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]; 926*3561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 927*3561b9dbSLiu Bo cur->bytenr) { 928*3561b9dbSLiu Bo btrfs_err(root->fs_info, 929*3561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 930*3561b9dbSLiu Bo cur->bytenr, level - 1, root->objectid, 931*3561b9dbSLiu Bo node_key->objectid, node_key->type, 932*3561b9dbSLiu Bo node_key->offset); 933*3561b9dbSLiu Bo err = -ENOENT; 934*3561b9dbSLiu Bo goto out; 935*3561b9dbSLiu Bo } 9365d4f98a2SYan Zheng lower = cur; 937b6c60c80SJosef Bacik need_check = true; 9385d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9395d4f98a2SYan Zheng if (!path2->nodes[level]) { 94075bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9415d4f98a2SYan Zheng lower->bytenr); 9423fd0a558SYan, Zheng if (should_ignore_root(root)) 9433fd0a558SYan, Zheng list_add(&lower->list, &useless); 9443fd0a558SYan, Zheng else 9455d4f98a2SYan Zheng lower->root = root; 9465d4f98a2SYan Zheng break; 9475d4f98a2SYan Zheng } 9485d4f98a2SYan Zheng 9493fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9505d4f98a2SYan Zheng if (!edge) { 9515d4f98a2SYan Zheng err = -ENOMEM; 9525d4f98a2SYan Zheng goto out; 9535d4f98a2SYan Zheng } 9545d4f98a2SYan Zheng 9555d4f98a2SYan Zheng eb = path2->nodes[level]; 9565d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9575d4f98a2SYan Zheng if (!rb_node) { 9583fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9595d4f98a2SYan Zheng if (!upper) { 9603fd0a558SYan, Zheng free_backref_edge(cache, edge); 9615d4f98a2SYan Zheng err = -ENOMEM; 9625d4f98a2SYan Zheng goto out; 9635d4f98a2SYan Zheng } 9645d4f98a2SYan Zheng upper->bytenr = eb->start; 9655d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9665d4f98a2SYan Zheng upper->level = lower->level + 1; 96727cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 96827cdeb70SMiao Xie &root->state)) 9693fd0a558SYan, Zheng upper->cowonly = 1; 9705d4f98a2SYan Zheng 9715d4f98a2SYan Zheng /* 9725d4f98a2SYan Zheng * if we know the block isn't shared 9735d4f98a2SYan Zheng * we can void checking its backrefs. 9745d4f98a2SYan Zheng */ 9755d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9765d4f98a2SYan Zheng upper->checked = 0; 9775d4f98a2SYan Zheng else 9785d4f98a2SYan Zheng upper->checked = 1; 9795d4f98a2SYan Zheng 9805d4f98a2SYan Zheng /* 9815d4f98a2SYan Zheng * add the block to pending list if we 982b6c60c80SJosef Bacik * need check its backrefs, we only do this once 983b6c60c80SJosef Bacik * while walking up a tree as we will catch 984b6c60c80SJosef Bacik * anything else later on. 9855d4f98a2SYan Zheng */ 986b6c60c80SJosef Bacik if (!upper->checked && need_check) { 987b6c60c80SJosef Bacik need_check = false; 9885d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9895d4f98a2SYan Zheng &list); 990bbe90514SJosef Bacik } else { 991bbe90514SJosef Bacik if (upper->checked) 992bbe90514SJosef Bacik need_check = true; 9935d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 994bbe90514SJosef Bacik } 9955d4f98a2SYan Zheng } else { 9965d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9975d4f98a2SYan Zheng rb_node); 99875bfb9afSJosef Bacik ASSERT(upper->checked); 9995d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 10003fd0a558SYan, Zheng if (!upper->owner) 10013fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 10025d4f98a2SYan Zheng } 10035d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 10045d4f98a2SYan Zheng edge->node[LOWER] = lower; 10053fd0a558SYan, Zheng edge->node[UPPER] = upper; 10065d4f98a2SYan Zheng 10075d4f98a2SYan Zheng if (rb_node) 10085d4f98a2SYan Zheng break; 10095d4f98a2SYan Zheng lower = upper; 10105d4f98a2SYan Zheng upper = NULL; 10115d4f98a2SYan Zheng } 1012b3b4aa74SDavid Sterba btrfs_release_path(path2); 10135d4f98a2SYan Zheng next: 10145d4f98a2SYan Zheng if (ptr < end) { 10155d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 10165d4f98a2SYan Zheng if (ptr >= end) { 10175d4f98a2SYan Zheng WARN_ON(ptr > end); 10185d4f98a2SYan Zheng ptr = 0; 10195d4f98a2SYan Zheng end = 0; 10205d4f98a2SYan Zheng } 10215d4f98a2SYan Zheng } 10225d4f98a2SYan Zheng if (ptr >= end) 10235d4f98a2SYan Zheng path1->slots[0]++; 10245d4f98a2SYan Zheng } 1025b3b4aa74SDavid Sterba btrfs_release_path(path1); 10265d4f98a2SYan Zheng 10275d4f98a2SYan Zheng cur->checked = 1; 10285d4f98a2SYan Zheng WARN_ON(exist); 10295d4f98a2SYan Zheng 10305d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10315d4f98a2SYan Zheng if (!list_empty(&list)) { 10325d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10335d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10345d4f98a2SYan Zheng cur = edge->node[UPPER]; 10355d4f98a2SYan Zheng goto again; 10365d4f98a2SYan Zheng } 10375d4f98a2SYan Zheng 10385d4f98a2SYan Zheng /* 10395d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10405d4f98a2SYan Zheng * into the cache. 10415d4f98a2SYan Zheng */ 104275bfb9afSJosef Bacik ASSERT(node->checked); 10433fd0a558SYan, Zheng cowonly = node->cowonly; 10443fd0a558SYan, Zheng if (!cowonly) { 10453fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10463fd0a558SYan, Zheng &node->rb_node); 104743c04fb1SJeff Mahoney if (rb_node) 104843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10493fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10503fd0a558SYan, Zheng } 10515d4f98a2SYan Zheng 10525d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10535d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10545d4f98a2SYan Zheng 10555d4f98a2SYan Zheng while (!list_empty(&list)) { 10565d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10575d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10585d4f98a2SYan Zheng upper = edge->node[UPPER]; 10593fd0a558SYan, Zheng if (upper->detached) { 10603fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10613fd0a558SYan, Zheng lower = edge->node[LOWER]; 10623fd0a558SYan, Zheng free_backref_edge(cache, edge); 10633fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10643fd0a558SYan, Zheng list_add(&lower->list, &useless); 10653fd0a558SYan, Zheng continue; 10663fd0a558SYan, Zheng } 10675d4f98a2SYan Zheng 10685d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10695d4f98a2SYan Zheng if (upper->lowest) { 10705d4f98a2SYan Zheng list_del_init(&upper->lower); 10715d4f98a2SYan Zheng upper->lowest = 0; 10725d4f98a2SYan Zheng } 10735d4f98a2SYan Zheng 10745d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10755d4f98a2SYan Zheng continue; 10765d4f98a2SYan Zheng } 10775d4f98a2SYan Zheng 107875bfb9afSJosef Bacik if (!upper->checked) { 107975bfb9afSJosef Bacik /* 108075bfb9afSJosef Bacik * Still want to blow up for developers since this is a 108175bfb9afSJosef Bacik * logic bug. 108275bfb9afSJosef Bacik */ 108375bfb9afSJosef Bacik ASSERT(0); 108475bfb9afSJosef Bacik err = -EINVAL; 108575bfb9afSJosef Bacik goto out; 108675bfb9afSJosef Bacik } 108775bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 108875bfb9afSJosef Bacik ASSERT(0); 108975bfb9afSJosef Bacik err = -EINVAL; 109075bfb9afSJosef Bacik goto out; 109175bfb9afSJosef Bacik } 109275bfb9afSJosef Bacik 10933fd0a558SYan, Zheng if (!cowonly) { 10945d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10955d4f98a2SYan Zheng &upper->rb_node); 109643c04fb1SJeff Mahoney if (rb_node) 109743c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 109843c04fb1SJeff Mahoney upper->bytenr); 10993fd0a558SYan, Zheng } 11005d4f98a2SYan Zheng 11015d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11025d4f98a2SYan Zheng 11035d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 11045d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 11055d4f98a2SYan Zheng } 11063fd0a558SYan, Zheng /* 11073fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 11083fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 11093fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 11103fd0a558SYan, Zheng * lookup. 11113fd0a558SYan, Zheng */ 11123fd0a558SYan, Zheng while (!list_empty(&useless)) { 11133fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 11143fd0a558SYan, Zheng list_del_init(&upper->list); 111575bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 11163fd0a558SYan, Zheng if (upper == node) 11173fd0a558SYan, Zheng node = NULL; 11183fd0a558SYan, Zheng if (upper->lowest) { 11193fd0a558SYan, Zheng list_del_init(&upper->lower); 11203fd0a558SYan, Zheng upper->lowest = 0; 11213fd0a558SYan, Zheng } 11223fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 11233fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 11243fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 11253fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11263fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11273fd0a558SYan, Zheng lower = edge->node[LOWER]; 11283fd0a558SYan, Zheng free_backref_edge(cache, edge); 11293fd0a558SYan, Zheng 11303fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11313fd0a558SYan, Zheng list_add(&lower->list, &useless); 11323fd0a558SYan, Zheng } 11333fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11343fd0a558SYan, Zheng if (upper->level > 0) { 11353fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11363fd0a558SYan, Zheng upper->detached = 1; 11373fd0a558SYan, Zheng } else { 11383fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11393fd0a558SYan, Zheng free_backref_node(cache, upper); 11403fd0a558SYan, Zheng } 11413fd0a558SYan, Zheng } 11425d4f98a2SYan Zheng out: 11435d4f98a2SYan Zheng btrfs_free_path(path1); 11445d4f98a2SYan Zheng btrfs_free_path(path2); 11455d4f98a2SYan Zheng if (err) { 11463fd0a558SYan, Zheng while (!list_empty(&useless)) { 11473fd0a558SYan, Zheng lower = list_entry(useless.next, 114875bfb9afSJosef Bacik struct backref_node, list); 114975bfb9afSJosef Bacik list_del_init(&lower->list); 11503fd0a558SYan, Zheng } 115175bfb9afSJosef Bacik while (!list_empty(&list)) { 115275bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 115375bfb9afSJosef Bacik list[UPPER]); 115475bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11553fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 115675bfb9afSJosef Bacik lower = edge->node[LOWER]; 11575d4f98a2SYan Zheng upper = edge->node[UPPER]; 11583fd0a558SYan, Zheng free_backref_edge(cache, edge); 115975bfb9afSJosef Bacik 116075bfb9afSJosef Bacik /* 116175bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 116275bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 116375bfb9afSJosef Bacik */ 116475bfb9afSJosef Bacik if (list_empty(&lower->upper) && 116575bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 116675bfb9afSJosef Bacik list_add(&lower->list, &useless); 116775bfb9afSJosef Bacik 116875bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 116975bfb9afSJosef Bacik continue; 117075bfb9afSJosef Bacik 117101327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 117275bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 117375bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 117475bfb9afSJosef Bacik if (list_empty(&upper->upper)) 117575bfb9afSJosef Bacik list_add(&upper->list, &useless); 117675bfb9afSJosef Bacik } 117775bfb9afSJosef Bacik 117875bfb9afSJosef Bacik while (!list_empty(&useless)) { 117975bfb9afSJosef Bacik lower = list_entry(useless.next, 118075bfb9afSJosef Bacik struct backref_node, list); 118175bfb9afSJosef Bacik list_del_init(&lower->list); 11820fd8c3daSLiu Bo if (lower == node) 11830fd8c3daSLiu Bo node = NULL; 118475bfb9afSJosef Bacik free_backref_node(cache, lower); 11855d4f98a2SYan Zheng } 11860fd8c3daSLiu Bo 11870fd8c3daSLiu Bo free_backref_node(cache, node); 11885d4f98a2SYan Zheng return ERR_PTR(err); 11895d4f98a2SYan Zheng } 119075bfb9afSJosef Bacik ASSERT(!node || !node->detached); 11915d4f98a2SYan Zheng return node; 11925d4f98a2SYan Zheng } 11935d4f98a2SYan Zheng 11945d4f98a2SYan Zheng /* 11953fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11963fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11973fd0a558SYan, Zheng * corresponds to root of source tree 11983fd0a558SYan, Zheng */ 11993fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 12003fd0a558SYan, Zheng struct reloc_control *rc, 12013fd0a558SYan, Zheng struct btrfs_root *src, 12023fd0a558SYan, Zheng struct btrfs_root *dest) 12033fd0a558SYan, Zheng { 12043fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 12053fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 12063fd0a558SYan, Zheng struct backref_node *node = NULL; 12073fd0a558SYan, Zheng struct backref_node *new_node; 12083fd0a558SYan, Zheng struct backref_edge *edge; 12093fd0a558SYan, Zheng struct backref_edge *new_edge; 12103fd0a558SYan, Zheng struct rb_node *rb_node; 12113fd0a558SYan, Zheng 12123fd0a558SYan, Zheng if (cache->last_trans > 0) 12133fd0a558SYan, Zheng update_backref_cache(trans, cache); 12143fd0a558SYan, Zheng 12153fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 12163fd0a558SYan, Zheng if (rb_node) { 12173fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 12183fd0a558SYan, Zheng if (node->detached) 12193fd0a558SYan, Zheng node = NULL; 12203fd0a558SYan, Zheng else 12213fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 12223fd0a558SYan, Zheng } 12233fd0a558SYan, Zheng 12243fd0a558SYan, Zheng if (!node) { 12253fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 12263fd0a558SYan, Zheng reloc_root->commit_root->start); 12273fd0a558SYan, Zheng if (rb_node) { 12283fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 12293fd0a558SYan, Zheng rb_node); 12303fd0a558SYan, Zheng BUG_ON(node->detached); 12313fd0a558SYan, Zheng } 12323fd0a558SYan, Zheng } 12333fd0a558SYan, Zheng 12343fd0a558SYan, Zheng if (!node) 12353fd0a558SYan, Zheng return 0; 12363fd0a558SYan, Zheng 12373fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12383fd0a558SYan, Zheng if (!new_node) 12393fd0a558SYan, Zheng return -ENOMEM; 12403fd0a558SYan, Zheng 12413fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12423fd0a558SYan, Zheng new_node->level = node->level; 12433fd0a558SYan, Zheng new_node->lowest = node->lowest; 12446848ad64SYan, Zheng new_node->checked = 1; 12453fd0a558SYan, Zheng new_node->root = dest; 12463fd0a558SYan, Zheng 12473fd0a558SYan, Zheng if (!node->lowest) { 12483fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12493fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12503fd0a558SYan, Zheng if (!new_edge) 12513fd0a558SYan, Zheng goto fail; 12523fd0a558SYan, Zheng 12533fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12543fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12553fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12563fd0a558SYan, Zheng &new_node->lower); 12573fd0a558SYan, Zheng } 125876b9e23dSMiao Xie } else { 125976b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12603fd0a558SYan, Zheng } 12613fd0a558SYan, Zheng 12623fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12633fd0a558SYan, Zheng &new_node->rb_node); 126443c04fb1SJeff Mahoney if (rb_node) 126543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12663fd0a558SYan, Zheng 12673fd0a558SYan, Zheng if (!new_node->lowest) { 12683fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12693fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12703fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12713fd0a558SYan, Zheng } 12723fd0a558SYan, Zheng } 12733fd0a558SYan, Zheng return 0; 12743fd0a558SYan, Zheng fail: 12753fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12763fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12773fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12783fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12793fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12803fd0a558SYan, Zheng } 12813fd0a558SYan, Zheng free_backref_node(cache, new_node); 12823fd0a558SYan, Zheng return -ENOMEM; 12833fd0a558SYan, Zheng } 12843fd0a558SYan, Zheng 12853fd0a558SYan, Zheng /* 12865d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12875d4f98a2SYan Zheng */ 1288ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12895d4f98a2SYan Zheng { 12905d4f98a2SYan Zheng struct rb_node *rb_node; 12915d4f98a2SYan Zheng struct mapping_node *node; 12925d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 12935d4f98a2SYan Zheng 12945d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1295ffd7b339SJeff Mahoney if (!node) 1296ffd7b339SJeff Mahoney return -ENOMEM; 12975d4f98a2SYan Zheng 12985d4f98a2SYan Zheng node->bytenr = root->node->start; 12995d4f98a2SYan Zheng node->data = root; 13005d4f98a2SYan Zheng 13015d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13025d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13035d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13045d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1305ffd7b339SJeff Mahoney if (rb_node) { 1306ffd7b339SJeff Mahoney btrfs_panic(root->fs_info, -EEXIST, "Duplicate root found " 1307ffd7b339SJeff Mahoney "for start=%llu while inserting into relocation " 1308351fd353SDavid Sterba "tree", node->bytenr); 130923291a04SDan Carpenter kfree(node); 131023291a04SDan Carpenter return -EEXIST; 1311ffd7b339SJeff Mahoney } 13125d4f98a2SYan Zheng 13135d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 13145d4f98a2SYan Zheng return 0; 13155d4f98a2SYan Zheng } 13165d4f98a2SYan Zheng 13175d4f98a2SYan Zheng /* 1318c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 13195d4f98a2SYan Zheng * mapping 13205d4f98a2SYan Zheng */ 1321c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 13225d4f98a2SYan Zheng { 13235d4f98a2SYan Zheng struct rb_node *rb_node; 13245d4f98a2SYan Zheng struct mapping_node *node = NULL; 13255d4f98a2SYan Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 13265d4f98a2SYan Zheng 13275d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13285d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1329c974c464SWang Shilong root->node->start); 1330c974c464SWang Shilong if (rb_node) { 1331c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1332c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1333c974c464SWang Shilong } 1334c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1335c974c464SWang Shilong 1336c974c464SWang Shilong if (!node) 1337c974c464SWang Shilong return; 1338c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1339c974c464SWang Shilong 1340c974c464SWang Shilong spin_lock(&root->fs_info->trans_lock); 1341c974c464SWang Shilong list_del_init(&root->root_list); 1342c974c464SWang Shilong spin_unlock(&root->fs_info->trans_lock); 1343c974c464SWang Shilong kfree(node); 1344c974c464SWang Shilong } 1345c974c464SWang Shilong 1346c974c464SWang Shilong /* 1347c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1348c974c464SWang Shilong * mapping 1349c974c464SWang Shilong */ 1350c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1351c974c464SWang Shilong { 1352c974c464SWang Shilong struct rb_node *rb_node; 1353c974c464SWang Shilong struct mapping_node *node = NULL; 1354c974c464SWang Shilong struct reloc_control *rc = root->fs_info->reloc_ctl; 1355c974c464SWang Shilong 1356c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1357c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1358c974c464SWang Shilong root->node->start); 13595d4f98a2SYan Zheng if (rb_node) { 13605d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13615d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13625d4f98a2SYan Zheng } 13635d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13645d4f98a2SYan Zheng 13658f71f3e0SLiu Bo if (!node) 13668f71f3e0SLiu Bo return 0; 13675d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13685d4f98a2SYan Zheng 13695d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1370c974c464SWang Shilong node->bytenr = new_bytenr; 13715d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13725d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13735d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 137443c04fb1SJeff Mahoney if (rb_node) 137543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13765d4f98a2SYan Zheng return 0; 13775d4f98a2SYan Zheng } 13785d4f98a2SYan Zheng 13793fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13803fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13815d4f98a2SYan Zheng { 13825d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13835d4f98a2SYan Zheng struct extent_buffer *eb; 13845d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13855d4f98a2SYan Zheng struct btrfs_key root_key; 13865bc7247aSMiao Xie u64 last_snap = 0; 13875d4f98a2SYan Zheng int ret; 13885d4f98a2SYan Zheng 13895d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13905d4f98a2SYan Zheng BUG_ON(!root_item); 13915d4f98a2SYan Zheng 13925d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13935d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13943fd0a558SYan, Zheng root_key.offset = objectid; 13955d4f98a2SYan Zheng 13963fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 13973fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 13985d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 13995d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 14005d4f98a2SYan Zheng BUG_ON(ret); 14015d4f98a2SYan Zheng 14025bc7247aSMiao Xie last_snap = btrfs_root_last_snapshot(&root->root_item); 14033fd0a558SYan, Zheng btrfs_set_root_last_snapshot(&root->root_item, 14043fd0a558SYan, Zheng trans->transid - 1); 14053fd0a558SYan, Zheng } else { 14063fd0a558SYan, Zheng /* 14073fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 14083fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 14093fd0a558SYan, Zheng * modified after it was created have RELOC flag 14103fd0a558SYan, Zheng * set in their headers. so it's OK to not update 14113fd0a558SYan, Zheng * the 'last_snapshot'. 14123fd0a558SYan, Zheng */ 14133fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 14143fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 14153fd0a558SYan, Zheng BUG_ON(ret); 14163fd0a558SYan, Zheng } 14173fd0a558SYan, Zheng 14185d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 14195d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14205d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14215d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14223fd0a558SYan, Zheng 14233fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14243fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14253fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14263fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14275d4f98a2SYan Zheng root_item->drop_level = 0; 14285bc7247aSMiao Xie /* 14295bc7247aSMiao Xie * abuse rtransid, it is safe because it is impossible to 14305bc7247aSMiao Xie * receive data into a relocation tree. 14315bc7247aSMiao Xie */ 14325bc7247aSMiao Xie btrfs_set_root_rtransid(root_item, last_snap); 14335bc7247aSMiao Xie btrfs_set_root_otransid(root_item, trans->transid); 14343fd0a558SYan, Zheng } 14355d4f98a2SYan Zheng 14365d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14375d4f98a2SYan Zheng free_extent_buffer(eb); 14385d4f98a2SYan Zheng 14395d4f98a2SYan Zheng ret = btrfs_insert_root(trans, root->fs_info->tree_root, 14405d4f98a2SYan Zheng &root_key, root_item); 14415d4f98a2SYan Zheng BUG_ON(ret); 14425d4f98a2SYan Zheng kfree(root_item); 14435d4f98a2SYan Zheng 1444cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root->fs_info->tree_root, &root_key); 14455d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14465d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14473fd0a558SYan, Zheng return reloc_root; 14483fd0a558SYan, Zheng } 14493fd0a558SYan, Zheng 14503fd0a558SYan, Zheng /* 14513fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 14523fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 14533fd0a558SYan, Zheng */ 14543fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 14553fd0a558SYan, Zheng struct btrfs_root *root) 14563fd0a558SYan, Zheng { 14573fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14583fd0a558SYan, Zheng struct reloc_control *rc = root->fs_info->reloc_ctl; 145920dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14603fd0a558SYan, Zheng int clear_rsv = 0; 1461ffd7b339SJeff Mahoney int ret; 14623fd0a558SYan, Zheng 14633fd0a558SYan, Zheng if (root->reloc_root) { 14643fd0a558SYan, Zheng reloc_root = root->reloc_root; 14653fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14663fd0a558SYan, Zheng return 0; 14673fd0a558SYan, Zheng } 14683fd0a558SYan, Zheng 14693fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14703fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14713fd0a558SYan, Zheng return 0; 14723fd0a558SYan, Zheng 147320dd2cbfSMiao Xie if (!trans->reloc_reserved) { 147420dd2cbfSMiao Xie rsv = trans->block_rsv; 14753fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14763fd0a558SYan, Zheng clear_rsv = 1; 14773fd0a558SYan, Zheng } 14783fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14793fd0a558SYan, Zheng if (clear_rsv) 148020dd2cbfSMiao Xie trans->block_rsv = rsv; 14815d4f98a2SYan Zheng 1482ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1483ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14845d4f98a2SYan Zheng root->reloc_root = reloc_root; 14855d4f98a2SYan Zheng return 0; 14865d4f98a2SYan Zheng } 14875d4f98a2SYan Zheng 14885d4f98a2SYan Zheng /* 14895d4f98a2SYan Zheng * update root item of reloc tree 14905d4f98a2SYan Zheng */ 14915d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 14925d4f98a2SYan Zheng struct btrfs_root *root) 14935d4f98a2SYan Zheng { 14945d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14955d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14965d4f98a2SYan Zheng int ret; 14975d4f98a2SYan Zheng 14985d4f98a2SYan Zheng if (!root->reloc_root) 14997585717fSChris Mason goto out; 15005d4f98a2SYan Zheng 15015d4f98a2SYan Zheng reloc_root = root->reloc_root; 15025d4f98a2SYan Zheng root_item = &reloc_root->root_item; 15035d4f98a2SYan Zheng 15043fd0a558SYan, Zheng if (root->fs_info->reloc_ctl->merge_reloc_tree && 15053fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 15065d4f98a2SYan Zheng root->reloc_root = NULL; 1507c974c464SWang Shilong __del_reloc_root(reloc_root); 15085d4f98a2SYan Zheng } 15095d4f98a2SYan Zheng 15105d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 15115d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 15125d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 15135d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 15145d4f98a2SYan Zheng } 15155d4f98a2SYan Zheng 15165d4f98a2SYan Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 15175d4f98a2SYan Zheng &reloc_root->root_key, root_item); 15185d4f98a2SYan Zheng BUG_ON(ret); 15197585717fSChris Mason 15207585717fSChris Mason out: 15215d4f98a2SYan Zheng return 0; 15225d4f98a2SYan Zheng } 15235d4f98a2SYan Zheng 15245d4f98a2SYan Zheng /* 15255d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15265d4f98a2SYan Zheng * in a subvolume 15275d4f98a2SYan Zheng */ 15285d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15295d4f98a2SYan Zheng { 15305d4f98a2SYan Zheng struct rb_node *node; 15315d4f98a2SYan Zheng struct rb_node *prev; 15325d4f98a2SYan Zheng struct btrfs_inode *entry; 15335d4f98a2SYan Zheng struct inode *inode; 15345d4f98a2SYan Zheng 15355d4f98a2SYan Zheng spin_lock(&root->inode_lock); 15365d4f98a2SYan Zheng again: 15375d4f98a2SYan Zheng node = root->inode_tree.rb_node; 15385d4f98a2SYan Zheng prev = NULL; 15395d4f98a2SYan Zheng while (node) { 15405d4f98a2SYan Zheng prev = node; 15415d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15425d4f98a2SYan Zheng 154333345d01SLi Zefan if (objectid < btrfs_ino(&entry->vfs_inode)) 15445d4f98a2SYan Zheng node = node->rb_left; 154533345d01SLi Zefan else if (objectid > btrfs_ino(&entry->vfs_inode)) 15465d4f98a2SYan Zheng node = node->rb_right; 15475d4f98a2SYan Zheng else 15485d4f98a2SYan Zheng break; 15495d4f98a2SYan Zheng } 15505d4f98a2SYan Zheng if (!node) { 15515d4f98a2SYan Zheng while (prev) { 15525d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 155333345d01SLi Zefan if (objectid <= btrfs_ino(&entry->vfs_inode)) { 15545d4f98a2SYan Zheng node = prev; 15555d4f98a2SYan Zheng break; 15565d4f98a2SYan Zheng } 15575d4f98a2SYan Zheng prev = rb_next(prev); 15585d4f98a2SYan Zheng } 15595d4f98a2SYan Zheng } 15605d4f98a2SYan Zheng while (node) { 15615d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15625d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15635d4f98a2SYan Zheng if (inode) { 15645d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15655d4f98a2SYan Zheng return inode; 15665d4f98a2SYan Zheng } 15675d4f98a2SYan Zheng 156833345d01SLi Zefan objectid = btrfs_ino(&entry->vfs_inode) + 1; 15695d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15705d4f98a2SYan Zheng goto again; 15715d4f98a2SYan Zheng 15725d4f98a2SYan Zheng node = rb_next(node); 15735d4f98a2SYan Zheng } 15745d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15755d4f98a2SYan Zheng return NULL; 15765d4f98a2SYan Zheng } 15775d4f98a2SYan Zheng 15785d4f98a2SYan Zheng static int in_block_group(u64 bytenr, 15795d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group) 15805d4f98a2SYan Zheng { 15815d4f98a2SYan Zheng if (bytenr >= block_group->key.objectid && 15825d4f98a2SYan Zheng bytenr < block_group->key.objectid + block_group->key.offset) 15835d4f98a2SYan Zheng return 1; 15845d4f98a2SYan Zheng return 0; 15855d4f98a2SYan Zheng } 15865d4f98a2SYan Zheng 15875d4f98a2SYan Zheng /* 15885d4f98a2SYan Zheng * get new location of data 15895d4f98a2SYan Zheng */ 15905d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 15915d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 15925d4f98a2SYan Zheng { 15935d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 15945d4f98a2SYan Zheng struct btrfs_path *path; 15955d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15965d4f98a2SYan Zheng struct extent_buffer *leaf; 15975d4f98a2SYan Zheng int ret; 15985d4f98a2SYan Zheng 15995d4f98a2SYan Zheng path = btrfs_alloc_path(); 16005d4f98a2SYan Zheng if (!path) 16015d4f98a2SYan Zheng return -ENOMEM; 16025d4f98a2SYan Zheng 16035d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 160433345d01SLi Zefan ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode), 16055d4f98a2SYan Zheng bytenr, 0); 16065d4f98a2SYan Zheng if (ret < 0) 16075d4f98a2SYan Zheng goto out; 16085d4f98a2SYan Zheng if (ret > 0) { 16095d4f98a2SYan Zheng ret = -ENOENT; 16105d4f98a2SYan Zheng goto out; 16115d4f98a2SYan Zheng } 16125d4f98a2SYan Zheng 16135d4f98a2SYan Zheng leaf = path->nodes[0]; 16145d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 16155d4f98a2SYan Zheng struct btrfs_file_extent_item); 16165d4f98a2SYan Zheng 16175d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 16185d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 16195d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 16205d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 16215d4f98a2SYan Zheng 16225d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 162383d4cfd4SJosef Bacik ret = -EINVAL; 16245d4f98a2SYan Zheng goto out; 16255d4f98a2SYan Zheng } 16265d4f98a2SYan Zheng 16275d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16285d4f98a2SYan Zheng ret = 0; 16295d4f98a2SYan Zheng out: 16305d4f98a2SYan Zheng btrfs_free_path(path); 16315d4f98a2SYan Zheng return ret; 16325d4f98a2SYan Zheng } 16335d4f98a2SYan Zheng 16345d4f98a2SYan Zheng /* 16355d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16365d4f98a2SYan Zheng * the new locations. 16375d4f98a2SYan Zheng */ 16383fd0a558SYan, Zheng static noinline_for_stack 16393fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 16405d4f98a2SYan Zheng struct reloc_control *rc, 16415d4f98a2SYan Zheng struct btrfs_root *root, 16423fd0a558SYan, Zheng struct extent_buffer *leaf) 16435d4f98a2SYan Zheng { 16445d4f98a2SYan Zheng struct btrfs_key key; 16455d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16465d4f98a2SYan Zheng struct inode *inode = NULL; 16475d4f98a2SYan Zheng u64 parent; 16485d4f98a2SYan Zheng u64 bytenr; 16493fd0a558SYan, Zheng u64 new_bytenr = 0; 16505d4f98a2SYan Zheng u64 num_bytes; 16515d4f98a2SYan Zheng u64 end; 16525d4f98a2SYan Zheng u32 nritems; 16535d4f98a2SYan Zheng u32 i; 165483d4cfd4SJosef Bacik int ret = 0; 16555d4f98a2SYan Zheng int first = 1; 16565d4f98a2SYan Zheng int dirty = 0; 16575d4f98a2SYan Zheng 16585d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16595d4f98a2SYan Zheng return 0; 16605d4f98a2SYan Zheng 16615d4f98a2SYan Zheng /* reloc trees always use full backref */ 16625d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16635d4f98a2SYan Zheng parent = leaf->start; 16645d4f98a2SYan Zheng else 16655d4f98a2SYan Zheng parent = 0; 16665d4f98a2SYan Zheng 16675d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16685d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 16695d4f98a2SYan Zheng cond_resched(); 16705d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16715d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16725d4f98a2SYan Zheng continue; 16735d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16745d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16755d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16765d4f98a2SYan Zheng continue; 16775d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16785d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16795d4f98a2SYan Zheng if (bytenr == 0) 16805d4f98a2SYan Zheng continue; 16815d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16825d4f98a2SYan Zheng continue; 16835d4f98a2SYan Zheng 16845d4f98a2SYan Zheng /* 16855d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16865d4f98a2SYan Zheng * to complete and drop the extent cache 16875d4f98a2SYan Zheng */ 16885d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16895d4f98a2SYan Zheng if (first) { 16905d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16915d4f98a2SYan Zheng first = 0; 169233345d01SLi Zefan } else if (inode && btrfs_ino(inode) < key.objectid) { 16933fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16945d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16955d4f98a2SYan Zheng } 169633345d01SLi Zefan if (inode && btrfs_ino(inode) == key.objectid) { 16975d4f98a2SYan Zheng end = key.offset + 16985d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 16995d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 17005d4f98a2SYan Zheng root->sectorsize)); 17015d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 17025d4f98a2SYan Zheng end--; 17035d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1704d0082371SJeff Mahoney key.offset, end); 17055d4f98a2SYan Zheng if (!ret) 17065d4f98a2SYan Zheng continue; 17075d4f98a2SYan Zheng 17085d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, key.offset, end, 17095d4f98a2SYan Zheng 1); 17105d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1711d0082371SJeff Mahoney key.offset, end); 17125d4f98a2SYan Zheng } 17135d4f98a2SYan Zheng } 17145d4f98a2SYan Zheng 17155d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 17165d4f98a2SYan Zheng bytenr, num_bytes); 171783d4cfd4SJosef Bacik if (ret) { 171883d4cfd4SJosef Bacik /* 171983d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 172083d4cfd4SJosef Bacik * in the file extent yet. 172183d4cfd4SJosef Bacik */ 172283d4cfd4SJosef Bacik break; 17233fd0a558SYan, Zheng } 17245d4f98a2SYan Zheng 17255d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17265d4f98a2SYan Zheng dirty = 1; 17275d4f98a2SYan Zheng 17285d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 17295d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, new_bytenr, 17305d4f98a2SYan Zheng num_bytes, parent, 17315d4f98a2SYan Zheng btrfs_header_owner(leaf), 1732b06c4bf5SFilipe Manana key.objectid, key.offset); 173383d4cfd4SJosef Bacik if (ret) { 173466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 173583d4cfd4SJosef Bacik break; 173683d4cfd4SJosef Bacik } 17375d4f98a2SYan Zheng 17385d4f98a2SYan Zheng ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 17395d4f98a2SYan Zheng parent, btrfs_header_owner(leaf), 1740b06c4bf5SFilipe Manana key.objectid, key.offset); 174183d4cfd4SJosef Bacik if (ret) { 174266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 174383d4cfd4SJosef Bacik break; 174483d4cfd4SJosef Bacik } 17455d4f98a2SYan Zheng } 17465d4f98a2SYan Zheng if (dirty) 17475d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17483fd0a558SYan, Zheng if (inode) 17493fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 175083d4cfd4SJosef Bacik return ret; 17515d4f98a2SYan Zheng } 17525d4f98a2SYan Zheng 17535d4f98a2SYan Zheng static noinline_for_stack 17545d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17555d4f98a2SYan Zheng struct btrfs_path *path, int level) 17565d4f98a2SYan Zheng { 17575d4f98a2SYan Zheng struct btrfs_disk_key key1; 17585d4f98a2SYan Zheng struct btrfs_disk_key key2; 17595d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17605d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17615d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17625d4f98a2SYan Zheng } 17635d4f98a2SYan Zheng 17645d4f98a2SYan Zheng /* 17655d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17665d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17675d4f98a2SYan Zheng * reloc tree was create can be replaced. 17685d4f98a2SYan Zheng * 17695d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17705d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17715d4f98a2SYan Zheng * errors, a negative error number is returned. 17725d4f98a2SYan Zheng */ 17733fd0a558SYan, Zheng static noinline_for_stack 17743fd0a558SYan, Zheng int replace_path(struct btrfs_trans_handle *trans, 17755d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17765d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17775d4f98a2SYan Zheng int lowest_level, int max_level) 17785d4f98a2SYan Zheng { 17795d4f98a2SYan Zheng struct extent_buffer *eb; 17805d4f98a2SYan Zheng struct extent_buffer *parent; 17815d4f98a2SYan Zheng struct btrfs_key key; 17825d4f98a2SYan Zheng u64 old_bytenr; 17835d4f98a2SYan Zheng u64 new_bytenr; 17845d4f98a2SYan Zheng u64 old_ptr_gen; 17855d4f98a2SYan Zheng u64 new_ptr_gen; 17865d4f98a2SYan Zheng u64 last_snapshot; 17875d4f98a2SYan Zheng u32 blocksize; 17883fd0a558SYan, Zheng int cow = 0; 17895d4f98a2SYan Zheng int level; 17905d4f98a2SYan Zheng int ret; 17915d4f98a2SYan Zheng int slot; 17925d4f98a2SYan Zheng 17935d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17945d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17955d4f98a2SYan Zheng 17965d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 17973fd0a558SYan, Zheng again: 17985d4f98a2SYan Zheng slot = path->slots[lowest_level]; 17995d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 18005d4f98a2SYan Zheng 18015d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 18025d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 18035d4f98a2SYan Zheng level = btrfs_header_level(eb); 18045d4f98a2SYan Zheng 18055d4f98a2SYan Zheng if (level < lowest_level) { 18065d4f98a2SYan Zheng btrfs_tree_unlock(eb); 18075d4f98a2SYan Zheng free_extent_buffer(eb); 18085d4f98a2SYan Zheng return 0; 18095d4f98a2SYan Zheng } 18105d4f98a2SYan Zheng 18113fd0a558SYan, Zheng if (cow) { 18125d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18135d4f98a2SYan Zheng BUG_ON(ret); 18143fd0a558SYan, Zheng } 18155d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 18165d4f98a2SYan Zheng 18175d4f98a2SYan Zheng if (next_key) { 18185d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18195d4f98a2SYan Zheng next_key->type = (u8)-1; 18205d4f98a2SYan Zheng next_key->offset = (u64)-1; 18215d4f98a2SYan Zheng } 18225d4f98a2SYan Zheng 18235d4f98a2SYan Zheng parent = eb; 18245d4f98a2SYan Zheng while (1) { 18255d4f98a2SYan Zheng level = btrfs_header_level(parent); 18265d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18275d4f98a2SYan Zheng 18285d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 18295d4f98a2SYan Zheng if (ret && slot > 0) 18305d4f98a2SYan Zheng slot--; 18315d4f98a2SYan Zheng 18325d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18335d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18345d4f98a2SYan Zheng 18355d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 1836707e8a07SDavid Sterba blocksize = dest->nodesize; 18375d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 18385d4f98a2SYan Zheng 18395d4f98a2SYan Zheng if (level <= max_level) { 18405d4f98a2SYan Zheng eb = path->nodes[level]; 18415d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18425d4f98a2SYan Zheng path->slots[level]); 18435d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18445d4f98a2SYan Zheng path->slots[level]); 18455d4f98a2SYan Zheng } else { 18465d4f98a2SYan Zheng new_bytenr = 0; 18475d4f98a2SYan Zheng new_ptr_gen = 0; 18485d4f98a2SYan Zheng } 18495d4f98a2SYan Zheng 1850fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18515d4f98a2SYan Zheng ret = level; 18525d4f98a2SYan Zheng break; 18535d4f98a2SYan Zheng } 18545d4f98a2SYan Zheng 18555d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18565d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18573fd0a558SYan, Zheng if (level <= lowest_level) { 18585d4f98a2SYan Zheng ret = 0; 18595d4f98a2SYan Zheng break; 18605d4f98a2SYan Zheng } 18615d4f98a2SYan Zheng 1862ce86cd59SDavid Sterba eb = read_tree_block(dest, old_bytenr, old_ptr_gen); 186364c043deSLiu Bo if (IS_ERR(eb)) { 186464c043deSLiu Bo ret = PTR_ERR(eb); 1865264813acSLiu Bo break; 186664c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 186764c043deSLiu Bo ret = -EIO; 1868416bc658SJosef Bacik free_extent_buffer(eb); 1869379cde74SStefan Behrens break; 1870416bc658SJosef Bacik } 18715d4f98a2SYan Zheng btrfs_tree_lock(eb); 18723fd0a558SYan, Zheng if (cow) { 18735d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18745d4f98a2SYan Zheng slot, &eb); 18755d4f98a2SYan Zheng BUG_ON(ret); 18765d4f98a2SYan Zheng } 18773fd0a558SYan, Zheng btrfs_set_lock_blocking(eb); 18785d4f98a2SYan Zheng 18795d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18805d4f98a2SYan Zheng free_extent_buffer(parent); 18815d4f98a2SYan Zheng 18825d4f98a2SYan Zheng parent = eb; 18835d4f98a2SYan Zheng continue; 18845d4f98a2SYan Zheng } 18855d4f98a2SYan Zheng 18863fd0a558SYan, Zheng if (!cow) { 18873fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18883fd0a558SYan, Zheng free_extent_buffer(parent); 18893fd0a558SYan, Zheng cow = 1; 18903fd0a558SYan, Zheng goto again; 18913fd0a558SYan, Zheng } 18923fd0a558SYan, Zheng 18935d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18945d4f98a2SYan Zheng path->slots[level]); 1895b3b4aa74SDavid Sterba btrfs_release_path(path); 18965d4f98a2SYan Zheng 18975d4f98a2SYan Zheng path->lowest_level = level; 18985d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 18995d4f98a2SYan Zheng path->lowest_level = 0; 19005d4f98a2SYan Zheng BUG_ON(ret); 19015d4f98a2SYan Zheng 19025d4f98a2SYan Zheng /* 19035d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 19045d4f98a2SYan Zheng */ 19055d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 19065d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19075d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19085d4f98a2SYan Zheng 19095d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19105d4f98a2SYan Zheng path->slots[level], old_bytenr); 19115d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19125d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19135d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19145d4f98a2SYan Zheng 19155d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize, 19165d4f98a2SYan Zheng path->nodes[level]->start, 1917b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19185d4f98a2SYan Zheng BUG_ON(ret); 19195d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize, 19205d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 1921b06c4bf5SFilipe Manana 0); 19225d4f98a2SYan Zheng BUG_ON(ret); 19235d4f98a2SYan Zheng 19245d4f98a2SYan Zheng ret = btrfs_free_extent(trans, src, new_bytenr, blocksize, 19255d4f98a2SYan Zheng path->nodes[level]->start, 1926b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19275d4f98a2SYan Zheng BUG_ON(ret); 19285d4f98a2SYan Zheng 19295d4f98a2SYan Zheng ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize, 19305d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 1931b06c4bf5SFilipe Manana 0); 19325d4f98a2SYan Zheng BUG_ON(ret); 19335d4f98a2SYan Zheng 19345d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19355d4f98a2SYan Zheng 19365d4f98a2SYan Zheng ret = level; 19375d4f98a2SYan Zheng break; 19385d4f98a2SYan Zheng } 19395d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19405d4f98a2SYan Zheng free_extent_buffer(parent); 19415d4f98a2SYan Zheng return ret; 19425d4f98a2SYan Zheng } 19435d4f98a2SYan Zheng 19445d4f98a2SYan Zheng /* 19455d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19465d4f98a2SYan Zheng */ 19475d4f98a2SYan Zheng static noinline_for_stack 19485d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19495d4f98a2SYan Zheng int *level) 19505d4f98a2SYan Zheng { 19515d4f98a2SYan Zheng struct extent_buffer *eb; 19525d4f98a2SYan Zheng int i; 19535d4f98a2SYan Zheng u64 last_snapshot; 19545d4f98a2SYan Zheng u32 nritems; 19555d4f98a2SYan Zheng 19565d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19575d4f98a2SYan Zheng 19585d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19595d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19605d4f98a2SYan Zheng path->nodes[i] = NULL; 19615d4f98a2SYan Zheng } 19625d4f98a2SYan Zheng 19635d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 19645d4f98a2SYan Zheng eb = path->nodes[i]; 19655d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19665d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 19675d4f98a2SYan Zheng path->slots[i]++; 19685d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 19695d4f98a2SYan Zheng last_snapshot) 19705d4f98a2SYan Zheng continue; 19715d4f98a2SYan Zheng 19725d4f98a2SYan Zheng *level = i; 19735d4f98a2SYan Zheng return 0; 19745d4f98a2SYan Zheng } 19755d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19765d4f98a2SYan Zheng path->nodes[i] = NULL; 19775d4f98a2SYan Zheng } 19785d4f98a2SYan Zheng return 1; 19795d4f98a2SYan Zheng } 19805d4f98a2SYan Zheng 19815d4f98a2SYan Zheng /* 19825d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 19835d4f98a2SYan Zheng */ 19845d4f98a2SYan Zheng static noinline_for_stack 19855d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19865d4f98a2SYan Zheng int *level) 19875d4f98a2SYan Zheng { 19885d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 19895d4f98a2SYan Zheng int i; 19905d4f98a2SYan Zheng u64 bytenr; 19915d4f98a2SYan Zheng u64 ptr_gen = 0; 19925d4f98a2SYan Zheng u64 last_snapshot; 19935d4f98a2SYan Zheng u32 nritems; 19945d4f98a2SYan Zheng 19955d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19965d4f98a2SYan Zheng 19975d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 19985d4f98a2SYan Zheng eb = path->nodes[i]; 19995d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20005d4f98a2SYan Zheng while (path->slots[i] < nritems) { 20015d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 20025d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 20035d4f98a2SYan Zheng break; 20045d4f98a2SYan Zheng path->slots[i]++; 20055d4f98a2SYan Zheng } 20065d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20075d4f98a2SYan Zheng if (i == *level) 20085d4f98a2SYan Zheng break; 20095d4f98a2SYan Zheng *level = i + 1; 20105d4f98a2SYan Zheng return 0; 20115d4f98a2SYan Zheng } 20125d4f98a2SYan Zheng if (i == 1) { 20135d4f98a2SYan Zheng *level = i; 20145d4f98a2SYan Zheng return 0; 20155d4f98a2SYan Zheng } 20165d4f98a2SYan Zheng 20175d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2018ce86cd59SDavid Sterba eb = read_tree_block(root, bytenr, ptr_gen); 201964c043deSLiu Bo if (IS_ERR(eb)) { 202064c043deSLiu Bo return PTR_ERR(eb); 202164c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2022416bc658SJosef Bacik free_extent_buffer(eb); 2023416bc658SJosef Bacik return -EIO; 2024416bc658SJosef Bacik } 20255d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20265d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20275d4f98a2SYan Zheng path->slots[i - 1] = 0; 20285d4f98a2SYan Zheng } 20295d4f98a2SYan Zheng return 1; 20305d4f98a2SYan Zheng } 20315d4f98a2SYan Zheng 20325d4f98a2SYan Zheng /* 20335d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 20345d4f98a2SYan Zheng * [min_key, max_key) 20355d4f98a2SYan Zheng */ 20365d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 20375d4f98a2SYan Zheng struct btrfs_key *min_key, 20385d4f98a2SYan Zheng struct btrfs_key *max_key) 20395d4f98a2SYan Zheng { 20405d4f98a2SYan Zheng struct inode *inode = NULL; 20415d4f98a2SYan Zheng u64 objectid; 20425d4f98a2SYan Zheng u64 start, end; 204333345d01SLi Zefan u64 ino; 20445d4f98a2SYan Zheng 20455d4f98a2SYan Zheng objectid = min_key->objectid; 20465d4f98a2SYan Zheng while (1) { 20475d4f98a2SYan Zheng cond_resched(); 20485d4f98a2SYan Zheng iput(inode); 20495d4f98a2SYan Zheng 20505d4f98a2SYan Zheng if (objectid > max_key->objectid) 20515d4f98a2SYan Zheng break; 20525d4f98a2SYan Zheng 20535d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20545d4f98a2SYan Zheng if (!inode) 20555d4f98a2SYan Zheng break; 205633345d01SLi Zefan ino = btrfs_ino(inode); 20575d4f98a2SYan Zheng 205833345d01SLi Zefan if (ino > max_key->objectid) { 20595d4f98a2SYan Zheng iput(inode); 20605d4f98a2SYan Zheng break; 20615d4f98a2SYan Zheng } 20625d4f98a2SYan Zheng 206333345d01SLi Zefan objectid = ino + 1; 20645d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 20655d4f98a2SYan Zheng continue; 20665d4f98a2SYan Zheng 206733345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 20685d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 20695d4f98a2SYan Zheng continue; 20705d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 20715d4f98a2SYan Zheng start = 0; 20725d4f98a2SYan Zheng else { 20735d4f98a2SYan Zheng start = min_key->offset; 20745d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(start, root->sectorsize)); 20755d4f98a2SYan Zheng } 20765d4f98a2SYan Zheng } else { 20775d4f98a2SYan Zheng start = 0; 20785d4f98a2SYan Zheng } 20795d4f98a2SYan Zheng 208033345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 20815d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 20825d4f98a2SYan Zheng continue; 20835d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 20845d4f98a2SYan Zheng end = (u64)-1; 20855d4f98a2SYan Zheng } else { 20865d4f98a2SYan Zheng if (max_key->offset == 0) 20875d4f98a2SYan Zheng continue; 20885d4f98a2SYan Zheng end = max_key->offset; 20895d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(end, root->sectorsize)); 20905d4f98a2SYan Zheng end--; 20915d4f98a2SYan Zheng } 20925d4f98a2SYan Zheng } else { 20935d4f98a2SYan Zheng end = (u64)-1; 20945d4f98a2SYan Zheng } 20955d4f98a2SYan Zheng 20965d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2097d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 20985d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 1); 2099d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 21005d4f98a2SYan Zheng } 21015d4f98a2SYan Zheng return 0; 21025d4f98a2SYan Zheng } 21035d4f98a2SYan Zheng 21045d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 21055d4f98a2SYan Zheng struct btrfs_key *key) 21065d4f98a2SYan Zheng 21075d4f98a2SYan Zheng { 21085d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21095d4f98a2SYan Zheng if (!path->nodes[level]) 21105d4f98a2SYan Zheng break; 21115d4f98a2SYan Zheng if (path->slots[level] + 1 < 21125d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21135d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21145d4f98a2SYan Zheng path->slots[level] + 1); 21155d4f98a2SYan Zheng return 0; 21165d4f98a2SYan Zheng } 21175d4f98a2SYan Zheng level++; 21185d4f98a2SYan Zheng } 21195d4f98a2SYan Zheng return 1; 21205d4f98a2SYan Zheng } 21215d4f98a2SYan Zheng 21225d4f98a2SYan Zheng /* 21235d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 21245d4f98a2SYan Zheng * fs tree. 21255d4f98a2SYan Zheng */ 21265d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 21275d4f98a2SYan Zheng struct btrfs_root *root) 21285d4f98a2SYan Zheng { 21295d4f98a2SYan Zheng LIST_HEAD(inode_list); 21305d4f98a2SYan Zheng struct btrfs_key key; 21315d4f98a2SYan Zheng struct btrfs_key next_key; 21329e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 21335d4f98a2SYan Zheng struct btrfs_root *reloc_root; 21345d4f98a2SYan Zheng struct btrfs_root_item *root_item; 21355d4f98a2SYan Zheng struct btrfs_path *path; 21363fd0a558SYan, Zheng struct extent_buffer *leaf; 21375d4f98a2SYan Zheng int level; 21385d4f98a2SYan Zheng int max_level; 21395d4f98a2SYan Zheng int replaced = 0; 21405d4f98a2SYan Zheng int ret; 21415d4f98a2SYan Zheng int err = 0; 21423fd0a558SYan, Zheng u32 min_reserved; 21435d4f98a2SYan Zheng 21445d4f98a2SYan Zheng path = btrfs_alloc_path(); 21455d4f98a2SYan Zheng if (!path) 21465d4f98a2SYan Zheng return -ENOMEM; 2147e4058b54SDavid Sterba path->reada = READA_FORWARD; 21485d4f98a2SYan Zheng 21495d4f98a2SYan Zheng reloc_root = root->reloc_root; 21505d4f98a2SYan Zheng root_item = &reloc_root->root_item; 21515d4f98a2SYan Zheng 21525d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 21535d4f98a2SYan Zheng level = btrfs_root_level(root_item); 21545d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 21555d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 21565d4f98a2SYan Zheng path->slots[level] = 0; 21575d4f98a2SYan Zheng } else { 21585d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 21595d4f98a2SYan Zheng 21605d4f98a2SYan Zheng level = root_item->drop_level; 21615d4f98a2SYan Zheng BUG_ON(level == 0); 21625d4f98a2SYan Zheng path->lowest_level = level; 21635d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 216433c66f43SYan Zheng path->lowest_level = 0; 21655d4f98a2SYan Zheng if (ret < 0) { 21665d4f98a2SYan Zheng btrfs_free_path(path); 21675d4f98a2SYan Zheng return ret; 21685d4f98a2SYan Zheng } 21695d4f98a2SYan Zheng 21705d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 21715d4f98a2SYan Zheng path->slots[level]); 21725d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 21735d4f98a2SYan Zheng 21745d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 21755d4f98a2SYan Zheng } 21765d4f98a2SYan Zheng 21773fd0a558SYan, Zheng min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 21785d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 21795d4f98a2SYan Zheng 21805d4f98a2SYan Zheng while (1) { 218108e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 218208e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 21833fd0a558SYan, Zheng if (ret) { 21849e6a0c52SJosef Bacik err = ret; 21859e6a0c52SJosef Bacik goto out; 21863fd0a558SYan, Zheng } 21879e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 21889e6a0c52SJosef Bacik if (IS_ERR(trans)) { 21899e6a0c52SJosef Bacik err = PTR_ERR(trans); 21909e6a0c52SJosef Bacik trans = NULL; 21919e6a0c52SJosef Bacik goto out; 21929e6a0c52SJosef Bacik } 21939e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 21943fd0a558SYan, Zheng 21953fd0a558SYan, Zheng replaced = 0; 21965d4f98a2SYan Zheng max_level = level; 21975d4f98a2SYan Zheng 21985d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 21995d4f98a2SYan Zheng if (ret < 0) { 22005d4f98a2SYan Zheng err = ret; 22015d4f98a2SYan Zheng goto out; 22025d4f98a2SYan Zheng } 22035d4f98a2SYan Zheng if (ret > 0) 22045d4f98a2SYan Zheng break; 22055d4f98a2SYan Zheng 22065d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 22075d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 22085d4f98a2SYan Zheng ret = 0; 22095d4f98a2SYan Zheng } else { 22103fd0a558SYan, Zheng ret = replace_path(trans, root, reloc_root, path, 22113fd0a558SYan, Zheng &next_key, level, max_level); 22125d4f98a2SYan Zheng } 22135d4f98a2SYan Zheng if (ret < 0) { 22145d4f98a2SYan Zheng err = ret; 22155d4f98a2SYan Zheng goto out; 22165d4f98a2SYan Zheng } 22175d4f98a2SYan Zheng 22185d4f98a2SYan Zheng if (ret > 0) { 22195d4f98a2SYan Zheng level = ret; 22205d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 22215d4f98a2SYan Zheng path->slots[level]); 22225d4f98a2SYan Zheng replaced = 1; 22235d4f98a2SYan Zheng } 22245d4f98a2SYan Zheng 22255d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 22265d4f98a2SYan Zheng if (ret > 0) 22275d4f98a2SYan Zheng break; 22285d4f98a2SYan Zheng 22295d4f98a2SYan Zheng BUG_ON(level == 0); 22305d4f98a2SYan Zheng /* 22315d4f98a2SYan Zheng * save the merging progress in the drop_progress. 22325d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 22335d4f98a2SYan Zheng */ 22345d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 22355d4f98a2SYan Zheng path->slots[level]); 22365d4f98a2SYan Zheng root_item->drop_level = level; 22375d4f98a2SYan Zheng 22383fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 22399e6a0c52SJosef Bacik trans = NULL; 22405d4f98a2SYan Zheng 2241b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 22425d4f98a2SYan Zheng 22435d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22445d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22455d4f98a2SYan Zheng } 22465d4f98a2SYan Zheng 22475d4f98a2SYan Zheng /* 22485d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 22495d4f98a2SYan Zheng * relocated and the block is tree root. 22505d4f98a2SYan Zheng */ 22515d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 22525d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 22535d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 22545d4f98a2SYan Zheng free_extent_buffer(leaf); 22555d4f98a2SYan Zheng if (ret < 0) 22565d4f98a2SYan Zheng err = ret; 22575d4f98a2SYan Zheng out: 22585d4f98a2SYan Zheng btrfs_free_path(path); 22595d4f98a2SYan Zheng 22605d4f98a2SYan Zheng if (err == 0) { 22615d4f98a2SYan Zheng memset(&root_item->drop_progress, 0, 22625d4f98a2SYan Zheng sizeof(root_item->drop_progress)); 22635d4f98a2SYan Zheng root_item->drop_level = 0; 22645d4f98a2SYan Zheng btrfs_set_root_refs(root_item, 0); 22653fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 22665d4f98a2SYan Zheng } 22675d4f98a2SYan Zheng 22689e6a0c52SJosef Bacik if (trans) 22693fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, root); 22705d4f98a2SYan Zheng 2271b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 22725d4f98a2SYan Zheng 22735d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22745d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22755d4f98a2SYan Zheng 22765d4f98a2SYan Zheng return err; 22775d4f98a2SYan Zheng } 22785d4f98a2SYan Zheng 22793fd0a558SYan, Zheng static noinline_for_stack 22803fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 22815d4f98a2SYan Zheng { 22823fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 22833fd0a558SYan, Zheng struct btrfs_root *reloc_root; 22845d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 22853fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 22863fd0a558SYan, Zheng u64 num_bytes = 0; 22873fd0a558SYan, Zheng int ret; 22883fd0a558SYan, Zheng 22897585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 22903fd0a558SYan, Zheng rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22913fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 22927585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 22937585717fSChris Mason 22943fd0a558SYan, Zheng again: 22953fd0a558SYan, Zheng if (!err) { 22963fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 229708e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 229808e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22993fd0a558SYan, Zheng if (ret) 23003fd0a558SYan, Zheng err = ret; 23013fd0a558SYan, Zheng } 23023fd0a558SYan, Zheng 23037a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 23043612b495STsutomu Itoh if (IS_ERR(trans)) { 23053612b495STsutomu Itoh if (!err) 23063612b495STsutomu Itoh btrfs_block_rsv_release(rc->extent_root, 23073612b495STsutomu Itoh rc->block_rsv, num_bytes); 23083612b495STsutomu Itoh return PTR_ERR(trans); 23093612b495STsutomu Itoh } 23103fd0a558SYan, Zheng 23113fd0a558SYan, Zheng if (!err) { 23123fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 23133fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 23143fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, 23153fd0a558SYan, Zheng rc->block_rsv, num_bytes); 23163fd0a558SYan, Zheng goto again; 23173fd0a558SYan, Zheng } 23183fd0a558SYan, Zheng } 23193fd0a558SYan, Zheng 23203fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 23213fd0a558SYan, Zheng 23223fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 23233fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 23243fd0a558SYan, Zheng struct btrfs_root, root_list); 23253fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 23263fd0a558SYan, Zheng 23273fd0a558SYan, Zheng root = read_fs_root(reloc_root->fs_info, 23283fd0a558SYan, Zheng reloc_root->root_key.offset); 23293fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 23303fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 23313fd0a558SYan, Zheng 23323fd0a558SYan, Zheng /* 23333fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 23343fd0a558SYan, Zheng * knows it should resumes merging 23353fd0a558SYan, Zheng */ 23363fd0a558SYan, Zheng if (!err) 23373fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 23383fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 23393fd0a558SYan, Zheng 23403fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 23413fd0a558SYan, Zheng } 23423fd0a558SYan, Zheng 23433fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 23443fd0a558SYan, Zheng 23453fd0a558SYan, Zheng if (!err) 23463fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 23473fd0a558SYan, Zheng else 23483fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 23493fd0a558SYan, Zheng return err; 23503fd0a558SYan, Zheng } 23513fd0a558SYan, Zheng 23523fd0a558SYan, Zheng static noinline_for_stack 2353aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2354aca1bba6SLiu Bo { 2355aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2356aca1bba6SLiu Bo 2357aca1bba6SLiu Bo while (!list_empty(list)) { 2358aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2359aca1bba6SLiu Bo root_list); 2360c974c464SWang Shilong __del_reloc_root(reloc_root); 2361aca1bba6SLiu Bo } 2362aca1bba6SLiu Bo } 2363aca1bba6SLiu Bo 2364aca1bba6SLiu Bo static noinline_for_stack 236594404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 23663fd0a558SYan, Zheng { 23675d4f98a2SYan Zheng struct btrfs_root *root; 23685d4f98a2SYan Zheng struct btrfs_root *reloc_root; 23695bc7247aSMiao Xie u64 last_snap; 23705bc7247aSMiao Xie u64 otransid; 23715bc7247aSMiao Xie u64 objectid; 23723fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23733fd0a558SYan, Zheng int found = 0; 2374aca1bba6SLiu Bo int ret = 0; 23753fd0a558SYan, Zheng again: 23763fd0a558SYan, Zheng root = rc->extent_root; 23777585717fSChris Mason 23787585717fSChris Mason /* 23797585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 23807585717fSChris Mason * we have to make sure nobody is in the middle of 23817585717fSChris Mason * adding their roots to the list while we are 23827585717fSChris Mason * doing this splice 23837585717fSChris Mason */ 23847585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 23853fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 23867585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 23875d4f98a2SYan Zheng 23883fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 23893fd0a558SYan, Zheng found = 1; 23903fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 23913fd0a558SYan, Zheng struct btrfs_root, root_list); 23925d4f98a2SYan Zheng 23935d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 23945d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, 23955d4f98a2SYan Zheng reloc_root->root_key.offset); 23965d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 23975d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 23985d4f98a2SYan Zheng 23993fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2400b37b39cdSJosef Bacik if (ret) { 240125e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 240225e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 240325e293c2SWang Shilong &reloc_roots); 2404aca1bba6SLiu Bo goto out; 2405b37b39cdSJosef Bacik } 24063fd0a558SYan, Zheng } else { 24073fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24083fd0a558SYan, Zheng } 24095bc7247aSMiao Xie 24105bc7247aSMiao Xie /* 241101327610SNicholas D Steeves * we keep the old last snapshot transid in rtranid when we 24125bc7247aSMiao Xie * created the relocation tree. 24135bc7247aSMiao Xie */ 24145bc7247aSMiao Xie last_snap = btrfs_root_rtransid(&reloc_root->root_item); 24155bc7247aSMiao Xie otransid = btrfs_root_otransid(&reloc_root->root_item); 24165bc7247aSMiao Xie objectid = reloc_root->root_key.offset; 24175bc7247aSMiao Xie 24182c536799SJeff Mahoney ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1); 2419aca1bba6SLiu Bo if (ret < 0) { 2420aca1bba6SLiu Bo if (list_empty(&reloc_root->root_list)) 2421aca1bba6SLiu Bo list_add_tail(&reloc_root->root_list, 2422aca1bba6SLiu Bo &reloc_roots); 2423aca1bba6SLiu Bo goto out; 2424aca1bba6SLiu Bo } 24255d4f98a2SYan Zheng } 24265d4f98a2SYan Zheng 24273fd0a558SYan, Zheng if (found) { 24283fd0a558SYan, Zheng found = 0; 24293fd0a558SYan, Zheng goto again; 24305d4f98a2SYan Zheng } 2431aca1bba6SLiu Bo out: 2432aca1bba6SLiu Bo if (ret) { 243334d97007SAnand Jain btrfs_handle_fs_error(root->fs_info, ret, NULL); 2434aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2435aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2436467bb1d2SWang Shilong 2437467bb1d2SWang Shilong /* new reloc root may be added */ 2438467bb1d2SWang Shilong mutex_lock(&root->fs_info->reloc_mutex); 2439467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 2440467bb1d2SWang Shilong mutex_unlock(&root->fs_info->reloc_mutex); 2441467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2442467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2443aca1bba6SLiu Bo } 2444aca1bba6SLiu Bo 24455d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 24465d4f98a2SYan Zheng } 24475d4f98a2SYan Zheng 24485d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 24495d4f98a2SYan Zheng { 24505d4f98a2SYan Zheng struct tree_block *block; 24515d4f98a2SYan Zheng struct rb_node *rb_node; 24525d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 24535d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 24545d4f98a2SYan Zheng rb_erase(rb_node, blocks); 24555d4f98a2SYan Zheng kfree(block); 24565d4f98a2SYan Zheng } 24575d4f98a2SYan Zheng } 24585d4f98a2SYan Zheng 24595d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 24605d4f98a2SYan Zheng struct btrfs_root *reloc_root) 24615d4f98a2SYan Zheng { 24625d4f98a2SYan Zheng struct btrfs_root *root; 24635d4f98a2SYan Zheng 24645d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 24655d4f98a2SYan Zheng return 0; 24665d4f98a2SYan Zheng 24675d4f98a2SYan Zheng root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset); 24685d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24695d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24705d4f98a2SYan Zheng 24715d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 24725d4f98a2SYan Zheng } 24735d4f98a2SYan Zheng 24743fd0a558SYan, Zheng static noinline_for_stack 24753fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 24763fd0a558SYan, Zheng struct reloc_control *rc, 24775d4f98a2SYan Zheng struct backref_node *node, 2478dc4103f9SWang Shilong struct backref_edge *edges[]) 24795d4f98a2SYan Zheng { 24805d4f98a2SYan Zheng struct backref_node *next; 24815d4f98a2SYan Zheng struct btrfs_root *root; 24823fd0a558SYan, Zheng int index = 0; 24833fd0a558SYan, Zheng 24845d4f98a2SYan Zheng next = node; 24855d4f98a2SYan Zheng while (1) { 24865d4f98a2SYan Zheng cond_resched(); 24875d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 24885d4f98a2SYan Zheng root = next->root; 24893fd0a558SYan, Zheng BUG_ON(!root); 249027cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 24915d4f98a2SYan Zheng 24925d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 24935d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 24945d4f98a2SYan Zheng break; 24955d4f98a2SYan Zheng } 24965d4f98a2SYan Zheng 24975d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 24983fd0a558SYan, Zheng root = root->reloc_root; 24993fd0a558SYan, Zheng 25003fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 25013fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 25023fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 25033fd0a558SYan, Zheng next->new_bytenr = root->node->start; 25043fd0a558SYan, Zheng next->root = root; 25053fd0a558SYan, Zheng list_add_tail(&next->list, 25063fd0a558SYan, Zheng &rc->backref_cache.changed); 25073fd0a558SYan, Zheng __mark_block_processed(rc, next); 25085d4f98a2SYan Zheng break; 25095d4f98a2SYan Zheng } 25105d4f98a2SYan Zheng 25113fd0a558SYan, Zheng WARN_ON(1); 25125d4f98a2SYan Zheng root = NULL; 25135d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25145d4f98a2SYan Zheng if (!next || next->level <= node->level) 25155d4f98a2SYan Zheng break; 25165d4f98a2SYan Zheng } 25173fd0a558SYan, Zheng if (!root) 25183fd0a558SYan, Zheng return NULL; 25195d4f98a2SYan Zheng 25203fd0a558SYan, Zheng next = node; 25213fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 25223fd0a558SYan, Zheng while (1) { 25233fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 25243fd0a558SYan, Zheng if (--index < 0) 25253fd0a558SYan, Zheng break; 25263fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 25273fd0a558SYan, Zheng } 25285d4f98a2SYan Zheng return root; 25295d4f98a2SYan Zheng } 25305d4f98a2SYan Zheng 25313fd0a558SYan, Zheng /* 25323fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 25333fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 25343fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 25353fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 25363fd0a558SYan, Zheng */ 25375d4f98a2SYan Zheng static noinline_for_stack 2538147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 25395d4f98a2SYan Zheng { 25403fd0a558SYan, Zheng struct backref_node *next; 25413fd0a558SYan, Zheng struct btrfs_root *root; 25423fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 25435d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25443fd0a558SYan, Zheng int index = 0; 25453fd0a558SYan, Zheng 25463fd0a558SYan, Zheng next = node; 25473fd0a558SYan, Zheng while (1) { 25483fd0a558SYan, Zheng cond_resched(); 25493fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 25503fd0a558SYan, Zheng root = next->root; 25513fd0a558SYan, Zheng BUG_ON(!root); 25523fd0a558SYan, Zheng 255325985edcSLucas De Marchi /* no other choice for non-references counted tree */ 255427cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 25553fd0a558SYan, Zheng return root; 25563fd0a558SYan, Zheng 25573fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 25583fd0a558SYan, Zheng fs_root = root; 25593fd0a558SYan, Zheng 25603fd0a558SYan, Zheng if (next != node) 25613fd0a558SYan, Zheng return NULL; 25623fd0a558SYan, Zheng 25633fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 25643fd0a558SYan, Zheng if (!next || next->level <= node->level) 25653fd0a558SYan, Zheng break; 25663fd0a558SYan, Zheng } 25673fd0a558SYan, Zheng 25683fd0a558SYan, Zheng if (!fs_root) 25693fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 25703fd0a558SYan, Zheng return fs_root; 25715d4f98a2SYan Zheng } 25725d4f98a2SYan Zheng 25735d4f98a2SYan Zheng static noinline_for_stack 25743fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 25753fd0a558SYan, Zheng struct backref_node *node, int reserve) 25765d4f98a2SYan Zheng { 25773fd0a558SYan, Zheng struct backref_node *next = node; 25783fd0a558SYan, Zheng struct backref_edge *edge; 25793fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25803fd0a558SYan, Zheng u64 num_bytes = 0; 25813fd0a558SYan, Zheng int index = 0; 25825d4f98a2SYan Zheng 25833fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 25843fd0a558SYan, Zheng 25853fd0a558SYan, Zheng while (next) { 25863fd0a558SYan, Zheng cond_resched(); 25875d4f98a2SYan Zheng while (1) { 25883fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 25895d4f98a2SYan Zheng break; 25905d4f98a2SYan Zheng 2591707e8a07SDavid Sterba num_bytes += rc->extent_root->nodesize; 25923fd0a558SYan, Zheng 25933fd0a558SYan, Zheng if (list_empty(&next->upper)) 25943fd0a558SYan, Zheng break; 25953fd0a558SYan, Zheng 25963fd0a558SYan, Zheng edge = list_entry(next->upper.next, 25973fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 25983fd0a558SYan, Zheng edges[index++] = edge; 25993fd0a558SYan, Zheng next = edge->node[UPPER]; 26005d4f98a2SYan Zheng } 26013fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26023fd0a558SYan, Zheng } 26033fd0a558SYan, Zheng return num_bytes; 26043fd0a558SYan, Zheng } 26053fd0a558SYan, Zheng 26063fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 26073fd0a558SYan, Zheng struct reloc_control *rc, 26083fd0a558SYan, Zheng struct backref_node *node) 26093fd0a558SYan, Zheng { 26103fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 26113fd0a558SYan, Zheng u64 num_bytes; 26123fd0a558SYan, Zheng int ret; 26130647bf56SWang Shilong u64 tmp; 26143fd0a558SYan, Zheng 26153fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 26163fd0a558SYan, Zheng 26173fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 26180647bf56SWang Shilong rc->reserved_bytes += num_bytes; 26198ca17f0fSJosef Bacik 26208ca17f0fSJosef Bacik /* 26218ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 26228ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 26238ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 26248ca17f0fSJosef Bacik */ 26250647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 26268ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 26273fd0a558SYan, Zheng if (ret) { 26288ca17f0fSJosef Bacik tmp = rc->extent_root->nodesize * RELOCATION_RESERVED_NODES; 26290647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 26300647bf56SWang Shilong tmp <<= 1; 26310647bf56SWang Shilong /* 26320647bf56SWang Shilong * only one thread can access block_rsv at this point, 26330647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 26340647bf56SWang Shilong * we expand more reservation size here to allow enough 26358ca17f0fSJosef Bacik * space for relocation and we will return eailer in 26360647bf56SWang Shilong * enospc case. 26370647bf56SWang Shilong */ 26380647bf56SWang Shilong rc->block_rsv->size = tmp + rc->extent_root->nodesize * 26390647bf56SWang Shilong RELOCATION_RESERVED_NODES; 26408ca17f0fSJosef Bacik return -EAGAIN; 26413fd0a558SYan, Zheng } 26423fd0a558SYan, Zheng 26433fd0a558SYan, Zheng return 0; 26443fd0a558SYan, Zheng } 26453fd0a558SYan, Zheng 26465d4f98a2SYan Zheng /* 26475d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 26485d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 26495d4f98a2SYan Zheng * 26505d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 26515d4f98a2SYan Zheng * in that case this function just updates pointers. 26525d4f98a2SYan Zheng */ 26535d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 26543fd0a558SYan, Zheng struct reloc_control *rc, 26555d4f98a2SYan Zheng struct backref_node *node, 26565d4f98a2SYan Zheng struct btrfs_key *key, 26575d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 26585d4f98a2SYan Zheng { 26595d4f98a2SYan Zheng struct backref_node *upper; 26605d4f98a2SYan Zheng struct backref_edge *edge; 26615d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26625d4f98a2SYan Zheng struct btrfs_root *root; 26635d4f98a2SYan Zheng struct extent_buffer *eb; 26645d4f98a2SYan Zheng u32 blocksize; 26655d4f98a2SYan Zheng u64 bytenr; 26665d4f98a2SYan Zheng u64 generation; 26675d4f98a2SYan Zheng int slot; 26685d4f98a2SYan Zheng int ret; 26695d4f98a2SYan Zheng int err = 0; 26705d4f98a2SYan Zheng 26715d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 26725d4f98a2SYan Zheng 26735d4f98a2SYan Zheng path->lowest_level = node->level + 1; 26743fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 26755d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 26765d4f98a2SYan Zheng cond_resched(); 26775d4f98a2SYan Zheng 26785d4f98a2SYan Zheng upper = edge->node[UPPER]; 2679dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 26803fd0a558SYan, Zheng BUG_ON(!root); 26815d4f98a2SYan Zheng 26823fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 26833fd0a558SYan, Zheng if (!lowest) { 26843fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 26853fd0a558SYan, Zheng upper->level, &slot); 26863fd0a558SYan, Zheng BUG_ON(ret); 26873fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 26883fd0a558SYan, Zheng if (node->eb->start == bytenr) 26893fd0a558SYan, Zheng goto next; 26903fd0a558SYan, Zheng } 26915d4f98a2SYan Zheng drop_node_buffer(upper); 26923fd0a558SYan, Zheng } 26935d4f98a2SYan Zheng 26945d4f98a2SYan Zheng if (!upper->eb) { 26955d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2696*3561b9dbSLiu Bo if (ret) { 2697*3561b9dbSLiu Bo if (ret < 0) 26985d4f98a2SYan Zheng err = ret; 2699*3561b9dbSLiu Bo else 2700*3561b9dbSLiu Bo err = -ENOENT; 2701*3561b9dbSLiu Bo 2702*3561b9dbSLiu Bo btrfs_release_path(path); 27035d4f98a2SYan Zheng break; 27045d4f98a2SYan Zheng } 27055d4f98a2SYan Zheng 27063fd0a558SYan, Zheng if (!upper->eb) { 27073fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 27083fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 27093fd0a558SYan, Zheng } else { 27103fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 27113fd0a558SYan, Zheng } 27123fd0a558SYan, Zheng 27133fd0a558SYan, Zheng upper->locked = 1; 27143fd0a558SYan, Zheng path->locks[upper->level] = 0; 27153fd0a558SYan, Zheng 27165d4f98a2SYan Zheng slot = path->slots[upper->level]; 2717b3b4aa74SDavid Sterba btrfs_release_path(path); 27185d4f98a2SYan Zheng } else { 27195d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 27205d4f98a2SYan Zheng &slot); 27215d4f98a2SYan Zheng BUG_ON(ret); 27225d4f98a2SYan Zheng } 27235d4f98a2SYan Zheng 27245d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27253fd0a558SYan, Zheng if (lowest) { 27263fd0a558SYan, Zheng BUG_ON(bytenr != node->bytenr); 27275d4f98a2SYan Zheng } else { 27283fd0a558SYan, Zheng if (node->eb->start == bytenr) 27293fd0a558SYan, Zheng goto next; 27305d4f98a2SYan Zheng } 27315d4f98a2SYan Zheng 2732707e8a07SDavid Sterba blocksize = root->nodesize; 27335d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2734ce86cd59SDavid Sterba eb = read_tree_block(root, bytenr, generation); 273564c043deSLiu Bo if (IS_ERR(eb)) { 273664c043deSLiu Bo err = PTR_ERR(eb); 273764c043deSLiu Bo goto next; 273864c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2739416bc658SJosef Bacik free_extent_buffer(eb); 274097d9a8a4STsutomu Itoh err = -EIO; 274197d9a8a4STsutomu Itoh goto next; 274297d9a8a4STsutomu Itoh } 27435d4f98a2SYan Zheng btrfs_tree_lock(eb); 27445d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 27455d4f98a2SYan Zheng 27465d4f98a2SYan Zheng if (!node->eb) { 27475d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 27485d4f98a2SYan Zheng slot, &eb); 27493fd0a558SYan, Zheng btrfs_tree_unlock(eb); 27503fd0a558SYan, Zheng free_extent_buffer(eb); 27515d4f98a2SYan Zheng if (ret < 0) { 27525d4f98a2SYan Zheng err = ret; 27533fd0a558SYan, Zheng goto next; 27545d4f98a2SYan Zheng } 27553fd0a558SYan, Zheng BUG_ON(node->eb != eb); 27565d4f98a2SYan Zheng } else { 27575d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 27585d4f98a2SYan Zheng node->eb->start); 27595d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 27605d4f98a2SYan Zheng trans->transid); 27615d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 27625d4f98a2SYan Zheng 27635d4f98a2SYan Zheng ret = btrfs_inc_extent_ref(trans, root, 27645d4f98a2SYan Zheng node->eb->start, blocksize, 27655d4f98a2SYan Zheng upper->eb->start, 27665d4f98a2SYan Zheng btrfs_header_owner(upper->eb), 2767b06c4bf5SFilipe Manana node->level, 0); 27685d4f98a2SYan Zheng BUG_ON(ret); 27695d4f98a2SYan Zheng 27705d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 27715d4f98a2SYan Zheng BUG_ON(ret); 27725d4f98a2SYan Zheng } 27733fd0a558SYan, Zheng next: 27743fd0a558SYan, Zheng if (!upper->pending) 27753fd0a558SYan, Zheng drop_node_buffer(upper); 27763fd0a558SYan, Zheng else 27773fd0a558SYan, Zheng unlock_node_buffer(upper); 27783fd0a558SYan, Zheng if (err) 27793fd0a558SYan, Zheng break; 27805d4f98a2SYan Zheng } 27813fd0a558SYan, Zheng 27823fd0a558SYan, Zheng if (!err && node->pending) { 27833fd0a558SYan, Zheng drop_node_buffer(node); 27843fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 27853fd0a558SYan, Zheng node->pending = 0; 27865d4f98a2SYan Zheng } 27873fd0a558SYan, Zheng 27885d4f98a2SYan Zheng path->lowest_level = 0; 27893fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 27905d4f98a2SYan Zheng return err; 27915d4f98a2SYan Zheng } 27925d4f98a2SYan Zheng 27935d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 27943fd0a558SYan, Zheng struct reloc_control *rc, 27955d4f98a2SYan Zheng struct backref_node *node, 27965d4f98a2SYan Zheng struct btrfs_path *path) 27975d4f98a2SYan Zheng { 27985d4f98a2SYan Zheng struct btrfs_key key; 27995d4f98a2SYan Zheng 28005d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 28013fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 28025d4f98a2SYan Zheng } 28035d4f98a2SYan Zheng 28045d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 28053fd0a558SYan, Zheng struct reloc_control *rc, 28063fd0a558SYan, Zheng struct btrfs_path *path, int err) 28075d4f98a2SYan Zheng { 28083fd0a558SYan, Zheng LIST_HEAD(list); 28093fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 28105d4f98a2SYan Zheng struct backref_node *node; 28115d4f98a2SYan Zheng int level; 28125d4f98a2SYan Zheng int ret; 28135d4f98a2SYan Zheng 28145d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 28155d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 28165d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 28173fd0a558SYan, Zheng struct backref_node, list); 28183fd0a558SYan, Zheng list_move_tail(&node->list, &list); 28193fd0a558SYan, Zheng BUG_ON(!node->pending); 28205d4f98a2SYan Zheng 28213fd0a558SYan, Zheng if (!err) { 28223fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 28235d4f98a2SYan Zheng if (ret < 0) 28245d4f98a2SYan Zheng err = ret; 28255d4f98a2SYan Zheng } 28265d4f98a2SYan Zheng } 28273fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 28283fd0a558SYan, Zheng } 28295d4f98a2SYan Zheng return err; 28305d4f98a2SYan Zheng } 28315d4f98a2SYan Zheng 28325d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 28333fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 28343fd0a558SYan, Zheng { 28353fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2836ceeb0ae7SDavid Sterba EXTENT_DIRTY); 28373fd0a558SYan, Zheng } 28383fd0a558SYan, Zheng 28393fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 28405d4f98a2SYan Zheng struct backref_node *node) 28415d4f98a2SYan Zheng { 28425d4f98a2SYan Zheng u32 blocksize; 28435d4f98a2SYan Zheng if (node->level == 0 || 28445d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2845707e8a07SDavid Sterba blocksize = rc->extent_root->nodesize; 28463fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 28475d4f98a2SYan Zheng } 28485d4f98a2SYan Zheng node->processed = 1; 28495d4f98a2SYan Zheng } 28505d4f98a2SYan Zheng 28515d4f98a2SYan Zheng /* 28525d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 28535d4f98a2SYan Zheng * as processed. 28545d4f98a2SYan Zheng */ 28555d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 28565d4f98a2SYan Zheng struct backref_node *node) 28575d4f98a2SYan Zheng { 28585d4f98a2SYan Zheng struct backref_node *next = node; 28595d4f98a2SYan Zheng struct backref_edge *edge; 28605d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 28615d4f98a2SYan Zheng int index = 0; 28625d4f98a2SYan Zheng 28635d4f98a2SYan Zheng while (next) { 28645d4f98a2SYan Zheng cond_resched(); 28655d4f98a2SYan Zheng while (1) { 28665d4f98a2SYan Zheng if (next->processed) 28675d4f98a2SYan Zheng break; 28685d4f98a2SYan Zheng 28693fd0a558SYan, Zheng __mark_block_processed(rc, next); 28705d4f98a2SYan Zheng 28715d4f98a2SYan Zheng if (list_empty(&next->upper)) 28725d4f98a2SYan Zheng break; 28735d4f98a2SYan Zheng 28745d4f98a2SYan Zheng edge = list_entry(next->upper.next, 28755d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 28765d4f98a2SYan Zheng edges[index++] = edge; 28775d4f98a2SYan Zheng next = edge->node[UPPER]; 28785d4f98a2SYan Zheng } 28795d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 28805d4f98a2SYan Zheng } 28815d4f98a2SYan Zheng } 28825d4f98a2SYan Zheng 28837476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 28845d4f98a2SYan Zheng { 28857476dfdaSDavid Sterba u32 blocksize = rc->extent_root->nodesize; 28867476dfdaSDavid Sterba 28875d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 28889655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 28895d4f98a2SYan Zheng return 1; 28905d4f98a2SYan Zheng return 0; 28915d4f98a2SYan Zheng } 28925d4f98a2SYan Zheng 28935d4f98a2SYan Zheng static int get_tree_block_key(struct reloc_control *rc, 28945d4f98a2SYan Zheng struct tree_block *block) 28955d4f98a2SYan Zheng { 28965d4f98a2SYan Zheng struct extent_buffer *eb; 28975d4f98a2SYan Zheng 28985d4f98a2SYan Zheng BUG_ON(block->key_ready); 28995d4f98a2SYan Zheng eb = read_tree_block(rc->extent_root, block->bytenr, 2900ce86cd59SDavid Sterba block->key.offset); 290164c043deSLiu Bo if (IS_ERR(eb)) { 290264c043deSLiu Bo return PTR_ERR(eb); 290364c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2904416bc658SJosef Bacik free_extent_buffer(eb); 2905416bc658SJosef Bacik return -EIO; 2906416bc658SJosef Bacik } 29075d4f98a2SYan Zheng WARN_ON(btrfs_header_level(eb) != block->level); 29085d4f98a2SYan Zheng if (block->level == 0) 29095d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 29105d4f98a2SYan Zheng else 29115d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 29125d4f98a2SYan Zheng free_extent_buffer(eb); 29135d4f98a2SYan Zheng block->key_ready = 1; 29145d4f98a2SYan Zheng return 0; 29155d4f98a2SYan Zheng } 29165d4f98a2SYan Zheng 29175d4f98a2SYan Zheng /* 29185d4f98a2SYan Zheng * helper function to relocate a tree block 29195d4f98a2SYan Zheng */ 29205d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 29215d4f98a2SYan Zheng struct reloc_control *rc, 29225d4f98a2SYan Zheng struct backref_node *node, 29235d4f98a2SYan Zheng struct btrfs_key *key, 29245d4f98a2SYan Zheng struct btrfs_path *path) 29255d4f98a2SYan Zheng { 29265d4f98a2SYan Zheng struct btrfs_root *root; 29273fd0a558SYan, Zheng int ret = 0; 29285d4f98a2SYan Zheng 29293fd0a558SYan, Zheng if (!node) 29305d4f98a2SYan Zheng return 0; 29313fd0a558SYan, Zheng 29323fd0a558SYan, Zheng BUG_ON(node->processed); 2933147d256eSZhaolei root = select_one_root(node); 29343fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 29353fd0a558SYan, Zheng update_processed_blocks(rc, node); 29363fd0a558SYan, Zheng goto out; 29375d4f98a2SYan Zheng } 29385d4f98a2SYan Zheng 293927cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29403fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 29413fd0a558SYan, Zheng if (ret) 29425d4f98a2SYan Zheng goto out; 29435d4f98a2SYan Zheng } 29443fd0a558SYan, Zheng 29453fd0a558SYan, Zheng if (root) { 294627cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29473fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 29483fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 29493fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 29503fd0a558SYan, Zheng root = root->reloc_root; 29513fd0a558SYan, Zheng node->new_bytenr = root->node->start; 29523fd0a558SYan, Zheng node->root = root; 29533fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 29543fd0a558SYan, Zheng } else { 29555d4f98a2SYan Zheng path->lowest_level = node->level; 29565d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2957b3b4aa74SDavid Sterba btrfs_release_path(path); 29583fd0a558SYan, Zheng if (ret > 0) 29595d4f98a2SYan Zheng ret = 0; 29603fd0a558SYan, Zheng } 29613fd0a558SYan, Zheng if (!ret) 29623fd0a558SYan, Zheng update_processed_blocks(rc, node); 29633fd0a558SYan, Zheng } else { 29643fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 29653fd0a558SYan, Zheng } 29665d4f98a2SYan Zheng out: 29670647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 29683fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 29695d4f98a2SYan Zheng return ret; 29705d4f98a2SYan Zheng } 29715d4f98a2SYan Zheng 29725d4f98a2SYan Zheng /* 29735d4f98a2SYan Zheng * relocate a list of blocks 29745d4f98a2SYan Zheng */ 29755d4f98a2SYan Zheng static noinline_for_stack 29765d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 29775d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 29785d4f98a2SYan Zheng { 29795d4f98a2SYan Zheng struct backref_node *node; 29805d4f98a2SYan Zheng struct btrfs_path *path; 29815d4f98a2SYan Zheng struct tree_block *block; 29825d4f98a2SYan Zheng struct rb_node *rb_node; 29835d4f98a2SYan Zheng int ret; 29845d4f98a2SYan Zheng int err = 0; 29855d4f98a2SYan Zheng 29865d4f98a2SYan Zheng path = btrfs_alloc_path(); 2987e1a12670SLiu Bo if (!path) { 2988e1a12670SLiu Bo err = -ENOMEM; 298934c2b290SDavid Sterba goto out_free_blocks; 2990e1a12670SLiu Bo } 29915d4f98a2SYan Zheng 29925d4f98a2SYan Zheng rb_node = rb_first(blocks); 29935d4f98a2SYan Zheng while (rb_node) { 29945d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 29955d4f98a2SYan Zheng if (!block->key_ready) 2996d3e46feaSDavid Sterba readahead_tree_block(rc->extent_root, block->bytenr); 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); 300334c2b290SDavid Sterba if (!block->key_ready) { 300434c2b290SDavid Sterba err = get_tree_block_key(rc, block); 300534c2b290SDavid Sterba if (err) 300634c2b290SDavid Sterba goto out_free_path; 300734c2b290SDavid Sterba } 30085d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30095d4f98a2SYan Zheng } 30105d4f98a2SYan Zheng 30115d4f98a2SYan Zheng rb_node = rb_first(blocks); 30125d4f98a2SYan Zheng while (rb_node) { 30135d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 30145d4f98a2SYan Zheng 30153fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 30165d4f98a2SYan Zheng block->level, block->bytenr); 30175d4f98a2SYan Zheng if (IS_ERR(node)) { 30185d4f98a2SYan Zheng err = PTR_ERR(node); 30195d4f98a2SYan Zheng goto out; 30205d4f98a2SYan Zheng } 30215d4f98a2SYan Zheng 30225d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 30235d4f98a2SYan Zheng path); 30245d4f98a2SYan Zheng if (ret < 0) { 30253fd0a558SYan, Zheng if (ret != -EAGAIN || rb_node == rb_first(blocks)) 30265d4f98a2SYan Zheng err = ret; 30275d4f98a2SYan Zheng goto out; 30285d4f98a2SYan Zheng } 30295d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30305d4f98a2SYan Zheng } 30315d4f98a2SYan Zheng out: 30323fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 30335d4f98a2SYan Zheng 303434c2b290SDavid Sterba out_free_path: 30355d4f98a2SYan Zheng btrfs_free_path(path); 303634c2b290SDavid Sterba out_free_blocks: 3037e1a12670SLiu Bo free_block_list(blocks); 30385d4f98a2SYan Zheng return err; 30395d4f98a2SYan Zheng } 30405d4f98a2SYan Zheng 30415d4f98a2SYan Zheng static noinline_for_stack 3042efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3043efa56464SYan, Zheng struct file_extent_cluster *cluster) 3044efa56464SYan, Zheng { 3045efa56464SYan, Zheng u64 alloc_hint = 0; 3046efa56464SYan, Zheng u64 start; 3047efa56464SYan, Zheng u64 end; 3048efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3049efa56464SYan, Zheng u64 num_bytes; 3050efa56464SYan, Zheng int nr = 0; 3051efa56464SYan, Zheng int ret = 0; 3052dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3053dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 305418513091SWang Xiaoguang u64 cur_offset; 3055efa56464SYan, Zheng 3056efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 30575955102cSAl Viro inode_lock(inode); 3058efa56464SYan, Zheng 3059dcb40c19SWang Xiaoguang ret = btrfs_check_data_free_space(inode, prealloc_start, 3060dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3061efa56464SYan, Zheng if (ret) 3062efa56464SYan, Zheng goto out; 3063efa56464SYan, Zheng 306418513091SWang Xiaoguang cur_offset = prealloc_start; 3065efa56464SYan, Zheng while (nr < cluster->nr) { 3066efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3067efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3068efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3069efa56464SYan, Zheng else 3070efa56464SYan, Zheng end = cluster->end - offset; 3071efa56464SYan, Zheng 3072d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3073efa56464SYan, Zheng num_bytes = end + 1 - start; 307418513091SWang Xiaoguang if (cur_offset < start) 307518513091SWang Xiaoguang btrfs_free_reserved_data_space(inode, cur_offset, 307618513091SWang Xiaoguang start - cur_offset); 3077efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3078efa56464SYan, Zheng num_bytes, num_bytes, 3079efa56464SYan, Zheng end + 1, &alloc_hint); 308018513091SWang Xiaoguang cur_offset = end + 1; 3081d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3082efa56464SYan, Zheng if (ret) 3083efa56464SYan, Zheng break; 3084efa56464SYan, Zheng nr++; 3085efa56464SYan, Zheng } 308618513091SWang Xiaoguang if (cur_offset < prealloc_end) 308718513091SWang Xiaoguang btrfs_free_reserved_data_space(inode, cur_offset, 308818513091SWang Xiaoguang prealloc_end + 1 - cur_offset); 3089efa56464SYan, Zheng out: 30905955102cSAl Viro inode_unlock(inode); 3091efa56464SYan, Zheng return ret; 3092efa56464SYan, Zheng } 3093efa56464SYan, Zheng 3094efa56464SYan, Zheng static noinline_for_stack 30950257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 30960257bb82SYan, Zheng u64 block_start) 30975d4f98a2SYan Zheng { 30985d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 30995d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 31005d4f98a2SYan Zheng struct extent_map *em; 31010257bb82SYan, Zheng int ret = 0; 31025d4f98a2SYan Zheng 3103172ddd60SDavid Sterba em = alloc_extent_map(); 31040257bb82SYan, Zheng if (!em) 31050257bb82SYan, Zheng return -ENOMEM; 31060257bb82SYan, Zheng 31075d4f98a2SYan Zheng em->start = start; 31080257bb82SYan, Zheng em->len = end + 1 - start; 31090257bb82SYan, Zheng em->block_len = em->len; 31100257bb82SYan, Zheng em->block_start = block_start; 31115d4f98a2SYan Zheng em->bdev = root->fs_info->fs_devices->latest_bdev; 31125d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 31135d4f98a2SYan Zheng 3114d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 31155d4f98a2SYan Zheng while (1) { 3116890871beSChris Mason write_lock(&em_tree->lock); 311709a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3118890871beSChris Mason write_unlock(&em_tree->lock); 31195d4f98a2SYan Zheng if (ret != -EEXIST) { 31205d4f98a2SYan Zheng free_extent_map(em); 31215d4f98a2SYan Zheng break; 31225d4f98a2SYan Zheng } 31235d4f98a2SYan Zheng btrfs_drop_extent_cache(inode, start, end, 0); 31245d4f98a2SYan Zheng } 3125d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 31260257bb82SYan, Zheng return ret; 31270257bb82SYan, Zheng } 31285d4f98a2SYan Zheng 31290257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 31300257bb82SYan, Zheng struct file_extent_cluster *cluster) 31310257bb82SYan, Zheng { 31320257bb82SYan, Zheng u64 page_start; 31330257bb82SYan, Zheng u64 page_end; 31340257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 31350257bb82SYan, Zheng unsigned long index; 31360257bb82SYan, Zheng unsigned long last_index; 31370257bb82SYan, Zheng struct page *page; 31380257bb82SYan, Zheng struct file_ra_state *ra; 31393b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 31400257bb82SYan, Zheng int nr = 0; 31410257bb82SYan, Zheng int ret = 0; 31420257bb82SYan, Zheng 31430257bb82SYan, Zheng if (!cluster->nr) 31440257bb82SYan, Zheng return 0; 31450257bb82SYan, Zheng 31460257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 31470257bb82SYan, Zheng if (!ra) 31480257bb82SYan, Zheng return -ENOMEM; 31490257bb82SYan, Zheng 3150efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 31510257bb82SYan, Zheng if (ret) 3152efa56464SYan, Zheng goto out; 31530257bb82SYan, Zheng 31540257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 31550257bb82SYan, Zheng 3156efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3157efa56464SYan, Zheng cluster->end - offset, cluster->start); 3158efa56464SYan, Zheng if (ret) 3159efa56464SYan, Zheng goto out; 3160efa56464SYan, Zheng 316109cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 316209cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 31630257bb82SYan, Zheng while (index <= last_index) { 316409cbfeafSKirill A. Shutemov ret = btrfs_delalloc_reserve_metadata(inode, PAGE_SIZE); 3165efa56464SYan, Zheng if (ret) 3166efa56464SYan, Zheng goto out; 3167efa56464SYan, Zheng 31680257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 31690257bb82SYan, Zheng if (!page) { 31700257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 31710257bb82SYan, Zheng ra, NULL, index, 31720257bb82SYan, Zheng last_index + 1 - index); 3173a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 31743b16a4e3SJosef Bacik mask); 31750257bb82SYan, Zheng if (!page) { 3176efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 317709cbfeafSKirill A. Shutemov PAGE_SIZE); 31780257bb82SYan, Zheng ret = -ENOMEM; 3179efa56464SYan, Zheng goto out; 31800257bb82SYan, Zheng } 31810257bb82SYan, Zheng } 31820257bb82SYan, Zheng 31830257bb82SYan, Zheng if (PageReadahead(page)) { 31840257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 31850257bb82SYan, Zheng ra, NULL, page, index, 31860257bb82SYan, Zheng last_index + 1 - index); 31870257bb82SYan, Zheng } 31880257bb82SYan, Zheng 31890257bb82SYan, Zheng if (!PageUptodate(page)) { 31900257bb82SYan, Zheng btrfs_readpage(NULL, page); 31910257bb82SYan, Zheng lock_page(page); 31920257bb82SYan, Zheng if (!PageUptodate(page)) { 31930257bb82SYan, Zheng unlock_page(page); 319409cbfeafSKirill A. Shutemov put_page(page); 3195efa56464SYan, Zheng btrfs_delalloc_release_metadata(inode, 319609cbfeafSKirill A. Shutemov PAGE_SIZE); 31970257bb82SYan, Zheng ret = -EIO; 3198efa56464SYan, Zheng goto out; 31990257bb82SYan, Zheng } 32000257bb82SYan, Zheng } 32010257bb82SYan, Zheng 32024eee4fa4SMiao Xie page_start = page_offset(page); 320309cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 32040257bb82SYan, Zheng 3205d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 32060257bb82SYan, Zheng 32070257bb82SYan, Zheng set_page_extent_mapped(page); 32080257bb82SYan, Zheng 32090257bb82SYan, Zheng if (nr < cluster->nr && 32100257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 32110257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 32120257bb82SYan, Zheng page_start, page_end, 3213ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 32140257bb82SYan, Zheng nr++; 32150257bb82SYan, Zheng } 32160257bb82SYan, Zheng 3217ba8b04c1SQu Wenruo btrfs_set_extent_delalloc(inode, page_start, page_end, NULL, 0); 32180257bb82SYan, Zheng set_page_dirty(page); 32190257bb82SYan, Zheng 32200257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3221d0082371SJeff Mahoney page_start, page_end); 32220257bb82SYan, Zheng unlock_page(page); 322309cbfeafSKirill A. Shutemov put_page(page); 32240257bb82SYan, Zheng 32250257bb82SYan, Zheng index++; 3226efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 3227efa56464SYan, Zheng btrfs_throttle(BTRFS_I(inode)->root); 32280257bb82SYan, Zheng } 32290257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3230efa56464SYan, Zheng out: 32310257bb82SYan, Zheng kfree(ra); 32320257bb82SYan, Zheng return ret; 32330257bb82SYan, Zheng } 32340257bb82SYan, Zheng 32350257bb82SYan, Zheng static noinline_for_stack 32360257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 32370257bb82SYan, Zheng struct file_extent_cluster *cluster) 32380257bb82SYan, Zheng { 32390257bb82SYan, Zheng int ret; 32400257bb82SYan, Zheng 32410257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 32420257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 32430257bb82SYan, Zheng if (ret) 32440257bb82SYan, Zheng return ret; 32450257bb82SYan, Zheng cluster->nr = 0; 32460257bb82SYan, Zheng } 32470257bb82SYan, Zheng 32480257bb82SYan, Zheng if (!cluster->nr) 32490257bb82SYan, Zheng cluster->start = extent_key->objectid; 32500257bb82SYan, Zheng else 32510257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 32520257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 32530257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 32540257bb82SYan, Zheng cluster->nr++; 32550257bb82SYan, Zheng 32560257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 32570257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 32580257bb82SYan, Zheng if (ret) 32590257bb82SYan, Zheng return ret; 32600257bb82SYan, Zheng cluster->nr = 0; 32610257bb82SYan, Zheng } 32620257bb82SYan, Zheng return 0; 32635d4f98a2SYan Zheng } 32645d4f98a2SYan Zheng 32655d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 32665d4f98a2SYan Zheng static int get_ref_objectid_v0(struct reloc_control *rc, 32675d4f98a2SYan Zheng struct btrfs_path *path, 32685d4f98a2SYan Zheng struct btrfs_key *extent_key, 32695d4f98a2SYan Zheng u64 *ref_objectid, int *path_change) 32705d4f98a2SYan Zheng { 32715d4f98a2SYan Zheng struct btrfs_key key; 32725d4f98a2SYan Zheng struct extent_buffer *leaf; 32735d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 32745d4f98a2SYan Zheng int ret; 32755d4f98a2SYan Zheng int slot; 32765d4f98a2SYan Zheng 32775d4f98a2SYan Zheng leaf = path->nodes[0]; 32785d4f98a2SYan Zheng slot = path->slots[0]; 32795d4f98a2SYan Zheng while (1) { 32805d4f98a2SYan Zheng if (slot >= btrfs_header_nritems(leaf)) { 32815d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 32825d4f98a2SYan Zheng if (ret < 0) 32835d4f98a2SYan Zheng return ret; 32845d4f98a2SYan Zheng BUG_ON(ret > 0); 32855d4f98a2SYan Zheng leaf = path->nodes[0]; 32865d4f98a2SYan Zheng slot = path->slots[0]; 32875d4f98a2SYan Zheng if (path_change) 32885d4f98a2SYan Zheng *path_change = 1; 32895d4f98a2SYan Zheng } 32905d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 32915d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 32925d4f98a2SYan Zheng return -ENOENT; 32935d4f98a2SYan Zheng 32945d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_REF_V0_KEY) { 32955d4f98a2SYan Zheng slot++; 32965d4f98a2SYan Zheng continue; 32975d4f98a2SYan Zheng } 32985d4f98a2SYan Zheng ref0 = btrfs_item_ptr(leaf, slot, 32995d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 33005d4f98a2SYan Zheng *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0); 33015d4f98a2SYan Zheng break; 33025d4f98a2SYan Zheng } 33035d4f98a2SYan Zheng return 0; 33045d4f98a2SYan Zheng } 33055d4f98a2SYan Zheng #endif 33065d4f98a2SYan Zheng 33075d4f98a2SYan Zheng /* 33085d4f98a2SYan Zheng * helper to add a tree block to the list. 33095d4f98a2SYan Zheng * the major work is getting the generation and level of the block 33105d4f98a2SYan Zheng */ 33115d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 33125d4f98a2SYan Zheng struct btrfs_key *extent_key, 33135d4f98a2SYan Zheng struct btrfs_path *path, 33145d4f98a2SYan Zheng struct rb_root *blocks) 33155d4f98a2SYan Zheng { 33165d4f98a2SYan Zheng struct extent_buffer *eb; 33175d4f98a2SYan Zheng struct btrfs_extent_item *ei; 33185d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 33195d4f98a2SYan Zheng struct tree_block *block; 33205d4f98a2SYan Zheng struct rb_node *rb_node; 33215d4f98a2SYan Zheng u32 item_size; 33225d4f98a2SYan Zheng int level = -1; 33237fdf4b60SWang Shilong u64 generation; 33245d4f98a2SYan Zheng 33255d4f98a2SYan Zheng eb = path->nodes[0]; 33265d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 33275d4f98a2SYan Zheng 33283173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 33293173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 33305d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 33315d4f98a2SYan Zheng struct btrfs_extent_item); 33323173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 33335d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 33345d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 33355d4f98a2SYan Zheng } else { 33363173a18fSJosef Bacik level = (int)extent_key->offset; 33373173a18fSJosef Bacik } 33383173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 33393173a18fSJosef Bacik } else { 33405d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 33415d4f98a2SYan Zheng u64 ref_owner; 33425d4f98a2SYan Zheng int ret; 33435d4f98a2SYan Zheng 33445d4f98a2SYan Zheng BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 33455d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, extent_key, 33465d4f98a2SYan Zheng &ref_owner, NULL); 3347411fc6bcSAndi Kleen if (ret < 0) 3348411fc6bcSAndi Kleen return ret; 33495d4f98a2SYan Zheng BUG_ON(ref_owner >= BTRFS_MAX_LEVEL); 33505d4f98a2SYan Zheng level = (int)ref_owner; 33515d4f98a2SYan Zheng /* FIXME: get real generation */ 33525d4f98a2SYan Zheng generation = 0; 33535d4f98a2SYan Zheng #else 33545d4f98a2SYan Zheng BUG(); 33555d4f98a2SYan Zheng #endif 33565d4f98a2SYan Zheng } 33575d4f98a2SYan Zheng 3358b3b4aa74SDavid Sterba btrfs_release_path(path); 33595d4f98a2SYan Zheng 33605d4f98a2SYan Zheng BUG_ON(level == -1); 33615d4f98a2SYan Zheng 33625d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 33635d4f98a2SYan Zheng if (!block) 33645d4f98a2SYan Zheng return -ENOMEM; 33655d4f98a2SYan Zheng 33665d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3367707e8a07SDavid Sterba block->key.objectid = rc->extent_root->nodesize; 33685d4f98a2SYan Zheng block->key.offset = generation; 33695d4f98a2SYan Zheng block->level = level; 33705d4f98a2SYan Zheng block->key_ready = 0; 33715d4f98a2SYan Zheng 33725d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 337343c04fb1SJeff Mahoney if (rb_node) 337443c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 33755d4f98a2SYan Zheng 33765d4f98a2SYan Zheng return 0; 33775d4f98a2SYan Zheng } 33785d4f98a2SYan Zheng 33795d4f98a2SYan Zheng /* 33805d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 33815d4f98a2SYan Zheng */ 33825d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 33835d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 33845d4f98a2SYan Zheng struct rb_root *blocks) 33855d4f98a2SYan Zheng { 33865d4f98a2SYan Zheng struct btrfs_path *path; 33875d4f98a2SYan Zheng struct btrfs_key key; 33885d4f98a2SYan Zheng int ret; 3389aee68ee5SJosef Bacik bool skinny = btrfs_fs_incompat(rc->extent_root->fs_info, 3390aee68ee5SJosef Bacik SKINNY_METADATA); 33915d4f98a2SYan Zheng 33927476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 33935d4f98a2SYan Zheng return 0; 33945d4f98a2SYan Zheng 33955d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 33965d4f98a2SYan Zheng return 0; 33975d4f98a2SYan Zheng 33985d4f98a2SYan Zheng path = btrfs_alloc_path(); 33995d4f98a2SYan Zheng if (!path) 34005d4f98a2SYan Zheng return -ENOMEM; 3401aee68ee5SJosef Bacik again: 34025d4f98a2SYan Zheng key.objectid = bytenr; 3403aee68ee5SJosef Bacik if (skinny) { 3404aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3405aee68ee5SJosef Bacik key.offset = (u64)-1; 3406aee68ee5SJosef Bacik } else { 34075d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34085d4f98a2SYan Zheng key.offset = blocksize; 3409aee68ee5SJosef Bacik } 34105d4f98a2SYan Zheng 34115d4f98a2SYan Zheng path->search_commit_root = 1; 34125d4f98a2SYan Zheng path->skip_locking = 1; 34135d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 34145d4f98a2SYan Zheng if (ret < 0) 34155d4f98a2SYan Zheng goto out; 34165d4f98a2SYan Zheng 3417aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3418aee68ee5SJosef Bacik if (path->slots[0]) { 3419aee68ee5SJosef Bacik path->slots[0]--; 3420aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3421aee68ee5SJosef Bacik path->slots[0]); 34223173a18fSJosef Bacik if (key.objectid == bytenr && 3423aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3424aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3425aee68ee5SJosef Bacik key.offset == blocksize))) 34263173a18fSJosef Bacik ret = 0; 34273173a18fSJosef Bacik } 3428aee68ee5SJosef Bacik 3429aee68ee5SJosef Bacik if (ret) { 3430aee68ee5SJosef Bacik skinny = false; 3431aee68ee5SJosef Bacik btrfs_release_path(path); 3432aee68ee5SJosef Bacik goto again; 3433aee68ee5SJosef Bacik } 3434aee68ee5SJosef Bacik } 34353173a18fSJosef Bacik BUG_ON(ret); 34363173a18fSJosef Bacik 34375d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 34385d4f98a2SYan Zheng out: 34395d4f98a2SYan Zheng btrfs_free_path(path); 34405d4f98a2SYan Zheng return ret; 34415d4f98a2SYan Zheng } 34425d4f98a2SYan Zheng 34435d4f98a2SYan Zheng /* 34445d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 34455d4f98a2SYan Zheng */ 34465d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 34475d4f98a2SYan Zheng struct extent_buffer *eb) 34485d4f98a2SYan Zheng { 34495d4f98a2SYan Zheng u64 flags; 34505d4f98a2SYan Zheng int ret; 34515d4f98a2SYan Zheng 34525d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 34535d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 34545d4f98a2SYan Zheng return 1; 34555d4f98a2SYan Zheng 34563fd0a558SYan, Zheng ret = btrfs_lookup_extent_info(NULL, rc->extent_root, 34573173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 34583173a18fSJosef Bacik NULL, &flags); 34595d4f98a2SYan Zheng BUG_ON(ret); 34605d4f98a2SYan Zheng 34615d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 34625d4f98a2SYan Zheng ret = 1; 34635d4f98a2SYan Zheng else 34645d4f98a2SYan Zheng ret = 0; 34655d4f98a2SYan Zheng return ret; 34665d4f98a2SYan Zheng } 34675d4f98a2SYan Zheng 34680af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 34691bbc621eSChris Mason struct btrfs_block_group_cache *block_group, 34701bbc621eSChris Mason struct inode *inode, 34711bbc621eSChris Mason u64 ino) 34720af3d00bSJosef Bacik { 34730af3d00bSJosef Bacik struct btrfs_key key; 34740af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 34750af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 34760af3d00bSJosef Bacik int ret = 0; 34770af3d00bSJosef Bacik 34780af3d00bSJosef Bacik if (inode) 34790af3d00bSJosef Bacik goto truncate; 34800af3d00bSJosef Bacik 34810af3d00bSJosef Bacik key.objectid = ino; 34820af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 34830af3d00bSJosef Bacik key.offset = 0; 34840af3d00bSJosef Bacik 34850af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 3486f54fb859STsutomu Itoh if (IS_ERR(inode) || is_bad_inode(inode)) { 3487f54fb859STsutomu Itoh if (!IS_ERR(inode)) 34880af3d00bSJosef Bacik iput(inode); 34890af3d00bSJosef Bacik return -ENOENT; 34900af3d00bSJosef Bacik } 34910af3d00bSJosef Bacik 34920af3d00bSJosef Bacik truncate: 34937b61cd92SMiao Xie ret = btrfs_check_trunc_cache_free_space(root, 34947b61cd92SMiao Xie &fs_info->global_block_rsv); 34957b61cd92SMiao Xie if (ret) 34967b61cd92SMiao Xie goto out; 34977b61cd92SMiao Xie 34987a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 34990af3d00bSJosef Bacik if (IS_ERR(trans)) { 35003612b495STsutomu Itoh ret = PTR_ERR(trans); 35010af3d00bSJosef Bacik goto out; 35020af3d00bSJosef Bacik } 35030af3d00bSJosef Bacik 35041bbc621eSChris Mason ret = btrfs_truncate_free_space_cache(root, trans, block_group, inode); 35050af3d00bSJosef Bacik 35060af3d00bSJosef Bacik btrfs_end_transaction(trans, root); 3507b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 35080af3d00bSJosef Bacik out: 35090af3d00bSJosef Bacik iput(inode); 35100af3d00bSJosef Bacik return ret; 35110af3d00bSJosef Bacik } 35120af3d00bSJosef Bacik 35135d4f98a2SYan Zheng /* 35145d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 35155d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 35165d4f98a2SYan Zheng */ 35175d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 35185d4f98a2SYan Zheng struct btrfs_key *extent_key, 35195d4f98a2SYan Zheng struct extent_buffer *leaf, 35205d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 35215d4f98a2SYan Zheng struct rb_root *blocks) 35225d4f98a2SYan Zheng { 35235d4f98a2SYan Zheng struct btrfs_path *path; 35245d4f98a2SYan Zheng struct tree_block *block; 35255d4f98a2SYan Zheng struct btrfs_root *root; 35265d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 35275d4f98a2SYan Zheng struct rb_node *rb_node; 35285d4f98a2SYan Zheng struct btrfs_key key; 35295d4f98a2SYan Zheng u64 ref_root; 35305d4f98a2SYan Zheng u64 ref_objectid; 35315d4f98a2SYan Zheng u64 ref_offset; 35325d4f98a2SYan Zheng u32 ref_count; 35335d4f98a2SYan Zheng u32 nritems; 35345d4f98a2SYan Zheng int err = 0; 35355d4f98a2SYan Zheng int added = 0; 35365d4f98a2SYan Zheng int counted; 35375d4f98a2SYan Zheng int ret; 35385d4f98a2SYan Zheng 35395d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 35405d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 35415d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 35425d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 35435d4f98a2SYan Zheng 35440af3d00bSJosef Bacik /* 35450af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 35460af3d00bSJosef Bacik * it and redo the search. 35470af3d00bSJosef Bacik */ 35480af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 35490af3d00bSJosef Bacik ret = delete_block_group_cache(rc->extent_root->fs_info, 35501bbc621eSChris Mason rc->block_group, 35510af3d00bSJosef Bacik NULL, ref_objectid); 35520af3d00bSJosef Bacik if (ret != -ENOENT) 35530af3d00bSJosef Bacik return ret; 35540af3d00bSJosef Bacik ret = 0; 35550af3d00bSJosef Bacik } 35560af3d00bSJosef Bacik 35570af3d00bSJosef Bacik path = btrfs_alloc_path(); 35580af3d00bSJosef Bacik if (!path) 35590af3d00bSJosef Bacik return -ENOMEM; 3560e4058b54SDavid Sterba path->reada = READA_FORWARD; 35610af3d00bSJosef Bacik 35625d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, ref_root); 35635d4f98a2SYan Zheng if (IS_ERR(root)) { 35645d4f98a2SYan Zheng err = PTR_ERR(root); 35655d4f98a2SYan Zheng goto out; 35665d4f98a2SYan Zheng } 35675d4f98a2SYan Zheng 35685d4f98a2SYan Zheng key.objectid = ref_objectid; 35695d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 357084850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 357184850e8dSYan, Zheng key.offset = 0; 357284850e8dSYan, Zheng else 357384850e8dSYan, Zheng key.offset = ref_offset; 35745d4f98a2SYan Zheng 35755d4f98a2SYan Zheng path->search_commit_root = 1; 35765d4f98a2SYan Zheng path->skip_locking = 1; 35775d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 35785d4f98a2SYan Zheng if (ret < 0) { 35795d4f98a2SYan Zheng err = ret; 35805d4f98a2SYan Zheng goto out; 35815d4f98a2SYan Zheng } 35825d4f98a2SYan Zheng 35835d4f98a2SYan Zheng leaf = path->nodes[0]; 35845d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 35855d4f98a2SYan Zheng /* 35865d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 35875d4f98a2SYan Zheng * are not counted in 35885d4f98a2SYan Zheng */ 35895d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 35905d4f98a2SYan Zheng counted = 0; 35915d4f98a2SYan Zheng else 35925d4f98a2SYan Zheng counted = 1; 35935d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 35945d4f98a2SYan Zheng if (rb_node) { 35955d4f98a2SYan Zheng if (counted) 35965d4f98a2SYan Zheng added = 1; 35975d4f98a2SYan Zheng else 35985d4f98a2SYan Zheng path->slots[0] = nritems; 35995d4f98a2SYan Zheng } 36005d4f98a2SYan Zheng 36015d4f98a2SYan Zheng while (ref_count > 0) { 36025d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 36035d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 36045d4f98a2SYan Zheng if (ret < 0) { 36055d4f98a2SYan Zheng err = ret; 36065d4f98a2SYan Zheng goto out; 36075d4f98a2SYan Zheng } 3608fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 36095d4f98a2SYan Zheng goto out; 36105d4f98a2SYan Zheng 36115d4f98a2SYan Zheng leaf = path->nodes[0]; 36125d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36135d4f98a2SYan Zheng added = 0; 36145d4f98a2SYan Zheng 36155d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36165d4f98a2SYan Zheng counted = 0; 36175d4f98a2SYan Zheng else 36185d4f98a2SYan Zheng counted = 1; 36195d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36205d4f98a2SYan Zheng if (rb_node) { 36215d4f98a2SYan Zheng if (counted) 36225d4f98a2SYan Zheng added = 1; 36235d4f98a2SYan Zheng else 36245d4f98a2SYan Zheng path->slots[0] = nritems; 36255d4f98a2SYan Zheng } 36265d4f98a2SYan Zheng } 36275d4f98a2SYan Zheng 36285d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3629fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3630fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 36315d4f98a2SYan Zheng break; 36325d4f98a2SYan Zheng 36335d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 36345d4f98a2SYan Zheng struct btrfs_file_extent_item); 36355d4f98a2SYan Zheng 36365d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 36375d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 36385d4f98a2SYan Zheng goto next; 36395d4f98a2SYan Zheng 36405d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 36415d4f98a2SYan Zheng extent_key->objectid) 36425d4f98a2SYan Zheng goto next; 36435d4f98a2SYan Zheng 36445d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 36455d4f98a2SYan Zheng if (key.offset != ref_offset) 36465d4f98a2SYan Zheng goto next; 36475d4f98a2SYan Zheng 36485d4f98a2SYan Zheng if (counted) 36495d4f98a2SYan Zheng ref_count--; 36505d4f98a2SYan Zheng if (added) 36515d4f98a2SYan Zheng goto next; 36525d4f98a2SYan Zheng 36537476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 36545d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 36555d4f98a2SYan Zheng if (!block) { 36565d4f98a2SYan Zheng err = -ENOMEM; 36575d4f98a2SYan Zheng break; 36585d4f98a2SYan Zheng } 36595d4f98a2SYan Zheng block->bytenr = leaf->start; 36605d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 36615d4f98a2SYan Zheng block->level = 0; 36625d4f98a2SYan Zheng block->key_ready = 1; 36635d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 36645d4f98a2SYan Zheng &block->rb_node); 366543c04fb1SJeff Mahoney if (rb_node) 366643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 366743c04fb1SJeff Mahoney block->bytenr); 36685d4f98a2SYan Zheng } 36695d4f98a2SYan Zheng if (counted) 36705d4f98a2SYan Zheng added = 1; 36715d4f98a2SYan Zheng else 36725d4f98a2SYan Zheng path->slots[0] = nritems; 36735d4f98a2SYan Zheng next: 36745d4f98a2SYan Zheng path->slots[0]++; 36755d4f98a2SYan Zheng 36765d4f98a2SYan Zheng } 36775d4f98a2SYan Zheng out: 36785d4f98a2SYan Zheng btrfs_free_path(path); 36795d4f98a2SYan Zheng return err; 36805d4f98a2SYan Zheng } 36815d4f98a2SYan Zheng 36825d4f98a2SYan Zheng /* 36832c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 36845d4f98a2SYan Zheng */ 36855d4f98a2SYan Zheng static noinline_for_stack 36865d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 36875d4f98a2SYan Zheng struct btrfs_key *extent_key, 36885d4f98a2SYan Zheng struct btrfs_path *path, 36895d4f98a2SYan Zheng struct rb_root *blocks) 36905d4f98a2SYan Zheng { 36915d4f98a2SYan Zheng struct btrfs_key key; 36925d4f98a2SYan Zheng struct extent_buffer *eb; 36935d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 36945d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 36955d4f98a2SYan Zheng unsigned long ptr; 36965d4f98a2SYan Zheng unsigned long end; 3697707e8a07SDavid Sterba u32 blocksize = rc->extent_root->nodesize; 3698647f63bdSFilipe David Borba Manana int ret = 0; 36995d4f98a2SYan Zheng int err = 0; 37005d4f98a2SYan Zheng 37015d4f98a2SYan Zheng eb = path->nodes[0]; 37025d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 37035d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 37045d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 37055d4f98a2SYan Zheng if (ptr + sizeof(struct btrfs_extent_item_v0) == end) 37065d4f98a2SYan Zheng ptr = end; 37075d4f98a2SYan Zheng else 37085d4f98a2SYan Zheng #endif 37095d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 37105d4f98a2SYan Zheng 37115d4f98a2SYan Zheng while (ptr < end) { 37125d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 37135d4f98a2SYan Zheng key.type = btrfs_extent_inline_ref_type(eb, iref); 37145d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37155d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 37165d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37175d4f98a2SYan Zheng blocks); 37185d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37195d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 37205d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37215d4f98a2SYan Zheng eb, dref, blocks); 37225d4f98a2SYan Zheng } else { 37235d4f98a2SYan Zheng BUG(); 37245d4f98a2SYan Zheng } 3725647f63bdSFilipe David Borba Manana if (ret) { 3726647f63bdSFilipe David Borba Manana err = ret; 3727647f63bdSFilipe David Borba Manana goto out; 3728647f63bdSFilipe David Borba Manana } 37295d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 37305d4f98a2SYan Zheng } 37315d4f98a2SYan Zheng WARN_ON(ptr > end); 37325d4f98a2SYan Zheng 37335d4f98a2SYan Zheng while (1) { 37345d4f98a2SYan Zheng cond_resched(); 37355d4f98a2SYan Zheng eb = path->nodes[0]; 37365d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 37375d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 37385d4f98a2SYan Zheng if (ret < 0) { 37395d4f98a2SYan Zheng err = ret; 37405d4f98a2SYan Zheng break; 37415d4f98a2SYan Zheng } 37425d4f98a2SYan Zheng if (ret > 0) 37435d4f98a2SYan Zheng break; 37445d4f98a2SYan Zheng eb = path->nodes[0]; 37455d4f98a2SYan Zheng } 37465d4f98a2SYan Zheng 37475d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 37485d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 37495d4f98a2SYan Zheng break; 37505d4f98a2SYan Zheng 37515d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 37525d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY || 37535d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 37545d4f98a2SYan Zheng #else 37555d4f98a2SYan Zheng BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); 37565d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37575d4f98a2SYan Zheng #endif 37585d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37595d4f98a2SYan Zheng blocks); 37605d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37615d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 37625d4f98a2SYan Zheng struct btrfs_extent_data_ref); 37635d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37645d4f98a2SYan Zheng eb, dref, blocks); 37655d4f98a2SYan Zheng } else { 37665d4f98a2SYan Zheng ret = 0; 37675d4f98a2SYan Zheng } 37685d4f98a2SYan Zheng if (ret) { 37695d4f98a2SYan Zheng err = ret; 37705d4f98a2SYan Zheng break; 37715d4f98a2SYan Zheng } 37725d4f98a2SYan Zheng path->slots[0]++; 37735d4f98a2SYan Zheng } 3774647f63bdSFilipe David Borba Manana out: 3775b3b4aa74SDavid Sterba btrfs_release_path(path); 37765d4f98a2SYan Zheng if (err) 37775d4f98a2SYan Zheng free_block_list(blocks); 37785d4f98a2SYan Zheng return err; 37795d4f98a2SYan Zheng } 37805d4f98a2SYan Zheng 37815d4f98a2SYan Zheng /* 37822c016dc2SLiu Bo * helper to find next unprocessed extent 37835d4f98a2SYan Zheng */ 37845d4f98a2SYan Zheng static noinline_for_stack 3785147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 37863fd0a558SYan, Zheng struct btrfs_key *extent_key) 37875d4f98a2SYan Zheng { 37885d4f98a2SYan Zheng struct btrfs_key key; 37895d4f98a2SYan Zheng struct extent_buffer *leaf; 37905d4f98a2SYan Zheng u64 start, end, last; 37915d4f98a2SYan Zheng int ret; 37925d4f98a2SYan Zheng 37935d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 37945d4f98a2SYan Zheng while (1) { 37955d4f98a2SYan Zheng cond_resched(); 37965d4f98a2SYan Zheng if (rc->search_start >= last) { 37975d4f98a2SYan Zheng ret = 1; 37985d4f98a2SYan Zheng break; 37995d4f98a2SYan Zheng } 38005d4f98a2SYan Zheng 38015d4f98a2SYan Zheng key.objectid = rc->search_start; 38025d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 38035d4f98a2SYan Zheng key.offset = 0; 38045d4f98a2SYan Zheng 38055d4f98a2SYan Zheng path->search_commit_root = 1; 38065d4f98a2SYan Zheng path->skip_locking = 1; 38075d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 38085d4f98a2SYan Zheng 0, 0); 38095d4f98a2SYan Zheng if (ret < 0) 38105d4f98a2SYan Zheng break; 38115d4f98a2SYan Zheng next: 38125d4f98a2SYan Zheng leaf = path->nodes[0]; 38135d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38145d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38155d4f98a2SYan Zheng if (ret != 0) 38165d4f98a2SYan Zheng break; 38175d4f98a2SYan Zheng leaf = path->nodes[0]; 38185d4f98a2SYan Zheng } 38195d4f98a2SYan Zheng 38205d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38215d4f98a2SYan Zheng if (key.objectid >= last) { 38225d4f98a2SYan Zheng ret = 1; 38235d4f98a2SYan Zheng break; 38245d4f98a2SYan Zheng } 38255d4f98a2SYan Zheng 38263173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38273173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38283173a18fSJosef Bacik path->slots[0]++; 38293173a18fSJosef Bacik goto next; 38303173a18fSJosef Bacik } 38313173a18fSJosef Bacik 38323173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 38335d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 38345d4f98a2SYan Zheng path->slots[0]++; 38355d4f98a2SYan Zheng goto next; 38365d4f98a2SYan Zheng } 38375d4f98a2SYan Zheng 38383173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 3839707e8a07SDavid Sterba key.objectid + rc->extent_root->nodesize <= 38403173a18fSJosef Bacik rc->search_start) { 38413173a18fSJosef Bacik path->slots[0]++; 38423173a18fSJosef Bacik goto next; 38433173a18fSJosef Bacik } 38443173a18fSJosef Bacik 38455d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 38465d4f98a2SYan Zheng key.objectid, &start, &end, 3847e6138876SJosef Bacik EXTENT_DIRTY, NULL); 38485d4f98a2SYan Zheng 38495d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3850b3b4aa74SDavid Sterba btrfs_release_path(path); 38515d4f98a2SYan Zheng rc->search_start = end + 1; 38525d4f98a2SYan Zheng } else { 38533173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 38545d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 38553173a18fSJosef Bacik else 38563173a18fSJosef Bacik rc->search_start = key.objectid + 3857707e8a07SDavid Sterba rc->extent_root->nodesize; 38583fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 38595d4f98a2SYan Zheng return 0; 38605d4f98a2SYan Zheng } 38615d4f98a2SYan Zheng } 3862b3b4aa74SDavid Sterba btrfs_release_path(path); 38635d4f98a2SYan Zheng return ret; 38645d4f98a2SYan Zheng } 38655d4f98a2SYan Zheng 38665d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 38675d4f98a2SYan Zheng { 38685d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38697585717fSChris Mason 38707585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38715d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 38727585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38735d4f98a2SYan Zheng } 38745d4f98a2SYan Zheng 38755d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 38765d4f98a2SYan Zheng { 38775d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38787585717fSChris Mason 38797585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38805d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 38817585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38825d4f98a2SYan Zheng } 38835d4f98a2SYan Zheng 38845d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 38855d4f98a2SYan Zheng { 38865d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38875d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38885d4f98a2SYan Zheng return 1; 38895d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 38905d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38915d4f98a2SYan Zheng return 1; 38925d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38935d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 38945d4f98a2SYan Zheng return 1; 38955d4f98a2SYan Zheng return 0; 38965d4f98a2SYan Zheng } 38975d4f98a2SYan Zheng 38983fd0a558SYan, Zheng static noinline_for_stack 38993fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 39003fd0a558SYan, Zheng { 39013fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3902ac2fabacSJosef Bacik int ret; 39033fd0a558SYan, Zheng 390466d8f3ddSMiao Xie rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root, 390566d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 39063fd0a558SYan, Zheng if (!rc->block_rsv) 39073fd0a558SYan, Zheng return -ENOMEM; 39083fd0a558SYan, Zheng 39093fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 39103fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 39113fd0a558SYan, Zheng rc->extents_found = 0; 39123fd0a558SYan, Zheng rc->nodes_relocated = 0; 39133fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39140647bf56SWang Shilong rc->reserved_bytes = 0; 39150647bf56SWang Shilong rc->block_rsv->size = rc->extent_root->nodesize * 39160647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3917ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3918ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3919ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3920ac2fabacSJosef Bacik if (ret) 3921ac2fabacSJosef Bacik return ret; 39223fd0a558SYan, Zheng 39233fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39243fd0a558SYan, Zheng set_reloc_control(rc); 39253fd0a558SYan, Zheng 39267a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 392728818947SLiu Bo if (IS_ERR(trans)) { 392828818947SLiu Bo unset_reloc_control(rc); 392928818947SLiu Bo /* 393028818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 393128818947SLiu Bo * cleanup. And callers are responsible to free the above 393228818947SLiu Bo * block rsv. 393328818947SLiu Bo */ 393428818947SLiu Bo return PTR_ERR(trans); 393528818947SLiu Bo } 39363fd0a558SYan, Zheng btrfs_commit_transaction(trans, rc->extent_root); 39373fd0a558SYan, Zheng return 0; 39383fd0a558SYan, Zheng } 393976dda93cSYan, Zheng 394062b99540SQu Wenruo /* 394162b99540SQu Wenruo * Qgroup fixer for data chunk relocation. 394262b99540SQu Wenruo * The data relocation is done in the following steps 394362b99540SQu Wenruo * 1) Copy data extents into data reloc tree 394462b99540SQu Wenruo * 2) Create tree reloc tree(special snapshot) for related subvolumes 394562b99540SQu Wenruo * 3) Modify file extents in tree reloc tree 394662b99540SQu Wenruo * 4) Merge tree reloc tree with original fs tree, by swapping tree blocks 394762b99540SQu Wenruo * 394862b99540SQu Wenruo * The problem is, data and tree reloc tree are not accounted to qgroup, 394962b99540SQu Wenruo * and 4) will only info qgroup to track tree blocks change, not file extents 395062b99540SQu Wenruo * in the tree blocks. 395162b99540SQu Wenruo * 395262b99540SQu Wenruo * The good news is, related data extents are all in data reloc tree, so we 395362b99540SQu Wenruo * only need to info qgroup to track all file extents in data reloc tree 395462b99540SQu Wenruo * before commit trans. 395562b99540SQu Wenruo */ 395662b99540SQu Wenruo static int qgroup_fix_relocated_data_extents(struct btrfs_trans_handle *trans, 395762b99540SQu Wenruo struct reloc_control *rc) 395862b99540SQu Wenruo { 395962b99540SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 396062b99540SQu Wenruo struct inode *inode = rc->data_inode; 396162b99540SQu Wenruo struct btrfs_root *data_reloc_root = BTRFS_I(inode)->root; 396262b99540SQu Wenruo struct btrfs_path *path; 396362b99540SQu Wenruo struct btrfs_key key; 396462b99540SQu Wenruo int ret = 0; 396562b99540SQu Wenruo 3966afcdd129SJosef Bacik if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 396762b99540SQu Wenruo return 0; 396862b99540SQu Wenruo 396962b99540SQu Wenruo /* 397062b99540SQu Wenruo * Only for stage where we update data pointers the qgroup fix is 397162b99540SQu Wenruo * valid. 397262b99540SQu Wenruo * For MOVING_DATA stage, we will miss the timing of swapping tree 397362b99540SQu Wenruo * blocks, and won't fix it. 397462b99540SQu Wenruo */ 397562b99540SQu Wenruo if (!(rc->stage == UPDATE_DATA_PTRS && rc->extents_found)) 397662b99540SQu Wenruo return 0; 397762b99540SQu Wenruo 397862b99540SQu Wenruo path = btrfs_alloc_path(); 397962b99540SQu Wenruo if (!path) 398062b99540SQu Wenruo return -ENOMEM; 398162b99540SQu Wenruo key.objectid = btrfs_ino(inode); 398262b99540SQu Wenruo key.type = BTRFS_EXTENT_DATA_KEY; 398362b99540SQu Wenruo key.offset = 0; 398462b99540SQu Wenruo 398562b99540SQu Wenruo ret = btrfs_search_slot(NULL, data_reloc_root, &key, path, 0, 0); 398662b99540SQu Wenruo if (ret < 0) 398762b99540SQu Wenruo goto out; 398862b99540SQu Wenruo 398962b99540SQu Wenruo lock_extent(&BTRFS_I(inode)->io_tree, 0, (u64)-1); 399062b99540SQu Wenruo while (1) { 399162b99540SQu Wenruo struct btrfs_file_extent_item *fi; 399262b99540SQu Wenruo 399362b99540SQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 399462b99540SQu Wenruo if (key.objectid > btrfs_ino(inode)) 399562b99540SQu Wenruo break; 399662b99540SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 399762b99540SQu Wenruo goto next; 399862b99540SQu Wenruo fi = btrfs_item_ptr(path->nodes[0], path->slots[0], 399962b99540SQu Wenruo struct btrfs_file_extent_item); 400062b99540SQu Wenruo if (btrfs_file_extent_type(path->nodes[0], fi) != 400162b99540SQu Wenruo BTRFS_FILE_EXTENT_REG) 400262b99540SQu Wenruo goto next; 400362b99540SQu Wenruo ret = btrfs_qgroup_insert_dirty_extent(trans, fs_info, 400462b99540SQu Wenruo btrfs_file_extent_disk_bytenr(path->nodes[0], fi), 400562b99540SQu Wenruo btrfs_file_extent_disk_num_bytes(path->nodes[0], fi), 400662b99540SQu Wenruo GFP_NOFS); 400762b99540SQu Wenruo if (ret < 0) 400862b99540SQu Wenruo break; 400962b99540SQu Wenruo next: 401062b99540SQu Wenruo ret = btrfs_next_item(data_reloc_root, path); 401162b99540SQu Wenruo if (ret < 0) 401262b99540SQu Wenruo break; 401362b99540SQu Wenruo if (ret > 0) { 401462b99540SQu Wenruo ret = 0; 401562b99540SQu Wenruo break; 401662b99540SQu Wenruo } 401762b99540SQu Wenruo } 401862b99540SQu Wenruo unlock_extent(&BTRFS_I(inode)->io_tree, 0 , (u64)-1); 401962b99540SQu Wenruo out: 402062b99540SQu Wenruo btrfs_free_path(path); 402162b99540SQu Wenruo return ret; 402262b99540SQu Wenruo } 402362b99540SQu Wenruo 40245d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 40255d4f98a2SYan Zheng { 40265d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 40275d4f98a2SYan Zheng struct btrfs_key key; 40285d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 40295d4f98a2SYan Zheng struct btrfs_path *path; 40305d4f98a2SYan Zheng struct btrfs_extent_item *ei; 40315d4f98a2SYan Zheng u64 flags; 40325d4f98a2SYan Zheng u32 item_size; 40335d4f98a2SYan Zheng int ret; 40345d4f98a2SYan Zheng int err = 0; 4035c87f08caSChris Mason int progress = 0; 40365d4f98a2SYan Zheng 40375d4f98a2SYan Zheng path = btrfs_alloc_path(); 40383fd0a558SYan, Zheng if (!path) 40395d4f98a2SYan Zheng return -ENOMEM; 4040e4058b54SDavid Sterba path->reada = READA_FORWARD; 40413fd0a558SYan, Zheng 40423fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 40433fd0a558SYan, Zheng if (ret) { 40443fd0a558SYan, Zheng err = ret; 40453fd0a558SYan, Zheng goto out_free; 40462423fdfbSJiri Slaby } 40475d4f98a2SYan Zheng 40485d4f98a2SYan Zheng while (1) { 40490647bf56SWang Shilong rc->reserved_bytes = 0; 40500647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40510647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40520647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40530647bf56SWang Shilong if (ret) { 40540647bf56SWang Shilong err = ret; 40550647bf56SWang Shilong break; 40560647bf56SWang Shilong } 4057c87f08caSChris Mason progress++; 4058a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40590f788c58SLiu Bo if (IS_ERR(trans)) { 40600f788c58SLiu Bo err = PTR_ERR(trans); 40610f788c58SLiu Bo trans = NULL; 40620f788c58SLiu Bo break; 40630f788c58SLiu Bo } 4064c87f08caSChris Mason restart: 40653fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40663fd0a558SYan, Zheng btrfs_end_transaction(trans, rc->extent_root); 40673fd0a558SYan, Zheng continue; 40683fd0a558SYan, Zheng } 40693fd0a558SYan, Zheng 4070147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40715d4f98a2SYan Zheng if (ret < 0) 40725d4f98a2SYan Zheng err = ret; 40735d4f98a2SYan Zheng if (ret != 0) 40745d4f98a2SYan Zheng break; 40755d4f98a2SYan Zheng 40765d4f98a2SYan Zheng rc->extents_found++; 40775d4f98a2SYan Zheng 40785d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40795d4f98a2SYan Zheng struct btrfs_extent_item); 40803fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40815d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40825d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40835d4f98a2SYan Zheng ret = check_extent_flags(flags); 40845d4f98a2SYan Zheng BUG_ON(ret); 40855d4f98a2SYan Zheng 40865d4f98a2SYan Zheng } else { 40875d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 40885d4f98a2SYan Zheng u64 ref_owner; 40895d4f98a2SYan Zheng int path_change = 0; 40905d4f98a2SYan Zheng 40915d4f98a2SYan Zheng BUG_ON(item_size != 40925d4f98a2SYan Zheng sizeof(struct btrfs_extent_item_v0)); 40935d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, &key, &ref_owner, 40945d4f98a2SYan Zheng &path_change); 40954b3576e4SZhaolei if (ret < 0) { 40964b3576e4SZhaolei err = ret; 40974b3576e4SZhaolei break; 40984b3576e4SZhaolei } 40995d4f98a2SYan Zheng if (ref_owner < BTRFS_FIRST_FREE_OBJECTID) 41005d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_TREE_BLOCK; 41015d4f98a2SYan Zheng else 41025d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_DATA; 41035d4f98a2SYan Zheng 41045d4f98a2SYan Zheng if (path_change) { 4105b3b4aa74SDavid Sterba btrfs_release_path(path); 41065d4f98a2SYan Zheng 41075d4f98a2SYan Zheng path->search_commit_root = 1; 41085d4f98a2SYan Zheng path->skip_locking = 1; 41095d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, 41105d4f98a2SYan Zheng &key, path, 0, 0); 41115d4f98a2SYan Zheng if (ret < 0) { 41125d4f98a2SYan Zheng err = ret; 41135d4f98a2SYan Zheng break; 41145d4f98a2SYan Zheng } 41155d4f98a2SYan Zheng BUG_ON(ret > 0); 41165d4f98a2SYan Zheng } 41175d4f98a2SYan Zheng #else 41185d4f98a2SYan Zheng BUG(); 41195d4f98a2SYan Zheng #endif 41205d4f98a2SYan Zheng } 41215d4f98a2SYan Zheng 41225d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 41235d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 41245d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 41255d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41265d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 41275d4f98a2SYan Zheng } else { 4128b3b4aa74SDavid Sterba btrfs_release_path(path); 41295d4f98a2SYan Zheng ret = 0; 41305d4f98a2SYan Zheng } 41315d4f98a2SYan Zheng if (ret < 0) { 41323fd0a558SYan, Zheng err = ret; 41335d4f98a2SYan Zheng break; 41345d4f98a2SYan Zheng } 41355d4f98a2SYan Zheng 41365d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 41375d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 41385d4f98a2SYan Zheng if (ret < 0) { 41391708cc57SWang Shilong /* 41401708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 41411708cc57SWang Shilong * backref cache when committing transaction. 41421708cc57SWang Shilong */ 41431708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 41441708cc57SWang Shilong 41453fd0a558SYan, Zheng if (ret != -EAGAIN) { 41465d4f98a2SYan Zheng err = ret; 41475d4f98a2SYan Zheng break; 41485d4f98a2SYan Zheng } 41493fd0a558SYan, Zheng rc->extents_found--; 41503fd0a558SYan, Zheng rc->search_start = key.objectid; 41513fd0a558SYan, Zheng } 41525d4f98a2SYan Zheng } 41535d4f98a2SYan Zheng 41543fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 4155b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 41563fd0a558SYan, Zheng trans = NULL; 41575d4f98a2SYan Zheng 41585d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 41595d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41605d4f98a2SYan Zheng rc->found_file_extent = 1; 41610257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 41623fd0a558SYan, Zheng &key, &rc->cluster); 41635d4f98a2SYan Zheng if (ret < 0) { 41645d4f98a2SYan Zheng err = ret; 41655d4f98a2SYan Zheng break; 41665d4f98a2SYan Zheng } 41675d4f98a2SYan Zheng } 41685d4f98a2SYan Zheng } 4169c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 4170c87f08caSChris Mason ret = btrfs_force_chunk_alloc(trans, rc->extent_root, 4171c87f08caSChris Mason rc->block_group->flags); 41729689457bSShilong Wang if (ret == 1) { 4173c87f08caSChris Mason err = 0; 4174c87f08caSChris Mason progress = 0; 4175c87f08caSChris Mason goto restart; 4176c87f08caSChris Mason } 4177c87f08caSChris Mason } 41783fd0a558SYan, Zheng 4179b3b4aa74SDavid Sterba btrfs_release_path(path); 418091166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41815d4f98a2SYan Zheng 41825d4f98a2SYan Zheng if (trans) { 41833fd0a558SYan, Zheng btrfs_end_transaction_throttle(trans, rc->extent_root); 4184b53d3f5dSLiu Bo btrfs_btree_balance_dirty(rc->extent_root); 41855d4f98a2SYan Zheng } 41865d4f98a2SYan Zheng 41870257bb82SYan, Zheng if (!err) { 41883fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 41893fd0a558SYan, Zheng &rc->cluster); 41900257bb82SYan, Zheng if (ret < 0) 41910257bb82SYan, Zheng err = ret; 41920257bb82SYan, Zheng } 41930257bb82SYan, Zheng 41943fd0a558SYan, Zheng rc->create_reloc_tree = 0; 41953fd0a558SYan, Zheng set_reloc_control(rc); 41960257bb82SYan, Zheng 41973fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 41983fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 41995d4f98a2SYan Zheng 42003fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 42015d4f98a2SYan Zheng 42025d4f98a2SYan Zheng merge_reloc_roots(rc); 42035d4f98a2SYan Zheng 42043fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 42055d4f98a2SYan Zheng unset_reloc_control(rc); 42063fd0a558SYan, Zheng btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1); 42075d4f98a2SYan Zheng 42085d4f98a2SYan Zheng /* get rid of pinned extents */ 42097a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 421062b99540SQu Wenruo if (IS_ERR(trans)) { 42113612b495STsutomu Itoh err = PTR_ERR(trans); 421262b99540SQu Wenruo goto out_free; 421362b99540SQu Wenruo } 4214a9b1fc85SLiu Bo ret = qgroup_fix_relocated_data_extents(trans, rc); 4215a9b1fc85SLiu Bo if (ret < 0) { 4216a9b1fc85SLiu Bo btrfs_abort_transaction(trans, ret); 4217a9b1fc85SLiu Bo if (!err) 4218a9b1fc85SLiu Bo err = ret; 421962b99540SQu Wenruo goto out_free; 422062b99540SQu Wenruo } 42215d4f98a2SYan Zheng btrfs_commit_transaction(trans, rc->extent_root); 42223fd0a558SYan, Zheng out_free: 42233fd0a558SYan, Zheng btrfs_free_block_rsv(rc->extent_root, rc->block_rsv); 42243fd0a558SYan, Zheng btrfs_free_path(path); 42255d4f98a2SYan Zheng return err; 42265d4f98a2SYan Zheng } 42275d4f98a2SYan Zheng 42285d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 42290257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 42305d4f98a2SYan Zheng { 42315d4f98a2SYan Zheng struct btrfs_path *path; 42325d4f98a2SYan Zheng struct btrfs_inode_item *item; 42335d4f98a2SYan Zheng struct extent_buffer *leaf; 42345d4f98a2SYan Zheng int ret; 42355d4f98a2SYan Zheng 42365d4f98a2SYan Zheng path = btrfs_alloc_path(); 42375d4f98a2SYan Zheng if (!path) 42385d4f98a2SYan Zheng return -ENOMEM; 42395d4f98a2SYan Zheng 42405d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 42415d4f98a2SYan Zheng if (ret) 42425d4f98a2SYan Zheng goto out; 42435d4f98a2SYan Zheng 42445d4f98a2SYan Zheng leaf = path->nodes[0]; 42455d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 42465d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); 42475d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 42480257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 42495d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 42503fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 42513fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 42525d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 42535d4f98a2SYan Zheng out: 42545d4f98a2SYan Zheng btrfs_free_path(path); 42555d4f98a2SYan Zheng return ret; 42565d4f98a2SYan Zheng } 42575d4f98a2SYan Zheng 42585d4f98a2SYan Zheng /* 42595d4f98a2SYan Zheng * helper to create inode for data relocation. 42605d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 42615d4f98a2SYan Zheng */ 42623fd0a558SYan, Zheng static noinline_for_stack 42633fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 42645d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 42655d4f98a2SYan Zheng { 42665d4f98a2SYan Zheng struct inode *inode = NULL; 42675d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42685d4f98a2SYan Zheng struct btrfs_root *root; 42695d4f98a2SYan Zheng struct btrfs_key key; 42704624900dSZhaolei u64 objectid; 42715d4f98a2SYan Zheng int err = 0; 42725d4f98a2SYan Zheng 42735d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 42745d4f98a2SYan Zheng if (IS_ERR(root)) 42755d4f98a2SYan Zheng return ERR_CAST(root); 42765d4f98a2SYan Zheng 4277a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 42783fd0a558SYan, Zheng if (IS_ERR(trans)) 42793fd0a558SYan, Zheng return ERR_CAST(trans); 42805d4f98a2SYan Zheng 4281581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42825d4f98a2SYan Zheng if (err) 42835d4f98a2SYan Zheng goto out; 42845d4f98a2SYan Zheng 42850257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42865d4f98a2SYan Zheng BUG_ON(err); 42875d4f98a2SYan Zheng 42885d4f98a2SYan Zheng key.objectid = objectid; 42895d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 42905d4f98a2SYan Zheng key.offset = 0; 429173f73415SJosef Bacik inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); 42925d4f98a2SYan Zheng BUG_ON(IS_ERR(inode) || is_bad_inode(inode)); 42935d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 42945d4f98a2SYan Zheng 42955d4f98a2SYan Zheng err = btrfs_orphan_add(trans, inode); 42965d4f98a2SYan Zheng out: 42975d4f98a2SYan Zheng btrfs_end_transaction(trans, root); 4298b53d3f5dSLiu Bo btrfs_btree_balance_dirty(root); 42995d4f98a2SYan Zheng if (err) { 43005d4f98a2SYan Zheng if (inode) 43015d4f98a2SYan Zheng iput(inode); 43025d4f98a2SYan Zheng inode = ERR_PTR(err); 43035d4f98a2SYan Zheng } 43045d4f98a2SYan Zheng return inode; 43055d4f98a2SYan Zheng } 43065d4f98a2SYan Zheng 4307a9995eecSJosef Bacik static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 43083fd0a558SYan, Zheng { 43093fd0a558SYan, Zheng struct reloc_control *rc; 43103fd0a558SYan, Zheng 43113fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 43123fd0a558SYan, Zheng if (!rc) 43133fd0a558SYan, Zheng return NULL; 43143fd0a558SYan, Zheng 43153fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 43163fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 43173fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 4318a9995eecSJosef Bacik extent_io_tree_init(&rc->processed_blocks, 4319a9995eecSJosef Bacik fs_info->btree_inode->i_mapping); 43203fd0a558SYan, Zheng return rc; 43213fd0a558SYan, Zheng } 43223fd0a558SYan, Zheng 43235d4f98a2SYan Zheng /* 43245d4f98a2SYan Zheng * function to relocate all extents in a block group. 43255d4f98a2SYan Zheng */ 43265d4f98a2SYan Zheng int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start) 43275d4f98a2SYan Zheng { 43285d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = extent_root->fs_info; 43295d4f98a2SYan Zheng struct reloc_control *rc; 43300af3d00bSJosef Bacik struct inode *inode; 43310af3d00bSJosef Bacik struct btrfs_path *path; 43325d4f98a2SYan Zheng int ret; 4333f0486c68SYan, Zheng int rw = 0; 43345d4f98a2SYan Zheng int err = 0; 43355d4f98a2SYan Zheng 4336a9995eecSJosef Bacik rc = alloc_reloc_control(fs_info); 43375d4f98a2SYan Zheng if (!rc) 43385d4f98a2SYan Zheng return -ENOMEM; 43395d4f98a2SYan Zheng 4340f0486c68SYan, Zheng rc->extent_root = extent_root; 43413fd0a558SYan, Zheng 43425d4f98a2SYan Zheng rc->block_group = btrfs_lookup_block_group(fs_info, group_start); 43435d4f98a2SYan Zheng BUG_ON(!rc->block_group); 43445d4f98a2SYan Zheng 4345868f401aSZhaolei ret = btrfs_inc_block_group_ro(extent_root, rc->block_group); 4346f0486c68SYan, Zheng if (ret) { 4347f0486c68SYan, Zheng err = ret; 4348f0486c68SYan, Zheng goto out; 4349f0486c68SYan, Zheng } 4350f0486c68SYan, Zheng rw = 1; 4351f0486c68SYan, Zheng 43520af3d00bSJosef Bacik path = btrfs_alloc_path(); 43530af3d00bSJosef Bacik if (!path) { 43540af3d00bSJosef Bacik err = -ENOMEM; 43550af3d00bSJosef Bacik goto out; 43560af3d00bSJosef Bacik } 43570af3d00bSJosef Bacik 43580af3d00bSJosef Bacik inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group, 43590af3d00bSJosef Bacik path); 43600af3d00bSJosef Bacik btrfs_free_path(path); 43610af3d00bSJosef Bacik 43620af3d00bSJosef Bacik if (!IS_ERR(inode)) 43631bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43640af3d00bSJosef Bacik else 43650af3d00bSJosef Bacik ret = PTR_ERR(inode); 43660af3d00bSJosef Bacik 43670af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 43680af3d00bSJosef Bacik err = ret; 43690af3d00bSJosef Bacik goto out; 43700af3d00bSJosef Bacik } 43710af3d00bSJosef Bacik 43725d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 43735d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 43745d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 43755d4f98a2SYan Zheng rc->data_inode = NULL; 43765d4f98a2SYan Zheng goto out; 43775d4f98a2SYan Zheng } 43785d4f98a2SYan Zheng 4379efe120a0SFrank Holton btrfs_info(extent_root->fs_info, "relocating block group %llu flags %llu", 4380c1c9ff7cSGeert Uytterhoeven rc->block_group->key.objectid, rc->block_group->flags); 43815d4f98a2SYan Zheng 43829cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4383f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 4384578def7cSFilipe Manana btrfs_wait_ordered_roots(fs_info, -1, 4385578def7cSFilipe Manana rc->block_group->key.objectid, 4386578def7cSFilipe Manana rc->block_group->key.offset); 43875d4f98a2SYan Zheng 43885d4f98a2SYan Zheng while (1) { 438976dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 43905d4f98a2SYan Zheng ret = relocate_block_group(rc); 439176dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 43925d4f98a2SYan Zheng if (ret < 0) { 43935d4f98a2SYan Zheng err = ret; 43943fd0a558SYan, Zheng goto out; 43955d4f98a2SYan Zheng } 43965d4f98a2SYan Zheng 43975d4f98a2SYan Zheng if (rc->extents_found == 0) 43985d4f98a2SYan Zheng break; 43995d4f98a2SYan Zheng 4400efe120a0SFrank Holton btrfs_info(extent_root->fs_info, "found %llu extents", 4401c1c9ff7cSGeert Uytterhoeven rc->extents_found); 44025d4f98a2SYan Zheng 44035d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 44040ef8b726SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 44050ef8b726SJosef Bacik (u64)-1); 44060ef8b726SJosef Bacik if (ret) { 44070ef8b726SJosef Bacik err = ret; 44080ef8b726SJosef Bacik goto out; 44090ef8b726SJosef Bacik } 44105d4f98a2SYan Zheng invalidate_mapping_pages(rc->data_inode->i_mapping, 44115d4f98a2SYan Zheng 0, -1); 44125d4f98a2SYan Zheng rc->stage = UPDATE_DATA_PTRS; 44135d4f98a2SYan Zheng } 44145d4f98a2SYan Zheng } 44155d4f98a2SYan Zheng 44165d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 44175d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 44185d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 44195d4f98a2SYan Zheng out: 4420f0486c68SYan, Zheng if (err && rw) 4421868f401aSZhaolei btrfs_dec_block_group_ro(extent_root, rc->block_group); 44225d4f98a2SYan Zheng iput(rc->data_inode); 44235d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 44245d4f98a2SYan Zheng kfree(rc); 44255d4f98a2SYan Zheng return err; 44265d4f98a2SYan Zheng } 44275d4f98a2SYan Zheng 442876dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 442976dda93cSYan, Zheng { 443076dda93cSYan, Zheng struct btrfs_trans_handle *trans; 443179787eaaSJeff Mahoney int ret, err; 443276dda93cSYan, Zheng 4433a22285a6SYan, Zheng trans = btrfs_start_transaction(root->fs_info->tree_root, 0); 443479787eaaSJeff Mahoney if (IS_ERR(trans)) 443579787eaaSJeff Mahoney return PTR_ERR(trans); 443676dda93cSYan, Zheng 443776dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 443876dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 443976dda93cSYan, Zheng root->root_item.drop_level = 0; 444076dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 444176dda93cSYan, Zheng ret = btrfs_update_root(trans, root->fs_info->tree_root, 444276dda93cSYan, Zheng &root->root_key, &root->root_item); 444376dda93cSYan, Zheng 444479787eaaSJeff Mahoney err = btrfs_end_transaction(trans, root->fs_info->tree_root); 444579787eaaSJeff Mahoney if (err) 444679787eaaSJeff Mahoney return err; 444779787eaaSJeff Mahoney return ret; 444876dda93cSYan, Zheng } 444976dda93cSYan, Zheng 44505d4f98a2SYan Zheng /* 44515d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44525d4f98a2SYan Zheng * 44535d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44545d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44555d4f98a2SYan Zheng */ 44565d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44575d4f98a2SYan Zheng { 44585d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 44595d4f98a2SYan Zheng struct btrfs_key key; 44605d4f98a2SYan Zheng struct btrfs_root *fs_root; 44615d4f98a2SYan Zheng struct btrfs_root *reloc_root; 44625d4f98a2SYan Zheng struct btrfs_path *path; 44635d4f98a2SYan Zheng struct extent_buffer *leaf; 44645d4f98a2SYan Zheng struct reloc_control *rc = NULL; 44655d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 44665d4f98a2SYan Zheng int ret; 44675d4f98a2SYan Zheng int err = 0; 44685d4f98a2SYan Zheng 44695d4f98a2SYan Zheng path = btrfs_alloc_path(); 44705d4f98a2SYan Zheng if (!path) 44715d4f98a2SYan Zheng return -ENOMEM; 4472e4058b54SDavid Sterba path->reada = READA_BACK; 44735d4f98a2SYan Zheng 44745d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 44755d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 44765d4f98a2SYan Zheng key.offset = (u64)-1; 44775d4f98a2SYan Zheng 44785d4f98a2SYan Zheng while (1) { 44795d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, 44805d4f98a2SYan Zheng path, 0, 0); 44815d4f98a2SYan Zheng if (ret < 0) { 44825d4f98a2SYan Zheng err = ret; 44835d4f98a2SYan Zheng goto out; 44845d4f98a2SYan Zheng } 44855d4f98a2SYan Zheng if (ret > 0) { 44865d4f98a2SYan Zheng if (path->slots[0] == 0) 44875d4f98a2SYan Zheng break; 44885d4f98a2SYan Zheng path->slots[0]--; 44895d4f98a2SYan Zheng } 44905d4f98a2SYan Zheng leaf = path->nodes[0]; 44915d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4492b3b4aa74SDavid Sterba btrfs_release_path(path); 44935d4f98a2SYan Zheng 44945d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 44955d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 44965d4f98a2SYan Zheng break; 44975d4f98a2SYan Zheng 4498cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 44995d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 45005d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 45015d4f98a2SYan Zheng goto out; 45025d4f98a2SYan Zheng } 45035d4f98a2SYan Zheng 45045d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 45055d4f98a2SYan Zheng 45065d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 45075d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 45085d4f98a2SYan Zheng reloc_root->root_key.offset); 45095d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 451076dda93cSYan, Zheng ret = PTR_ERR(fs_root); 451176dda93cSYan, Zheng if (ret != -ENOENT) { 451276dda93cSYan, Zheng err = ret; 45135d4f98a2SYan Zheng goto out; 45145d4f98a2SYan Zheng } 451579787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 451679787eaaSJeff Mahoney if (ret < 0) { 451779787eaaSJeff Mahoney err = ret; 451879787eaaSJeff Mahoney goto out; 451979787eaaSJeff Mahoney } 452076dda93cSYan, Zheng } 45215d4f98a2SYan Zheng } 45225d4f98a2SYan Zheng 45235d4f98a2SYan Zheng if (key.offset == 0) 45245d4f98a2SYan Zheng break; 45255d4f98a2SYan Zheng 45265d4f98a2SYan Zheng key.offset--; 45275d4f98a2SYan Zheng } 4528b3b4aa74SDavid Sterba btrfs_release_path(path); 45295d4f98a2SYan Zheng 45305d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 45315d4f98a2SYan Zheng goto out; 45325d4f98a2SYan Zheng 4533a9995eecSJosef Bacik rc = alloc_reloc_control(root->fs_info); 45345d4f98a2SYan Zheng if (!rc) { 45355d4f98a2SYan Zheng err = -ENOMEM; 45365d4f98a2SYan Zheng goto out; 45375d4f98a2SYan Zheng } 45385d4f98a2SYan Zheng 45395d4f98a2SYan Zheng rc->extent_root = root->fs_info->extent_root; 45405d4f98a2SYan Zheng 45415d4f98a2SYan Zheng set_reloc_control(rc); 45425d4f98a2SYan Zheng 45437a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45443612b495STsutomu Itoh if (IS_ERR(trans)) { 45453612b495STsutomu Itoh unset_reloc_control(rc); 45463612b495STsutomu Itoh err = PTR_ERR(trans); 45473612b495STsutomu Itoh goto out_free; 45483612b495STsutomu Itoh } 45493fd0a558SYan, Zheng 45503fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45513fd0a558SYan, Zheng 45525d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45535d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45545d4f98a2SYan Zheng struct btrfs_root, root_list); 45555d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45565d4f98a2SYan Zheng 45575d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 45585d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 45595d4f98a2SYan Zheng &rc->reloc_roots); 45605d4f98a2SYan Zheng continue; 45615d4f98a2SYan Zheng } 45625d4f98a2SYan Zheng 45635d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 45645d4f98a2SYan Zheng reloc_root->root_key.offset); 456579787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 456679787eaaSJeff Mahoney err = PTR_ERR(fs_root); 456779787eaaSJeff Mahoney goto out_free; 456879787eaaSJeff Mahoney } 45695d4f98a2SYan Zheng 4570ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 457179787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 45725d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 45735d4f98a2SYan Zheng } 45745d4f98a2SYan Zheng 457579787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 457679787eaaSJeff Mahoney if (err) 457779787eaaSJeff Mahoney goto out_free; 45785d4f98a2SYan Zheng 45795d4f98a2SYan Zheng merge_reloc_roots(rc); 45805d4f98a2SYan Zheng 45815d4f98a2SYan Zheng unset_reloc_control(rc); 45825d4f98a2SYan Zheng 45837a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 458462b99540SQu Wenruo if (IS_ERR(trans)) { 45853612b495STsutomu Itoh err = PTR_ERR(trans); 458662b99540SQu Wenruo goto out_free; 458762b99540SQu Wenruo } 458862b99540SQu Wenruo err = qgroup_fix_relocated_data_extents(trans, rc); 458962b99540SQu Wenruo if (err < 0) { 459062b99540SQu Wenruo btrfs_abort_transaction(trans, err); 459162b99540SQu Wenruo goto out_free; 459262b99540SQu Wenruo } 459379787eaaSJeff Mahoney err = btrfs_commit_transaction(trans, rc->extent_root); 45943612b495STsutomu Itoh out_free: 45955d4f98a2SYan Zheng kfree(rc); 45963612b495STsutomu Itoh out: 4597aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4598aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4599aca1bba6SLiu Bo 46005d4f98a2SYan Zheng btrfs_free_path(path); 46015d4f98a2SYan Zheng 46025d4f98a2SYan Zheng if (err == 0) { 46035d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 46045d4f98a2SYan Zheng fs_root = read_fs_root(root->fs_info, 46055d4f98a2SYan Zheng BTRFS_DATA_RELOC_TREE_OBJECTID); 46065d4f98a2SYan Zheng if (IS_ERR(fs_root)) 46075d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4608d7ce5843SMiao Xie else 460966b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 46105d4f98a2SYan Zheng } 46115d4f98a2SYan Zheng return err; 46125d4f98a2SYan Zheng } 46135d4f98a2SYan Zheng 46145d4f98a2SYan Zheng /* 46155d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 46165d4f98a2SYan Zheng * 46175d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 46185d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 46195d4f98a2SYan Zheng */ 46205d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 46215d4f98a2SYan Zheng { 46225d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 46235d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 46245d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(inode)->root; 46255d4f98a2SYan Zheng int ret; 46265d4f98a2SYan Zheng u64 disk_bytenr; 46274577b014SJosef Bacik u64 new_bytenr; 46285d4f98a2SYan Zheng LIST_HEAD(list); 46295d4f98a2SYan Zheng 46305d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 46315d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 46325d4f98a2SYan Zheng 46335d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46345d4f98a2SYan Zheng ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr, 4635a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 463679787eaaSJeff Mahoney if (ret) 463779787eaaSJeff Mahoney goto out; 46385d4f98a2SYan Zheng 46395d4f98a2SYan Zheng while (!list_empty(&list)) { 46405d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46415d4f98a2SYan Zheng list_del_init(&sums->list); 46425d4f98a2SYan Zheng 46434577b014SJosef Bacik /* 46444577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46454577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46464577b014SJosef Bacik * extents but we may have written to say the middle of the 46474577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46484577b014SJosef Bacik * the right disk offset. 46494577b014SJosef Bacik * 46504577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46514577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46524577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46534577b014SJosef Bacik * disk length. 46544577b014SJosef Bacik */ 46554577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 46564577b014SJosef Bacik sums->bytenr = new_bytenr; 46575d4f98a2SYan Zheng 46585d4f98a2SYan Zheng btrfs_add_ordered_sum(inode, ordered, sums); 46595d4f98a2SYan Zheng } 466079787eaaSJeff Mahoney out: 46615d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4662411fc6bcSAndi Kleen return ret; 46635d4f98a2SYan Zheng } 46643fd0a558SYan, Zheng 466583d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 46663fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 46673fd0a558SYan, Zheng struct extent_buffer *cow) 46683fd0a558SYan, Zheng { 46693fd0a558SYan, Zheng struct reloc_control *rc; 46703fd0a558SYan, Zheng struct backref_node *node; 46713fd0a558SYan, Zheng int first_cow = 0; 46723fd0a558SYan, Zheng int level; 467383d4cfd4SJosef Bacik int ret = 0; 46743fd0a558SYan, Zheng 46753fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 46763fd0a558SYan, Zheng if (!rc) 467783d4cfd4SJosef Bacik return 0; 46783fd0a558SYan, Zheng 46793fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46803fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46813fd0a558SYan, Zheng 4682c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4683c974c464SWang Shilong if (buf == root->node) 4684c974c464SWang Shilong __update_reloc_root(root, cow->start); 4685c974c464SWang Shilong } 4686c974c464SWang Shilong 46873fd0a558SYan, Zheng level = btrfs_header_level(buf); 46883fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46893fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 46903fd0a558SYan, Zheng first_cow = 1; 46913fd0a558SYan, Zheng 46923fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 46933fd0a558SYan, Zheng rc->create_reloc_tree) { 46943fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 46953fd0a558SYan, Zheng 46963fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 46973fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 46983fd0a558SYan, Zheng node->new_bytenr != buf->start); 46993fd0a558SYan, Zheng 47003fd0a558SYan, Zheng drop_node_buffer(node); 47013fd0a558SYan, Zheng extent_buffer_get(cow); 47023fd0a558SYan, Zheng node->eb = cow; 47033fd0a558SYan, Zheng node->new_bytenr = cow->start; 47043fd0a558SYan, Zheng 47053fd0a558SYan, Zheng if (!node->pending) { 47063fd0a558SYan, Zheng list_move_tail(&node->list, 47073fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 47083fd0a558SYan, Zheng node->pending = 1; 47093fd0a558SYan, Zheng } 47103fd0a558SYan, Zheng 47113fd0a558SYan, Zheng if (first_cow) 47123fd0a558SYan, Zheng __mark_block_processed(rc, node); 47133fd0a558SYan, Zheng 47143fd0a558SYan, Zheng if (first_cow && level > 0) 47153fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 47163fd0a558SYan, Zheng } 47173fd0a558SYan, Zheng 471883d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 47193fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 472083d4cfd4SJosef Bacik return ret; 47213fd0a558SYan, Zheng } 47223fd0a558SYan, Zheng 47233fd0a558SYan, Zheng /* 47243fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 472501327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 47263fd0a558SYan, Zheng */ 4727147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 47283fd0a558SYan, Zheng u64 *bytes_to_reserve) 47293fd0a558SYan, Zheng { 47303fd0a558SYan, Zheng struct btrfs_root *root; 47313fd0a558SYan, Zheng struct reloc_control *rc; 47323fd0a558SYan, Zheng 47333fd0a558SYan, Zheng root = pending->root; 47343fd0a558SYan, Zheng if (!root->reloc_root) 47353fd0a558SYan, Zheng return; 47363fd0a558SYan, Zheng 47373fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47383fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47393fd0a558SYan, Zheng return; 47403fd0a558SYan, Zheng 47413fd0a558SYan, Zheng root = root->reloc_root; 47423fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47433fd0a558SYan, Zheng /* 47443fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47453fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47463fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47473fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47483fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47493fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 47503fd0a558SYan, Zheng * tree may use more space than it frees. so we need 47513fd0a558SYan, Zheng * reserve extra space. 47523fd0a558SYan, Zheng */ 47533fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47543fd0a558SYan, Zheng } 47553fd0a558SYan, Zheng 47563fd0a558SYan, Zheng /* 47573fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47583fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 47593fd0a558SYan, Zheng */ 476049b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 47613fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 47623fd0a558SYan, Zheng { 47633fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 47643fd0a558SYan, Zheng struct btrfs_root *reloc_root; 47653fd0a558SYan, Zheng struct btrfs_root *new_root; 47663fd0a558SYan, Zheng struct reloc_control *rc; 47673fd0a558SYan, Zheng int ret; 47683fd0a558SYan, Zheng 47693fd0a558SYan, Zheng if (!root->reloc_root) 477049b25e05SJeff Mahoney return 0; 47713fd0a558SYan, Zheng 47723fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47733fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 47743fd0a558SYan, Zheng 47753fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 47763fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 47773fd0a558SYan, Zheng rc->block_rsv, 477825d609f8SJosef Bacik rc->nodes_relocated, 1); 477949b25e05SJeff Mahoney if (ret) 478049b25e05SJeff Mahoney return ret; 47813fd0a558SYan, Zheng } 47823fd0a558SYan, Zheng 47833fd0a558SYan, Zheng new_root = pending->snap; 47843fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47853fd0a558SYan, Zheng new_root->root_key.objectid); 478649b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 478749b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47883fd0a558SYan, Zheng 4789ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4790ffd7b339SJeff Mahoney BUG_ON(ret < 0); 47913fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 47923fd0a558SYan, Zheng 479349b25e05SJeff Mahoney if (rc->create_reloc_tree) 47943fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 479549b25e05SJeff Mahoney return ret; 47963fd0a558SYan, Zheng } 4797