 20925d4f98a2SYan Zheng } 20935d4f98a2SYan Zheng } else { 20945d4f98a2SYan Zheng end = (u64)-1; 20955d4f98a2SYan Zheng } 20965d4f98a2SYan Zheng 20975d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2098d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 20995d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 1); 2100d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 21015d4f98a2SYan Zheng } 21025d4f98a2SYan Zheng return 0; 21035d4f98a2SYan Zheng } 21045d4f98a2SYan Zheng 21055d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 21065d4f98a2SYan Zheng struct btrfs_key *key) 21075d4f98a2SYan Zheng 21085d4f98a2SYan Zheng { 21095d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21105d4f98a2SYan Zheng if (!path->nodes[level]) 21115d4f98a2SYan Zheng break; 21125d4f98a2SYan Zheng if (path->slots[level] + 1 < 21135d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21145d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21155d4f98a2SYan Zheng path->slots[level] + 1); 21165d4f98a2SYan Zheng return 0; 21175d4f98a2SYan Zheng } 21185d4f98a2SYan Zheng level++; 21195d4f98a2SYan Zheng } 21205d4f98a2SYan Zheng return 1; 21215d4f98a2SYan Zheng } 21225d4f98a2SYan Zheng 21235d4f98a2SYan Zheng /* 21245d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 21255d4f98a2SYan Zheng * fs tree. 21265d4f98a2SYan Zheng */ 21275d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 21285d4f98a2SYan Zheng struct btrfs_root *root) 21295d4f98a2SYan Zheng { 21305d4f98a2SYan Zheng LIST_HEAD(inode_list); 21315d4f98a2SYan Zheng struct btrfs_key key; 21325d4f98a2SYan Zheng struct btrfs_key next_key; 21339e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 21345d4f98a2SYan Zheng struct btrfs_root *reloc_root; 21355d4f98a2SYan Zheng struct btrfs_root_item *root_item; 21365d4f98a2SYan Zheng struct btrfs_path *path; 21373fd0a558SYan, Zheng struct extent_buffer *leaf; 21385d4f98a2SYan Zheng int level; 21395d4f98a2SYan Zheng int max_level; 21405d4f98a2SYan Zheng int replaced = 0; 21415d4f98a2SYan Zheng int ret; 21425d4f98a2SYan Zheng int err = 0; 21433fd0a558SYan, Zheng u32 min_reserved; 21445d4f98a2SYan Zheng 21455d4f98a2SYan Zheng path = btrfs_alloc_path(); 21465d4f98a2SYan Zheng if (!path) 21475d4f98a2SYan Zheng return -ENOMEM; 2148e4058b54SDavid Sterba path->reada = READA_FORWARD; 21495d4f98a2SYan Zheng 21505d4f98a2SYan Zheng reloc_root = root->reloc_root; 21515d4f98a2SYan Zheng root_item = &reloc_root->root_item; 21525d4f98a2SYan Zheng 21535d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 21545d4f98a2SYan Zheng level = btrfs_root_level(root_item); 21555d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 21565d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 21575d4f98a2SYan Zheng path->slots[level] = 0; 21585d4f98a2SYan Zheng } else { 21595d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 21605d4f98a2SYan Zheng 21615d4f98a2SYan Zheng level = root_item->drop_level; 21625d4f98a2SYan Zheng BUG_ON(level == 0); 21635d4f98a2SYan Zheng path->lowest_level = level; 21645d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 216533c66f43SYan Zheng path->lowest_level = 0; 21665d4f98a2SYan Zheng if (ret < 0) { 21675d4f98a2SYan Zheng btrfs_free_path(path); 21685d4f98a2SYan Zheng return ret; 21695d4f98a2SYan Zheng } 21705d4f98a2SYan Zheng 21715d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 21725d4f98a2SYan Zheng path->slots[level]); 21735d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 21745d4f98a2SYan Zheng 21755d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 21765d4f98a2SYan Zheng } 21775d4f98a2SYan Zheng 21783fd0a558SYan, Zheng min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 21795d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 21805d4f98a2SYan Zheng 21815d4f98a2SYan Zheng while (1) { 218208e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 218308e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 21843fd0a558SYan, Zheng if (ret) { 21859e6a0c52SJosef Bacik err = ret; 21869e6a0c52SJosef Bacik goto out; 21873fd0a558SYan, Zheng } 21889e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 21899e6a0c52SJosef Bacik if (IS_ERR(trans)) { 21909e6a0c52SJosef Bacik err = PTR_ERR(trans); 21919e6a0c52SJosef Bacik trans = NULL; 21929e6a0c52SJosef Bacik goto out; 21939e6a0c52SJosef Bacik } 21949e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 21953fd0a558SYan, Zheng 21963fd0a558SYan, Zheng replaced = 0; 21975d4f98a2SYan Zheng max_level = level; 21985d4f98a2SYan Zheng 21995d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 22005d4f98a2SYan Zheng if (ret < 0) { 22015d4f98a2SYan Zheng err = ret; 22025d4f98a2SYan Zheng goto out; 22035d4f98a2SYan Zheng } 22045d4f98a2SYan Zheng if (ret > 0) 22055d4f98a2SYan Zheng break; 22065d4f98a2SYan Zheng 22075d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 22085d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 22095d4f98a2SYan Zheng ret = 0; 22105d4f98a2SYan Zheng } else { 22113fd0a558SYan, Zheng ret = replace_path(trans, root, reloc_root, path, 22123fd0a558SYan, Zheng &next_key, level, max_level); 22135d4f98a2SYan Zheng } 22145d4f98a2SYan Zheng if (ret < 0) { 22155d4f98a2SYan Zheng err = ret; 22165d4f98a2SYan Zheng goto out; 22175d4f98a2SYan Zheng } 22185d4f98a2SYan Zheng 22195d4f98a2SYan Zheng if (ret > 0) { 22205d4f98a2SYan Zheng level = ret; 22215d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 22225d4f98a2SYan Zheng path->slots[level]); 22235d4f98a2SYan Zheng replaced = 1; 22245d4f98a2SYan Zheng } 22255d4f98a2SYan Zheng 22265d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 22275d4f98a2SYan Zheng if (ret > 0) 22285d4f98a2SYan Zheng break; 22295d4f98a2SYan Zheng 22305d4f98a2SYan Zheng BUG_ON(level == 0); 22315d4f98a2SYan Zheng /* 22325d4f98a2SYan Zheng * save the merging progress in the drop_progress. 22335d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 22345d4f98a2SYan Zheng */ 22355d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 22365d4f98a2SYan Zheng path->slots[level]); 22375d4f98a2SYan Zheng root_item->drop_level = level; 22385d4f98a2SYan Zheng 22393fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 22409e6a0c52SJosef Bacik trans = NULL; 22415d4f98a2SYan Zheng 2242b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 22435d4f98a2SYan Zheng 22445d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22455d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22465d4f98a2SYan Zheng } 22475d4f98a2SYan Zheng 22485d4f98a2SYan Zheng /* 22495d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 22505d4f98a2SYan Zheng * relocated and the block is tree root. 22515d4f98a2SYan Zheng */ 22525d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 22535d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 22545d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 22555d4f98a2SYan Zheng free_extent_buffer(leaf); 22565d4f98a2SYan Zheng if (ret < 0) 22575d4f98a2SYan Zheng err = ret; 22585d4f98a2SYan Zheng out: 22595d4f98a2SYan Zheng btrfs_free_path(path); 22605d4f98a2SYan Zheng 22615d4f98a2SYan Zheng if (err == 0) { 22625d4f98a2SYan Zheng memset(&root_item->drop_progress, 0, 22635d4f98a2SYan Zheng sizeof(root_item->drop_progress)); 22645d4f98a2SYan Zheng root_item->drop_level = 0; 22655d4f98a2SYan Zheng btrfs_set_root_refs(root_item, 0); 22663fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 22675d4f98a2SYan Zheng } 22685d4f98a2SYan Zheng 22699e6a0c52SJosef Bacik if (trans) 22703fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 22715d4f98a2SYan Zheng 2272b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 22735d4f98a2SYan Zheng 22745d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22755d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22765d4f98a2SYan Zheng 22775d4f98a2SYan Zheng return err; 22785d4f98a2SYan Zheng } 22795d4f98a2SYan Zheng 22803fd0a558SYan, Zheng static noinline_for_stack 22813fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 22825d4f98a2SYan Zheng { 22833fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 22843fd0a558SYan, Zheng struct btrfs_root *reloc_root; 22855d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 22863fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 22873fd0a558SYan, Zheng u64 num_bytes = 0; 22883fd0a558SYan, Zheng int ret; 22893fd0a558SYan, Zheng 22907585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 22913fd0a558SYan, Zheng rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22923fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 22937585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 22947585717fSChris Mason 22953fd0a558SYan, Zheng again: 22963fd0a558SYan, Zheng if (!err) { 22973fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 229808e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 229908e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23003fd0a558SYan, Zheng if (ret) 23013fd0a558SYan, Zheng err = ret; 23023fd0a558SYan, Zheng } 23033fd0a558SYan, Zheng 23047a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 23053612b495STsutomu Itoh if (IS_ERR(trans)) { 23063612b495STsutomu Itoh if (!err) 23073612b495STsutomu Itoh btrfs_block_rsv_release(rc->extent_root, 23083612b495STsutomu Itoh rc->block_rsv, num_bytes); 23093612b495STsutomu Itoh return PTR_ERR(trans); 23103612b495STsutomu Itoh } 23113fd0a558SYan, Zheng 23123fd0a558SYan, Zheng if (!err) { 23133fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 23143fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 23153fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, 23163fd0a558SYan, Zheng rc->block_rsv, num_bytes); 23173fd0a558SYan, Zheng goto again; 23183fd0a558SYan, Zheng } 23193fd0a558SYan, Zheng } 23203fd0a558SYan, Zheng 23213fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 23223fd0a558SYan, Zheng 23233fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 23243fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 23253fd0a558SYan, Zheng struct btrfs_root, root_list); 23263fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 23273fd0a558SYan, Zheng 23283fd0a558SYan, Zheng root = read_fs_root(reloc_root->fs_info, 23293fd0a558SYan, Zheng reloc_root->root_key.offset); 23303fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 23313fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 23323fd0a558SYan, Zheng 23333fd0a558SYan, Zheng /* 23343fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 23353fd0a558SYan, Zheng * knows it should resumes merging 23363fd0a558SYan, Zheng */ 23373fd0a558SYan, Zheng if (!err) 23383fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 23393fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 23403fd0a558SYan, Zheng 23413fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 23423fd0a558SYan, Zheng } 23433fd0a558SYan, Zheng 23443fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 23453fd0a558SYan, Zheng 23463fd0a558SYan, Zheng if (!err) 23473fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 23483fd0a558SYan, Zheng else 23493fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 23503fd0a558SYan, Zheng return err; 23513fd0a558SYan, Zheng } 23523fd0a558SYan, Zheng 23533fd0a558SYan, Zheng static noinline_for_stack 2354aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2355aca1bba6SLiu Bo { 2356aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2357aca1bba6SLiu Bo 2358aca1bba6SLiu Bo while (!list_empty(list)) { 2359aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2360aca1bba6SLiu Bo root_list); 23616bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 23626bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 23636bdf131fSJosef Bacik reloc_root->node = NULL; 23646bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2365c974c464SWang Shilong __del_reloc_root(reloc_root); 2366aca1bba6SLiu Bo } 2367aca1bba6SLiu Bo } 2368aca1bba6SLiu Bo 2369aca1bba6SLiu Bo static noinline_for_stack 237094404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 23713fd0a558SYan, Zheng { 23725d4f98a2SYan Zheng struct btrfs_root *root; 23735d4f98a2SYan Zheng struct btrfs_root *reloc_root; 23745bc7247aSMiao Xie u64 last_snap; 23755bc7247aSMiao Xie u64 otransid; 23765bc7247aSMiao Xie u64 objectid; 23773fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23783fd0a558SYan, Zheng int found = 0; 2379aca1bba6SLiu Bo int ret = 0; 23803fd0a558SYan, Zheng again: 23813fd0a558SYan, Zheng root = rc->extent_root; 23827585717fSChris Mason 23837585717fSChris Mason /* 23847585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 23857585717fSChris Mason * we have to make sure nobody is in the middle of 23867585717fSChris Mason * adding their roots to the list while we are 23877585717fSChris Mason * doing this splice 23887585717fSChris Mason */ 23897585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 23903fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 23917585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 23925d4f98a2SYan Zheng 23933fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 23943fd0a558SYan, Zheng found = 1; 23953fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 23963fd0a558SYan, Zheng struct btrfs_root, root_list); 23975d4f98a2SYan Zheng 23985d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 23995d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, 24005d4f98a2SYan Zheng reloc_root->root_key.offset); 24015d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24025d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24035d4f98a2SYan Zheng 24043fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2405b37b39cdSJosef Bacik if (ret) { 240625e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 240725e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 240825e293c2SWang Shilong &reloc_roots); 2409aca1bba6SLiu Bo goto out; 2410b37b39cdSJosef Bacik } 24113fd0a558SYan, Zheng } else { 24123fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24133fd0a558SYan, Zheng } 24145bc7247aSMiao Xie 24155bc7247aSMiao Xie /* 241601327610SNicholas D Steeves * we keep the old last snapshot transid in rtranid when we 24175bc7247aSMiao Xie * created the relocation tree. 24185bc7247aSMiao Xie */ 24195bc7247aSMiao Xie last_snap = btrfs_root_rtransid(&reloc_root->root_item); 24205bc7247aSMiao Xie otransid = btrfs_root_otransid(&reloc_root->root_item); 24215bc7247aSMiao Xie objectid = reloc_root->root_key.offset; 24225bc7247aSMiao Xie 24232c536799SJeff Mahoney ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1); 2424aca1bba6SLiu Bo if (ret < 0) { 2425aca1bba6SLiu Bo if (list_empty(&reloc_root->root_list)) 2426aca1bba6SLiu Bo list_add_tail(&reloc_root->root_list, 2427aca1bba6SLiu Bo &reloc_roots); 2428aca1bba6SLiu Bo goto out; 2429aca1bba6SLiu Bo } 24305d4f98a2SYan Zheng } 24315d4f98a2SYan Zheng 24323fd0a558SYan, Zheng if (found) { 24333fd0a558SYan, Zheng found = 0; 24343fd0a558SYan, Zheng goto again; 24355d4f98a2SYan Zheng } 2436aca1bba6SLiu Bo out: 2437aca1bba6SLiu Bo if (ret) { 243834d97007SAnand Jain btrfs_handle_fs_error(root->fs_info, ret, NULL); 2439aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2440aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2441467bb1d2SWang Shilong 2442467bb1d2SWang Shilong /* new reloc root may be added */ 2443467bb1d2SWang Shilong mutex_lock(&root->fs_info->reloc_mutex); 2444467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 2445467bb1d2SWang Shilong mutex_unlock(&root->fs_info->reloc_mutex); 2446467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2447467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2448aca1bba6SLiu Bo } 2449aca1bba6SLiu Bo 24505d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 24515d4f98a2SYan Zheng } 24525d4f98a2SYan Zheng 24535d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 24545d4f98a2SYan Zheng { 24555d4f98a2SYan Zheng struct tree_block *block; 24565d4f98a2SYan Zheng struct rb_node *rb_node; 24575d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 24585d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 24595d4f98a2SYan Zheng rb_erase(rb_node, blocks); 24605d4f98a2SYan Zheng kfree(block); 24615d4f98a2SYan Zheng } 24625d4f98a2SYan Zheng } 24635d4f98a2SYan Zheng 24645d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 24655d4f98a2SYan Zheng struct btrfs_root *reloc_root) 24665d4f98a2SYan Zheng { 24675d4f98a2SYan Zheng struct btrfs_root *root; 24685d4f98a2SYan Zheng 24695d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 24705d4f98a2SYan Zheng return 0; 24715d4f98a2SYan Zheng 24725d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset); 24735d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24745d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24755d4f98a2SYan Zheng 24765d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 24775d4f98a2SYan Zheng } 24785d4f98a2SYan Zheng 24793fd0a558SYan, Zheng static noinline_for_stack 24803fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 24813fd0a558SYan, Zheng struct reloc_control *rc, 24825d4f98a2SYan Zheng struct backref_node *node, 2483dc4103f9SWang Shilong struct backref_edge *edges[]) 24845d4f98a2SYan Zheng { 24855d4f98a2SYan Zheng struct backref_node *next; 24865d4f98a2SYan Zheng struct btrfs_root *root; 24873fd0a558SYan, Zheng int index = 0; 24883fd0a558SYan, Zheng 24895d4f98a2SYan Zheng next = node; 24905d4f98a2SYan Zheng while (1) { 24915d4f98a2SYan Zheng cond_resched(); 24925d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 24935d4f98a2SYan Zheng root = next->root; 24943fd0a558SYan, Zheng BUG_ON(!root); 249527cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 24965d4f98a2SYan Zheng 24975d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 24985d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 24995d4f98a2SYan Zheng break; 25005d4f98a2SYan Zheng } 25015d4f98a2SYan Zheng 25025d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 25033fd0a558SYan, Zheng root = root->reloc_root; 25043fd0a558SYan, Zheng 25053fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 25063fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 25073fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 25083fd0a558SYan, Zheng next->new_bytenr = root->node->start; 25093fd0a558SYan, Zheng next->root = root; 25103fd0a558SYan, Zheng list_add_tail(&next->list, 25113fd0a558SYan, Zheng &rc->backref_cache.changed); 25123fd0a558SYan, Zheng __mark_block_processed(rc, next); 25135d4f98a2SYan Zheng break; 25145d4f98a2SYan Zheng } 25155d4f98a2SYan Zheng 25163fd0a558SYan, Zheng WARN_ON(1); 25175d4f98a2SYan Zheng root = NULL; 25185d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25195d4f98a2SYan Zheng if (!next || next->level <= node->level) 25205d4f98a2SYan Zheng break; 25215d4f98a2SYan Zheng } 25223fd0a558SYan, Zheng if (!root) 25233fd0a558SYan, Zheng return NULL; 25245d4f98a2SYan Zheng 25253fd0a558SYan, Zheng next = node; 25263fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 25273fd0a558SYan, Zheng while (1) { 25283fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 25293fd0a558SYan, Zheng if (--index < 0) 25303fd0a558SYan, Zheng break; 25313fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 25323fd0a558SYan, Zheng } 25335d4f98a2SYan Zheng return root; 25345d4f98a2SYan Zheng } 25355d4f98a2SYan Zheng 25363fd0a558SYan, Zheng /* 25373fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 25383fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 25393fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 25403fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 25413fd0a558SYan, Zheng */ 25425d4f98a2SYan Zheng static noinline_for_stack 2543147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 25445d4f98a2SYan Zheng { 25453fd0a558SYan, Zheng struct backref_node *next; 25463fd0a558SYan, Zheng struct btrfs_root *root; 25473fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 25485d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25493fd0a558SYan, Zheng int index = 0; 25503fd0a558SYan, Zheng 25513fd0a558SYan, Zheng next = node; 25523fd0a558SYan, Zheng while (1) { 25533fd0a558SYan, Zheng cond_resched(); 25543fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 25553fd0a558SYan, Zheng root = next->root; 25563fd0a558SYan, Zheng BUG_ON(!root); 25573fd0a558SYan, Zheng 255825985edcSLucas De Marchi /* no other choice for non-references counted tree */ 255927cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 25603fd0a558SYan, Zheng return root; 25613fd0a558SYan, Zheng 25623fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 25633fd0a558SYan, Zheng fs_root = root; 25643fd0a558SYan, Zheng 25653fd0a558SYan, Zheng if (next != node) 25663fd0a558SYan, Zheng return NULL; 25673fd0a558SYan, Zheng 25683fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 25693fd0a558SYan, Zheng if (!next || next->level <= node->level) 25703fd0a558SYan, Zheng break; 25713fd0a558SYan, Zheng } 25723fd0a558SYan, Zheng 25733fd0a558SYan, Zheng if (!fs_root) 25743fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 25753fd0a558SYan, Zheng return fs_root; 25765d4f98a2SYan Zheng } 25775d4f98a2SYan Zheng 25785d4f98a2SYan Zheng static noinline_for_stack 25793fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 25803fd0a558SYan, Zheng struct backref_node *node, int reserve) 25815d4f98a2SYan Zheng { 25823fd0a558SYan, Zheng struct backref_node *next = node; 25833fd0a558SYan, Zheng struct backref_edge *edge; 25843fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25853fd0a558SYan, Zheng u64 num_bytes = 0; 25863fd0a558SYan, Zheng int index = 0; 25875d4f98a2SYan Zheng 25883fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 25893fd0a558SYan, Zheng 25903fd0a558SYan, Zheng while (next) { 25913fd0a558SYan, Zheng cond_resched(); 25925d4f98a2SYan Zheng while (1) { 25933fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 25945d4f98a2SYan Zheng break; 25955d4f98a2SYan Zheng 2596707e8a07SDavid Sterba num_bytes += rc->extent_root->nodesize; 25973fd0a558SYan, Zheng 25983fd0a558SYan, Zheng if (list_empty(&next->upper)) 25993fd0a558SYan, Zheng break; 26003fd0a558SYan, Zheng 26013fd0a558SYan, Zheng edge = list_entry(next->upper.next, 26023fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 26033fd0a558SYan, Zheng edges[index++] = edge; 26043fd0a558SYan, Zheng next = edge->node[UPPER]; 26055d4f98a2SYan Zheng } 26063fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26073fd0a558SYan, Zheng } 26083fd0a558SYan, Zheng return num_bytes; 26093fd0a558SYan, Zheng } 26103fd0a558SYan, Zheng 26113fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 26123fd0a558SYan, Zheng struct reloc_control *rc, 26133fd0a558SYan, Zheng struct backref_node *node) 26143fd0a558SYan, Zheng { 26153fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 26163fd0a558SYan, Zheng u64 num_bytes; 26173fd0a558SYan, Zheng int ret; 26180647bf56SWang Shilong u64 tmp; 26193fd0a558SYan, Zheng 26203fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 26213fd0a558SYan, Zheng 26223fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 26230647bf56SWang Shilong rc->reserved_bytes += num_bytes; 26248ca17f0fSJosef Bacik 26258ca17f0fSJosef Bacik /* 26268ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 26278ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 26288ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 26298ca17f0fSJosef Bacik */ 26300647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 26318ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 26323fd0a558SYan, Zheng if (ret) { 26338ca17f0fSJosef Bacik tmp = rc->extent_root->nodesize * RELOCATION_RESERVED_NODES; 26340647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 26350647bf56SWang Shilong tmp <<= 1; 26360647bf56SWang Shilong /* 26370647bf56SWang Shilong * only one thread can access block_rsv at this point, 26380647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 26390647bf56SWang Shilong * we expand more reservation size here to allow enough 26408ca17f0fSJosef Bacik * space for relocation and we will return eailer in 26410647bf56SWang Shilong * enospc case. 26420647bf56SWang Shilong */ 26430647bf56SWang Shilong rc->block_rsv->size = tmp + rc->extent_root->nodesize * 26440647bf56SWang Shilong RELOCATION_RESERVED_NODES; 26458ca17f0fSJosef Bacik return -EAGAIN; 26463fd0a558SYan, Zheng } 26473fd0a558SYan, Zheng 26483fd0a558SYan, Zheng return 0; 26493fd0a558SYan, Zheng } 26503fd0a558SYan, Zheng 26515d4f98a2SYan Zheng /* 26525d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 26535d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 26545d4f98a2SYan Zheng * 26555d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 26565d4f98a2SYan Zheng * in that case this function just updates pointers. 26575d4f98a2SYan Zheng */ 26585d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 26593fd0a558SYan, Zheng struct reloc_control *rc, 26605d4f98a2SYan Zheng struct backref_node *node, 26615d4f98a2SYan Zheng struct btrfs_key *key, 26625d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 26635d4f98a2SYan Zheng { 26645d4f98a2SYan Zheng struct backref_node *upper; 26655d4f98a2SYan Zheng struct backref_edge *edge; 26665d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26675d4f98a2SYan Zheng struct btrfs_root *root; 26685d4f98a2SYan Zheng struct extent_buffer *eb; 26695d4f98a2SYan Zheng u32 blocksize; 26705d4f98a2SYan Zheng u64 bytenr; 26715d4f98a2SYan Zheng u64 generation; 26725d4f98a2SYan Zheng int slot; 26735d4f98a2SYan Zheng int ret; 26745d4f98a2SYan Zheng int err = 0; 26755d4f98a2SYan Zheng 26765d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 26775d4f98a2SYan Zheng 26785d4f98a2SYan Zheng path->lowest_level = node->level + 1; 26793fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 26805d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 26815d4f98a2SYan Zheng cond_resched(); 26825d4f98a2SYan Zheng 26835d4f98a2SYan Zheng upper = edge->node[UPPER]; 2684dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 26853fd0a558SYan, Zheng BUG_ON(!root); 26865d4f98a2SYan Zheng 26873fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 26883fd0a558SYan, Zheng if (!lowest) { 26893fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 26903fd0a558SYan, Zheng upper->level, &slot); 26913fd0a558SYan, Zheng BUG_ON(ret); 26923fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 26933fd0a558SYan, Zheng if (node->eb->start == bytenr) 26943fd0a558SYan, Zheng goto next; 26953fd0a558SYan, Zheng } 26965d4f98a2SYan Zheng drop_node_buffer(upper); 26973fd0a558SYan, Zheng } 26985d4f98a2SYan Zheng 26995d4f98a2SYan Zheng if (!upper->eb) { 27005d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 27013561b9dbSLiu Bo if (ret) { 27023561b9dbSLiu Bo if (ret < 0) 27035d4f98a2SYan Zheng err = ret; 27043561b9dbSLiu Bo else 27053561b9dbSLiu Bo err = -ENOENT; 27063561b9dbSLiu Bo 27073561b9dbSLiu Bo btrfs_release_path(path); 27085d4f98a2SYan Zheng break; 27095d4f98a2SYan Zheng } 27105d4f98a2SYan Zheng 27113fd0a558SYan, Zheng if (!upper->eb) { 27123fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 27133fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 27143fd0a558SYan, Zheng } else { 27153fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 27163fd0a558SYan, Zheng } 27173fd0a558SYan, Zheng 27183fd0a558SYan, Zheng upper->locked = 1; 27193fd0a558SYan, Zheng path->locks[upper->level] = 0; 27203fd0a558SYan, Zheng 27215d4f98a2SYan Zheng slot = path->slots[upper->level]; 2722b3b4aa74SDavid Sterba btrfs_release_path(path); 27235d4f98a2SYan Zheng } else { 27245d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 27255d4f98a2SYan Zheng &slot); 27265d4f98a2SYan Zheng BUG_ON(ret); 27275d4f98a2SYan Zheng } 27285d4f98a2SYan Zheng 27295d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27303fd0a558SYan, Zheng if (lowest) { 27314547f4d8SLiu Bo if (bytenr != node->bytenr) { 27324547f4d8SLiu Bo btrfs_err(root->fs_info, 27334547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 27344547f4d8SLiu Bo bytenr, node->bytenr, slot, 27354547f4d8SLiu Bo upper->eb->start); 27364547f4d8SLiu Bo err = -EIO; 27374547f4d8SLiu Bo goto next; 27384547f4d8SLiu Bo } 27395d4f98a2SYan Zheng } else { 27403fd0a558SYan, Zheng if (node->eb->start == bytenr) 27413fd0a558SYan, Zheng goto next; 27425d4f98a2SYan Zheng } 27435d4f98a2SYan Zheng 2744707e8a07SDavid Sterba blocksize = root->nodesize; 27455d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2746ce86cd59SDavid Sterba eb = read_tree_block(root, bytenr, generation); 274764c043deSLiu Bo if (IS_ERR(eb)) { 274864c043deSLiu Bo err = PTR_ERR(eb); 274964c043deSLiu Bo goto next; 275064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2751416bc658SJosef Bacik free_extent_buffer(eb); 275297d9a8a4STsutomu Itoh err = -EIO; 275397d9a8a4STsutomu Itoh goto next; 275497d9a8a4STsutomu Itoh } 27555d4f98a2SYan Zheng btrfs_tree_lock(eb); 27565d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 27575d4f98a2SYan Zheng 27585d4f98a2SYan Zheng if (!node->eb) { 27595d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 27605d4f98a2SYan Zheng slot, &eb); 27613fd0a558SYan, Zheng btrfs_tree_unlock(eb); 27623fd0a558SYan, Zheng free_extent_buffer(eb); 27635d4f98a2SYan Zheng if (ret < 0) { 27645d4f98a2SYan Zheng err = ret; 27653fd0a558SYan, Zheng goto next; 27665d4f98a2SYan Zheng } 27673fd0a558SYan, Zheng BUG_ON(node->eb != eb); 27685d4f98a2SYan Zheng } else { 27695d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 27705d4f98a2SYan Zheng node->eb->start); 27715d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 27725d4f98a2SYan Zheng trans->transid); 27735d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 27745d4f98a2SYan Zheng 27755d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, 27765d4f98a2SYan Zheng node->eb->start, blocksize, 27775d4f98a2SYan Zheng upper->eb->start, 27785d4f98a2SYan Zheng btrfs_header_owner(upper->eb), 2779b06c4bf5SFilipe Manana node->level, 0); 27805d4f98a2SYan Zheng BUG_ON(ret); 27815d4f98a2SYan Zheng 27825d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 27835d4f98a2SYan Zheng BUG_ON(ret); 27845d4f98a2SYan Zheng } 27853fd0a558SYan, Zheng next: 27863fd0a558SYan, Zheng if (!upper->pending) 27873fd0a558SYan, Zheng drop_node_buffer(upper); 27883fd0a558SYan, Zheng else 27893fd0a558SYan, Zheng unlock_node_buffer(upper); 27903fd0a558SYan, Zheng if (err) 27913fd0a558SYan, Zheng break; 27925d4f98a2SYan Zheng } 27933fd0a558SYan, Zheng 27943fd0a558SYan, Zheng if (!err && node->pending) { 27953fd0a558SYan, Zheng drop_node_buffer(node); 27963fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 27973fd0a558SYan, Zheng node->pending = 0; 27985d4f98a2SYan Zheng } 27993fd0a558SYan, Zheng 28005d4f98a2SYan Zheng path->lowest_level = 0; 28013fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 28025d4f98a2SYan Zheng return err; 28035d4f98a2SYan Zheng } 28045d4f98a2SYan Zheng 28055d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 28063fd0a558SYan, Zheng struct reloc_control *rc, 28075d4f98a2SYan Zheng struct backref_node *node, 28085d4f98a2SYan Zheng struct btrfs_path *path) 28095d4f98a2SYan Zheng { 28105d4f98a2SYan Zheng struct btrfs_key key; 28115d4f98a2SYan Zheng 28125d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 28133fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 28145d4f98a2SYan Zheng } 28155d4f98a2SYan Zheng 28165d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 28173fd0a558SYan, Zheng struct reloc_control *rc, 28183fd0a558SYan, Zheng struct btrfs_path *path, int err) 28195d4f98a2SYan Zheng { 28203fd0a558SYan, Zheng LIST_HEAD(list); 28213fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 28225d4f98a2SYan Zheng struct backref_node *node; 28235d4f98a2SYan Zheng int level; 28245d4f98a2SYan Zheng int ret; 28255d4f98a2SYan Zheng 28265d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 28275d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 28285d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 28293fd0a558SYan, Zheng struct backref_node, list); 28303fd0a558SYan, Zheng list_move_tail(&node->list, &list); 28313fd0a558SYan, Zheng BUG_ON(!node->pending); 28325d4f98a2SYan Zheng 28333fd0a558SYan, Zheng if (!err) { 28343fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 28355d4f98a2SYan Zheng if (ret < 0) 28365d4f98a2SYan Zheng err = ret; 28375d4f98a2SYan Zheng } 28385d4f98a2SYan Zheng } 28393fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 28403fd0a558SYan, Zheng } 28415d4f98a2SYan Zheng return err; 28425d4f98a2SYan Zheng } 28435d4f98a2SYan Zheng 28445d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 28453fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 28463fd0a558SYan, Zheng { 28473fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2848ceeb0ae7SDavid Sterba EXTENT_DIRTY); 28493fd0a558SYan, Zheng } 28503fd0a558SYan, Zheng 28513fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 28525d4f98a2SYan Zheng struct backref_node *node) 28535d4f98a2SYan Zheng { 28545d4f98a2SYan Zheng u32 blocksize; 28555d4f98a2SYan Zheng if (node->level == 0 || 28565d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2857707e8a07SDavid Sterba blocksize = rc->extent_root->nodesize; 28583fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 28595d4f98a2SYan Zheng } 28605d4f98a2SYan Zheng node->processed = 1; 28615d4f98a2SYan Zheng } 28625d4f98a2SYan Zheng 28635d4f98a2SYan Zheng /* 28645d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 28655d4f98a2SYan Zheng * as processed. 28665d4f98a2SYan Zheng */ 28675d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 28685d4f98a2SYan Zheng struct backref_node *node) 28695d4f98a2SYan Zheng { 28705d4f98a2SYan Zheng struct backref_node *next = node; 28715d4f98a2SYan Zheng struct backref_edge *edge; 28725d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 28735d4f98a2SYan Zheng int index = 0; 28745d4f98a2SYan Zheng 28755d4f98a2SYan Zheng while (next) { 28765d4f98a2SYan Zheng cond_resched(); 28775d4f98a2SYan Zheng while (1) { 28785d4f98a2SYan Zheng if (next->processed) 28795d4f98a2SYan Zheng break; 28805d4f98a2SYan Zheng 28813fd0a558SYan, Zheng __mark_block_processed(rc, next); 28825d4f98a2SYan Zheng 28835d4f98a2SYan Zheng if (list_empty(&next->upper)) 28845d4f98a2SYan Zheng break; 28855d4f98a2SYan Zheng 28865d4f98a2SYan Zheng edge = list_entry(next->upper.next, 28875d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 28885d4f98a2SYan Zheng edges[index++] = edge; 28895d4f98a2SYan Zheng next = edge->node[UPPER]; 28905d4f98a2SYan Zheng } 28915d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 28925d4f98a2SYan Zheng } 28935d4f98a2SYan Zheng } 28945d4f98a2SYan Zheng 28957476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 28965d4f98a2SYan Zheng { 28977476dfdaSDavid Sterba u32 blocksize = rc->extent_root->nodesize; 28987476dfdaSDavid Sterba 28995d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 29009655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 29015d4f98a2SYan Zheng return 1; 29025d4f98a2SYan Zheng return 0; 29035d4f98a2SYan Zheng } 29045d4f98a2SYan Zheng 29055d4f98a2SYan Zheng static int get_tree_block_key(struct reloc_control *rc, 29065d4f98a2SYan Zheng struct tree_block *block) 29075d4f98a2SYan Zheng { 29085d4f98a2SYan Zheng struct extent_buffer *eb; 29095d4f98a2SYan Zheng 29105d4f98a2SYan Zheng BUG_ON(block->key_ready); 29115d4f98a2SYan Zheng eb = read_tree_block(rc->extent_root, block->bytenr, 2912ce86cd59SDavid Sterba block->key.offset); 291364c043deSLiu Bo if (IS_ERR(eb)) { 291464c043deSLiu Bo return PTR_ERR(eb); 291564c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2916416bc658SJosef Bacik free_extent_buffer(eb); 2917416bc658SJosef Bacik return -EIO; 2918416bc658SJosef Bacik } 29195d4f98a2SYan Zheng WARN_ON(btrfs_header_level(eb) != block->level); 29205d4f98a2SYan Zheng if (block->level == 0) 29215d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 29225d4f98a2SYan Zheng else 29235d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 29245d4f98a2SYan Zheng free_extent_buffer(eb); 29255d4f98a2SYan Zheng block->key_ready = 1; 29265d4f98a2SYan Zheng return 0; 29275d4f98a2SYan Zheng } 29285d4f98a2SYan Zheng 29295d4f98a2SYan Zheng /* 29305d4f98a2SYan Zheng * helper function to relocate a tree block 29315d4f98a2SYan Zheng */ 29325d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 29335d4f98a2SYan Zheng struct reloc_control *rc, 29345d4f98a2SYan Zheng struct backref_node *node, 29355d4f98a2SYan Zheng struct btrfs_key *key, 29365d4f98a2SYan Zheng struct btrfs_path *path) 29375d4f98a2SYan Zheng { 29385d4f98a2SYan Zheng struct btrfs_root *root; 29393fd0a558SYan, Zheng int ret = 0; 29405d4f98a2SYan Zheng 29413fd0a558SYan, Zheng if (!node) 29425d4f98a2SYan Zheng return 0; 29433fd0a558SYan, Zheng 29443fd0a558SYan, Zheng BUG_ON(node->processed); 2945147d256eSZhaolei root = select_one_root(node); 29463fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 29473fd0a558SYan, Zheng update_processed_blocks(rc, node); 29483fd0a558SYan, Zheng goto out; 29495d4f98a2SYan Zheng } 29505d4f98a2SYan Zheng 295127cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29523fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 29533fd0a558SYan, Zheng if (ret) 29545d4f98a2SYan Zheng goto out; 29555d4f98a2SYan Zheng } 29563fd0a558SYan, Zheng 29573fd0a558SYan, Zheng if (root) { 295827cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29593fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 29603fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 29613fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 29623fd0a558SYan, Zheng root = root->reloc_root; 29633fd0a558SYan, Zheng node->new_bytenr = root->node->start; 29643fd0a558SYan, Zheng node->root = root; 29653fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 29663fd0a558SYan, Zheng } else { 29675d4f98a2SYan Zheng path->lowest_level = node->level; 29685d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2969b3b4aa74SDavid Sterba btrfs_release_path(path); 29703fd0a558SYan, Zheng if (ret > 0) 29715d4f98a2SYan Zheng ret = 0; 29723fd0a558SYan, Zheng } 29733fd0a558SYan, Zheng if (!ret) 29743fd0a558SYan, Zheng update_processed_blocks(rc, node); 29753fd0a558SYan, Zheng } else { 29763fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 29773fd0a558SYan, Zheng } 29785d4f98a2SYan Zheng out: 29790647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 29803fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 29815d4f98a2SYan Zheng return ret; 29825d4f98a2SYan Zheng } 29835d4f98a2SYan Zheng 29845d4f98a2SYan Zheng /* 29855d4f98a2SYan Zheng * relocate a list of blocks 29865d4f98a2SYan Zheng */ 29875d4f98a2SYan Zheng static noinline_for_stack 29885d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 29895d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 29905d4f98a2SYan Zheng { 29915d4f98a2SYan Zheng struct backref_node *node; 29925d4f98a2SYan Zheng struct btrfs_path *path; 29935d4f98a2SYan Zheng struct tree_block *block; 29945d4f98a2SYan Zheng struct rb_node *rb_node; 29955d4f98a2SYan Zheng int ret; 29965d4f98a2SYan Zheng int err = 0; 29975d4f98a2SYan Zheng 29985d4f98a2SYan Zheng path = btrfs_alloc_path(); 2999e1a12670SLiu Bo if (!path) { 3000e1a12670SLiu Bo err = -ENOMEM; 300134c2b290SDavid Sterba goto out_free_blocks; 3002e1a12670SLiu Bo } 30035d4f98a2SYan Zheng 30045d4f98a2SYan Zheng rb_node = rb_first(blocks); 30055d4f98a2SYan Zheng while (rb_node) { 30065d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 30075d4f98a2SYan Zheng if (!block->key_ready) 3008d3e46feaSDavid Sterba readahead_tree_block(rc->extent_root, block->bytenr); 30095d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30105d4f98a2SYan Zheng } 30115d4f98a2SYan Zheng 30125d4f98a2SYan Zheng rb_node = rb_first(blocks); 30135d4f98a2SYan Zheng while (rb_node) { 30145d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 301534c2b290SDavid Sterba if (!block->key_ready) { 301634c2b290SDavid Sterba err = get_tree_block_key(rc, block); 301734c2b290SDavid Sterba if (err) 301834c2b290SDavid Sterba goto out_free_path; 301934c2b290SDavid Sterba } 30205d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30215d4f98a2SYan Zheng } 30225d4f98a2SYan Zheng 30235d4f98a2SYan Zheng rb_node = rb_first(blocks); 30245d4f98a2SYan Zheng while (rb_node) { 30255d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 30265d4f98a2SYan Zheng 30273fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 30285d4f98a2SYan Zheng block->level, block->bytenr); 30295d4f98a2SYan Zheng if (IS_ERR(node)) { 30305d4f98a2SYan Zheng err = PTR_ERR(node); 30315d4f98a2SYan Zheng goto out; 30325d4f98a2SYan Zheng } 30335d4f98a2SYan Zheng 30345d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 30355d4f98a2SYan Zheng path); 30365d4f98a2SYan Zheng if (ret < 0) { 30373fd0a558SYan, Zheng if (ret != -EAGAIN || rb_node == rb_first(blocks)) 30385d4f98a2SYan Zheng err = ret; 30395d4f98a2SYan Zheng goto out; 30405d4f98a2SYan Zheng } 30415d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30425d4f98a2SYan Zheng } 30435d4f98a2SYan Zheng out: 30443fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 30455d4f98a2SYan Zheng 304634c2b290SDavid Sterba out_free_path: 30475d4f98a2SYan Zheng btrfs_free_path(path); 304834c2b290SDavid Sterba out_free_blocks: 3049e1a12670SLiu Bo free_block_list(blocks); 30505d4f98a2SYan Zheng return err; 30515d4f98a2SYan Zheng } 30525d4f98a2SYan Zheng 30535d4f98a2SYan Zheng static noinline_for_stack 3054efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3055efa56464SYan, Zheng struct file_extent_cluster *cluster) 3056efa56464SYan, Zheng { 3057efa56464SYan, Zheng u64 alloc_hint = 0; 3058efa56464SYan, Zheng u64 start; 3059efa56464SYan, Zheng u64 end; 3060efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3061efa56464SYan, Zheng u64 num_bytes; 3062efa56464SYan, Zheng int nr = 0; 3063efa56464SYan, Zheng int ret = 0; 3064dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3065dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 306618513091SWang Xiaoguang u64 cur_offset; 3067efa56464SYan, Zheng 3068efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 30695955102cSAl Viro inode_lock(inode); 3070efa56464SYan, Zheng 3071dcb40c19SWang Xiaoguang ret = btrfs_check_data_free_space(inode, prealloc_start, 3072dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3073efa56464SYan, Zheng if (ret) 3074efa56464SYan, Zheng goto out; 3075efa56464SYan, Zheng 307618513091SWang Xiaoguang cur_offset = prealloc_start; 3077efa56464SYan, Zheng while (nr < cluster->nr) { 3078efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3079efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3080efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3081efa56464SYan, Zheng else 3082efa56464SYan, Zheng end = cluster->end - offset; 3083efa56464SYan, Zheng 3084d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3085efa56464SYan, Zheng num_bytes = end + 1 - start; 308618513091SWang Xiaoguang if (cur_offset < start) 308718513091SWang Xiaoguang btrfs_free_reserved_data_space(inode, cur_offset, 308818513091SWang Xiaoguang start - cur_offset); 3089efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3090efa56464SYan, Zheng num_bytes, num_bytes, 3091efa56464SYan, Zheng end + 1, &alloc_hint); 309218513091SWang Xiaoguang cur_offset = end + 1; 3093d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3094efa56464SYan, Zheng if (ret) 3095efa56464SYan, Zheng break; 3096efa56464SYan, Zheng nr++; 3097efa56464SYan, Zheng } 309818513091SWang Xiaoguang if (cur_offset < prealloc_end) 309918513091SWang Xiaoguang btrfs_free_reserved_data_space(inode, cur_offset, 310018513091SWang Xiaoguang prealloc_end + 1 - cur_offset); 3101efa56464SYan, Zheng out: 31025955102cSAl Viro inode_unlock(inode); 3103efa56464SYan, Zheng return ret; 3104efa56464SYan, Zheng } 3105efa56464SYan, Zheng 3106efa56464SYan, Zheng static noinline_for_stack 31070257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 31080257bb82SYan, Zheng u64 block_start) 31095d4f98a2SYan Zheng { 31105d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 31115d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 31125d4f98a2SYan Zheng struct extent_map *em; 31130257bb82SYan, Zheng int ret = 0; 31145d4f98a2SYan Zheng 3115172ddd60SDavid Sterba em = alloc_extent_map(); 31160257bb82SYan, Zheng if (!em) 31170257bb82SYan, Zheng return -ENOMEM; 31180257bb82SYan, Zheng 31195d4f98a2SYan Zheng em->start = start; 31200257bb82SYan, Zheng em->len = end + 1 - start; 31210257bb82SYan, Zheng em->block_len = em->len; 31220257bb82SYan, Zheng em->block_start = block_start; 31235d4f98a2SYan Zheng em->bdev = root->fs_info->fs_devices->latest_bdev; 31245d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 31255d4f98a2SYan Zheng 3126d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 31275d4f98a2SYan Zheng while (1) { 3128890871beSChris Mason write_lock(&em_tree->lock); 312909a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3130890871beSChris Mason write_unlock(&em_tree->lock); 31315d4f98a2SYan Zheng if (ret != -EEXIST) { 31325d4f98a2SYan Zheng free_extent_map(em); 31335d4f98a2SYan Zheng break; 31345d4f98a2SYan Zheng } 31355d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 0); 31365d4f98a2SYan Zheng } 3137d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 31380257bb82SYan, Zheng return ret; 31390257bb82SYan, Zheng } 31405d4f98a2SYan Zheng 31410257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 31420257bb82SYan, Zheng struct file_extent_cluster *cluster) 31430257bb82SYan, Zheng { 31440257bb82SYan, Zheng u64 page_start; 31450257bb82SYan, Zheng u64 page_end; 31460257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 31470257bb82SYan, Zheng unsigned long index; 31480257bb82SYan, Zheng unsigned long last_index; 31490257bb82SYan, Zheng struct page *page; 31500257bb82SYan, Zheng struct file_ra_state *ra; 31513b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 31520257bb82SYan, Zheng int nr = 0; 31530257bb82SYan, Zheng int ret = 0; 31540257bb82SYan, Zheng 31550257bb82SYan, Zheng if (!cluster->nr) 31560257bb82SYan, Zheng return 0; 31570257bb82SYan, Zheng 31580257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 31590257bb82SYan, Zheng if (!ra) 31600257bb82SYan, Zheng return -ENOMEM; 31610257bb82SYan, Zheng 3162efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 31630257bb82SYan, Zheng if (ret) 3164efa56464SYan, Zheng goto out; 31650257bb82SYan, Zheng 31660257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 31670257bb82SYan, Zheng 3168efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3169efa56464SYan, Zheng cluster->end - offset, cluster->start); 3170efa56464SYan, Zheng if (ret) 3171efa56464SYan, Zheng goto out; 3172efa56464SYan, Zheng 317309cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 317409cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 31750257bb82SYan, Zheng while (index <= last_index) { 317609cbfeafSKirill A. Shutemov ret = btrfs_delalloc_reserve_metadata(inode, PAGE_SIZE); 3177efa56464SYan, Zheng if (ret) 3178efa56464SYan, Zheng goto out; 3179efa56464SYan, Zheng 31800257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 31810257bb82SYan, Zheng if (!page) { 31820257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 31830257bb82SYan, Zheng ra, NULL, index, 31840257bb82SYan, Zheng last_index + 1 - index); 3185a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 31863b16a4e3SJosef Bacik mask); 31870257bb82SYan, Zheng if (!page) { 3188efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 318909cbfeafSKirill A. Shutemov PAGE_SIZE); 31900257bb82SYan, Zheng ret = -ENOMEM; 3191efa56464SYan, Zheng goto out; 31920257bb82SYan, Zheng } 31930257bb82SYan, Zheng } 31940257bb82SYan, Zheng 31950257bb82SYan, Zheng if (PageReadahead(page)) { 31960257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 31970257bb82SYan, Zheng ra, NULL, page, index, 31980257bb82SYan, Zheng last_index + 1 - index); 31990257bb82SYan, Zheng } 32000257bb82SYan, Zheng 32010257bb82SYan, Zheng if (!PageUptodate(page)) { 32020257bb82SYan, Zheng btrfs_readpage(NULL, page); 32030257bb82SYan, Zheng lock_page(page); 32040257bb82SYan, Zheng if (!PageUptodate(page)) { 32050257bb82SYan, Zheng unlock_page(page); 320609cbfeafSKirill A. Shutemov put_page(page); 3207efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 320809cbfeafSKirill A. Shutemov PAGE_SIZE); 32090257bb82SYan, Zheng ret = -EIO; 3210efa56464SYan, Zheng goto out; 32110257bb82SYan, Zheng } 32120257bb82SYan, Zheng } 32130257bb82SYan, Zheng 32144eee4fa4SMiao Xie page_start = page_offset(page); 321509cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 32160257bb82SYan, Zheng 3217d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 32180257bb82SYan, Zheng 32190257bb82SYan, Zheng set_page_extent_mapped(page); 32200257bb82SYan, Zheng 32210257bb82SYan, Zheng if (nr < cluster->nr && 32220257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 32230257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 32240257bb82SYan, Zheng page_start, page_end, 3225ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 32260257bb82SYan, Zheng nr++; 32270257bb82SYan, Zheng } 32280257bb82SYan, Zheng 3229ba8b04c1SQu Wenruo btrfs_set_extent_delalloc(inode, page_start, page_end, NULL, 0); 32300257bb82SYan, Zheng set_page_dirty(page); 32310257bb82SYan, Zheng 32320257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3233d0082371SJeff Mahoney page_start, page_end); 32340257bb82SYan, Zheng unlock_page(page); 323509cbfeafSKirill A. Shutemov put_page(page); 32360257bb82SYan, Zheng 32370257bb82SYan, Zheng index++; 3238efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 3239efa56464SYan, Zheng btrfs_throttle(BTRFS_I(inode)->root); 32400257bb82SYan, Zheng } 32410257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3242efa56464SYan, Zheng out: 32430257bb82SYan, Zheng kfree(ra); 32440257bb82SYan, Zheng return ret; 32450257bb82SYan, Zheng } 32460257bb82SYan, Zheng 32470257bb82SYan, Zheng static noinline_for_stack 32480257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 32490257bb82SYan, Zheng struct file_extent_cluster *cluster) 32500257bb82SYan, Zheng { 32510257bb82SYan, Zheng int ret; 32520257bb82SYan, Zheng 32530257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 32540257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 32550257bb82SYan, Zheng if (ret) 32560257bb82SYan, Zheng return ret; 32570257bb82SYan, Zheng cluster->nr = 0; 32580257bb82SYan, Zheng } 32590257bb82SYan, Zheng 32600257bb82SYan, Zheng if (!cluster->nr) 32610257bb82SYan, Zheng cluster->start = extent_key->objectid; 32620257bb82SYan, Zheng else 32630257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 32640257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 32650257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 32660257bb82SYan, Zheng cluster->nr++; 32670257bb82SYan, Zheng 32680257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 32690257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 32700257bb82SYan, Zheng if (ret) 32710257bb82SYan, Zheng return ret; 32720257bb82SYan, Zheng cluster->nr = 0; 32730257bb82SYan, Zheng } 32740257bb82SYan, Zheng return 0; 32755d4f98a2SYan Zheng } 32765d4f98a2SYan Zheng 32775d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 32785d4f98a2SYan Zheng static int get_ref_objectid_v0(struct reloc_control *rc, 32795d4f98a2SYan Zheng struct btrfs_path *path, 32805d4f98a2SYan Zheng struct btrfs_key *extent_key, 32815d4f98a2SYan Zheng u64 *ref_objectid, int *path_change) 32825d4f98a2SYan Zheng { 32835d4f98a2SYan Zheng struct btrfs_key key; 32845d4f98a2SYan Zheng struct extent_buffer *leaf; 32855d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 32865d4f98a2SYan Zheng int ret; 32875d4f98a2SYan Zheng int slot; 32885d4f98a2SYan Zheng 32895d4f98a2SYan Zheng leaf = path->nodes[0]; 32905d4f98a2SYan Zheng slot = path->slots[0]; 32915d4f98a2SYan Zheng while (1) { 32925d4f98a2SYan Zheng if (slot >= btrfs_header_nritems(leaf)) { 32935d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 32945d4f98a2SYan Zheng if (ret < 0) 32955d4f98a2SYan Zheng return ret; 32965d4f98a2SYan Zheng BUG_ON(ret > 0); 32975d4f98a2SYan Zheng leaf = path->nodes[0]; 32985d4f98a2SYan Zheng slot = path->slots[0]; 32995d4f98a2SYan Zheng if (path_change) 33005d4f98a2SYan Zheng *path_change = 1; 33015d4f98a2SYan Zheng } 33025d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 33035d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 33045d4f98a2SYan Zheng return -ENOENT; 33055d4f98a2SYan Zheng 33065d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_REF_V0_KEY) { 33075d4f98a2SYan Zheng slot++; 33085d4f98a2SYan Zheng continue; 33095d4f98a2SYan Zheng } 33105d4f98a2SYan Zheng ref0 = btrfs_item_ptr(leaf, slot, 33115d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 33125d4f98a2SYan Zheng *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0); 33135d4f98a2SYan Zheng break; 33145d4f98a2SYan Zheng } 33155d4f98a2SYan Zheng return 0; 33165d4f98a2SYan Zheng } 33175d4f98a2SYan Zheng #endif 33185d4f98a2SYan Zheng 33195d4f98a2SYan Zheng /* 33205d4f98a2SYan Zheng * helper to add a tree block to the list. 33215d4f98a2SYan Zheng * the major work is getting the generation and level of the block 33225d4f98a2SYan Zheng */ 33235d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 33245d4f98a2SYan Zheng struct btrfs_key *extent_key, 33255d4f98a2SYan Zheng struct btrfs_path *path, 33265d4f98a2SYan Zheng struct rb_root *blocks) 33275d4f98a2SYan Zheng { 33285d4f98a2SYan Zheng struct extent_buffer *eb; 33295d4f98a2SYan Zheng struct btrfs_extent_item *ei; 33305d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 33315d4f98a2SYan Zheng struct tree_block *block; 33325d4f98a2SYan Zheng struct rb_node *rb_node; 33335d4f98a2SYan Zheng u32 item_size; 33345d4f98a2SYan Zheng int level = -1; 33357fdf4b60SWang Shilong u64 generation; 33365d4f98a2SYan Zheng 33375d4f98a2SYan Zheng eb = path->nodes[0]; 33385d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 33395d4f98a2SYan Zheng 33403173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 33413173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 33425d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 33435d4f98a2SYan Zheng struct btrfs_extent_item); 33443173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 33455d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 33465d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 33475d4f98a2SYan Zheng } else { 33483173a18fSJosef Bacik level = (int)extent_key->offset; 33493173a18fSJosef Bacik } 33503173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 33513173a18fSJosef Bacik } else { 33525d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 33535d4f98a2SYan Zheng u64 ref_owner; 33545d4f98a2SYan Zheng int ret; 33555d4f98a2SYan Zheng 33565d4f98a2SYan Zheng BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 33575d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, extent_key, 33585d4f98a2SYan Zheng &ref_owner, NULL); 3359411fc6bcSAndi Kleen if (ret < 0) 3360411fc6bcSAndi Kleen return ret; 33615d4f98a2SYan Zheng BUG_ON(ref_owner >= BTRFS_MAX_LEVEL); 33625d4f98a2SYan Zheng level = (int)ref_owner; 33635d4f98a2SYan Zheng /* FIXME: get real generation */ 33645d4f98a2SYan Zheng generation = 0; 33655d4f98a2SYan Zheng #else 33665d4f98a2SYan Zheng BUG(); 33675d4f98a2SYan Zheng #endif 33685d4f98a2SYan Zheng } 33695d4f98a2SYan Zheng 3370b3b4aa74SDavid Sterba btrfs_release_path(path); 33715d4f98a2SYan Zheng 33725d4f98a2SYan Zheng BUG_ON(level == -1); 33735d4f98a2SYan Zheng 33745d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 33755d4f98a2SYan Zheng if (!block) 33765d4f98a2SYan Zheng return -ENOMEM; 33775d4f98a2SYan Zheng 33785d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3379707e8a07SDavid Sterba block->key.objectid = rc->extent_root->nodesize; 33805d4f98a2SYan Zheng block->key.offset = generation; 33815d4f98a2SYan Zheng block->level = level; 33825d4f98a2SYan Zheng block->key_ready = 0; 33835d4f98a2SYan Zheng 33845d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 338543c04fb1SJeff Mahoney if (rb_node) 338643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 33875d4f98a2SYan Zheng 33885d4f98a2SYan Zheng return 0; 33895d4f98a2SYan Zheng } 33905d4f98a2SYan Zheng 33915d4f98a2SYan Zheng /* 33925d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 33935d4f98a2SYan Zheng */ 33945d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 33955d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 33965d4f98a2SYan Zheng struct rb_root *blocks) 33975d4f98a2SYan Zheng { 33985d4f98a2SYan Zheng struct btrfs_path *path; 33995d4f98a2SYan Zheng struct btrfs_key key; 34005d4f98a2SYan Zheng int ret; 3401aee68ee5SJosef Bacik bool skinny = btrfs_fs_incompat(rc->extent_root->fs_info, 3402aee68ee5SJosef Bacik SKINNY_METADATA); 34035d4f98a2SYan Zheng 34047476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 34055d4f98a2SYan Zheng return 0; 34065d4f98a2SYan Zheng 34075d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 34085d4f98a2SYan Zheng return 0; 34095d4f98a2SYan Zheng 34105d4f98a2SYan Zheng path = btrfs_alloc_path(); 34115d4f98a2SYan Zheng if (!path) 34125d4f98a2SYan Zheng return -ENOMEM; 3413aee68ee5SJosef Bacik again: 34145d4f98a2SYan Zheng key.objectid = bytenr; 3415aee68ee5SJosef Bacik if (skinny) { 3416aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3417aee68ee5SJosef Bacik key.offset = (u64)-1; 3418aee68ee5SJosef Bacik } else { 34195d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34205d4f98a2SYan Zheng key.offset = blocksize; 3421aee68ee5SJosef Bacik } 34225d4f98a2SYan Zheng 34235d4f98a2SYan Zheng path->search_commit_root = 1; 34245d4f98a2SYan Zheng path->skip_locking = 1; 34255d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 34265d4f98a2SYan Zheng if (ret < 0) 34275d4f98a2SYan Zheng goto out; 34285d4f98a2SYan Zheng 3429aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3430aee68ee5SJosef Bacik if (path->slots[0]) { 3431aee68ee5SJosef Bacik path->slots[0]--; 3432aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3433aee68ee5SJosef Bacik path->slots[0]); 34343173a18fSJosef Bacik if (key.objectid == bytenr && 3435aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3436aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3437aee68ee5SJosef Bacik key.offset == blocksize))) 34383173a18fSJosef Bacik ret = 0; 34393173a18fSJosef Bacik } 3440aee68ee5SJosef Bacik 3441aee68ee5SJosef Bacik if (ret) { 3442aee68ee5SJosef Bacik skinny = false; 3443aee68ee5SJosef Bacik btrfs_release_path(path); 3444aee68ee5SJosef Bacik goto again; 3445aee68ee5SJosef Bacik } 3446aee68ee5SJosef Bacik } 34473173a18fSJosef Bacik BUG_ON(ret); 34483173a18fSJosef Bacik 34495d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 34505d4f98a2SYan Zheng out: 34515d4f98a2SYan Zheng btrfs_free_path(path); 34525d4f98a2SYan Zheng return ret; 34535d4f98a2SYan Zheng } 34545d4f98a2SYan Zheng 34555d4f98a2SYan Zheng /* 34565d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 34575d4f98a2SYan Zheng */ 34585d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 34595d4f98a2SYan Zheng struct extent_buffer *eb) 34605d4f98a2SYan Zheng { 34615d4f98a2SYan Zheng u64 flags; 34625d4f98a2SYan Zheng int ret; 34635d4f98a2SYan Zheng 34645d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 34655d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 34665d4f98a2SYan Zheng return 1; 34675d4f98a2SYan Zheng 34683fd0a558SYan, Zheng ret = btrfs_lookup_extent_info(NULL, rc->extent_root, 34693173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 34703173a18fSJosef Bacik NULL, &flags); 34715d4f98a2SYan Zheng BUG_ON(ret); 34725d4f98a2SYan Zheng 34735d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 34745d4f98a2SYan Zheng ret = 1; 34755d4f98a2SYan Zheng else 34765d4f98a2SYan Zheng ret = 0; 34775d4f98a2SYan Zheng return ret; 34785d4f98a2SYan Zheng } 34795d4f98a2SYan Zheng 34800af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 34811bbc621eSChris Mason struct btrfs_block_group_cache *block_group, 34821bbc621eSChris Mason struct inode *inode, 34831bbc621eSChris Mason u64 ino) 34840af3d00bSJosef Bacik { 34850af3d00bSJosef Bacik struct btrfs_key key; 34860af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 34870af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 34880af3d00bSJosef Bacik int ret = 0; 34890af3d00bSJosef Bacik 34900af3d00bSJosef Bacik if (inode) 34910af3d00bSJosef Bacik goto truncate; 34920af3d00bSJosef Bacik 34930af3d00bSJosef Bacik key.objectid = ino; 34940af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 34950af3d00bSJosef Bacik key.offset = 0; 34960af3d00bSJosef Bacik 34970af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 3498f54fb859STsutomu Itoh if (IS_ERR(inode) || is_bad_inode(inode)) { 3499f54fb859STsutomu Itoh if (!IS_ERR(inode)) 35000af3d00bSJosef Bacik iput(inode); 35010af3d00bSJosef Bacik return -ENOENT; 35020af3d00bSJosef Bacik } 35030af3d00bSJosef Bacik 35040af3d00bSJosef Bacik truncate: 35057b61cd92SMiao Xie ret = btrfs_check_trunc_cache_free_space(root, 35067b61cd92SMiao Xie &fs_info->global_block_rsv); 35077b61cd92SMiao Xie if (ret) 35087b61cd92SMiao Xie goto out; 35097b61cd92SMiao Xie 35107a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 35110af3d00bSJosef Bacik if (IS_ERR(trans)) { 35123612b495STsutomu Itoh ret = PTR_ERR(trans); 35130af3d00bSJosef Bacik goto out; 35140af3d00bSJosef Bacik } 35150af3d00bSJosef Bacik 35161bbc621eSChris Mason ret = btrfs_truncate_free_space_cache(root, trans, block_group, inode); 35170af3d00bSJosef Bacik 35180af3d00bSJosef Bacik btrfs_end_transaction(trans, root); 3519b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 35200af3d00bSJosef Bacik out: 35210af3d00bSJosef Bacik iput(inode); 35220af3d00bSJosef Bacik return ret; 35230af3d00bSJosef Bacik } 35240af3d00bSJosef Bacik 35255d4f98a2SYan Zheng /* 35265d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 35275d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 35285d4f98a2SYan Zheng */ 35295d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 35305d4f98a2SYan Zheng struct btrfs_key *extent_key, 35315d4f98a2SYan Zheng struct extent_buffer *leaf, 35325d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 35335d4f98a2SYan Zheng struct rb_root *blocks) 35345d4f98a2SYan Zheng { 35355d4f98a2SYan Zheng struct btrfs_path *path; 35365d4f98a2SYan Zheng struct tree_block *block; 35375d4f98a2SYan Zheng struct btrfs_root *root; 35385d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 35395d4f98a2SYan Zheng struct rb_node *rb_node; 35405d4f98a2SYan Zheng struct btrfs_key key; 35415d4f98a2SYan Zheng u64 ref_root; 35425d4f98a2SYan Zheng u64 ref_objectid; 35435d4f98a2SYan Zheng u64 ref_offset; 35445d4f98a2SYan Zheng u32 ref_count; 35455d4f98a2SYan Zheng u32 nritems; 35465d4f98a2SYan Zheng int err = 0; 35475d4f98a2SYan Zheng int added = 0; 35485d4f98a2SYan Zheng int counted; 35495d4f98a2SYan Zheng int ret; 35505d4f98a2SYan Zheng 35515d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 35525d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 35535d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 35545d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 35555d4f98a2SYan Zheng 35560af3d00bSJosef Bacik /* 35570af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 35580af3d00bSJosef Bacik * it and redo the search. 35590af3d00bSJosef Bacik */ 35600af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 35610af3d00bSJosef Bacik ret = delete_block_group_cache(rc->extent_root->fs_info, 35621bbc621eSChris Mason rc->block_group, 35630af3d00bSJosef Bacik NULL, ref_objectid); 35640af3d00bSJosef Bacik if (ret != -ENOENT) 35650af3d00bSJosef Bacik return ret; 35660af3d00bSJosef Bacik ret = 0; 35670af3d00bSJosef Bacik } 35680af3d00bSJosef Bacik 35690af3d00bSJosef Bacik path = btrfs_alloc_path(); 35700af3d00bSJosef Bacik if (!path) 35710af3d00bSJosef Bacik return -ENOMEM; 3572e4058b54SDavid Sterba path->reada = READA_FORWARD; 35730af3d00bSJosef Bacik 35745d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, ref_root); 35755d4f98a2SYan Zheng if (IS_ERR(root)) { 35765d4f98a2SYan Zheng err = PTR_ERR(root); 35775d4f98a2SYan Zheng goto out; 35785d4f98a2SYan Zheng } 35795d4f98a2SYan Zheng 35805d4f98a2SYan Zheng key.objectid = ref_objectid; 35815d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 358284850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 358384850e8dSYan, Zheng key.offset = 0; 358484850e8dSYan, Zheng else 358584850e8dSYan, Zheng key.offset = ref_offset; 35865d4f98a2SYan Zheng 35875d4f98a2SYan Zheng path->search_commit_root = 1; 35885d4f98a2SYan Zheng path->skip_locking = 1; 35895d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 35905d4f98a2SYan Zheng if (ret < 0) { 35915d4f98a2SYan Zheng err = ret; 35925d4f98a2SYan Zheng goto out; 35935d4f98a2SYan Zheng } 35945d4f98a2SYan Zheng 35955d4f98a2SYan Zheng leaf = path->nodes[0]; 35965d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 35975d4f98a2SYan Zheng /* 35985d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 35995d4f98a2SYan Zheng * are not counted in 36005d4f98a2SYan Zheng */ 36015d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36025d4f98a2SYan Zheng counted = 0; 36035d4f98a2SYan Zheng else 36045d4f98a2SYan Zheng counted = 1; 36055d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36065d4f98a2SYan Zheng if (rb_node) { 36075d4f98a2SYan Zheng if (counted) 36085d4f98a2SYan Zheng added = 1; 36095d4f98a2SYan Zheng else 36105d4f98a2SYan Zheng path->slots[0] = nritems; 36115d4f98a2SYan Zheng } 36125d4f98a2SYan Zheng 36135d4f98a2SYan Zheng while (ref_count > 0) { 36145d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 36155d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 36165d4f98a2SYan Zheng if (ret < 0) { 36175d4f98a2SYan Zheng err = ret; 36185d4f98a2SYan Zheng goto out; 36195d4f98a2SYan Zheng } 3620fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 36215d4f98a2SYan Zheng goto out; 36225d4f98a2SYan Zheng 36235d4f98a2SYan Zheng leaf = path->nodes[0]; 36245d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36255d4f98a2SYan Zheng added = 0; 36265d4f98a2SYan Zheng 36275d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36285d4f98a2SYan Zheng counted = 0; 36295d4f98a2SYan Zheng else 36305d4f98a2SYan Zheng counted = 1; 36315d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36325d4f98a2SYan Zheng if (rb_node) { 36335d4f98a2SYan Zheng if (counted) 36345d4f98a2SYan Zheng added = 1; 36355d4f98a2SYan Zheng else 36365d4f98a2SYan Zheng path->slots[0] = nritems; 36375d4f98a2SYan Zheng } 36385d4f98a2SYan Zheng } 36395d4f98a2SYan Zheng 36405d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3641fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3642fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 36435d4f98a2SYan Zheng break; 36445d4f98a2SYan Zheng 36455d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 36465d4f98a2SYan Zheng struct btrfs_file_extent_item); 36475d4f98a2SYan Zheng 36485d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 36495d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 36505d4f98a2SYan Zheng goto next; 36515d4f98a2SYan Zheng 36525d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 36535d4f98a2SYan Zheng extent_key->objectid) 36545d4f98a2SYan Zheng goto next; 36555d4f98a2SYan Zheng 36565d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 36575d4f98a2SYan Zheng if (key.offset != ref_offset) 36585d4f98a2SYan Zheng goto next; 36595d4f98a2SYan Zheng 36605d4f98a2SYan Zheng if (counted) 36615d4f98a2SYan Zheng ref_count--; 36625d4f98a2SYan Zheng if (added) 36635d4f98a2SYan Zheng goto next; 36645d4f98a2SYan Zheng 36657476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 36665d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 36675d4f98a2SYan Zheng if (!block) { 36685d4f98a2SYan Zheng err = -ENOMEM; 36695d4f98a2SYan Zheng break; 36705d4f98a2SYan Zheng } 36715d4f98a2SYan Zheng block->bytenr = leaf->start; 36725d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 36735d4f98a2SYan Zheng block->level = 0; 36745d4f98a2SYan Zheng block->key_ready = 1; 36755d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 36765d4f98a2SYan Zheng &block->rb_node); 367743c04fb1SJeff Mahoney if (rb_node) 367843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 367943c04fb1SJeff Mahoney block->bytenr); 36805d4f98a2SYan Zheng } 36815d4f98a2SYan Zheng if (counted) 36825d4f98a2SYan Zheng added = 1; 36835d4f98a2SYan Zheng else 36845d4f98a2SYan Zheng path->slots[0] = nritems; 36855d4f98a2SYan Zheng next: 36865d4f98a2SYan Zheng path->slots[0]++; 36875d4f98a2SYan Zheng 36885d4f98a2SYan Zheng } 36895d4f98a2SYan Zheng out: 36905d4f98a2SYan Zheng btrfs_free_path(path); 36915d4f98a2SYan Zheng return err; 36925d4f98a2SYan Zheng } 36935d4f98a2SYan Zheng 36945d4f98a2SYan Zheng /* 36952c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 36965d4f98a2SYan Zheng */ 36975d4f98a2SYan Zheng static noinline_for_stack 36985d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 36995d4f98a2SYan Zheng struct btrfs_key *extent_key, 37005d4f98a2SYan Zheng struct btrfs_path *path, 37015d4f98a2SYan Zheng struct rb_root *blocks) 37025d4f98a2SYan Zheng { 37035d4f98a2SYan Zheng struct btrfs_key key; 37045d4f98a2SYan Zheng struct extent_buffer *eb; 37055d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 37065d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 37075d4f98a2SYan Zheng unsigned long ptr; 37085d4f98a2SYan Zheng unsigned long end; 3709707e8a07SDavid Sterba u32 blocksize = rc->extent_root->nodesize; 3710647f63bdSFilipe David Borba Manana int ret = 0; 37115d4f98a2SYan Zheng int err = 0; 37125d4f98a2SYan Zheng 37135d4f98a2SYan Zheng eb = path->nodes[0]; 37145d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 37155d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 37165d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 37175d4f98a2SYan Zheng if (ptr + sizeof(struct btrfs_extent_item_v0) == end) 37185d4f98a2SYan Zheng ptr = end; 37195d4f98a2SYan Zheng else 37205d4f98a2SYan Zheng #endif 37215d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 37225d4f98a2SYan Zheng 37235d4f98a2SYan Zheng while (ptr < end) { 37245d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 37255d4f98a2SYan Zheng key.type = btrfs_extent_inline_ref_type(eb, iref); 37265d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37275d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 37285d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37295d4f98a2SYan Zheng blocks); 37305d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37315d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 37325d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37335d4f98a2SYan Zheng eb, dref, blocks); 37345d4f98a2SYan Zheng } else { 37355d4f98a2SYan Zheng BUG(); 37365d4f98a2SYan Zheng } 3737647f63bdSFilipe David Borba Manana if (ret) { 3738647f63bdSFilipe David Borba Manana err = ret; 3739647f63bdSFilipe David Borba Manana goto out; 3740647f63bdSFilipe David Borba Manana } 37415d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 37425d4f98a2SYan Zheng } 37435d4f98a2SYan Zheng WARN_ON(ptr > end); 37445d4f98a2SYan Zheng 37455d4f98a2SYan Zheng while (1) { 37465d4f98a2SYan Zheng cond_resched(); 37475d4f98a2SYan Zheng eb = path->nodes[0]; 37485d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 37495d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 37505d4f98a2SYan Zheng if (ret < 0) { 37515d4f98a2SYan Zheng err = ret; 37525d4f98a2SYan Zheng break; 37535d4f98a2SYan Zheng } 37545d4f98a2SYan Zheng if (ret > 0) 37555d4f98a2SYan Zheng break; 37565d4f98a2SYan Zheng eb = path->nodes[0]; 37575d4f98a2SYan Zheng } 37585d4f98a2SYan Zheng 37595d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 37605d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 37615d4f98a2SYan Zheng break; 37625d4f98a2SYan Zheng 37635d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 37645d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY || 37655d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 37665d4f98a2SYan Zheng #else 37675d4f98a2SYan Zheng BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); 37685d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37695d4f98a2SYan Zheng #endif 37705d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37715d4f98a2SYan Zheng blocks); 37725d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37735d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 37745d4f98a2SYan Zheng struct btrfs_extent_data_ref); 37755d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37765d4f98a2SYan Zheng eb, dref, blocks); 37775d4f98a2SYan Zheng } else { 37785d4f98a2SYan Zheng ret = 0; 37795d4f98a2SYan Zheng } 37805d4f98a2SYan Zheng if (ret) { 37815d4f98a2SYan Zheng err = ret; 37825d4f98a2SYan Zheng break; 37835d4f98a2SYan Zheng } 37845d4f98a2SYan Zheng path->slots[0]++; 37855d4f98a2SYan Zheng } 3786647f63bdSFilipe David Borba Manana out: 3787b3b4aa74SDavid Sterba btrfs_release_path(path); 37885d4f98a2SYan Zheng if (err) 37895d4f98a2SYan Zheng free_block_list(blocks); 37905d4f98a2SYan Zheng return err; 37915d4f98a2SYan Zheng } 37925d4f98a2SYan Zheng 37935d4f98a2SYan Zheng /* 37942c016dc2SLiu Bo * helper to find next unprocessed extent 37955d4f98a2SYan Zheng */ 37965d4f98a2SYan Zheng static noinline_for_stack 3797147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 37983fd0a558SYan, Zheng struct btrfs_key *extent_key) 37995d4f98a2SYan Zheng { 38005d4f98a2SYan Zheng struct btrfs_key key; 38015d4f98a2SYan Zheng struct extent_buffer *leaf; 38025d4f98a2SYan Zheng u64 start, end, last; 38035d4f98a2SYan Zheng int ret; 38045d4f98a2SYan Zheng 38055d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 38065d4f98a2SYan Zheng while (1) { 38075d4f98a2SYan Zheng cond_resched(); 38085d4f98a2SYan Zheng if (rc->search_start >= last) { 38095d4f98a2SYan Zheng ret = 1; 38105d4f98a2SYan Zheng break; 38115d4f98a2SYan Zheng } 38125d4f98a2SYan Zheng 38135d4f98a2SYan Zheng key.objectid = rc->search_start; 38145d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 38155d4f98a2SYan Zheng key.offset = 0; 38165d4f98a2SYan Zheng 38175d4f98a2SYan Zheng path->search_commit_root = 1; 38185d4f98a2SYan Zheng path->skip_locking = 1; 38195d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 38205d4f98a2SYan Zheng 0, 0); 38215d4f98a2SYan Zheng if (ret < 0) 38225d4f98a2SYan Zheng break; 38235d4f98a2SYan Zheng next: 38245d4f98a2SYan Zheng leaf = path->nodes[0]; 38255d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38265d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38275d4f98a2SYan Zheng if (ret != 0) 38285d4f98a2SYan Zheng break; 38295d4f98a2SYan Zheng leaf = path->nodes[0]; 38305d4f98a2SYan Zheng } 38315d4f98a2SYan Zheng 38325d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38335d4f98a2SYan Zheng if (key.objectid >= last) { 38345d4f98a2SYan Zheng ret = 1; 38355d4f98a2SYan Zheng break; 38365d4f98a2SYan Zheng } 38375d4f98a2SYan Zheng 38383173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38393173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38403173a18fSJosef Bacik path->slots[0]++; 38413173a18fSJosef Bacik goto next; 38423173a18fSJosef Bacik } 38433173a18fSJosef Bacik 38443173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 38455d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 38465d4f98a2SYan Zheng path->slots[0]++; 38475d4f98a2SYan Zheng goto next; 38485d4f98a2SYan Zheng } 38495d4f98a2SYan Zheng 38503173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 3851707e8a07SDavid Sterba key.objectid + rc->extent_root->nodesize <= 38523173a18fSJosef Bacik rc->search_start) { 38533173a18fSJosef Bacik path->slots[0]++; 38543173a18fSJosef Bacik goto next; 38553173a18fSJosef Bacik } 38563173a18fSJosef Bacik 38575d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 38585d4f98a2SYan Zheng key.objectid, &start, &end, 3859e6138876SJosef Bacik EXTENT_DIRTY, NULL); 38605d4f98a2SYan Zheng 38615d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3862b3b4aa74SDavid Sterba btrfs_release_path(path); 38635d4f98a2SYan Zheng rc->search_start = end + 1; 38645d4f98a2SYan Zheng } else { 38653173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 38665d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 38673173a18fSJosef Bacik else 38683173a18fSJosef Bacik rc->search_start = key.objectid + 3869707e8a07SDavid Sterba rc->extent_root->nodesize; 38703fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 38715d4f98a2SYan Zheng return 0; 38725d4f98a2SYan Zheng } 38735d4f98a2SYan Zheng } 3874b3b4aa74SDavid Sterba btrfs_release_path(path); 38755d4f98a2SYan Zheng return ret; 38765d4f98a2SYan Zheng } 38775d4f98a2SYan Zheng 38785d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 38795d4f98a2SYan Zheng { 38805d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38817585717fSChris Mason 38827585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38835d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 38847585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38855d4f98a2SYan Zheng } 38865d4f98a2SYan Zheng 38875d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 38885d4f98a2SYan Zheng { 38895d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38907585717fSChris Mason 38917585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38925d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 38937585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38945d4f98a2SYan Zheng } 38955d4f98a2SYan Zheng 38965d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 38975d4f98a2SYan Zheng { 38985d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38995d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39005d4f98a2SYan Zheng return 1; 39015d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 39025d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39035d4f98a2SYan Zheng return 1; 39045d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39055d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 39065d4f98a2SYan Zheng return 1; 39075d4f98a2SYan Zheng return 0; 39085d4f98a2SYan Zheng } 39095d4f98a2SYan Zheng 39103fd0a558SYan, Zheng static noinline_for_stack 39113fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 39123fd0a558SYan, Zheng { 39133fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3914ac2fabacSJosef Bacik int ret; 39153fd0a558SYan, Zheng 391666d8f3ddSMiao Xie rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root, 391766d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 39183fd0a558SYan, Zheng if (!rc->block_rsv) 39193fd0a558SYan, Zheng return -ENOMEM; 39203fd0a558SYan, Zheng 39213fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 39223fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 39233fd0a558SYan, Zheng rc->extents_found = 0; 39243fd0a558SYan, Zheng rc->nodes_relocated = 0; 39253fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39260647bf56SWang Shilong rc->reserved_bytes = 0; 39270647bf56SWang Shilong rc->block_rsv->size = rc->extent_root->nodesize * 39280647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3929ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3930ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3931ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3932ac2fabacSJosef Bacik if (ret) 3933ac2fabacSJosef Bacik return ret; 39343fd0a558SYan, Zheng 39353fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39363fd0a558SYan, Zheng set_reloc_control(rc); 39373fd0a558SYan, Zheng 39387a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 393928818947SLiu Bo if (IS_ERR(trans)) { 394028818947SLiu Bo unset_reloc_control(rc); 394128818947SLiu Bo /* 394228818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 394328818947SLiu Bo * cleanup. And callers are responsible to free the above 394428818947SLiu Bo * block rsv. 394528818947SLiu Bo */ 394628818947SLiu Bo return PTR_ERR(trans); 394728818947SLiu Bo } 39483fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 39493fd0a558SYan, Zheng return 0; 39503fd0a558SYan, Zheng } 395176dda93cSYan, Zheng 395262b99540SQu Wenruo /* 395362b99540SQu Wenruo * Qgroup fixer for data chunk relocation. 395462b99540SQu Wenruo * The data relocation is done in the following steps 395562b99540SQu Wenruo * 1) Copy data extents into data reloc tree 395662b99540SQu Wenruo * 2) Create tree reloc tree(special snapshot) for related subvolumes 395762b99540SQu Wenruo * 3) Modify file extents in tree reloc tree 395862b99540SQu Wenruo * 4) Merge tree reloc tree with original fs tree, by swapping tree blocks 395962b99540SQu Wenruo * 396062b99540SQu Wenruo * The problem is, data and tree reloc tree are not accounted to qgroup, 396162b99540SQu Wenruo * and 4) will only info qgroup to track tree blocks change, not file extents 396262b99540SQu Wenruo * in the tree blocks. 396362b99540SQu Wenruo * 396462b99540SQu Wenruo * The good news is, related data extents are all in data reloc tree, so we 396562b99540SQu Wenruo * only need to info qgroup to track all file extents in data reloc tree 396662b99540SQu Wenruo * before commit trans. 396762b99540SQu Wenruo */ 396862b99540SQu Wenruo static int qgroup_fix_relocated_data_extents(struct btrfs_trans_handle *trans, 396962b99540SQu Wenruo struct reloc_control *rc) 397062b99540SQu Wenruo { 397162b99540SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 397262b99540SQu Wenruo struct inode *inode = rc->data_inode; 397362b99540SQu Wenruo struct btrfs_root *data_reloc_root = BTRFS_I(inode)->root; 397462b99540SQu Wenruo struct btrfs_path *path; 397562b99540SQu Wenruo struct btrfs_key key; 397662b99540SQu Wenruo int ret = 0; 397762b99540SQu Wenruo 3978afcdd129SJosef Bacik if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 397962b99540SQu Wenruo return 0; 398062b99540SQu Wenruo 398162b99540SQu Wenruo /* 398262b99540SQu Wenruo * Only for stage where we update data pointers the qgroup fix is 398362b99540SQu Wenruo * valid. 398462b99540SQu Wenruo * For MOVING_DATA stage, we will miss the timing of swapping tree 398562b99540SQu Wenruo * blocks, and won't fix it. 398662b99540SQu Wenruo */ 398762b99540SQu Wenruo if (!(rc->stage == UPDATE_DATA_PTRS && rc->extents_found)) 398862b99540SQu Wenruo return 0; 398962b99540SQu Wenruo 399062b99540SQu Wenruo path = btrfs_alloc_path(); 399162b99540SQu Wenruo if (!path) 399262b99540SQu Wenruo return -ENOMEM; 399362b99540SQu Wenruo key.objectid = btrfs_ino(inode); 399462b99540SQu Wenruo key.type = BTRFS_EXTENT_DATA_KEY; 399562b99540SQu Wenruo key.offset = 0; 399662b99540SQu Wenruo 399762b99540SQu Wenruo ret = btrfs_search_slot(NULL, data_reloc_root, &key, path, 0, 0); 399862b99540SQu Wenruo if (ret < 0) 399962b99540SQu Wenruo goto out; 400062b99540SQu Wenruo 400162b99540SQu Wenruo lock_extent(&BTRFS_I(inode)->io_tree, 0, (u64)-1); 400262b99540SQu Wenruo while (1) { 400362b99540SQu Wenruo struct btrfs_file_extent_item *fi; 400462b99540SQu Wenruo 400562b99540SQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 400662b99540SQu Wenruo if (key.objectid > btrfs_ino(inode)) 400762b99540SQu Wenruo break; 400862b99540SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 400962b99540SQu Wenruo goto next; 401062b99540SQu Wenruo fi = btrfs_item_ptr(path->nodes[0], path->slots[0], 401162b99540SQu Wenruo struct btrfs_file_extent_item); 401262b99540SQu Wenruo if (btrfs_file_extent_type(path->nodes[0], fi) != 401362b99540SQu Wenruo BTRFS_FILE_EXTENT_REG) 401462b99540SQu Wenruo goto next; 401562b99540SQu Wenruo ret = btrfs_qgroup_insert_dirty_extent(trans, fs_info, 401662b99540SQu Wenruo btrfs_file_extent_disk_bytenr(path->nodes[0], fi), 401762b99540SQu Wenruo btrfs_file_extent_disk_num_bytes(path->nodes[0], fi), 401862b99540SQu Wenruo GFP_NOFS); 401962b99540SQu Wenruo if (ret < 0) 402062b99540SQu Wenruo break; 402162b99540SQu Wenruo next: 402262b99540SQu Wenruo ret = btrfs_next_item(data_reloc_root, path); 402362b99540SQu Wenruo if (ret < 0) 402462b99540SQu Wenruo break; 402562b99540SQu Wenruo if (ret > 0) { 402662b99540SQu Wenruo ret = 0; 402762b99540SQu Wenruo break; 402862b99540SQu Wenruo } 402962b99540SQu Wenruo } 403062b99540SQu Wenruo unlock_extent(&BTRFS_I(inode)->io_tree, 0 , (u64)-1); 403162b99540SQu Wenruo out: 403262b99540SQu Wenruo btrfs_free_path(path); 403362b99540SQu Wenruo return ret; 403462b99540SQu Wenruo } 403562b99540SQu Wenruo 40365d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 40375d4f98a2SYan Zheng { 40385d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 40395d4f98a2SYan Zheng struct btrfs_key key; 40405d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 40415d4f98a2SYan Zheng struct btrfs_path *path; 40425d4f98a2SYan Zheng struct btrfs_extent_item *ei; 40435d4f98a2SYan Zheng u64 flags; 40445d4f98a2SYan Zheng u32 item_size; 40455d4f98a2SYan Zheng int ret; 40465d4f98a2SYan Zheng int err = 0; 4047c87f08caSChris Mason int progress = 0; 40485d4f98a2SYan Zheng 40495d4f98a2SYan Zheng path = btrfs_alloc_path(); 40503fd0a558SYan, Zheng if (!path) 40515d4f98a2SYan Zheng return -ENOMEM; 4052e4058b54SDavid Sterba path->reada = READA_FORWARD; 40533fd0a558SYan, Zheng 40543fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 40553fd0a558SYan, Zheng if (ret) { 40563fd0a558SYan, Zheng err = ret; 40573fd0a558SYan, Zheng goto out_free; 40582423fdfbSJiri Slaby } 40595d4f98a2SYan Zheng 40605d4f98a2SYan Zheng while (1) { 40610647bf56SWang Shilong rc->reserved_bytes = 0; 40620647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40630647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40640647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40650647bf56SWang Shilong if (ret) { 40660647bf56SWang Shilong err = ret; 40670647bf56SWang Shilong break; 40680647bf56SWang Shilong } 4069c87f08caSChris Mason progress++; 4070a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40710f788c58SLiu Bo if (IS_ERR(trans)) { 40720f788c58SLiu Bo err = PTR_ERR(trans); 40730f788c58SLiu Bo trans = NULL; 40740f788c58SLiu Bo break; 40750f788c58SLiu Bo } 4076c87f08caSChris Mason restart: 40773fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40783fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 40793fd0a558SYan, Zheng continue; 40803fd0a558SYan, Zheng } 40813fd0a558SYan, Zheng 4082147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40835d4f98a2SYan Zheng if (ret < 0) 40845d4f98a2SYan Zheng err = ret; 40855d4f98a2SYan Zheng if (ret != 0) 40865d4f98a2SYan Zheng break; 40875d4f98a2SYan Zheng 40885d4f98a2SYan Zheng rc->extents_found++; 40895d4f98a2SYan Zheng 40905d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40915d4f98a2SYan Zheng struct btrfs_extent_item); 40923fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40935d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40945d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40955d4f98a2SYan Zheng ret = check_extent_flags(flags); 40965d4f98a2SYan Zheng BUG_ON(ret); 40975d4f98a2SYan Zheng 40985d4f98a2SYan Zheng } else { 40995d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 41005d4f98a2SYan Zheng u64 ref_owner; 41015d4f98a2SYan Zheng int path_change = 0; 41025d4f98a2SYan Zheng 41035d4f98a2SYan Zheng BUG_ON(item_size != 41045d4f98a2SYan Zheng sizeof(struct btrfs_extent_item_v0)); 41055d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, &key, &ref_owner, 41065d4f98a2SYan Zheng &path_change); 41074b3576e4SZhaolei if (ret < 0) { 41084b3576e4SZhaolei err = ret; 41094b3576e4SZhaolei break; 41104b3576e4SZhaolei } 41115d4f98a2SYan Zheng if (ref_owner < BTRFS_FIRST_FREE_OBJECTID) 41125d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_TREE_BLOCK; 41135d4f98a2SYan Zheng else 41145d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_DATA; 41155d4f98a2SYan Zheng 41165d4f98a2SYan Zheng if (path_change) { 4117b3b4aa74SDavid Sterba btrfs_release_path(path); 41185d4f98a2SYan Zheng 41195d4f98a2SYan Zheng path->search_commit_root = 1; 41205d4f98a2SYan Zheng path->skip_locking = 1; 41215d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, 41225d4f98a2SYan Zheng &key, path, 0, 0); 41235d4f98a2SYan Zheng if (ret < 0) { 41245d4f98a2SYan Zheng err = ret; 41255d4f98a2SYan Zheng break; 41265d4f98a2SYan Zheng } 41275d4f98a2SYan Zheng BUG_ON(ret > 0); 41285d4f98a2SYan Zheng } 41295d4f98a2SYan Zheng #else 41305d4f98a2SYan Zheng BUG(); 41315d4f98a2SYan Zheng #endif 41325d4f98a2SYan Zheng } 41335d4f98a2SYan Zheng 41345d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 41355d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 41365d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 41375d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41385d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 41395d4f98a2SYan Zheng } else { 4140b3b4aa74SDavid Sterba btrfs_release_path(path); 41415d4f98a2SYan Zheng ret = 0; 41425d4f98a2SYan Zheng } 41435d4f98a2SYan Zheng if (ret < 0) { 41443fd0a558SYan, Zheng err = ret; 41455d4f98a2SYan Zheng break; 41465d4f98a2SYan Zheng } 41475d4f98a2SYan Zheng 41485d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 41495d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 41505d4f98a2SYan Zheng if (ret < 0) { 41511708cc57SWang Shilong /* 41521708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 41531708cc57SWang Shilong * backref cache when committing transaction. 41541708cc57SWang Shilong */ 41551708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 41561708cc57SWang Shilong 41573fd0a558SYan, Zheng if (ret != -EAGAIN) { 41585d4f98a2SYan Zheng err = ret; 41595d4f98a2SYan Zheng break; 41605d4f98a2SYan Zheng } 41613fd0a558SYan, Zheng rc->extents_found--; 41623fd0a558SYan, Zheng rc->search_start = key.objectid; 41633fd0a558SYan, Zheng } 41645d4f98a2SYan Zheng } 41655d4f98a2SYan Zheng 41663fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 4167b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 41683fd0a558SYan, Zheng trans = NULL; 41695d4f98a2SYan Zheng 41705d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 41715d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41725d4f98a2SYan Zheng rc->found_file_extent = 1; 41730257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 41743fd0a558SYan, Zheng &key, &rc->cluster); 41755d4f98a2SYan Zheng if (ret < 0) { 41765d4f98a2SYan Zheng err = ret; 41775d4f98a2SYan Zheng break; 41785d4f98a2SYan Zheng } 41795d4f98a2SYan Zheng } 41805d4f98a2SYan Zheng } 4181c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 4182c87f08caSChris Mason ret = btrfs_force_chunk_alloc(trans, rc->extent_root, 4183c87f08caSChris Mason rc->block_group->flags); 41849689457bSShilong Wang if (ret == 1) { 4185c87f08caSChris Mason err = 0; 4186c87f08caSChris Mason progress = 0; 4187c87f08caSChris Mason goto restart; 4188c87f08caSChris Mason } 4189c87f08caSChris Mason } 41903fd0a558SYan, Zheng 4191b3b4aa74SDavid Sterba btrfs_release_path(path); 419291166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41935d4f98a2SYan Zheng 41945d4f98a2SYan Zheng if (trans) { 41953fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 4196b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 41975d4f98a2SYan Zheng } 41985d4f98a2SYan Zheng 41990257bb82SYan, Zheng if (!err) { 42003fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 42013fd0a558SYan, Zheng &rc->cluster); 42020257bb82SYan, Zheng if (ret < 0) 42030257bb82SYan, Zheng err = ret; 42040257bb82SYan, Zheng } 42050257bb82SYan, Zheng 42063fd0a558SYan, Zheng rc->create_reloc_tree = 0; 42073fd0a558SYan, Zheng set_reloc_control(rc); 42080257bb82SYan, Zheng 42093fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 42103fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 42115d4f98a2SYan Zheng 42123fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 42135d4f98a2SYan Zheng 42145d4f98a2SYan Zheng merge_reloc_roots(rc); 42155d4f98a2SYan Zheng 42163fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 42175d4f98a2SYan Zheng unset_reloc_control(rc); 42183fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 42195d4f98a2SYan Zheng 42205d4f98a2SYan Zheng /* get rid of pinned extents */ 42217a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 422262b99540SQu Wenruo if (IS_ERR(trans)) { 42233612b495STsutomu Itoh err = PTR_ERR(trans); 422462b99540SQu Wenruo goto out_free; 422562b99540SQu Wenruo } 4226a9b1fc85SLiu Bo ret = qgroup_fix_relocated_data_extents(trans, rc); 4227a9b1fc85SLiu Bo if (ret < 0) { 4228a9b1fc85SLiu Bo btrfs_abort_transaction(trans, ret); 4229a9b1fc85SLiu Bo if (!err) 4230a9b1fc85SLiu Bo err = ret; 423162b99540SQu Wenruo goto out_free; 423262b99540SQu Wenruo } 42335d4f98a2SYan Zheng btrfs_commit_transaction(trans, rc->extent_root); 42343fd0a558SYan, Zheng out_free: 42353fd0a558SYan, Zheng btrfs_free_block_rsv(rc->extent_root, rc->block_rsv); 42363fd0a558SYan, Zheng btrfs_free_path(path); 42375d4f98a2SYan Zheng return err; 42385d4f98a2SYan Zheng } 42395d4f98a2SYan Zheng 42405d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 42410257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 42425d4f98a2SYan Zheng { 42435d4f98a2SYan Zheng struct btrfs_path *path; 42445d4f98a2SYan Zheng struct btrfs_inode_item *item; 42455d4f98a2SYan Zheng struct extent_buffer *leaf; 42465d4f98a2SYan Zheng int ret; 42475d4f98a2SYan Zheng 42485d4f98a2SYan Zheng path = btrfs_alloc_path(); 42495d4f98a2SYan Zheng if (!path) 42505d4f98a2SYan Zheng return -ENOMEM; 42515d4f98a2SYan Zheng 42525d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 42535d4f98a2SYan Zheng if (ret) 42545d4f98a2SYan Zheng goto out; 42555d4f98a2SYan Zheng 42565d4f98a2SYan Zheng leaf = path->nodes[0]; 42575d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4258b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 42595d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 42600257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 42615d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 42623fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 42633fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 42645d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 42655d4f98a2SYan Zheng out: 42665d4f98a2SYan Zheng btrfs_free_path(path); 42675d4f98a2SYan Zheng return ret; 42685d4f98a2SYan Zheng } 42695d4f98a2SYan Zheng 42705d4f98a2SYan Zheng /* 42715d4f98a2SYan Zheng * helper to create inode for data relocation. 42725d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 42735d4f98a2SYan Zheng */ 42743fd0a558SYan, Zheng static noinline_for_stack 42753fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 42765d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 42775d4f98a2SYan Zheng { 42785d4f98a2SYan Zheng struct inode *inode = NULL; 42795d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42805d4f98a2SYan Zheng struct btrfs_root *root; 42815d4f98a2SYan Zheng struct btrfs_key key; 42824624900dSZhaolei u64 objectid; 42835d4f98a2SYan Zheng int err = 0; 42845d4f98a2SYan Zheng 42855d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 42865d4f98a2SYan Zheng if (IS_ERR(root)) 42875d4f98a2SYan Zheng return ERR_CAST(root); 42885d4f98a2SYan Zheng 4289a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 42903fd0a558SYan, Zheng if (IS_ERR(trans)) 42913fd0a558SYan, Zheng return ERR_CAST(trans); 42925d4f98a2SYan Zheng 4293581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42945d4f98a2SYan Zheng if (err) 42955d4f98a2SYan Zheng goto out; 42965d4f98a2SYan Zheng 42970257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42985d4f98a2SYan Zheng BUG_ON(err); 42995d4f98a2SYan Zheng 43005d4f98a2SYan Zheng key.objectid = objectid; 43015d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 43025d4f98a2SYan Zheng key.offset = 0; 430373f73415SJosef Bacik inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); 43045d4f98a2SYan Zheng BUG_ON(IS_ERR(inode) || is_bad_inode(inode)); 43055d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 43065d4f98a2SYan Zheng 43075d4f98a2SYan Zheng err = btrfs_orphan_add(trans, inode); 43085d4f98a2SYan Zheng out: 43095d4f98a2SYan Zheng btrfs_end_transaction(trans, root); 4310b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 43115d4f98a2SYan Zheng if (err) { 43125d4f98a2SYan Zheng if (inode) 43135d4f98a2SYan Zheng iput(inode); 43145d4f98a2SYan Zheng inode = ERR_PTR(err); 43155d4f98a2SYan Zheng } 43165d4f98a2SYan Zheng return inode; 43175d4f98a2SYan Zheng } 43185d4f98a2SYan Zheng 4319a9995eecSJosef Bacik static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 43203fd0a558SYan, Zheng { 43213fd0a558SYan, Zheng struct reloc_control *rc; 43223fd0a558SYan, Zheng 43233fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 43243fd0a558SYan, Zheng if (!rc) 43253fd0a558SYan, Zheng return NULL; 43263fd0a558SYan, Zheng 43273fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 43283fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 43293fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 4330a9995eecSJosef Bacik extent_io_tree_init(&rc->processed_blocks, 4331a9995eecSJosef Bacik fs_info->btree_inode->i_mapping); 43323fd0a558SYan, Zheng return rc; 43333fd0a558SYan, Zheng } 43343fd0a558SYan, Zheng 43355d4f98a2SYan Zheng /* 4336ebce0e01SAdam Borowski * Print the block group being relocated 4337ebce0e01SAdam Borowski */ 4338ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 4339ebce0e01SAdam Borowski struct btrfs_block_group_cache *block_group) 4340ebce0e01SAdam Borowski { 4341ebce0e01SAdam Borowski char buf[128]; /* prefixed by a '|' that'll be dropped */ 4342ebce0e01SAdam Borowski u64 flags = block_group->flags; 4343ebce0e01SAdam Borowski 4344ebce0e01SAdam Borowski /* Shouldn't happen */ 4345ebce0e01SAdam Borowski if (!flags) { 4346ebce0e01SAdam Borowski strcpy(buf, "|NONE"); 4347ebce0e01SAdam Borowski } else { 4348ebce0e01SAdam Borowski char *bp = buf; 4349ebce0e01SAdam Borowski 4350ebce0e01SAdam Borowski #define DESCRIBE_FLAG(f, d) \ 4351ebce0e01SAdam Borowski if (flags & BTRFS_BLOCK_GROUP_##f) { \ 4352ebce0e01SAdam Borowski bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \ 4353ebce0e01SAdam Borowski flags &= ~BTRFS_BLOCK_GROUP_##f; \ 4354ebce0e01SAdam Borowski } 4355ebce0e01SAdam Borowski DESCRIBE_FLAG(DATA, "data"); 4356ebce0e01SAdam Borowski DESCRIBE_FLAG(SYSTEM, "system"); 4357ebce0e01SAdam Borowski DESCRIBE_FLAG(METADATA, "metadata"); 4358ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID0, "raid0"); 4359ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID1, "raid1"); 4360ebce0e01SAdam Borowski DESCRIBE_FLAG(DUP, "dup"); 4361ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID10, "raid10"); 4362ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID5, "raid5"); 4363ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID6, "raid6"); 4364ebce0e01SAdam Borowski if (flags) 4365ebce0e01SAdam Borowski snprintf(buf, buf - bp + sizeof(buf), "|0x%llx", flags); 4366ebce0e01SAdam Borowski #undef DESCRIBE_FLAG 4367ebce0e01SAdam Borowski } 4368ebce0e01SAdam Borowski 4369ebce0e01SAdam Borowski btrfs_info(fs_info, 4370ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4371ebce0e01SAdam Borowski block_group->key.objectid, buf + 1); 4372ebce0e01SAdam Borowski } 4373ebce0e01SAdam Borowski 4374ebce0e01SAdam Borowski /* 43755d4f98a2SYan Zheng * function to relocate all extents in a block group. 43765d4f98a2SYan Zheng */ 43775d4f98a2SYan Zheng int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start) 43785d4f98a2SYan Zheng { 43795d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = extent_root->fs_info; 43805d4f98a2SYan Zheng struct reloc_control *rc; 43810af3d00bSJosef Bacik struct inode *inode; 43820af3d00bSJosef Bacik struct btrfs_path *path; 43835d4f98a2SYan Zheng int ret; 4384f0486c68SYan, Zheng int rw = 0; 43855d4f98a2SYan Zheng int err = 0; 43865d4f98a2SYan Zheng 4387a9995eecSJosef Bacik rc = alloc_reloc_control(fs_info); 43885d4f98a2SYan Zheng if (!rc) 43895d4f98a2SYan Zheng return -ENOMEM; 43905d4f98a2SYan Zheng 4391f0486c68SYan, Zheng rc->extent_root = extent_root; 43923fd0a558SYan, Zheng 43935d4f98a2SYan Zheng rc->block_group = btrfs_lookup_block_group(fs_info, group_start); 43945d4f98a2SYan Zheng BUG_ON(!rc->block_group); 43955d4f98a2SYan Zheng 4396868f401aSZhaolei ret = btrfs_inc_block_group_ro(extent_root, rc->block_group); 4397f0486c68SYan, Zheng if (ret) { 4398f0486c68SYan, Zheng err = ret; 4399f0486c68SYan, Zheng goto out; 4400f0486c68SYan, Zheng } 4401f0486c68SYan, Zheng rw = 1; 4402f0486c68SYan, Zheng 44030af3d00bSJosef Bacik path = btrfs_alloc_path(); 44040af3d00bSJosef Bacik if (!path) { 44050af3d00bSJosef Bacik err = -ENOMEM; 44060af3d00bSJosef Bacik goto out; 44070af3d00bSJosef Bacik } 44080af3d00bSJosef Bacik 44090af3d00bSJosef Bacik inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group, 44100af3d00bSJosef Bacik path); 44110af3d00bSJosef Bacik btrfs_free_path(path); 44120af3d00bSJosef Bacik 44130af3d00bSJosef Bacik if (!IS_ERR(inode)) 44141bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 44150af3d00bSJosef Bacik else 44160af3d00bSJosef Bacik ret = PTR_ERR(inode); 44170af3d00bSJosef Bacik 44180af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 44190af3d00bSJosef Bacik err = ret; 44200af3d00bSJosef Bacik goto out; 44210af3d00bSJosef Bacik } 44220af3d00bSJosef Bacik 44235d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 44245d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 44255d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 44265d4f98a2SYan Zheng rc->data_inode = NULL; 44275d4f98a2SYan Zheng goto out; 44285d4f98a2SYan Zheng } 44295d4f98a2SYan Zheng 4430ebce0e01SAdam Borowski describe_relocation(extent_root->fs_info, rc->block_group); 44315d4f98a2SYan Zheng 44329cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4433f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 4434578def7cSFilipe Manana btrfs_wait_ordered_roots(fs_info, -1, 4435578def7cSFilipe Manana rc->block_group->key.objectid, 4436578def7cSFilipe Manana rc->block_group->key.offset); 44375d4f98a2SYan Zheng 44385d4f98a2SYan Zheng while (1) { 443976dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 44405d4f98a2SYan Zheng ret = relocate_block_group(rc); 444176dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 44425d4f98a2SYan Zheng if (ret < 0) { 44435d4f98a2SYan Zheng err = ret; 44443fd0a558SYan, Zheng goto out; 44455d4f98a2SYan Zheng } 44465d4f98a2SYan Zheng 44475d4f98a2SYan Zheng if (rc->extents_found == 0) 44485d4f98a2SYan Zheng break; 44495d4f98a2SYan Zheng 4450efe120a0SFrank Holton btrfs_info(extent_root->fs_info, "found %llu extents", 4451c1c9ff7cSGeert Uytterhoeven rc->extents_found); 44525d4f98a2SYan Zheng 44535d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 44540ef8b726SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 44550ef8b726SJosef Bacik (u64)-1); 44560ef8b726SJosef Bacik if (ret) { 44570ef8b726SJosef Bacik err = ret; 44580ef8b726SJosef Bacik goto out; 44590ef8b726SJosef Bacik } 44605d4f98a2SYan Zheng invalidate_mapping_pages(rc->data_inode->i_mapping, 44615d4f98a2SYan Zheng 0, -1); 44625d4f98a2SYan Zheng rc->stage = UPDATE_DATA_PTRS; 44635d4f98a2SYan Zheng } 44645d4f98a2SYan Zheng } 44655d4f98a2SYan Zheng 44665d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 44675d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 44685d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 44695d4f98a2SYan Zheng out: 4470f0486c68SYan, Zheng if (err && rw) 4471868f401aSZhaolei btrfs_dec_block_group_ro(extent_root, rc->block_group); 44725d4f98a2SYan Zheng iput(rc->data_inode); 44735d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 44745d4f98a2SYan Zheng kfree(rc); 44755d4f98a2SYan Zheng return err; 44765d4f98a2SYan Zheng } 44775d4f98a2SYan Zheng 447876dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 447976dda93cSYan, Zheng { 448076dda93cSYan, Zheng struct btrfs_trans_handle *trans; 448179787eaaSJeff Mahoney int ret, err; 448276dda93cSYan, Zheng 4483a22285a6SYan, Zheng trans = btrfs_start_transaction(root->fs_info->tree_root, 0); 448479787eaaSJeff Mahoney if (IS_ERR(trans)) 448579787eaaSJeff Mahoney return PTR_ERR(trans); 448676dda93cSYan, Zheng 448776dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 448876dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 448976dda93cSYan, Zheng root->root_item.drop_level = 0; 449076dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 449176dda93cSYan, Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 449276dda93cSYan, Zheng &root->root_key, &root->root_item); 449376dda93cSYan, Zheng 449479787eaaSJeff Mahoney err = btrfs_end_transaction(trans, root->fs_info->tree_root); 449579787eaaSJeff Mahoney if (err) 449679787eaaSJeff Mahoney return err; 449779787eaaSJeff Mahoney return ret; 449876dda93cSYan, Zheng } 449976dda93cSYan, Zheng 45005d4f98a2SYan Zheng /* 45015d4f98a2SYan Zheng * recover relocation interrupted by system crash. 45025d4f98a2SYan Zheng * 45035d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 45045d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 45055d4f98a2SYan Zheng */ 45065d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 45075d4f98a2SYan Zheng { 45085d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 45095d4f98a2SYan Zheng struct btrfs_key key; 45105d4f98a2SYan Zheng struct btrfs_root *fs_root; 45115d4f98a2SYan Zheng struct btrfs_root *reloc_root; 45125d4f98a2SYan Zheng struct btrfs_path *path; 45135d4f98a2SYan Zheng struct extent_buffer *leaf; 45145d4f98a2SYan Zheng struct reloc_control *rc = NULL; 45155d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 45165d4f98a2SYan Zheng int ret; 45175d4f98a2SYan Zheng int err = 0; 45185d4f98a2SYan Zheng 45195d4f98a2SYan Zheng path = btrfs_alloc_path(); 45205d4f98a2SYan Zheng if (!path) 45215d4f98a2SYan Zheng return -ENOMEM; 4522e4058b54SDavid Sterba path->reada = READA_BACK; 45235d4f98a2SYan Zheng 45245d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 45255d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 45265d4f98a2SYan Zheng key.offset = (u64)-1; 45275d4f98a2SYan Zheng 45285d4f98a2SYan Zheng while (1) { 45295d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, 45305d4f98a2SYan Zheng path, 0, 0); 45315d4f98a2SYan Zheng if (ret < 0) { 45325d4f98a2SYan Zheng err = ret; 45335d4f98a2SYan Zheng goto out; 45345d4f98a2SYan Zheng } 45355d4f98a2SYan Zheng if (ret > 0) { 45365d4f98a2SYan Zheng if (path->slots[0] == 0) 45375d4f98a2SYan Zheng break; 45385d4f98a2SYan Zheng path->slots[0]--; 45395d4f98a2SYan Zheng } 45405d4f98a2SYan Zheng leaf = path->nodes[0]; 45415d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4542b3b4aa74SDavid Sterba btrfs_release_path(path); 45435d4f98a2SYan Zheng 45445d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 45455d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 45465d4f98a2SYan Zheng break; 45475d4f98a2SYan Zheng 4548cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 45495d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 45505d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 45515d4f98a2SYan Zheng goto out; 45525d4f98a2SYan Zheng } 45535d4f98a2SYan Zheng 45545d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 45555d4f98a2SYan Zheng 45565d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 45575d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 45585d4f98a2SYan Zheng reloc_root->root_key.offset); 45595d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 456076dda93cSYan, Zheng ret = PTR_ERR(fs_root); 456176dda93cSYan, Zheng if (ret != -ENOENT) { 456276dda93cSYan, Zheng err = ret; 45635d4f98a2SYan Zheng goto out; 45645d4f98a2SYan Zheng } 456579787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 456679787eaaSJeff Mahoney if (ret < 0) { 456779787eaaSJeff Mahoney err = ret; 456879787eaaSJeff Mahoney goto out; 456979787eaaSJeff Mahoney } 457076dda93cSYan, Zheng } 45715d4f98a2SYan Zheng } 45725d4f98a2SYan Zheng 45735d4f98a2SYan Zheng if (key.offset == 0) 45745d4f98a2SYan Zheng break; 45755d4f98a2SYan Zheng 45765d4f98a2SYan Zheng key.offset--; 45775d4f98a2SYan Zheng } 4578b3b4aa74SDavid Sterba btrfs_release_path(path); 45795d4f98a2SYan Zheng 45805d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 45815d4f98a2SYan Zheng goto out; 45825d4f98a2SYan Zheng 4583a9995eecSJosef Bacik rc = alloc_reloc_control(root->fs_info); 45845d4f98a2SYan Zheng if (!rc) { 45855d4f98a2SYan Zheng err = -ENOMEM; 45865d4f98a2SYan Zheng goto out; 45875d4f98a2SYan Zheng } 45885d4f98a2SYan Zheng 45895d4f98a2SYan Zheng rc->extent_root = root->fs_info->extent_root; 45905d4f98a2SYan Zheng 45915d4f98a2SYan Zheng set_reloc_control(rc); 45925d4f98a2SYan Zheng 45937a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45943612b495STsutomu Itoh if (IS_ERR(trans)) { 45953612b495STsutomu Itoh unset_reloc_control(rc); 45963612b495STsutomu Itoh err = PTR_ERR(trans); 45973612b495STsutomu Itoh goto out_free; 45983612b495STsutomu Itoh } 45993fd0a558SYan, Zheng 46003fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 46013fd0a558SYan, Zheng 46025d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 46035d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 46045d4f98a2SYan Zheng struct btrfs_root, root_list); 46055d4f98a2SYan Zheng list_del(&reloc_root->root_list); 46065d4f98a2SYan Zheng 46075d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 46085d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 46095d4f98a2SYan Zheng &rc->reloc_roots); 46105d4f98a2SYan Zheng continue; 46115d4f98a2SYan Zheng } 46125d4f98a2SYan Zheng 46135d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 46145d4f98a2SYan Zheng reloc_root->root_key.offset); 461579787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 461679787eaaSJeff Mahoney err = PTR_ERR(fs_root); 461779787eaaSJeff Mahoney goto out_free; 461879787eaaSJeff Mahoney } 46195d4f98a2SYan Zheng 4620ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 462179787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 46225d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 46235d4f98a2SYan Zheng } 46245d4f98a2SYan Zheng 462579787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 462679787eaaSJeff Mahoney if (err) 462779787eaaSJeff Mahoney goto out_free; 46285d4f98a2SYan Zheng 46295d4f98a2SYan Zheng merge_reloc_roots(rc); 46305d4f98a2SYan Zheng 46315d4f98a2SYan Zheng unset_reloc_control(rc); 46325d4f98a2SYan Zheng 46337a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 463462b99540SQu Wenruo if (IS_ERR(trans)) { 46353612b495STsutomu Itoh err = PTR_ERR(trans); 463662b99540SQu Wenruo goto out_free; 463762b99540SQu Wenruo } 463862b99540SQu Wenruo err = qgroup_fix_relocated_data_extents(trans, rc); 463962b99540SQu Wenruo if (err < 0) { 464062b99540SQu Wenruo btrfs_abort_transaction(trans, err); 464162b99540SQu Wenruo goto out_free; 464262b99540SQu Wenruo } 464379787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 46443612b495STsutomu Itoh out_free: 46455d4f98a2SYan Zheng kfree(rc); 46463612b495STsutomu Itoh out: 4647aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4648aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4649aca1bba6SLiu Bo 46505d4f98a2SYan Zheng btrfs_free_path(path); 46515d4f98a2SYan Zheng 46525d4f98a2SYan Zheng if (err == 0) { 46535d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 46545d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 46555d4f98a2SYan Zheng BTRFS_DATA_RELOC_TREE_OBJECTID); 46565d4f98a2SYan Zheng if (IS_ERR(fs_root)) 46575d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4658d7ce5843SMiao Xie else 465966b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 46605d4f98a2SYan Zheng } 46615d4f98a2SYan Zheng return err; 46625d4f98a2SYan Zheng } 46635d4f98a2SYan Zheng 46645d4f98a2SYan Zheng /* 46655d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 46665d4f98a2SYan Zheng * 46675d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 46685d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 46695d4f98a2SYan Zheng */ 46705d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 46715d4f98a2SYan Zheng { 46725d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 46735d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 46745d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 46755d4f98a2SYan Zheng int ret; 46765d4f98a2SYan Zheng u64 disk_bytenr; 46774577b014SJosef Bacik u64 new_bytenr; 46785d4f98a2SYan Zheng LIST_HEAD(list); 46795d4f98a2SYan Zheng 46805d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 46815d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 46825d4f98a2SYan Zheng 46835d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46845d4f98a2SYan Zheng ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr, 4685a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 468679787eaaSJeff Mahoney if (ret) 468779787eaaSJeff Mahoney goto out; 46885d4f98a2SYan Zheng 46895d4f98a2SYan Zheng while (!list_empty(&list)) { 46905d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46915d4f98a2SYan Zheng list_del_init(&sums->list); 46925d4f98a2SYan Zheng 46934577b014SJosef Bacik /* 46944577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46954577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46964577b014SJosef Bacik * extents but we may have written to say the middle of the 46974577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46984577b014SJosef Bacik * the right disk offset. 46994577b014SJosef Bacik * 47004577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 47014577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 47024577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 47034577b014SJosef Bacik * disk length. 47044577b014SJosef Bacik */ 47054577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 47064577b014SJosef Bacik sums->bytenr = new_bytenr; 47075d4f98a2SYan Zheng 47085d4f98a2SYan Zheng btrfs_add_ordered_sum(inode, ordered, sums); 47095d4f98a2SYan Zheng } 471079787eaaSJeff Mahoney out: 47115d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4712411fc6bcSAndi Kleen return ret; 47135d4f98a2SYan Zheng } 47143fd0a558SYan, Zheng 471583d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 47163fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 47173fd0a558SYan, Zheng struct extent_buffer *cow) 47183fd0a558SYan, Zheng { 47193fd0a558SYan, Zheng struct reloc_control *rc; 47203fd0a558SYan, Zheng struct backref_node *node; 47213fd0a558SYan, Zheng int first_cow = 0; 47223fd0a558SYan, Zheng int level; 472383d4cfd4SJosef Bacik int ret = 0; 47243fd0a558SYan, Zheng 47253fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47263fd0a558SYan, Zheng if (!rc) 472783d4cfd4SJosef Bacik return 0; 47283fd0a558SYan, Zheng 47293fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 47303fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 47313fd0a558SYan, Zheng 4732c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4733c974c464SWang Shilong if (buf == root->node) 4734c974c464SWang Shilong __update_reloc_root(root, cow->start); 4735c974c464SWang Shilong } 4736c974c464SWang Shilong 47373fd0a558SYan, Zheng level = btrfs_header_level(buf); 47383fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 47393fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 47403fd0a558SYan, Zheng first_cow = 1; 47413fd0a558SYan, Zheng 47423fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 47433fd0a558SYan, Zheng rc->create_reloc_tree) { 47443fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 47453fd0a558SYan, Zheng 47463fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 47473fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 47483fd0a558SYan, Zheng node->new_bytenr != buf->start); 47493fd0a558SYan, Zheng 47503fd0a558SYan, Zheng drop_node_buffer(node); 47513fd0a558SYan, Zheng extent_buffer_get(cow); 47523fd0a558SYan, Zheng node->eb = cow; 47533fd0a558SYan, Zheng node->new_bytenr = cow->start; 47543fd0a558SYan, Zheng 47553fd0a558SYan, Zheng if (!node->pending) { 47563fd0a558SYan, Zheng list_move_tail(&node->list, 47573fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 47583fd0a558SYan, Zheng node->pending = 1; 47593fd0a558SYan, Zheng } 47603fd0a558SYan, Zheng 47613fd0a558SYan, Zheng if (first_cow) 47623fd0a558SYan, Zheng __mark_block_processed(rc, node); 47633fd0a558SYan, Zheng 47643fd0a558SYan, Zheng if (first_cow && level > 0) 47653fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 47663fd0a558SYan, Zheng } 47673fd0a558SYan, Zheng 476883d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 47693fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 477083d4cfd4SJosef Bacik return ret; 47713fd0a558SYan, Zheng } 47723fd0a558SYan, Zheng 47733fd0a558SYan, Zheng /* 47743fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 477501327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 47763fd0a558SYan, Zheng */ 4777147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 47783fd0a558SYan, Zheng u64 *bytes_to_reserve) 47793fd0a558SYan, Zheng { 47803fd0a558SYan, Zheng struct btrfs_root *root; 47813fd0a558SYan, Zheng struct reloc_control *rc; 47823fd0a558SYan, Zheng 47833fd0a558SYan, Zheng root = pending->root; 47843fd0a558SYan, Zheng if (!root->reloc_root) 47853fd0a558SYan, Zheng return; 47863fd0a558SYan, Zheng 47873fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47883fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47893fd0a558SYan, Zheng return; 47903fd0a558SYan, Zheng 47913fd0a558SYan, Zheng root = root->reloc_root; 47923fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47933fd0a558SYan, Zheng /* 47943fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47953fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47963fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47973fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47983fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47993fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 48003fd0a558SYan, Zheng * tree may use more space than it frees. so we need 48013fd0a558SYan, Zheng * reserve extra space. 48023fd0a558SYan, Zheng */ 48033fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 48043fd0a558SYan, Zheng } 48053fd0a558SYan, Zheng 48063fd0a558SYan, Zheng /* 48073fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 48083fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 48093fd0a558SYan, Zheng */ 481049b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 48113fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 48123fd0a558SYan, Zheng { 48133fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 48143fd0a558SYan, Zheng struct btrfs_root *reloc_root; 48153fd0a558SYan, Zheng struct btrfs_root *new_root; 48163fd0a558SYan, Zheng struct reloc_control *rc; 48173fd0a558SYan, Zheng int ret; 48183fd0a558SYan, Zheng 48193fd0a558SYan, Zheng if (!root->reloc_root) 482049b25e05SJeff Mahoney return 0; 48213fd0a558SYan, Zheng 48223fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 48233fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 48243fd0a558SYan, Zheng 48253fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 48263fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 48273fd0a558SYan, Zheng rc->block_rsv, 482825d609f8SJosef Bacik rc->nodes_relocated, 1); 482949b25e05SJeff Mahoney if (ret) 483049b25e05SJeff Mahoney return ret; 48313fd0a558SYan, Zheng } 48323fd0a558SYan, Zheng 48333fd0a558SYan, Zheng new_root = pending->snap; 48343fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 48353fd0a558SYan, Zheng new_root->root_key.objectid); 483649b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 483749b25e05SJeff Mahoney return PTR_ERR(reloc_root); 48383fd0a558SYan, Zheng 4839ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4840ffd7b339SJeff Mahoney BUG_ON(ret < 0); 48413fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 48423fd0a558SYan, Zheng 484349b25e05SJeff Mahoney if (rc->create_reloc_tree) 48443fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 484549b25e05SJeff Mahoney return ret; 48463fd0a558SYan, Zheng } 4847