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; 34043c04fb1SJeff Mahoney btrfs_panic(fs_info, errno, "Inconsistency in backref cache " 341351fd353SDavid Sterba "found at offset %llu", bytenr); 34243c04fb1SJeff Mahoney } 34343c04fb1SJeff Mahoney 3445d4f98a2SYan Zheng /* 3455d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3465d4f98a2SYan Zheng */ 3475d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3485d4f98a2SYan Zheng struct backref_edge *edges[], 3495d4f98a2SYan Zheng int *index) 3505d4f98a2SYan Zheng { 3515d4f98a2SYan Zheng struct backref_edge *edge; 3525d4f98a2SYan Zheng int idx = *index; 3535d4f98a2SYan Zheng 3545d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3555d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3565d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3575d4f98a2SYan Zheng edges[idx++] = edge; 3585d4f98a2SYan Zheng node = edge->node[UPPER]; 3595d4f98a2SYan Zheng } 3603fd0a558SYan, Zheng BUG_ON(node->detached); 3615d4f98a2SYan Zheng *index = idx; 3625d4f98a2SYan Zheng return node; 3635d4f98a2SYan Zheng } 3645d4f98a2SYan Zheng 3655d4f98a2SYan Zheng /* 3665d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 3675d4f98a2SYan Zheng */ 3685d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 3695d4f98a2SYan Zheng int *index) 3705d4f98a2SYan Zheng { 3715d4f98a2SYan Zheng struct backref_edge *edge; 3725d4f98a2SYan Zheng struct backref_node *lower; 3735d4f98a2SYan Zheng int idx = *index; 3745d4f98a2SYan Zheng 3755d4f98a2SYan Zheng while (idx > 0) { 3765d4f98a2SYan Zheng edge = edges[idx - 1]; 3775d4f98a2SYan Zheng lower = edge->node[LOWER]; 3785d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 3795d4f98a2SYan Zheng idx--; 3805d4f98a2SYan Zheng continue; 3815d4f98a2SYan Zheng } 3825d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 3835d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3845d4f98a2SYan Zheng edges[idx - 1] = edge; 3855d4f98a2SYan Zheng *index = idx; 3865d4f98a2SYan Zheng return edge->node[UPPER]; 3875d4f98a2SYan Zheng } 3885d4f98a2SYan Zheng *index = 0; 3895d4f98a2SYan Zheng return NULL; 3905d4f98a2SYan Zheng } 3915d4f98a2SYan Zheng 3923fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 3935d4f98a2SYan Zheng { 3945d4f98a2SYan Zheng if (node->locked) { 3955d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 3965d4f98a2SYan Zheng node->locked = 0; 3975d4f98a2SYan Zheng } 3983fd0a558SYan, Zheng } 3993fd0a558SYan, Zheng 4003fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 4013fd0a558SYan, Zheng { 4023fd0a558SYan, Zheng if (node->eb) { 4033fd0a558SYan, Zheng unlock_node_buffer(node); 4045d4f98a2SYan Zheng free_extent_buffer(node->eb); 4055d4f98a2SYan Zheng node->eb = NULL; 4065d4f98a2SYan Zheng } 4075d4f98a2SYan Zheng } 4085d4f98a2SYan Zheng 4095d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4105d4f98a2SYan Zheng struct backref_node *node) 4115d4f98a2SYan Zheng { 4125d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4135d4f98a2SYan Zheng 4145d4f98a2SYan Zheng drop_node_buffer(node); 4153fd0a558SYan, Zheng list_del(&node->list); 4165d4f98a2SYan Zheng list_del(&node->lower); 4173fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4185d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4193fd0a558SYan, Zheng free_backref_node(tree, node); 4205d4f98a2SYan Zheng } 4215d4f98a2SYan Zheng 4225d4f98a2SYan Zheng /* 4235d4f98a2SYan Zheng * remove a backref node from the backref cache 4245d4f98a2SYan Zheng */ 4255d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4265d4f98a2SYan Zheng struct backref_node *node) 4275d4f98a2SYan Zheng { 4285d4f98a2SYan Zheng struct backref_node *upper; 4295d4f98a2SYan Zheng struct backref_edge *edge; 4305d4f98a2SYan Zheng 4315d4f98a2SYan Zheng if (!node) 4325d4f98a2SYan Zheng return; 4335d4f98a2SYan Zheng 4343fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4355d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4365d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4375d4f98a2SYan Zheng list[LOWER]); 4385d4f98a2SYan Zheng upper = edge->node[UPPER]; 4395d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4405d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4413fd0a558SYan, Zheng free_backref_edge(cache, edge); 4423fd0a558SYan, Zheng 4433fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4443fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4453fd0a558SYan, Zheng drop_backref_node(cache, node); 4463fd0a558SYan, Zheng node = upper; 4473fd0a558SYan, Zheng node->lowest = 1; 4483fd0a558SYan, Zheng continue; 4493fd0a558SYan, Zheng } 4505d4f98a2SYan Zheng /* 4513fd0a558SYan, Zheng * add the node to leaf node list if no other 4525d4f98a2SYan Zheng * child block cached. 4535d4f98a2SYan Zheng */ 4545d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4553fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4565d4f98a2SYan Zheng upper->lowest = 1; 4575d4f98a2SYan Zheng } 4585d4f98a2SYan Zheng } 4593fd0a558SYan, Zheng 4605d4f98a2SYan Zheng drop_backref_node(cache, node); 4615d4f98a2SYan Zheng } 4625d4f98a2SYan Zheng 4633fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 4643fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 4653fd0a558SYan, Zheng { 4663fd0a558SYan, Zheng struct rb_node *rb_node; 4673fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 4683fd0a558SYan, Zheng node->bytenr = bytenr; 4693fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 47043c04fb1SJeff Mahoney if (rb_node) 47143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 4723fd0a558SYan, Zheng } 4733fd0a558SYan, Zheng 4743fd0a558SYan, Zheng /* 4753fd0a558SYan, Zheng * update backref cache after a transaction commit 4763fd0a558SYan, Zheng */ 4773fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 4783fd0a558SYan, Zheng struct backref_cache *cache) 4793fd0a558SYan, Zheng { 4803fd0a558SYan, Zheng struct backref_node *node; 4813fd0a558SYan, Zheng int level = 0; 4823fd0a558SYan, Zheng 4833fd0a558SYan, Zheng if (cache->last_trans == 0) { 4843fd0a558SYan, Zheng cache->last_trans = trans->transid; 4853fd0a558SYan, Zheng return 0; 4863fd0a558SYan, Zheng } 4873fd0a558SYan, Zheng 4883fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 4893fd0a558SYan, Zheng return 0; 4903fd0a558SYan, Zheng 4913fd0a558SYan, Zheng /* 4923fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 4933fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 4943fd0a558SYan, Zheng * so the detached nodes are no longer useful. 4953fd0a558SYan, Zheng */ 4963fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 4973fd0a558SYan, Zheng node = list_entry(cache->detached.next, 4983fd0a558SYan, Zheng struct backref_node, list); 4993fd0a558SYan, Zheng remove_backref_node(cache, node); 5003fd0a558SYan, Zheng } 5013fd0a558SYan, Zheng 5023fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 5033fd0a558SYan, Zheng node = list_entry(cache->changed.next, 5043fd0a558SYan, Zheng struct backref_node, list); 5053fd0a558SYan, Zheng list_del_init(&node->list); 5063fd0a558SYan, Zheng BUG_ON(node->pending); 5073fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5083fd0a558SYan, Zheng } 5093fd0a558SYan, Zheng 5103fd0a558SYan, Zheng /* 5113fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5123fd0a558SYan, Zheng * errors during processing the pending nodes. 5133fd0a558SYan, Zheng */ 5143fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5153fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5163fd0a558SYan, Zheng BUG_ON(!node->pending); 5173fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5183fd0a558SYan, Zheng continue; 5193fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5203fd0a558SYan, Zheng } 5213fd0a558SYan, Zheng } 5223fd0a558SYan, Zheng 5233fd0a558SYan, Zheng cache->last_trans = 0; 5243fd0a558SYan, Zheng return 1; 5253fd0a558SYan, Zheng } 5263fd0a558SYan, Zheng 527f2a97a9dSDavid Sterba 5283fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5293fd0a558SYan, Zheng { 5303fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5313fd0a558SYan, Zheng 53227cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 5333fd0a558SYan, Zheng return 0; 5343fd0a558SYan, Zheng 5353fd0a558SYan, Zheng reloc_root = root->reloc_root; 5363fd0a558SYan, Zheng if (!reloc_root) 5373fd0a558SYan, Zheng return 0; 5383fd0a558SYan, Zheng 5393fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5403fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5413fd0a558SYan, Zheng return 0; 5423fd0a558SYan, Zheng /* 5433fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5443fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5453fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5463fd0a558SYan, Zheng * relocation. 5473fd0a558SYan, Zheng */ 5483fd0a558SYan, Zheng return 1; 5493fd0a558SYan, Zheng } 5505d4f98a2SYan Zheng /* 5515d4f98a2SYan Zheng * find reloc tree by address of tree root 5525d4f98a2SYan Zheng */ 5535d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5545d4f98a2SYan Zheng u64 bytenr) 5555d4f98a2SYan Zheng { 5565d4f98a2SYan Zheng struct rb_node *rb_node; 5575d4f98a2SYan Zheng struct mapping_node *node; 5585d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5595d4f98a2SYan Zheng 5605d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5615d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5625d4f98a2SYan Zheng if (rb_node) { 5635d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5645d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5655d4f98a2SYan Zheng } 5665d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5675d4f98a2SYan Zheng return root; 5685d4f98a2SYan Zheng } 5695d4f98a2SYan Zheng 5705d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5715d4f98a2SYan Zheng { 5725d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5735d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5745d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 5755d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 5765d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 57766463748SWang Shilong root_objectid == BTRFS_CSUM_TREE_OBJECTID || 57866463748SWang Shilong root_objectid == BTRFS_UUID_TREE_OBJECTID || 5793e4c5efbSDavid Sterba root_objectid == BTRFS_QUOTA_TREE_OBJECTID || 5803e4c5efbSDavid Sterba root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 5815d4f98a2SYan Zheng return 1; 5825d4f98a2SYan Zheng return 0; 5835d4f98a2SYan Zheng } 5845d4f98a2SYan Zheng 5855d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 5865d4f98a2SYan Zheng u64 root_objectid) 5875d4f98a2SYan Zheng { 5885d4f98a2SYan Zheng struct btrfs_key key; 5895d4f98a2SYan Zheng 5905d4f98a2SYan Zheng key.objectid = root_objectid; 5915d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 5925d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 5935d4f98a2SYan Zheng key.offset = 0; 5945d4f98a2SYan Zheng else 5955d4f98a2SYan Zheng key.offset = (u64)-1; 5965d4f98a2SYan Zheng 597c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 5985d4f98a2SYan Zheng } 5995d4f98a2SYan Zheng 6005d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 6015d4f98a2SYan Zheng static noinline_for_stack 6025d4f98a2SYan Zheng struct btrfs_root *find_tree_root(struct reloc_control *rc, 6035d4f98a2SYan Zheng struct extent_buffer *leaf, 6045d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0) 6055d4f98a2SYan Zheng { 6065d4f98a2SYan Zheng struct btrfs_root *root; 6075d4f98a2SYan Zheng u64 root_objectid = btrfs_ref_root_v0(leaf, ref0); 6085d4f98a2SYan Zheng u64 generation = btrfs_ref_generation_v0(leaf, ref0); 6095d4f98a2SYan Zheng 6105d4f98a2SYan Zheng BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID); 6115d4f98a2SYan Zheng 6125d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, root_objectid); 6135d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 6145d4f98a2SYan Zheng 61527cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && 6165d4f98a2SYan Zheng generation != btrfs_root_generation(&root->root_item)) 6175d4f98a2SYan Zheng return NULL; 6185d4f98a2SYan Zheng 6195d4f98a2SYan Zheng return root; 6205d4f98a2SYan Zheng } 6215d4f98a2SYan Zheng #endif 6225d4f98a2SYan Zheng 6235d4f98a2SYan Zheng static noinline_for_stack 6245d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 6255d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 6265d4f98a2SYan Zheng { 6273173a18fSJosef Bacik struct btrfs_key key; 6285d4f98a2SYan Zheng struct btrfs_extent_item *ei; 6295d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6305d4f98a2SYan Zheng u32 item_size; 6315d4f98a2SYan Zheng 6323173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6333173a18fSJosef Bacik 6345d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 6355d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 6365d4f98a2SYan Zheng if (item_size < sizeof(*ei)) { 6375d4f98a2SYan Zheng WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 6385d4f98a2SYan Zheng return 1; 6395d4f98a2SYan Zheng } 6405d4f98a2SYan Zheng #endif 6415d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6425d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6435d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6445d4f98a2SYan Zheng 6453173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6463173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6475d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6485d4f98a2SYan Zheng return 1; 6495d4f98a2SYan Zheng } 650d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 651d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 652d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 653d062d13cSJosef Bacik return 1; 654d062d13cSJosef Bacik } 6555d4f98a2SYan Zheng 6563173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6575d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6585d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6593173a18fSJosef Bacik } else { 6603173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6613173a18fSJosef Bacik } 6625d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6635d4f98a2SYan Zheng return 0; 6645d4f98a2SYan Zheng } 6655d4f98a2SYan Zheng 6665d4f98a2SYan Zheng /* 6675d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6685d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6695d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6705d4f98a2SYan Zheng * 6715d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 67201327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 67301327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 6745d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6755d4f98a2SYan Zheng * 6765d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6775d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6785d4f98a2SYan Zheng * block are also cached. 6795d4f98a2SYan Zheng */ 6803fd0a558SYan, Zheng static noinline_for_stack 6813fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6825d4f98a2SYan Zheng struct btrfs_key *node_key, 6835d4f98a2SYan Zheng int level, u64 bytenr) 6845d4f98a2SYan Zheng { 6853fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 6865d4f98a2SYan Zheng struct btrfs_path *path1; 6875d4f98a2SYan Zheng struct btrfs_path *path2; 6885d4f98a2SYan Zheng struct extent_buffer *eb; 6895d4f98a2SYan Zheng struct btrfs_root *root; 6905d4f98a2SYan Zheng struct backref_node *cur; 6915d4f98a2SYan Zheng struct backref_node *upper; 6925d4f98a2SYan Zheng struct backref_node *lower; 6935d4f98a2SYan Zheng struct backref_node *node = NULL; 6945d4f98a2SYan Zheng struct backref_node *exist = NULL; 6955d4f98a2SYan Zheng struct backref_edge *edge; 6965d4f98a2SYan Zheng struct rb_node *rb_node; 6975d4f98a2SYan Zheng struct btrfs_key key; 6985d4f98a2SYan Zheng unsigned long end; 6995d4f98a2SYan Zheng unsigned long ptr; 7005d4f98a2SYan Zheng LIST_HEAD(list); 7013fd0a558SYan, Zheng LIST_HEAD(useless); 7023fd0a558SYan, Zheng int cowonly; 7035d4f98a2SYan Zheng int ret; 7045d4f98a2SYan Zheng int err = 0; 705b6c60c80SJosef Bacik bool need_check = true; 7065d4f98a2SYan Zheng 7075d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 7085d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 7095d4f98a2SYan Zheng if (!path1 || !path2) { 7105d4f98a2SYan Zheng err = -ENOMEM; 7115d4f98a2SYan Zheng goto out; 7125d4f98a2SYan Zheng } 713e4058b54SDavid Sterba path1->reada = READA_FORWARD; 714e4058b54SDavid Sterba path2->reada = READA_FORWARD; 7155d4f98a2SYan Zheng 7163fd0a558SYan, Zheng node = alloc_backref_node(cache); 7175d4f98a2SYan Zheng if (!node) { 7185d4f98a2SYan Zheng err = -ENOMEM; 7195d4f98a2SYan Zheng goto out; 7205d4f98a2SYan Zheng } 7215d4f98a2SYan Zheng 7225d4f98a2SYan Zheng node->bytenr = bytenr; 7235d4f98a2SYan Zheng node->level = level; 7245d4f98a2SYan Zheng node->lowest = 1; 7255d4f98a2SYan Zheng cur = node; 7265d4f98a2SYan Zheng again: 7275d4f98a2SYan Zheng end = 0; 7285d4f98a2SYan Zheng ptr = 0; 7295d4f98a2SYan Zheng key.objectid = cur->bytenr; 7303173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7315d4f98a2SYan Zheng key.offset = (u64)-1; 7325d4f98a2SYan Zheng 7335d4f98a2SYan Zheng path1->search_commit_root = 1; 7345d4f98a2SYan Zheng path1->skip_locking = 1; 7355d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7365d4f98a2SYan Zheng 0, 0); 7375d4f98a2SYan Zheng if (ret < 0) { 7385d4f98a2SYan Zheng err = ret; 7395d4f98a2SYan Zheng goto out; 7405d4f98a2SYan Zheng } 74175bfb9afSJosef Bacik ASSERT(ret); 74275bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7435d4f98a2SYan Zheng 7445d4f98a2SYan Zheng path1->slots[0]--; 7455d4f98a2SYan Zheng 7465d4f98a2SYan Zheng WARN_ON(cur->checked); 7475d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7485d4f98a2SYan Zheng /* 74970f23fd6SJustin P. Mattock * the backref was added previously when processing 7505d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7515d4f98a2SYan Zheng */ 75275bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7535d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7545d4f98a2SYan Zheng list[LOWER]); 75575bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7565d4f98a2SYan Zheng exist = edge->node[UPPER]; 7575d4f98a2SYan Zheng /* 7585d4f98a2SYan Zheng * add the upper level block to pending list if we need 7595d4f98a2SYan Zheng * check its backrefs 7605d4f98a2SYan Zheng */ 7615d4f98a2SYan Zheng if (!exist->checked) 7625d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7635d4f98a2SYan Zheng } else { 7645d4f98a2SYan Zheng exist = NULL; 7655d4f98a2SYan Zheng } 7665d4f98a2SYan Zheng 7675d4f98a2SYan Zheng while (1) { 7685d4f98a2SYan Zheng cond_resched(); 7695d4f98a2SYan Zheng eb = path1->nodes[0]; 7705d4f98a2SYan Zheng 7715d4f98a2SYan Zheng if (ptr >= end) { 7725d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7735d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7745d4f98a2SYan Zheng if (ret < 0) { 7755d4f98a2SYan Zheng err = ret; 7765d4f98a2SYan Zheng goto out; 7775d4f98a2SYan Zheng } 7785d4f98a2SYan Zheng if (ret > 0) 7795d4f98a2SYan Zheng break; 7805d4f98a2SYan Zheng eb = path1->nodes[0]; 7815d4f98a2SYan Zheng } 7825d4f98a2SYan Zheng 7835d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7845d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7855d4f98a2SYan Zheng WARN_ON(exist); 7865d4f98a2SYan Zheng break; 7875d4f98a2SYan Zheng } 7885d4f98a2SYan Zheng 7893173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7903173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7915d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7925d4f98a2SYan Zheng &ptr, &end); 7935d4f98a2SYan Zheng if (ret) 7945d4f98a2SYan Zheng goto next; 7955d4f98a2SYan Zheng } 7965d4f98a2SYan Zheng } 7975d4f98a2SYan Zheng 7985d4f98a2SYan Zheng if (ptr < end) { 7995d4f98a2SYan Zheng /* update key for inline back ref */ 8005d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 8015d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 8025d4f98a2SYan Zheng key.type = btrfs_extent_inline_ref_type(eb, iref); 8035d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 8045d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 8055d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 8065d4f98a2SYan Zheng } 8075d4f98a2SYan Zheng 8085d4f98a2SYan Zheng if (exist && 8095d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8105d4f98a2SYan Zheng exist->owner == key.offset) || 8115d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8125d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8135d4f98a2SYan Zheng exist = NULL; 8145d4f98a2SYan Zheng goto next; 8155d4f98a2SYan Zheng } 8165d4f98a2SYan Zheng 8175d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 8185d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY || 8195d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 8203fd0a558SYan, Zheng if (key.type == BTRFS_EXTENT_REF_V0_KEY) { 8215d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 8225d4f98a2SYan Zheng ref0 = btrfs_item_ptr(eb, path1->slots[0], 8235d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 8243fd0a558SYan, Zheng if (key.objectid == key.offset) { 825046f264fSYan, Zheng root = find_tree_root(rc, eb, ref0); 8263fd0a558SYan, Zheng if (root && !should_ignore_root(root)) 8275d4f98a2SYan Zheng cur->root = root; 8285d4f98a2SYan Zheng else 8293fd0a558SYan, Zheng list_add(&cur->list, &useless); 8305d4f98a2SYan Zheng break; 8315d4f98a2SYan Zheng } 832046f264fSYan, Zheng if (is_cowonly_root(btrfs_ref_root_v0(eb, 833046f264fSYan, Zheng ref0))) 834046f264fSYan, Zheng cur->cowonly = 1; 8353fd0a558SYan, Zheng } 8365d4f98a2SYan Zheng #else 83775bfb9afSJosef Bacik ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY); 8385d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8395d4f98a2SYan Zheng #endif 8405d4f98a2SYan Zheng if (key.objectid == key.offset) { 8415d4f98a2SYan Zheng /* 8425d4f98a2SYan Zheng * only root blocks of reloc trees use 8435d4f98a2SYan Zheng * backref of this type. 8445d4f98a2SYan Zheng */ 8455d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 84675bfb9afSJosef Bacik ASSERT(root); 8475d4f98a2SYan Zheng cur->root = root; 8485d4f98a2SYan Zheng break; 8495d4f98a2SYan Zheng } 8505d4f98a2SYan Zheng 8513fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8525d4f98a2SYan Zheng if (!edge) { 8535d4f98a2SYan Zheng err = -ENOMEM; 8545d4f98a2SYan Zheng goto out; 8555d4f98a2SYan Zheng } 8565d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8575d4f98a2SYan Zheng if (!rb_node) { 8583fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8595d4f98a2SYan Zheng if (!upper) { 8603fd0a558SYan, Zheng free_backref_edge(cache, edge); 8615d4f98a2SYan Zheng err = -ENOMEM; 8625d4f98a2SYan Zheng goto out; 8635d4f98a2SYan Zheng } 8645d4f98a2SYan Zheng upper->bytenr = key.offset; 8655d4f98a2SYan Zheng upper->level = cur->level + 1; 8665d4f98a2SYan Zheng /* 8675d4f98a2SYan Zheng * backrefs for the upper level block isn't 8685d4f98a2SYan Zheng * cached, add the block to pending list 8695d4f98a2SYan Zheng */ 8705d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8715d4f98a2SYan Zheng } else { 8725d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8735d4f98a2SYan Zheng rb_node); 87475bfb9afSJosef Bacik ASSERT(upper->checked); 8755d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8765d4f98a2SYan Zheng } 8773fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8785d4f98a2SYan Zheng edge->node[LOWER] = cur; 8793fd0a558SYan, Zheng edge->node[UPPER] = upper; 8805d4f98a2SYan Zheng 8815d4f98a2SYan Zheng goto next; 8825d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8835d4f98a2SYan Zheng goto next; 8845d4f98a2SYan Zheng } 8855d4f98a2SYan Zheng 8865d4f98a2SYan Zheng /* key.type == BTRFS_TREE_BLOCK_REF_KEY */ 8875d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8885d4f98a2SYan Zheng if (IS_ERR(root)) { 8895d4f98a2SYan Zheng err = PTR_ERR(root); 8905d4f98a2SYan Zheng goto out; 8915d4f98a2SYan Zheng } 8925d4f98a2SYan Zheng 89327cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 8943fd0a558SYan, Zheng cur->cowonly = 1; 8953fd0a558SYan, Zheng 8965d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 8975d4f98a2SYan Zheng /* tree root */ 89875bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 8995d4f98a2SYan Zheng cur->bytenr); 9003fd0a558SYan, Zheng if (should_ignore_root(root)) 9013fd0a558SYan, Zheng list_add(&cur->list, &useless); 9023fd0a558SYan, Zheng else 9035d4f98a2SYan Zheng cur->root = root; 9045d4f98a2SYan Zheng break; 9055d4f98a2SYan Zheng } 9065d4f98a2SYan Zheng 9075d4f98a2SYan Zheng level = cur->level + 1; 9085d4f98a2SYan Zheng 9095d4f98a2SYan Zheng /* 9105d4f98a2SYan Zheng * searching the tree to find upper level blocks 9115d4f98a2SYan Zheng * reference the block. 9125d4f98a2SYan Zheng */ 9135d4f98a2SYan Zheng path2->search_commit_root = 1; 9145d4f98a2SYan Zheng path2->skip_locking = 1; 9155d4f98a2SYan Zheng path2->lowest_level = level; 9165d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 9175d4f98a2SYan Zheng path2->lowest_level = 0; 9185d4f98a2SYan Zheng if (ret < 0) { 9195d4f98a2SYan Zheng err = ret; 9205d4f98a2SYan Zheng goto out; 9215d4f98a2SYan Zheng } 92233c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 92333c66f43SYan Zheng path2->slots[level]--; 9245d4f98a2SYan Zheng 9255d4f98a2SYan Zheng eb = path2->nodes[level]; 9265d4f98a2SYan Zheng WARN_ON(btrfs_node_blockptr(eb, path2->slots[level]) != 9275d4f98a2SYan Zheng cur->bytenr); 9285d4f98a2SYan Zheng 9295d4f98a2SYan Zheng lower = cur; 930b6c60c80SJosef Bacik need_check = true; 9315d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9325d4f98a2SYan Zheng if (!path2->nodes[level]) { 93375bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9345d4f98a2SYan Zheng lower->bytenr); 9353fd0a558SYan, Zheng if (should_ignore_root(root)) 9363fd0a558SYan, Zheng list_add(&lower->list, &useless); 9373fd0a558SYan, Zheng else 9385d4f98a2SYan Zheng lower->root = root; 9395d4f98a2SYan Zheng break; 9405d4f98a2SYan Zheng } 9415d4f98a2SYan Zheng 9423fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9435d4f98a2SYan Zheng if (!edge) { 9445d4f98a2SYan Zheng err = -ENOMEM; 9455d4f98a2SYan Zheng goto out; 9465d4f98a2SYan Zheng } 9475d4f98a2SYan Zheng 9485d4f98a2SYan Zheng eb = path2->nodes[level]; 9495d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9505d4f98a2SYan Zheng if (!rb_node) { 9513fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9525d4f98a2SYan Zheng if (!upper) { 9533fd0a558SYan, Zheng free_backref_edge(cache, edge); 9545d4f98a2SYan Zheng err = -ENOMEM; 9555d4f98a2SYan Zheng goto out; 9565d4f98a2SYan Zheng } 9575d4f98a2SYan Zheng upper->bytenr = eb->start; 9585d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9595d4f98a2SYan Zheng upper->level = lower->level + 1; 96027cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 96127cdeb70SMiao Xie &root->state)) 9623fd0a558SYan, Zheng upper->cowonly = 1; 9635d4f98a2SYan Zheng 9645d4f98a2SYan Zheng /* 9655d4f98a2SYan Zheng * if we know the block isn't shared 9665d4f98a2SYan Zheng * we can void checking its backrefs. 9675d4f98a2SYan Zheng */ 9685d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9695d4f98a2SYan Zheng upper->checked = 0; 9705d4f98a2SYan Zheng else 9715d4f98a2SYan Zheng upper->checked = 1; 9725d4f98a2SYan Zheng 9735d4f98a2SYan Zheng /* 9745d4f98a2SYan Zheng * add the block to pending list if we 975b6c60c80SJosef Bacik * need check its backrefs, we only do this once 976b6c60c80SJosef Bacik * while walking up a tree as we will catch 977b6c60c80SJosef Bacik * anything else later on. 9785d4f98a2SYan Zheng */ 979b6c60c80SJosef Bacik if (!upper->checked && need_check) { 980b6c60c80SJosef Bacik need_check = false; 9815d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9825d4f98a2SYan Zheng &list); 983bbe90514SJosef Bacik } else { 984bbe90514SJosef Bacik if (upper->checked) 985bbe90514SJosef Bacik need_check = true; 9865d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 987bbe90514SJosef Bacik } 9885d4f98a2SYan Zheng } else { 9895d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9905d4f98a2SYan Zheng rb_node); 99175bfb9afSJosef Bacik ASSERT(upper->checked); 9925d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9933fd0a558SYan, Zheng if (!upper->owner) 9943fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 9955d4f98a2SYan Zheng } 9965d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 9975d4f98a2SYan Zheng edge->node[LOWER] = lower; 9983fd0a558SYan, Zheng edge->node[UPPER] = upper; 9995d4f98a2SYan Zheng 10005d4f98a2SYan Zheng if (rb_node) 10015d4f98a2SYan Zheng break; 10025d4f98a2SYan Zheng lower = upper; 10035d4f98a2SYan Zheng upper = NULL; 10045d4f98a2SYan Zheng } 1005b3b4aa74SDavid Sterba btrfs_release_path(path2); 10065d4f98a2SYan Zheng next: 10075d4f98a2SYan Zheng if (ptr < end) { 10085d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 10095d4f98a2SYan Zheng if (ptr >= end) { 10105d4f98a2SYan Zheng WARN_ON(ptr > end); 10115d4f98a2SYan Zheng ptr = 0; 10125d4f98a2SYan Zheng end = 0; 10135d4f98a2SYan Zheng } 10145d4f98a2SYan Zheng } 10155d4f98a2SYan Zheng if (ptr >= end) 10165d4f98a2SYan Zheng path1->slots[0]++; 10175d4f98a2SYan Zheng } 1018b3b4aa74SDavid Sterba btrfs_release_path(path1); 10195d4f98a2SYan Zheng 10205d4f98a2SYan Zheng cur->checked = 1; 10215d4f98a2SYan Zheng WARN_ON(exist); 10225d4f98a2SYan Zheng 10235d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10245d4f98a2SYan Zheng if (!list_empty(&list)) { 10255d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10265d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10275d4f98a2SYan Zheng cur = edge->node[UPPER]; 10285d4f98a2SYan Zheng goto again; 10295d4f98a2SYan Zheng } 10305d4f98a2SYan Zheng 10315d4f98a2SYan Zheng /* 10325d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10335d4f98a2SYan Zheng * into the cache. 10345d4f98a2SYan Zheng */ 103575bfb9afSJosef Bacik ASSERT(node->checked); 10363fd0a558SYan, Zheng cowonly = node->cowonly; 10373fd0a558SYan, Zheng if (!cowonly) { 10383fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10393fd0a558SYan, Zheng &node->rb_node); 104043c04fb1SJeff Mahoney if (rb_node) 104143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10423fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10433fd0a558SYan, Zheng } 10445d4f98a2SYan Zheng 10455d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10465d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10475d4f98a2SYan Zheng 10485d4f98a2SYan Zheng while (!list_empty(&list)) { 10495d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10505d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10515d4f98a2SYan Zheng upper = edge->node[UPPER]; 10523fd0a558SYan, Zheng if (upper->detached) { 10533fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10543fd0a558SYan, Zheng lower = edge->node[LOWER]; 10553fd0a558SYan, Zheng free_backref_edge(cache, edge); 10563fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10573fd0a558SYan, Zheng list_add(&lower->list, &useless); 10583fd0a558SYan, Zheng continue; 10593fd0a558SYan, Zheng } 10605d4f98a2SYan Zheng 10615d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10625d4f98a2SYan Zheng if (upper->lowest) { 10635d4f98a2SYan Zheng list_del_init(&upper->lower); 10645d4f98a2SYan Zheng upper->lowest = 0; 10655d4f98a2SYan Zheng } 10665d4f98a2SYan Zheng 10675d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10685d4f98a2SYan Zheng continue; 10695d4f98a2SYan Zheng } 10705d4f98a2SYan Zheng 107175bfb9afSJosef Bacik if (!upper->checked) { 107275bfb9afSJosef Bacik /* 107375bfb9afSJosef Bacik * Still want to blow up for developers since this is a 107475bfb9afSJosef Bacik * logic bug. 107575bfb9afSJosef Bacik */ 107675bfb9afSJosef Bacik ASSERT(0); 107775bfb9afSJosef Bacik err = -EINVAL; 107875bfb9afSJosef Bacik goto out; 107975bfb9afSJosef Bacik } 108075bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 108175bfb9afSJosef Bacik ASSERT(0); 108275bfb9afSJosef Bacik err = -EINVAL; 108375bfb9afSJosef Bacik goto out; 108475bfb9afSJosef Bacik } 108575bfb9afSJosef Bacik 10863fd0a558SYan, Zheng if (!cowonly) { 10875d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10885d4f98a2SYan Zheng &upper->rb_node); 108943c04fb1SJeff Mahoney if (rb_node) 109043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 109143c04fb1SJeff Mahoney upper->bytenr); 10923fd0a558SYan, Zheng } 10935d4f98a2SYan Zheng 10945d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10955d4f98a2SYan Zheng 10965d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 10975d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10985d4f98a2SYan Zheng } 10993fd0a558SYan, Zheng /* 11003fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 11013fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 11023fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 11033fd0a558SYan, Zheng * lookup. 11043fd0a558SYan, Zheng */ 11053fd0a558SYan, Zheng while (!list_empty(&useless)) { 11063fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 11073fd0a558SYan, Zheng list_del_init(&upper->list); 110875bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 11093fd0a558SYan, Zheng if (upper == node) 11103fd0a558SYan, Zheng node = NULL; 11113fd0a558SYan, Zheng if (upper->lowest) { 11123fd0a558SYan, Zheng list_del_init(&upper->lower); 11133fd0a558SYan, Zheng upper->lowest = 0; 11143fd0a558SYan, Zheng } 11153fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 11163fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 11173fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 11183fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11193fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11203fd0a558SYan, Zheng lower = edge->node[LOWER]; 11213fd0a558SYan, Zheng free_backref_edge(cache, edge); 11223fd0a558SYan, Zheng 11233fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11243fd0a558SYan, Zheng list_add(&lower->list, &useless); 11253fd0a558SYan, Zheng } 11263fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11273fd0a558SYan, Zheng if (upper->level > 0) { 11283fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11293fd0a558SYan, Zheng upper->detached = 1; 11303fd0a558SYan, Zheng } else { 11313fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11323fd0a558SYan, Zheng free_backref_node(cache, upper); 11333fd0a558SYan, Zheng } 11343fd0a558SYan, Zheng } 11355d4f98a2SYan Zheng out: 11365d4f98a2SYan Zheng btrfs_free_path(path1); 11375d4f98a2SYan Zheng btrfs_free_path(path2); 11385d4f98a2SYan Zheng if (err) { 11393fd0a558SYan, Zheng while (!list_empty(&useless)) { 11403fd0a558SYan, Zheng lower = list_entry(useless.next, 114175bfb9afSJosef Bacik struct backref_node, list); 114275bfb9afSJosef Bacik list_del_init(&lower->list); 11433fd0a558SYan, Zheng } 114475bfb9afSJosef Bacik while (!list_empty(&list)) { 114575bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 114675bfb9afSJosef Bacik list[UPPER]); 114775bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11483fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 114975bfb9afSJosef Bacik lower = edge->node[LOWER]; 11505d4f98a2SYan Zheng upper = edge->node[UPPER]; 11513fd0a558SYan, Zheng free_backref_edge(cache, edge); 115275bfb9afSJosef Bacik 115375bfb9afSJosef Bacik /* 115475bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 115575bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 115675bfb9afSJosef Bacik */ 115775bfb9afSJosef Bacik if (list_empty(&lower->upper) && 115875bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 115975bfb9afSJosef Bacik list_add(&lower->list, &useless); 116075bfb9afSJosef Bacik 116175bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 116275bfb9afSJosef Bacik continue; 116375bfb9afSJosef Bacik 116401327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 116575bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 116675bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 116775bfb9afSJosef Bacik if (list_empty(&upper->upper)) 116875bfb9afSJosef Bacik list_add(&upper->list, &useless); 116975bfb9afSJosef Bacik } 117075bfb9afSJosef Bacik 117175bfb9afSJosef Bacik while (!list_empty(&useless)) { 117275bfb9afSJosef Bacik lower = list_entry(useless.next, 117375bfb9afSJosef Bacik struct backref_node, list); 117475bfb9afSJosef Bacik list_del_init(&lower->list); 11750fd8c3daSLiu Bo if (lower == node) 11760fd8c3daSLiu Bo node = NULL; 117775bfb9afSJosef Bacik free_backref_node(cache, lower); 11785d4f98a2SYan Zheng } 11790fd8c3daSLiu Bo 11800fd8c3daSLiu Bo free_backref_node(cache, node); 11815d4f98a2SYan Zheng return ERR_PTR(err); 11825d4f98a2SYan Zheng } 118375bfb9afSJosef Bacik ASSERT(!node || !node->detached); 11845d4f98a2SYan Zheng return node; 11855d4f98a2SYan Zheng } 11865d4f98a2SYan Zheng 11875d4f98a2SYan Zheng /* 11883fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11893fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11903fd0a558SYan, Zheng * corresponds to root of source tree 11913fd0a558SYan, Zheng */ 11923fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 11933fd0a558SYan, Zheng struct reloc_control *rc, 11943fd0a558SYan, Zheng struct btrfs_root *src, 11953fd0a558SYan, Zheng struct btrfs_root *dest) 11963fd0a558SYan, Zheng { 11973fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 11983fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 11993fd0a558SYan, Zheng struct backref_node *node = NULL; 12003fd0a558SYan, Zheng struct backref_node *new_node; 12013fd0a558SYan, Zheng struct backref_edge *edge; 12023fd0a558SYan, Zheng struct backref_edge *new_edge; 12033fd0a558SYan, Zheng struct rb_node *rb_node; 12043fd0a558SYan, Zheng 12053fd0a558SYan, Zheng if (cache->last_trans > 0) 12063fd0a558SYan, Zheng update_backref_cache(trans, cache); 12073fd0a558SYan, Zheng 12083fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 12093fd0a558SYan, Zheng if (rb_node) { 12103fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 12113fd0a558SYan, Zheng if (node->detached) 12123fd0a558SYan, Zheng node = NULL; 12133fd0a558SYan, Zheng else 12143fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 12153fd0a558SYan, Zheng } 12163fd0a558SYan, Zheng 12173fd0a558SYan, Zheng if (!node) { 12183fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 12193fd0a558SYan, Zheng reloc_root->commit_root->start); 12203fd0a558SYan, Zheng if (rb_node) { 12213fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 12223fd0a558SYan, Zheng rb_node); 12233fd0a558SYan, Zheng BUG_ON(node->detached); 12243fd0a558SYan, Zheng } 12253fd0a558SYan, Zheng } 12263fd0a558SYan, Zheng 12273fd0a558SYan, Zheng if (!node) 12283fd0a558SYan, Zheng return 0; 12293fd0a558SYan, Zheng 12303fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12313fd0a558SYan, Zheng if (!new_node) 12323fd0a558SYan, Zheng return -ENOMEM; 12333fd0a558SYan, Zheng 12343fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12353fd0a558SYan, Zheng new_node->level = node->level; 12363fd0a558SYan, Zheng new_node->lowest = node->lowest; 12376848ad64SYan, Zheng new_node->checked = 1; 12383fd0a558SYan, Zheng new_node->root = dest; 12393fd0a558SYan, Zheng 12403fd0a558SYan, Zheng if (!node->lowest) { 12413fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12423fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12433fd0a558SYan, Zheng if (!new_edge) 12443fd0a558SYan, Zheng goto fail; 12453fd0a558SYan, Zheng 12463fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12473fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12483fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12493fd0a558SYan, Zheng &new_node->lower); 12503fd0a558SYan, Zheng } 125176b9e23dSMiao Xie } else { 125276b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12533fd0a558SYan, Zheng } 12543fd0a558SYan, Zheng 12553fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12563fd0a558SYan, Zheng &new_node->rb_node); 125743c04fb1SJeff Mahoney if (rb_node) 125843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12593fd0a558SYan, Zheng 12603fd0a558SYan, Zheng if (!new_node->lowest) { 12613fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12623fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12633fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12643fd0a558SYan, Zheng } 12653fd0a558SYan, Zheng } 12663fd0a558SYan, Zheng return 0; 12673fd0a558SYan, Zheng fail: 12683fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12693fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12703fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12713fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12723fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12733fd0a558SYan, Zheng } 12743fd0a558SYan, Zheng free_backref_node(cache, new_node); 12753fd0a558SYan, Zheng return -ENOMEM; 12763fd0a558SYan, Zheng } 12773fd0a558SYan, Zheng 12783fd0a558SYan, Zheng /* 12795d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12805d4f98a2SYan Zheng */ 1281ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12825d4f98a2SYan Zheng { 12835d4f98a2SYan Zheng struct rb_node *rb_node; 12845d4f98a2SYan Zheng struct mapping_node *node; 12855d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 12865d4f98a2SYan Zheng 12875d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1288ffd7b339SJeff Mahoney if (!node) 1289ffd7b339SJeff Mahoney return -ENOMEM; 12905d4f98a2SYan Zheng 12915d4f98a2SYan Zheng node->bytenr = root->node->start; 12925d4f98a2SYan Zheng node->data = root; 12935d4f98a2SYan Zheng 12945d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12955d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 12965d4f98a2SYan Zheng node->bytenr, &node->rb_node); 12975d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1298ffd7b339SJeff Mahoney if (rb_node) { 1299ffd7b339SJeff Mahoney btrfs_panic(root->fs_info, -EEXIST, "Duplicate root found " 1300ffd7b339SJeff Mahoney "for start=%llu while inserting into relocation " 1301351fd353SDavid Sterba "tree", node->bytenr); 130223291a04SDan Carpenter kfree(node); 130323291a04SDan Carpenter return -EEXIST; 1304ffd7b339SJeff Mahoney } 13055d4f98a2SYan Zheng 13065d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 13075d4f98a2SYan Zheng return 0; 13085d4f98a2SYan Zheng } 13095d4f98a2SYan Zheng 13105d4f98a2SYan Zheng /* 1311c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 13125d4f98a2SYan Zheng * mapping 13135d4f98a2SYan Zheng */ 1314c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 13155d4f98a2SYan Zheng { 13165d4f98a2SYan Zheng struct rb_node *rb_node; 13175d4f98a2SYan Zheng struct mapping_node *node = NULL; 13185d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 13195d4f98a2SYan Zheng 13205d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13215d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1322c974c464SWang Shilong root->node->start); 1323c974c464SWang Shilong if (rb_node) { 1324c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1325c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1326c974c464SWang Shilong } 1327c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1328c974c464SWang Shilong 1329c974c464SWang Shilong if (!node) 1330c974c464SWang Shilong return; 1331c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1332c974c464SWang Shilong 1333c974c464SWang Shilong spin_lock(&root->fs_info->trans_lock); 1334c974c464SWang Shilong list_del_init(&root->root_list); 1335c974c464SWang Shilong spin_unlock(&root->fs_info->trans_lock); 1336c974c464SWang Shilong kfree(node); 1337c974c464SWang Shilong } 1338c974c464SWang Shilong 1339c974c464SWang Shilong /* 1340c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1341c974c464SWang Shilong * mapping 1342c974c464SWang Shilong */ 1343c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1344c974c464SWang Shilong { 1345c974c464SWang Shilong struct rb_node *rb_node; 1346c974c464SWang Shilong struct mapping_node *node = NULL; 1347c974c464SWang Shilong struct reloc_control *rc = root->fs_info->reloc_ctl; 1348c974c464SWang Shilong 1349c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1350c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1351c974c464SWang Shilong root->node->start); 13525d4f98a2SYan Zheng if (rb_node) { 13535d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13545d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13555d4f98a2SYan Zheng } 13565d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13575d4f98a2SYan Zheng 13588f71f3e0SLiu Bo if (!node) 13598f71f3e0SLiu Bo return 0; 13605d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13615d4f98a2SYan Zheng 13625d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1363c974c464SWang Shilong node->bytenr = new_bytenr; 13645d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13655d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13665d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 136743c04fb1SJeff Mahoney if (rb_node) 136843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13695d4f98a2SYan Zheng return 0; 13705d4f98a2SYan Zheng } 13715d4f98a2SYan Zheng 13723fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13733fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13745d4f98a2SYan Zheng { 13755d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13765d4f98a2SYan Zheng struct extent_buffer *eb; 13775d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13785d4f98a2SYan Zheng struct btrfs_key root_key; 13795bc7247aSMiao Xie u64 last_snap = 0; 13805d4f98a2SYan Zheng int ret; 13815d4f98a2SYan Zheng 13825d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13835d4f98a2SYan Zheng BUG_ON(!root_item); 13845d4f98a2SYan Zheng 13855d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13865d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13873fd0a558SYan, Zheng root_key.offset = objectid; 13885d4f98a2SYan Zheng 13893fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 13903fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 13915d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 13925d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 13935d4f98a2SYan Zheng BUG_ON(ret); 13945d4f98a2SYan Zheng 13955bc7247aSMiao Xie last_snap = btrfs_root_last_snapshot(&root->root_item); 13963fd0a558SYan, Zheng btrfs_set_root_last_snapshot(&root->root_item, 13973fd0a558SYan, Zheng trans->transid - 1); 13983fd0a558SYan, Zheng } else { 13993fd0a558SYan, Zheng /* 14003fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 14013fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 14023fd0a558SYan, Zheng * modified after it was created have RELOC flag 14033fd0a558SYan, Zheng * set in their headers. so it's OK to not update 14043fd0a558SYan, Zheng * the 'last_snapshot'. 14053fd0a558SYan, Zheng */ 14063fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 14073fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 14083fd0a558SYan, Zheng BUG_ON(ret); 14093fd0a558SYan, Zheng } 14103fd0a558SYan, Zheng 14115d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 14125d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14135d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14145d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14153fd0a558SYan, Zheng 14163fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14173fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14183fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14193fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14205d4f98a2SYan Zheng root_item->drop_level = 0; 14215bc7247aSMiao Xie /* 14225bc7247aSMiao Xie * abuse rtransid, it is safe because it is impossible to 14235bc7247aSMiao Xie * receive data into a relocation tree. 14245bc7247aSMiao Xie */ 14255bc7247aSMiao Xie btrfs_set_root_rtransid(root_item, last_snap); 14265bc7247aSMiao Xie btrfs_set_root_otransid(root_item, trans->transid); 14273fd0a558SYan, Zheng } 14285d4f98a2SYan Zheng 14295d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14305d4f98a2SYan Zheng free_extent_buffer(eb); 14315d4f98a2SYan Zheng 14325d4f98a2SYan Zheng ret = btrfs_insert_root(trans, root->fs_info->tree_root, 14335d4f98a2SYan Zheng &root_key, root_item); 14345d4f98a2SYan Zheng BUG_ON(ret); 14355d4f98a2SYan Zheng kfree(root_item); 14365d4f98a2SYan Zheng 1437cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root->fs_info->tree_root, &root_key); 14385d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14395d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14403fd0a558SYan, Zheng return reloc_root; 14413fd0a558SYan, Zheng } 14423fd0a558SYan, Zheng 14433fd0a558SYan, Zheng /* 14443fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 14453fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 14463fd0a558SYan, Zheng */ 14473fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 14483fd0a558SYan, Zheng struct btrfs_root *root) 14493fd0a558SYan, Zheng { 14503fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14513fd0a558SYan, Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 145220dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14533fd0a558SYan, Zheng int clear_rsv = 0; 1454ffd7b339SJeff Mahoney int ret; 14553fd0a558SYan, Zheng 14563fd0a558SYan, Zheng if (root->reloc_root) { 14573fd0a558SYan, Zheng reloc_root = root->reloc_root; 14583fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14593fd0a558SYan, Zheng return 0; 14603fd0a558SYan, Zheng } 14613fd0a558SYan, Zheng 14623fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14633fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14643fd0a558SYan, Zheng return 0; 14653fd0a558SYan, Zheng 146620dd2cbfSMiao Xie if (!trans->reloc_reserved) { 146720dd2cbfSMiao Xie rsv = trans->block_rsv; 14683fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14693fd0a558SYan, Zheng clear_rsv = 1; 14703fd0a558SYan, Zheng } 14713fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14723fd0a558SYan, Zheng if (clear_rsv) 147320dd2cbfSMiao Xie trans->block_rsv = rsv; 14745d4f98a2SYan Zheng 1475ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1476ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14775d4f98a2SYan Zheng root->reloc_root = reloc_root; 14785d4f98a2SYan Zheng return 0; 14795d4f98a2SYan Zheng } 14805d4f98a2SYan Zheng 14815d4f98a2SYan Zheng /* 14825d4f98a2SYan Zheng * update root item of reloc tree 14835d4f98a2SYan Zheng */ 14845d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 14855d4f98a2SYan Zheng struct btrfs_root *root) 14865d4f98a2SYan Zheng { 14875d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14885d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14895d4f98a2SYan Zheng int ret; 14905d4f98a2SYan Zheng 14915d4f98a2SYan Zheng if (!root->reloc_root) 14927585717fSChris Mason goto out; 14935d4f98a2SYan Zheng 14945d4f98a2SYan Zheng reloc_root = root->reloc_root; 14955d4f98a2SYan Zheng root_item = &reloc_root->root_item; 14965d4f98a2SYan Zheng 14973fd0a558SYan, Zheng if (root->fs_info->reloc_ctl->merge_reloc_tree && 14983fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 14995d4f98a2SYan Zheng root->reloc_root = NULL; 1500c974c464SWang Shilong __del_reloc_root(reloc_root); 15015d4f98a2SYan Zheng } 15025d4f98a2SYan Zheng 15035d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 15045d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 15055d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 15065d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 15075d4f98a2SYan Zheng } 15085d4f98a2SYan Zheng 15095d4f98a2SYan Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 15105d4f98a2SYan Zheng &reloc_root->root_key, root_item); 15115d4f98a2SYan Zheng BUG_ON(ret); 15127585717fSChris Mason 15137585717fSChris Mason out: 15145d4f98a2SYan Zheng return 0; 15155d4f98a2SYan Zheng } 15165d4f98a2SYan Zheng 15175d4f98a2SYan Zheng /* 15185d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15195d4f98a2SYan Zheng * in a subvolume 15205d4f98a2SYan Zheng */ 15215d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15225d4f98a2SYan Zheng { 15235d4f98a2SYan Zheng struct rb_node *node; 15245d4f98a2SYan Zheng struct rb_node *prev; 15255d4f98a2SYan Zheng struct btrfs_inode *entry; 15265d4f98a2SYan Zheng struct inode *inode; 15275d4f98a2SYan Zheng 15285d4f98a2SYan Zheng spin_lock(&root->inode_lock); 15295d4f98a2SYan Zheng again: 15305d4f98a2SYan Zheng node = root->inode_tree.rb_node; 15315d4f98a2SYan Zheng prev = NULL; 15325d4f98a2SYan Zheng while (node) { 15335d4f98a2SYan Zheng prev = node; 15345d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15355d4f98a2SYan Zheng 153633345d01SLi Zefan if (objectid < btrfs_ino(&entry->vfs_inode)) 15375d4f98a2SYan Zheng node = node->rb_left; 153833345d01SLi Zefan else if (objectid > btrfs_ino(&entry->vfs_inode)) 15395d4f98a2SYan Zheng node = node->rb_right; 15405d4f98a2SYan Zheng else 15415d4f98a2SYan Zheng break; 15425d4f98a2SYan Zheng } 15435d4f98a2SYan Zheng if (!node) { 15445d4f98a2SYan Zheng while (prev) { 15455d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 154633345d01SLi Zefan if (objectid <= btrfs_ino(&entry->vfs_inode)) { 15475d4f98a2SYan Zheng node = prev; 15485d4f98a2SYan Zheng break; 15495d4f98a2SYan Zheng } 15505d4f98a2SYan Zheng prev = rb_next(prev); 15515d4f98a2SYan Zheng } 15525d4f98a2SYan Zheng } 15535d4f98a2SYan Zheng while (node) { 15545d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15555d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15565d4f98a2SYan Zheng if (inode) { 15575d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15585d4f98a2SYan Zheng return inode; 15595d4f98a2SYan Zheng } 15605d4f98a2SYan Zheng 156133345d01SLi Zefan objectid = btrfs_ino(&entry->vfs_inode) + 1; 15625d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15635d4f98a2SYan Zheng goto again; 15645d4f98a2SYan Zheng 15655d4f98a2SYan Zheng node = rb_next(node); 15665d4f98a2SYan Zheng } 15675d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15685d4f98a2SYan Zheng return NULL; 15695d4f98a2SYan Zheng } 15705d4f98a2SYan Zheng 15715d4f98a2SYan Zheng static int in_block_group(u64 bytenr, 15725d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group) 15735d4f98a2SYan Zheng { 15745d4f98a2SYan Zheng if (bytenr >= block_group->key.objectid && 15755d4f98a2SYan Zheng bytenr < block_group->key.objectid + block_group->key.offset) 15765d4f98a2SYan Zheng return 1; 15775d4f98a2SYan Zheng return 0; 15785d4f98a2SYan Zheng } 15795d4f98a2SYan Zheng 15805d4f98a2SYan Zheng /* 15815d4f98a2SYan Zheng * get new location of data 15825d4f98a2SYan Zheng */ 15835d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 15845d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 15855d4f98a2SYan Zheng { 15865d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 15875d4f98a2SYan Zheng struct btrfs_path *path; 15885d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15895d4f98a2SYan Zheng struct extent_buffer *leaf; 15905d4f98a2SYan Zheng int ret; 15915d4f98a2SYan Zheng 15925d4f98a2SYan Zheng path = btrfs_alloc_path(); 15935d4f98a2SYan Zheng if (!path) 15945d4f98a2SYan Zheng return -ENOMEM; 15955d4f98a2SYan Zheng 15965d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 159733345d01SLi Zefan ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode), 15985d4f98a2SYan Zheng bytenr, 0); 15995d4f98a2SYan Zheng if (ret < 0) 16005d4f98a2SYan Zheng goto out; 16015d4f98a2SYan Zheng if (ret > 0) { 16025d4f98a2SYan Zheng ret = -ENOENT; 16035d4f98a2SYan Zheng goto out; 16045d4f98a2SYan Zheng } 16055d4f98a2SYan Zheng 16065d4f98a2SYan Zheng leaf = path->nodes[0]; 16075d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 16085d4f98a2SYan Zheng struct btrfs_file_extent_item); 16095d4f98a2SYan Zheng 16105d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 16115d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 16125d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 16135d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 16145d4f98a2SYan Zheng 16155d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 161683d4cfd4SJosef Bacik ret = -EINVAL; 16175d4f98a2SYan Zheng goto out; 16185d4f98a2SYan Zheng } 16195d4f98a2SYan Zheng 16205d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16215d4f98a2SYan Zheng ret = 0; 16225d4f98a2SYan Zheng out: 16235d4f98a2SYan Zheng btrfs_free_path(path); 16245d4f98a2SYan Zheng return ret; 16255d4f98a2SYan Zheng } 16265d4f98a2SYan Zheng 16275d4f98a2SYan Zheng /* 16285d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16295d4f98a2SYan Zheng * the new locations. 16305d4f98a2SYan Zheng */ 16313fd0a558SYan, Zheng static noinline_for_stack 16323fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 16335d4f98a2SYan Zheng struct reloc_control *rc, 16345d4f98a2SYan Zheng struct btrfs_root *root, 16353fd0a558SYan, Zheng struct extent_buffer *leaf) 16365d4f98a2SYan Zheng { 16375d4f98a2SYan Zheng struct btrfs_key key; 16385d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16395d4f98a2SYan Zheng struct inode *inode = NULL; 16405d4f98a2SYan Zheng u64 parent; 16415d4f98a2SYan Zheng u64 bytenr; 16423fd0a558SYan, Zheng u64 new_bytenr = 0; 16435d4f98a2SYan Zheng u64 num_bytes; 16445d4f98a2SYan Zheng u64 end; 16455d4f98a2SYan Zheng u32 nritems; 16465d4f98a2SYan Zheng u32 i; 164783d4cfd4SJosef Bacik int ret = 0; 16485d4f98a2SYan Zheng int first = 1; 16495d4f98a2SYan Zheng int dirty = 0; 16505d4f98a2SYan Zheng 16515d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16525d4f98a2SYan Zheng return 0; 16535d4f98a2SYan Zheng 16545d4f98a2SYan Zheng /* reloc trees always use full backref */ 16555d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16565d4f98a2SYan Zheng parent = leaf->start; 16575d4f98a2SYan Zheng else 16585d4f98a2SYan Zheng parent = 0; 16595d4f98a2SYan Zheng 16605d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16615d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 16625d4f98a2SYan Zheng cond_resched(); 16635d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16645d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16655d4f98a2SYan Zheng continue; 16665d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16675d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16685d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16695d4f98a2SYan Zheng continue; 16705d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16715d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16725d4f98a2SYan Zheng if (bytenr == 0) 16735d4f98a2SYan Zheng continue; 16745d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16755d4f98a2SYan Zheng continue; 16765d4f98a2SYan Zheng 16775d4f98a2SYan Zheng /* 16785d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16795d4f98a2SYan Zheng * to complete and drop the extent cache 16805d4f98a2SYan Zheng */ 16815d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16825d4f98a2SYan Zheng if (first) { 16835d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16845d4f98a2SYan Zheng first = 0; 168533345d01SLi Zefan } else if (inode && btrfs_ino(inode) < key.objectid) { 16863fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16875d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16885d4f98a2SYan Zheng } 168933345d01SLi Zefan if (inode && btrfs_ino(inode) == key.objectid) { 16905d4f98a2SYan Zheng end = key.offset + 16915d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 16925d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 16935d4f98a2SYan Zheng root->sectorsize)); 16945d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 16955d4f98a2SYan Zheng end--; 16965d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1697d0082371SJeff Mahoney key.offset, end); 16985d4f98a2SYan Zheng if (!ret) 16995d4f98a2SYan Zheng continue; 17005d4f98a2SYan Zheng 17015d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, key.offset, end, 17025d4f98a2SYan Zheng 1); 17035d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1704d0082371SJeff Mahoney key.offset, end); 17055d4f98a2SYan Zheng } 17065d4f98a2SYan Zheng } 17075d4f98a2SYan Zheng 17085d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 17095d4f98a2SYan Zheng bytenr, num_bytes); 171083d4cfd4SJosef Bacik if (ret) { 171183d4cfd4SJosef Bacik /* 171283d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 171383d4cfd4SJosef Bacik * in the file extent yet. 171483d4cfd4SJosef Bacik */ 171583d4cfd4SJosef Bacik break; 17163fd0a558SYan, Zheng } 17175d4f98a2SYan Zheng 17185d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17195d4f98a2SYan Zheng dirty = 1; 17205d4f98a2SYan Zheng 17215d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 17225d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, new_bytenr, 17235d4f98a2SYan Zheng num_bytes, parent, 17245d4f98a2SYan Zheng btrfs_header_owner(leaf), 1725b06c4bf5SFilipe Manana key.objectid, key.offset); 172683d4cfd4SJosef Bacik if (ret) { 172766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 172883d4cfd4SJosef Bacik break; 172983d4cfd4SJosef Bacik } 17305d4f98a2SYan Zheng 17315d4f98a2SYan Zheng ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 17325d4f98a2SYan Zheng parent, 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 if (dirty) 17405d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17413fd0a558SYan, Zheng if (inode) 17423fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 174383d4cfd4SJosef Bacik return ret; 17445d4f98a2SYan Zheng } 17455d4f98a2SYan Zheng 17465d4f98a2SYan Zheng static noinline_for_stack 17475d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17485d4f98a2SYan Zheng struct btrfs_path *path, int level) 17495d4f98a2SYan Zheng { 17505d4f98a2SYan Zheng struct btrfs_disk_key key1; 17515d4f98a2SYan Zheng struct btrfs_disk_key key2; 17525d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17535d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17545d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17555d4f98a2SYan Zheng } 17565d4f98a2SYan Zheng 17575d4f98a2SYan Zheng /* 17585d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17595d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17605d4f98a2SYan Zheng * reloc tree was create can be replaced. 17615d4f98a2SYan Zheng * 17625d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17635d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17645d4f98a2SYan Zheng * errors, a negative error number is returned. 17655d4f98a2SYan Zheng */ 17663fd0a558SYan, Zheng static noinline_for_stack 17673fd0a558SYan, Zheng int replace_path(struct btrfs_trans_handle *trans, 17685d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17695d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17705d4f98a2SYan Zheng int lowest_level, int max_level) 17715d4f98a2SYan Zheng { 17725d4f98a2SYan Zheng struct extent_buffer *eb; 17735d4f98a2SYan Zheng struct extent_buffer *parent; 17745d4f98a2SYan Zheng struct btrfs_key key; 17755d4f98a2SYan Zheng u64 old_bytenr; 17765d4f98a2SYan Zheng u64 new_bytenr; 17775d4f98a2SYan Zheng u64 old_ptr_gen; 17785d4f98a2SYan Zheng u64 new_ptr_gen; 17795d4f98a2SYan Zheng u64 last_snapshot; 17805d4f98a2SYan Zheng u32 blocksize; 17813fd0a558SYan, Zheng int cow = 0; 17825d4f98a2SYan Zheng int level; 17835d4f98a2SYan Zheng int ret; 17845d4f98a2SYan Zheng int slot; 17855d4f98a2SYan Zheng 17865d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17875d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17885d4f98a2SYan Zheng 17895d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 17903fd0a558SYan, Zheng again: 17915d4f98a2SYan Zheng slot = path->slots[lowest_level]; 17925d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 17935d4f98a2SYan Zheng 17945d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 17955d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 17965d4f98a2SYan Zheng level = btrfs_header_level(eb); 17975d4f98a2SYan Zheng 17985d4f98a2SYan Zheng if (level < lowest_level) { 17995d4f98a2SYan Zheng btrfs_tree_unlock(eb); 18005d4f98a2SYan Zheng free_extent_buffer(eb); 18015d4f98a2SYan Zheng return 0; 18025d4f98a2SYan Zheng } 18035d4f98a2SYan Zheng 18043fd0a558SYan, Zheng if (cow) { 18055d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18065d4f98a2SYan Zheng BUG_ON(ret); 18073fd0a558SYan, Zheng } 18085d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 18095d4f98a2SYan Zheng 18105d4f98a2SYan Zheng if (next_key) { 18115d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18125d4f98a2SYan Zheng next_key->type = (u8)-1; 18135d4f98a2SYan Zheng next_key->offset = (u64)-1; 18145d4f98a2SYan Zheng } 18155d4f98a2SYan Zheng 18165d4f98a2SYan Zheng parent = eb; 18175d4f98a2SYan Zheng while (1) { 18185d4f98a2SYan Zheng level = btrfs_header_level(parent); 18195d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18205d4f98a2SYan Zheng 18215d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 18225d4f98a2SYan Zheng if (ret && slot > 0) 18235d4f98a2SYan Zheng slot--; 18245d4f98a2SYan Zheng 18255d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18265d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18275d4f98a2SYan Zheng 18285d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 1829707e8a07SDavid Sterba blocksize = dest->nodesize; 18305d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 18315d4f98a2SYan Zheng 18325d4f98a2SYan Zheng if (level <= max_level) { 18335d4f98a2SYan Zheng eb = path->nodes[level]; 18345d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18355d4f98a2SYan Zheng path->slots[level]); 18365d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18375d4f98a2SYan Zheng path->slots[level]); 18385d4f98a2SYan Zheng } else { 18395d4f98a2SYan Zheng new_bytenr = 0; 18405d4f98a2SYan Zheng new_ptr_gen = 0; 18415d4f98a2SYan Zheng } 18425d4f98a2SYan Zheng 1843fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18445d4f98a2SYan Zheng ret = level; 18455d4f98a2SYan Zheng break; 18465d4f98a2SYan Zheng } 18475d4f98a2SYan Zheng 18485d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18495d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18503fd0a558SYan, Zheng if (level <= lowest_level) { 18515d4f98a2SYan Zheng ret = 0; 18525d4f98a2SYan Zheng break; 18535d4f98a2SYan Zheng } 18545d4f98a2SYan Zheng 1855ce86cd59SDavid Sterba eb = read_tree_block(dest, old_bytenr, old_ptr_gen); 185664c043deSLiu Bo if (IS_ERR(eb)) { 185764c043deSLiu Bo ret = PTR_ERR(eb); 1858264813acSLiu Bo break; 185964c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 186064c043deSLiu Bo ret = -EIO; 1861416bc658SJosef Bacik free_extent_buffer(eb); 1862379cde74SStefan Behrens break; 1863416bc658SJosef Bacik } 18645d4f98a2SYan Zheng btrfs_tree_lock(eb); 18653fd0a558SYan, Zheng if (cow) { 18665d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18675d4f98a2SYan Zheng slot, &eb); 18685d4f98a2SYan Zheng BUG_ON(ret); 18695d4f98a2SYan Zheng } 18703fd0a558SYan, Zheng btrfs_set_lock_blocking(eb); 18715d4f98a2SYan Zheng 18725d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18735d4f98a2SYan Zheng free_extent_buffer(parent); 18745d4f98a2SYan Zheng 18755d4f98a2SYan Zheng parent = eb; 18765d4f98a2SYan Zheng continue; 18775d4f98a2SYan Zheng } 18785d4f98a2SYan Zheng 18793fd0a558SYan, Zheng if (!cow) { 18803fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18813fd0a558SYan, Zheng free_extent_buffer(parent); 18823fd0a558SYan, Zheng cow = 1; 18833fd0a558SYan, Zheng goto again; 18843fd0a558SYan, Zheng } 18853fd0a558SYan, Zheng 18865d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18875d4f98a2SYan Zheng path->slots[level]); 1888b3b4aa74SDavid Sterba btrfs_release_path(path); 18895d4f98a2SYan Zheng 18905d4f98a2SYan Zheng path->lowest_level = level; 18915d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 18925d4f98a2SYan Zheng path->lowest_level = 0; 18935d4f98a2SYan Zheng BUG_ON(ret); 18945d4f98a2SYan Zheng 18955d4f98a2SYan Zheng /* 18965d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 18975d4f98a2SYan Zheng */ 18985d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 18995d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19005d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19015d4f98a2SYan Zheng 19025d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19035d4f98a2SYan Zheng path->slots[level], old_bytenr); 19045d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19055d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19065d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19075d4f98a2SYan Zheng 19085d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize, 19095d4f98a2SYan Zheng path->nodes[level]->start, 1910b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19115d4f98a2SYan Zheng BUG_ON(ret); 19125d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize, 19135d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 1914b06c4bf5SFilipe Manana 0); 19155d4f98a2SYan Zheng BUG_ON(ret); 19165d4f98a2SYan Zheng 19175d4f98a2SYan Zheng ret = btrfs_free_extent(trans, src, new_bytenr, blocksize, 19185d4f98a2SYan Zheng path->nodes[level]->start, 1919b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19205d4f98a2SYan Zheng BUG_ON(ret); 19215d4f98a2SYan Zheng 19225d4f98a2SYan Zheng ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize, 19235d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 1924b06c4bf5SFilipe Manana 0); 19255d4f98a2SYan Zheng BUG_ON(ret); 19265d4f98a2SYan Zheng 19275d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19285d4f98a2SYan Zheng 19295d4f98a2SYan Zheng ret = level; 19305d4f98a2SYan Zheng break; 19315d4f98a2SYan Zheng } 19325d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19335d4f98a2SYan Zheng free_extent_buffer(parent); 19345d4f98a2SYan Zheng return ret; 19355d4f98a2SYan Zheng } 19365d4f98a2SYan Zheng 19375d4f98a2SYan Zheng /* 19385d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19395d4f98a2SYan Zheng */ 19405d4f98a2SYan Zheng static noinline_for_stack 19415d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19425d4f98a2SYan Zheng int *level) 19435d4f98a2SYan Zheng { 19445d4f98a2SYan Zheng struct extent_buffer *eb; 19455d4f98a2SYan Zheng int i; 19465d4f98a2SYan Zheng u64 last_snapshot; 19475d4f98a2SYan Zheng u32 nritems; 19485d4f98a2SYan Zheng 19495d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19505d4f98a2SYan Zheng 19515d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19525d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19535d4f98a2SYan Zheng path->nodes[i] = NULL; 19545d4f98a2SYan Zheng } 19555d4f98a2SYan Zheng 19565d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 19575d4f98a2SYan Zheng eb = path->nodes[i]; 19585d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19595d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 19605d4f98a2SYan Zheng path->slots[i]++; 19615d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 19625d4f98a2SYan Zheng last_snapshot) 19635d4f98a2SYan Zheng continue; 19645d4f98a2SYan Zheng 19655d4f98a2SYan Zheng *level = i; 19665d4f98a2SYan Zheng return 0; 19675d4f98a2SYan Zheng } 19685d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19695d4f98a2SYan Zheng path->nodes[i] = NULL; 19705d4f98a2SYan Zheng } 19715d4f98a2SYan Zheng return 1; 19725d4f98a2SYan Zheng } 19735d4f98a2SYan Zheng 19745d4f98a2SYan Zheng /* 19755d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 19765d4f98a2SYan Zheng */ 19775d4f98a2SYan Zheng static noinline_for_stack 19785d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19795d4f98a2SYan Zheng int *level) 19805d4f98a2SYan Zheng { 19815d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 19825d4f98a2SYan Zheng int i; 19835d4f98a2SYan Zheng u64 bytenr; 19845d4f98a2SYan Zheng u64 ptr_gen = 0; 19855d4f98a2SYan Zheng u64 last_snapshot; 19865d4f98a2SYan Zheng u32 nritems; 19875d4f98a2SYan Zheng 19885d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19895d4f98a2SYan Zheng 19905d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 19915d4f98a2SYan Zheng eb = path->nodes[i]; 19925d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19935d4f98a2SYan Zheng while (path->slots[i] < nritems) { 19945d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 19955d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 19965d4f98a2SYan Zheng break; 19975d4f98a2SYan Zheng path->slots[i]++; 19985d4f98a2SYan Zheng } 19995d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20005d4f98a2SYan Zheng if (i == *level) 20015d4f98a2SYan Zheng break; 20025d4f98a2SYan Zheng *level = i + 1; 20035d4f98a2SYan Zheng return 0; 20045d4f98a2SYan Zheng } 20055d4f98a2SYan Zheng if (i == 1) { 20065d4f98a2SYan Zheng *level = i; 20075d4f98a2SYan Zheng return 0; 20085d4f98a2SYan Zheng } 20095d4f98a2SYan Zheng 20105d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2011ce86cd59SDavid Sterba eb = read_tree_block(root, bytenr, ptr_gen); 201264c043deSLiu Bo if (IS_ERR(eb)) { 201364c043deSLiu Bo return PTR_ERR(eb); 201464c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2015416bc658SJosef Bacik free_extent_buffer(eb); 2016416bc658SJosef Bacik return -EIO; 2017416bc658SJosef Bacik } 20185d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20195d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20205d4f98a2SYan Zheng path->slots[i - 1] = 0; 20215d4f98a2SYan Zheng } 20225d4f98a2SYan Zheng return 1; 20235d4f98a2SYan Zheng } 20245d4f98a2SYan Zheng 20255d4f98a2SYan Zheng /* 20265d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 20275d4f98a2SYan Zheng * [min_key, max_key) 20285d4f98a2SYan Zheng */ 20295d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 20305d4f98a2SYan Zheng struct btrfs_key *min_key, 20315d4f98a2SYan Zheng struct btrfs_key *max_key) 20325d4f98a2SYan Zheng { 20335d4f98a2SYan Zheng struct inode *inode = NULL; 20345d4f98a2SYan Zheng u64 objectid; 20355d4f98a2SYan Zheng u64 start, end; 203633345d01SLi Zefan u64 ino; 20375d4f98a2SYan Zheng 20385d4f98a2SYan Zheng objectid = min_key->objectid; 20395d4f98a2SYan Zheng while (1) { 20405d4f98a2SYan Zheng cond_resched(); 20415d4f98a2SYan Zheng iput(inode); 20425d4f98a2SYan Zheng 20435d4f98a2SYan Zheng if (objectid > max_key->objectid) 20445d4f98a2SYan Zheng break; 20455d4f98a2SYan Zheng 20465d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20475d4f98a2SYan Zheng if (!inode) 20485d4f98a2SYan Zheng break; 204933345d01SLi Zefan ino = btrfs_ino(inode); 20505d4f98a2SYan Zheng 205133345d01SLi Zefan if (ino > max_key->objectid) { 20525d4f98a2SYan Zheng iput(inode); 20535d4f98a2SYan Zheng break; 20545d4f98a2SYan Zheng } 20555d4f98a2SYan Zheng 205633345d01SLi Zefan objectid = ino + 1; 20575d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 20585d4f98a2SYan Zheng continue; 20595d4f98a2SYan Zheng 206033345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 20615d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 20625d4f98a2SYan Zheng continue; 20635d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 20645d4f98a2SYan Zheng start = 0; 20655d4f98a2SYan Zheng else { 20665d4f98a2SYan Zheng start = min_key->offset; 20675d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(start, root->sectorsize)); 20685d4f98a2SYan Zheng } 20695d4f98a2SYan Zheng } else { 20705d4f98a2SYan Zheng start = 0; 20715d4f98a2SYan Zheng } 20725d4f98a2SYan Zheng 207333345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 20745d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 20755d4f98a2SYan Zheng continue; 20765d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 20775d4f98a2SYan Zheng end = (u64)-1; 20785d4f98a2SYan Zheng } else { 20795d4f98a2SYan Zheng if (max_key->offset == 0) 20805d4f98a2SYan Zheng continue; 20815d4f98a2SYan Zheng end = max_key->offset; 20825d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 20835d4f98a2SYan Zheng end--; 20845d4f98a2SYan Zheng } 20855d4f98a2SYan Zheng } else { 20865d4f98a2SYan Zheng end = (u64)-1; 20875d4f98a2SYan Zheng } 20885d4f98a2SYan Zheng 20895d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2090d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 20915d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 1); 2092d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 20935d4f98a2SYan Zheng } 20945d4f98a2SYan Zheng return 0; 20955d4f98a2SYan Zheng } 20965d4f98a2SYan Zheng 20975d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 20985d4f98a2SYan Zheng struct btrfs_key *key) 20995d4f98a2SYan Zheng 21005d4f98a2SYan Zheng { 21015d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21025d4f98a2SYan Zheng if (!path->nodes[level]) 21035d4f98a2SYan Zheng break; 21045d4f98a2SYan Zheng if (path->slots[level] + 1 < 21055d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21065d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21075d4f98a2SYan Zheng path->slots[level] + 1); 21085d4f98a2SYan Zheng return 0; 21095d4f98a2SYan Zheng } 21105d4f98a2SYan Zheng level++; 21115d4f98a2SYan Zheng } 21125d4f98a2SYan Zheng return 1; 21135d4f98a2SYan Zheng } 21145d4f98a2SYan Zheng 21155d4f98a2SYan Zheng /* 21165d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 21175d4f98a2SYan Zheng * fs tree. 21185d4f98a2SYan Zheng */ 21195d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 21205d4f98a2SYan Zheng struct btrfs_root *root) 21215d4f98a2SYan Zheng { 21225d4f98a2SYan Zheng LIST_HEAD(inode_list); 21235d4f98a2SYan Zheng struct btrfs_key key; 21245d4f98a2SYan Zheng struct btrfs_key next_key; 21259e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 21265d4f98a2SYan Zheng struct btrfs_root *reloc_root; 21275d4f98a2SYan Zheng struct btrfs_root_item *root_item; 21285d4f98a2SYan Zheng struct btrfs_path *path; 21293fd0a558SYan, Zheng struct extent_buffer *leaf; 21305d4f98a2SYan Zheng int level; 21315d4f98a2SYan Zheng int max_level; 21325d4f98a2SYan Zheng int replaced = 0; 21335d4f98a2SYan Zheng int ret; 21345d4f98a2SYan Zheng int err = 0; 21353fd0a558SYan, Zheng u32 min_reserved; 21365d4f98a2SYan Zheng 21375d4f98a2SYan Zheng path = btrfs_alloc_path(); 21385d4f98a2SYan Zheng if (!path) 21395d4f98a2SYan Zheng return -ENOMEM; 2140e4058b54SDavid Sterba path->reada = READA_FORWARD; 21415d4f98a2SYan Zheng 21425d4f98a2SYan Zheng reloc_root = root->reloc_root; 21435d4f98a2SYan Zheng root_item = &reloc_root->root_item; 21445d4f98a2SYan Zheng 21455d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 21465d4f98a2SYan Zheng level = btrfs_root_level(root_item); 21475d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 21485d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 21495d4f98a2SYan Zheng path->slots[level] = 0; 21505d4f98a2SYan Zheng } else { 21515d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 21525d4f98a2SYan Zheng 21535d4f98a2SYan Zheng level = root_item->drop_level; 21545d4f98a2SYan Zheng BUG_ON(level == 0); 21555d4f98a2SYan Zheng path->lowest_level = level; 21565d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 215733c66f43SYan Zheng path->lowest_level = 0; 21585d4f98a2SYan Zheng if (ret < 0) { 21595d4f98a2SYan Zheng btrfs_free_path(path); 21605d4f98a2SYan Zheng return ret; 21615d4f98a2SYan Zheng } 21625d4f98a2SYan Zheng 21635d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 21645d4f98a2SYan Zheng path->slots[level]); 21655d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 21665d4f98a2SYan Zheng 21675d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 21685d4f98a2SYan Zheng } 21695d4f98a2SYan Zheng 21703fd0a558SYan, Zheng min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 21715d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 21725d4f98a2SYan Zheng 21735d4f98a2SYan Zheng while (1) { 217408e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 217508e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 21763fd0a558SYan, Zheng if (ret) { 21779e6a0c52SJosef Bacik err = ret; 21789e6a0c52SJosef Bacik goto out; 21793fd0a558SYan, Zheng } 21809e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 21819e6a0c52SJosef Bacik if (IS_ERR(trans)) { 21829e6a0c52SJosef Bacik err = PTR_ERR(trans); 21839e6a0c52SJosef Bacik trans = NULL; 21849e6a0c52SJosef Bacik goto out; 21859e6a0c52SJosef Bacik } 21869e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 21873fd0a558SYan, Zheng 21883fd0a558SYan, Zheng replaced = 0; 21895d4f98a2SYan Zheng max_level = level; 21905d4f98a2SYan Zheng 21915d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 21925d4f98a2SYan Zheng if (ret < 0) { 21935d4f98a2SYan Zheng err = ret; 21945d4f98a2SYan Zheng goto out; 21955d4f98a2SYan Zheng } 21965d4f98a2SYan Zheng if (ret > 0) 21975d4f98a2SYan Zheng break; 21985d4f98a2SYan Zheng 21995d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 22005d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 22015d4f98a2SYan Zheng ret = 0; 22025d4f98a2SYan Zheng } else { 22033fd0a558SYan, Zheng ret = replace_path(trans, root, reloc_root, path, 22043fd0a558SYan, Zheng &next_key, level, max_level); 22055d4f98a2SYan Zheng } 22065d4f98a2SYan Zheng if (ret < 0) { 22075d4f98a2SYan Zheng err = ret; 22085d4f98a2SYan Zheng goto out; 22095d4f98a2SYan Zheng } 22105d4f98a2SYan Zheng 22115d4f98a2SYan Zheng if (ret > 0) { 22125d4f98a2SYan Zheng level = ret; 22135d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 22145d4f98a2SYan Zheng path->slots[level]); 22155d4f98a2SYan Zheng replaced = 1; 22165d4f98a2SYan Zheng } 22175d4f98a2SYan Zheng 22185d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 22195d4f98a2SYan Zheng if (ret > 0) 22205d4f98a2SYan Zheng break; 22215d4f98a2SYan Zheng 22225d4f98a2SYan Zheng BUG_ON(level == 0); 22235d4f98a2SYan Zheng /* 22245d4f98a2SYan Zheng * save the merging progress in the drop_progress. 22255d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 22265d4f98a2SYan Zheng */ 22275d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 22285d4f98a2SYan Zheng path->slots[level]); 22295d4f98a2SYan Zheng root_item->drop_level = level; 22305d4f98a2SYan Zheng 22313fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 22329e6a0c52SJosef Bacik trans = NULL; 22335d4f98a2SYan Zheng 2234b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 22355d4f98a2SYan Zheng 22365d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22375d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22385d4f98a2SYan Zheng } 22395d4f98a2SYan Zheng 22405d4f98a2SYan Zheng /* 22415d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 22425d4f98a2SYan Zheng * relocated and the block is tree root. 22435d4f98a2SYan Zheng */ 22445d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 22455d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 22465d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 22475d4f98a2SYan Zheng free_extent_buffer(leaf); 22485d4f98a2SYan Zheng if (ret < 0) 22495d4f98a2SYan Zheng err = ret; 22505d4f98a2SYan Zheng out: 22515d4f98a2SYan Zheng btrfs_free_path(path); 22525d4f98a2SYan Zheng 22535d4f98a2SYan Zheng if (err == 0) { 22545d4f98a2SYan Zheng memset(&root_item->drop_progress, 0, 22555d4f98a2SYan Zheng sizeof(root_item->drop_progress)); 22565d4f98a2SYan Zheng root_item->drop_level = 0; 22575d4f98a2SYan Zheng btrfs_set_root_refs(root_item, 0); 22583fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 22595d4f98a2SYan Zheng } 22605d4f98a2SYan Zheng 22619e6a0c52SJosef Bacik if (trans) 22623fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 22635d4f98a2SYan Zheng 2264b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 22655d4f98a2SYan Zheng 22665d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22675d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22685d4f98a2SYan Zheng 22695d4f98a2SYan Zheng return err; 22705d4f98a2SYan Zheng } 22715d4f98a2SYan Zheng 22723fd0a558SYan, Zheng static noinline_for_stack 22733fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 22745d4f98a2SYan Zheng { 22753fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 22763fd0a558SYan, Zheng struct btrfs_root *reloc_root; 22775d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 22783fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 22793fd0a558SYan, Zheng u64 num_bytes = 0; 22803fd0a558SYan, Zheng int ret; 22813fd0a558SYan, Zheng 22827585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 22833fd0a558SYan, Zheng rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22843fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 22857585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 22867585717fSChris Mason 22873fd0a558SYan, Zheng again: 22883fd0a558SYan, Zheng if (!err) { 22893fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 229008e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 229108e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22923fd0a558SYan, Zheng if (ret) 22933fd0a558SYan, Zheng err = ret; 22943fd0a558SYan, Zheng } 22953fd0a558SYan, Zheng 22967a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 22973612b495STsutomu Itoh if (IS_ERR(trans)) { 22983612b495STsutomu Itoh if (!err) 22993612b495STsutomu Itoh btrfs_block_rsv_release(rc->extent_root, 23003612b495STsutomu Itoh rc->block_rsv, num_bytes); 23013612b495STsutomu Itoh return PTR_ERR(trans); 23023612b495STsutomu Itoh } 23033fd0a558SYan, Zheng 23043fd0a558SYan, Zheng if (!err) { 23053fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 23063fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 23073fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, 23083fd0a558SYan, Zheng rc->block_rsv, num_bytes); 23093fd0a558SYan, Zheng goto again; 23103fd0a558SYan, Zheng } 23113fd0a558SYan, Zheng } 23123fd0a558SYan, Zheng 23133fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 23143fd0a558SYan, Zheng 23153fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 23163fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 23173fd0a558SYan, Zheng struct btrfs_root, root_list); 23183fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 23193fd0a558SYan, Zheng 23203fd0a558SYan, Zheng root = read_fs_root(reloc_root->fs_info, 23213fd0a558SYan, Zheng reloc_root->root_key.offset); 23223fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 23233fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 23243fd0a558SYan, Zheng 23253fd0a558SYan, Zheng /* 23263fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 23273fd0a558SYan, Zheng * knows it should resumes merging 23283fd0a558SYan, Zheng */ 23293fd0a558SYan, Zheng if (!err) 23303fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 23313fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 23323fd0a558SYan, Zheng 23333fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 23343fd0a558SYan, Zheng } 23353fd0a558SYan, Zheng 23363fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 23373fd0a558SYan, Zheng 23383fd0a558SYan, Zheng if (!err) 23393fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 23403fd0a558SYan, Zheng else 23413fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 23423fd0a558SYan, Zheng return err; 23433fd0a558SYan, Zheng } 23443fd0a558SYan, Zheng 23453fd0a558SYan, Zheng static noinline_for_stack 2346aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2347aca1bba6SLiu Bo { 2348aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2349aca1bba6SLiu Bo 2350aca1bba6SLiu Bo while (!list_empty(list)) { 2351aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2352aca1bba6SLiu Bo root_list); 2353c974c464SWang Shilong __del_reloc_root(reloc_root); 2354aca1bba6SLiu Bo } 2355aca1bba6SLiu Bo } 2356aca1bba6SLiu Bo 2357aca1bba6SLiu Bo static noinline_for_stack 235894404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 23593fd0a558SYan, Zheng { 23605d4f98a2SYan Zheng struct btrfs_root *root; 23615d4f98a2SYan Zheng struct btrfs_root *reloc_root; 23625bc7247aSMiao Xie u64 last_snap; 23635bc7247aSMiao Xie u64 otransid; 23645bc7247aSMiao Xie u64 objectid; 23653fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23663fd0a558SYan, Zheng int found = 0; 2367aca1bba6SLiu Bo int ret = 0; 23683fd0a558SYan, Zheng again: 23693fd0a558SYan, Zheng root = rc->extent_root; 23707585717fSChris Mason 23717585717fSChris Mason /* 23727585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 23737585717fSChris Mason * we have to make sure nobody is in the middle of 23747585717fSChris Mason * adding their roots to the list while we are 23757585717fSChris Mason * doing this splice 23767585717fSChris Mason */ 23777585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 23783fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 23797585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 23805d4f98a2SYan Zheng 23813fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 23823fd0a558SYan, Zheng found = 1; 23833fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 23843fd0a558SYan, Zheng struct btrfs_root, root_list); 23855d4f98a2SYan Zheng 23865d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 23875d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, 23885d4f98a2SYan Zheng reloc_root->root_key.offset); 23895d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 23905d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 23915d4f98a2SYan Zheng 23923fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2393b37b39cdSJosef Bacik if (ret) { 239425e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 239525e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 239625e293c2SWang Shilong &reloc_roots); 2397aca1bba6SLiu Bo goto out; 2398b37b39cdSJosef Bacik } 23993fd0a558SYan, Zheng } else { 24003fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24013fd0a558SYan, Zheng } 24025bc7247aSMiao Xie 24035bc7247aSMiao Xie /* 240401327610SNicholas D Steeves * we keep the old last snapshot transid in rtranid when we 24055bc7247aSMiao Xie * created the relocation tree. 24065bc7247aSMiao Xie */ 24075bc7247aSMiao Xie last_snap = btrfs_root_rtransid(&reloc_root->root_item); 24085bc7247aSMiao Xie otransid = btrfs_root_otransid(&reloc_root->root_item); 24095bc7247aSMiao Xie objectid = reloc_root->root_key.offset; 24105bc7247aSMiao Xie 24112c536799SJeff Mahoney ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1); 2412aca1bba6SLiu Bo if (ret < 0) { 2413aca1bba6SLiu Bo if (list_empty(&reloc_root->root_list)) 2414aca1bba6SLiu Bo list_add_tail(&reloc_root->root_list, 2415aca1bba6SLiu Bo &reloc_roots); 2416aca1bba6SLiu Bo goto out; 2417aca1bba6SLiu Bo } 24185d4f98a2SYan Zheng } 24195d4f98a2SYan Zheng 24203fd0a558SYan, Zheng if (found) { 24213fd0a558SYan, Zheng found = 0; 24223fd0a558SYan, Zheng goto again; 24235d4f98a2SYan Zheng } 2424aca1bba6SLiu Bo out: 2425aca1bba6SLiu Bo if (ret) { 242634d97007SAnand Jain btrfs_handle_fs_error(root->fs_info, ret, NULL); 2427aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2428aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2429467bb1d2SWang Shilong 2430467bb1d2SWang Shilong /* new reloc root may be added */ 2431467bb1d2SWang Shilong mutex_lock(&root->fs_info->reloc_mutex); 2432467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 2433467bb1d2SWang Shilong mutex_unlock(&root->fs_info->reloc_mutex); 2434467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2435467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2436aca1bba6SLiu Bo } 2437aca1bba6SLiu Bo 24385d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 24395d4f98a2SYan Zheng } 24405d4f98a2SYan Zheng 24415d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 24425d4f98a2SYan Zheng { 24435d4f98a2SYan Zheng struct tree_block *block; 24445d4f98a2SYan Zheng struct rb_node *rb_node; 24455d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 24465d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 24475d4f98a2SYan Zheng rb_erase(rb_node, blocks); 24485d4f98a2SYan Zheng kfree(block); 24495d4f98a2SYan Zheng } 24505d4f98a2SYan Zheng } 24515d4f98a2SYan Zheng 24525d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 24535d4f98a2SYan Zheng struct btrfs_root *reloc_root) 24545d4f98a2SYan Zheng { 24555d4f98a2SYan Zheng struct btrfs_root *root; 24565d4f98a2SYan Zheng 24575d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 24585d4f98a2SYan Zheng return 0; 24595d4f98a2SYan Zheng 24605d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset); 24615d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24625d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24635d4f98a2SYan Zheng 24645d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 24655d4f98a2SYan Zheng } 24665d4f98a2SYan Zheng 24673fd0a558SYan, Zheng static noinline_for_stack 24683fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 24693fd0a558SYan, Zheng struct reloc_control *rc, 24705d4f98a2SYan Zheng struct backref_node *node, 2471dc4103f9SWang Shilong struct backref_edge *edges[]) 24725d4f98a2SYan Zheng { 24735d4f98a2SYan Zheng struct backref_node *next; 24745d4f98a2SYan Zheng struct btrfs_root *root; 24753fd0a558SYan, Zheng int index = 0; 24763fd0a558SYan, Zheng 24775d4f98a2SYan Zheng next = node; 24785d4f98a2SYan Zheng while (1) { 24795d4f98a2SYan Zheng cond_resched(); 24805d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 24815d4f98a2SYan Zheng root = next->root; 24823fd0a558SYan, Zheng BUG_ON(!root); 248327cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 24845d4f98a2SYan Zheng 24855d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 24865d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 24875d4f98a2SYan Zheng break; 24885d4f98a2SYan Zheng } 24895d4f98a2SYan Zheng 24905d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 24913fd0a558SYan, Zheng root = root->reloc_root; 24923fd0a558SYan, Zheng 24933fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 24943fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 24953fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 24963fd0a558SYan, Zheng next->new_bytenr = root->node->start; 24973fd0a558SYan, Zheng next->root = root; 24983fd0a558SYan, Zheng list_add_tail(&next->list, 24993fd0a558SYan, Zheng &rc->backref_cache.changed); 25003fd0a558SYan, Zheng __mark_block_processed(rc, next); 25015d4f98a2SYan Zheng break; 25025d4f98a2SYan Zheng } 25035d4f98a2SYan Zheng 25043fd0a558SYan, Zheng WARN_ON(1); 25055d4f98a2SYan Zheng root = NULL; 25065d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25075d4f98a2SYan Zheng if (!next || next->level <= node->level) 25085d4f98a2SYan Zheng break; 25095d4f98a2SYan Zheng } 25103fd0a558SYan, Zheng if (!root) 25113fd0a558SYan, Zheng return NULL; 25125d4f98a2SYan Zheng 25133fd0a558SYan, Zheng next = node; 25143fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 25153fd0a558SYan, Zheng while (1) { 25163fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 25173fd0a558SYan, Zheng if (--index < 0) 25183fd0a558SYan, Zheng break; 25193fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 25203fd0a558SYan, Zheng } 25215d4f98a2SYan Zheng return root; 25225d4f98a2SYan Zheng } 25235d4f98a2SYan Zheng 25243fd0a558SYan, Zheng /* 25253fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 25263fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 25273fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 25283fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 25293fd0a558SYan, Zheng */ 25305d4f98a2SYan Zheng static noinline_for_stack 2531147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 25325d4f98a2SYan Zheng { 25333fd0a558SYan, Zheng struct backref_node *next; 25343fd0a558SYan, Zheng struct btrfs_root *root; 25353fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 25365d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25373fd0a558SYan, Zheng int index = 0; 25383fd0a558SYan, Zheng 25393fd0a558SYan, Zheng next = node; 25403fd0a558SYan, Zheng while (1) { 25413fd0a558SYan, Zheng cond_resched(); 25423fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 25433fd0a558SYan, Zheng root = next->root; 25443fd0a558SYan, Zheng BUG_ON(!root); 25453fd0a558SYan, Zheng 254625985edcSLucas De Marchi /* no other choice for non-references counted tree */ 254727cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 25483fd0a558SYan, Zheng return root; 25493fd0a558SYan, Zheng 25503fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 25513fd0a558SYan, Zheng fs_root = root; 25523fd0a558SYan, Zheng 25533fd0a558SYan, Zheng if (next != node) 25543fd0a558SYan, Zheng return NULL; 25553fd0a558SYan, Zheng 25563fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 25573fd0a558SYan, Zheng if (!next || next->level <= node->level) 25583fd0a558SYan, Zheng break; 25593fd0a558SYan, Zheng } 25603fd0a558SYan, Zheng 25613fd0a558SYan, Zheng if (!fs_root) 25623fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 25633fd0a558SYan, Zheng return fs_root; 25645d4f98a2SYan Zheng } 25655d4f98a2SYan Zheng 25665d4f98a2SYan Zheng static noinline_for_stack 25673fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 25683fd0a558SYan, Zheng struct backref_node *node, int reserve) 25695d4f98a2SYan Zheng { 25703fd0a558SYan, Zheng struct backref_node *next = node; 25713fd0a558SYan, Zheng struct backref_edge *edge; 25723fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25733fd0a558SYan, Zheng u64 num_bytes = 0; 25743fd0a558SYan, Zheng int index = 0; 25755d4f98a2SYan Zheng 25763fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 25773fd0a558SYan, Zheng 25783fd0a558SYan, Zheng while (next) { 25793fd0a558SYan, Zheng cond_resched(); 25805d4f98a2SYan Zheng while (1) { 25813fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 25825d4f98a2SYan Zheng break; 25835d4f98a2SYan Zheng 2584707e8a07SDavid Sterba num_bytes += rc->extent_root->nodesize; 25853fd0a558SYan, Zheng 25863fd0a558SYan, Zheng if (list_empty(&next->upper)) 25873fd0a558SYan, Zheng break; 25883fd0a558SYan, Zheng 25893fd0a558SYan, Zheng edge = list_entry(next->upper.next, 25903fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 25913fd0a558SYan, Zheng edges[index++] = edge; 25923fd0a558SYan, Zheng next = edge->node[UPPER]; 25935d4f98a2SYan Zheng } 25943fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 25953fd0a558SYan, Zheng } 25963fd0a558SYan, Zheng return num_bytes; 25973fd0a558SYan, Zheng } 25983fd0a558SYan, Zheng 25993fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 26003fd0a558SYan, Zheng struct reloc_control *rc, 26013fd0a558SYan, Zheng struct backref_node *node) 26023fd0a558SYan, Zheng { 26033fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 26043fd0a558SYan, Zheng u64 num_bytes; 26053fd0a558SYan, Zheng int ret; 26060647bf56SWang Shilong u64 tmp; 26073fd0a558SYan, Zheng 26083fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 26093fd0a558SYan, Zheng 26103fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 26110647bf56SWang Shilong rc->reserved_bytes += num_bytes; 26128ca17f0fSJosef Bacik 26138ca17f0fSJosef Bacik /* 26148ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 26158ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 26168ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 26178ca17f0fSJosef Bacik */ 26180647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 26198ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 26203fd0a558SYan, Zheng if (ret) { 26218ca17f0fSJosef Bacik tmp = rc->extent_root->nodesize * RELOCATION_RESERVED_NODES; 26220647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 26230647bf56SWang Shilong tmp <<= 1; 26240647bf56SWang Shilong /* 26250647bf56SWang Shilong * only one thread can access block_rsv at this point, 26260647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 26270647bf56SWang Shilong * we expand more reservation size here to allow enough 26288ca17f0fSJosef Bacik * space for relocation and we will return eailer in 26290647bf56SWang Shilong * enospc case. 26300647bf56SWang Shilong */ 26310647bf56SWang Shilong rc->block_rsv->size = tmp + rc->extent_root->nodesize * 26320647bf56SWang Shilong RELOCATION_RESERVED_NODES; 26338ca17f0fSJosef Bacik return -EAGAIN; 26343fd0a558SYan, Zheng } 26353fd0a558SYan, Zheng 26363fd0a558SYan, Zheng return 0; 26373fd0a558SYan, Zheng } 26383fd0a558SYan, Zheng 26395d4f98a2SYan Zheng /* 26405d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 26415d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 26425d4f98a2SYan Zheng * 26435d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 26445d4f98a2SYan Zheng * in that case this function just updates pointers. 26455d4f98a2SYan Zheng */ 26465d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 26473fd0a558SYan, Zheng struct reloc_control *rc, 26485d4f98a2SYan Zheng struct backref_node *node, 26495d4f98a2SYan Zheng struct btrfs_key *key, 26505d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 26515d4f98a2SYan Zheng { 26525d4f98a2SYan Zheng struct backref_node *upper; 26535d4f98a2SYan Zheng struct backref_edge *edge; 26545d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26555d4f98a2SYan Zheng struct btrfs_root *root; 26565d4f98a2SYan Zheng struct extent_buffer *eb; 26575d4f98a2SYan Zheng u32 blocksize; 26585d4f98a2SYan Zheng u64 bytenr; 26595d4f98a2SYan Zheng u64 generation; 26605d4f98a2SYan Zheng int slot; 26615d4f98a2SYan Zheng int ret; 26625d4f98a2SYan Zheng int err = 0; 26635d4f98a2SYan Zheng 26645d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 26655d4f98a2SYan Zheng 26665d4f98a2SYan Zheng path->lowest_level = node->level + 1; 26673fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 26685d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 26695d4f98a2SYan Zheng cond_resched(); 26705d4f98a2SYan Zheng 26715d4f98a2SYan Zheng upper = edge->node[UPPER]; 2672dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 26733fd0a558SYan, Zheng BUG_ON(!root); 26745d4f98a2SYan Zheng 26753fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 26763fd0a558SYan, Zheng if (!lowest) { 26773fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 26783fd0a558SYan, Zheng upper->level, &slot); 26793fd0a558SYan, Zheng BUG_ON(ret); 26803fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 26813fd0a558SYan, Zheng if (node->eb->start == bytenr) 26823fd0a558SYan, Zheng goto next; 26833fd0a558SYan, Zheng } 26845d4f98a2SYan Zheng drop_node_buffer(upper); 26853fd0a558SYan, Zheng } 26865d4f98a2SYan Zheng 26875d4f98a2SYan Zheng if (!upper->eb) { 26885d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 26895d4f98a2SYan Zheng if (ret < 0) { 26905d4f98a2SYan Zheng err = ret; 26915d4f98a2SYan Zheng break; 26925d4f98a2SYan Zheng } 26935d4f98a2SYan Zheng BUG_ON(ret > 0); 26945d4f98a2SYan Zheng 26953fd0a558SYan, Zheng if (!upper->eb) { 26963fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 26973fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 26983fd0a558SYan, Zheng } else { 26993fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 27003fd0a558SYan, Zheng } 27013fd0a558SYan, Zheng 27023fd0a558SYan, Zheng upper->locked = 1; 27033fd0a558SYan, Zheng path->locks[upper->level] = 0; 27043fd0a558SYan, Zheng 27055d4f98a2SYan Zheng slot = path->slots[upper->level]; 2706b3b4aa74SDavid Sterba btrfs_release_path(path); 27075d4f98a2SYan Zheng } else { 27085d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 27095d4f98a2SYan Zheng &slot); 27105d4f98a2SYan Zheng BUG_ON(ret); 27115d4f98a2SYan Zheng } 27125d4f98a2SYan Zheng 27135d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27143fd0a558SYan, Zheng if (lowest) { 27153fd0a558SYan, Zheng BUG_ON(bytenr != node->bytenr); 27165d4f98a2SYan Zheng } else { 27173fd0a558SYan, Zheng if (node->eb->start == bytenr) 27183fd0a558SYan, Zheng goto next; 27195d4f98a2SYan Zheng } 27205d4f98a2SYan Zheng 2721707e8a07SDavid Sterba blocksize = root->nodesize; 27225d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2723ce86cd59SDavid Sterba eb = read_tree_block(root, bytenr, generation); 272464c043deSLiu Bo if (IS_ERR(eb)) { 272564c043deSLiu Bo err = PTR_ERR(eb); 272664c043deSLiu Bo goto next; 272764c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2728416bc658SJosef Bacik free_extent_buffer(eb); 272997d9a8a4STsutomu Itoh err = -EIO; 273097d9a8a4STsutomu Itoh goto next; 273197d9a8a4STsutomu Itoh } 27325d4f98a2SYan Zheng btrfs_tree_lock(eb); 27335d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 27345d4f98a2SYan Zheng 27355d4f98a2SYan Zheng if (!node->eb) { 27365d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 27375d4f98a2SYan Zheng slot, &eb); 27383fd0a558SYan, Zheng btrfs_tree_unlock(eb); 27393fd0a558SYan, Zheng free_extent_buffer(eb); 27405d4f98a2SYan Zheng if (ret < 0) { 27415d4f98a2SYan Zheng err = ret; 27423fd0a558SYan, Zheng goto next; 27435d4f98a2SYan Zheng } 27443fd0a558SYan, Zheng BUG_ON(node->eb != eb); 27455d4f98a2SYan Zheng } else { 27465d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 27475d4f98a2SYan Zheng node->eb->start); 27485d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 27495d4f98a2SYan Zheng trans->transid); 27505d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 27515d4f98a2SYan Zheng 27525d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, 27535d4f98a2SYan Zheng node->eb->start, blocksize, 27545d4f98a2SYan Zheng upper->eb->start, 27555d4f98a2SYan Zheng btrfs_header_owner(upper->eb), 2756b06c4bf5SFilipe Manana node->level, 0); 27575d4f98a2SYan Zheng BUG_ON(ret); 27585d4f98a2SYan Zheng 27595d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 27605d4f98a2SYan Zheng BUG_ON(ret); 27615d4f98a2SYan Zheng } 27623fd0a558SYan, Zheng next: 27633fd0a558SYan, Zheng if (!upper->pending) 27643fd0a558SYan, Zheng drop_node_buffer(upper); 27653fd0a558SYan, Zheng else 27663fd0a558SYan, Zheng unlock_node_buffer(upper); 27673fd0a558SYan, Zheng if (err) 27683fd0a558SYan, Zheng break; 27695d4f98a2SYan Zheng } 27703fd0a558SYan, Zheng 27713fd0a558SYan, Zheng if (!err && node->pending) { 27723fd0a558SYan, Zheng drop_node_buffer(node); 27733fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 27743fd0a558SYan, Zheng node->pending = 0; 27755d4f98a2SYan Zheng } 27763fd0a558SYan, Zheng 27775d4f98a2SYan Zheng path->lowest_level = 0; 27783fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 27795d4f98a2SYan Zheng return err; 27805d4f98a2SYan Zheng } 27815d4f98a2SYan Zheng 27825d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 27833fd0a558SYan, Zheng struct reloc_control *rc, 27845d4f98a2SYan Zheng struct backref_node *node, 27855d4f98a2SYan Zheng struct btrfs_path *path) 27865d4f98a2SYan Zheng { 27875d4f98a2SYan Zheng struct btrfs_key key; 27885d4f98a2SYan Zheng 27895d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 27903fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 27915d4f98a2SYan Zheng } 27925d4f98a2SYan Zheng 27935d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 27943fd0a558SYan, Zheng struct reloc_control *rc, 27953fd0a558SYan, Zheng struct btrfs_path *path, int err) 27965d4f98a2SYan Zheng { 27973fd0a558SYan, Zheng LIST_HEAD(list); 27983fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 27995d4f98a2SYan Zheng struct backref_node *node; 28005d4f98a2SYan Zheng int level; 28015d4f98a2SYan Zheng int ret; 28025d4f98a2SYan Zheng 28035d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 28045d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 28055d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 28063fd0a558SYan, Zheng struct backref_node, list); 28073fd0a558SYan, Zheng list_move_tail(&node->list, &list); 28083fd0a558SYan, Zheng BUG_ON(!node->pending); 28095d4f98a2SYan Zheng 28103fd0a558SYan, Zheng if (!err) { 28113fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 28125d4f98a2SYan Zheng if (ret < 0) 28135d4f98a2SYan Zheng err = ret; 28145d4f98a2SYan Zheng } 28155d4f98a2SYan Zheng } 28163fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 28173fd0a558SYan, Zheng } 28185d4f98a2SYan Zheng return err; 28195d4f98a2SYan Zheng } 28205d4f98a2SYan Zheng 28215d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 28223fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 28233fd0a558SYan, Zheng { 28243fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2825ceeb0ae7SDavid Sterba EXTENT_DIRTY); 28263fd0a558SYan, Zheng } 28273fd0a558SYan, Zheng 28283fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 28295d4f98a2SYan Zheng struct backref_node *node) 28305d4f98a2SYan Zheng { 28315d4f98a2SYan Zheng u32 blocksize; 28325d4f98a2SYan Zheng if (node->level == 0 || 28335d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2834707e8a07SDavid Sterba blocksize = rc->extent_root->nodesize; 28353fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 28365d4f98a2SYan Zheng } 28375d4f98a2SYan Zheng node->processed = 1; 28385d4f98a2SYan Zheng } 28395d4f98a2SYan Zheng 28405d4f98a2SYan Zheng /* 28415d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 28425d4f98a2SYan Zheng * as processed. 28435d4f98a2SYan Zheng */ 28445d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 28455d4f98a2SYan Zheng struct backref_node *node) 28465d4f98a2SYan Zheng { 28475d4f98a2SYan Zheng struct backref_node *next = node; 28485d4f98a2SYan Zheng struct backref_edge *edge; 28495d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 28505d4f98a2SYan Zheng int index = 0; 28515d4f98a2SYan Zheng 28525d4f98a2SYan Zheng while (next) { 28535d4f98a2SYan Zheng cond_resched(); 28545d4f98a2SYan Zheng while (1) { 28555d4f98a2SYan Zheng if (next->processed) 28565d4f98a2SYan Zheng break; 28575d4f98a2SYan Zheng 28583fd0a558SYan, Zheng __mark_block_processed(rc, next); 28595d4f98a2SYan Zheng 28605d4f98a2SYan Zheng if (list_empty(&next->upper)) 28615d4f98a2SYan Zheng break; 28625d4f98a2SYan Zheng 28635d4f98a2SYan Zheng edge = list_entry(next->upper.next, 28645d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 28655d4f98a2SYan Zheng edges[index++] = edge; 28665d4f98a2SYan Zheng next = edge->node[UPPER]; 28675d4f98a2SYan Zheng } 28685d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 28695d4f98a2SYan Zheng } 28705d4f98a2SYan Zheng } 28715d4f98a2SYan Zheng 28727476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 28735d4f98a2SYan Zheng { 28747476dfdaSDavid Sterba u32 blocksize = rc->extent_root->nodesize; 28757476dfdaSDavid Sterba 28765d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 28779655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 28785d4f98a2SYan Zheng return 1; 28795d4f98a2SYan Zheng return 0; 28805d4f98a2SYan Zheng } 28815d4f98a2SYan Zheng 28825d4f98a2SYan Zheng static int get_tree_block_key(struct reloc_control *rc, 28835d4f98a2SYan Zheng struct tree_block *block) 28845d4f98a2SYan Zheng { 28855d4f98a2SYan Zheng struct extent_buffer *eb; 28865d4f98a2SYan Zheng 28875d4f98a2SYan Zheng BUG_ON(block->key_ready); 28885d4f98a2SYan Zheng eb = read_tree_block(rc->extent_root, block->bytenr, 2889ce86cd59SDavid Sterba block->key.offset); 289064c043deSLiu Bo if (IS_ERR(eb)) { 289164c043deSLiu Bo return PTR_ERR(eb); 289264c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2893416bc658SJosef Bacik free_extent_buffer(eb); 2894416bc658SJosef Bacik return -EIO; 2895416bc658SJosef Bacik } 28965d4f98a2SYan Zheng WARN_ON(btrfs_header_level(eb) != block->level); 28975d4f98a2SYan Zheng if (block->level == 0) 28985d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 28995d4f98a2SYan Zheng else 29005d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 29015d4f98a2SYan Zheng free_extent_buffer(eb); 29025d4f98a2SYan Zheng block->key_ready = 1; 29035d4f98a2SYan Zheng return 0; 29045d4f98a2SYan Zheng } 29055d4f98a2SYan Zheng 29065d4f98a2SYan Zheng /* 29075d4f98a2SYan Zheng * helper function to relocate a tree block 29085d4f98a2SYan Zheng */ 29095d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 29105d4f98a2SYan Zheng struct reloc_control *rc, 29115d4f98a2SYan Zheng struct backref_node *node, 29125d4f98a2SYan Zheng struct btrfs_key *key, 29135d4f98a2SYan Zheng struct btrfs_path *path) 29145d4f98a2SYan Zheng { 29155d4f98a2SYan Zheng struct btrfs_root *root; 29163fd0a558SYan, Zheng int ret = 0; 29175d4f98a2SYan Zheng 29183fd0a558SYan, Zheng if (!node) 29195d4f98a2SYan Zheng return 0; 29203fd0a558SYan, Zheng 29213fd0a558SYan, Zheng BUG_ON(node->processed); 2922147d256eSZhaolei root = select_one_root(node); 29233fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 29243fd0a558SYan, Zheng update_processed_blocks(rc, node); 29253fd0a558SYan, Zheng goto out; 29265d4f98a2SYan Zheng } 29275d4f98a2SYan Zheng 292827cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29293fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 29303fd0a558SYan, Zheng if (ret) 29315d4f98a2SYan Zheng goto out; 29325d4f98a2SYan Zheng } 29333fd0a558SYan, Zheng 29343fd0a558SYan, Zheng if (root) { 293527cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29363fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 29373fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 29383fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 29393fd0a558SYan, Zheng root = root->reloc_root; 29403fd0a558SYan, Zheng node->new_bytenr = root->node->start; 29413fd0a558SYan, Zheng node->root = root; 29423fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 29433fd0a558SYan, Zheng } else { 29445d4f98a2SYan Zheng path->lowest_level = node->level; 29455d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2946b3b4aa74SDavid Sterba btrfs_release_path(path); 29473fd0a558SYan, Zheng if (ret > 0) 29485d4f98a2SYan Zheng ret = 0; 29493fd0a558SYan, Zheng } 29503fd0a558SYan, Zheng if (!ret) 29513fd0a558SYan, Zheng update_processed_blocks(rc, node); 29523fd0a558SYan, Zheng } else { 29533fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 29543fd0a558SYan, Zheng } 29555d4f98a2SYan Zheng out: 29560647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 29573fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 29585d4f98a2SYan Zheng return ret; 29595d4f98a2SYan Zheng } 29605d4f98a2SYan Zheng 29615d4f98a2SYan Zheng /* 29625d4f98a2SYan Zheng * relocate a list of blocks 29635d4f98a2SYan Zheng */ 29645d4f98a2SYan Zheng static noinline_for_stack 29655d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 29665d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 29675d4f98a2SYan Zheng { 29685d4f98a2SYan Zheng struct backref_node *node; 29695d4f98a2SYan Zheng struct btrfs_path *path; 29705d4f98a2SYan Zheng struct tree_block *block; 29715d4f98a2SYan Zheng struct rb_node *rb_node; 29725d4f98a2SYan Zheng int ret; 29735d4f98a2SYan Zheng int err = 0; 29745d4f98a2SYan Zheng 29755d4f98a2SYan Zheng path = btrfs_alloc_path(); 2976e1a12670SLiu Bo if (!path) { 2977e1a12670SLiu Bo err = -ENOMEM; 297834c2b290SDavid Sterba goto out_free_blocks; 2979e1a12670SLiu Bo } 29805d4f98a2SYan Zheng 29815d4f98a2SYan Zheng rb_node = rb_first(blocks); 29825d4f98a2SYan Zheng while (rb_node) { 29835d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 29845d4f98a2SYan Zheng if (!block->key_ready) 2985d3e46feaSDavid Sterba readahead_tree_block(rc->extent_root, block->bytenr); 29865d4f98a2SYan Zheng rb_node = rb_next(rb_node); 29875d4f98a2SYan Zheng } 29885d4f98a2SYan Zheng 29895d4f98a2SYan Zheng rb_node = rb_first(blocks); 29905d4f98a2SYan Zheng while (rb_node) { 29915d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 299234c2b290SDavid Sterba if (!block->key_ready) { 299334c2b290SDavid Sterba err = get_tree_block_key(rc, block); 299434c2b290SDavid Sterba if (err) 299534c2b290SDavid Sterba goto out_free_path; 299634c2b290SDavid Sterba } 29975d4f98a2SYan Zheng rb_node = rb_next(rb_node); 29985d4f98a2SYan Zheng } 29995d4f98a2SYan Zheng 30005d4f98a2SYan Zheng rb_node = rb_first(blocks); 30015d4f98a2SYan Zheng while (rb_node) { 30025d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 30035d4f98a2SYan Zheng 30043fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 30055d4f98a2SYan Zheng block->level, block->bytenr); 30065d4f98a2SYan Zheng if (IS_ERR(node)) { 30075d4f98a2SYan Zheng err = PTR_ERR(node); 30085d4f98a2SYan Zheng goto out; 30095d4f98a2SYan Zheng } 30105d4f98a2SYan Zheng 30115d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 30125d4f98a2SYan Zheng path); 30135d4f98a2SYan Zheng if (ret < 0) { 30143fd0a558SYan, Zheng if (ret != -EAGAIN || rb_node == rb_first(blocks)) 30155d4f98a2SYan Zheng err = ret; 30165d4f98a2SYan Zheng goto out; 30175d4f98a2SYan Zheng } 30185d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30195d4f98a2SYan Zheng } 30205d4f98a2SYan Zheng out: 30213fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 30225d4f98a2SYan Zheng 302334c2b290SDavid Sterba out_free_path: 30245d4f98a2SYan Zheng btrfs_free_path(path); 302534c2b290SDavid Sterba out_free_blocks: 3026e1a12670SLiu Bo free_block_list(blocks); 30275d4f98a2SYan Zheng return err; 30285d4f98a2SYan Zheng } 30295d4f98a2SYan Zheng 30305d4f98a2SYan Zheng static noinline_for_stack 3031efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3032efa56464SYan, Zheng struct file_extent_cluster *cluster) 3033efa56464SYan, Zheng { 3034efa56464SYan, Zheng u64 alloc_hint = 0; 3035efa56464SYan, Zheng u64 start; 3036efa56464SYan, Zheng u64 end; 3037efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3038efa56464SYan, Zheng u64 num_bytes; 3039efa56464SYan, Zheng int nr = 0; 3040efa56464SYan, Zheng int ret = 0; 3041dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3042dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 304318513091SWang Xiaoguang u64 cur_offset; 3044efa56464SYan, Zheng 3045efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 30465955102cSAl Viro inode_lock(inode); 3047efa56464SYan, Zheng 3048dcb40c19SWang Xiaoguang ret = btrfs_check_data_free_space(inode, prealloc_start, 3049dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3050efa56464SYan, Zheng if (ret) 3051efa56464SYan, Zheng goto out; 3052efa56464SYan, Zheng 305318513091SWang Xiaoguang cur_offset = prealloc_start; 3054efa56464SYan, Zheng while (nr < cluster->nr) { 3055efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3056efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3057efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3058efa56464SYan, Zheng else 3059efa56464SYan, Zheng end = cluster->end - offset; 3060efa56464SYan, Zheng 3061d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3062efa56464SYan, Zheng num_bytes = end + 1 - start; 306318513091SWang Xiaoguang if (cur_offset < start) 306418513091SWang Xiaoguang btrfs_free_reserved_data_space(inode, cur_offset, 306518513091SWang Xiaoguang start - cur_offset); 3066efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3067efa56464SYan, Zheng num_bytes, num_bytes, 3068efa56464SYan, Zheng end + 1, &alloc_hint); 306918513091SWang Xiaoguang cur_offset = end + 1; 3070d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3071efa56464SYan, Zheng if (ret) 3072efa56464SYan, Zheng break; 3073efa56464SYan, Zheng nr++; 3074efa56464SYan, Zheng } 307518513091SWang Xiaoguang if (cur_offset < prealloc_end) 307618513091SWang Xiaoguang btrfs_free_reserved_data_space(inode, cur_offset, 307718513091SWang Xiaoguang prealloc_end + 1 - cur_offset); 3078efa56464SYan, Zheng out: 30795955102cSAl Viro inode_unlock(inode); 3080efa56464SYan, Zheng return ret; 3081efa56464SYan, Zheng } 3082efa56464SYan, Zheng 3083efa56464SYan, Zheng static noinline_for_stack 30840257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 30850257bb82SYan, Zheng u64 block_start) 30865d4f98a2SYan Zheng { 30875d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 30885d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 30895d4f98a2SYan Zheng struct extent_map *em; 30900257bb82SYan, Zheng int ret = 0; 30915d4f98a2SYan Zheng 3092172ddd60SDavid Sterba em = alloc_extent_map(); 30930257bb82SYan, Zheng if (!em) 30940257bb82SYan, Zheng return -ENOMEM; 30950257bb82SYan, Zheng 30965d4f98a2SYan Zheng em->start = start; 30970257bb82SYan, Zheng em->len = end + 1 - start; 30980257bb82SYan, Zheng em->block_len = em->len; 30990257bb82SYan, Zheng em->block_start = block_start; 31005d4f98a2SYan Zheng em->bdev = root->fs_info->fs_devices->latest_bdev; 31015d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 31025d4f98a2SYan Zheng 3103d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 31045d4f98a2SYan Zheng while (1) { 3105890871beSChris Mason write_lock(&em_tree->lock); 310609a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3107890871beSChris Mason write_unlock(&em_tree->lock); 31085d4f98a2SYan Zheng if (ret != -EEXIST) { 31095d4f98a2SYan Zheng free_extent_map(em); 31105d4f98a2SYan Zheng break; 31115d4f98a2SYan Zheng } 31125d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 0); 31135d4f98a2SYan Zheng } 3114d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 31150257bb82SYan, Zheng return ret; 31160257bb82SYan, Zheng } 31175d4f98a2SYan Zheng 31180257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 31190257bb82SYan, Zheng struct file_extent_cluster *cluster) 31200257bb82SYan, Zheng { 31210257bb82SYan, Zheng u64 page_start; 31220257bb82SYan, Zheng u64 page_end; 31230257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 31240257bb82SYan, Zheng unsigned long index; 31250257bb82SYan, Zheng unsigned long last_index; 31260257bb82SYan, Zheng struct page *page; 31270257bb82SYan, Zheng struct file_ra_state *ra; 31283b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 31290257bb82SYan, Zheng int nr = 0; 31300257bb82SYan, Zheng int ret = 0; 31310257bb82SYan, Zheng 31320257bb82SYan, Zheng if (!cluster->nr) 31330257bb82SYan, Zheng return 0; 31340257bb82SYan, Zheng 31350257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 31360257bb82SYan, Zheng if (!ra) 31370257bb82SYan, Zheng return -ENOMEM; 31380257bb82SYan, Zheng 3139efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 31400257bb82SYan, Zheng if (ret) 3141efa56464SYan, Zheng goto out; 31420257bb82SYan, Zheng 31430257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 31440257bb82SYan, Zheng 3145efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3146efa56464SYan, Zheng cluster->end - offset, cluster->start); 3147efa56464SYan, Zheng if (ret) 3148efa56464SYan, Zheng goto out; 3149efa56464SYan, Zheng 315009cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 315109cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 31520257bb82SYan, Zheng while (index <= last_index) { 315309cbfeafSKirill A. Shutemov ret = btrfs_delalloc_reserve_metadata(inode, PAGE_SIZE); 3154efa56464SYan, Zheng if (ret) 3155efa56464SYan, Zheng goto out; 3156efa56464SYan, Zheng 31570257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 31580257bb82SYan, Zheng if (!page) { 31590257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 31600257bb82SYan, Zheng ra, NULL, index, 31610257bb82SYan, Zheng last_index + 1 - index); 3162a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 31633b16a4e3SJosef Bacik mask); 31640257bb82SYan, Zheng if (!page) { 3165efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 316609cbfeafSKirill A. Shutemov PAGE_SIZE); 31670257bb82SYan, Zheng ret = -ENOMEM; 3168efa56464SYan, Zheng goto out; 31690257bb82SYan, Zheng } 31700257bb82SYan, Zheng } 31710257bb82SYan, Zheng 31720257bb82SYan, Zheng if (PageReadahead(page)) { 31730257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 31740257bb82SYan, Zheng ra, NULL, page, index, 31750257bb82SYan, Zheng last_index + 1 - index); 31760257bb82SYan, Zheng } 31770257bb82SYan, Zheng 31780257bb82SYan, Zheng if (!PageUptodate(page)) { 31790257bb82SYan, Zheng btrfs_readpage(NULL, page); 31800257bb82SYan, Zheng lock_page(page); 31810257bb82SYan, Zheng if (!PageUptodate(page)) { 31820257bb82SYan, Zheng unlock_page(page); 318309cbfeafSKirill A. Shutemov put_page(page); 3184efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 318509cbfeafSKirill A. Shutemov PAGE_SIZE); 31860257bb82SYan, Zheng ret = -EIO; 3187efa56464SYan, Zheng goto out; 31880257bb82SYan, Zheng } 31890257bb82SYan, Zheng } 31900257bb82SYan, Zheng 31914eee4fa4SMiao Xie page_start = page_offset(page); 319209cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 31930257bb82SYan, Zheng 3194d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 31950257bb82SYan, Zheng 31960257bb82SYan, Zheng set_page_extent_mapped(page); 31970257bb82SYan, Zheng 31980257bb82SYan, Zheng if (nr < cluster->nr && 31990257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 32000257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 32010257bb82SYan, Zheng page_start, page_end, 3202ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 32030257bb82SYan, Zheng nr++; 32040257bb82SYan, Zheng } 32050257bb82SYan, Zheng 3206efa56464SYan, Zheng btrfs_set_extent_delalloc(inode, page_start, page_end, NULL); 32070257bb82SYan, Zheng set_page_dirty(page); 32080257bb82SYan, Zheng 32090257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3210d0082371SJeff Mahoney page_start, page_end); 32110257bb82SYan, Zheng unlock_page(page); 321209cbfeafSKirill A. Shutemov put_page(page); 32130257bb82SYan, Zheng 32140257bb82SYan, Zheng index++; 3215efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 3216efa56464SYan, Zheng btrfs_throttle(BTRFS_I(inode)->root); 32170257bb82SYan, Zheng } 32180257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3219efa56464SYan, Zheng out: 32200257bb82SYan, Zheng kfree(ra); 32210257bb82SYan, Zheng return ret; 32220257bb82SYan, Zheng } 32230257bb82SYan, Zheng 32240257bb82SYan, Zheng static noinline_for_stack 32250257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 32260257bb82SYan, Zheng struct file_extent_cluster *cluster) 32270257bb82SYan, Zheng { 32280257bb82SYan, Zheng int ret; 32290257bb82SYan, Zheng 32300257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 32310257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 32320257bb82SYan, Zheng if (ret) 32330257bb82SYan, Zheng return ret; 32340257bb82SYan, Zheng cluster->nr = 0; 32350257bb82SYan, Zheng } 32360257bb82SYan, Zheng 32370257bb82SYan, Zheng if (!cluster->nr) 32380257bb82SYan, Zheng cluster->start = extent_key->objectid; 32390257bb82SYan, Zheng else 32400257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 32410257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 32420257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 32430257bb82SYan, Zheng cluster->nr++; 32440257bb82SYan, Zheng 32450257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 32460257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 32470257bb82SYan, Zheng if (ret) 32480257bb82SYan, Zheng return ret; 32490257bb82SYan, Zheng cluster->nr = 0; 32500257bb82SYan, Zheng } 32510257bb82SYan, Zheng return 0; 32525d4f98a2SYan Zheng } 32535d4f98a2SYan Zheng 32545d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 32555d4f98a2SYan Zheng static int get_ref_objectid_v0(struct reloc_control *rc, 32565d4f98a2SYan Zheng struct btrfs_path *path, 32575d4f98a2SYan Zheng struct btrfs_key *extent_key, 32585d4f98a2SYan Zheng u64 *ref_objectid, int *path_change) 32595d4f98a2SYan Zheng { 32605d4f98a2SYan Zheng struct btrfs_key key; 32615d4f98a2SYan Zheng struct extent_buffer *leaf; 32625d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 32635d4f98a2SYan Zheng int ret; 32645d4f98a2SYan Zheng int slot; 32655d4f98a2SYan Zheng 32665d4f98a2SYan Zheng leaf = path->nodes[0]; 32675d4f98a2SYan Zheng slot = path->slots[0]; 32685d4f98a2SYan Zheng while (1) { 32695d4f98a2SYan Zheng if (slot >= btrfs_header_nritems(leaf)) { 32705d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 32715d4f98a2SYan Zheng if (ret < 0) 32725d4f98a2SYan Zheng return ret; 32735d4f98a2SYan Zheng BUG_ON(ret > 0); 32745d4f98a2SYan Zheng leaf = path->nodes[0]; 32755d4f98a2SYan Zheng slot = path->slots[0]; 32765d4f98a2SYan Zheng if (path_change) 32775d4f98a2SYan Zheng *path_change = 1; 32785d4f98a2SYan Zheng } 32795d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 32805d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 32815d4f98a2SYan Zheng return -ENOENT; 32825d4f98a2SYan Zheng 32835d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_REF_V0_KEY) { 32845d4f98a2SYan Zheng slot++; 32855d4f98a2SYan Zheng continue; 32865d4f98a2SYan Zheng } 32875d4f98a2SYan Zheng ref0 = btrfs_item_ptr(leaf, slot, 32885d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 32895d4f98a2SYan Zheng *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0); 32905d4f98a2SYan Zheng break; 32915d4f98a2SYan Zheng } 32925d4f98a2SYan Zheng return 0; 32935d4f98a2SYan Zheng } 32945d4f98a2SYan Zheng #endif 32955d4f98a2SYan Zheng 32965d4f98a2SYan Zheng /* 32975d4f98a2SYan Zheng * helper to add a tree block to the list. 32985d4f98a2SYan Zheng * the major work is getting the generation and level of the block 32995d4f98a2SYan Zheng */ 33005d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 33015d4f98a2SYan Zheng struct btrfs_key *extent_key, 33025d4f98a2SYan Zheng struct btrfs_path *path, 33035d4f98a2SYan Zheng struct rb_root *blocks) 33045d4f98a2SYan Zheng { 33055d4f98a2SYan Zheng struct extent_buffer *eb; 33065d4f98a2SYan Zheng struct btrfs_extent_item *ei; 33075d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 33085d4f98a2SYan Zheng struct tree_block *block; 33095d4f98a2SYan Zheng struct rb_node *rb_node; 33105d4f98a2SYan Zheng u32 item_size; 33115d4f98a2SYan Zheng int level = -1; 33127fdf4b60SWang Shilong u64 generation; 33135d4f98a2SYan Zheng 33145d4f98a2SYan Zheng eb = path->nodes[0]; 33155d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 33165d4f98a2SYan Zheng 33173173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 33183173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 33195d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 33205d4f98a2SYan Zheng struct btrfs_extent_item); 33213173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 33225d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 33235d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 33245d4f98a2SYan Zheng } else { 33253173a18fSJosef Bacik level = (int)extent_key->offset; 33263173a18fSJosef Bacik } 33273173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 33283173a18fSJosef Bacik } else { 33295d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 33305d4f98a2SYan Zheng u64 ref_owner; 33315d4f98a2SYan Zheng int ret; 33325d4f98a2SYan Zheng 33335d4f98a2SYan Zheng BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 33345d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, extent_key, 33355d4f98a2SYan Zheng &ref_owner, NULL); 3336411fc6bcSAndi Kleen if (ret < 0) 3337411fc6bcSAndi Kleen return ret; 33385d4f98a2SYan Zheng BUG_ON(ref_owner >= BTRFS_MAX_LEVEL); 33395d4f98a2SYan Zheng level = (int)ref_owner; 33405d4f98a2SYan Zheng /* FIXME: get real generation */ 33415d4f98a2SYan Zheng generation = 0; 33425d4f98a2SYan Zheng #else 33435d4f98a2SYan Zheng BUG(); 33445d4f98a2SYan Zheng #endif 33455d4f98a2SYan Zheng } 33465d4f98a2SYan Zheng 3347b3b4aa74SDavid Sterba btrfs_release_path(path); 33485d4f98a2SYan Zheng 33495d4f98a2SYan Zheng BUG_ON(level == -1); 33505d4f98a2SYan Zheng 33515d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 33525d4f98a2SYan Zheng if (!block) 33535d4f98a2SYan Zheng return -ENOMEM; 33545d4f98a2SYan Zheng 33555d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3356707e8a07SDavid Sterba block->key.objectid = rc->extent_root->nodesize; 33575d4f98a2SYan Zheng block->key.offset = generation; 33585d4f98a2SYan Zheng block->level = level; 33595d4f98a2SYan Zheng block->key_ready = 0; 33605d4f98a2SYan Zheng 33615d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 336243c04fb1SJeff Mahoney if (rb_node) 336343c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 33645d4f98a2SYan Zheng 33655d4f98a2SYan Zheng return 0; 33665d4f98a2SYan Zheng } 33675d4f98a2SYan Zheng 33685d4f98a2SYan Zheng /* 33695d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 33705d4f98a2SYan Zheng */ 33715d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 33725d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 33735d4f98a2SYan Zheng struct rb_root *blocks) 33745d4f98a2SYan Zheng { 33755d4f98a2SYan Zheng struct btrfs_path *path; 33765d4f98a2SYan Zheng struct btrfs_key key; 33775d4f98a2SYan Zheng int ret; 3378aee68ee5SJosef Bacik bool skinny = btrfs_fs_incompat(rc->extent_root->fs_info, 3379aee68ee5SJosef Bacik SKINNY_METADATA); 33805d4f98a2SYan Zheng 33817476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 33825d4f98a2SYan Zheng return 0; 33835d4f98a2SYan Zheng 33845d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 33855d4f98a2SYan Zheng return 0; 33865d4f98a2SYan Zheng 33875d4f98a2SYan Zheng path = btrfs_alloc_path(); 33885d4f98a2SYan Zheng if (!path) 33895d4f98a2SYan Zheng return -ENOMEM; 3390aee68ee5SJosef Bacik again: 33915d4f98a2SYan Zheng key.objectid = bytenr; 3392aee68ee5SJosef Bacik if (skinny) { 3393aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3394aee68ee5SJosef Bacik key.offset = (u64)-1; 3395aee68ee5SJosef Bacik } else { 33965d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 33975d4f98a2SYan Zheng key.offset = blocksize; 3398aee68ee5SJosef Bacik } 33995d4f98a2SYan Zheng 34005d4f98a2SYan Zheng path->search_commit_root = 1; 34015d4f98a2SYan Zheng path->skip_locking = 1; 34025d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 34035d4f98a2SYan Zheng if (ret < 0) 34045d4f98a2SYan Zheng goto out; 34055d4f98a2SYan Zheng 3406aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3407aee68ee5SJosef Bacik if (path->slots[0]) { 3408aee68ee5SJosef Bacik path->slots[0]--; 3409aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3410aee68ee5SJosef Bacik path->slots[0]); 34113173a18fSJosef Bacik if (key.objectid == bytenr && 3412aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3413aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3414aee68ee5SJosef Bacik key.offset == blocksize))) 34153173a18fSJosef Bacik ret = 0; 34163173a18fSJosef Bacik } 3417aee68ee5SJosef Bacik 3418aee68ee5SJosef Bacik if (ret) { 3419aee68ee5SJosef Bacik skinny = false; 3420aee68ee5SJosef Bacik btrfs_release_path(path); 3421aee68ee5SJosef Bacik goto again; 3422aee68ee5SJosef Bacik } 3423aee68ee5SJosef Bacik } 34243173a18fSJosef Bacik BUG_ON(ret); 34253173a18fSJosef Bacik 34265d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 34275d4f98a2SYan Zheng out: 34285d4f98a2SYan Zheng btrfs_free_path(path); 34295d4f98a2SYan Zheng return ret; 34305d4f98a2SYan Zheng } 34315d4f98a2SYan Zheng 34325d4f98a2SYan Zheng /* 34335d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 34345d4f98a2SYan Zheng */ 34355d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 34365d4f98a2SYan Zheng struct extent_buffer *eb) 34375d4f98a2SYan Zheng { 34385d4f98a2SYan Zheng u64 flags; 34395d4f98a2SYan Zheng int ret; 34405d4f98a2SYan Zheng 34415d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 34425d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 34435d4f98a2SYan Zheng return 1; 34445d4f98a2SYan Zheng 34453fd0a558SYan, Zheng ret = btrfs_lookup_extent_info(NULL, rc->extent_root, 34463173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 34473173a18fSJosef Bacik NULL, &flags); 34485d4f98a2SYan Zheng BUG_ON(ret); 34495d4f98a2SYan Zheng 34505d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 34515d4f98a2SYan Zheng ret = 1; 34525d4f98a2SYan Zheng else 34535d4f98a2SYan Zheng ret = 0; 34545d4f98a2SYan Zheng return ret; 34555d4f98a2SYan Zheng } 34565d4f98a2SYan Zheng 34570af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 34581bbc621eSChris Mason struct btrfs_block_group_cache *block_group, 34591bbc621eSChris Mason struct inode *inode, 34601bbc621eSChris Mason u64 ino) 34610af3d00bSJosef Bacik { 34620af3d00bSJosef Bacik struct btrfs_key key; 34630af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 34640af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 34650af3d00bSJosef Bacik int ret = 0; 34660af3d00bSJosef Bacik 34670af3d00bSJosef Bacik if (inode) 34680af3d00bSJosef Bacik goto truncate; 34690af3d00bSJosef Bacik 34700af3d00bSJosef Bacik key.objectid = ino; 34710af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 34720af3d00bSJosef Bacik key.offset = 0; 34730af3d00bSJosef Bacik 34740af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 3475f54fb859STsutomu Itoh if (IS_ERR(inode) || is_bad_inode(inode)) { 3476f54fb859STsutomu Itoh if (!IS_ERR(inode)) 34770af3d00bSJosef Bacik iput(inode); 34780af3d00bSJosef Bacik return -ENOENT; 34790af3d00bSJosef Bacik } 34800af3d00bSJosef Bacik 34810af3d00bSJosef Bacik truncate: 34827b61cd92SMiao Xie ret = btrfs_check_trunc_cache_free_space(root, 34837b61cd92SMiao Xie &fs_info->global_block_rsv); 34847b61cd92SMiao Xie if (ret) 34857b61cd92SMiao Xie goto out; 34867b61cd92SMiao Xie 34877a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 34880af3d00bSJosef Bacik if (IS_ERR(trans)) { 34893612b495STsutomu Itoh ret = PTR_ERR(trans); 34900af3d00bSJosef Bacik goto out; 34910af3d00bSJosef Bacik } 34920af3d00bSJosef Bacik 34931bbc621eSChris Mason ret = btrfs_truncate_free_space_cache(root, trans, block_group, inode); 34940af3d00bSJosef Bacik 34950af3d00bSJosef Bacik btrfs_end_transaction(trans, root); 3496b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 34970af3d00bSJosef Bacik out: 34980af3d00bSJosef Bacik iput(inode); 34990af3d00bSJosef Bacik return ret; 35000af3d00bSJosef Bacik } 35010af3d00bSJosef Bacik 35025d4f98a2SYan Zheng /* 35035d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 35045d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 35055d4f98a2SYan Zheng */ 35065d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 35075d4f98a2SYan Zheng struct btrfs_key *extent_key, 35085d4f98a2SYan Zheng struct extent_buffer *leaf, 35095d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 35105d4f98a2SYan Zheng struct rb_root *blocks) 35115d4f98a2SYan Zheng { 35125d4f98a2SYan Zheng struct btrfs_path *path; 35135d4f98a2SYan Zheng struct tree_block *block; 35145d4f98a2SYan Zheng struct btrfs_root *root; 35155d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 35165d4f98a2SYan Zheng struct rb_node *rb_node; 35175d4f98a2SYan Zheng struct btrfs_key key; 35185d4f98a2SYan Zheng u64 ref_root; 35195d4f98a2SYan Zheng u64 ref_objectid; 35205d4f98a2SYan Zheng u64 ref_offset; 35215d4f98a2SYan Zheng u32 ref_count; 35225d4f98a2SYan Zheng u32 nritems; 35235d4f98a2SYan Zheng int err = 0; 35245d4f98a2SYan Zheng int added = 0; 35255d4f98a2SYan Zheng int counted; 35265d4f98a2SYan Zheng int ret; 35275d4f98a2SYan Zheng 35285d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 35295d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 35305d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 35315d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 35325d4f98a2SYan Zheng 35330af3d00bSJosef Bacik /* 35340af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 35350af3d00bSJosef Bacik * it and redo the search. 35360af3d00bSJosef Bacik */ 35370af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 35380af3d00bSJosef Bacik ret = delete_block_group_cache(rc->extent_root->fs_info, 35391bbc621eSChris Mason rc->block_group, 35400af3d00bSJosef Bacik NULL, ref_objectid); 35410af3d00bSJosef Bacik if (ret != -ENOENT) 35420af3d00bSJosef Bacik return ret; 35430af3d00bSJosef Bacik ret = 0; 35440af3d00bSJosef Bacik } 35450af3d00bSJosef Bacik 35460af3d00bSJosef Bacik path = btrfs_alloc_path(); 35470af3d00bSJosef Bacik if (!path) 35480af3d00bSJosef Bacik return -ENOMEM; 3549e4058b54SDavid Sterba path->reada = READA_FORWARD; 35500af3d00bSJosef Bacik 35515d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, ref_root); 35525d4f98a2SYan Zheng if (IS_ERR(root)) { 35535d4f98a2SYan Zheng err = PTR_ERR(root); 35545d4f98a2SYan Zheng goto out; 35555d4f98a2SYan Zheng } 35565d4f98a2SYan Zheng 35575d4f98a2SYan Zheng key.objectid = ref_objectid; 35585d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 355984850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 356084850e8dSYan, Zheng key.offset = 0; 356184850e8dSYan, Zheng else 356284850e8dSYan, Zheng key.offset = ref_offset; 35635d4f98a2SYan Zheng 35645d4f98a2SYan Zheng path->search_commit_root = 1; 35655d4f98a2SYan Zheng path->skip_locking = 1; 35665d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 35675d4f98a2SYan Zheng if (ret < 0) { 35685d4f98a2SYan Zheng err = ret; 35695d4f98a2SYan Zheng goto out; 35705d4f98a2SYan Zheng } 35715d4f98a2SYan Zheng 35725d4f98a2SYan Zheng leaf = path->nodes[0]; 35735d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 35745d4f98a2SYan Zheng /* 35755d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 35765d4f98a2SYan Zheng * are not counted in 35775d4f98a2SYan Zheng */ 35785d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 35795d4f98a2SYan Zheng counted = 0; 35805d4f98a2SYan Zheng else 35815d4f98a2SYan Zheng counted = 1; 35825d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 35835d4f98a2SYan Zheng if (rb_node) { 35845d4f98a2SYan Zheng if (counted) 35855d4f98a2SYan Zheng added = 1; 35865d4f98a2SYan Zheng else 35875d4f98a2SYan Zheng path->slots[0] = nritems; 35885d4f98a2SYan Zheng } 35895d4f98a2SYan Zheng 35905d4f98a2SYan Zheng while (ref_count > 0) { 35915d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 35925d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 35935d4f98a2SYan Zheng if (ret < 0) { 35945d4f98a2SYan Zheng err = ret; 35955d4f98a2SYan Zheng goto out; 35965d4f98a2SYan Zheng } 3597fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 35985d4f98a2SYan Zheng goto out; 35995d4f98a2SYan Zheng 36005d4f98a2SYan Zheng leaf = path->nodes[0]; 36015d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36025d4f98a2SYan Zheng added = 0; 36035d4f98a2SYan Zheng 36045d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36055d4f98a2SYan Zheng counted = 0; 36065d4f98a2SYan Zheng else 36075d4f98a2SYan Zheng counted = 1; 36085d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36095d4f98a2SYan Zheng if (rb_node) { 36105d4f98a2SYan Zheng if (counted) 36115d4f98a2SYan Zheng added = 1; 36125d4f98a2SYan Zheng else 36135d4f98a2SYan Zheng path->slots[0] = nritems; 36145d4f98a2SYan Zheng } 36155d4f98a2SYan Zheng } 36165d4f98a2SYan Zheng 36175d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3618fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3619fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 36205d4f98a2SYan Zheng break; 36215d4f98a2SYan Zheng 36225d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 36235d4f98a2SYan Zheng struct btrfs_file_extent_item); 36245d4f98a2SYan Zheng 36255d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 36265d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 36275d4f98a2SYan Zheng goto next; 36285d4f98a2SYan Zheng 36295d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 36305d4f98a2SYan Zheng extent_key->objectid) 36315d4f98a2SYan Zheng goto next; 36325d4f98a2SYan Zheng 36335d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 36345d4f98a2SYan Zheng if (key.offset != ref_offset) 36355d4f98a2SYan Zheng goto next; 36365d4f98a2SYan Zheng 36375d4f98a2SYan Zheng if (counted) 36385d4f98a2SYan Zheng ref_count--; 36395d4f98a2SYan Zheng if (added) 36405d4f98a2SYan Zheng goto next; 36415d4f98a2SYan Zheng 36427476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 36435d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 36445d4f98a2SYan Zheng if (!block) { 36455d4f98a2SYan Zheng err = -ENOMEM; 36465d4f98a2SYan Zheng break; 36475d4f98a2SYan Zheng } 36485d4f98a2SYan Zheng block->bytenr = leaf->start; 36495d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 36505d4f98a2SYan Zheng block->level = 0; 36515d4f98a2SYan Zheng block->key_ready = 1; 36525d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 36535d4f98a2SYan Zheng &block->rb_node); 365443c04fb1SJeff Mahoney if (rb_node) 365543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 365643c04fb1SJeff Mahoney block->bytenr); 36575d4f98a2SYan Zheng } 36585d4f98a2SYan Zheng if (counted) 36595d4f98a2SYan Zheng added = 1; 36605d4f98a2SYan Zheng else 36615d4f98a2SYan Zheng path->slots[0] = nritems; 36625d4f98a2SYan Zheng next: 36635d4f98a2SYan Zheng path->slots[0]++; 36645d4f98a2SYan Zheng 36655d4f98a2SYan Zheng } 36665d4f98a2SYan Zheng out: 36675d4f98a2SYan Zheng btrfs_free_path(path); 36685d4f98a2SYan Zheng return err; 36695d4f98a2SYan Zheng } 36705d4f98a2SYan Zheng 36715d4f98a2SYan Zheng /* 36722c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 36735d4f98a2SYan Zheng */ 36745d4f98a2SYan Zheng static noinline_for_stack 36755d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 36765d4f98a2SYan Zheng struct btrfs_key *extent_key, 36775d4f98a2SYan Zheng struct btrfs_path *path, 36785d4f98a2SYan Zheng struct rb_root *blocks) 36795d4f98a2SYan Zheng { 36805d4f98a2SYan Zheng struct btrfs_key key; 36815d4f98a2SYan Zheng struct extent_buffer *eb; 36825d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 36835d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 36845d4f98a2SYan Zheng unsigned long ptr; 36855d4f98a2SYan Zheng unsigned long end; 3686707e8a07SDavid Sterba u32 blocksize = rc->extent_root->nodesize; 3687647f63bdSFilipe David Borba Manana int ret = 0; 36885d4f98a2SYan Zheng int err = 0; 36895d4f98a2SYan Zheng 36905d4f98a2SYan Zheng eb = path->nodes[0]; 36915d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 36925d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 36935d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 36945d4f98a2SYan Zheng if (ptr + sizeof(struct btrfs_extent_item_v0) == end) 36955d4f98a2SYan Zheng ptr = end; 36965d4f98a2SYan Zheng else 36975d4f98a2SYan Zheng #endif 36985d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 36995d4f98a2SYan Zheng 37005d4f98a2SYan Zheng while (ptr < end) { 37015d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 37025d4f98a2SYan Zheng key.type = btrfs_extent_inline_ref_type(eb, iref); 37035d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37045d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 37055d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37065d4f98a2SYan Zheng blocks); 37075d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37085d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 37095d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37105d4f98a2SYan Zheng eb, dref, blocks); 37115d4f98a2SYan Zheng } else { 37125d4f98a2SYan Zheng BUG(); 37135d4f98a2SYan Zheng } 3714647f63bdSFilipe David Borba Manana if (ret) { 3715647f63bdSFilipe David Borba Manana err = ret; 3716647f63bdSFilipe David Borba Manana goto out; 3717647f63bdSFilipe David Borba Manana } 37185d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 37195d4f98a2SYan Zheng } 37205d4f98a2SYan Zheng WARN_ON(ptr > end); 37215d4f98a2SYan Zheng 37225d4f98a2SYan Zheng while (1) { 37235d4f98a2SYan Zheng cond_resched(); 37245d4f98a2SYan Zheng eb = path->nodes[0]; 37255d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 37265d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 37275d4f98a2SYan Zheng if (ret < 0) { 37285d4f98a2SYan Zheng err = ret; 37295d4f98a2SYan Zheng break; 37305d4f98a2SYan Zheng } 37315d4f98a2SYan Zheng if (ret > 0) 37325d4f98a2SYan Zheng break; 37335d4f98a2SYan Zheng eb = path->nodes[0]; 37345d4f98a2SYan Zheng } 37355d4f98a2SYan Zheng 37365d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 37375d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 37385d4f98a2SYan Zheng break; 37395d4f98a2SYan Zheng 37405d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 37415d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY || 37425d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 37435d4f98a2SYan Zheng #else 37445d4f98a2SYan Zheng BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); 37455d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37465d4f98a2SYan Zheng #endif 37475d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37485d4f98a2SYan Zheng blocks); 37495d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37505d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 37515d4f98a2SYan Zheng struct btrfs_extent_data_ref); 37525d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37535d4f98a2SYan Zheng eb, dref, blocks); 37545d4f98a2SYan Zheng } else { 37555d4f98a2SYan Zheng ret = 0; 37565d4f98a2SYan Zheng } 37575d4f98a2SYan Zheng if (ret) { 37585d4f98a2SYan Zheng err = ret; 37595d4f98a2SYan Zheng break; 37605d4f98a2SYan Zheng } 37615d4f98a2SYan Zheng path->slots[0]++; 37625d4f98a2SYan Zheng } 3763647f63bdSFilipe David Borba Manana out: 3764b3b4aa74SDavid Sterba btrfs_release_path(path); 37655d4f98a2SYan Zheng if (err) 37665d4f98a2SYan Zheng free_block_list(blocks); 37675d4f98a2SYan Zheng return err; 37685d4f98a2SYan Zheng } 37695d4f98a2SYan Zheng 37705d4f98a2SYan Zheng /* 37712c016dc2SLiu Bo * helper to find next unprocessed extent 37725d4f98a2SYan Zheng */ 37735d4f98a2SYan Zheng static noinline_for_stack 3774147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 37753fd0a558SYan, Zheng struct btrfs_key *extent_key) 37765d4f98a2SYan Zheng { 37775d4f98a2SYan Zheng struct btrfs_key key; 37785d4f98a2SYan Zheng struct extent_buffer *leaf; 37795d4f98a2SYan Zheng u64 start, end, last; 37805d4f98a2SYan Zheng int ret; 37815d4f98a2SYan Zheng 37825d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 37835d4f98a2SYan Zheng while (1) { 37845d4f98a2SYan Zheng cond_resched(); 37855d4f98a2SYan Zheng if (rc->search_start >= last) { 37865d4f98a2SYan Zheng ret = 1; 37875d4f98a2SYan Zheng break; 37885d4f98a2SYan Zheng } 37895d4f98a2SYan Zheng 37905d4f98a2SYan Zheng key.objectid = rc->search_start; 37915d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 37925d4f98a2SYan Zheng key.offset = 0; 37935d4f98a2SYan Zheng 37945d4f98a2SYan Zheng path->search_commit_root = 1; 37955d4f98a2SYan Zheng path->skip_locking = 1; 37965d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 37975d4f98a2SYan Zheng 0, 0); 37985d4f98a2SYan Zheng if (ret < 0) 37995d4f98a2SYan Zheng break; 38005d4f98a2SYan Zheng next: 38015d4f98a2SYan Zheng leaf = path->nodes[0]; 38025d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38035d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38045d4f98a2SYan Zheng if (ret != 0) 38055d4f98a2SYan Zheng break; 38065d4f98a2SYan Zheng leaf = path->nodes[0]; 38075d4f98a2SYan Zheng } 38085d4f98a2SYan Zheng 38095d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38105d4f98a2SYan Zheng if (key.objectid >= last) { 38115d4f98a2SYan Zheng ret = 1; 38125d4f98a2SYan Zheng break; 38135d4f98a2SYan Zheng } 38145d4f98a2SYan Zheng 38153173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38163173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38173173a18fSJosef Bacik path->slots[0]++; 38183173a18fSJosef Bacik goto next; 38193173a18fSJosef Bacik } 38203173a18fSJosef Bacik 38213173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 38225d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 38235d4f98a2SYan Zheng path->slots[0]++; 38245d4f98a2SYan Zheng goto next; 38255d4f98a2SYan Zheng } 38265d4f98a2SYan Zheng 38273173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 3828707e8a07SDavid Sterba key.objectid + rc->extent_root->nodesize <= 38293173a18fSJosef Bacik rc->search_start) { 38303173a18fSJosef Bacik path->slots[0]++; 38313173a18fSJosef Bacik goto next; 38323173a18fSJosef Bacik } 38333173a18fSJosef Bacik 38345d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 38355d4f98a2SYan Zheng key.objectid, &start, &end, 3836e6138876SJosef Bacik EXTENT_DIRTY, NULL); 38375d4f98a2SYan Zheng 38385d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3839b3b4aa74SDavid Sterba btrfs_release_path(path); 38405d4f98a2SYan Zheng rc->search_start = end + 1; 38415d4f98a2SYan Zheng } else { 38423173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 38435d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 38443173a18fSJosef Bacik else 38453173a18fSJosef Bacik rc->search_start = key.objectid + 3846707e8a07SDavid Sterba rc->extent_root->nodesize; 38473fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 38485d4f98a2SYan Zheng return 0; 38495d4f98a2SYan Zheng } 38505d4f98a2SYan Zheng } 3851b3b4aa74SDavid Sterba btrfs_release_path(path); 38525d4f98a2SYan Zheng return ret; 38535d4f98a2SYan Zheng } 38545d4f98a2SYan Zheng 38555d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 38565d4f98a2SYan Zheng { 38575d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38587585717fSChris Mason 38597585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38605d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 38617585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38625d4f98a2SYan Zheng } 38635d4f98a2SYan Zheng 38645d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 38655d4f98a2SYan Zheng { 38665d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38677585717fSChris Mason 38687585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38695d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 38707585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38715d4f98a2SYan Zheng } 38725d4f98a2SYan Zheng 38735d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 38745d4f98a2SYan Zheng { 38755d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38765d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38775d4f98a2SYan Zheng return 1; 38785d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 38795d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38805d4f98a2SYan Zheng return 1; 38815d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38825d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 38835d4f98a2SYan Zheng return 1; 38845d4f98a2SYan Zheng return 0; 38855d4f98a2SYan Zheng } 38865d4f98a2SYan Zheng 38873fd0a558SYan, Zheng static noinline_for_stack 38883fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 38893fd0a558SYan, Zheng { 38903fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3891ac2fabacSJosef Bacik int ret; 38923fd0a558SYan, Zheng 389366d8f3ddSMiao Xie rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root, 389466d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 38953fd0a558SYan, Zheng if (!rc->block_rsv) 38963fd0a558SYan, Zheng return -ENOMEM; 38973fd0a558SYan, Zheng 38983fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 38993fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 39003fd0a558SYan, Zheng rc->extents_found = 0; 39013fd0a558SYan, Zheng rc->nodes_relocated = 0; 39023fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39030647bf56SWang Shilong rc->reserved_bytes = 0; 39040647bf56SWang Shilong rc->block_rsv->size = rc->extent_root->nodesize * 39050647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3906ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3907ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3908ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3909ac2fabacSJosef Bacik if (ret) 3910ac2fabacSJosef Bacik return ret; 39113fd0a558SYan, Zheng 39123fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39133fd0a558SYan, Zheng set_reloc_control(rc); 39143fd0a558SYan, Zheng 39157a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 391628818947SLiu Bo if (IS_ERR(trans)) { 391728818947SLiu Bo unset_reloc_control(rc); 391828818947SLiu Bo /* 391928818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 392028818947SLiu Bo * cleanup. And callers are responsible to free the above 392128818947SLiu Bo * block rsv. 392228818947SLiu Bo */ 392328818947SLiu Bo return PTR_ERR(trans); 392428818947SLiu Bo } 39253fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 39263fd0a558SYan, Zheng return 0; 39273fd0a558SYan, Zheng } 392876dda93cSYan, Zheng 392962b99540SQu Wenruo /* 393062b99540SQu Wenruo * Qgroup fixer for data chunk relocation. 393162b99540SQu Wenruo * The data relocation is done in the following steps 393262b99540SQu Wenruo * 1) Copy data extents into data reloc tree 393362b99540SQu Wenruo * 2) Create tree reloc tree(special snapshot) for related subvolumes 393462b99540SQu Wenruo * 3) Modify file extents in tree reloc tree 393562b99540SQu Wenruo * 4) Merge tree reloc tree with original fs tree, by swapping tree blocks 393662b99540SQu Wenruo * 393762b99540SQu Wenruo * The problem is, data and tree reloc tree are not accounted to qgroup, 393862b99540SQu Wenruo * and 4) will only info qgroup to track tree blocks change, not file extents 393962b99540SQu Wenruo * in the tree blocks. 394062b99540SQu Wenruo * 394162b99540SQu Wenruo * The good news is, related data extents are all in data reloc tree, so we 394262b99540SQu Wenruo * only need to info qgroup to track all file extents in data reloc tree 394362b99540SQu Wenruo * before commit trans. 394462b99540SQu Wenruo */ 394562b99540SQu Wenruo static int qgroup_fix_relocated_data_extents(struct btrfs_trans_handle *trans, 394662b99540SQu Wenruo struct reloc_control *rc) 394762b99540SQu Wenruo { 394862b99540SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 394962b99540SQu Wenruo struct inode *inode = rc->data_inode; 395062b99540SQu Wenruo struct btrfs_root *data_reloc_root = BTRFS_I(inode)->root; 395162b99540SQu Wenruo struct btrfs_path *path; 395262b99540SQu Wenruo struct btrfs_key key; 395362b99540SQu Wenruo int ret = 0; 395462b99540SQu Wenruo 395562b99540SQu Wenruo if (!fs_info->quota_enabled) 395662b99540SQu Wenruo return 0; 395762b99540SQu Wenruo 395862b99540SQu Wenruo /* 395962b99540SQu Wenruo * Only for stage where we update data pointers the qgroup fix is 396062b99540SQu Wenruo * valid. 396162b99540SQu Wenruo * For MOVING_DATA stage, we will miss the timing of swapping tree 396262b99540SQu Wenruo * blocks, and won't fix it. 396362b99540SQu Wenruo */ 396462b99540SQu Wenruo if (!(rc->stage == UPDATE_DATA_PTRS && rc->extents_found)) 396562b99540SQu Wenruo return 0; 396662b99540SQu Wenruo 396762b99540SQu Wenruo path = btrfs_alloc_path(); 396862b99540SQu Wenruo if (!path) 396962b99540SQu Wenruo return -ENOMEM; 397062b99540SQu Wenruo key.objectid = btrfs_ino(inode); 397162b99540SQu Wenruo key.type = BTRFS_EXTENT_DATA_KEY; 397262b99540SQu Wenruo key.offset = 0; 397362b99540SQu Wenruo 397462b99540SQu Wenruo ret = btrfs_search_slot(NULL, data_reloc_root, &key, path, 0, 0); 397562b99540SQu Wenruo if (ret < 0) 397662b99540SQu Wenruo goto out; 397762b99540SQu Wenruo 397862b99540SQu Wenruo lock_extent(&BTRFS_I(inode)->io_tree, 0, (u64)-1); 397962b99540SQu Wenruo while (1) { 398062b99540SQu Wenruo struct btrfs_file_extent_item *fi; 398162b99540SQu Wenruo 398262b99540SQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 398362b99540SQu Wenruo if (key.objectid > btrfs_ino(inode)) 398462b99540SQu Wenruo break; 398562b99540SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 398662b99540SQu Wenruo goto next; 398762b99540SQu Wenruo fi = btrfs_item_ptr(path->nodes[0], path->slots[0], 398862b99540SQu Wenruo struct btrfs_file_extent_item); 398962b99540SQu Wenruo if (btrfs_file_extent_type(path->nodes[0], fi) != 399062b99540SQu Wenruo BTRFS_FILE_EXTENT_REG) 399162b99540SQu Wenruo goto next; 399262b99540SQu Wenruo ret = btrfs_qgroup_insert_dirty_extent(trans, fs_info, 399362b99540SQu Wenruo btrfs_file_extent_disk_bytenr(path->nodes[0], fi), 399462b99540SQu Wenruo btrfs_file_extent_disk_num_bytes(path->nodes[0], fi), 399562b99540SQu Wenruo GFP_NOFS); 399662b99540SQu Wenruo if (ret < 0) 399762b99540SQu Wenruo break; 399862b99540SQu Wenruo next: 399962b99540SQu Wenruo ret = btrfs_next_item(data_reloc_root, path); 400062b99540SQu Wenruo if (ret < 0) 400162b99540SQu Wenruo break; 400262b99540SQu Wenruo if (ret > 0) { 400362b99540SQu Wenruo ret = 0; 400462b99540SQu Wenruo break; 400562b99540SQu Wenruo } 400662b99540SQu Wenruo } 400762b99540SQu Wenruo unlock_extent(&BTRFS_I(inode)->io_tree, 0 , (u64)-1); 400862b99540SQu Wenruo out: 400962b99540SQu Wenruo btrfs_free_path(path); 401062b99540SQu Wenruo return ret; 401162b99540SQu Wenruo } 401262b99540SQu Wenruo 40135d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 40145d4f98a2SYan Zheng { 40155d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 40165d4f98a2SYan Zheng struct btrfs_key key; 40175d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 40185d4f98a2SYan Zheng struct btrfs_path *path; 40195d4f98a2SYan Zheng struct btrfs_extent_item *ei; 40205d4f98a2SYan Zheng u64 flags; 40215d4f98a2SYan Zheng u32 item_size; 40225d4f98a2SYan Zheng int ret; 40235d4f98a2SYan Zheng int err = 0; 4024c87f08caSChris Mason int progress = 0; 40255d4f98a2SYan Zheng 40265d4f98a2SYan Zheng path = btrfs_alloc_path(); 40273fd0a558SYan, Zheng if (!path) 40285d4f98a2SYan Zheng return -ENOMEM; 4029e4058b54SDavid Sterba path->reada = READA_FORWARD; 40303fd0a558SYan, Zheng 40313fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 40323fd0a558SYan, Zheng if (ret) { 40333fd0a558SYan, Zheng err = ret; 40343fd0a558SYan, Zheng goto out_free; 40352423fdfbSJiri Slaby } 40365d4f98a2SYan Zheng 40375d4f98a2SYan Zheng while (1) { 40380647bf56SWang Shilong rc->reserved_bytes = 0; 40390647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40400647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40410647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40420647bf56SWang Shilong if (ret) { 40430647bf56SWang Shilong err = ret; 40440647bf56SWang Shilong break; 40450647bf56SWang Shilong } 4046c87f08caSChris Mason progress++; 4047a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40480f788c58SLiu Bo if (IS_ERR(trans)) { 40490f788c58SLiu Bo err = PTR_ERR(trans); 40500f788c58SLiu Bo trans = NULL; 40510f788c58SLiu Bo break; 40520f788c58SLiu Bo } 4053c87f08caSChris Mason restart: 40543fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40553fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 40563fd0a558SYan, Zheng continue; 40573fd0a558SYan, Zheng } 40583fd0a558SYan, Zheng 4059147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40605d4f98a2SYan Zheng if (ret < 0) 40615d4f98a2SYan Zheng err = ret; 40625d4f98a2SYan Zheng if (ret != 0) 40635d4f98a2SYan Zheng break; 40645d4f98a2SYan Zheng 40655d4f98a2SYan Zheng rc->extents_found++; 40665d4f98a2SYan Zheng 40675d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40685d4f98a2SYan Zheng struct btrfs_extent_item); 40693fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40705d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40715d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40725d4f98a2SYan Zheng ret = check_extent_flags(flags); 40735d4f98a2SYan Zheng BUG_ON(ret); 40745d4f98a2SYan Zheng 40755d4f98a2SYan Zheng } else { 40765d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 40775d4f98a2SYan Zheng u64 ref_owner; 40785d4f98a2SYan Zheng int path_change = 0; 40795d4f98a2SYan Zheng 40805d4f98a2SYan Zheng BUG_ON(item_size != 40815d4f98a2SYan Zheng sizeof(struct btrfs_extent_item_v0)); 40825d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, &key, &ref_owner, 40835d4f98a2SYan Zheng &path_change); 40844b3576e4SZhaolei if (ret < 0) { 40854b3576e4SZhaolei err = ret; 40864b3576e4SZhaolei break; 40874b3576e4SZhaolei } 40885d4f98a2SYan Zheng if (ref_owner < BTRFS_FIRST_FREE_OBJECTID) 40895d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_TREE_BLOCK; 40905d4f98a2SYan Zheng else 40915d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_DATA; 40925d4f98a2SYan Zheng 40935d4f98a2SYan Zheng if (path_change) { 4094b3b4aa74SDavid Sterba btrfs_release_path(path); 40955d4f98a2SYan Zheng 40965d4f98a2SYan Zheng path->search_commit_root = 1; 40975d4f98a2SYan Zheng path->skip_locking = 1; 40985d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, 40995d4f98a2SYan Zheng &key, path, 0, 0); 41005d4f98a2SYan Zheng if (ret < 0) { 41015d4f98a2SYan Zheng err = ret; 41025d4f98a2SYan Zheng break; 41035d4f98a2SYan Zheng } 41045d4f98a2SYan Zheng BUG_ON(ret > 0); 41055d4f98a2SYan Zheng } 41065d4f98a2SYan Zheng #else 41075d4f98a2SYan Zheng BUG(); 41085d4f98a2SYan Zheng #endif 41095d4f98a2SYan Zheng } 41105d4f98a2SYan Zheng 41115d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 41125d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 41135d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 41145d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41155d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 41165d4f98a2SYan Zheng } else { 4117b3b4aa74SDavid Sterba btrfs_release_path(path); 41185d4f98a2SYan Zheng ret = 0; 41195d4f98a2SYan Zheng } 41205d4f98a2SYan Zheng if (ret < 0) { 41213fd0a558SYan, Zheng err = ret; 41225d4f98a2SYan Zheng break; 41235d4f98a2SYan Zheng } 41245d4f98a2SYan Zheng 41255d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 41265d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 41275d4f98a2SYan Zheng if (ret < 0) { 41281708cc57SWang Shilong /* 41291708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 41301708cc57SWang Shilong * backref cache when committing transaction. 41311708cc57SWang Shilong */ 41321708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 41331708cc57SWang Shilong 41343fd0a558SYan, Zheng if (ret != -EAGAIN) { 41355d4f98a2SYan Zheng err = ret; 41365d4f98a2SYan Zheng break; 41375d4f98a2SYan Zheng } 41383fd0a558SYan, Zheng rc->extents_found--; 41393fd0a558SYan, Zheng rc->search_start = key.objectid; 41403fd0a558SYan, Zheng } 41415d4f98a2SYan Zheng } 41425d4f98a2SYan Zheng 41433fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 4144b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 41453fd0a558SYan, Zheng trans = NULL; 41465d4f98a2SYan Zheng 41475d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 41485d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41495d4f98a2SYan Zheng rc->found_file_extent = 1; 41500257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 41513fd0a558SYan, Zheng &key, &rc->cluster); 41525d4f98a2SYan Zheng if (ret < 0) { 41535d4f98a2SYan Zheng err = ret; 41545d4f98a2SYan Zheng break; 41555d4f98a2SYan Zheng } 41565d4f98a2SYan Zheng } 41575d4f98a2SYan Zheng } 4158c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 4159c87f08caSChris Mason ret = btrfs_force_chunk_alloc(trans, rc->extent_root, 4160c87f08caSChris Mason rc->block_group->flags); 41619689457bSShilong Wang if (ret == 1) { 4162c87f08caSChris Mason err = 0; 4163c87f08caSChris Mason progress = 0; 4164c87f08caSChris Mason goto restart; 4165c87f08caSChris Mason } 4166c87f08caSChris Mason } 41673fd0a558SYan, Zheng 4168b3b4aa74SDavid Sterba btrfs_release_path(path); 416991166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41705d4f98a2SYan Zheng 41715d4f98a2SYan Zheng if (trans) { 41723fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 4173b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 41745d4f98a2SYan Zheng } 41755d4f98a2SYan Zheng 41760257bb82SYan, Zheng if (!err) { 41773fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 41783fd0a558SYan, Zheng &rc->cluster); 41790257bb82SYan, Zheng if (ret < 0) 41800257bb82SYan, Zheng err = ret; 41810257bb82SYan, Zheng } 41820257bb82SYan, Zheng 41833fd0a558SYan, Zheng rc->create_reloc_tree = 0; 41843fd0a558SYan, Zheng set_reloc_control(rc); 41850257bb82SYan, Zheng 41863fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 41873fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 41885d4f98a2SYan Zheng 41893fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 41905d4f98a2SYan Zheng 41915d4f98a2SYan Zheng merge_reloc_roots(rc); 41925d4f98a2SYan Zheng 41933fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 41945d4f98a2SYan Zheng unset_reloc_control(rc); 41953fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 41965d4f98a2SYan Zheng 41975d4f98a2SYan Zheng /* get rid of pinned extents */ 41987a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 419962b99540SQu Wenruo if (IS_ERR(trans)) { 42003612b495STsutomu Itoh err = PTR_ERR(trans); 420162b99540SQu Wenruo goto out_free; 420262b99540SQu Wenruo } 4203a9b1fc85SLiu Bo ret = qgroup_fix_relocated_data_extents(trans, rc); 4204a9b1fc85SLiu Bo if (ret < 0) { 4205a9b1fc85SLiu Bo btrfs_abort_transaction(trans, ret); 4206a9b1fc85SLiu Bo if (!err) 4207a9b1fc85SLiu Bo err = ret; 420862b99540SQu Wenruo goto out_free; 420962b99540SQu Wenruo } 42105d4f98a2SYan Zheng btrfs_commit_transaction(trans, rc->extent_root); 42113fd0a558SYan, Zheng out_free: 42123fd0a558SYan, Zheng btrfs_free_block_rsv(rc->extent_root, rc->block_rsv); 42133fd0a558SYan, Zheng btrfs_free_path(path); 42145d4f98a2SYan Zheng return err; 42155d4f98a2SYan Zheng } 42165d4f98a2SYan Zheng 42175d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 42180257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 42195d4f98a2SYan Zheng { 42205d4f98a2SYan Zheng struct btrfs_path *path; 42215d4f98a2SYan Zheng struct btrfs_inode_item *item; 42225d4f98a2SYan Zheng struct extent_buffer *leaf; 42235d4f98a2SYan Zheng int ret; 42245d4f98a2SYan Zheng 42255d4f98a2SYan Zheng path = btrfs_alloc_path(); 42265d4f98a2SYan Zheng if (!path) 42275d4f98a2SYan Zheng return -ENOMEM; 42285d4f98a2SYan Zheng 42295d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 42305d4f98a2SYan Zheng if (ret) 42315d4f98a2SYan Zheng goto out; 42325d4f98a2SYan Zheng 42335d4f98a2SYan Zheng leaf = path->nodes[0]; 42345d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 42355d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); 42365d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 42370257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 42385d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 42393fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 42403fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 42415d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 42425d4f98a2SYan Zheng out: 42435d4f98a2SYan Zheng btrfs_free_path(path); 42445d4f98a2SYan Zheng return ret; 42455d4f98a2SYan Zheng } 42465d4f98a2SYan Zheng 42475d4f98a2SYan Zheng /* 42485d4f98a2SYan Zheng * helper to create inode for data relocation. 42495d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 42505d4f98a2SYan Zheng */ 42513fd0a558SYan, Zheng static noinline_for_stack 42523fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 42535d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 42545d4f98a2SYan Zheng { 42555d4f98a2SYan Zheng struct inode *inode = NULL; 42565d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42575d4f98a2SYan Zheng struct btrfs_root *root; 42585d4f98a2SYan Zheng struct btrfs_key key; 42594624900dSZhaolei u64 objectid; 42605d4f98a2SYan Zheng int err = 0; 42615d4f98a2SYan Zheng 42625d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 42635d4f98a2SYan Zheng if (IS_ERR(root)) 42645d4f98a2SYan Zheng return ERR_CAST(root); 42655d4f98a2SYan Zheng 4266a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 42673fd0a558SYan, Zheng if (IS_ERR(trans)) 42683fd0a558SYan, Zheng return ERR_CAST(trans); 42695d4f98a2SYan Zheng 4270581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42715d4f98a2SYan Zheng if (err) 42725d4f98a2SYan Zheng goto out; 42735d4f98a2SYan Zheng 42740257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42755d4f98a2SYan Zheng BUG_ON(err); 42765d4f98a2SYan Zheng 42775d4f98a2SYan Zheng key.objectid = objectid; 42785d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 42795d4f98a2SYan Zheng key.offset = 0; 428073f73415SJosef Bacik inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); 42815d4f98a2SYan Zheng BUG_ON(IS_ERR(inode) || is_bad_inode(inode)); 42825d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 42835d4f98a2SYan Zheng 42845d4f98a2SYan Zheng err = btrfs_orphan_add(trans, inode); 42855d4f98a2SYan Zheng out: 42865d4f98a2SYan Zheng btrfs_end_transaction(trans, root); 4287b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 42885d4f98a2SYan Zheng if (err) { 42895d4f98a2SYan Zheng if (inode) 42905d4f98a2SYan Zheng iput(inode); 42915d4f98a2SYan Zheng inode = ERR_PTR(err); 42925d4f98a2SYan Zheng } 42935d4f98a2SYan Zheng return inode; 42945d4f98a2SYan Zheng } 42955d4f98a2SYan Zheng 4296a9995eecSJosef Bacik static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 42973fd0a558SYan, Zheng { 42983fd0a558SYan, Zheng struct reloc_control *rc; 42993fd0a558SYan, Zheng 43003fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 43013fd0a558SYan, Zheng if (!rc) 43023fd0a558SYan, Zheng return NULL; 43033fd0a558SYan, Zheng 43043fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 43053fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 43063fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 4307a9995eecSJosef Bacik extent_io_tree_init(&rc->processed_blocks, 4308a9995eecSJosef Bacik fs_info->btree_inode->i_mapping); 43093fd0a558SYan, Zheng return rc; 43103fd0a558SYan, Zheng } 43113fd0a558SYan, Zheng 43125d4f98a2SYan Zheng /* 43135d4f98a2SYan Zheng * function to relocate all extents in a block group. 43145d4f98a2SYan Zheng */ 43155d4f98a2SYan Zheng int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start) 43165d4f98a2SYan Zheng { 43175d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = extent_root->fs_info; 43185d4f98a2SYan Zheng struct reloc_control *rc; 43190af3d00bSJosef Bacik struct inode *inode; 43200af3d00bSJosef Bacik struct btrfs_path *path; 43215d4f98a2SYan Zheng int ret; 4322f0486c68SYan, Zheng int rw = 0; 43235d4f98a2SYan Zheng int err = 0; 43245d4f98a2SYan Zheng 4325a9995eecSJosef Bacik rc = alloc_reloc_control(fs_info); 43265d4f98a2SYan Zheng if (!rc) 43275d4f98a2SYan Zheng return -ENOMEM; 43285d4f98a2SYan Zheng 4329f0486c68SYan, Zheng rc->extent_root = extent_root; 43303fd0a558SYan, Zheng 43315d4f98a2SYan Zheng rc->block_group = btrfs_lookup_block_group(fs_info, group_start); 43325d4f98a2SYan Zheng BUG_ON(!rc->block_group); 43335d4f98a2SYan Zheng 4334868f401aSZhaolei ret = btrfs_inc_block_group_ro(extent_root, rc->block_group); 4335f0486c68SYan, Zheng if (ret) { 4336f0486c68SYan, Zheng err = ret; 4337f0486c68SYan, Zheng goto out; 4338f0486c68SYan, Zheng } 4339f0486c68SYan, Zheng rw = 1; 4340f0486c68SYan, Zheng 43410af3d00bSJosef Bacik path = btrfs_alloc_path(); 43420af3d00bSJosef Bacik if (!path) { 43430af3d00bSJosef Bacik err = -ENOMEM; 43440af3d00bSJosef Bacik goto out; 43450af3d00bSJosef Bacik } 43460af3d00bSJosef Bacik 43470af3d00bSJosef Bacik inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group, 43480af3d00bSJosef Bacik path); 43490af3d00bSJosef Bacik btrfs_free_path(path); 43500af3d00bSJosef Bacik 43510af3d00bSJosef Bacik if (!IS_ERR(inode)) 43521bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43530af3d00bSJosef Bacik else 43540af3d00bSJosef Bacik ret = PTR_ERR(inode); 43550af3d00bSJosef Bacik 43560af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 43570af3d00bSJosef Bacik err = ret; 43580af3d00bSJosef Bacik goto out; 43590af3d00bSJosef Bacik } 43600af3d00bSJosef Bacik 43615d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 43625d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 43635d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 43645d4f98a2SYan Zheng rc->data_inode = NULL; 43655d4f98a2SYan Zheng goto out; 43665d4f98a2SYan Zheng } 43675d4f98a2SYan Zheng 4368efe120a0SFrank Holton btrfs_info(extent_root->fs_info, "relocating block group %llu flags %llu", 4369c1c9ff7cSGeert Uytterhoeven rc->block_group->key.objectid, rc->block_group->flags); 43705d4f98a2SYan Zheng 43719cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4372f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 4373578def7cSFilipe Manana btrfs_wait_ordered_roots(fs_info, -1, 4374578def7cSFilipe Manana rc->block_group->key.objectid, 4375578def7cSFilipe Manana rc->block_group->key.offset); 43765d4f98a2SYan Zheng 43775d4f98a2SYan Zheng while (1) { 437876dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 43795d4f98a2SYan Zheng ret = relocate_block_group(rc); 438076dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 43815d4f98a2SYan Zheng if (ret < 0) { 43825d4f98a2SYan Zheng err = ret; 43833fd0a558SYan, Zheng goto out; 43845d4f98a2SYan Zheng } 43855d4f98a2SYan Zheng 43865d4f98a2SYan Zheng if (rc->extents_found == 0) 43875d4f98a2SYan Zheng break; 43885d4f98a2SYan Zheng 4389efe120a0SFrank Holton btrfs_info(extent_root->fs_info, "found %llu extents", 4390c1c9ff7cSGeert Uytterhoeven rc->extents_found); 43915d4f98a2SYan Zheng 43925d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 43930ef8b726SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 43940ef8b726SJosef Bacik (u64)-1); 43950ef8b726SJosef Bacik if (ret) { 43960ef8b726SJosef Bacik err = ret; 43970ef8b726SJosef Bacik goto out; 43980ef8b726SJosef Bacik } 43995d4f98a2SYan Zheng invalidate_mapping_pages(rc->data_inode->i_mapping, 44005d4f98a2SYan Zheng 0, -1); 44015d4f98a2SYan Zheng rc->stage = UPDATE_DATA_PTRS; 44025d4f98a2SYan Zheng } 44035d4f98a2SYan Zheng } 44045d4f98a2SYan Zheng 44055d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 44065d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 44075d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 44085d4f98a2SYan Zheng out: 4409f0486c68SYan, Zheng if (err && rw) 4410868f401aSZhaolei btrfs_dec_block_group_ro(extent_root, rc->block_group); 44115d4f98a2SYan Zheng iput(rc->data_inode); 44125d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 44135d4f98a2SYan Zheng kfree(rc); 44145d4f98a2SYan Zheng return err; 44155d4f98a2SYan Zheng } 44165d4f98a2SYan Zheng 441776dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 441876dda93cSYan, Zheng { 441976dda93cSYan, Zheng struct btrfs_trans_handle *trans; 442079787eaaSJeff Mahoney int ret, err; 442176dda93cSYan, Zheng 4422a22285a6SYan, Zheng trans = btrfs_start_transaction(root->fs_info->tree_root, 0); 442379787eaaSJeff Mahoney if (IS_ERR(trans)) 442479787eaaSJeff Mahoney return PTR_ERR(trans); 442576dda93cSYan, Zheng 442676dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 442776dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 442876dda93cSYan, Zheng root->root_item.drop_level = 0; 442976dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 443076dda93cSYan, Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 443176dda93cSYan, Zheng &root->root_key, &root->root_item); 443276dda93cSYan, Zheng 443379787eaaSJeff Mahoney err = btrfs_end_transaction(trans, root->fs_info->tree_root); 443479787eaaSJeff Mahoney if (err) 443579787eaaSJeff Mahoney return err; 443679787eaaSJeff Mahoney return ret; 443776dda93cSYan, Zheng } 443876dda93cSYan, Zheng 44395d4f98a2SYan Zheng /* 44405d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44415d4f98a2SYan Zheng * 44425d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44435d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44445d4f98a2SYan Zheng */ 44455d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44465d4f98a2SYan Zheng { 44475d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 44485d4f98a2SYan Zheng struct btrfs_key key; 44495d4f98a2SYan Zheng struct btrfs_root *fs_root; 44505d4f98a2SYan Zheng struct btrfs_root *reloc_root; 44515d4f98a2SYan Zheng struct btrfs_path *path; 44525d4f98a2SYan Zheng struct extent_buffer *leaf; 44535d4f98a2SYan Zheng struct reloc_control *rc = NULL; 44545d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 44555d4f98a2SYan Zheng int ret; 44565d4f98a2SYan Zheng int err = 0; 44575d4f98a2SYan Zheng 44585d4f98a2SYan Zheng path = btrfs_alloc_path(); 44595d4f98a2SYan Zheng if (!path) 44605d4f98a2SYan Zheng return -ENOMEM; 4461e4058b54SDavid Sterba path->reada = READA_BACK; 44625d4f98a2SYan Zheng 44635d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 44645d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 44655d4f98a2SYan Zheng key.offset = (u64)-1; 44665d4f98a2SYan Zheng 44675d4f98a2SYan Zheng while (1) { 44685d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, 44695d4f98a2SYan Zheng path, 0, 0); 44705d4f98a2SYan Zheng if (ret < 0) { 44715d4f98a2SYan Zheng err = ret; 44725d4f98a2SYan Zheng goto out; 44735d4f98a2SYan Zheng } 44745d4f98a2SYan Zheng if (ret > 0) { 44755d4f98a2SYan Zheng if (path->slots[0] == 0) 44765d4f98a2SYan Zheng break; 44775d4f98a2SYan Zheng path->slots[0]--; 44785d4f98a2SYan Zheng } 44795d4f98a2SYan Zheng leaf = path->nodes[0]; 44805d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4481b3b4aa74SDavid Sterba btrfs_release_path(path); 44825d4f98a2SYan Zheng 44835d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 44845d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 44855d4f98a2SYan Zheng break; 44865d4f98a2SYan Zheng 4487cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 44885d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 44895d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 44905d4f98a2SYan Zheng goto out; 44915d4f98a2SYan Zheng } 44925d4f98a2SYan Zheng 44935d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 44945d4f98a2SYan Zheng 44955d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 44965d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 44975d4f98a2SYan Zheng reloc_root->root_key.offset); 44985d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 449976dda93cSYan, Zheng ret = PTR_ERR(fs_root); 450076dda93cSYan, Zheng if (ret != -ENOENT) { 450176dda93cSYan, Zheng err = ret; 45025d4f98a2SYan Zheng goto out; 45035d4f98a2SYan Zheng } 450479787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 450579787eaaSJeff Mahoney if (ret < 0) { 450679787eaaSJeff Mahoney err = ret; 450779787eaaSJeff Mahoney goto out; 450879787eaaSJeff Mahoney } 450976dda93cSYan, Zheng } 45105d4f98a2SYan Zheng } 45115d4f98a2SYan Zheng 45125d4f98a2SYan Zheng if (key.offset == 0) 45135d4f98a2SYan Zheng break; 45145d4f98a2SYan Zheng 45155d4f98a2SYan Zheng key.offset--; 45165d4f98a2SYan Zheng } 4517b3b4aa74SDavid Sterba btrfs_release_path(path); 45185d4f98a2SYan Zheng 45195d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 45205d4f98a2SYan Zheng goto out; 45215d4f98a2SYan Zheng 4522a9995eecSJosef Bacik rc = alloc_reloc_control(root->fs_info); 45235d4f98a2SYan Zheng if (!rc) { 45245d4f98a2SYan Zheng err = -ENOMEM; 45255d4f98a2SYan Zheng goto out; 45265d4f98a2SYan Zheng } 45275d4f98a2SYan Zheng 45285d4f98a2SYan Zheng rc->extent_root = root->fs_info->extent_root; 45295d4f98a2SYan Zheng 45305d4f98a2SYan Zheng set_reloc_control(rc); 45315d4f98a2SYan Zheng 45327a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45333612b495STsutomu Itoh if (IS_ERR(trans)) { 45343612b495STsutomu Itoh unset_reloc_control(rc); 45353612b495STsutomu Itoh err = PTR_ERR(trans); 45363612b495STsutomu Itoh goto out_free; 45373612b495STsutomu Itoh } 45383fd0a558SYan, Zheng 45393fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45403fd0a558SYan, Zheng 45415d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45425d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45435d4f98a2SYan Zheng struct btrfs_root, root_list); 45445d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45455d4f98a2SYan Zheng 45465d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 45475d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 45485d4f98a2SYan Zheng &rc->reloc_roots); 45495d4f98a2SYan Zheng continue; 45505d4f98a2SYan Zheng } 45515d4f98a2SYan Zheng 45525d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 45535d4f98a2SYan Zheng reloc_root->root_key.offset); 455479787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 455579787eaaSJeff Mahoney err = PTR_ERR(fs_root); 455679787eaaSJeff Mahoney goto out_free; 455779787eaaSJeff Mahoney } 45585d4f98a2SYan Zheng 4559ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 456079787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 45615d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 45625d4f98a2SYan Zheng } 45635d4f98a2SYan Zheng 456479787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 456579787eaaSJeff Mahoney if (err) 456679787eaaSJeff Mahoney goto out_free; 45675d4f98a2SYan Zheng 45685d4f98a2SYan Zheng merge_reloc_roots(rc); 45695d4f98a2SYan Zheng 45705d4f98a2SYan Zheng unset_reloc_control(rc); 45715d4f98a2SYan Zheng 45727a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 457362b99540SQu Wenruo if (IS_ERR(trans)) { 45743612b495STsutomu Itoh err = PTR_ERR(trans); 457562b99540SQu Wenruo goto out_free; 457662b99540SQu Wenruo } 457762b99540SQu Wenruo err = qgroup_fix_relocated_data_extents(trans, rc); 457862b99540SQu Wenruo if (err < 0) { 457962b99540SQu Wenruo btrfs_abort_transaction(trans, err); 458062b99540SQu Wenruo goto out_free; 458162b99540SQu Wenruo } 458279787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 45833612b495STsutomu Itoh out_free: 45845d4f98a2SYan Zheng kfree(rc); 45853612b495STsutomu Itoh out: 4586aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4587aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4588aca1bba6SLiu Bo 45895d4f98a2SYan Zheng btrfs_free_path(path); 45905d4f98a2SYan Zheng 45915d4f98a2SYan Zheng if (err == 0) { 45925d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 45935d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 45945d4f98a2SYan Zheng BTRFS_DATA_RELOC_TREE_OBJECTID); 45955d4f98a2SYan Zheng if (IS_ERR(fs_root)) 45965d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4597d7ce5843SMiao Xie else 459866b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 45995d4f98a2SYan Zheng } 46005d4f98a2SYan Zheng return err; 46015d4f98a2SYan Zheng } 46025d4f98a2SYan Zheng 46035d4f98a2SYan Zheng /* 46045d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 46055d4f98a2SYan Zheng * 46065d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 46075d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 46085d4f98a2SYan Zheng */ 46095d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 46105d4f98a2SYan Zheng { 46115d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 46125d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 46135d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 46145d4f98a2SYan Zheng int ret; 46155d4f98a2SYan Zheng u64 disk_bytenr; 46164577b014SJosef Bacik u64 new_bytenr; 46175d4f98a2SYan Zheng LIST_HEAD(list); 46185d4f98a2SYan Zheng 46195d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 46205d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 46215d4f98a2SYan Zheng 46225d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46235d4f98a2SYan Zheng ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr, 4624a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 462579787eaaSJeff Mahoney if (ret) 462679787eaaSJeff Mahoney goto out; 46275d4f98a2SYan Zheng 46285d4f98a2SYan Zheng while (!list_empty(&list)) { 46295d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46305d4f98a2SYan Zheng list_del_init(&sums->list); 46315d4f98a2SYan Zheng 46324577b014SJosef Bacik /* 46334577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46344577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46354577b014SJosef Bacik * extents but we may have written to say the middle of the 46364577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46374577b014SJosef Bacik * the right disk offset. 46384577b014SJosef Bacik * 46394577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46404577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46414577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46424577b014SJosef Bacik * disk length. 46434577b014SJosef Bacik */ 46444577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 46454577b014SJosef Bacik sums->bytenr = new_bytenr; 46465d4f98a2SYan Zheng 46475d4f98a2SYan Zheng btrfs_add_ordered_sum(inode, ordered, sums); 46485d4f98a2SYan Zheng } 464979787eaaSJeff Mahoney out: 46505d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4651411fc6bcSAndi Kleen return ret; 46525d4f98a2SYan Zheng } 46533fd0a558SYan, Zheng 465483d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 46553fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 46563fd0a558SYan, Zheng struct extent_buffer *cow) 46573fd0a558SYan, Zheng { 46583fd0a558SYan, Zheng struct reloc_control *rc; 46593fd0a558SYan, Zheng struct backref_node *node; 46603fd0a558SYan, Zheng int first_cow = 0; 46613fd0a558SYan, Zheng int level; 466283d4cfd4SJosef Bacik int ret = 0; 46633fd0a558SYan, Zheng 46643fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 46653fd0a558SYan, Zheng if (!rc) 466683d4cfd4SJosef Bacik return 0; 46673fd0a558SYan, Zheng 46683fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46693fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46703fd0a558SYan, Zheng 4671c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4672c974c464SWang Shilong if (buf == root->node) 4673c974c464SWang Shilong __update_reloc_root(root, cow->start); 4674c974c464SWang Shilong } 4675c974c464SWang Shilong 46763fd0a558SYan, Zheng level = btrfs_header_level(buf); 46773fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46783fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 46793fd0a558SYan, Zheng first_cow = 1; 46803fd0a558SYan, Zheng 46813fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 46823fd0a558SYan, Zheng rc->create_reloc_tree) { 46833fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 46843fd0a558SYan, Zheng 46853fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 46863fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 46873fd0a558SYan, Zheng node->new_bytenr != buf->start); 46883fd0a558SYan, Zheng 46893fd0a558SYan, Zheng drop_node_buffer(node); 46903fd0a558SYan, Zheng extent_buffer_get(cow); 46913fd0a558SYan, Zheng node->eb = cow; 46923fd0a558SYan, Zheng node->new_bytenr = cow->start; 46933fd0a558SYan, Zheng 46943fd0a558SYan, Zheng if (!node->pending) { 46953fd0a558SYan, Zheng list_move_tail(&node->list, 46963fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 46973fd0a558SYan, Zheng node->pending = 1; 46983fd0a558SYan, Zheng } 46993fd0a558SYan, Zheng 47003fd0a558SYan, Zheng if (first_cow) 47013fd0a558SYan, Zheng __mark_block_processed(rc, node); 47023fd0a558SYan, Zheng 47033fd0a558SYan, Zheng if (first_cow && level > 0) 47043fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 47053fd0a558SYan, Zheng } 47063fd0a558SYan, Zheng 470783d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 47083fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 470983d4cfd4SJosef Bacik return ret; 47103fd0a558SYan, Zheng } 47113fd0a558SYan, Zheng 47123fd0a558SYan, Zheng /* 47133fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 471401327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 47153fd0a558SYan, Zheng */ 4716147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 47173fd0a558SYan, Zheng u64 *bytes_to_reserve) 47183fd0a558SYan, Zheng { 47193fd0a558SYan, Zheng struct btrfs_root *root; 47203fd0a558SYan, Zheng struct reloc_control *rc; 47213fd0a558SYan, Zheng 47223fd0a558SYan, Zheng root = pending->root; 47233fd0a558SYan, Zheng if (!root->reloc_root) 47243fd0a558SYan, Zheng return; 47253fd0a558SYan, Zheng 47263fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47273fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47283fd0a558SYan, Zheng return; 47293fd0a558SYan, Zheng 47303fd0a558SYan, Zheng root = root->reloc_root; 47313fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47323fd0a558SYan, Zheng /* 47333fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47343fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47353fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47363fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47373fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47383fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 47393fd0a558SYan, Zheng * tree may use more space than it frees. so we need 47403fd0a558SYan, Zheng * reserve extra space. 47413fd0a558SYan, Zheng */ 47423fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47433fd0a558SYan, Zheng } 47443fd0a558SYan, Zheng 47453fd0a558SYan, Zheng /* 47463fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47473fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 47483fd0a558SYan, Zheng */ 474949b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 47503fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 47513fd0a558SYan, Zheng { 47523fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 47533fd0a558SYan, Zheng struct btrfs_root *reloc_root; 47543fd0a558SYan, Zheng struct btrfs_root *new_root; 47553fd0a558SYan, Zheng struct reloc_control *rc; 47563fd0a558SYan, Zheng int ret; 47573fd0a558SYan, Zheng 47583fd0a558SYan, Zheng if (!root->reloc_root) 475949b25e05SJeff Mahoney return 0; 47603fd0a558SYan, Zheng 47613fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47623fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 47633fd0a558SYan, Zheng 47643fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 47653fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 47663fd0a558SYan, Zheng rc->block_rsv, 476725d609f8SJosef Bacik rc->nodes_relocated, 1); 476849b25e05SJeff Mahoney if (ret) 476949b25e05SJeff Mahoney return ret; 47703fd0a558SYan, Zheng } 47713fd0a558SYan, Zheng 47723fd0a558SYan, Zheng new_root = pending->snap; 47733fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47743fd0a558SYan, Zheng new_root->root_key.objectid); 477549b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 477649b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47773fd0a558SYan, Zheng 4778ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4779ffd7b339SJeff Mahoney BUG_ON(ret < 0); 47803fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 47813fd0a558SYan, Zheng 478249b25e05SJeff Mahoney if (rc->create_reloc_tree) 47833fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 478449b25e05SJeff Mahoney return ret; 47853fd0a558SYan, Zheng } 4786