1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 25d4f98a2SYan Zheng /* 35d4f98a2SYan Zheng * Copyright (C) 2009 Oracle. All rights reserved. 45d4f98a2SYan Zheng */ 55d4f98a2SYan Zheng 65d4f98a2SYan Zheng #include <linux/sched.h> 75d4f98a2SYan Zheng #include <linux/pagemap.h> 85d4f98a2SYan Zheng #include <linux/writeback.h> 95d4f98a2SYan Zheng #include <linux/blkdev.h> 105d4f98a2SYan Zheng #include <linux/rbtree.h> 115a0e3ad6STejun Heo #include <linux/slab.h> 125d4f98a2SYan Zheng #include "ctree.h" 135d4f98a2SYan Zheng #include "disk-io.h" 145d4f98a2SYan Zheng #include "transaction.h" 155d4f98a2SYan Zheng #include "volumes.h" 165d4f98a2SYan Zheng #include "locking.h" 175d4f98a2SYan Zheng #include "btrfs_inode.h" 185d4f98a2SYan Zheng #include "async-thread.h" 190af3d00bSJosef Bacik #include "free-space-cache.h" 20581bb050SLi Zefan #include "inode-map.h" 2162b99540SQu Wenruo #include "qgroup.h" 22cdccee99SLiu Bo #include "print-tree.h" 235d4f98a2SYan Zheng 245d4f98a2SYan Zheng /* 255d4f98a2SYan Zheng * backref_node, mapping_node and tree_block start with this 265d4f98a2SYan Zheng */ 275d4f98a2SYan Zheng struct tree_entry { 285d4f98a2SYan Zheng struct rb_node rb_node; 295d4f98a2SYan Zheng u64 bytenr; 305d4f98a2SYan Zheng }; 315d4f98a2SYan Zheng 325d4f98a2SYan Zheng /* 335d4f98a2SYan Zheng * present a tree block in the backref cache 345d4f98a2SYan Zheng */ 355d4f98a2SYan Zheng struct backref_node { 365d4f98a2SYan Zheng struct rb_node rb_node; 375d4f98a2SYan Zheng u64 bytenr; 383fd0a558SYan, Zheng 393fd0a558SYan, Zheng u64 new_bytenr; 403fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 415d4f98a2SYan Zheng u64 owner; 423fd0a558SYan, Zheng /* link to pending, changed or detached list */ 433fd0a558SYan, Zheng struct list_head list; 445d4f98a2SYan Zheng /* list of upper level blocks reference this block */ 455d4f98a2SYan Zheng struct list_head upper; 465d4f98a2SYan Zheng /* list of child blocks in the cache */ 475d4f98a2SYan Zheng struct list_head lower; 485d4f98a2SYan Zheng /* NULL if this node is not tree root */ 495d4f98a2SYan Zheng struct btrfs_root *root; 505d4f98a2SYan Zheng /* extent buffer got by COW the block */ 515d4f98a2SYan Zheng struct extent_buffer *eb; 525d4f98a2SYan Zheng /* level of tree block */ 535d4f98a2SYan Zheng unsigned int level:8; 543fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 553fd0a558SYan, Zheng unsigned int cowonly:1; 563fd0a558SYan, Zheng /* 1 if no child node in the cache */ 575d4f98a2SYan Zheng unsigned int lowest:1; 585d4f98a2SYan Zheng /* is the extent buffer locked */ 595d4f98a2SYan Zheng unsigned int locked:1; 605d4f98a2SYan Zheng /* has the block been processed */ 615d4f98a2SYan Zheng unsigned int processed:1; 625d4f98a2SYan Zheng /* have backrefs of this block been checked */ 635d4f98a2SYan Zheng unsigned int checked:1; 643fd0a558SYan, Zheng /* 653fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 663fd0a558SYan, Zheng * level block pointers may not point to the new location 673fd0a558SYan, Zheng */ 683fd0a558SYan, Zheng unsigned int pending:1; 693fd0a558SYan, Zheng /* 703fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 713fd0a558SYan, Zheng * backref node. 723fd0a558SYan, Zheng */ 733fd0a558SYan, Zheng unsigned int detached:1; 745d4f98a2SYan Zheng }; 755d4f98a2SYan Zheng 765d4f98a2SYan Zheng /* 775d4f98a2SYan Zheng * present a block pointer in the backref cache 785d4f98a2SYan Zheng */ 795d4f98a2SYan Zheng struct backref_edge { 805d4f98a2SYan Zheng struct list_head list[2]; 815d4f98a2SYan Zheng struct backref_node *node[2]; 825d4f98a2SYan Zheng }; 835d4f98a2SYan Zheng 845d4f98a2SYan Zheng #define LOWER 0 855d4f98a2SYan Zheng #define UPPER 1 860647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 875d4f98a2SYan Zheng 885d4f98a2SYan Zheng struct backref_cache { 895d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 905d4f98a2SYan Zheng struct rb_root rb_root; 913fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 923fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 933fd0a558SYan, Zheng /* 943fd0a558SYan, Zheng * list of blocks that have been cowed but some block 953fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 963fd0a558SYan, Zheng * new location 973fd0a558SYan, Zheng */ 985d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 993fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1003fd0a558SYan, Zheng struct list_head leaves; 1013fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1023fd0a558SYan, Zheng struct list_head changed; 1033fd0a558SYan, Zheng /* list of detached backref node. */ 1043fd0a558SYan, Zheng struct list_head detached; 1053fd0a558SYan, Zheng 1063fd0a558SYan, Zheng u64 last_trans; 1073fd0a558SYan, Zheng 1083fd0a558SYan, Zheng int nr_nodes; 1093fd0a558SYan, Zheng int nr_edges; 1105d4f98a2SYan Zheng }; 1115d4f98a2SYan Zheng 1125d4f98a2SYan Zheng /* 1135d4f98a2SYan Zheng * map address of tree root to tree 1145d4f98a2SYan Zheng */ 1155d4f98a2SYan Zheng struct mapping_node { 1165d4f98a2SYan Zheng struct rb_node rb_node; 1175d4f98a2SYan Zheng u64 bytenr; 1185d4f98a2SYan Zheng void *data; 1195d4f98a2SYan Zheng }; 1205d4f98a2SYan Zheng 1215d4f98a2SYan Zheng struct mapping_tree { 1225d4f98a2SYan Zheng struct rb_root rb_root; 1235d4f98a2SYan Zheng spinlock_t lock; 1245d4f98a2SYan Zheng }; 1255d4f98a2SYan Zheng 1265d4f98a2SYan Zheng /* 1275d4f98a2SYan Zheng * present a tree block to process 1285d4f98a2SYan Zheng */ 1295d4f98a2SYan Zheng struct tree_block { 1305d4f98a2SYan Zheng struct rb_node rb_node; 1315d4f98a2SYan Zheng u64 bytenr; 1325d4f98a2SYan Zheng struct btrfs_key key; 1335d4f98a2SYan Zheng unsigned int level:8; 1345d4f98a2SYan Zheng unsigned int key_ready:1; 1355d4f98a2SYan Zheng }; 1365d4f98a2SYan Zheng 1370257bb82SYan, Zheng #define MAX_EXTENTS 128 1380257bb82SYan, Zheng 1390257bb82SYan, Zheng struct file_extent_cluster { 1400257bb82SYan, Zheng u64 start; 1410257bb82SYan, Zheng u64 end; 1420257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1430257bb82SYan, Zheng unsigned int nr; 1440257bb82SYan, Zheng }; 1450257bb82SYan, Zheng 1465d4f98a2SYan Zheng struct reloc_control { 1475d4f98a2SYan Zheng /* block group to relocate */ 1485d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group; 1495d4f98a2SYan Zheng /* extent tree */ 1505d4f98a2SYan Zheng struct btrfs_root *extent_root; 1515d4f98a2SYan Zheng /* inode for moving data */ 1525d4f98a2SYan Zheng struct inode *data_inode; 1533fd0a558SYan, Zheng 1543fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1553fd0a558SYan, Zheng 1563fd0a558SYan, Zheng struct backref_cache backref_cache; 1573fd0a558SYan, Zheng 1583fd0a558SYan, Zheng struct file_extent_cluster cluster; 1595d4f98a2SYan Zheng /* tree blocks have been processed */ 1605d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1615d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1625d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1635d4f98a2SYan Zheng /* list of reloc trees */ 1645d4f98a2SYan Zheng struct list_head reloc_roots; 1653fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1663fd0a558SYan, Zheng u64 merging_rsv_size; 1673fd0a558SYan, Zheng /* size of relocated tree nodes */ 1683fd0a558SYan, Zheng u64 nodes_relocated; 1690647bf56SWang Shilong /* reserved size for block group relocation*/ 1700647bf56SWang Shilong u64 reserved_bytes; 1713fd0a558SYan, Zheng 1725d4f98a2SYan Zheng u64 search_start; 1735d4f98a2SYan Zheng u64 extents_found; 1743fd0a558SYan, Zheng 1753fd0a558SYan, Zheng unsigned int stage:8; 1763fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1773fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1785d4f98a2SYan Zheng unsigned int found_file_extent:1; 1795d4f98a2SYan Zheng }; 1805d4f98a2SYan Zheng 1815d4f98a2SYan Zheng /* stages of data relocation */ 1825d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1835d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1845d4f98a2SYan Zheng 1853fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 1863fd0a558SYan, Zheng struct backref_node *node); 1873fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 1883fd0a558SYan, Zheng struct backref_node *node); 1895d4f98a2SYan Zheng 1905d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1915d4f98a2SYan Zheng { 1926bef4d31SEric Paris tree->rb_root = RB_ROOT; 1935d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1945d4f98a2SYan Zheng } 1955d4f98a2SYan Zheng 1965d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 1975d4f98a2SYan Zheng { 1985d4f98a2SYan Zheng int i; 1996bef4d31SEric Paris cache->rb_root = RB_ROOT; 2005d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2015d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2023fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2033fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2043fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2055d4f98a2SYan Zheng } 2065d4f98a2SYan Zheng 2073fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2085d4f98a2SYan Zheng { 2093fd0a558SYan, Zheng struct backref_node *node; 2103fd0a558SYan, Zheng int i; 2113fd0a558SYan, Zheng 2123fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2133fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2143fd0a558SYan, Zheng struct backref_node, list); 2153fd0a558SYan, Zheng remove_backref_node(cache, node); 2163fd0a558SYan, Zheng } 2173fd0a558SYan, Zheng 2183fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2193fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2203fd0a558SYan, Zheng struct backref_node, lower); 2213fd0a558SYan, Zheng remove_backref_node(cache, node); 2223fd0a558SYan, Zheng } 2233fd0a558SYan, Zheng 2243fd0a558SYan, Zheng cache->last_trans = 0; 2253fd0a558SYan, Zheng 2263fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 227f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 228f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 229f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 230f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 231f4907095SLiu Bo ASSERT(!cache->nr_nodes); 232f4907095SLiu Bo ASSERT(!cache->nr_edges); 2333fd0a558SYan, Zheng } 2343fd0a558SYan, Zheng 2353fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2363fd0a558SYan, Zheng { 2373fd0a558SYan, Zheng struct backref_node *node; 2383fd0a558SYan, Zheng 2393fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2403fd0a558SYan, Zheng if (node) { 2413fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2425d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2435d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2445d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2453fd0a558SYan, Zheng cache->nr_nodes++; 2463fd0a558SYan, Zheng } 2473fd0a558SYan, Zheng return node; 2483fd0a558SYan, Zheng } 2493fd0a558SYan, Zheng 2503fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 2513fd0a558SYan, Zheng struct backref_node *node) 2523fd0a558SYan, Zheng { 2533fd0a558SYan, Zheng if (node) { 2543fd0a558SYan, Zheng cache->nr_nodes--; 2553fd0a558SYan, Zheng kfree(node); 2563fd0a558SYan, Zheng } 2573fd0a558SYan, Zheng } 2583fd0a558SYan, Zheng 2593fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 2603fd0a558SYan, Zheng { 2613fd0a558SYan, Zheng struct backref_edge *edge; 2623fd0a558SYan, Zheng 2633fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 2643fd0a558SYan, Zheng if (edge) 2653fd0a558SYan, Zheng cache->nr_edges++; 2663fd0a558SYan, Zheng return edge; 2673fd0a558SYan, Zheng } 2683fd0a558SYan, Zheng 2693fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 2703fd0a558SYan, Zheng struct backref_edge *edge) 2713fd0a558SYan, Zheng { 2723fd0a558SYan, Zheng if (edge) { 2733fd0a558SYan, Zheng cache->nr_edges--; 2743fd0a558SYan, Zheng kfree(edge); 2753fd0a558SYan, Zheng } 2765d4f98a2SYan Zheng } 2775d4f98a2SYan Zheng 2785d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 2795d4f98a2SYan Zheng struct rb_node *node) 2805d4f98a2SYan Zheng { 2815d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 2825d4f98a2SYan Zheng struct rb_node *parent = NULL; 2835d4f98a2SYan Zheng struct tree_entry *entry; 2845d4f98a2SYan Zheng 2855d4f98a2SYan Zheng while (*p) { 2865d4f98a2SYan Zheng parent = *p; 2875d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 2885d4f98a2SYan Zheng 2895d4f98a2SYan Zheng if (bytenr < entry->bytenr) 2905d4f98a2SYan Zheng p = &(*p)->rb_left; 2915d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 2925d4f98a2SYan Zheng p = &(*p)->rb_right; 2935d4f98a2SYan Zheng else 2945d4f98a2SYan Zheng return parent; 2955d4f98a2SYan Zheng } 2965d4f98a2SYan Zheng 2975d4f98a2SYan Zheng rb_link_node(node, parent, p); 2985d4f98a2SYan Zheng rb_insert_color(node, root); 2995d4f98a2SYan Zheng return NULL; 3005d4f98a2SYan Zheng } 3015d4f98a2SYan Zheng 3025d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3035d4f98a2SYan Zheng { 3045d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3055d4f98a2SYan Zheng struct tree_entry *entry; 3065d4f98a2SYan Zheng 3075d4f98a2SYan Zheng while (n) { 3085d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3095d4f98a2SYan Zheng 3105d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3115d4f98a2SYan Zheng n = n->rb_left; 3125d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3135d4f98a2SYan Zheng n = n->rb_right; 3145d4f98a2SYan Zheng else 3155d4f98a2SYan Zheng return n; 3165d4f98a2SYan Zheng } 3175d4f98a2SYan Zheng return NULL; 3185d4f98a2SYan Zheng } 3195d4f98a2SYan Zheng 32048a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 32143c04fb1SJeff Mahoney { 32243c04fb1SJeff Mahoney 32343c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 32443c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 32543c04fb1SJeff Mahoney rb_node); 32643c04fb1SJeff Mahoney if (bnode->root) 32743c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 3285d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 3295d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 3305d163e0eSJeff Mahoney bytenr); 33143c04fb1SJeff Mahoney } 33243c04fb1SJeff Mahoney 3335d4f98a2SYan Zheng /* 3345d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3355d4f98a2SYan Zheng */ 3365d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3375d4f98a2SYan Zheng struct backref_edge *edges[], 3385d4f98a2SYan Zheng int *index) 3395d4f98a2SYan Zheng { 3405d4f98a2SYan Zheng struct backref_edge *edge; 3415d4f98a2SYan Zheng int idx = *index; 3425d4f98a2SYan Zheng 3435d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3445d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3455d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3465d4f98a2SYan Zheng edges[idx++] = edge; 3475d4f98a2SYan Zheng node = edge->node[UPPER]; 3485d4f98a2SYan Zheng } 3493fd0a558SYan, Zheng BUG_ON(node->detached); 3505d4f98a2SYan Zheng *index = idx; 3515d4f98a2SYan Zheng return node; 3525d4f98a2SYan Zheng } 3535d4f98a2SYan Zheng 3545d4f98a2SYan Zheng /* 3555d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 3565d4f98a2SYan Zheng */ 3575d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 3585d4f98a2SYan Zheng int *index) 3595d4f98a2SYan Zheng { 3605d4f98a2SYan Zheng struct backref_edge *edge; 3615d4f98a2SYan Zheng struct backref_node *lower; 3625d4f98a2SYan Zheng int idx = *index; 3635d4f98a2SYan Zheng 3645d4f98a2SYan Zheng while (idx > 0) { 3655d4f98a2SYan Zheng edge = edges[idx - 1]; 3665d4f98a2SYan Zheng lower = edge->node[LOWER]; 3675d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 3685d4f98a2SYan Zheng idx--; 3695d4f98a2SYan Zheng continue; 3705d4f98a2SYan Zheng } 3715d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 3725d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3735d4f98a2SYan Zheng edges[idx - 1] = edge; 3745d4f98a2SYan Zheng *index = idx; 3755d4f98a2SYan Zheng return edge->node[UPPER]; 3765d4f98a2SYan Zheng } 3775d4f98a2SYan Zheng *index = 0; 3785d4f98a2SYan Zheng return NULL; 3795d4f98a2SYan Zheng } 3805d4f98a2SYan Zheng 3813fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 3825d4f98a2SYan Zheng { 3835d4f98a2SYan Zheng if (node->locked) { 3845d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 3855d4f98a2SYan Zheng node->locked = 0; 3865d4f98a2SYan Zheng } 3873fd0a558SYan, Zheng } 3883fd0a558SYan, Zheng 3893fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 3903fd0a558SYan, Zheng { 3913fd0a558SYan, Zheng if (node->eb) { 3923fd0a558SYan, Zheng unlock_node_buffer(node); 3935d4f98a2SYan Zheng free_extent_buffer(node->eb); 3945d4f98a2SYan Zheng node->eb = NULL; 3955d4f98a2SYan Zheng } 3965d4f98a2SYan Zheng } 3975d4f98a2SYan Zheng 3985d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 3995d4f98a2SYan Zheng struct backref_node *node) 4005d4f98a2SYan Zheng { 4015d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4025d4f98a2SYan Zheng 4035d4f98a2SYan Zheng drop_node_buffer(node); 4043fd0a558SYan, Zheng list_del(&node->list); 4055d4f98a2SYan Zheng list_del(&node->lower); 4063fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4075d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4083fd0a558SYan, Zheng free_backref_node(tree, node); 4095d4f98a2SYan Zheng } 4105d4f98a2SYan Zheng 4115d4f98a2SYan Zheng /* 4125d4f98a2SYan Zheng * remove a backref node from the backref cache 4135d4f98a2SYan Zheng */ 4145d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4155d4f98a2SYan Zheng struct backref_node *node) 4165d4f98a2SYan Zheng { 4175d4f98a2SYan Zheng struct backref_node *upper; 4185d4f98a2SYan Zheng struct backref_edge *edge; 4195d4f98a2SYan Zheng 4205d4f98a2SYan Zheng if (!node) 4215d4f98a2SYan Zheng return; 4225d4f98a2SYan Zheng 4233fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4245d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4255d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4265d4f98a2SYan Zheng list[LOWER]); 4275d4f98a2SYan Zheng upper = edge->node[UPPER]; 4285d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4295d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4303fd0a558SYan, Zheng free_backref_edge(cache, edge); 4313fd0a558SYan, Zheng 4323fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4333fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4343fd0a558SYan, Zheng drop_backref_node(cache, node); 4353fd0a558SYan, Zheng node = upper; 4363fd0a558SYan, Zheng node->lowest = 1; 4373fd0a558SYan, Zheng continue; 4383fd0a558SYan, Zheng } 4395d4f98a2SYan Zheng /* 4403fd0a558SYan, Zheng * add the node to leaf node list if no other 4415d4f98a2SYan Zheng * child block cached. 4425d4f98a2SYan Zheng */ 4435d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4443fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4455d4f98a2SYan Zheng upper->lowest = 1; 4465d4f98a2SYan Zheng } 4475d4f98a2SYan Zheng } 4483fd0a558SYan, Zheng 4495d4f98a2SYan Zheng drop_backref_node(cache, node); 4505d4f98a2SYan Zheng } 4515d4f98a2SYan Zheng 4523fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 4533fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 4543fd0a558SYan, Zheng { 4553fd0a558SYan, Zheng struct rb_node *rb_node; 4563fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 4573fd0a558SYan, Zheng node->bytenr = bytenr; 4583fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 45943c04fb1SJeff Mahoney if (rb_node) 46043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 4613fd0a558SYan, Zheng } 4623fd0a558SYan, Zheng 4633fd0a558SYan, Zheng /* 4643fd0a558SYan, Zheng * update backref cache after a transaction commit 4653fd0a558SYan, Zheng */ 4663fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 4673fd0a558SYan, Zheng struct backref_cache *cache) 4683fd0a558SYan, Zheng { 4693fd0a558SYan, Zheng struct backref_node *node; 4703fd0a558SYan, Zheng int level = 0; 4713fd0a558SYan, Zheng 4723fd0a558SYan, Zheng if (cache->last_trans == 0) { 4733fd0a558SYan, Zheng cache->last_trans = trans->transid; 4743fd0a558SYan, Zheng return 0; 4753fd0a558SYan, Zheng } 4763fd0a558SYan, Zheng 4773fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 4783fd0a558SYan, Zheng return 0; 4793fd0a558SYan, Zheng 4803fd0a558SYan, Zheng /* 4813fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 4823fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 4833fd0a558SYan, Zheng * so the detached nodes are no longer useful. 4843fd0a558SYan, Zheng */ 4853fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 4863fd0a558SYan, Zheng node = list_entry(cache->detached.next, 4873fd0a558SYan, Zheng struct backref_node, list); 4883fd0a558SYan, Zheng remove_backref_node(cache, node); 4893fd0a558SYan, Zheng } 4903fd0a558SYan, Zheng 4913fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 4923fd0a558SYan, Zheng node = list_entry(cache->changed.next, 4933fd0a558SYan, Zheng struct backref_node, list); 4943fd0a558SYan, Zheng list_del_init(&node->list); 4953fd0a558SYan, Zheng BUG_ON(node->pending); 4963fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 4973fd0a558SYan, Zheng } 4983fd0a558SYan, Zheng 4993fd0a558SYan, Zheng /* 5003fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5013fd0a558SYan, Zheng * errors during processing the pending nodes. 5023fd0a558SYan, Zheng */ 5033fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5043fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5053fd0a558SYan, Zheng BUG_ON(!node->pending); 5063fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5073fd0a558SYan, Zheng continue; 5083fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5093fd0a558SYan, Zheng } 5103fd0a558SYan, Zheng } 5113fd0a558SYan, Zheng 5123fd0a558SYan, Zheng cache->last_trans = 0; 5133fd0a558SYan, Zheng return 1; 5143fd0a558SYan, Zheng } 5153fd0a558SYan, Zheng 516f2a97a9dSDavid Sterba 5173fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5183fd0a558SYan, Zheng { 5193fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5203fd0a558SYan, Zheng 52127cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 5223fd0a558SYan, Zheng return 0; 5233fd0a558SYan, Zheng 5243fd0a558SYan, Zheng reloc_root = root->reloc_root; 5253fd0a558SYan, Zheng if (!reloc_root) 5263fd0a558SYan, Zheng return 0; 5273fd0a558SYan, Zheng 5283fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5293fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5303fd0a558SYan, Zheng return 0; 5313fd0a558SYan, Zheng /* 5323fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5333fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5343fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5353fd0a558SYan, Zheng * relocation. 5363fd0a558SYan, Zheng */ 5373fd0a558SYan, Zheng return 1; 5383fd0a558SYan, Zheng } 5395d4f98a2SYan Zheng /* 5405d4f98a2SYan Zheng * find reloc tree by address of tree root 5415d4f98a2SYan Zheng */ 5425d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5435d4f98a2SYan Zheng u64 bytenr) 5445d4f98a2SYan Zheng { 5455d4f98a2SYan Zheng struct rb_node *rb_node; 5465d4f98a2SYan Zheng struct mapping_node *node; 5475d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5485d4f98a2SYan Zheng 5495d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5505d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5515d4f98a2SYan Zheng if (rb_node) { 5525d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5535d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5545d4f98a2SYan Zheng } 5555d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5565d4f98a2SYan Zheng return root; 5575d4f98a2SYan Zheng } 5585d4f98a2SYan Zheng 5595d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5605d4f98a2SYan Zheng { 5615d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5625d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5635d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 5645d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 5655d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 56666463748SWang Shilong root_objectid == BTRFS_CSUM_TREE_OBJECTID || 56766463748SWang Shilong root_objectid == BTRFS_UUID_TREE_OBJECTID || 5683e4c5efbSDavid Sterba root_objectid == BTRFS_QUOTA_TREE_OBJECTID || 5693e4c5efbSDavid Sterba root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 5705d4f98a2SYan Zheng return 1; 5715d4f98a2SYan Zheng return 0; 5725d4f98a2SYan Zheng } 5735d4f98a2SYan Zheng 5745d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 5755d4f98a2SYan Zheng u64 root_objectid) 5765d4f98a2SYan Zheng { 5775d4f98a2SYan Zheng struct btrfs_key key; 5785d4f98a2SYan Zheng 5795d4f98a2SYan Zheng key.objectid = root_objectid; 5805d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 5815d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 5825d4f98a2SYan Zheng key.offset = 0; 5835d4f98a2SYan Zheng else 5845d4f98a2SYan Zheng key.offset = (u64)-1; 5855d4f98a2SYan Zheng 586c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 5875d4f98a2SYan Zheng } 5885d4f98a2SYan Zheng 5895d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 5905d4f98a2SYan Zheng static noinline_for_stack 5915d4f98a2SYan Zheng struct btrfs_root *find_tree_root(struct reloc_control *rc, 5925d4f98a2SYan Zheng struct extent_buffer *leaf, 5935d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0) 5945d4f98a2SYan Zheng { 5955d4f98a2SYan Zheng struct btrfs_root *root; 5965d4f98a2SYan Zheng u64 root_objectid = btrfs_ref_root_v0(leaf, ref0); 5975d4f98a2SYan Zheng u64 generation = btrfs_ref_generation_v0(leaf, ref0); 5985d4f98a2SYan Zheng 5995d4f98a2SYan Zheng BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID); 6005d4f98a2SYan Zheng 6015d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, root_objectid); 6025d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 6035d4f98a2SYan Zheng 60427cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && 6055d4f98a2SYan Zheng generation != btrfs_root_generation(&root->root_item)) 6065d4f98a2SYan Zheng return NULL; 6075d4f98a2SYan Zheng 6085d4f98a2SYan Zheng return root; 6095d4f98a2SYan Zheng } 6105d4f98a2SYan Zheng #endif 6115d4f98a2SYan Zheng 6125d4f98a2SYan Zheng static noinline_for_stack 6135d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 6145d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 6155d4f98a2SYan Zheng { 6163173a18fSJosef Bacik struct btrfs_key key; 6175d4f98a2SYan Zheng struct btrfs_extent_item *ei; 6185d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6195d4f98a2SYan Zheng u32 item_size; 6205d4f98a2SYan Zheng 6213173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6223173a18fSJosef Bacik 6235d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 6245d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 6255d4f98a2SYan Zheng if (item_size < sizeof(*ei)) { 6265d4f98a2SYan Zheng WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 6275d4f98a2SYan Zheng return 1; 6285d4f98a2SYan Zheng } 6295d4f98a2SYan Zheng #endif 6305d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6315d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6325d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6335d4f98a2SYan Zheng 6343173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6353173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6365d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6375d4f98a2SYan Zheng return 1; 6385d4f98a2SYan Zheng } 639d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 640d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 641d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 642d062d13cSJosef Bacik return 1; 643d062d13cSJosef Bacik } 6445d4f98a2SYan Zheng 6453173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6465d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6475d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6483173a18fSJosef Bacik } else { 6493173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6503173a18fSJosef Bacik } 6515d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6525d4f98a2SYan Zheng return 0; 6535d4f98a2SYan Zheng } 6545d4f98a2SYan Zheng 6555d4f98a2SYan Zheng /* 6565d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6575d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6585d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6595d4f98a2SYan Zheng * 6605d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 66101327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 66201327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 6635d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6645d4f98a2SYan Zheng * 6655d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6665d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6675d4f98a2SYan Zheng * block are also cached. 6685d4f98a2SYan Zheng */ 6693fd0a558SYan, Zheng static noinline_for_stack 6703fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6715d4f98a2SYan Zheng struct btrfs_key *node_key, 6725d4f98a2SYan Zheng int level, u64 bytenr) 6735d4f98a2SYan Zheng { 6743fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 6755d4f98a2SYan Zheng struct btrfs_path *path1; 6765d4f98a2SYan Zheng struct btrfs_path *path2; 6775d4f98a2SYan Zheng struct extent_buffer *eb; 6785d4f98a2SYan Zheng struct btrfs_root *root; 6795d4f98a2SYan Zheng struct backref_node *cur; 6805d4f98a2SYan Zheng struct backref_node *upper; 6815d4f98a2SYan Zheng struct backref_node *lower; 6825d4f98a2SYan Zheng struct backref_node *node = NULL; 6835d4f98a2SYan Zheng struct backref_node *exist = NULL; 6845d4f98a2SYan Zheng struct backref_edge *edge; 6855d4f98a2SYan Zheng struct rb_node *rb_node; 6865d4f98a2SYan Zheng struct btrfs_key key; 6875d4f98a2SYan Zheng unsigned long end; 6885d4f98a2SYan Zheng unsigned long ptr; 6895d4f98a2SYan Zheng LIST_HEAD(list); 6903fd0a558SYan, Zheng LIST_HEAD(useless); 6913fd0a558SYan, Zheng int cowonly; 6925d4f98a2SYan Zheng int ret; 6935d4f98a2SYan Zheng int err = 0; 694b6c60c80SJosef Bacik bool need_check = true; 6955d4f98a2SYan Zheng 6965d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 6975d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 6985d4f98a2SYan Zheng if (!path1 || !path2) { 6995d4f98a2SYan Zheng err = -ENOMEM; 7005d4f98a2SYan Zheng goto out; 7015d4f98a2SYan Zheng } 702e4058b54SDavid Sterba path1->reada = READA_FORWARD; 703e4058b54SDavid Sterba path2->reada = READA_FORWARD; 7045d4f98a2SYan Zheng 7053fd0a558SYan, Zheng node = alloc_backref_node(cache); 7065d4f98a2SYan Zheng if (!node) { 7075d4f98a2SYan Zheng err = -ENOMEM; 7085d4f98a2SYan Zheng goto out; 7095d4f98a2SYan Zheng } 7105d4f98a2SYan Zheng 7115d4f98a2SYan Zheng node->bytenr = bytenr; 7125d4f98a2SYan Zheng node->level = level; 7135d4f98a2SYan Zheng node->lowest = 1; 7145d4f98a2SYan Zheng cur = node; 7155d4f98a2SYan Zheng again: 7165d4f98a2SYan Zheng end = 0; 7175d4f98a2SYan Zheng ptr = 0; 7185d4f98a2SYan Zheng key.objectid = cur->bytenr; 7193173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7205d4f98a2SYan Zheng key.offset = (u64)-1; 7215d4f98a2SYan Zheng 7225d4f98a2SYan Zheng path1->search_commit_root = 1; 7235d4f98a2SYan Zheng path1->skip_locking = 1; 7245d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7255d4f98a2SYan Zheng 0, 0); 7265d4f98a2SYan Zheng if (ret < 0) { 7275d4f98a2SYan Zheng err = ret; 7285d4f98a2SYan Zheng goto out; 7295d4f98a2SYan Zheng } 73075bfb9afSJosef Bacik ASSERT(ret); 73175bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7325d4f98a2SYan Zheng 7335d4f98a2SYan Zheng path1->slots[0]--; 7345d4f98a2SYan Zheng 7355d4f98a2SYan Zheng WARN_ON(cur->checked); 7365d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7375d4f98a2SYan Zheng /* 73870f23fd6SJustin P. Mattock * the backref was added previously when processing 7395d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7405d4f98a2SYan Zheng */ 74175bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7425d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7435d4f98a2SYan Zheng list[LOWER]); 74475bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7455d4f98a2SYan Zheng exist = edge->node[UPPER]; 7465d4f98a2SYan Zheng /* 7475d4f98a2SYan Zheng * add the upper level block to pending list if we need 7485d4f98a2SYan Zheng * check its backrefs 7495d4f98a2SYan Zheng */ 7505d4f98a2SYan Zheng if (!exist->checked) 7515d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7525d4f98a2SYan Zheng } else { 7535d4f98a2SYan Zheng exist = NULL; 7545d4f98a2SYan Zheng } 7555d4f98a2SYan Zheng 7565d4f98a2SYan Zheng while (1) { 7575d4f98a2SYan Zheng cond_resched(); 7585d4f98a2SYan Zheng eb = path1->nodes[0]; 7595d4f98a2SYan Zheng 7605d4f98a2SYan Zheng if (ptr >= end) { 7615d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7625d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7635d4f98a2SYan Zheng if (ret < 0) { 7645d4f98a2SYan Zheng err = ret; 7655d4f98a2SYan Zheng goto out; 7665d4f98a2SYan Zheng } 7675d4f98a2SYan Zheng if (ret > 0) 7685d4f98a2SYan Zheng break; 7695d4f98a2SYan Zheng eb = path1->nodes[0]; 7705d4f98a2SYan Zheng } 7715d4f98a2SYan Zheng 7725d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7735d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7745d4f98a2SYan Zheng WARN_ON(exist); 7755d4f98a2SYan Zheng break; 7765d4f98a2SYan Zheng } 7775d4f98a2SYan Zheng 7783173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7793173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7805d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7815d4f98a2SYan Zheng &ptr, &end); 7825d4f98a2SYan Zheng if (ret) 7835d4f98a2SYan Zheng goto next; 7845d4f98a2SYan Zheng } 7855d4f98a2SYan Zheng } 7865d4f98a2SYan Zheng 7875d4f98a2SYan Zheng if (ptr < end) { 7885d4f98a2SYan Zheng /* update key for inline back ref */ 7895d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 7903de28d57SLiu Bo int type; 7915d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 7923de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 7933de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 7943de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 7953de28d57SLiu Bo err = -EINVAL; 7963de28d57SLiu Bo goto out; 7973de28d57SLiu Bo } 7983de28d57SLiu Bo key.type = type; 7995d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 8003de28d57SLiu Bo 8015d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 8025d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 8035d4f98a2SYan Zheng } 8045d4f98a2SYan Zheng 8055d4f98a2SYan Zheng if (exist && 8065d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8075d4f98a2SYan Zheng exist->owner == key.offset) || 8085d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8095d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8105d4f98a2SYan Zheng exist = NULL; 8115d4f98a2SYan Zheng goto next; 8125d4f98a2SYan Zheng } 8135d4f98a2SYan Zheng 8145d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 8155d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY || 8165d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 8173fd0a558SYan, Zheng if (key.type == BTRFS_EXTENT_REF_V0_KEY) { 8185d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 8195d4f98a2SYan Zheng ref0 = btrfs_item_ptr(eb, path1->slots[0], 8205d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 8213fd0a558SYan, Zheng if (key.objectid == key.offset) { 822046f264fSYan, Zheng root = find_tree_root(rc, eb, ref0); 8233fd0a558SYan, Zheng if (root && !should_ignore_root(root)) 8245d4f98a2SYan Zheng cur->root = root; 8255d4f98a2SYan Zheng else 8263fd0a558SYan, Zheng list_add(&cur->list, &useless); 8275d4f98a2SYan Zheng break; 8285d4f98a2SYan Zheng } 829046f264fSYan, Zheng if (is_cowonly_root(btrfs_ref_root_v0(eb, 830046f264fSYan, Zheng ref0))) 831046f264fSYan, Zheng cur->cowonly = 1; 8323fd0a558SYan, Zheng } 8335d4f98a2SYan Zheng #else 83475bfb9afSJosef Bacik ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY); 8355d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8365d4f98a2SYan Zheng #endif 8375d4f98a2SYan Zheng if (key.objectid == key.offset) { 8385d4f98a2SYan Zheng /* 8395d4f98a2SYan Zheng * only root blocks of reloc trees use 8405d4f98a2SYan Zheng * backref of this type. 8415d4f98a2SYan Zheng */ 8425d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 84375bfb9afSJosef Bacik ASSERT(root); 8445d4f98a2SYan Zheng cur->root = root; 8455d4f98a2SYan Zheng break; 8465d4f98a2SYan Zheng } 8475d4f98a2SYan Zheng 8483fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8495d4f98a2SYan Zheng if (!edge) { 8505d4f98a2SYan Zheng err = -ENOMEM; 8515d4f98a2SYan Zheng goto out; 8525d4f98a2SYan Zheng } 8535d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8545d4f98a2SYan Zheng if (!rb_node) { 8553fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8565d4f98a2SYan Zheng if (!upper) { 8573fd0a558SYan, Zheng free_backref_edge(cache, edge); 8585d4f98a2SYan Zheng err = -ENOMEM; 8595d4f98a2SYan Zheng goto out; 8605d4f98a2SYan Zheng } 8615d4f98a2SYan Zheng upper->bytenr = key.offset; 8625d4f98a2SYan Zheng upper->level = cur->level + 1; 8635d4f98a2SYan Zheng /* 8645d4f98a2SYan Zheng * backrefs for the upper level block isn't 8655d4f98a2SYan Zheng * cached, add the block to pending list 8665d4f98a2SYan Zheng */ 8675d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8685d4f98a2SYan Zheng } else { 8695d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8705d4f98a2SYan Zheng rb_node); 87175bfb9afSJosef Bacik ASSERT(upper->checked); 8725d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8735d4f98a2SYan Zheng } 8743fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8755d4f98a2SYan Zheng edge->node[LOWER] = cur; 8763fd0a558SYan, Zheng edge->node[UPPER] = upper; 8775d4f98a2SYan Zheng 8785d4f98a2SYan Zheng goto next; 8795d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8805d4f98a2SYan Zheng goto next; 8815d4f98a2SYan Zheng } 8825d4f98a2SYan Zheng 8835d4f98a2SYan Zheng /* key.type == BTRFS_TREE_BLOCK_REF_KEY */ 8845d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8855d4f98a2SYan Zheng if (IS_ERR(root)) { 8865d4f98a2SYan Zheng err = PTR_ERR(root); 8875d4f98a2SYan Zheng goto out; 8885d4f98a2SYan Zheng } 8895d4f98a2SYan Zheng 89027cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 8913fd0a558SYan, Zheng cur->cowonly = 1; 8923fd0a558SYan, Zheng 8935d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 8945d4f98a2SYan Zheng /* tree root */ 89575bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 8965d4f98a2SYan Zheng cur->bytenr); 8973fd0a558SYan, Zheng if (should_ignore_root(root)) 8983fd0a558SYan, Zheng list_add(&cur->list, &useless); 8993fd0a558SYan, Zheng else 9005d4f98a2SYan Zheng cur->root = root; 9015d4f98a2SYan Zheng break; 9025d4f98a2SYan Zheng } 9035d4f98a2SYan Zheng 9045d4f98a2SYan Zheng level = cur->level + 1; 9055d4f98a2SYan Zheng 9065d4f98a2SYan Zheng /* 9075d4f98a2SYan Zheng * searching the tree to find upper level blocks 9085d4f98a2SYan Zheng * reference the block. 9095d4f98a2SYan Zheng */ 9105d4f98a2SYan Zheng path2->search_commit_root = 1; 9115d4f98a2SYan Zheng path2->skip_locking = 1; 9125d4f98a2SYan Zheng path2->lowest_level = level; 9135d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 9145d4f98a2SYan Zheng path2->lowest_level = 0; 9155d4f98a2SYan Zheng if (ret < 0) { 9165d4f98a2SYan Zheng err = ret; 9175d4f98a2SYan Zheng goto out; 9185d4f98a2SYan Zheng } 91933c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 92033c66f43SYan Zheng path2->slots[level]--; 9215d4f98a2SYan Zheng 9225d4f98a2SYan Zheng eb = path2->nodes[level]; 9233561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 9243561b9dbSLiu Bo cur->bytenr) { 9253561b9dbSLiu Bo btrfs_err(root->fs_info, 9263561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 9273561b9dbSLiu Bo cur->bytenr, level - 1, root->objectid, 9283561b9dbSLiu Bo node_key->objectid, node_key->type, 9293561b9dbSLiu Bo node_key->offset); 9303561b9dbSLiu Bo err = -ENOENT; 9313561b9dbSLiu Bo goto out; 9323561b9dbSLiu Bo } 9335d4f98a2SYan Zheng lower = cur; 934b6c60c80SJosef Bacik need_check = true; 9355d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9365d4f98a2SYan Zheng if (!path2->nodes[level]) { 93775bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9385d4f98a2SYan Zheng lower->bytenr); 9393fd0a558SYan, Zheng if (should_ignore_root(root)) 9403fd0a558SYan, Zheng list_add(&lower->list, &useless); 9413fd0a558SYan, Zheng else 9425d4f98a2SYan Zheng lower->root = root; 9435d4f98a2SYan Zheng break; 9445d4f98a2SYan Zheng } 9455d4f98a2SYan Zheng 9463fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9475d4f98a2SYan Zheng if (!edge) { 9485d4f98a2SYan Zheng err = -ENOMEM; 9495d4f98a2SYan Zheng goto out; 9505d4f98a2SYan Zheng } 9515d4f98a2SYan Zheng 9525d4f98a2SYan Zheng eb = path2->nodes[level]; 9535d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9545d4f98a2SYan Zheng if (!rb_node) { 9553fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9565d4f98a2SYan Zheng if (!upper) { 9573fd0a558SYan, Zheng free_backref_edge(cache, edge); 9585d4f98a2SYan Zheng err = -ENOMEM; 9595d4f98a2SYan Zheng goto out; 9605d4f98a2SYan Zheng } 9615d4f98a2SYan Zheng upper->bytenr = eb->start; 9625d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9635d4f98a2SYan Zheng upper->level = lower->level + 1; 96427cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 96527cdeb70SMiao Xie &root->state)) 9663fd0a558SYan, Zheng upper->cowonly = 1; 9675d4f98a2SYan Zheng 9685d4f98a2SYan Zheng /* 9695d4f98a2SYan Zheng * if we know the block isn't shared 9705d4f98a2SYan Zheng * we can void checking its backrefs. 9715d4f98a2SYan Zheng */ 9725d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9735d4f98a2SYan Zheng upper->checked = 0; 9745d4f98a2SYan Zheng else 9755d4f98a2SYan Zheng upper->checked = 1; 9765d4f98a2SYan Zheng 9775d4f98a2SYan Zheng /* 9785d4f98a2SYan Zheng * add the block to pending list if we 979b6c60c80SJosef Bacik * need check its backrefs, we only do this once 980b6c60c80SJosef Bacik * while walking up a tree as we will catch 981b6c60c80SJosef Bacik * anything else later on. 9825d4f98a2SYan Zheng */ 983b6c60c80SJosef Bacik if (!upper->checked && need_check) { 984b6c60c80SJosef Bacik need_check = false; 9855d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9865d4f98a2SYan Zheng &list); 987bbe90514SJosef Bacik } else { 988bbe90514SJosef Bacik if (upper->checked) 989bbe90514SJosef Bacik need_check = true; 9905d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 991bbe90514SJosef Bacik } 9925d4f98a2SYan Zheng } else { 9935d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9945d4f98a2SYan Zheng rb_node); 99575bfb9afSJosef Bacik ASSERT(upper->checked); 9965d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9973fd0a558SYan, Zheng if (!upper->owner) 9983fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 9995d4f98a2SYan Zheng } 10005d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 10015d4f98a2SYan Zheng edge->node[LOWER] = lower; 10023fd0a558SYan, Zheng edge->node[UPPER] = upper; 10035d4f98a2SYan Zheng 10045d4f98a2SYan Zheng if (rb_node) 10055d4f98a2SYan Zheng break; 10065d4f98a2SYan Zheng lower = upper; 10075d4f98a2SYan Zheng upper = NULL; 10085d4f98a2SYan Zheng } 1009b3b4aa74SDavid Sterba btrfs_release_path(path2); 10105d4f98a2SYan Zheng next: 10115d4f98a2SYan Zheng if (ptr < end) { 10125d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 10135d4f98a2SYan Zheng if (ptr >= end) { 10145d4f98a2SYan Zheng WARN_ON(ptr > end); 10155d4f98a2SYan Zheng ptr = 0; 10165d4f98a2SYan Zheng end = 0; 10175d4f98a2SYan Zheng } 10185d4f98a2SYan Zheng } 10195d4f98a2SYan Zheng if (ptr >= end) 10205d4f98a2SYan Zheng path1->slots[0]++; 10215d4f98a2SYan Zheng } 1022b3b4aa74SDavid Sterba btrfs_release_path(path1); 10235d4f98a2SYan Zheng 10245d4f98a2SYan Zheng cur->checked = 1; 10255d4f98a2SYan Zheng WARN_ON(exist); 10265d4f98a2SYan Zheng 10275d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10285d4f98a2SYan Zheng if (!list_empty(&list)) { 10295d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10305d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10315d4f98a2SYan Zheng cur = edge->node[UPPER]; 10325d4f98a2SYan Zheng goto again; 10335d4f98a2SYan Zheng } 10345d4f98a2SYan Zheng 10355d4f98a2SYan Zheng /* 10365d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10375d4f98a2SYan Zheng * into the cache. 10385d4f98a2SYan Zheng */ 103975bfb9afSJosef Bacik ASSERT(node->checked); 10403fd0a558SYan, Zheng cowonly = node->cowonly; 10413fd0a558SYan, Zheng if (!cowonly) { 10423fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10433fd0a558SYan, Zheng &node->rb_node); 104443c04fb1SJeff Mahoney if (rb_node) 104543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10463fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10473fd0a558SYan, Zheng } 10485d4f98a2SYan Zheng 10495d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10505d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10515d4f98a2SYan Zheng 10525d4f98a2SYan Zheng while (!list_empty(&list)) { 10535d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10545d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10555d4f98a2SYan Zheng upper = edge->node[UPPER]; 10563fd0a558SYan, Zheng if (upper->detached) { 10573fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10583fd0a558SYan, Zheng lower = edge->node[LOWER]; 10593fd0a558SYan, Zheng free_backref_edge(cache, edge); 10603fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10613fd0a558SYan, Zheng list_add(&lower->list, &useless); 10623fd0a558SYan, Zheng continue; 10633fd0a558SYan, Zheng } 10645d4f98a2SYan Zheng 10655d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10665d4f98a2SYan Zheng if (upper->lowest) { 10675d4f98a2SYan Zheng list_del_init(&upper->lower); 10685d4f98a2SYan Zheng upper->lowest = 0; 10695d4f98a2SYan Zheng } 10705d4f98a2SYan Zheng 10715d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10725d4f98a2SYan Zheng continue; 10735d4f98a2SYan Zheng } 10745d4f98a2SYan Zheng 107575bfb9afSJosef Bacik if (!upper->checked) { 107675bfb9afSJosef Bacik /* 107775bfb9afSJosef Bacik * Still want to blow up for developers since this is a 107875bfb9afSJosef Bacik * logic bug. 107975bfb9afSJosef Bacik */ 108075bfb9afSJosef Bacik ASSERT(0); 108175bfb9afSJosef Bacik err = -EINVAL; 108275bfb9afSJosef Bacik goto out; 108375bfb9afSJosef Bacik } 108475bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 108575bfb9afSJosef Bacik ASSERT(0); 108675bfb9afSJosef Bacik err = -EINVAL; 108775bfb9afSJosef Bacik goto out; 108875bfb9afSJosef Bacik } 108975bfb9afSJosef Bacik 10903fd0a558SYan, Zheng if (!cowonly) { 10915d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10925d4f98a2SYan Zheng &upper->rb_node); 109343c04fb1SJeff Mahoney if (rb_node) 109443c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 109543c04fb1SJeff Mahoney upper->bytenr); 10963fd0a558SYan, Zheng } 10975d4f98a2SYan Zheng 10985d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10995d4f98a2SYan Zheng 11005d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 11015d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 11025d4f98a2SYan Zheng } 11033fd0a558SYan, Zheng /* 11043fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 11053fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 11063fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 11073fd0a558SYan, Zheng * lookup. 11083fd0a558SYan, Zheng */ 11093fd0a558SYan, Zheng while (!list_empty(&useless)) { 11103fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 11113fd0a558SYan, Zheng list_del_init(&upper->list); 111275bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 11133fd0a558SYan, Zheng if (upper == node) 11143fd0a558SYan, Zheng node = NULL; 11153fd0a558SYan, Zheng if (upper->lowest) { 11163fd0a558SYan, Zheng list_del_init(&upper->lower); 11173fd0a558SYan, Zheng upper->lowest = 0; 11183fd0a558SYan, Zheng } 11193fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 11203fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 11213fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 11223fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11233fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11243fd0a558SYan, Zheng lower = edge->node[LOWER]; 11253fd0a558SYan, Zheng free_backref_edge(cache, edge); 11263fd0a558SYan, Zheng 11273fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11283fd0a558SYan, Zheng list_add(&lower->list, &useless); 11293fd0a558SYan, Zheng } 11303fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11313fd0a558SYan, Zheng if (upper->level > 0) { 11323fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11333fd0a558SYan, Zheng upper->detached = 1; 11343fd0a558SYan, Zheng } else { 11353fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11363fd0a558SYan, Zheng free_backref_node(cache, upper); 11373fd0a558SYan, Zheng } 11383fd0a558SYan, Zheng } 11395d4f98a2SYan Zheng out: 11405d4f98a2SYan Zheng btrfs_free_path(path1); 11415d4f98a2SYan Zheng btrfs_free_path(path2); 11425d4f98a2SYan Zheng if (err) { 11433fd0a558SYan, Zheng while (!list_empty(&useless)) { 11443fd0a558SYan, Zheng lower = list_entry(useless.next, 114575bfb9afSJosef Bacik struct backref_node, list); 114675bfb9afSJosef Bacik list_del_init(&lower->list); 11473fd0a558SYan, Zheng } 114875bfb9afSJosef Bacik while (!list_empty(&list)) { 114975bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 115075bfb9afSJosef Bacik list[UPPER]); 115175bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11523fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 115375bfb9afSJosef Bacik lower = edge->node[LOWER]; 11545d4f98a2SYan Zheng upper = edge->node[UPPER]; 11553fd0a558SYan, Zheng free_backref_edge(cache, edge); 115675bfb9afSJosef Bacik 115775bfb9afSJosef Bacik /* 115875bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 115975bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 116075bfb9afSJosef Bacik */ 116175bfb9afSJosef Bacik if (list_empty(&lower->upper) && 116275bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 116375bfb9afSJosef Bacik list_add(&lower->list, &useless); 116475bfb9afSJosef Bacik 116575bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 116675bfb9afSJosef Bacik continue; 116775bfb9afSJosef Bacik 116801327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 116975bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 117075bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 117175bfb9afSJosef Bacik if (list_empty(&upper->upper)) 117275bfb9afSJosef Bacik list_add(&upper->list, &useless); 117375bfb9afSJosef Bacik } 117475bfb9afSJosef Bacik 117575bfb9afSJosef Bacik while (!list_empty(&useless)) { 117675bfb9afSJosef Bacik lower = list_entry(useless.next, 117775bfb9afSJosef Bacik struct backref_node, list); 117875bfb9afSJosef Bacik list_del_init(&lower->list); 11790fd8c3daSLiu Bo if (lower == node) 11800fd8c3daSLiu Bo node = NULL; 118175bfb9afSJosef Bacik free_backref_node(cache, lower); 11825d4f98a2SYan Zheng } 11830fd8c3daSLiu Bo 11840fd8c3daSLiu Bo free_backref_node(cache, node); 11855d4f98a2SYan Zheng return ERR_PTR(err); 11865d4f98a2SYan Zheng } 118775bfb9afSJosef Bacik ASSERT(!node || !node->detached); 11885d4f98a2SYan Zheng return node; 11895d4f98a2SYan Zheng } 11905d4f98a2SYan Zheng 11915d4f98a2SYan Zheng /* 11923fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11933fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11943fd0a558SYan, Zheng * corresponds to root of source tree 11953fd0a558SYan, Zheng */ 11963fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 11973fd0a558SYan, Zheng struct reloc_control *rc, 11983fd0a558SYan, Zheng struct btrfs_root *src, 11993fd0a558SYan, Zheng struct btrfs_root *dest) 12003fd0a558SYan, Zheng { 12013fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 12023fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 12033fd0a558SYan, Zheng struct backref_node *node = NULL; 12043fd0a558SYan, Zheng struct backref_node *new_node; 12053fd0a558SYan, Zheng struct backref_edge *edge; 12063fd0a558SYan, Zheng struct backref_edge *new_edge; 12073fd0a558SYan, Zheng struct rb_node *rb_node; 12083fd0a558SYan, Zheng 12093fd0a558SYan, Zheng if (cache->last_trans > 0) 12103fd0a558SYan, Zheng update_backref_cache(trans, cache); 12113fd0a558SYan, Zheng 12123fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 12133fd0a558SYan, Zheng if (rb_node) { 12143fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 12153fd0a558SYan, Zheng if (node->detached) 12163fd0a558SYan, Zheng node = NULL; 12173fd0a558SYan, Zheng else 12183fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 12193fd0a558SYan, Zheng } 12203fd0a558SYan, Zheng 12213fd0a558SYan, Zheng if (!node) { 12223fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 12233fd0a558SYan, Zheng reloc_root->commit_root->start); 12243fd0a558SYan, Zheng if (rb_node) { 12253fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 12263fd0a558SYan, Zheng rb_node); 12273fd0a558SYan, Zheng BUG_ON(node->detached); 12283fd0a558SYan, Zheng } 12293fd0a558SYan, Zheng } 12303fd0a558SYan, Zheng 12313fd0a558SYan, Zheng if (!node) 12323fd0a558SYan, Zheng return 0; 12333fd0a558SYan, Zheng 12343fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12353fd0a558SYan, Zheng if (!new_node) 12363fd0a558SYan, Zheng return -ENOMEM; 12373fd0a558SYan, Zheng 12383fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12393fd0a558SYan, Zheng new_node->level = node->level; 12403fd0a558SYan, Zheng new_node->lowest = node->lowest; 12416848ad64SYan, Zheng new_node->checked = 1; 12423fd0a558SYan, Zheng new_node->root = dest; 12433fd0a558SYan, Zheng 12443fd0a558SYan, Zheng if (!node->lowest) { 12453fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12463fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12473fd0a558SYan, Zheng if (!new_edge) 12483fd0a558SYan, Zheng goto fail; 12493fd0a558SYan, Zheng 12503fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12513fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12523fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12533fd0a558SYan, Zheng &new_node->lower); 12543fd0a558SYan, Zheng } 125576b9e23dSMiao Xie } else { 125676b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12573fd0a558SYan, Zheng } 12583fd0a558SYan, Zheng 12593fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12603fd0a558SYan, Zheng &new_node->rb_node); 126143c04fb1SJeff Mahoney if (rb_node) 126243c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12633fd0a558SYan, Zheng 12643fd0a558SYan, Zheng if (!new_node->lowest) { 12653fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12663fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12673fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12683fd0a558SYan, Zheng } 12693fd0a558SYan, Zheng } 12703fd0a558SYan, Zheng return 0; 12713fd0a558SYan, Zheng fail: 12723fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12733fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12743fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12753fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12763fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12773fd0a558SYan, Zheng } 12783fd0a558SYan, Zheng free_backref_node(cache, new_node); 12793fd0a558SYan, Zheng return -ENOMEM; 12803fd0a558SYan, Zheng } 12813fd0a558SYan, Zheng 12823fd0a558SYan, Zheng /* 12835d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12845d4f98a2SYan Zheng */ 1285ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12865d4f98a2SYan Zheng { 12870b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12885d4f98a2SYan Zheng struct rb_node *rb_node; 12895d4f98a2SYan Zheng struct mapping_node *node; 12900b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 12915d4f98a2SYan Zheng 12925d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1293ffd7b339SJeff Mahoney if (!node) 1294ffd7b339SJeff Mahoney return -ENOMEM; 12955d4f98a2SYan Zheng 12965d4f98a2SYan Zheng node->bytenr = root->node->start; 12975d4f98a2SYan Zheng node->data = root; 12985d4f98a2SYan Zheng 12995d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13005d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13015d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13025d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1303ffd7b339SJeff Mahoney if (rb_node) { 13040b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 13055d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 13065d163e0eSJeff Mahoney node->bytenr); 1307ffd7b339SJeff Mahoney } 13085d4f98a2SYan Zheng 13095d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 13105d4f98a2SYan Zheng return 0; 13115d4f98a2SYan Zheng } 13125d4f98a2SYan Zheng 13135d4f98a2SYan Zheng /* 1314c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 13155d4f98a2SYan Zheng * mapping 13165d4f98a2SYan Zheng */ 1317c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 13185d4f98a2SYan Zheng { 13190b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13205d4f98a2SYan Zheng struct rb_node *rb_node; 13215d4f98a2SYan Zheng struct mapping_node *node = NULL; 13220b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 13235d4f98a2SYan Zheng 13245d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13255d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1326c974c464SWang Shilong root->node->start); 1327c974c464SWang Shilong if (rb_node) { 1328c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1329c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1330c974c464SWang Shilong } 1331c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1332c974c464SWang Shilong 1333c974c464SWang Shilong if (!node) 1334c974c464SWang Shilong return; 1335c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1336c974c464SWang Shilong 13370b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1338c974c464SWang Shilong list_del_init(&root->root_list); 13390b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1340c974c464SWang Shilong kfree(node); 1341c974c464SWang Shilong } 1342c974c464SWang Shilong 1343c974c464SWang Shilong /* 1344c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1345c974c464SWang Shilong * mapping 1346c974c464SWang Shilong */ 1347c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1348c974c464SWang Shilong { 13490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1350c974c464SWang Shilong struct rb_node *rb_node; 1351c974c464SWang Shilong struct mapping_node *node = NULL; 13520b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1353c974c464SWang Shilong 1354c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1355c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1356c974c464SWang Shilong root->node->start); 13575d4f98a2SYan Zheng if (rb_node) { 13585d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13595d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13605d4f98a2SYan Zheng } 13615d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13625d4f98a2SYan Zheng 13638f71f3e0SLiu Bo if (!node) 13648f71f3e0SLiu Bo return 0; 13655d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13665d4f98a2SYan Zheng 13675d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1368c974c464SWang Shilong node->bytenr = new_bytenr; 13695d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13705d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13715d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 137243c04fb1SJeff Mahoney if (rb_node) 137343c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13745d4f98a2SYan Zheng return 0; 13755d4f98a2SYan Zheng } 13765d4f98a2SYan Zheng 13773fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13783fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13795d4f98a2SYan Zheng { 13800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13815d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13825d4f98a2SYan Zheng struct extent_buffer *eb; 13835d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13845d4f98a2SYan Zheng struct btrfs_key root_key; 13855d4f98a2SYan Zheng int ret; 13865d4f98a2SYan Zheng 13875d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13885d4f98a2SYan Zheng BUG_ON(!root_item); 13895d4f98a2SYan Zheng 13905d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13915d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13923fd0a558SYan, Zheng root_key.offset = objectid; 13935d4f98a2SYan Zheng 13943fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1395054570a1SFilipe Manana u64 commit_root_gen; 1396054570a1SFilipe Manana 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); 1401054570a1SFilipe Manana /* 1402054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1403054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1404054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1405054570a1SFilipe Manana * either inside the critical section of a transaction commit 1406054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1407054570a1SFilipe Manana * it's created before the transaction commit is started. 1408054570a1SFilipe Manana */ 1409054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1410054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 14113fd0a558SYan, Zheng } else { 14123fd0a558SYan, Zheng /* 14133fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 14143fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 14153fd0a558SYan, Zheng * modified after it was created have RELOC flag 14163fd0a558SYan, Zheng * set in their headers. so it's OK to not update 14173fd0a558SYan, Zheng * the 'last_snapshot'. 14183fd0a558SYan, Zheng */ 14193fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 14203fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 14213fd0a558SYan, Zheng BUG_ON(ret); 14223fd0a558SYan, Zheng } 14233fd0a558SYan, Zheng 14245d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 14255d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14265d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14275d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14283fd0a558SYan, Zheng 14293fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14303fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14313fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14323fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14335d4f98a2SYan Zheng root_item->drop_level = 0; 14343fd0a558SYan, Zheng } 14355d4f98a2SYan Zheng 14365d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14375d4f98a2SYan Zheng free_extent_buffer(eb); 14385d4f98a2SYan Zheng 14390b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 14405d4f98a2SYan Zheng &root_key, root_item); 14415d4f98a2SYan Zheng BUG_ON(ret); 14425d4f98a2SYan Zheng kfree(root_item); 14435d4f98a2SYan Zheng 14440b246afaSJeff Mahoney reloc_root = btrfs_read_fs_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 { 14570b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14583fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14590b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 146020dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14613fd0a558SYan, Zheng int clear_rsv = 0; 1462ffd7b339SJeff Mahoney int ret; 14633fd0a558SYan, Zheng 14643fd0a558SYan, Zheng if (root->reloc_root) { 14653fd0a558SYan, Zheng reloc_root = root->reloc_root; 14663fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14673fd0a558SYan, Zheng return 0; 14683fd0a558SYan, Zheng } 14693fd0a558SYan, Zheng 14703fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14713fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14723fd0a558SYan, Zheng return 0; 14733fd0a558SYan, Zheng 147420dd2cbfSMiao Xie if (!trans->reloc_reserved) { 147520dd2cbfSMiao Xie rsv = trans->block_rsv; 14763fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14773fd0a558SYan, Zheng clear_rsv = 1; 14783fd0a558SYan, Zheng } 14793fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14803fd0a558SYan, Zheng if (clear_rsv) 148120dd2cbfSMiao Xie trans->block_rsv = rsv; 14825d4f98a2SYan Zheng 1483ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1484ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14855d4f98a2SYan Zheng root->reloc_root = reloc_root; 14865d4f98a2SYan Zheng return 0; 14875d4f98a2SYan Zheng } 14885d4f98a2SYan Zheng 14895d4f98a2SYan Zheng /* 14905d4f98a2SYan Zheng * update root item of reloc tree 14915d4f98a2SYan Zheng */ 14925d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 14935d4f98a2SYan Zheng struct btrfs_root *root) 14945d4f98a2SYan Zheng { 14950b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14965d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14975d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14985d4f98a2SYan Zheng int ret; 14995d4f98a2SYan Zheng 15005d4f98a2SYan Zheng if (!root->reloc_root) 15017585717fSChris Mason goto out; 15025d4f98a2SYan Zheng 15035d4f98a2SYan Zheng reloc_root = root->reloc_root; 15045d4f98a2SYan Zheng root_item = &reloc_root->root_item; 15055d4f98a2SYan Zheng 15060b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 15073fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 15085d4f98a2SYan Zheng root->reloc_root = NULL; 1509c974c464SWang Shilong __del_reloc_root(reloc_root); 15105d4f98a2SYan Zheng } 15115d4f98a2SYan Zheng 15125d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 15135d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 15145d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 15155d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 15165d4f98a2SYan Zheng } 15175d4f98a2SYan Zheng 15180b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 15195d4f98a2SYan Zheng &reloc_root->root_key, root_item); 15205d4f98a2SYan Zheng BUG_ON(ret); 15217585717fSChris Mason 15227585717fSChris Mason out: 15235d4f98a2SYan Zheng return 0; 15245d4f98a2SYan Zheng } 15255d4f98a2SYan Zheng 15265d4f98a2SYan Zheng /* 15275d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15285d4f98a2SYan Zheng * in a subvolume 15295d4f98a2SYan Zheng */ 15305d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15315d4f98a2SYan Zheng { 15325d4f98a2SYan Zheng struct rb_node *node; 15335d4f98a2SYan Zheng struct rb_node *prev; 15345d4f98a2SYan Zheng struct btrfs_inode *entry; 15355d4f98a2SYan Zheng struct inode *inode; 15365d4f98a2SYan Zheng 15375d4f98a2SYan Zheng spin_lock(&root->inode_lock); 15385d4f98a2SYan Zheng again: 15395d4f98a2SYan Zheng node = root->inode_tree.rb_node; 15405d4f98a2SYan Zheng prev = NULL; 15415d4f98a2SYan Zheng while (node) { 15425d4f98a2SYan Zheng prev = node; 15435d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15445d4f98a2SYan Zheng 15454a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 15465d4f98a2SYan Zheng node = node->rb_left; 15474a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 15485d4f98a2SYan Zheng node = node->rb_right; 15495d4f98a2SYan Zheng else 15505d4f98a2SYan Zheng break; 15515d4f98a2SYan Zheng } 15525d4f98a2SYan Zheng if (!node) { 15535d4f98a2SYan Zheng while (prev) { 15545d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 15554a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 15565d4f98a2SYan Zheng node = prev; 15575d4f98a2SYan Zheng break; 15585d4f98a2SYan Zheng } 15595d4f98a2SYan Zheng prev = rb_next(prev); 15605d4f98a2SYan Zheng } 15615d4f98a2SYan Zheng } 15625d4f98a2SYan Zheng while (node) { 15635d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15645d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15655d4f98a2SYan Zheng if (inode) { 15665d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15675d4f98a2SYan Zheng return inode; 15685d4f98a2SYan Zheng } 15695d4f98a2SYan Zheng 15704a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 15715d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15725d4f98a2SYan Zheng goto again; 15735d4f98a2SYan Zheng 15745d4f98a2SYan Zheng node = rb_next(node); 15755d4f98a2SYan Zheng } 15765d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15775d4f98a2SYan Zheng return NULL; 15785d4f98a2SYan Zheng } 15795d4f98a2SYan Zheng 15805d4f98a2SYan Zheng static int in_block_group(u64 bytenr, 15815d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group) 15825d4f98a2SYan Zheng { 15835d4f98a2SYan Zheng if (bytenr >= block_group->key.objectid && 15845d4f98a2SYan Zheng bytenr < block_group->key.objectid + block_group->key.offset) 15855d4f98a2SYan Zheng return 1; 15865d4f98a2SYan Zheng return 0; 15875d4f98a2SYan Zheng } 15885d4f98a2SYan Zheng 15895d4f98a2SYan Zheng /* 15905d4f98a2SYan Zheng * get new location of data 15915d4f98a2SYan Zheng */ 15925d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 15935d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 15945d4f98a2SYan Zheng { 15955d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 15965d4f98a2SYan Zheng struct btrfs_path *path; 15975d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15985d4f98a2SYan Zheng struct extent_buffer *leaf; 15995d4f98a2SYan Zheng int ret; 16005d4f98a2SYan Zheng 16015d4f98a2SYan Zheng path = btrfs_alloc_path(); 16025d4f98a2SYan Zheng if (!path) 16035d4f98a2SYan Zheng return -ENOMEM; 16045d4f98a2SYan Zheng 16055d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1606f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1607f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 16085d4f98a2SYan Zheng if (ret < 0) 16095d4f98a2SYan Zheng goto out; 16105d4f98a2SYan Zheng if (ret > 0) { 16115d4f98a2SYan Zheng ret = -ENOENT; 16125d4f98a2SYan Zheng goto out; 16135d4f98a2SYan Zheng } 16145d4f98a2SYan Zheng 16155d4f98a2SYan Zheng leaf = path->nodes[0]; 16165d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 16175d4f98a2SYan Zheng struct btrfs_file_extent_item); 16185d4f98a2SYan Zheng 16195d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 16205d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 16215d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 16225d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 16235d4f98a2SYan Zheng 16245d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 162583d4cfd4SJosef Bacik ret = -EINVAL; 16265d4f98a2SYan Zheng goto out; 16275d4f98a2SYan Zheng } 16285d4f98a2SYan Zheng 16295d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16305d4f98a2SYan Zheng ret = 0; 16315d4f98a2SYan Zheng out: 16325d4f98a2SYan Zheng btrfs_free_path(path); 16335d4f98a2SYan Zheng return ret; 16345d4f98a2SYan Zheng } 16355d4f98a2SYan Zheng 16365d4f98a2SYan Zheng /* 16375d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16385d4f98a2SYan Zheng * the new locations. 16395d4f98a2SYan Zheng */ 16403fd0a558SYan, Zheng static noinline_for_stack 16413fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 16425d4f98a2SYan Zheng struct reloc_control *rc, 16435d4f98a2SYan Zheng struct btrfs_root *root, 16443fd0a558SYan, Zheng struct extent_buffer *leaf) 16455d4f98a2SYan Zheng { 16460b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 16475d4f98a2SYan Zheng struct btrfs_key key; 16485d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16495d4f98a2SYan Zheng struct inode *inode = NULL; 16505d4f98a2SYan Zheng u64 parent; 16515d4f98a2SYan Zheng u64 bytenr; 16523fd0a558SYan, Zheng u64 new_bytenr = 0; 16535d4f98a2SYan Zheng u64 num_bytes; 16545d4f98a2SYan Zheng u64 end; 16555d4f98a2SYan Zheng u32 nritems; 16565d4f98a2SYan Zheng u32 i; 165783d4cfd4SJosef Bacik int ret = 0; 16585d4f98a2SYan Zheng int first = 1; 16595d4f98a2SYan Zheng int dirty = 0; 16605d4f98a2SYan Zheng 16615d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16625d4f98a2SYan Zheng return 0; 16635d4f98a2SYan Zheng 16645d4f98a2SYan Zheng /* reloc trees always use full backref */ 16655d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16665d4f98a2SYan Zheng parent = leaf->start; 16675d4f98a2SYan Zheng else 16685d4f98a2SYan Zheng parent = 0; 16695d4f98a2SYan Zheng 16705d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16715d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 16725d4f98a2SYan Zheng cond_resched(); 16735d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16745d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16755d4f98a2SYan Zheng continue; 16765d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16775d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16785d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16795d4f98a2SYan Zheng continue; 16805d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16815d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16825d4f98a2SYan Zheng if (bytenr == 0) 16835d4f98a2SYan Zheng continue; 16845d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16855d4f98a2SYan Zheng continue; 16865d4f98a2SYan Zheng 16875d4f98a2SYan Zheng /* 16885d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16895d4f98a2SYan Zheng * to complete and drop the extent cache 16905d4f98a2SYan Zheng */ 16915d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16925d4f98a2SYan Zheng if (first) { 16935d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16945d4f98a2SYan Zheng first = 0; 16954a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 16963fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16975d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16985d4f98a2SYan Zheng } 16994a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 17005d4f98a2SYan Zheng end = key.offset + 17015d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 17025d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 17030b246afaSJeff Mahoney fs_info->sectorsize)); 17040b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 17055d4f98a2SYan Zheng end--; 17065d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1707d0082371SJeff Mahoney key.offset, end); 17085d4f98a2SYan Zheng if (!ret) 17095d4f98a2SYan Zheng continue; 17105d4f98a2SYan Zheng 1711dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1712dcdbc059SNikolay Borisov key.offset, end, 1); 17135d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1714d0082371SJeff Mahoney key.offset, end); 17155d4f98a2SYan Zheng } 17165d4f98a2SYan Zheng } 17175d4f98a2SYan Zheng 17185d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 17195d4f98a2SYan Zheng bytenr, num_bytes); 172083d4cfd4SJosef Bacik if (ret) { 172183d4cfd4SJosef Bacik /* 172283d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 172383d4cfd4SJosef Bacik * in the file extent yet. 172483d4cfd4SJosef Bacik */ 172583d4cfd4SJosef Bacik break; 17263fd0a558SYan, Zheng } 17275d4f98a2SYan Zheng 17285d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17295d4f98a2SYan Zheng dirty = 1; 17305d4f98a2SYan Zheng 17315d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 173284f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, root, new_bytenr, 17335d4f98a2SYan Zheng num_bytes, parent, 17345d4f98a2SYan Zheng btrfs_header_owner(leaf), 1735b06c4bf5SFilipe Manana key.objectid, key.offset); 173683d4cfd4SJosef Bacik if (ret) { 173766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 173883d4cfd4SJosef Bacik break; 173983d4cfd4SJosef Bacik } 17405d4f98a2SYan Zheng 174184f7d8e6SJosef Bacik ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 17425d4f98a2SYan Zheng parent, btrfs_header_owner(leaf), 1743b06c4bf5SFilipe Manana key.objectid, key.offset); 174483d4cfd4SJosef Bacik if (ret) { 174566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 174683d4cfd4SJosef Bacik break; 174783d4cfd4SJosef Bacik } 17485d4f98a2SYan Zheng } 17495d4f98a2SYan Zheng if (dirty) 17505d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17513fd0a558SYan, Zheng if (inode) 17523fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 175383d4cfd4SJosef Bacik return ret; 17545d4f98a2SYan Zheng } 17555d4f98a2SYan Zheng 17565d4f98a2SYan Zheng static noinline_for_stack 17575d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17585d4f98a2SYan Zheng struct btrfs_path *path, int level) 17595d4f98a2SYan Zheng { 17605d4f98a2SYan Zheng struct btrfs_disk_key key1; 17615d4f98a2SYan Zheng struct btrfs_disk_key key2; 17625d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17635d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17645d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17655d4f98a2SYan Zheng } 17665d4f98a2SYan Zheng 17675d4f98a2SYan Zheng /* 17685d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17695d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17705d4f98a2SYan Zheng * reloc tree was create can be replaced. 17715d4f98a2SYan Zheng * 17725d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17735d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17745d4f98a2SYan Zheng * errors, a negative error number is returned. 17755d4f98a2SYan Zheng */ 17763fd0a558SYan, Zheng static noinline_for_stack 17773fd0a558SYan, Zheng int replace_path(struct btrfs_trans_handle *trans, 17785d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17795d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17805d4f98a2SYan Zheng int lowest_level, int max_level) 17815d4f98a2SYan Zheng { 17820b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 17835d4f98a2SYan Zheng struct extent_buffer *eb; 17845d4f98a2SYan Zheng struct extent_buffer *parent; 17855d4f98a2SYan Zheng struct btrfs_key key; 17865d4f98a2SYan Zheng u64 old_bytenr; 17875d4f98a2SYan Zheng u64 new_bytenr; 17885d4f98a2SYan Zheng u64 old_ptr_gen; 17895d4f98a2SYan Zheng u64 new_ptr_gen; 17905d4f98a2SYan Zheng u64 last_snapshot; 17915d4f98a2SYan Zheng u32 blocksize; 17923fd0a558SYan, Zheng int cow = 0; 17935d4f98a2SYan Zheng int level; 17945d4f98a2SYan Zheng int ret; 17955d4f98a2SYan Zheng int slot; 17965d4f98a2SYan Zheng 17975d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17985d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17995d4f98a2SYan Zheng 18005d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 18013fd0a558SYan, Zheng again: 18025d4f98a2SYan Zheng slot = path->slots[lowest_level]; 18035d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 18045d4f98a2SYan Zheng 18055d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 18065d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 18075d4f98a2SYan Zheng level = btrfs_header_level(eb); 18085d4f98a2SYan Zheng 18095d4f98a2SYan Zheng if (level < lowest_level) { 18105d4f98a2SYan Zheng btrfs_tree_unlock(eb); 18115d4f98a2SYan Zheng free_extent_buffer(eb); 18125d4f98a2SYan Zheng return 0; 18135d4f98a2SYan Zheng } 18145d4f98a2SYan Zheng 18153fd0a558SYan, Zheng if (cow) { 18165d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18175d4f98a2SYan Zheng BUG_ON(ret); 18183fd0a558SYan, Zheng } 18195d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 18205d4f98a2SYan Zheng 18215d4f98a2SYan Zheng if (next_key) { 18225d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18235d4f98a2SYan Zheng next_key->type = (u8)-1; 18245d4f98a2SYan Zheng next_key->offset = (u64)-1; 18255d4f98a2SYan Zheng } 18265d4f98a2SYan Zheng 18275d4f98a2SYan Zheng parent = eb; 18285d4f98a2SYan Zheng while (1) { 1829581c1760SQu Wenruo struct btrfs_key first_key; 1830581c1760SQu Wenruo 18315d4f98a2SYan Zheng level = btrfs_header_level(parent); 18325d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18335d4f98a2SYan Zheng 18345d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 18355d4f98a2SYan Zheng if (ret && slot > 0) 18365d4f98a2SYan Zheng slot--; 18375d4f98a2SYan Zheng 18385d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18395d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18405d4f98a2SYan Zheng 18415d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 18420b246afaSJeff Mahoney blocksize = fs_info->nodesize; 18435d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 184417515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 18455d4f98a2SYan Zheng 18465d4f98a2SYan Zheng if (level <= max_level) { 18475d4f98a2SYan Zheng eb = path->nodes[level]; 18485d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18495d4f98a2SYan Zheng path->slots[level]); 18505d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18515d4f98a2SYan Zheng path->slots[level]); 18525d4f98a2SYan Zheng } else { 18535d4f98a2SYan Zheng new_bytenr = 0; 18545d4f98a2SYan Zheng new_ptr_gen = 0; 18555d4f98a2SYan Zheng } 18565d4f98a2SYan Zheng 1857fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18585d4f98a2SYan Zheng ret = level; 18595d4f98a2SYan Zheng break; 18605d4f98a2SYan Zheng } 18615d4f98a2SYan Zheng 18625d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18635d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18643fd0a558SYan, Zheng if (level <= lowest_level) { 18655d4f98a2SYan Zheng ret = 0; 18665d4f98a2SYan Zheng break; 18675d4f98a2SYan Zheng } 18685d4f98a2SYan Zheng 1869581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 1870581c1760SQu Wenruo level - 1, &first_key); 187164c043deSLiu Bo if (IS_ERR(eb)) { 187264c043deSLiu Bo ret = PTR_ERR(eb); 1873264813acSLiu Bo break; 187464c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 187564c043deSLiu Bo ret = -EIO; 1876416bc658SJosef Bacik free_extent_buffer(eb); 1877379cde74SStefan Behrens break; 1878416bc658SJosef Bacik } 18795d4f98a2SYan Zheng btrfs_tree_lock(eb); 18803fd0a558SYan, Zheng if (cow) { 18815d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18825d4f98a2SYan Zheng slot, &eb); 18835d4f98a2SYan Zheng BUG_ON(ret); 18845d4f98a2SYan Zheng } 18853fd0a558SYan, Zheng btrfs_set_lock_blocking(eb); 18865d4f98a2SYan Zheng 18875d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18885d4f98a2SYan Zheng free_extent_buffer(parent); 18895d4f98a2SYan Zheng 18905d4f98a2SYan Zheng parent = eb; 18915d4f98a2SYan Zheng continue; 18925d4f98a2SYan Zheng } 18935d4f98a2SYan Zheng 18943fd0a558SYan, Zheng if (!cow) { 18953fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18963fd0a558SYan, Zheng free_extent_buffer(parent); 18973fd0a558SYan, Zheng cow = 1; 18983fd0a558SYan, Zheng goto again; 18993fd0a558SYan, Zheng } 19003fd0a558SYan, Zheng 19015d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 19025d4f98a2SYan Zheng path->slots[level]); 1903b3b4aa74SDavid Sterba btrfs_release_path(path); 19045d4f98a2SYan Zheng 19055d4f98a2SYan Zheng path->lowest_level = level; 19065d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 19075d4f98a2SYan Zheng path->lowest_level = 0; 19085d4f98a2SYan Zheng BUG_ON(ret); 19095d4f98a2SYan Zheng 19105d4f98a2SYan Zheng /* 1911824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1912824d8dffSQu Wenruo * 1913824d8dffSQu Wenruo * We must trace both trees. 1914824d8dffSQu Wenruo * 1) Tree reloc subtree 1915824d8dffSQu Wenruo * If not traced, we will leak data numbers 1916824d8dffSQu Wenruo * 2) Fs subtree 1917824d8dffSQu Wenruo * If not traced, we will double count old data 1918824d8dffSQu Wenruo * and tree block numbers, if current trans doesn't free 1919824d8dffSQu Wenruo * data reloc tree inode. 1920824d8dffSQu Wenruo */ 1921824d8dffSQu Wenruo ret = btrfs_qgroup_trace_subtree(trans, src, parent, 1922824d8dffSQu Wenruo btrfs_header_generation(parent), 1923824d8dffSQu Wenruo btrfs_header_level(parent)); 1924824d8dffSQu Wenruo if (ret < 0) 1925824d8dffSQu Wenruo break; 1926824d8dffSQu Wenruo ret = btrfs_qgroup_trace_subtree(trans, dest, 1927824d8dffSQu Wenruo path->nodes[level], 1928824d8dffSQu Wenruo btrfs_header_generation(path->nodes[level]), 1929824d8dffSQu Wenruo btrfs_header_level(path->nodes[level])); 1930824d8dffSQu Wenruo if (ret < 0) 1931824d8dffSQu Wenruo break; 1932824d8dffSQu Wenruo 1933824d8dffSQu Wenruo /* 19345d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 19355d4f98a2SYan Zheng */ 19365d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 19375d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19385d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19395d4f98a2SYan Zheng 19405d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19415d4f98a2SYan Zheng path->slots[level], old_bytenr); 19425d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19435d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19445d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19455d4f98a2SYan Zheng 194684f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, src, old_bytenr, 19472ff7e61eSJeff Mahoney blocksize, path->nodes[level]->start, 1948b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19495d4f98a2SYan Zheng BUG_ON(ret); 195084f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, 19512ff7e61eSJeff Mahoney blocksize, 0, dest->root_key.objectid, 19522ff7e61eSJeff Mahoney level - 1, 0); 19535d4f98a2SYan Zheng BUG_ON(ret); 19545d4f98a2SYan Zheng 195584f7d8e6SJosef Bacik ret = btrfs_free_extent(trans, src, new_bytenr, blocksize, 19565d4f98a2SYan Zheng path->nodes[level]->start, 1957b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19585d4f98a2SYan Zheng BUG_ON(ret); 19595d4f98a2SYan Zheng 196084f7d8e6SJosef Bacik ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize, 19615d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 1962b06c4bf5SFilipe Manana 0); 19635d4f98a2SYan Zheng BUG_ON(ret); 19645d4f98a2SYan Zheng 19655d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19665d4f98a2SYan Zheng 19675d4f98a2SYan Zheng ret = level; 19685d4f98a2SYan Zheng break; 19695d4f98a2SYan Zheng } 19705d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19715d4f98a2SYan Zheng free_extent_buffer(parent); 19725d4f98a2SYan Zheng return ret; 19735d4f98a2SYan Zheng } 19745d4f98a2SYan Zheng 19755d4f98a2SYan Zheng /* 19765d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19775d4f98a2SYan Zheng */ 19785d4f98a2SYan Zheng static noinline_for_stack 19795d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19805d4f98a2SYan Zheng int *level) 19815d4f98a2SYan Zheng { 19825d4f98a2SYan Zheng struct extent_buffer *eb; 19835d4f98a2SYan Zheng int i; 19845d4f98a2SYan Zheng u64 last_snapshot; 19855d4f98a2SYan Zheng u32 nritems; 19865d4f98a2SYan Zheng 19875d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19885d4f98a2SYan Zheng 19895d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19905d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19915d4f98a2SYan Zheng path->nodes[i] = NULL; 19925d4f98a2SYan Zheng } 19935d4f98a2SYan Zheng 19945d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 19955d4f98a2SYan Zheng eb = path->nodes[i]; 19965d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19975d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 19985d4f98a2SYan Zheng path->slots[i]++; 19995d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 20005d4f98a2SYan Zheng last_snapshot) 20015d4f98a2SYan Zheng continue; 20025d4f98a2SYan Zheng 20035d4f98a2SYan Zheng *level = i; 20045d4f98a2SYan Zheng return 0; 20055d4f98a2SYan Zheng } 20065d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20075d4f98a2SYan Zheng path->nodes[i] = NULL; 20085d4f98a2SYan Zheng } 20095d4f98a2SYan Zheng return 1; 20105d4f98a2SYan Zheng } 20115d4f98a2SYan Zheng 20125d4f98a2SYan Zheng /* 20135d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 20145d4f98a2SYan Zheng */ 20155d4f98a2SYan Zheng static noinline_for_stack 20165d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20175d4f98a2SYan Zheng int *level) 20185d4f98a2SYan Zheng { 20192ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20205d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 20215d4f98a2SYan Zheng int i; 20225d4f98a2SYan Zheng u64 bytenr; 20235d4f98a2SYan Zheng u64 ptr_gen = 0; 20245d4f98a2SYan Zheng u64 last_snapshot; 20255d4f98a2SYan Zheng u32 nritems; 20265d4f98a2SYan Zheng 20275d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20285d4f98a2SYan Zheng 20295d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2030581c1760SQu Wenruo struct btrfs_key first_key; 2031581c1760SQu Wenruo 20325d4f98a2SYan Zheng eb = path->nodes[i]; 20335d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20345d4f98a2SYan Zheng while (path->slots[i] < nritems) { 20355d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 20365d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 20375d4f98a2SYan Zheng break; 20385d4f98a2SYan Zheng path->slots[i]++; 20395d4f98a2SYan Zheng } 20405d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20415d4f98a2SYan Zheng if (i == *level) 20425d4f98a2SYan Zheng break; 20435d4f98a2SYan Zheng *level = i + 1; 20445d4f98a2SYan Zheng return 0; 20455d4f98a2SYan Zheng } 20465d4f98a2SYan Zheng if (i == 1) { 20475d4f98a2SYan Zheng *level = i; 20485d4f98a2SYan Zheng return 0; 20495d4f98a2SYan Zheng } 20505d4f98a2SYan Zheng 20515d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2052581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2053581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2054581c1760SQu Wenruo &first_key); 205564c043deSLiu Bo if (IS_ERR(eb)) { 205664c043deSLiu Bo return PTR_ERR(eb); 205764c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2058416bc658SJosef Bacik free_extent_buffer(eb); 2059416bc658SJosef Bacik return -EIO; 2060416bc658SJosef Bacik } 20615d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20625d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20635d4f98a2SYan Zheng path->slots[i - 1] = 0; 20645d4f98a2SYan Zheng } 20655d4f98a2SYan Zheng return 1; 20665d4f98a2SYan Zheng } 20675d4f98a2SYan Zheng 20685d4f98a2SYan Zheng /* 20695d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 20705d4f98a2SYan Zheng * [min_key, max_key) 20715d4f98a2SYan Zheng */ 20725d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 20735d4f98a2SYan Zheng struct btrfs_key *min_key, 20745d4f98a2SYan Zheng struct btrfs_key *max_key) 20755d4f98a2SYan Zheng { 20760b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20775d4f98a2SYan Zheng struct inode *inode = NULL; 20785d4f98a2SYan Zheng u64 objectid; 20795d4f98a2SYan Zheng u64 start, end; 208033345d01SLi Zefan u64 ino; 20815d4f98a2SYan Zheng 20825d4f98a2SYan Zheng objectid = min_key->objectid; 20835d4f98a2SYan Zheng while (1) { 20845d4f98a2SYan Zheng cond_resched(); 20855d4f98a2SYan Zheng iput(inode); 20865d4f98a2SYan Zheng 20875d4f98a2SYan Zheng if (objectid > max_key->objectid) 20885d4f98a2SYan Zheng break; 20895d4f98a2SYan Zheng 20905d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20915d4f98a2SYan Zheng if (!inode) 20925d4f98a2SYan Zheng break; 20934a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 20945d4f98a2SYan Zheng 209533345d01SLi Zefan if (ino > max_key->objectid) { 20965d4f98a2SYan Zheng iput(inode); 20975d4f98a2SYan Zheng break; 20985d4f98a2SYan Zheng } 20995d4f98a2SYan Zheng 210033345d01SLi Zefan objectid = ino + 1; 21015d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 21025d4f98a2SYan Zheng continue; 21035d4f98a2SYan Zheng 210433345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 21055d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 21065d4f98a2SYan Zheng continue; 21075d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 21085d4f98a2SYan Zheng start = 0; 21095d4f98a2SYan Zheng else { 21105d4f98a2SYan Zheng start = min_key->offset; 21110b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 21125d4f98a2SYan Zheng } 21135d4f98a2SYan Zheng } else { 21145d4f98a2SYan Zheng start = 0; 21155d4f98a2SYan Zheng } 21165d4f98a2SYan Zheng 211733345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 21185d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 21195d4f98a2SYan Zheng continue; 21205d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 21215d4f98a2SYan Zheng end = (u64)-1; 21225d4f98a2SYan Zheng } else { 21235d4f98a2SYan Zheng if (max_key->offset == 0) 21245d4f98a2SYan Zheng continue; 21255d4f98a2SYan Zheng end = max_key->offset; 21260b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 21275d4f98a2SYan Zheng end--; 21285d4f98a2SYan Zheng } 21295d4f98a2SYan Zheng } else { 21305d4f98a2SYan Zheng end = (u64)-1; 21315d4f98a2SYan Zheng } 21325d4f98a2SYan Zheng 21335d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2134d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2135dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2136d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 21375d4f98a2SYan Zheng } 21385d4f98a2SYan Zheng return 0; 21395d4f98a2SYan Zheng } 21405d4f98a2SYan Zheng 21415d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 21425d4f98a2SYan Zheng struct btrfs_key *key) 21435d4f98a2SYan Zheng 21445d4f98a2SYan Zheng { 21455d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21465d4f98a2SYan Zheng if (!path->nodes[level]) 21475d4f98a2SYan Zheng break; 21485d4f98a2SYan Zheng if (path->slots[level] + 1 < 21495d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21505d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21515d4f98a2SYan Zheng path->slots[level] + 1); 21525d4f98a2SYan Zheng return 0; 21535d4f98a2SYan Zheng } 21545d4f98a2SYan Zheng level++; 21555d4f98a2SYan Zheng } 21565d4f98a2SYan Zheng return 1; 21575d4f98a2SYan Zheng } 21585d4f98a2SYan Zheng 21595d4f98a2SYan Zheng /* 21605d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 21615d4f98a2SYan Zheng * fs tree. 21625d4f98a2SYan Zheng */ 21635d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 21645d4f98a2SYan Zheng struct btrfs_root *root) 21655d4f98a2SYan Zheng { 21660b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 21675d4f98a2SYan Zheng LIST_HEAD(inode_list); 21685d4f98a2SYan Zheng struct btrfs_key key; 21695d4f98a2SYan Zheng struct btrfs_key next_key; 21709e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 21715d4f98a2SYan Zheng struct btrfs_root *reloc_root; 21725d4f98a2SYan Zheng struct btrfs_root_item *root_item; 21735d4f98a2SYan Zheng struct btrfs_path *path; 21743fd0a558SYan, Zheng struct extent_buffer *leaf; 21755d4f98a2SYan Zheng int level; 21765d4f98a2SYan Zheng int max_level; 21775d4f98a2SYan Zheng int replaced = 0; 21785d4f98a2SYan Zheng int ret; 21795d4f98a2SYan Zheng int err = 0; 21803fd0a558SYan, Zheng u32 min_reserved; 21815d4f98a2SYan Zheng 21825d4f98a2SYan Zheng path = btrfs_alloc_path(); 21835d4f98a2SYan Zheng if (!path) 21845d4f98a2SYan Zheng return -ENOMEM; 2185e4058b54SDavid Sterba path->reada = READA_FORWARD; 21865d4f98a2SYan Zheng 21875d4f98a2SYan Zheng reloc_root = root->reloc_root; 21885d4f98a2SYan Zheng root_item = &reloc_root->root_item; 21895d4f98a2SYan Zheng 21905d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 21915d4f98a2SYan Zheng level = btrfs_root_level(root_item); 21925d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 21935d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 21945d4f98a2SYan Zheng path->slots[level] = 0; 21955d4f98a2SYan Zheng } else { 21965d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 21975d4f98a2SYan Zheng 21985d4f98a2SYan Zheng level = root_item->drop_level; 21995d4f98a2SYan Zheng BUG_ON(level == 0); 22005d4f98a2SYan Zheng path->lowest_level = level; 22015d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 220233c66f43SYan Zheng path->lowest_level = 0; 22035d4f98a2SYan Zheng if (ret < 0) { 22045d4f98a2SYan Zheng btrfs_free_path(path); 22055d4f98a2SYan Zheng return ret; 22065d4f98a2SYan Zheng } 22075d4f98a2SYan Zheng 22085d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 22095d4f98a2SYan Zheng path->slots[level]); 22105d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 22115d4f98a2SYan Zheng 22125d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 22135d4f98a2SYan Zheng } 22145d4f98a2SYan Zheng 22150b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22165d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 22175d4f98a2SYan Zheng 22185d4f98a2SYan Zheng while (1) { 221908e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 222008e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22213fd0a558SYan, Zheng if (ret) { 22229e6a0c52SJosef Bacik err = ret; 22239e6a0c52SJosef Bacik goto out; 22243fd0a558SYan, Zheng } 22259e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 22269e6a0c52SJosef Bacik if (IS_ERR(trans)) { 22279e6a0c52SJosef Bacik err = PTR_ERR(trans); 22289e6a0c52SJosef Bacik trans = NULL; 22299e6a0c52SJosef Bacik goto out; 22309e6a0c52SJosef Bacik } 22319e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 22323fd0a558SYan, Zheng 22333fd0a558SYan, Zheng replaced = 0; 22345d4f98a2SYan Zheng max_level = level; 22355d4f98a2SYan Zheng 22365d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 22375d4f98a2SYan Zheng if (ret < 0) { 22385d4f98a2SYan Zheng err = ret; 22395d4f98a2SYan Zheng goto out; 22405d4f98a2SYan Zheng } 22415d4f98a2SYan Zheng if (ret > 0) 22425d4f98a2SYan Zheng break; 22435d4f98a2SYan Zheng 22445d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 22455d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 22465d4f98a2SYan Zheng ret = 0; 22475d4f98a2SYan Zheng } else { 22483fd0a558SYan, Zheng ret = replace_path(trans, root, reloc_root, path, 22493fd0a558SYan, Zheng &next_key, level, max_level); 22505d4f98a2SYan Zheng } 22515d4f98a2SYan Zheng if (ret < 0) { 22525d4f98a2SYan Zheng err = ret; 22535d4f98a2SYan Zheng goto out; 22545d4f98a2SYan Zheng } 22555d4f98a2SYan Zheng 22565d4f98a2SYan Zheng if (ret > 0) { 22575d4f98a2SYan Zheng level = ret; 22585d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 22595d4f98a2SYan Zheng path->slots[level]); 22605d4f98a2SYan Zheng replaced = 1; 22615d4f98a2SYan Zheng } 22625d4f98a2SYan Zheng 22635d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 22645d4f98a2SYan Zheng if (ret > 0) 22655d4f98a2SYan Zheng break; 22665d4f98a2SYan Zheng 22675d4f98a2SYan Zheng BUG_ON(level == 0); 22685d4f98a2SYan Zheng /* 22695d4f98a2SYan Zheng * save the merging progress in the drop_progress. 22705d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 22715d4f98a2SYan Zheng */ 22725d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 22735d4f98a2SYan Zheng path->slots[level]); 22745d4f98a2SYan Zheng root_item->drop_level = level; 22755d4f98a2SYan Zheng 22763a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 22779e6a0c52SJosef Bacik trans = NULL; 22785d4f98a2SYan Zheng 22792ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 22805d4f98a2SYan Zheng 22815d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 22825d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 22835d4f98a2SYan Zheng } 22845d4f98a2SYan Zheng 22855d4f98a2SYan Zheng /* 22865d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 22875d4f98a2SYan Zheng * relocated and the block is tree root. 22885d4f98a2SYan Zheng */ 22895d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 22905d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 22915d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 22925d4f98a2SYan Zheng free_extent_buffer(leaf); 22935d4f98a2SYan Zheng if (ret < 0) 22945d4f98a2SYan Zheng err = ret; 22955d4f98a2SYan Zheng out: 22965d4f98a2SYan Zheng btrfs_free_path(path); 22975d4f98a2SYan Zheng 22985d4f98a2SYan Zheng if (err == 0) { 22995d4f98a2SYan Zheng memset(&root_item->drop_progress, 0, 23005d4f98a2SYan Zheng sizeof(root_item->drop_progress)); 23015d4f98a2SYan Zheng root_item->drop_level = 0; 23025d4f98a2SYan Zheng btrfs_set_root_refs(root_item, 0); 23033fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 23045d4f98a2SYan Zheng } 23055d4f98a2SYan Zheng 23069e6a0c52SJosef Bacik if (trans) 23073a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23085d4f98a2SYan Zheng 23092ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23105d4f98a2SYan Zheng 23115d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23125d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23135d4f98a2SYan Zheng 23145d4f98a2SYan Zheng return err; 23155d4f98a2SYan Zheng } 23165d4f98a2SYan Zheng 23173fd0a558SYan, Zheng static noinline_for_stack 23183fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 23195d4f98a2SYan Zheng { 23203fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 23210b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23223fd0a558SYan, Zheng struct btrfs_root *reloc_root; 23235d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 23243fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23253fd0a558SYan, Zheng u64 num_bytes = 0; 23263fd0a558SYan, Zheng int ret; 23273fd0a558SYan, Zheng 23280b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23290b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23303fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 23310b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 23327585717fSChris Mason 23333fd0a558SYan, Zheng again: 23343fd0a558SYan, Zheng if (!err) { 23353fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 233608e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 233708e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23383fd0a558SYan, Zheng if (ret) 23393fd0a558SYan, Zheng err = ret; 23403fd0a558SYan, Zheng } 23413fd0a558SYan, Zheng 23427a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 23433612b495STsutomu Itoh if (IS_ERR(trans)) { 23443612b495STsutomu Itoh if (!err) 23452ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23462ff7e61eSJeff Mahoney num_bytes); 23473612b495STsutomu Itoh return PTR_ERR(trans); 23483612b495STsutomu Itoh } 23493fd0a558SYan, Zheng 23503fd0a558SYan, Zheng if (!err) { 23513fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 23523a45bb20SJeff Mahoney btrfs_end_transaction(trans); 23532ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23542ff7e61eSJeff Mahoney num_bytes); 23553fd0a558SYan, Zheng goto again; 23563fd0a558SYan, Zheng } 23573fd0a558SYan, Zheng } 23583fd0a558SYan, Zheng 23593fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 23603fd0a558SYan, Zheng 23613fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 23623fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 23633fd0a558SYan, Zheng struct btrfs_root, root_list); 23643fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 23653fd0a558SYan, Zheng 23660b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 23673fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 23683fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 23693fd0a558SYan, Zheng 23703fd0a558SYan, Zheng /* 23713fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 23723fd0a558SYan, Zheng * knows it should resumes merging 23733fd0a558SYan, Zheng */ 23743fd0a558SYan, Zheng if (!err) 23753fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 23763fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 23773fd0a558SYan, Zheng 23783fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 23793fd0a558SYan, Zheng } 23803fd0a558SYan, Zheng 23813fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 23823fd0a558SYan, Zheng 23833fd0a558SYan, Zheng if (!err) 23843a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 23853fd0a558SYan, Zheng else 23863a45bb20SJeff Mahoney btrfs_end_transaction(trans); 23873fd0a558SYan, Zheng return err; 23883fd0a558SYan, Zheng } 23893fd0a558SYan, Zheng 23903fd0a558SYan, Zheng static noinline_for_stack 2391aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2392aca1bba6SLiu Bo { 2393aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2394aca1bba6SLiu Bo 2395aca1bba6SLiu Bo while (!list_empty(list)) { 2396aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2397aca1bba6SLiu Bo root_list); 2398bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 23996bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 24006bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 24016bdf131fSJosef Bacik reloc_root->node = NULL; 24026bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2403aca1bba6SLiu Bo } 2404aca1bba6SLiu Bo } 2405aca1bba6SLiu Bo 2406aca1bba6SLiu Bo static noinline_for_stack 240794404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 24083fd0a558SYan, Zheng { 24090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 24105d4f98a2SYan Zheng struct btrfs_root *root; 24115d4f98a2SYan Zheng struct btrfs_root *reloc_root; 24123fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24133fd0a558SYan, Zheng int found = 0; 2414aca1bba6SLiu Bo int ret = 0; 24153fd0a558SYan, Zheng again: 24163fd0a558SYan, Zheng root = rc->extent_root; 24177585717fSChris Mason 24187585717fSChris Mason /* 24197585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 24207585717fSChris Mason * we have to make sure nobody is in the middle of 24217585717fSChris Mason * adding their roots to the list while we are 24227585717fSChris Mason * doing this splice 24237585717fSChris Mason */ 24240b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24253fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 24260b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24275d4f98a2SYan Zheng 24283fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 24293fd0a558SYan, Zheng found = 1; 24303fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 24313fd0a558SYan, Zheng struct btrfs_root, root_list); 24325d4f98a2SYan Zheng 24335d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 24340b246afaSJeff Mahoney root = read_fs_root(fs_info, 24355d4f98a2SYan Zheng reloc_root->root_key.offset); 24365d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24375d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24385d4f98a2SYan Zheng 24393fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2440b37b39cdSJosef Bacik if (ret) { 244125e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 244225e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 244325e293c2SWang Shilong &reloc_roots); 2444aca1bba6SLiu Bo goto out; 2445b37b39cdSJosef Bacik } 24463fd0a558SYan, Zheng } else { 24473fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24483fd0a558SYan, Zheng } 24495bc7247aSMiao Xie 24502c536799SJeff Mahoney ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1); 2451aca1bba6SLiu Bo if (ret < 0) { 2452aca1bba6SLiu Bo if (list_empty(&reloc_root->root_list)) 2453aca1bba6SLiu Bo list_add_tail(&reloc_root->root_list, 2454aca1bba6SLiu Bo &reloc_roots); 2455aca1bba6SLiu Bo goto out; 2456aca1bba6SLiu Bo } 24575d4f98a2SYan Zheng } 24585d4f98a2SYan Zheng 24593fd0a558SYan, Zheng if (found) { 24603fd0a558SYan, Zheng found = 0; 24613fd0a558SYan, Zheng goto again; 24625d4f98a2SYan Zheng } 2463aca1bba6SLiu Bo out: 2464aca1bba6SLiu Bo if (ret) { 24650b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2466aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2467aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2468467bb1d2SWang Shilong 2469467bb1d2SWang Shilong /* new reloc root may be added */ 24700b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2471467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 24720b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2473467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2474467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2475aca1bba6SLiu Bo } 2476aca1bba6SLiu Bo 24775d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 24785d4f98a2SYan Zheng } 24795d4f98a2SYan Zheng 24805d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 24815d4f98a2SYan Zheng { 24825d4f98a2SYan Zheng struct tree_block *block; 24835d4f98a2SYan Zheng struct rb_node *rb_node; 24845d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 24855d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 24865d4f98a2SYan Zheng rb_erase(rb_node, blocks); 24875d4f98a2SYan Zheng kfree(block); 24885d4f98a2SYan Zheng } 24895d4f98a2SYan Zheng } 24905d4f98a2SYan Zheng 24915d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 24925d4f98a2SYan Zheng struct btrfs_root *reloc_root) 24935d4f98a2SYan Zheng { 24940b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 24955d4f98a2SYan Zheng struct btrfs_root *root; 24965d4f98a2SYan Zheng 24975d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 24985d4f98a2SYan Zheng return 0; 24995d4f98a2SYan Zheng 25000b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25015d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25025d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25035d4f98a2SYan Zheng 25045d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 25055d4f98a2SYan Zheng } 25065d4f98a2SYan Zheng 25073fd0a558SYan, Zheng static noinline_for_stack 25083fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 25093fd0a558SYan, Zheng struct reloc_control *rc, 25105d4f98a2SYan Zheng struct backref_node *node, 2511dc4103f9SWang Shilong struct backref_edge *edges[]) 25125d4f98a2SYan Zheng { 25135d4f98a2SYan Zheng struct backref_node *next; 25145d4f98a2SYan Zheng struct btrfs_root *root; 25153fd0a558SYan, Zheng int index = 0; 25163fd0a558SYan, Zheng 25175d4f98a2SYan Zheng next = node; 25185d4f98a2SYan Zheng while (1) { 25195d4f98a2SYan Zheng cond_resched(); 25205d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 25215d4f98a2SYan Zheng root = next->root; 25223fd0a558SYan, Zheng BUG_ON(!root); 252327cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 25245d4f98a2SYan Zheng 25255d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 25265d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 25275d4f98a2SYan Zheng break; 25285d4f98a2SYan Zheng } 25295d4f98a2SYan Zheng 25305d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 25313fd0a558SYan, Zheng root = root->reloc_root; 25323fd0a558SYan, Zheng 25333fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 25343fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 25353fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 25363fd0a558SYan, Zheng next->new_bytenr = root->node->start; 25373fd0a558SYan, Zheng next->root = root; 25383fd0a558SYan, Zheng list_add_tail(&next->list, 25393fd0a558SYan, Zheng &rc->backref_cache.changed); 25403fd0a558SYan, Zheng __mark_block_processed(rc, next); 25415d4f98a2SYan Zheng break; 25425d4f98a2SYan Zheng } 25435d4f98a2SYan Zheng 25443fd0a558SYan, Zheng WARN_ON(1); 25455d4f98a2SYan Zheng root = NULL; 25465d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25475d4f98a2SYan Zheng if (!next || next->level <= node->level) 25485d4f98a2SYan Zheng break; 25495d4f98a2SYan Zheng } 25503fd0a558SYan, Zheng if (!root) 25513fd0a558SYan, Zheng return NULL; 25525d4f98a2SYan Zheng 25533fd0a558SYan, Zheng next = node; 25543fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 25553fd0a558SYan, Zheng while (1) { 25563fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 25573fd0a558SYan, Zheng if (--index < 0) 25583fd0a558SYan, Zheng break; 25593fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 25603fd0a558SYan, Zheng } 25615d4f98a2SYan Zheng return root; 25625d4f98a2SYan Zheng } 25635d4f98a2SYan Zheng 25643fd0a558SYan, Zheng /* 25653fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 25663fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 25673fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 25683fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 25693fd0a558SYan, Zheng */ 25705d4f98a2SYan Zheng static noinline_for_stack 2571147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 25725d4f98a2SYan Zheng { 25733fd0a558SYan, Zheng struct backref_node *next; 25743fd0a558SYan, Zheng struct btrfs_root *root; 25753fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 25765d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25773fd0a558SYan, Zheng int index = 0; 25783fd0a558SYan, Zheng 25793fd0a558SYan, Zheng next = node; 25803fd0a558SYan, Zheng while (1) { 25813fd0a558SYan, Zheng cond_resched(); 25823fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 25833fd0a558SYan, Zheng root = next->root; 25843fd0a558SYan, Zheng BUG_ON(!root); 25853fd0a558SYan, Zheng 258625985edcSLucas De Marchi /* no other choice for non-references counted tree */ 258727cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 25883fd0a558SYan, Zheng return root; 25893fd0a558SYan, Zheng 25903fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 25913fd0a558SYan, Zheng fs_root = root; 25923fd0a558SYan, Zheng 25933fd0a558SYan, Zheng if (next != node) 25943fd0a558SYan, Zheng return NULL; 25953fd0a558SYan, Zheng 25963fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 25973fd0a558SYan, Zheng if (!next || next->level <= node->level) 25983fd0a558SYan, Zheng break; 25993fd0a558SYan, Zheng } 26003fd0a558SYan, Zheng 26013fd0a558SYan, Zheng if (!fs_root) 26023fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 26033fd0a558SYan, Zheng return fs_root; 26045d4f98a2SYan Zheng } 26055d4f98a2SYan Zheng 26065d4f98a2SYan Zheng static noinline_for_stack 26073fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 26083fd0a558SYan, Zheng struct backref_node *node, int reserve) 26095d4f98a2SYan Zheng { 26100b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 26113fd0a558SYan, Zheng struct backref_node *next = node; 26123fd0a558SYan, Zheng struct backref_edge *edge; 26133fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26143fd0a558SYan, Zheng u64 num_bytes = 0; 26153fd0a558SYan, Zheng int index = 0; 26165d4f98a2SYan Zheng 26173fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 26183fd0a558SYan, Zheng 26193fd0a558SYan, Zheng while (next) { 26203fd0a558SYan, Zheng cond_resched(); 26215d4f98a2SYan Zheng while (1) { 26223fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 26235d4f98a2SYan Zheng break; 26245d4f98a2SYan Zheng 26250b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 26263fd0a558SYan, Zheng 26273fd0a558SYan, Zheng if (list_empty(&next->upper)) 26283fd0a558SYan, Zheng break; 26293fd0a558SYan, Zheng 26303fd0a558SYan, Zheng edge = list_entry(next->upper.next, 26313fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 26323fd0a558SYan, Zheng edges[index++] = edge; 26333fd0a558SYan, Zheng next = edge->node[UPPER]; 26345d4f98a2SYan Zheng } 26353fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26363fd0a558SYan, Zheng } 26373fd0a558SYan, Zheng return num_bytes; 26383fd0a558SYan, Zheng } 26393fd0a558SYan, Zheng 26403fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 26413fd0a558SYan, Zheng struct reloc_control *rc, 26423fd0a558SYan, Zheng struct backref_node *node) 26433fd0a558SYan, Zheng { 26443fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2645da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 26463fd0a558SYan, Zheng u64 num_bytes; 26473fd0a558SYan, Zheng int ret; 26480647bf56SWang Shilong u64 tmp; 26493fd0a558SYan, Zheng 26503fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 26513fd0a558SYan, Zheng 26523fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 26530647bf56SWang Shilong rc->reserved_bytes += num_bytes; 26548ca17f0fSJosef Bacik 26558ca17f0fSJosef Bacik /* 26568ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 26578ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 26588ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 26598ca17f0fSJosef Bacik */ 26600647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 26618ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 26623fd0a558SYan, Zheng if (ret) { 2663da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 26640647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 26650647bf56SWang Shilong tmp <<= 1; 26660647bf56SWang Shilong /* 26670647bf56SWang Shilong * only one thread can access block_rsv at this point, 26680647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 26690647bf56SWang Shilong * we expand more reservation size here to allow enough 26708ca17f0fSJosef Bacik * space for relocation and we will return eailer in 26710647bf56SWang Shilong * enospc case. 26720647bf56SWang Shilong */ 2673da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 26740647bf56SWang Shilong RELOCATION_RESERVED_NODES; 26758ca17f0fSJosef Bacik return -EAGAIN; 26763fd0a558SYan, Zheng } 26773fd0a558SYan, Zheng 26783fd0a558SYan, Zheng return 0; 26793fd0a558SYan, Zheng } 26803fd0a558SYan, Zheng 26815d4f98a2SYan Zheng /* 26825d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 26835d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 26845d4f98a2SYan Zheng * 26855d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 26865d4f98a2SYan Zheng * in that case this function just updates pointers. 26875d4f98a2SYan Zheng */ 26885d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 26893fd0a558SYan, Zheng struct reloc_control *rc, 26905d4f98a2SYan Zheng struct backref_node *node, 26915d4f98a2SYan Zheng struct btrfs_key *key, 26925d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 26935d4f98a2SYan Zheng { 26942ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 26955d4f98a2SYan Zheng struct backref_node *upper; 26965d4f98a2SYan Zheng struct backref_edge *edge; 26975d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26985d4f98a2SYan Zheng struct btrfs_root *root; 26995d4f98a2SYan Zheng struct extent_buffer *eb; 27005d4f98a2SYan Zheng u32 blocksize; 27015d4f98a2SYan Zheng u64 bytenr; 27025d4f98a2SYan Zheng u64 generation; 27035d4f98a2SYan Zheng int slot; 27045d4f98a2SYan Zheng int ret; 27055d4f98a2SYan Zheng int err = 0; 27065d4f98a2SYan Zheng 27075d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 27085d4f98a2SYan Zheng 27095d4f98a2SYan Zheng path->lowest_level = node->level + 1; 27103fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 27115d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2712581c1760SQu Wenruo struct btrfs_key first_key; 2713581c1760SQu Wenruo 27145d4f98a2SYan Zheng cond_resched(); 27155d4f98a2SYan Zheng 27165d4f98a2SYan Zheng upper = edge->node[UPPER]; 2717dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 27183fd0a558SYan, Zheng BUG_ON(!root); 27195d4f98a2SYan Zheng 27203fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 27213fd0a558SYan, Zheng if (!lowest) { 27223fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 27233fd0a558SYan, Zheng upper->level, &slot); 27243fd0a558SYan, Zheng BUG_ON(ret); 27253fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27263fd0a558SYan, Zheng if (node->eb->start == bytenr) 27273fd0a558SYan, Zheng goto next; 27283fd0a558SYan, Zheng } 27295d4f98a2SYan Zheng drop_node_buffer(upper); 27303fd0a558SYan, Zheng } 27315d4f98a2SYan Zheng 27325d4f98a2SYan Zheng if (!upper->eb) { 27335d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 27343561b9dbSLiu Bo if (ret) { 27353561b9dbSLiu Bo if (ret < 0) 27365d4f98a2SYan Zheng err = ret; 27373561b9dbSLiu Bo else 27383561b9dbSLiu Bo err = -ENOENT; 27393561b9dbSLiu Bo 27403561b9dbSLiu Bo btrfs_release_path(path); 27415d4f98a2SYan Zheng break; 27425d4f98a2SYan Zheng } 27435d4f98a2SYan Zheng 27443fd0a558SYan, Zheng if (!upper->eb) { 27453fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 27463fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 27473fd0a558SYan, Zheng } else { 27483fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 27493fd0a558SYan, Zheng } 27503fd0a558SYan, Zheng 27513fd0a558SYan, Zheng upper->locked = 1; 27523fd0a558SYan, Zheng path->locks[upper->level] = 0; 27533fd0a558SYan, Zheng 27545d4f98a2SYan Zheng slot = path->slots[upper->level]; 2755b3b4aa74SDavid Sterba btrfs_release_path(path); 27565d4f98a2SYan Zheng } else { 27575d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 27585d4f98a2SYan Zheng &slot); 27595d4f98a2SYan Zheng BUG_ON(ret); 27605d4f98a2SYan Zheng } 27615d4f98a2SYan Zheng 27625d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27633fd0a558SYan, Zheng if (lowest) { 27644547f4d8SLiu Bo if (bytenr != node->bytenr) { 27654547f4d8SLiu Bo btrfs_err(root->fs_info, 27664547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 27674547f4d8SLiu Bo bytenr, node->bytenr, slot, 27684547f4d8SLiu Bo upper->eb->start); 27694547f4d8SLiu Bo err = -EIO; 27704547f4d8SLiu Bo goto next; 27714547f4d8SLiu Bo } 27725d4f98a2SYan Zheng } else { 27733fd0a558SYan, Zheng if (node->eb->start == bytenr) 27743fd0a558SYan, Zheng goto next; 27755d4f98a2SYan Zheng } 27765d4f98a2SYan Zheng 2777da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 27785d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2779581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2780581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2781581c1760SQu Wenruo upper->level - 1, &first_key); 278264c043deSLiu Bo if (IS_ERR(eb)) { 278364c043deSLiu Bo err = PTR_ERR(eb); 278464c043deSLiu Bo goto next; 278564c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2786416bc658SJosef Bacik free_extent_buffer(eb); 278797d9a8a4STsutomu Itoh err = -EIO; 278897d9a8a4STsutomu Itoh goto next; 278997d9a8a4STsutomu Itoh } 27905d4f98a2SYan Zheng btrfs_tree_lock(eb); 27915d4f98a2SYan Zheng btrfs_set_lock_blocking(eb); 27925d4f98a2SYan Zheng 27935d4f98a2SYan Zheng if (!node->eb) { 27945d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 27955d4f98a2SYan Zheng slot, &eb); 27963fd0a558SYan, Zheng btrfs_tree_unlock(eb); 27973fd0a558SYan, Zheng free_extent_buffer(eb); 27985d4f98a2SYan Zheng if (ret < 0) { 27995d4f98a2SYan Zheng err = ret; 28003fd0a558SYan, Zheng goto next; 28015d4f98a2SYan Zheng } 28023fd0a558SYan, Zheng BUG_ON(node->eb != eb); 28035d4f98a2SYan Zheng } else { 28045d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 28055d4f98a2SYan Zheng node->eb->start); 28065d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 28075d4f98a2SYan Zheng trans->transid); 28085d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 28095d4f98a2SYan Zheng 281084f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, root, 28115d4f98a2SYan Zheng node->eb->start, blocksize, 28125d4f98a2SYan Zheng upper->eb->start, 28135d4f98a2SYan Zheng btrfs_header_owner(upper->eb), 2814b06c4bf5SFilipe Manana node->level, 0); 28155d4f98a2SYan Zheng BUG_ON(ret); 28165d4f98a2SYan Zheng 28175d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 28185d4f98a2SYan Zheng BUG_ON(ret); 28195d4f98a2SYan Zheng } 28203fd0a558SYan, Zheng next: 28213fd0a558SYan, Zheng if (!upper->pending) 28223fd0a558SYan, Zheng drop_node_buffer(upper); 28233fd0a558SYan, Zheng else 28243fd0a558SYan, Zheng unlock_node_buffer(upper); 28253fd0a558SYan, Zheng if (err) 28263fd0a558SYan, Zheng break; 28275d4f98a2SYan Zheng } 28283fd0a558SYan, Zheng 28293fd0a558SYan, Zheng if (!err && node->pending) { 28303fd0a558SYan, Zheng drop_node_buffer(node); 28313fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 28323fd0a558SYan, Zheng node->pending = 0; 28335d4f98a2SYan Zheng } 28343fd0a558SYan, Zheng 28355d4f98a2SYan Zheng path->lowest_level = 0; 28363fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 28375d4f98a2SYan Zheng return err; 28385d4f98a2SYan Zheng } 28395d4f98a2SYan Zheng 28405d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 28413fd0a558SYan, Zheng struct reloc_control *rc, 28425d4f98a2SYan Zheng struct backref_node *node, 28435d4f98a2SYan Zheng struct btrfs_path *path) 28445d4f98a2SYan Zheng { 28455d4f98a2SYan Zheng struct btrfs_key key; 28465d4f98a2SYan Zheng 28475d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 28483fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 28495d4f98a2SYan Zheng } 28505d4f98a2SYan Zheng 28515d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 28523fd0a558SYan, Zheng struct reloc_control *rc, 28533fd0a558SYan, Zheng struct btrfs_path *path, int err) 28545d4f98a2SYan Zheng { 28553fd0a558SYan, Zheng LIST_HEAD(list); 28563fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 28575d4f98a2SYan Zheng struct backref_node *node; 28585d4f98a2SYan Zheng int level; 28595d4f98a2SYan Zheng int ret; 28605d4f98a2SYan Zheng 28615d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 28625d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 28635d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 28643fd0a558SYan, Zheng struct backref_node, list); 28653fd0a558SYan, Zheng list_move_tail(&node->list, &list); 28663fd0a558SYan, Zheng BUG_ON(!node->pending); 28675d4f98a2SYan Zheng 28683fd0a558SYan, Zheng if (!err) { 28693fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 28705d4f98a2SYan Zheng if (ret < 0) 28715d4f98a2SYan Zheng err = ret; 28725d4f98a2SYan Zheng } 28735d4f98a2SYan Zheng } 28743fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 28753fd0a558SYan, Zheng } 28765d4f98a2SYan Zheng return err; 28775d4f98a2SYan Zheng } 28785d4f98a2SYan Zheng 28795d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 28803fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 28813fd0a558SYan, Zheng { 28823fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2883ceeb0ae7SDavid Sterba EXTENT_DIRTY); 28843fd0a558SYan, Zheng } 28853fd0a558SYan, Zheng 28863fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 28875d4f98a2SYan Zheng struct backref_node *node) 28885d4f98a2SYan Zheng { 28895d4f98a2SYan Zheng u32 blocksize; 28905d4f98a2SYan Zheng if (node->level == 0 || 28915d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2892da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 28933fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 28945d4f98a2SYan Zheng } 28955d4f98a2SYan Zheng node->processed = 1; 28965d4f98a2SYan Zheng } 28975d4f98a2SYan Zheng 28985d4f98a2SYan Zheng /* 28995d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 29005d4f98a2SYan Zheng * as processed. 29015d4f98a2SYan Zheng */ 29025d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 29035d4f98a2SYan Zheng struct backref_node *node) 29045d4f98a2SYan Zheng { 29055d4f98a2SYan Zheng struct backref_node *next = node; 29065d4f98a2SYan Zheng struct backref_edge *edge; 29075d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 29085d4f98a2SYan Zheng int index = 0; 29095d4f98a2SYan Zheng 29105d4f98a2SYan Zheng while (next) { 29115d4f98a2SYan Zheng cond_resched(); 29125d4f98a2SYan Zheng while (1) { 29135d4f98a2SYan Zheng if (next->processed) 29145d4f98a2SYan Zheng break; 29155d4f98a2SYan Zheng 29163fd0a558SYan, Zheng __mark_block_processed(rc, next); 29175d4f98a2SYan Zheng 29185d4f98a2SYan Zheng if (list_empty(&next->upper)) 29195d4f98a2SYan Zheng break; 29205d4f98a2SYan Zheng 29215d4f98a2SYan Zheng edge = list_entry(next->upper.next, 29225d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 29235d4f98a2SYan Zheng edges[index++] = edge; 29245d4f98a2SYan Zheng next = edge->node[UPPER]; 29255d4f98a2SYan Zheng } 29265d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 29275d4f98a2SYan Zheng } 29285d4f98a2SYan Zheng } 29295d4f98a2SYan Zheng 29307476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 29315d4f98a2SYan Zheng { 2932da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 29337476dfdaSDavid Sterba 29345d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 29359655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 29365d4f98a2SYan Zheng return 1; 29375d4f98a2SYan Zheng return 0; 29385d4f98a2SYan Zheng } 29395d4f98a2SYan Zheng 29402ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 29415d4f98a2SYan Zheng struct tree_block *block) 29425d4f98a2SYan Zheng { 29435d4f98a2SYan Zheng struct extent_buffer *eb; 29445d4f98a2SYan Zheng 29455d4f98a2SYan Zheng BUG_ON(block->key_ready); 2946581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 2947581c1760SQu Wenruo block->level, NULL); 294864c043deSLiu Bo if (IS_ERR(eb)) { 294964c043deSLiu Bo return PTR_ERR(eb); 295064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2951416bc658SJosef Bacik free_extent_buffer(eb); 2952416bc658SJosef Bacik return -EIO; 2953416bc658SJosef Bacik } 29545d4f98a2SYan Zheng WARN_ON(btrfs_header_level(eb) != block->level); 29555d4f98a2SYan Zheng if (block->level == 0) 29565d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 29575d4f98a2SYan Zheng else 29585d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 29595d4f98a2SYan Zheng free_extent_buffer(eb); 29605d4f98a2SYan Zheng block->key_ready = 1; 29615d4f98a2SYan Zheng return 0; 29625d4f98a2SYan Zheng } 29635d4f98a2SYan Zheng 29645d4f98a2SYan Zheng /* 29655d4f98a2SYan Zheng * helper function to relocate a tree block 29665d4f98a2SYan Zheng */ 29675d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 29685d4f98a2SYan Zheng struct reloc_control *rc, 29695d4f98a2SYan Zheng struct backref_node *node, 29705d4f98a2SYan Zheng struct btrfs_key *key, 29715d4f98a2SYan Zheng struct btrfs_path *path) 29725d4f98a2SYan Zheng { 29735d4f98a2SYan Zheng struct btrfs_root *root; 29743fd0a558SYan, Zheng int ret = 0; 29755d4f98a2SYan Zheng 29763fd0a558SYan, Zheng if (!node) 29775d4f98a2SYan Zheng return 0; 29783fd0a558SYan, Zheng 29793fd0a558SYan, Zheng BUG_ON(node->processed); 2980147d256eSZhaolei root = select_one_root(node); 29813fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 29823fd0a558SYan, Zheng update_processed_blocks(rc, node); 29833fd0a558SYan, Zheng goto out; 29845d4f98a2SYan Zheng } 29855d4f98a2SYan Zheng 298627cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29873fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 29883fd0a558SYan, Zheng if (ret) 29895d4f98a2SYan Zheng goto out; 29905d4f98a2SYan Zheng } 29913fd0a558SYan, Zheng 29923fd0a558SYan, Zheng if (root) { 299327cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 29943fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 29953fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 29963fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 29973fd0a558SYan, Zheng root = root->reloc_root; 29983fd0a558SYan, Zheng node->new_bytenr = root->node->start; 29993fd0a558SYan, Zheng node->root = root; 30003fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 30013fd0a558SYan, Zheng } else { 30025d4f98a2SYan Zheng path->lowest_level = node->level; 30035d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3004b3b4aa74SDavid Sterba btrfs_release_path(path); 30053fd0a558SYan, Zheng if (ret > 0) 30065d4f98a2SYan Zheng ret = 0; 30073fd0a558SYan, Zheng } 30083fd0a558SYan, Zheng if (!ret) 30093fd0a558SYan, Zheng update_processed_blocks(rc, node); 30103fd0a558SYan, Zheng } else { 30113fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 30123fd0a558SYan, Zheng } 30135d4f98a2SYan Zheng out: 30140647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 30153fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 30165d4f98a2SYan Zheng return ret; 30175d4f98a2SYan Zheng } 30185d4f98a2SYan Zheng 30195d4f98a2SYan Zheng /* 30205d4f98a2SYan Zheng * relocate a list of blocks 30215d4f98a2SYan Zheng */ 30225d4f98a2SYan Zheng static noinline_for_stack 30235d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 30245d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 30255d4f98a2SYan Zheng { 30262ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 30275d4f98a2SYan Zheng struct backref_node *node; 30285d4f98a2SYan Zheng struct btrfs_path *path; 30295d4f98a2SYan Zheng struct tree_block *block; 30305d4f98a2SYan Zheng struct rb_node *rb_node; 30315d4f98a2SYan Zheng int ret; 30325d4f98a2SYan Zheng int err = 0; 30335d4f98a2SYan Zheng 30345d4f98a2SYan Zheng path = btrfs_alloc_path(); 3035e1a12670SLiu Bo if (!path) { 3036e1a12670SLiu Bo err = -ENOMEM; 303734c2b290SDavid Sterba goto out_free_blocks; 3038e1a12670SLiu Bo } 30395d4f98a2SYan Zheng 30405d4f98a2SYan Zheng rb_node = rb_first(blocks); 30415d4f98a2SYan Zheng while (rb_node) { 30425d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 30435d4f98a2SYan Zheng if (!block->key_ready) 30442ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 30455d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30465d4f98a2SYan Zheng } 30475d4f98a2SYan Zheng 30485d4f98a2SYan Zheng rb_node = rb_first(blocks); 30495d4f98a2SYan Zheng while (rb_node) { 30505d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 305134c2b290SDavid Sterba if (!block->key_ready) { 30522ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 305334c2b290SDavid Sterba if (err) 305434c2b290SDavid Sterba goto out_free_path; 305534c2b290SDavid Sterba } 30565d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30575d4f98a2SYan Zheng } 30585d4f98a2SYan Zheng 30595d4f98a2SYan Zheng rb_node = rb_first(blocks); 30605d4f98a2SYan Zheng while (rb_node) { 30615d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 30625d4f98a2SYan Zheng 30633fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 30645d4f98a2SYan Zheng block->level, block->bytenr); 30655d4f98a2SYan Zheng if (IS_ERR(node)) { 30665d4f98a2SYan Zheng err = PTR_ERR(node); 30675d4f98a2SYan Zheng goto out; 30685d4f98a2SYan Zheng } 30695d4f98a2SYan Zheng 30705d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 30715d4f98a2SYan Zheng path); 30725d4f98a2SYan Zheng if (ret < 0) { 30733fd0a558SYan, Zheng if (ret != -EAGAIN || rb_node == rb_first(blocks)) 30745d4f98a2SYan Zheng err = ret; 30755d4f98a2SYan Zheng goto out; 30765d4f98a2SYan Zheng } 30775d4f98a2SYan Zheng rb_node = rb_next(rb_node); 30785d4f98a2SYan Zheng } 30795d4f98a2SYan Zheng out: 30803fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 30815d4f98a2SYan Zheng 308234c2b290SDavid Sterba out_free_path: 30835d4f98a2SYan Zheng btrfs_free_path(path); 308434c2b290SDavid Sterba out_free_blocks: 3085e1a12670SLiu Bo free_block_list(blocks); 30865d4f98a2SYan Zheng return err; 30875d4f98a2SYan Zheng } 30885d4f98a2SYan Zheng 30895d4f98a2SYan Zheng static noinline_for_stack 3090efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3091efa56464SYan, Zheng struct file_extent_cluster *cluster) 3092efa56464SYan, Zheng { 3093efa56464SYan, Zheng u64 alloc_hint = 0; 3094efa56464SYan, Zheng u64 start; 3095efa56464SYan, Zheng u64 end; 3096efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3097efa56464SYan, Zheng u64 num_bytes; 3098efa56464SYan, Zheng int nr = 0; 3099efa56464SYan, Zheng int ret = 0; 3100dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3101dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 310218513091SWang Xiaoguang u64 cur_offset; 3103364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3104efa56464SYan, Zheng 3105efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 31065955102cSAl Viro inode_lock(inode); 3107efa56464SYan, Zheng 3108364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3109dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3110efa56464SYan, Zheng if (ret) 3111efa56464SYan, Zheng goto out; 3112efa56464SYan, Zheng 311318513091SWang Xiaoguang cur_offset = prealloc_start; 3114efa56464SYan, Zheng while (nr < cluster->nr) { 3115efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3116efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3117efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3118efa56464SYan, Zheng else 3119efa56464SYan, Zheng end = cluster->end - offset; 3120efa56464SYan, Zheng 3121d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3122efa56464SYan, Zheng num_bytes = end + 1 - start; 312318513091SWang Xiaoguang if (cur_offset < start) 3124bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3125bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3126efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3127efa56464SYan, Zheng num_bytes, num_bytes, 3128efa56464SYan, Zheng end + 1, &alloc_hint); 312918513091SWang Xiaoguang cur_offset = end + 1; 3130d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3131efa56464SYan, Zheng if (ret) 3132efa56464SYan, Zheng break; 3133efa56464SYan, Zheng nr++; 3134efa56464SYan, Zheng } 313518513091SWang Xiaoguang if (cur_offset < prealloc_end) 3136bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3137bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3138efa56464SYan, Zheng out: 31395955102cSAl Viro inode_unlock(inode); 3140364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3141efa56464SYan, Zheng return ret; 3142efa56464SYan, Zheng } 3143efa56464SYan, Zheng 3144efa56464SYan, Zheng static noinline_for_stack 31450257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 31460257bb82SYan, Zheng u64 block_start) 31475d4f98a2SYan Zheng { 31480b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 31495d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 31505d4f98a2SYan Zheng struct extent_map *em; 31510257bb82SYan, Zheng int ret = 0; 31525d4f98a2SYan Zheng 3153172ddd60SDavid Sterba em = alloc_extent_map(); 31540257bb82SYan, Zheng if (!em) 31550257bb82SYan, Zheng return -ENOMEM; 31560257bb82SYan, Zheng 31575d4f98a2SYan Zheng em->start = start; 31580257bb82SYan, Zheng em->len = end + 1 - start; 31590257bb82SYan, Zheng em->block_len = em->len; 31600257bb82SYan, Zheng em->block_start = block_start; 31610b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 31625d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 31635d4f98a2SYan Zheng 3164d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 31655d4f98a2SYan Zheng while (1) { 3166890871beSChris Mason write_lock(&em_tree->lock); 316709a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3168890871beSChris Mason write_unlock(&em_tree->lock); 31695d4f98a2SYan Zheng if (ret != -EEXIST) { 31705d4f98a2SYan Zheng free_extent_map(em); 31715d4f98a2SYan Zheng break; 31725d4f98a2SYan Zheng } 3173dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 31745d4f98a2SYan Zheng } 3175d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 31760257bb82SYan, Zheng return ret; 31770257bb82SYan, Zheng } 31785d4f98a2SYan Zheng 31790257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 31800257bb82SYan, Zheng struct file_extent_cluster *cluster) 31810257bb82SYan, Zheng { 31822ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 31830257bb82SYan, Zheng u64 page_start; 31840257bb82SYan, Zheng u64 page_end; 31850257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 31860257bb82SYan, Zheng unsigned long index; 31870257bb82SYan, Zheng unsigned long last_index; 31880257bb82SYan, Zheng struct page *page; 31890257bb82SYan, Zheng struct file_ra_state *ra; 31903b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 31910257bb82SYan, Zheng int nr = 0; 31920257bb82SYan, Zheng int ret = 0; 31930257bb82SYan, Zheng 31940257bb82SYan, Zheng if (!cluster->nr) 31950257bb82SYan, Zheng return 0; 31960257bb82SYan, Zheng 31970257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 31980257bb82SYan, Zheng if (!ra) 31990257bb82SYan, Zheng return -ENOMEM; 32000257bb82SYan, Zheng 3201efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 32020257bb82SYan, Zheng if (ret) 3203efa56464SYan, Zheng goto out; 32040257bb82SYan, Zheng 32050257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 32060257bb82SYan, Zheng 3207efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3208efa56464SYan, Zheng cluster->end - offset, cluster->start); 3209efa56464SYan, Zheng if (ret) 3210efa56464SYan, Zheng goto out; 3211efa56464SYan, Zheng 321209cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 321309cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 32140257bb82SYan, Zheng while (index <= last_index) { 32159f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 32169f3db423SNikolay Borisov PAGE_SIZE); 3217efa56464SYan, Zheng if (ret) 3218efa56464SYan, Zheng goto out; 3219efa56464SYan, Zheng 32200257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 32210257bb82SYan, Zheng if (!page) { 32220257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 32230257bb82SYan, Zheng ra, NULL, index, 32240257bb82SYan, Zheng last_index + 1 - index); 3225a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 32263b16a4e3SJosef Bacik mask); 32270257bb82SYan, Zheng if (!page) { 3228691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 322943b18595SQu Wenruo PAGE_SIZE, true); 32300257bb82SYan, Zheng ret = -ENOMEM; 3231efa56464SYan, Zheng goto out; 32320257bb82SYan, Zheng } 32330257bb82SYan, Zheng } 32340257bb82SYan, Zheng 32350257bb82SYan, Zheng if (PageReadahead(page)) { 32360257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 32370257bb82SYan, Zheng ra, NULL, page, index, 32380257bb82SYan, Zheng last_index + 1 - index); 32390257bb82SYan, Zheng } 32400257bb82SYan, Zheng 32410257bb82SYan, Zheng if (!PageUptodate(page)) { 32420257bb82SYan, Zheng btrfs_readpage(NULL, page); 32430257bb82SYan, Zheng lock_page(page); 32440257bb82SYan, Zheng if (!PageUptodate(page)) { 32450257bb82SYan, Zheng unlock_page(page); 324609cbfeafSKirill A. Shutemov put_page(page); 3247691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 324843b18595SQu Wenruo PAGE_SIZE, true); 32498b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 325043b18595SQu Wenruo PAGE_SIZE, true); 32510257bb82SYan, Zheng ret = -EIO; 3252efa56464SYan, Zheng goto out; 32530257bb82SYan, Zheng } 32540257bb82SYan, Zheng } 32550257bb82SYan, Zheng 32564eee4fa4SMiao Xie page_start = page_offset(page); 325709cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 32580257bb82SYan, Zheng 3259d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 32600257bb82SYan, Zheng 32610257bb82SYan, Zheng set_page_extent_mapped(page); 32620257bb82SYan, Zheng 32630257bb82SYan, Zheng if (nr < cluster->nr && 32640257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 32650257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 32660257bb82SYan, Zheng page_start, page_end, 3267ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 32680257bb82SYan, Zheng nr++; 32690257bb82SYan, Zheng } 32700257bb82SYan, Zheng 3271765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3272765f3cebSNikolay Borisov NULL, 0); 3273765f3cebSNikolay Borisov if (ret) { 3274765f3cebSNikolay Borisov unlock_page(page); 3275765f3cebSNikolay Borisov put_page(page); 3276765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 327743b18595SQu Wenruo PAGE_SIZE, true); 3278765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 327943b18595SQu Wenruo PAGE_SIZE, true); 3280765f3cebSNikolay Borisov 3281765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3282765f3cebSNikolay Borisov page_start, page_end, 3283765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3284765f3cebSNikolay Borisov goto out; 3285765f3cebSNikolay Borisov 3286765f3cebSNikolay Borisov } 32870257bb82SYan, Zheng set_page_dirty(page); 32880257bb82SYan, Zheng 32890257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3290d0082371SJeff Mahoney page_start, page_end); 32910257bb82SYan, Zheng unlock_page(page); 329209cbfeafSKirill A. Shutemov put_page(page); 32930257bb82SYan, Zheng 32940257bb82SYan, Zheng index++; 329543b18595SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, 329643b18595SQu Wenruo false); 3297efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 32982ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 32990257bb82SYan, Zheng } 33000257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3301efa56464SYan, Zheng out: 33020257bb82SYan, Zheng kfree(ra); 33030257bb82SYan, Zheng return ret; 33040257bb82SYan, Zheng } 33050257bb82SYan, Zheng 33060257bb82SYan, Zheng static noinline_for_stack 33070257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 33080257bb82SYan, Zheng struct file_extent_cluster *cluster) 33090257bb82SYan, Zheng { 33100257bb82SYan, Zheng int ret; 33110257bb82SYan, Zheng 33120257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 33130257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33140257bb82SYan, Zheng if (ret) 33150257bb82SYan, Zheng return ret; 33160257bb82SYan, Zheng cluster->nr = 0; 33170257bb82SYan, Zheng } 33180257bb82SYan, Zheng 33190257bb82SYan, Zheng if (!cluster->nr) 33200257bb82SYan, Zheng cluster->start = extent_key->objectid; 33210257bb82SYan, Zheng else 33220257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 33230257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 33240257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 33250257bb82SYan, Zheng cluster->nr++; 33260257bb82SYan, Zheng 33270257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 33280257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33290257bb82SYan, Zheng if (ret) 33300257bb82SYan, Zheng return ret; 33310257bb82SYan, Zheng cluster->nr = 0; 33320257bb82SYan, Zheng } 33330257bb82SYan, Zheng return 0; 33345d4f98a2SYan Zheng } 33355d4f98a2SYan Zheng 33365d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 33375d4f98a2SYan Zheng static int get_ref_objectid_v0(struct reloc_control *rc, 33385d4f98a2SYan Zheng struct btrfs_path *path, 33395d4f98a2SYan Zheng struct btrfs_key *extent_key, 33405d4f98a2SYan Zheng u64 *ref_objectid, int *path_change) 33415d4f98a2SYan Zheng { 33425d4f98a2SYan Zheng struct btrfs_key key; 33435d4f98a2SYan Zheng struct extent_buffer *leaf; 33445d4f98a2SYan Zheng struct btrfs_extent_ref_v0 *ref0; 33455d4f98a2SYan Zheng int ret; 33465d4f98a2SYan Zheng int slot; 33475d4f98a2SYan Zheng 33485d4f98a2SYan Zheng leaf = path->nodes[0]; 33495d4f98a2SYan Zheng slot = path->slots[0]; 33505d4f98a2SYan Zheng while (1) { 33515d4f98a2SYan Zheng if (slot >= btrfs_header_nritems(leaf)) { 33525d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 33535d4f98a2SYan Zheng if (ret < 0) 33545d4f98a2SYan Zheng return ret; 33555d4f98a2SYan Zheng BUG_ON(ret > 0); 33565d4f98a2SYan Zheng leaf = path->nodes[0]; 33575d4f98a2SYan Zheng slot = path->slots[0]; 33585d4f98a2SYan Zheng if (path_change) 33595d4f98a2SYan Zheng *path_change = 1; 33605d4f98a2SYan Zheng } 33615d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 33625d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 33635d4f98a2SYan Zheng return -ENOENT; 33645d4f98a2SYan Zheng 33655d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_REF_V0_KEY) { 33665d4f98a2SYan Zheng slot++; 33675d4f98a2SYan Zheng continue; 33685d4f98a2SYan Zheng } 33695d4f98a2SYan Zheng ref0 = btrfs_item_ptr(leaf, slot, 33705d4f98a2SYan Zheng struct btrfs_extent_ref_v0); 33715d4f98a2SYan Zheng *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0); 33725d4f98a2SYan Zheng break; 33735d4f98a2SYan Zheng } 33745d4f98a2SYan Zheng return 0; 33755d4f98a2SYan Zheng } 33765d4f98a2SYan Zheng #endif 33775d4f98a2SYan Zheng 33785d4f98a2SYan Zheng /* 33795d4f98a2SYan Zheng * helper to add a tree block to the list. 33805d4f98a2SYan Zheng * the major work is getting the generation and level of the block 33815d4f98a2SYan Zheng */ 33825d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 33835d4f98a2SYan Zheng struct btrfs_key *extent_key, 33845d4f98a2SYan Zheng struct btrfs_path *path, 33855d4f98a2SYan Zheng struct rb_root *blocks) 33865d4f98a2SYan Zheng { 33875d4f98a2SYan Zheng struct extent_buffer *eb; 33885d4f98a2SYan Zheng struct btrfs_extent_item *ei; 33895d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 33905d4f98a2SYan Zheng struct tree_block *block; 33915d4f98a2SYan Zheng struct rb_node *rb_node; 33925d4f98a2SYan Zheng u32 item_size; 33935d4f98a2SYan Zheng int level = -1; 33947fdf4b60SWang Shilong u64 generation; 33955d4f98a2SYan Zheng 33965d4f98a2SYan Zheng eb = path->nodes[0]; 33975d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 33985d4f98a2SYan Zheng 33993173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 34003173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 34015d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 34025d4f98a2SYan Zheng struct btrfs_extent_item); 34033173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 34045d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 34055d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 34065d4f98a2SYan Zheng } else { 34073173a18fSJosef Bacik level = (int)extent_key->offset; 34083173a18fSJosef Bacik } 34093173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 34103173a18fSJosef Bacik } else { 34115d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 34125d4f98a2SYan Zheng u64 ref_owner; 34135d4f98a2SYan Zheng int ret; 34145d4f98a2SYan Zheng 34155d4f98a2SYan Zheng BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0)); 34165d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, extent_key, 34175d4f98a2SYan Zheng &ref_owner, NULL); 3418411fc6bcSAndi Kleen if (ret < 0) 3419411fc6bcSAndi Kleen return ret; 34205d4f98a2SYan Zheng BUG_ON(ref_owner >= BTRFS_MAX_LEVEL); 34215d4f98a2SYan Zheng level = (int)ref_owner; 34225d4f98a2SYan Zheng /* FIXME: get real generation */ 34235d4f98a2SYan Zheng generation = 0; 34245d4f98a2SYan Zheng #else 34255d4f98a2SYan Zheng BUG(); 34265d4f98a2SYan Zheng #endif 34275d4f98a2SYan Zheng } 34285d4f98a2SYan Zheng 3429b3b4aa74SDavid Sterba btrfs_release_path(path); 34305d4f98a2SYan Zheng 34315d4f98a2SYan Zheng BUG_ON(level == -1); 34325d4f98a2SYan Zheng 34335d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 34345d4f98a2SYan Zheng if (!block) 34355d4f98a2SYan Zheng return -ENOMEM; 34365d4f98a2SYan Zheng 34375d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3438da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 34395d4f98a2SYan Zheng block->key.offset = generation; 34405d4f98a2SYan Zheng block->level = level; 34415d4f98a2SYan Zheng block->key_ready = 0; 34425d4f98a2SYan Zheng 34435d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 344443c04fb1SJeff Mahoney if (rb_node) 344543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 34465d4f98a2SYan Zheng 34475d4f98a2SYan Zheng return 0; 34485d4f98a2SYan Zheng } 34495d4f98a2SYan Zheng 34505d4f98a2SYan Zheng /* 34515d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 34525d4f98a2SYan Zheng */ 34535d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 34545d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 34555d4f98a2SYan Zheng struct rb_root *blocks) 34565d4f98a2SYan Zheng { 34570b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34585d4f98a2SYan Zheng struct btrfs_path *path; 34595d4f98a2SYan Zheng struct btrfs_key key; 34605d4f98a2SYan Zheng int ret; 34610b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 34625d4f98a2SYan Zheng 34637476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 34645d4f98a2SYan Zheng return 0; 34655d4f98a2SYan Zheng 34665d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 34675d4f98a2SYan Zheng return 0; 34685d4f98a2SYan Zheng 34695d4f98a2SYan Zheng path = btrfs_alloc_path(); 34705d4f98a2SYan Zheng if (!path) 34715d4f98a2SYan Zheng return -ENOMEM; 3472aee68ee5SJosef Bacik again: 34735d4f98a2SYan Zheng key.objectid = bytenr; 3474aee68ee5SJosef Bacik if (skinny) { 3475aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3476aee68ee5SJosef Bacik key.offset = (u64)-1; 3477aee68ee5SJosef Bacik } else { 34785d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34795d4f98a2SYan Zheng key.offset = blocksize; 3480aee68ee5SJosef Bacik } 34815d4f98a2SYan Zheng 34825d4f98a2SYan Zheng path->search_commit_root = 1; 34835d4f98a2SYan Zheng path->skip_locking = 1; 34845d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 34855d4f98a2SYan Zheng if (ret < 0) 34865d4f98a2SYan Zheng goto out; 34875d4f98a2SYan Zheng 3488aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3489aee68ee5SJosef Bacik if (path->slots[0]) { 3490aee68ee5SJosef Bacik path->slots[0]--; 3491aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3492aee68ee5SJosef Bacik path->slots[0]); 34933173a18fSJosef Bacik if (key.objectid == bytenr && 3494aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3495aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3496aee68ee5SJosef Bacik key.offset == blocksize))) 34973173a18fSJosef Bacik ret = 0; 34983173a18fSJosef Bacik } 3499aee68ee5SJosef Bacik 3500aee68ee5SJosef Bacik if (ret) { 3501aee68ee5SJosef Bacik skinny = false; 3502aee68ee5SJosef Bacik btrfs_release_path(path); 3503aee68ee5SJosef Bacik goto again; 3504aee68ee5SJosef Bacik } 3505aee68ee5SJosef Bacik } 3506cdccee99SLiu Bo if (ret) { 3507cdccee99SLiu Bo ASSERT(ret == 1); 3508cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3509cdccee99SLiu Bo btrfs_err(fs_info, 3510cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3511cdccee99SLiu Bo bytenr); 3512cdccee99SLiu Bo WARN_ON(1); 3513cdccee99SLiu Bo ret = -EINVAL; 3514cdccee99SLiu Bo goto out; 3515cdccee99SLiu Bo } 35163173a18fSJosef Bacik 35175d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 35185d4f98a2SYan Zheng out: 35195d4f98a2SYan Zheng btrfs_free_path(path); 35205d4f98a2SYan Zheng return ret; 35215d4f98a2SYan Zheng } 35225d4f98a2SYan Zheng 35235d4f98a2SYan Zheng /* 35245d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 35255d4f98a2SYan Zheng */ 35265d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 35275d4f98a2SYan Zheng struct extent_buffer *eb) 35285d4f98a2SYan Zheng { 35295d4f98a2SYan Zheng u64 flags; 35305d4f98a2SYan Zheng int ret; 35315d4f98a2SYan Zheng 35325d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 35335d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 35345d4f98a2SYan Zheng return 1; 35355d4f98a2SYan Zheng 35362ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info, 35373173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 35383173a18fSJosef Bacik NULL, &flags); 35395d4f98a2SYan Zheng BUG_ON(ret); 35405d4f98a2SYan Zheng 35415d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 35425d4f98a2SYan Zheng ret = 1; 35435d4f98a2SYan Zheng else 35445d4f98a2SYan Zheng ret = 0; 35455d4f98a2SYan Zheng return ret; 35465d4f98a2SYan Zheng } 35475d4f98a2SYan Zheng 35480af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 35491bbc621eSChris Mason struct btrfs_block_group_cache *block_group, 35501bbc621eSChris Mason struct inode *inode, 35511bbc621eSChris Mason u64 ino) 35520af3d00bSJosef Bacik { 35530af3d00bSJosef Bacik struct btrfs_key key; 35540af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 35550af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 35560af3d00bSJosef Bacik int ret = 0; 35570af3d00bSJosef Bacik 35580af3d00bSJosef Bacik if (inode) 35590af3d00bSJosef Bacik goto truncate; 35600af3d00bSJosef Bacik 35610af3d00bSJosef Bacik key.objectid = ino; 35620af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 35630af3d00bSJosef Bacik key.offset = 0; 35640af3d00bSJosef Bacik 35650af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 3566f54fb859STsutomu Itoh if (IS_ERR(inode) || is_bad_inode(inode)) { 3567f54fb859STsutomu Itoh if (!IS_ERR(inode)) 35680af3d00bSJosef Bacik iput(inode); 35690af3d00bSJosef Bacik return -ENOENT; 35700af3d00bSJosef Bacik } 35710af3d00bSJosef Bacik 35720af3d00bSJosef Bacik truncate: 35732ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 35747b61cd92SMiao Xie &fs_info->global_block_rsv); 35757b61cd92SMiao Xie if (ret) 35767b61cd92SMiao Xie goto out; 35777b61cd92SMiao Xie 35787a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 35790af3d00bSJosef Bacik if (IS_ERR(trans)) { 35803612b495STsutomu Itoh ret = PTR_ERR(trans); 35810af3d00bSJosef Bacik goto out; 35820af3d00bSJosef Bacik } 35830af3d00bSJosef Bacik 358477ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 35850af3d00bSJosef Bacik 35863a45bb20SJeff Mahoney btrfs_end_transaction(trans); 35872ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 35880af3d00bSJosef Bacik out: 35890af3d00bSJosef Bacik iput(inode); 35900af3d00bSJosef Bacik return ret; 35910af3d00bSJosef Bacik } 35920af3d00bSJosef Bacik 35935d4f98a2SYan Zheng /* 35945d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 35955d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 35965d4f98a2SYan Zheng */ 35975d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 35985d4f98a2SYan Zheng struct btrfs_key *extent_key, 35995d4f98a2SYan Zheng struct extent_buffer *leaf, 36005d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 36015d4f98a2SYan Zheng struct rb_root *blocks) 36025d4f98a2SYan Zheng { 36030b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 36045d4f98a2SYan Zheng struct btrfs_path *path; 36055d4f98a2SYan Zheng struct tree_block *block; 36065d4f98a2SYan Zheng struct btrfs_root *root; 36075d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 36085d4f98a2SYan Zheng struct rb_node *rb_node; 36095d4f98a2SYan Zheng struct btrfs_key key; 36105d4f98a2SYan Zheng u64 ref_root; 36115d4f98a2SYan Zheng u64 ref_objectid; 36125d4f98a2SYan Zheng u64 ref_offset; 36135d4f98a2SYan Zheng u32 ref_count; 36145d4f98a2SYan Zheng u32 nritems; 36155d4f98a2SYan Zheng int err = 0; 36165d4f98a2SYan Zheng int added = 0; 36175d4f98a2SYan Zheng int counted; 36185d4f98a2SYan Zheng int ret; 36195d4f98a2SYan Zheng 36205d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 36215d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 36225d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 36235d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 36245d4f98a2SYan Zheng 36250af3d00bSJosef Bacik /* 36260af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 36270af3d00bSJosef Bacik * it and redo the search. 36280af3d00bSJosef Bacik */ 36290af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 36300b246afaSJeff Mahoney ret = delete_block_group_cache(fs_info, rc->block_group, 36310af3d00bSJosef Bacik NULL, ref_objectid); 36320af3d00bSJosef Bacik if (ret != -ENOENT) 36330af3d00bSJosef Bacik return ret; 36340af3d00bSJosef Bacik ret = 0; 36350af3d00bSJosef Bacik } 36360af3d00bSJosef Bacik 36370af3d00bSJosef Bacik path = btrfs_alloc_path(); 36380af3d00bSJosef Bacik if (!path) 36390af3d00bSJosef Bacik return -ENOMEM; 3640e4058b54SDavid Sterba path->reada = READA_FORWARD; 36410af3d00bSJosef Bacik 36420b246afaSJeff Mahoney root = read_fs_root(fs_info, ref_root); 36435d4f98a2SYan Zheng if (IS_ERR(root)) { 36445d4f98a2SYan Zheng err = PTR_ERR(root); 36455d4f98a2SYan Zheng goto out; 36465d4f98a2SYan Zheng } 36475d4f98a2SYan Zheng 36485d4f98a2SYan Zheng key.objectid = ref_objectid; 36495d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 365084850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 365184850e8dSYan, Zheng key.offset = 0; 365284850e8dSYan, Zheng else 365384850e8dSYan, Zheng key.offset = ref_offset; 36545d4f98a2SYan Zheng 36555d4f98a2SYan Zheng path->search_commit_root = 1; 36565d4f98a2SYan Zheng path->skip_locking = 1; 36575d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 36585d4f98a2SYan Zheng if (ret < 0) { 36595d4f98a2SYan Zheng err = ret; 36605d4f98a2SYan Zheng goto out; 36615d4f98a2SYan Zheng } 36625d4f98a2SYan Zheng 36635d4f98a2SYan Zheng leaf = path->nodes[0]; 36645d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36655d4f98a2SYan Zheng /* 36665d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 36675d4f98a2SYan Zheng * are not counted in 36685d4f98a2SYan Zheng */ 36695d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36705d4f98a2SYan Zheng counted = 0; 36715d4f98a2SYan Zheng else 36725d4f98a2SYan Zheng counted = 1; 36735d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36745d4f98a2SYan Zheng if (rb_node) { 36755d4f98a2SYan Zheng if (counted) 36765d4f98a2SYan Zheng added = 1; 36775d4f98a2SYan Zheng else 36785d4f98a2SYan Zheng path->slots[0] = nritems; 36795d4f98a2SYan Zheng } 36805d4f98a2SYan Zheng 36815d4f98a2SYan Zheng while (ref_count > 0) { 36825d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 36835d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 36845d4f98a2SYan Zheng if (ret < 0) { 36855d4f98a2SYan Zheng err = ret; 36865d4f98a2SYan Zheng goto out; 36875d4f98a2SYan Zheng } 3688fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 36895d4f98a2SYan Zheng goto out; 36905d4f98a2SYan Zheng 36915d4f98a2SYan Zheng leaf = path->nodes[0]; 36925d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36935d4f98a2SYan Zheng added = 0; 36945d4f98a2SYan Zheng 36955d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36965d4f98a2SYan Zheng counted = 0; 36975d4f98a2SYan Zheng else 36985d4f98a2SYan Zheng counted = 1; 36995d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 37005d4f98a2SYan Zheng if (rb_node) { 37015d4f98a2SYan Zheng if (counted) 37025d4f98a2SYan Zheng added = 1; 37035d4f98a2SYan Zheng else 37045d4f98a2SYan Zheng path->slots[0] = nritems; 37055d4f98a2SYan Zheng } 37065d4f98a2SYan Zheng } 37075d4f98a2SYan Zheng 37085d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3709fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3710fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 37115d4f98a2SYan Zheng break; 37125d4f98a2SYan Zheng 37135d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 37145d4f98a2SYan Zheng struct btrfs_file_extent_item); 37155d4f98a2SYan Zheng 37165d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 37175d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 37185d4f98a2SYan Zheng goto next; 37195d4f98a2SYan Zheng 37205d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 37215d4f98a2SYan Zheng extent_key->objectid) 37225d4f98a2SYan Zheng goto next; 37235d4f98a2SYan Zheng 37245d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 37255d4f98a2SYan Zheng if (key.offset != ref_offset) 37265d4f98a2SYan Zheng goto next; 37275d4f98a2SYan Zheng 37285d4f98a2SYan Zheng if (counted) 37295d4f98a2SYan Zheng ref_count--; 37305d4f98a2SYan Zheng if (added) 37315d4f98a2SYan Zheng goto next; 37325d4f98a2SYan Zheng 37337476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 37345d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 37355d4f98a2SYan Zheng if (!block) { 37365d4f98a2SYan Zheng err = -ENOMEM; 37375d4f98a2SYan Zheng break; 37385d4f98a2SYan Zheng } 37395d4f98a2SYan Zheng block->bytenr = leaf->start; 37405d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 37415d4f98a2SYan Zheng block->level = 0; 37425d4f98a2SYan Zheng block->key_ready = 1; 37435d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 37445d4f98a2SYan Zheng &block->rb_node); 374543c04fb1SJeff Mahoney if (rb_node) 374643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 374743c04fb1SJeff Mahoney block->bytenr); 37485d4f98a2SYan Zheng } 37495d4f98a2SYan Zheng if (counted) 37505d4f98a2SYan Zheng added = 1; 37515d4f98a2SYan Zheng else 37525d4f98a2SYan Zheng path->slots[0] = nritems; 37535d4f98a2SYan Zheng next: 37545d4f98a2SYan Zheng path->slots[0]++; 37555d4f98a2SYan Zheng 37565d4f98a2SYan Zheng } 37575d4f98a2SYan Zheng out: 37585d4f98a2SYan Zheng btrfs_free_path(path); 37595d4f98a2SYan Zheng return err; 37605d4f98a2SYan Zheng } 37615d4f98a2SYan Zheng 37625d4f98a2SYan Zheng /* 37632c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37645d4f98a2SYan Zheng */ 37655d4f98a2SYan Zheng static noinline_for_stack 37665d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 37675d4f98a2SYan Zheng struct btrfs_key *extent_key, 37685d4f98a2SYan Zheng struct btrfs_path *path, 37695d4f98a2SYan Zheng struct rb_root *blocks) 37705d4f98a2SYan Zheng { 37715d4f98a2SYan Zheng struct btrfs_key key; 37725d4f98a2SYan Zheng struct extent_buffer *eb; 37735d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 37745d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 37755d4f98a2SYan Zheng unsigned long ptr; 37765d4f98a2SYan Zheng unsigned long end; 3777da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 3778647f63bdSFilipe David Borba Manana int ret = 0; 37795d4f98a2SYan Zheng int err = 0; 37805d4f98a2SYan Zheng 37815d4f98a2SYan Zheng eb = path->nodes[0]; 37825d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 37835d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 37845d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 37855d4f98a2SYan Zheng if (ptr + sizeof(struct btrfs_extent_item_v0) == end) 37865d4f98a2SYan Zheng ptr = end; 37875d4f98a2SYan Zheng else 37885d4f98a2SYan Zheng #endif 37895d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 37905d4f98a2SYan Zheng 37915d4f98a2SYan Zheng while (ptr < end) { 37925d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 37933de28d57SLiu Bo key.type = btrfs_get_extent_inline_ref_type(eb, iref, 37943de28d57SLiu Bo BTRFS_REF_TYPE_DATA); 37955d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37965d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 37975d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37985d4f98a2SYan Zheng blocks); 37995d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 38005d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 38015d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 38025d4f98a2SYan Zheng eb, dref, blocks); 38035d4f98a2SYan Zheng } else { 3804b14c55a1SLiu Bo ret = -EINVAL; 3805b14c55a1SLiu Bo btrfs_err(rc->extent_root->fs_info, 3806b14c55a1SLiu Bo "extent %llu slot %d has an invalid inline ref type", 3807b14c55a1SLiu Bo eb->start, path->slots[0]); 38085d4f98a2SYan Zheng } 3809647f63bdSFilipe David Borba Manana if (ret) { 3810647f63bdSFilipe David Borba Manana err = ret; 3811647f63bdSFilipe David Borba Manana goto out; 3812647f63bdSFilipe David Borba Manana } 38135d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 38145d4f98a2SYan Zheng } 38155d4f98a2SYan Zheng WARN_ON(ptr > end); 38165d4f98a2SYan Zheng 38175d4f98a2SYan Zheng while (1) { 38185d4f98a2SYan Zheng cond_resched(); 38195d4f98a2SYan Zheng eb = path->nodes[0]; 38205d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 38215d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38225d4f98a2SYan Zheng if (ret < 0) { 38235d4f98a2SYan Zheng err = ret; 38245d4f98a2SYan Zheng break; 38255d4f98a2SYan Zheng } 38265d4f98a2SYan Zheng if (ret > 0) 38275d4f98a2SYan Zheng break; 38285d4f98a2SYan Zheng eb = path->nodes[0]; 38295d4f98a2SYan Zheng } 38305d4f98a2SYan Zheng 38315d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 38325d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 38335d4f98a2SYan Zheng break; 38345d4f98a2SYan Zheng 38355d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 38365d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY || 38375d4f98a2SYan Zheng key.type == BTRFS_EXTENT_REF_V0_KEY) { 38385d4f98a2SYan Zheng #else 38395d4f98a2SYan Zheng BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); 38405d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 38415d4f98a2SYan Zheng #endif 38425d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 38435d4f98a2SYan Zheng blocks); 38445d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 38455d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 38465d4f98a2SYan Zheng struct btrfs_extent_data_ref); 38475d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 38485d4f98a2SYan Zheng eb, dref, blocks); 38495d4f98a2SYan Zheng } else { 38505d4f98a2SYan Zheng ret = 0; 38515d4f98a2SYan Zheng } 38525d4f98a2SYan Zheng if (ret) { 38535d4f98a2SYan Zheng err = ret; 38545d4f98a2SYan Zheng break; 38555d4f98a2SYan Zheng } 38565d4f98a2SYan Zheng path->slots[0]++; 38575d4f98a2SYan Zheng } 3858647f63bdSFilipe David Borba Manana out: 3859b3b4aa74SDavid Sterba btrfs_release_path(path); 38605d4f98a2SYan Zheng if (err) 38615d4f98a2SYan Zheng free_block_list(blocks); 38625d4f98a2SYan Zheng return err; 38635d4f98a2SYan Zheng } 38645d4f98a2SYan Zheng 38655d4f98a2SYan Zheng /* 38662c016dc2SLiu Bo * helper to find next unprocessed extent 38675d4f98a2SYan Zheng */ 38685d4f98a2SYan Zheng static noinline_for_stack 3869147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 38703fd0a558SYan, Zheng struct btrfs_key *extent_key) 38715d4f98a2SYan Zheng { 38720b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38735d4f98a2SYan Zheng struct btrfs_key key; 38745d4f98a2SYan Zheng struct extent_buffer *leaf; 38755d4f98a2SYan Zheng u64 start, end, last; 38765d4f98a2SYan Zheng int ret; 38775d4f98a2SYan Zheng 38785d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 38795d4f98a2SYan Zheng while (1) { 38805d4f98a2SYan Zheng cond_resched(); 38815d4f98a2SYan Zheng if (rc->search_start >= last) { 38825d4f98a2SYan Zheng ret = 1; 38835d4f98a2SYan Zheng break; 38845d4f98a2SYan Zheng } 38855d4f98a2SYan Zheng 38865d4f98a2SYan Zheng key.objectid = rc->search_start; 38875d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 38885d4f98a2SYan Zheng key.offset = 0; 38895d4f98a2SYan Zheng 38905d4f98a2SYan Zheng path->search_commit_root = 1; 38915d4f98a2SYan Zheng path->skip_locking = 1; 38925d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 38935d4f98a2SYan Zheng 0, 0); 38945d4f98a2SYan Zheng if (ret < 0) 38955d4f98a2SYan Zheng break; 38965d4f98a2SYan Zheng next: 38975d4f98a2SYan Zheng leaf = path->nodes[0]; 38985d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38995d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 39005d4f98a2SYan Zheng if (ret != 0) 39015d4f98a2SYan Zheng break; 39025d4f98a2SYan Zheng leaf = path->nodes[0]; 39035d4f98a2SYan Zheng } 39045d4f98a2SYan Zheng 39055d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 39065d4f98a2SYan Zheng if (key.objectid >= last) { 39075d4f98a2SYan Zheng ret = 1; 39085d4f98a2SYan Zheng break; 39095d4f98a2SYan Zheng } 39105d4f98a2SYan Zheng 39113173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 39123173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 39133173a18fSJosef Bacik path->slots[0]++; 39143173a18fSJosef Bacik goto next; 39153173a18fSJosef Bacik } 39163173a18fSJosef Bacik 39173173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 39185d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 39195d4f98a2SYan Zheng path->slots[0]++; 39205d4f98a2SYan Zheng goto next; 39215d4f98a2SYan Zheng } 39225d4f98a2SYan Zheng 39233173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 39240b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 39253173a18fSJosef Bacik rc->search_start) { 39263173a18fSJosef Bacik path->slots[0]++; 39273173a18fSJosef Bacik goto next; 39283173a18fSJosef Bacik } 39293173a18fSJosef Bacik 39305d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 39315d4f98a2SYan Zheng key.objectid, &start, &end, 3932e6138876SJosef Bacik EXTENT_DIRTY, NULL); 39335d4f98a2SYan Zheng 39345d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3935b3b4aa74SDavid Sterba btrfs_release_path(path); 39365d4f98a2SYan Zheng rc->search_start = end + 1; 39375d4f98a2SYan Zheng } else { 39383173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 39395d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 39403173a18fSJosef Bacik else 39413173a18fSJosef Bacik rc->search_start = key.objectid + 39420b246afaSJeff Mahoney fs_info->nodesize; 39433fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 39445d4f98a2SYan Zheng return 0; 39455d4f98a2SYan Zheng } 39465d4f98a2SYan Zheng } 3947b3b4aa74SDavid Sterba btrfs_release_path(path); 39485d4f98a2SYan Zheng return ret; 39495d4f98a2SYan Zheng } 39505d4f98a2SYan Zheng 39515d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 39525d4f98a2SYan Zheng { 39535d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39547585717fSChris Mason 39557585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39565d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 39577585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39585d4f98a2SYan Zheng } 39595d4f98a2SYan Zheng 39605d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 39615d4f98a2SYan Zheng { 39625d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39637585717fSChris Mason 39647585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39655d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 39667585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39675d4f98a2SYan Zheng } 39685d4f98a2SYan Zheng 39695d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 39705d4f98a2SYan Zheng { 39715d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39725d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39735d4f98a2SYan Zheng return 1; 39745d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 39755d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39765d4f98a2SYan Zheng return 1; 39775d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39785d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 39795d4f98a2SYan Zheng return 1; 39805d4f98a2SYan Zheng return 0; 39815d4f98a2SYan Zheng } 39825d4f98a2SYan Zheng 39833fd0a558SYan, Zheng static noinline_for_stack 39843fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 39853fd0a558SYan, Zheng { 39863fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3987ac2fabacSJosef Bacik int ret; 39883fd0a558SYan, Zheng 39892ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 399066d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 39913fd0a558SYan, Zheng if (!rc->block_rsv) 39923fd0a558SYan, Zheng return -ENOMEM; 39933fd0a558SYan, Zheng 39943fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 39953fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 39963fd0a558SYan, Zheng rc->extents_found = 0; 39973fd0a558SYan, Zheng rc->nodes_relocated = 0; 39983fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39990647bf56SWang Shilong rc->reserved_bytes = 0; 4000da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 40010647bf56SWang Shilong RELOCATION_RESERVED_NODES; 4002ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 4003ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 4004ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 4005ac2fabacSJosef Bacik if (ret) 4006ac2fabacSJosef Bacik return ret; 40073fd0a558SYan, Zheng 40083fd0a558SYan, Zheng rc->create_reloc_tree = 1; 40093fd0a558SYan, Zheng set_reloc_control(rc); 40103fd0a558SYan, Zheng 40117a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 401228818947SLiu Bo if (IS_ERR(trans)) { 401328818947SLiu Bo unset_reloc_control(rc); 401428818947SLiu Bo /* 401528818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 401628818947SLiu Bo * cleanup. And callers are responsible to free the above 401728818947SLiu Bo * block rsv. 401828818947SLiu Bo */ 401928818947SLiu Bo return PTR_ERR(trans); 402028818947SLiu Bo } 40213a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 40223fd0a558SYan, Zheng return 0; 40233fd0a558SYan, Zheng } 402476dda93cSYan, Zheng 40255d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 40265d4f98a2SYan Zheng { 40272ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40285d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 40295d4f98a2SYan Zheng struct btrfs_key key; 40305d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 40315d4f98a2SYan Zheng struct btrfs_path *path; 40325d4f98a2SYan Zheng struct btrfs_extent_item *ei; 40335d4f98a2SYan Zheng u64 flags; 40345d4f98a2SYan Zheng u32 item_size; 40355d4f98a2SYan Zheng int ret; 40365d4f98a2SYan Zheng int err = 0; 4037c87f08caSChris Mason int progress = 0; 40385d4f98a2SYan Zheng 40395d4f98a2SYan Zheng path = btrfs_alloc_path(); 40403fd0a558SYan, Zheng if (!path) 40415d4f98a2SYan Zheng return -ENOMEM; 4042e4058b54SDavid Sterba path->reada = READA_FORWARD; 40433fd0a558SYan, Zheng 40443fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 40453fd0a558SYan, Zheng if (ret) { 40463fd0a558SYan, Zheng err = ret; 40473fd0a558SYan, Zheng goto out_free; 40482423fdfbSJiri Slaby } 40495d4f98a2SYan Zheng 40505d4f98a2SYan Zheng while (1) { 40510647bf56SWang Shilong rc->reserved_bytes = 0; 40520647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40530647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40540647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40550647bf56SWang Shilong if (ret) { 40560647bf56SWang Shilong err = ret; 40570647bf56SWang Shilong break; 40580647bf56SWang Shilong } 4059c87f08caSChris Mason progress++; 4060a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40610f788c58SLiu Bo if (IS_ERR(trans)) { 40620f788c58SLiu Bo err = PTR_ERR(trans); 40630f788c58SLiu Bo trans = NULL; 40640f788c58SLiu Bo break; 40650f788c58SLiu Bo } 4066c87f08caSChris Mason restart: 40673fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40683a45bb20SJeff Mahoney btrfs_end_transaction(trans); 40693fd0a558SYan, Zheng continue; 40703fd0a558SYan, Zheng } 40713fd0a558SYan, Zheng 4072147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40735d4f98a2SYan Zheng if (ret < 0) 40745d4f98a2SYan Zheng err = ret; 40755d4f98a2SYan Zheng if (ret != 0) 40765d4f98a2SYan Zheng break; 40775d4f98a2SYan Zheng 40785d4f98a2SYan Zheng rc->extents_found++; 40795d4f98a2SYan Zheng 40805d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40815d4f98a2SYan Zheng struct btrfs_extent_item); 40823fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40835d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40845d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40855d4f98a2SYan Zheng ret = check_extent_flags(flags); 40865d4f98a2SYan Zheng BUG_ON(ret); 40875d4f98a2SYan Zheng 40885d4f98a2SYan Zheng } else { 40895d4f98a2SYan Zheng #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 40905d4f98a2SYan Zheng u64 ref_owner; 40915d4f98a2SYan Zheng int path_change = 0; 40925d4f98a2SYan Zheng 40935d4f98a2SYan Zheng BUG_ON(item_size != 40945d4f98a2SYan Zheng sizeof(struct btrfs_extent_item_v0)); 40955d4f98a2SYan Zheng ret = get_ref_objectid_v0(rc, path, &key, &ref_owner, 40965d4f98a2SYan Zheng &path_change); 40974b3576e4SZhaolei if (ret < 0) { 40984b3576e4SZhaolei err = ret; 40994b3576e4SZhaolei break; 41004b3576e4SZhaolei } 41015d4f98a2SYan Zheng if (ref_owner < BTRFS_FIRST_FREE_OBJECTID) 41025d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_TREE_BLOCK; 41035d4f98a2SYan Zheng else 41045d4f98a2SYan Zheng flags = BTRFS_EXTENT_FLAG_DATA; 41055d4f98a2SYan Zheng 41065d4f98a2SYan Zheng if (path_change) { 4107b3b4aa74SDavid Sterba btrfs_release_path(path); 41085d4f98a2SYan Zheng 41095d4f98a2SYan Zheng path->search_commit_root = 1; 41105d4f98a2SYan Zheng path->skip_locking = 1; 41115d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, 41125d4f98a2SYan Zheng &key, path, 0, 0); 41135d4f98a2SYan Zheng if (ret < 0) { 41145d4f98a2SYan Zheng err = ret; 41155d4f98a2SYan Zheng break; 41165d4f98a2SYan Zheng } 41175d4f98a2SYan Zheng BUG_ON(ret > 0); 41185d4f98a2SYan Zheng } 41195d4f98a2SYan Zheng #else 41205d4f98a2SYan Zheng BUG(); 41215d4f98a2SYan Zheng #endif 41225d4f98a2SYan Zheng } 41235d4f98a2SYan Zheng 41245d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 41255d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 41265d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 41275d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41285d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 41295d4f98a2SYan Zheng } else { 4130b3b4aa74SDavid Sterba btrfs_release_path(path); 41315d4f98a2SYan Zheng ret = 0; 41325d4f98a2SYan Zheng } 41335d4f98a2SYan Zheng if (ret < 0) { 41343fd0a558SYan, Zheng err = ret; 41355d4f98a2SYan Zheng break; 41365d4f98a2SYan Zheng } 41375d4f98a2SYan Zheng 41385d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 41395d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 41405d4f98a2SYan Zheng if (ret < 0) { 41411708cc57SWang Shilong /* 41421708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 41431708cc57SWang Shilong * backref cache when committing transaction. 41441708cc57SWang Shilong */ 41451708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 41461708cc57SWang Shilong 41473fd0a558SYan, Zheng if (ret != -EAGAIN) { 41485d4f98a2SYan Zheng err = ret; 41495d4f98a2SYan Zheng break; 41505d4f98a2SYan Zheng } 41513fd0a558SYan, Zheng rc->extents_found--; 41523fd0a558SYan, Zheng rc->search_start = key.objectid; 41533fd0a558SYan, Zheng } 41545d4f98a2SYan Zheng } 41555d4f98a2SYan Zheng 41563a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41572ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41583fd0a558SYan, Zheng trans = NULL; 41595d4f98a2SYan Zheng 41605d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 41615d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41625d4f98a2SYan Zheng rc->found_file_extent = 1; 41630257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 41643fd0a558SYan, Zheng &key, &rc->cluster); 41655d4f98a2SYan Zheng if (ret < 0) { 41665d4f98a2SYan Zheng err = ret; 41675d4f98a2SYan Zheng break; 41685d4f98a2SYan Zheng } 41695d4f98a2SYan Zheng } 41705d4f98a2SYan Zheng } 4171c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 41722ff7e61eSJeff Mahoney ret = btrfs_force_chunk_alloc(trans, fs_info, 4173c87f08caSChris Mason rc->block_group->flags); 41749689457bSShilong Wang if (ret == 1) { 4175c87f08caSChris Mason err = 0; 4176c87f08caSChris Mason progress = 0; 4177c87f08caSChris Mason goto restart; 4178c87f08caSChris Mason } 4179c87f08caSChris Mason } 41803fd0a558SYan, Zheng 4181b3b4aa74SDavid Sterba btrfs_release_path(path); 418291166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41835d4f98a2SYan Zheng 41845d4f98a2SYan Zheng if (trans) { 41853a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41862ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41875d4f98a2SYan Zheng } 41885d4f98a2SYan Zheng 41890257bb82SYan, Zheng if (!err) { 41903fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 41913fd0a558SYan, Zheng &rc->cluster); 41920257bb82SYan, Zheng if (ret < 0) 41930257bb82SYan, Zheng err = ret; 41940257bb82SYan, Zheng } 41950257bb82SYan, Zheng 41963fd0a558SYan, Zheng rc->create_reloc_tree = 0; 41973fd0a558SYan, Zheng set_reloc_control(rc); 41980257bb82SYan, Zheng 41993fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 42002ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 42015d4f98a2SYan Zheng 42023fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 42035d4f98a2SYan Zheng 42045d4f98a2SYan Zheng merge_reloc_roots(rc); 42055d4f98a2SYan Zheng 42063fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 42075d4f98a2SYan Zheng unset_reloc_control(rc); 42082ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 42095d4f98a2SYan Zheng 42105d4f98a2SYan Zheng /* get rid of pinned extents */ 42117a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 421262b99540SQu Wenruo if (IS_ERR(trans)) { 42133612b495STsutomu Itoh err = PTR_ERR(trans); 421462b99540SQu Wenruo goto out_free; 421562b99540SQu Wenruo } 42163a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 42173fd0a558SYan, Zheng out_free: 42182ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 42193fd0a558SYan, Zheng btrfs_free_path(path); 42205d4f98a2SYan Zheng return err; 42215d4f98a2SYan Zheng } 42225d4f98a2SYan Zheng 42235d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 42240257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 42255d4f98a2SYan Zheng { 42265d4f98a2SYan Zheng struct btrfs_path *path; 42275d4f98a2SYan Zheng struct btrfs_inode_item *item; 42285d4f98a2SYan Zheng struct extent_buffer *leaf; 42295d4f98a2SYan Zheng int ret; 42305d4f98a2SYan Zheng 42315d4f98a2SYan Zheng path = btrfs_alloc_path(); 42325d4f98a2SYan Zheng if (!path) 42335d4f98a2SYan Zheng return -ENOMEM; 42345d4f98a2SYan Zheng 42355d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 42365d4f98a2SYan Zheng if (ret) 42375d4f98a2SYan Zheng goto out; 42385d4f98a2SYan Zheng 42395d4f98a2SYan Zheng leaf = path->nodes[0]; 42405d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4241b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 42425d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 42430257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 42445d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 42453fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 42463fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 42475d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 42485d4f98a2SYan Zheng out: 42495d4f98a2SYan Zheng btrfs_free_path(path); 42505d4f98a2SYan Zheng return ret; 42515d4f98a2SYan Zheng } 42525d4f98a2SYan Zheng 42535d4f98a2SYan Zheng /* 42545d4f98a2SYan Zheng * helper to create inode for data relocation. 42555d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 42565d4f98a2SYan Zheng */ 42573fd0a558SYan, Zheng static noinline_for_stack 42583fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 42595d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 42605d4f98a2SYan Zheng { 42615d4f98a2SYan Zheng struct inode *inode = NULL; 42625d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42635d4f98a2SYan Zheng struct btrfs_root *root; 42645d4f98a2SYan Zheng struct btrfs_key key; 42654624900dSZhaolei u64 objectid; 42665d4f98a2SYan Zheng int err = 0; 42675d4f98a2SYan Zheng 42685d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 42695d4f98a2SYan Zheng if (IS_ERR(root)) 42705d4f98a2SYan Zheng return ERR_CAST(root); 42715d4f98a2SYan Zheng 4272a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 42733fd0a558SYan, Zheng if (IS_ERR(trans)) 42743fd0a558SYan, Zheng return ERR_CAST(trans); 42755d4f98a2SYan Zheng 4276581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42775d4f98a2SYan Zheng if (err) 42785d4f98a2SYan Zheng goto out; 42795d4f98a2SYan Zheng 42800257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42815d4f98a2SYan Zheng BUG_ON(err); 42825d4f98a2SYan Zheng 42835d4f98a2SYan Zheng key.objectid = objectid; 42845d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 42855d4f98a2SYan Zheng key.offset = 0; 42860b246afaSJeff Mahoney inode = btrfs_iget(fs_info->sb, &key, root, NULL); 42875d4f98a2SYan Zheng BUG_ON(IS_ERR(inode) || is_bad_inode(inode)); 42885d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 42895d4f98a2SYan Zheng 429073f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 42915d4f98a2SYan Zheng out: 42923a45bb20SJeff Mahoney btrfs_end_transaction(trans); 42932ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42945d4f98a2SYan Zheng if (err) { 42955d4f98a2SYan Zheng if (inode) 42965d4f98a2SYan Zheng iput(inode); 42975d4f98a2SYan Zheng inode = ERR_PTR(err); 42985d4f98a2SYan Zheng } 42995d4f98a2SYan Zheng return inode; 43005d4f98a2SYan Zheng } 43015d4f98a2SYan Zheng 4302a9995eecSJosef Bacik static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 43033fd0a558SYan, Zheng { 43043fd0a558SYan, Zheng struct reloc_control *rc; 43053fd0a558SYan, Zheng 43063fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 43073fd0a558SYan, Zheng if (!rc) 43083fd0a558SYan, Zheng return NULL; 43093fd0a558SYan, Zheng 43103fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 43113fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 43123fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 4313c6100a4bSJosef Bacik extent_io_tree_init(&rc->processed_blocks, NULL); 43143fd0a558SYan, Zheng return rc; 43153fd0a558SYan, Zheng } 43163fd0a558SYan, Zheng 43175d4f98a2SYan Zheng /* 4318ebce0e01SAdam Borowski * Print the block group being relocated 4319ebce0e01SAdam Borowski */ 4320ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 4321ebce0e01SAdam Borowski struct btrfs_block_group_cache *block_group) 4322ebce0e01SAdam Borowski { 4323ebce0e01SAdam Borowski char buf[128]; /* prefixed by a '|' that'll be dropped */ 4324ebce0e01SAdam Borowski u64 flags = block_group->flags; 4325ebce0e01SAdam Borowski 4326ebce0e01SAdam Borowski /* Shouldn't happen */ 4327ebce0e01SAdam Borowski if (!flags) { 4328ebce0e01SAdam Borowski strcpy(buf, "|NONE"); 4329ebce0e01SAdam Borowski } else { 4330ebce0e01SAdam Borowski char *bp = buf; 4331ebce0e01SAdam Borowski 4332ebce0e01SAdam Borowski #define DESCRIBE_FLAG(f, d) \ 4333ebce0e01SAdam Borowski if (flags & BTRFS_BLOCK_GROUP_##f) { \ 4334ebce0e01SAdam Borowski bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \ 4335ebce0e01SAdam Borowski flags &= ~BTRFS_BLOCK_GROUP_##f; \ 4336ebce0e01SAdam Borowski } 4337ebce0e01SAdam Borowski DESCRIBE_FLAG(DATA, "data"); 4338ebce0e01SAdam Borowski DESCRIBE_FLAG(SYSTEM, "system"); 4339ebce0e01SAdam Borowski DESCRIBE_FLAG(METADATA, "metadata"); 4340ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID0, "raid0"); 4341ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID1, "raid1"); 4342ebce0e01SAdam Borowski DESCRIBE_FLAG(DUP, "dup"); 4343ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID10, "raid10"); 4344ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID5, "raid5"); 4345ebce0e01SAdam Borowski DESCRIBE_FLAG(RAID6, "raid6"); 4346ebce0e01SAdam Borowski if (flags) 4347ebce0e01SAdam Borowski snprintf(buf, buf - bp + sizeof(buf), "|0x%llx", flags); 4348ebce0e01SAdam Borowski #undef DESCRIBE_FLAG 4349ebce0e01SAdam Borowski } 4350ebce0e01SAdam Borowski 4351ebce0e01SAdam Borowski btrfs_info(fs_info, 4352ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4353ebce0e01SAdam Borowski block_group->key.objectid, buf + 1); 4354ebce0e01SAdam Borowski } 4355ebce0e01SAdam Borowski 4356ebce0e01SAdam Borowski /* 43575d4f98a2SYan Zheng * function to relocate all extents in a block group. 43585d4f98a2SYan Zheng */ 43596bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 43605d4f98a2SYan Zheng { 43616bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 43625d4f98a2SYan Zheng struct reloc_control *rc; 43630af3d00bSJosef Bacik struct inode *inode; 43640af3d00bSJosef Bacik struct btrfs_path *path; 43655d4f98a2SYan Zheng int ret; 4366f0486c68SYan, Zheng int rw = 0; 43675d4f98a2SYan Zheng int err = 0; 43685d4f98a2SYan Zheng 4369a9995eecSJosef Bacik rc = alloc_reloc_control(fs_info); 43705d4f98a2SYan Zheng if (!rc) 43715d4f98a2SYan Zheng return -ENOMEM; 43725d4f98a2SYan Zheng 4373f0486c68SYan, Zheng rc->extent_root = extent_root; 43743fd0a558SYan, Zheng 43755d4f98a2SYan Zheng rc->block_group = btrfs_lookup_block_group(fs_info, group_start); 43765d4f98a2SYan Zheng BUG_ON(!rc->block_group); 43775d4f98a2SYan Zheng 43785e00f193SJeff Mahoney ret = btrfs_inc_block_group_ro(fs_info, rc->block_group); 4379f0486c68SYan, Zheng if (ret) { 4380f0486c68SYan, Zheng err = ret; 4381f0486c68SYan, Zheng goto out; 4382f0486c68SYan, Zheng } 4383f0486c68SYan, Zheng rw = 1; 4384f0486c68SYan, Zheng 43850af3d00bSJosef Bacik path = btrfs_alloc_path(); 43860af3d00bSJosef Bacik if (!path) { 43870af3d00bSJosef Bacik err = -ENOMEM; 43880af3d00bSJosef Bacik goto out; 43890af3d00bSJosef Bacik } 43900af3d00bSJosef Bacik 439177ab86bfSJeff Mahoney inode = lookup_free_space_inode(fs_info, rc->block_group, path); 43920af3d00bSJosef Bacik btrfs_free_path(path); 43930af3d00bSJosef Bacik 43940af3d00bSJosef Bacik if (!IS_ERR(inode)) 43951bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43960af3d00bSJosef Bacik else 43970af3d00bSJosef Bacik ret = PTR_ERR(inode); 43980af3d00bSJosef Bacik 43990af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 44000af3d00bSJosef Bacik err = ret; 44010af3d00bSJosef Bacik goto out; 44020af3d00bSJosef Bacik } 44030af3d00bSJosef Bacik 44045d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 44055d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 44065d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 44075d4f98a2SYan Zheng rc->data_inode = NULL; 44085d4f98a2SYan Zheng goto out; 44095d4f98a2SYan Zheng } 44105d4f98a2SYan Zheng 44110b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 44125d4f98a2SYan Zheng 44139cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4414f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 44156374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4416578def7cSFilipe Manana rc->block_group->key.objectid, 4417578def7cSFilipe Manana rc->block_group->key.offset); 44185d4f98a2SYan Zheng 44195d4f98a2SYan Zheng while (1) { 442076dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 44215d4f98a2SYan Zheng ret = relocate_block_group(rc); 442276dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 44235d4f98a2SYan Zheng if (ret < 0) { 44245d4f98a2SYan Zheng err = ret; 44253fd0a558SYan, Zheng goto out; 44265d4f98a2SYan Zheng } 44275d4f98a2SYan Zheng 44285d4f98a2SYan Zheng if (rc->extents_found == 0) 44295d4f98a2SYan Zheng break; 44305d4f98a2SYan Zheng 44310b246afaSJeff Mahoney btrfs_info(fs_info, "found %llu extents", rc->extents_found); 44325d4f98a2SYan Zheng 44335d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 44340ef8b726SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 44350ef8b726SJosef Bacik (u64)-1); 44360ef8b726SJosef Bacik if (ret) { 44370ef8b726SJosef Bacik err = ret; 44380ef8b726SJosef Bacik goto out; 44390ef8b726SJosef Bacik } 44405d4f98a2SYan Zheng invalidate_mapping_pages(rc->data_inode->i_mapping, 44415d4f98a2SYan Zheng 0, -1); 44425d4f98a2SYan Zheng rc->stage = UPDATE_DATA_PTRS; 44435d4f98a2SYan Zheng } 44445d4f98a2SYan Zheng } 44455d4f98a2SYan Zheng 44465d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 44475d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 44485d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 44495d4f98a2SYan Zheng out: 4450f0486c68SYan, Zheng if (err && rw) 44512ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 44525d4f98a2SYan Zheng iput(rc->data_inode); 44535d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 44545d4f98a2SYan Zheng kfree(rc); 44555d4f98a2SYan Zheng return err; 44565d4f98a2SYan Zheng } 44575d4f98a2SYan Zheng 445876dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 445976dda93cSYan, Zheng { 44600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 446176dda93cSYan, Zheng struct btrfs_trans_handle *trans; 446279787eaaSJeff Mahoney int ret, err; 446376dda93cSYan, Zheng 44640b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 446579787eaaSJeff Mahoney if (IS_ERR(trans)) 446679787eaaSJeff Mahoney return PTR_ERR(trans); 446776dda93cSYan, Zheng 446876dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 446976dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 447076dda93cSYan, Zheng root->root_item.drop_level = 0; 447176dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 44720b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 447376dda93cSYan, Zheng &root->root_key, &root->root_item); 447476dda93cSYan, Zheng 44753a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 447679787eaaSJeff Mahoney if (err) 447779787eaaSJeff Mahoney return err; 447879787eaaSJeff Mahoney return ret; 447976dda93cSYan, Zheng } 448076dda93cSYan, Zheng 44815d4f98a2SYan Zheng /* 44825d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44835d4f98a2SYan Zheng * 44845d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44855d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44865d4f98a2SYan Zheng */ 44875d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44885d4f98a2SYan Zheng { 44890b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 44905d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 44915d4f98a2SYan Zheng struct btrfs_key key; 44925d4f98a2SYan Zheng struct btrfs_root *fs_root; 44935d4f98a2SYan Zheng struct btrfs_root *reloc_root; 44945d4f98a2SYan Zheng struct btrfs_path *path; 44955d4f98a2SYan Zheng struct extent_buffer *leaf; 44965d4f98a2SYan Zheng struct reloc_control *rc = NULL; 44975d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 44985d4f98a2SYan Zheng int ret; 44995d4f98a2SYan Zheng int err = 0; 45005d4f98a2SYan Zheng 45015d4f98a2SYan Zheng path = btrfs_alloc_path(); 45025d4f98a2SYan Zheng if (!path) 45035d4f98a2SYan Zheng return -ENOMEM; 4504e4058b54SDavid Sterba path->reada = READA_BACK; 45055d4f98a2SYan Zheng 45065d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 45075d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 45085d4f98a2SYan Zheng key.offset = (u64)-1; 45095d4f98a2SYan Zheng 45105d4f98a2SYan Zheng while (1) { 45110b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 45125d4f98a2SYan Zheng path, 0, 0); 45135d4f98a2SYan Zheng if (ret < 0) { 45145d4f98a2SYan Zheng err = ret; 45155d4f98a2SYan Zheng goto out; 45165d4f98a2SYan Zheng } 45175d4f98a2SYan Zheng if (ret > 0) { 45185d4f98a2SYan Zheng if (path->slots[0] == 0) 45195d4f98a2SYan Zheng break; 45205d4f98a2SYan Zheng path->slots[0]--; 45215d4f98a2SYan Zheng } 45225d4f98a2SYan Zheng leaf = path->nodes[0]; 45235d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4524b3b4aa74SDavid Sterba btrfs_release_path(path); 45255d4f98a2SYan Zheng 45265d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 45275d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 45285d4f98a2SYan Zheng break; 45295d4f98a2SYan Zheng 4530cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 45315d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 45325d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 45335d4f98a2SYan Zheng goto out; 45345d4f98a2SYan Zheng } 45355d4f98a2SYan Zheng 45365d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 45375d4f98a2SYan Zheng 45385d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 45390b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 45405d4f98a2SYan Zheng reloc_root->root_key.offset); 45415d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 454276dda93cSYan, Zheng ret = PTR_ERR(fs_root); 454376dda93cSYan, Zheng if (ret != -ENOENT) { 454476dda93cSYan, Zheng err = ret; 45455d4f98a2SYan Zheng goto out; 45465d4f98a2SYan Zheng } 454779787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 454879787eaaSJeff Mahoney if (ret < 0) { 454979787eaaSJeff Mahoney err = ret; 455079787eaaSJeff Mahoney goto out; 455179787eaaSJeff Mahoney } 455276dda93cSYan, Zheng } 45535d4f98a2SYan Zheng } 45545d4f98a2SYan Zheng 45555d4f98a2SYan Zheng if (key.offset == 0) 45565d4f98a2SYan Zheng break; 45575d4f98a2SYan Zheng 45585d4f98a2SYan Zheng key.offset--; 45595d4f98a2SYan Zheng } 4560b3b4aa74SDavid Sterba btrfs_release_path(path); 45615d4f98a2SYan Zheng 45625d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 45635d4f98a2SYan Zheng goto out; 45645d4f98a2SYan Zheng 45650b246afaSJeff Mahoney rc = alloc_reloc_control(fs_info); 45665d4f98a2SYan Zheng if (!rc) { 45675d4f98a2SYan Zheng err = -ENOMEM; 45685d4f98a2SYan Zheng goto out; 45695d4f98a2SYan Zheng } 45705d4f98a2SYan Zheng 45710b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 45725d4f98a2SYan Zheng 45735d4f98a2SYan Zheng set_reloc_control(rc); 45745d4f98a2SYan Zheng 45757a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45763612b495STsutomu Itoh if (IS_ERR(trans)) { 45773612b495STsutomu Itoh unset_reloc_control(rc); 45783612b495STsutomu Itoh err = PTR_ERR(trans); 45793612b495STsutomu Itoh goto out_free; 45803612b495STsutomu Itoh } 45813fd0a558SYan, Zheng 45823fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45833fd0a558SYan, Zheng 45845d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45855d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45865d4f98a2SYan Zheng struct btrfs_root, root_list); 45875d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45885d4f98a2SYan Zheng 45895d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 45905d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 45915d4f98a2SYan Zheng &rc->reloc_roots); 45925d4f98a2SYan Zheng continue; 45935d4f98a2SYan Zheng } 45945d4f98a2SYan Zheng 45950b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 459679787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 459779787eaaSJeff Mahoney err = PTR_ERR(fs_root); 459879787eaaSJeff Mahoney goto out_free; 459979787eaaSJeff Mahoney } 46005d4f98a2SYan Zheng 4601ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 460279787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 46035d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 46045d4f98a2SYan Zheng } 46055d4f98a2SYan Zheng 46063a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 460779787eaaSJeff Mahoney if (err) 460879787eaaSJeff Mahoney goto out_free; 46095d4f98a2SYan Zheng 46105d4f98a2SYan Zheng merge_reloc_roots(rc); 46115d4f98a2SYan Zheng 46125d4f98a2SYan Zheng unset_reloc_control(rc); 46135d4f98a2SYan Zheng 46147a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 461562b99540SQu Wenruo if (IS_ERR(trans)) { 46163612b495STsutomu Itoh err = PTR_ERR(trans); 461762b99540SQu Wenruo goto out_free; 461862b99540SQu Wenruo } 46193a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 46203612b495STsutomu Itoh out_free: 46215d4f98a2SYan Zheng kfree(rc); 46223612b495STsutomu Itoh out: 4623aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4624aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4625aca1bba6SLiu Bo 46265d4f98a2SYan Zheng btrfs_free_path(path); 46275d4f98a2SYan Zheng 46285d4f98a2SYan Zheng if (err == 0) { 46295d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 46300b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 46315d4f98a2SYan Zheng if (IS_ERR(fs_root)) 46325d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4633d7ce5843SMiao Xie else 463466b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 46355d4f98a2SYan Zheng } 46365d4f98a2SYan Zheng return err; 46375d4f98a2SYan Zheng } 46385d4f98a2SYan Zheng 46395d4f98a2SYan Zheng /* 46405d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 46415d4f98a2SYan Zheng * 46425d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 46435d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 46445d4f98a2SYan Zheng */ 46455d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 46465d4f98a2SYan Zheng { 46470b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 46485d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 46495d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 46505d4f98a2SYan Zheng int ret; 46515d4f98a2SYan Zheng u64 disk_bytenr; 46524577b014SJosef Bacik u64 new_bytenr; 46535d4f98a2SYan Zheng LIST_HEAD(list); 46545d4f98a2SYan Zheng 46555d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 46565d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 46575d4f98a2SYan Zheng 46585d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46590b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4660a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 466179787eaaSJeff Mahoney if (ret) 466279787eaaSJeff Mahoney goto out; 46635d4f98a2SYan Zheng 46645d4f98a2SYan Zheng while (!list_empty(&list)) { 46655d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46665d4f98a2SYan Zheng list_del_init(&sums->list); 46675d4f98a2SYan Zheng 46684577b014SJosef Bacik /* 46694577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46704577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46714577b014SJosef Bacik * extents but we may have written to say the middle of the 46724577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46734577b014SJosef Bacik * the right disk offset. 46744577b014SJosef Bacik * 46754577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46764577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46774577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46784577b014SJosef Bacik * disk length. 46794577b014SJosef Bacik */ 46804577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 46814577b014SJosef Bacik sums->bytenr = new_bytenr; 46825d4f98a2SYan Zheng 46835d4f98a2SYan Zheng btrfs_add_ordered_sum(inode, ordered, sums); 46845d4f98a2SYan Zheng } 468579787eaaSJeff Mahoney out: 46865d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4687411fc6bcSAndi Kleen return ret; 46885d4f98a2SYan Zheng } 46893fd0a558SYan, Zheng 469083d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 46913fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 46923fd0a558SYan, Zheng struct extent_buffer *cow) 46933fd0a558SYan, Zheng { 46940b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46953fd0a558SYan, Zheng struct reloc_control *rc; 46963fd0a558SYan, Zheng struct backref_node *node; 46973fd0a558SYan, Zheng int first_cow = 0; 46983fd0a558SYan, Zheng int level; 469983d4cfd4SJosef Bacik int ret = 0; 47003fd0a558SYan, Zheng 47010b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 47023fd0a558SYan, Zheng if (!rc) 470383d4cfd4SJosef Bacik return 0; 47043fd0a558SYan, Zheng 47053fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 47063fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 47073fd0a558SYan, Zheng 4708c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4709c974c464SWang Shilong if (buf == root->node) 4710c974c464SWang Shilong __update_reloc_root(root, cow->start); 4711c974c464SWang Shilong } 4712c974c464SWang Shilong 47133fd0a558SYan, Zheng level = btrfs_header_level(buf); 47143fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 47153fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 47163fd0a558SYan, Zheng first_cow = 1; 47173fd0a558SYan, Zheng 47183fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 47193fd0a558SYan, Zheng rc->create_reloc_tree) { 47203fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 47213fd0a558SYan, Zheng 47223fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 47233fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 47243fd0a558SYan, Zheng node->new_bytenr != buf->start); 47253fd0a558SYan, Zheng 47263fd0a558SYan, Zheng drop_node_buffer(node); 47273fd0a558SYan, Zheng extent_buffer_get(cow); 47283fd0a558SYan, Zheng node->eb = cow; 47293fd0a558SYan, Zheng node->new_bytenr = cow->start; 47303fd0a558SYan, Zheng 47313fd0a558SYan, Zheng if (!node->pending) { 47323fd0a558SYan, Zheng list_move_tail(&node->list, 47333fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 47343fd0a558SYan, Zheng node->pending = 1; 47353fd0a558SYan, Zheng } 47363fd0a558SYan, Zheng 47373fd0a558SYan, Zheng if (first_cow) 47383fd0a558SYan, Zheng __mark_block_processed(rc, node); 47393fd0a558SYan, Zheng 47403fd0a558SYan, Zheng if (first_cow && level > 0) 47413fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 47423fd0a558SYan, Zheng } 47433fd0a558SYan, Zheng 474483d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 47453fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 474683d4cfd4SJosef Bacik return ret; 47473fd0a558SYan, Zheng } 47483fd0a558SYan, Zheng 47493fd0a558SYan, Zheng /* 47503fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 475101327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 47523fd0a558SYan, Zheng */ 4753147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 47543fd0a558SYan, Zheng u64 *bytes_to_reserve) 47553fd0a558SYan, Zheng { 47563fd0a558SYan, Zheng struct btrfs_root *root; 47573fd0a558SYan, Zheng struct reloc_control *rc; 47583fd0a558SYan, Zheng 47593fd0a558SYan, Zheng root = pending->root; 47603fd0a558SYan, Zheng if (!root->reloc_root) 47613fd0a558SYan, Zheng return; 47623fd0a558SYan, Zheng 47633fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47643fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47653fd0a558SYan, Zheng return; 47663fd0a558SYan, Zheng 47673fd0a558SYan, Zheng root = root->reloc_root; 47683fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47693fd0a558SYan, Zheng /* 47703fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47713fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47723fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47733fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47743fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47753fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 47763fd0a558SYan, Zheng * tree may use more space than it frees. so we need 47773fd0a558SYan, Zheng * reserve extra space. 47783fd0a558SYan, Zheng */ 47793fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47803fd0a558SYan, Zheng } 47813fd0a558SYan, Zheng 47823fd0a558SYan, Zheng /* 47833fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47843fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 47853fd0a558SYan, Zheng */ 478649b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 47873fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 47883fd0a558SYan, Zheng { 47893fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 47903fd0a558SYan, Zheng struct btrfs_root *reloc_root; 47913fd0a558SYan, Zheng struct btrfs_root *new_root; 47923fd0a558SYan, Zheng struct reloc_control *rc; 47933fd0a558SYan, Zheng int ret; 47943fd0a558SYan, Zheng 47953fd0a558SYan, Zheng if (!root->reloc_root) 479649b25e05SJeff Mahoney return 0; 47973fd0a558SYan, Zheng 47983fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47993fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 48003fd0a558SYan, Zheng 48013fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 48023fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 48033fd0a558SYan, Zheng rc->block_rsv, 480425d609f8SJosef Bacik rc->nodes_relocated, 1); 480549b25e05SJeff Mahoney if (ret) 480649b25e05SJeff Mahoney return ret; 48073fd0a558SYan, Zheng } 48083fd0a558SYan, Zheng 48093fd0a558SYan, Zheng new_root = pending->snap; 48103fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 48113fd0a558SYan, Zheng new_root->root_key.objectid); 481249b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 481349b25e05SJeff Mahoney return PTR_ERR(reloc_root); 48143fd0a558SYan, Zheng 4815ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4816ffd7b339SJeff Mahoney BUG_ON(ret < 0); 48173fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 48183fd0a558SYan, Zheng 481949b25e05SJeff Mahoney if (rc->create_reloc_tree) 48203fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 482149b25e05SJeff Mahoney return ret; 48223fd0a558SYan, Zheng } 4823