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> 12726a3421SQu Wenruo #include <linux/error-injection.h> 135d4f98a2SYan Zheng #include "ctree.h" 145d4f98a2SYan Zheng #include "disk-io.h" 155d4f98a2SYan Zheng #include "transaction.h" 165d4f98a2SYan Zheng #include "volumes.h" 175d4f98a2SYan Zheng #include "locking.h" 185d4f98a2SYan Zheng #include "btrfs_inode.h" 195d4f98a2SYan Zheng #include "async-thread.h" 200af3d00bSJosef Bacik #include "free-space-cache.h" 2162b99540SQu Wenruo #include "qgroup.h" 22cdccee99SLiu Bo #include "print-tree.h" 2386736342SJosef Bacik #include "delalloc-space.h" 24aac0023cSJosef Bacik #include "block-group.h" 2519b546d7SQu Wenruo #include "backref.h" 26e9a28dc5SQu Wenruo #include "misc.h" 27c2832898SQu Wenruo #include "subpage.h" 287ae9bd18SNaohiro Aota #include "zoned.h" 2926c2c454SJosef Bacik #include "inode-item.h" 305d4f98a2SYan Zheng 315d4f98a2SYan Zheng /* 320c891389SQu Wenruo * Relocation overview 330c891389SQu Wenruo * 340c891389SQu Wenruo * [What does relocation do] 350c891389SQu Wenruo * 360c891389SQu Wenruo * The objective of relocation is to relocate all extents of the target block 370c891389SQu Wenruo * group to other block groups. 380c891389SQu Wenruo * This is utilized by resize (shrink only), profile converting, compacting 390c891389SQu Wenruo * space, or balance routine to spread chunks over devices. 400c891389SQu Wenruo * 410c891389SQu Wenruo * Before | After 420c891389SQu Wenruo * ------------------------------------------------------------------ 430c891389SQu Wenruo * BG A: 10 data extents | BG A: deleted 440c891389SQu Wenruo * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) 450c891389SQu Wenruo * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) 460c891389SQu Wenruo * 470c891389SQu Wenruo * [How does relocation work] 480c891389SQu Wenruo * 490c891389SQu Wenruo * 1. Mark the target block group read-only 500c891389SQu Wenruo * New extents won't be allocated from the target block group. 510c891389SQu Wenruo * 520c891389SQu Wenruo * 2.1 Record each extent in the target block group 530c891389SQu Wenruo * To build a proper map of extents to be relocated. 540c891389SQu Wenruo * 550c891389SQu Wenruo * 2.2 Build data reloc tree and reloc trees 560c891389SQu Wenruo * Data reloc tree will contain an inode, recording all newly relocated 570c891389SQu Wenruo * data extents. 580c891389SQu Wenruo * There will be only one data reloc tree for one data block group. 590c891389SQu Wenruo * 600c891389SQu Wenruo * Reloc tree will be a special snapshot of its source tree, containing 610c891389SQu Wenruo * relocated tree blocks. 620c891389SQu Wenruo * Each tree referring to a tree block in target block group will get its 630c891389SQu Wenruo * reloc tree built. 640c891389SQu Wenruo * 650c891389SQu Wenruo * 2.3 Swap source tree with its corresponding reloc tree 660c891389SQu Wenruo * Each involved tree only refers to new extents after swap. 670c891389SQu Wenruo * 680c891389SQu Wenruo * 3. Cleanup reloc trees and data reloc tree. 690c891389SQu Wenruo * As old extents in the target block group are still referenced by reloc 700c891389SQu Wenruo * trees, we need to clean them up before really freeing the target block 710c891389SQu Wenruo * group. 720c891389SQu Wenruo * 730c891389SQu Wenruo * The main complexity is in steps 2.2 and 2.3. 740c891389SQu Wenruo * 750c891389SQu Wenruo * The entry point of relocation is relocate_block_group() function. 760c891389SQu Wenruo */ 770c891389SQu Wenruo 780647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 792a979612SQu Wenruo /* 805d4f98a2SYan Zheng * map address of tree root to tree 815d4f98a2SYan Zheng */ 825d4f98a2SYan Zheng struct mapping_node { 83e9a28dc5SQu Wenruo struct { 845d4f98a2SYan Zheng struct rb_node rb_node; 855d4f98a2SYan Zheng u64 bytenr; 86e9a28dc5SQu Wenruo }; /* Use rb_simle_node for search/insert */ 875d4f98a2SYan Zheng void *data; 885d4f98a2SYan Zheng }; 895d4f98a2SYan Zheng 905d4f98a2SYan Zheng struct mapping_tree { 915d4f98a2SYan Zheng struct rb_root rb_root; 925d4f98a2SYan Zheng spinlock_t lock; 935d4f98a2SYan Zheng }; 945d4f98a2SYan Zheng 955d4f98a2SYan Zheng /* 965d4f98a2SYan Zheng * present a tree block to process 975d4f98a2SYan Zheng */ 985d4f98a2SYan Zheng struct tree_block { 99e9a28dc5SQu Wenruo struct { 1005d4f98a2SYan Zheng struct rb_node rb_node; 1015d4f98a2SYan Zheng u64 bytenr; 102e9a28dc5SQu Wenruo }; /* Use rb_simple_node for search/insert */ 103f7ba2d37SJosef Bacik u64 owner; 1045d4f98a2SYan Zheng struct btrfs_key key; 1055d4f98a2SYan Zheng unsigned int level:8; 1065d4f98a2SYan Zheng unsigned int key_ready:1; 1075d4f98a2SYan Zheng }; 1085d4f98a2SYan Zheng 1090257bb82SYan, Zheng #define MAX_EXTENTS 128 1100257bb82SYan, Zheng 1110257bb82SYan, Zheng struct file_extent_cluster { 1120257bb82SYan, Zheng u64 start; 1130257bb82SYan, Zheng u64 end; 1140257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1150257bb82SYan, Zheng unsigned int nr; 1160257bb82SYan, Zheng }; 1170257bb82SYan, Zheng 1185d4f98a2SYan Zheng struct reloc_control { 1195d4f98a2SYan Zheng /* block group to relocate */ 12032da5386SDavid Sterba struct btrfs_block_group *block_group; 1215d4f98a2SYan Zheng /* extent tree */ 1225d4f98a2SYan Zheng struct btrfs_root *extent_root; 1235d4f98a2SYan Zheng /* inode for moving data */ 1245d4f98a2SYan Zheng struct inode *data_inode; 1253fd0a558SYan, Zheng 1263fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1273fd0a558SYan, Zheng 128a26195a5SQu Wenruo struct btrfs_backref_cache backref_cache; 1293fd0a558SYan, Zheng 1303fd0a558SYan, Zheng struct file_extent_cluster cluster; 1315d4f98a2SYan Zheng /* tree blocks have been processed */ 1325d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1335d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1345d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1355d4f98a2SYan Zheng /* list of reloc trees */ 1365d4f98a2SYan Zheng struct list_head reloc_roots; 137d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 138d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 1393fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1403fd0a558SYan, Zheng u64 merging_rsv_size; 1413fd0a558SYan, Zheng /* size of relocated tree nodes */ 1423fd0a558SYan, Zheng u64 nodes_relocated; 1430647bf56SWang Shilong /* reserved size for block group relocation*/ 1440647bf56SWang Shilong u64 reserved_bytes; 1453fd0a558SYan, Zheng 1465d4f98a2SYan Zheng u64 search_start; 1475d4f98a2SYan Zheng u64 extents_found; 1483fd0a558SYan, Zheng 1493fd0a558SYan, Zheng unsigned int stage:8; 1503fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1513fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1525d4f98a2SYan Zheng unsigned int found_file_extent:1; 1535d4f98a2SYan Zheng }; 1545d4f98a2SYan Zheng 1555d4f98a2SYan Zheng /* stages of data relocation */ 1565d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1575d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1585d4f98a2SYan Zheng 1599569cc20SQu Wenruo static void mark_block_processed(struct reloc_control *rc, 160a26195a5SQu Wenruo struct btrfs_backref_node *node) 1619569cc20SQu Wenruo { 1629569cc20SQu Wenruo u32 blocksize; 1639569cc20SQu Wenruo 1649569cc20SQu Wenruo if (node->level == 0 || 1659569cc20SQu Wenruo in_range(node->bytenr, rc->block_group->start, 1669569cc20SQu Wenruo rc->block_group->length)) { 1679569cc20SQu Wenruo blocksize = rc->extent_root->fs_info->nodesize; 1689569cc20SQu Wenruo set_extent_bits(&rc->processed_blocks, node->bytenr, 1699569cc20SQu Wenruo node->bytenr + blocksize - 1, EXTENT_DIRTY); 1709569cc20SQu Wenruo } 1719569cc20SQu Wenruo node->processed = 1; 1729569cc20SQu Wenruo } 1739569cc20SQu Wenruo 1745d4f98a2SYan Zheng 1755d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1765d4f98a2SYan Zheng { 1776bef4d31SEric Paris tree->rb_root = RB_ROOT; 1785d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1795d4f98a2SYan Zheng } 1805d4f98a2SYan Zheng 1815d4f98a2SYan Zheng /* 1825d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 1835d4f98a2SYan Zheng */ 184a26195a5SQu Wenruo static struct btrfs_backref_node *walk_up_backref( 185a26195a5SQu Wenruo struct btrfs_backref_node *node, 186a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 1875d4f98a2SYan Zheng { 188a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 1895d4f98a2SYan Zheng int idx = *index; 1905d4f98a2SYan Zheng 1915d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 1925d4f98a2SYan Zheng edge = list_entry(node->upper.next, 193a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 1945d4f98a2SYan Zheng edges[idx++] = edge; 1955d4f98a2SYan Zheng node = edge->node[UPPER]; 1965d4f98a2SYan Zheng } 1973fd0a558SYan, Zheng BUG_ON(node->detached); 1985d4f98a2SYan Zheng *index = idx; 1995d4f98a2SYan Zheng return node; 2005d4f98a2SYan Zheng } 2015d4f98a2SYan Zheng 2025d4f98a2SYan Zheng /* 2035d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 2045d4f98a2SYan Zheng */ 205a26195a5SQu Wenruo static struct btrfs_backref_node *walk_down_backref( 206a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 2075d4f98a2SYan Zheng { 208a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 209a26195a5SQu Wenruo struct btrfs_backref_node *lower; 2105d4f98a2SYan Zheng int idx = *index; 2115d4f98a2SYan Zheng 2125d4f98a2SYan Zheng while (idx > 0) { 2135d4f98a2SYan Zheng edge = edges[idx - 1]; 2145d4f98a2SYan Zheng lower = edge->node[LOWER]; 2155d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 2165d4f98a2SYan Zheng idx--; 2175d4f98a2SYan Zheng continue; 2185d4f98a2SYan Zheng } 2195d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 220a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 2215d4f98a2SYan Zheng edges[idx - 1] = edge; 2225d4f98a2SYan Zheng *index = idx; 2235d4f98a2SYan Zheng return edge->node[UPPER]; 2245d4f98a2SYan Zheng } 2255d4f98a2SYan Zheng *index = 0; 2265d4f98a2SYan Zheng return NULL; 2275d4f98a2SYan Zheng } 2285d4f98a2SYan Zheng 229a26195a5SQu Wenruo static void update_backref_node(struct btrfs_backref_cache *cache, 230a26195a5SQu Wenruo struct btrfs_backref_node *node, u64 bytenr) 2313fd0a558SYan, Zheng { 2323fd0a558SYan, Zheng struct rb_node *rb_node; 2333fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 2343fd0a558SYan, Zheng node->bytenr = bytenr; 235e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node); 23643c04fb1SJeff Mahoney if (rb_node) 237982c92cbSQu Wenruo btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST); 2383fd0a558SYan, Zheng } 2393fd0a558SYan, Zheng 2403fd0a558SYan, Zheng /* 2413fd0a558SYan, Zheng * update backref cache after a transaction commit 2423fd0a558SYan, Zheng */ 2433fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 244a26195a5SQu Wenruo struct btrfs_backref_cache *cache) 2453fd0a558SYan, Zheng { 246a26195a5SQu Wenruo struct btrfs_backref_node *node; 2473fd0a558SYan, Zheng int level = 0; 2483fd0a558SYan, Zheng 2493fd0a558SYan, Zheng if (cache->last_trans == 0) { 2503fd0a558SYan, Zheng cache->last_trans = trans->transid; 2513fd0a558SYan, Zheng return 0; 2523fd0a558SYan, Zheng } 2533fd0a558SYan, Zheng 2543fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 2553fd0a558SYan, Zheng return 0; 2563fd0a558SYan, Zheng 2573fd0a558SYan, Zheng /* 2583fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 2593fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 2603fd0a558SYan, Zheng * so the detached nodes are no longer useful. 2613fd0a558SYan, Zheng */ 2623fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2633fd0a558SYan, Zheng node = list_entry(cache->detached.next, 264a26195a5SQu Wenruo struct btrfs_backref_node, list); 265023acb07SQu Wenruo btrfs_backref_cleanup_node(cache, node); 2663fd0a558SYan, Zheng } 2673fd0a558SYan, Zheng 2683fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 2693fd0a558SYan, Zheng node = list_entry(cache->changed.next, 270a26195a5SQu Wenruo struct btrfs_backref_node, list); 2713fd0a558SYan, Zheng list_del_init(&node->list); 2723fd0a558SYan, Zheng BUG_ON(node->pending); 2733fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2743fd0a558SYan, Zheng } 2753fd0a558SYan, Zheng 2763fd0a558SYan, Zheng /* 2773fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 2783fd0a558SYan, Zheng * errors during processing the pending nodes. 2793fd0a558SYan, Zheng */ 2803fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 2813fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 2823fd0a558SYan, Zheng BUG_ON(!node->pending); 2833fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 2843fd0a558SYan, Zheng continue; 2853fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2863fd0a558SYan, Zheng } 2873fd0a558SYan, Zheng } 2883fd0a558SYan, Zheng 2893fd0a558SYan, Zheng cache->last_trans = 0; 2903fd0a558SYan, Zheng return 1; 2913fd0a558SYan, Zheng } 2923fd0a558SYan, Zheng 2936282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 2946282675eSQu Wenruo { 2956282675eSQu Wenruo /* 2966282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 2976282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 2986282675eSQu Wenruo * trying to access reloc_root 2996282675eSQu Wenruo */ 3006282675eSQu Wenruo smp_rmb(); 3016282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 3026282675eSQu Wenruo return true; 3036282675eSQu Wenruo return false; 3046282675eSQu Wenruo } 3056282675eSQu Wenruo 3066282675eSQu Wenruo /* 3076282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 3086282675eSQu Wenruo * 3096282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 3106282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 31155465730SQu Wenruo * from no reloc root. But btrfs_should_ignore_reloc_root() below is a 31255465730SQu Wenruo * special case. 3136282675eSQu Wenruo */ 3146282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 3156282675eSQu Wenruo { 3166282675eSQu Wenruo if (reloc_root_is_dead(root)) 3176282675eSQu Wenruo return false; 3186282675eSQu Wenruo if (!root->reloc_root) 3196282675eSQu Wenruo return false; 3206282675eSQu Wenruo return true; 3216282675eSQu Wenruo } 322f2a97a9dSDavid Sterba 32355465730SQu Wenruo int btrfs_should_ignore_reloc_root(struct btrfs_root *root) 3243fd0a558SYan, Zheng { 3253fd0a558SYan, Zheng struct btrfs_root *reloc_root; 3263fd0a558SYan, Zheng 32792a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 3283fd0a558SYan, Zheng return 0; 3293fd0a558SYan, Zheng 3306282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 3316282675eSQu Wenruo if (reloc_root_is_dead(root)) 3326282675eSQu Wenruo return 1; 3336282675eSQu Wenruo 3343fd0a558SYan, Zheng reloc_root = root->reloc_root; 3353fd0a558SYan, Zheng if (!reloc_root) 3363fd0a558SYan, Zheng return 0; 3373fd0a558SYan, Zheng 3384d4225fcSJosef Bacik if (btrfs_header_generation(reloc_root->commit_root) == 3394d4225fcSJosef Bacik root->fs_info->running_transaction->transid) 3403fd0a558SYan, Zheng return 0; 3413fd0a558SYan, Zheng /* 3423fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 3433fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 3443fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 3453fd0a558SYan, Zheng * relocation. 3463fd0a558SYan, Zheng */ 3473fd0a558SYan, Zheng return 1; 3483fd0a558SYan, Zheng } 34955465730SQu Wenruo 3505d4f98a2SYan Zheng /* 3515d4f98a2SYan Zheng * find reloc tree by address of tree root 3525d4f98a2SYan Zheng */ 3532433bea5SQu Wenruo struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) 3545d4f98a2SYan Zheng { 3552433bea5SQu Wenruo struct reloc_control *rc = fs_info->reloc_ctl; 3565d4f98a2SYan Zheng struct rb_node *rb_node; 3575d4f98a2SYan Zheng struct mapping_node *node; 3585d4f98a2SYan Zheng struct btrfs_root *root = NULL; 3595d4f98a2SYan Zheng 3602433bea5SQu Wenruo ASSERT(rc); 3615d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 362e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr); 3635d4f98a2SYan Zheng if (rb_node) { 3645d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 3655d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 3665d4f98a2SYan Zheng } 3675d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 36800246528SJosef Bacik return btrfs_grab_root(root); 3695d4f98a2SYan Zheng } 3705d4f98a2SYan Zheng 3715d4f98a2SYan Zheng /* 37229db137bSQu Wenruo * For useless nodes, do two major clean ups: 37329db137bSQu Wenruo * 37429db137bSQu Wenruo * - Cleanup the children edges and nodes 37529db137bSQu Wenruo * If child node is also orphan (no parent) during cleanup, then the child 37629db137bSQu Wenruo * node will also be cleaned up. 37729db137bSQu Wenruo * 37829db137bSQu Wenruo * - Freeing up leaves (level 0), keeps nodes detached 37929db137bSQu Wenruo * For nodes, the node is still cached as "detached" 38029db137bSQu Wenruo * 38129db137bSQu Wenruo * Return false if @node is not in the @useless_nodes list. 38229db137bSQu Wenruo * Return true if @node is in the @useless_nodes list. 38329db137bSQu Wenruo */ 38429db137bSQu Wenruo static bool handle_useless_nodes(struct reloc_control *rc, 385a26195a5SQu Wenruo struct btrfs_backref_node *node) 38629db137bSQu Wenruo { 387a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 38829db137bSQu Wenruo struct list_head *useless_node = &cache->useless_node; 38929db137bSQu Wenruo bool ret = false; 39029db137bSQu Wenruo 39129db137bSQu Wenruo while (!list_empty(useless_node)) { 392a26195a5SQu Wenruo struct btrfs_backref_node *cur; 39329db137bSQu Wenruo 394a26195a5SQu Wenruo cur = list_first_entry(useless_node, struct btrfs_backref_node, 39529db137bSQu Wenruo list); 39629db137bSQu Wenruo list_del_init(&cur->list); 39729db137bSQu Wenruo 39829db137bSQu Wenruo /* Only tree root nodes can be added to @useless_nodes */ 39929db137bSQu Wenruo ASSERT(list_empty(&cur->upper)); 40029db137bSQu Wenruo 40129db137bSQu Wenruo if (cur == node) 40229db137bSQu Wenruo ret = true; 40329db137bSQu Wenruo 40429db137bSQu Wenruo /* The node is the lowest node */ 40529db137bSQu Wenruo if (cur->lowest) { 40629db137bSQu Wenruo list_del_init(&cur->lower); 40729db137bSQu Wenruo cur->lowest = 0; 40829db137bSQu Wenruo } 40929db137bSQu Wenruo 41029db137bSQu Wenruo /* Cleanup the lower edges */ 41129db137bSQu Wenruo while (!list_empty(&cur->lower)) { 412a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 413a26195a5SQu Wenruo struct btrfs_backref_node *lower; 41429db137bSQu Wenruo 41529db137bSQu Wenruo edge = list_entry(cur->lower.next, 416a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 41729db137bSQu Wenruo list_del(&edge->list[UPPER]); 41829db137bSQu Wenruo list_del(&edge->list[LOWER]); 41929db137bSQu Wenruo lower = edge->node[LOWER]; 420741188d3SQu Wenruo btrfs_backref_free_edge(cache, edge); 42129db137bSQu Wenruo 42229db137bSQu Wenruo /* Child node is also orphan, queue for cleanup */ 42329db137bSQu Wenruo if (list_empty(&lower->upper)) 42429db137bSQu Wenruo list_add(&lower->list, useless_node); 42529db137bSQu Wenruo } 42629db137bSQu Wenruo /* Mark this block processed for relocation */ 42729db137bSQu Wenruo mark_block_processed(rc, cur); 42829db137bSQu Wenruo 42929db137bSQu Wenruo /* 43029db137bSQu Wenruo * Backref nodes for tree leaves are deleted from the cache. 43129db137bSQu Wenruo * Backref nodes for upper level tree blocks are left in the 43229db137bSQu Wenruo * cache to avoid unnecessary backref lookup. 43329db137bSQu Wenruo */ 43429db137bSQu Wenruo if (cur->level > 0) { 43529db137bSQu Wenruo list_add(&cur->list, &cache->detached); 43629db137bSQu Wenruo cur->detached = 1; 43729db137bSQu Wenruo } else { 43829db137bSQu Wenruo rb_erase(&cur->rb_node, &cache->rb_root); 439741188d3SQu Wenruo btrfs_backref_free_node(cache, cur); 44029db137bSQu Wenruo } 44129db137bSQu Wenruo } 44229db137bSQu Wenruo return ret; 44329db137bSQu Wenruo } 44429db137bSQu Wenruo 44529db137bSQu Wenruo /* 446e7d571c7SQu Wenruo * Build backref tree for a given tree block. Root of the backref tree 447e7d571c7SQu Wenruo * corresponds the tree block, leaves of the backref tree correspond roots of 448e7d571c7SQu Wenruo * b-trees that reference the tree block. 449e7d571c7SQu Wenruo * 450e7d571c7SQu Wenruo * The basic idea of this function is check backrefs of a given block to find 451e7d571c7SQu Wenruo * upper level blocks that reference the block, and then check backrefs of 452e7d571c7SQu Wenruo * these upper level blocks recursively. The recursion stops when tree root is 453e7d571c7SQu Wenruo * reached or backrefs for the block is cached. 454e7d571c7SQu Wenruo * 455e7d571c7SQu Wenruo * NOTE: if we find that backrefs for a block are cached, we know backrefs for 456e7d571c7SQu Wenruo * all upper level blocks that directly/indirectly reference the block are also 457e7d571c7SQu Wenruo * cached. 458e7d571c7SQu Wenruo */ 459a26195a5SQu Wenruo static noinline_for_stack struct btrfs_backref_node *build_backref_tree( 460e7d571c7SQu Wenruo struct reloc_control *rc, struct btrfs_key *node_key, 461e7d571c7SQu Wenruo int level, u64 bytenr) 462e7d571c7SQu Wenruo { 463e7d571c7SQu Wenruo struct btrfs_backref_iter *iter; 464a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 465e7d571c7SQu Wenruo /* For searching parent of TREE_BLOCK_REF */ 466e7d571c7SQu Wenruo struct btrfs_path *path; 467a26195a5SQu Wenruo struct btrfs_backref_node *cur; 468a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 469a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 470e7d571c7SQu Wenruo int ret; 471e7d571c7SQu Wenruo int err = 0; 472e7d571c7SQu Wenruo 473e7d571c7SQu Wenruo iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS); 474e7d571c7SQu Wenruo if (!iter) 475e7d571c7SQu Wenruo return ERR_PTR(-ENOMEM); 476e7d571c7SQu Wenruo path = btrfs_alloc_path(); 477e7d571c7SQu Wenruo if (!path) { 478e7d571c7SQu Wenruo err = -ENOMEM; 479e7d571c7SQu Wenruo goto out; 480e7d571c7SQu Wenruo } 481e7d571c7SQu Wenruo 482b1818dabSQu Wenruo node = btrfs_backref_alloc_node(cache, bytenr, level); 483e7d571c7SQu Wenruo if (!node) { 484e7d571c7SQu Wenruo err = -ENOMEM; 485e7d571c7SQu Wenruo goto out; 486e7d571c7SQu Wenruo } 487e7d571c7SQu Wenruo 488e7d571c7SQu Wenruo node->lowest = 1; 489e7d571c7SQu Wenruo cur = node; 490e7d571c7SQu Wenruo 491e7d571c7SQu Wenruo /* Breadth-first search to build backref cache */ 492e7d571c7SQu Wenruo do { 4931b60d2ecSQu Wenruo ret = btrfs_backref_add_tree_node(cache, path, iter, node_key, 4941b60d2ecSQu Wenruo cur); 495e7d571c7SQu Wenruo if (ret < 0) { 496e7d571c7SQu Wenruo err = ret; 497e7d571c7SQu Wenruo goto out; 498e7d571c7SQu Wenruo } 499e7d571c7SQu Wenruo edge = list_first_entry_or_null(&cache->pending_edge, 500a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 501e7d571c7SQu Wenruo /* 502e7d571c7SQu Wenruo * The pending list isn't empty, take the first block to 503e7d571c7SQu Wenruo * process 504e7d571c7SQu Wenruo */ 505e7d571c7SQu Wenruo if (edge) { 5065d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 5075d4f98a2SYan Zheng cur = edge->node[UPPER]; 5085d4f98a2SYan Zheng } 509e7d571c7SQu Wenruo } while (edge); 5105d4f98a2SYan Zheng 5111f872924SQu Wenruo /* Finish the upper linkage of newly added edges/nodes */ 512fc997ed0SQu Wenruo ret = btrfs_backref_finish_upper_links(cache, node); 5131f872924SQu Wenruo if (ret < 0) { 5141f872924SQu Wenruo err = ret; 51575bfb9afSJosef Bacik goto out; 51675bfb9afSJosef Bacik } 51775bfb9afSJosef Bacik 51829db137bSQu Wenruo if (handle_useless_nodes(rc, node)) 5193fd0a558SYan, Zheng node = NULL; 5205d4f98a2SYan Zheng out: 52171f572a9SQu Wenruo btrfs_backref_iter_free(iter); 52271f572a9SQu Wenruo btrfs_free_path(path); 5235d4f98a2SYan Zheng if (err) { 5241b23ea18SQu Wenruo btrfs_backref_error_cleanup(cache, node); 5255d4f98a2SYan Zheng return ERR_PTR(err); 5265d4f98a2SYan Zheng } 52775bfb9afSJosef Bacik ASSERT(!node || !node->detached); 52884780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 52984780289SQu Wenruo list_empty(&cache->pending_edge)); 5305d4f98a2SYan Zheng return node; 5315d4f98a2SYan Zheng } 5325d4f98a2SYan Zheng 5335d4f98a2SYan Zheng /* 5343fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 5353fd0a558SYan, Zheng * the backref node is created by cloning backref node that 5363fd0a558SYan, Zheng * corresponds to root of source tree 5373fd0a558SYan, Zheng */ 5383fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 5393fd0a558SYan, Zheng struct reloc_control *rc, 5403fd0a558SYan, Zheng struct btrfs_root *src, 5413fd0a558SYan, Zheng struct btrfs_root *dest) 5423fd0a558SYan, Zheng { 5433fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 544a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 545a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 546a26195a5SQu Wenruo struct btrfs_backref_node *new_node; 547a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 548a26195a5SQu Wenruo struct btrfs_backref_edge *new_edge; 5493fd0a558SYan, Zheng struct rb_node *rb_node; 5503fd0a558SYan, Zheng 5513fd0a558SYan, Zheng if (cache->last_trans > 0) 5523fd0a558SYan, Zheng update_backref_cache(trans, cache); 5533fd0a558SYan, Zheng 554e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start); 5553fd0a558SYan, Zheng if (rb_node) { 556a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, rb_node); 5573fd0a558SYan, Zheng if (node->detached) 5583fd0a558SYan, Zheng node = NULL; 5593fd0a558SYan, Zheng else 5603fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 5613fd0a558SYan, Zheng } 5623fd0a558SYan, Zheng 5633fd0a558SYan, Zheng if (!node) { 564e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, 5653fd0a558SYan, Zheng reloc_root->commit_root->start); 5663fd0a558SYan, Zheng if (rb_node) { 567a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, 5683fd0a558SYan, Zheng rb_node); 5693fd0a558SYan, Zheng BUG_ON(node->detached); 5703fd0a558SYan, Zheng } 5713fd0a558SYan, Zheng } 5723fd0a558SYan, Zheng 5733fd0a558SYan, Zheng if (!node) 5743fd0a558SYan, Zheng return 0; 5753fd0a558SYan, Zheng 576b1818dabSQu Wenruo new_node = btrfs_backref_alloc_node(cache, dest->node->start, 577b1818dabSQu Wenruo node->level); 5783fd0a558SYan, Zheng if (!new_node) 5793fd0a558SYan, Zheng return -ENOMEM; 5803fd0a558SYan, Zheng 5813fd0a558SYan, Zheng new_node->lowest = node->lowest; 5826848ad64SYan, Zheng new_node->checked = 1; 58300246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 5840b530bc5SJosef Bacik ASSERT(new_node->root); 5853fd0a558SYan, Zheng 5863fd0a558SYan, Zheng if (!node->lowest) { 5873fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 58847254d07SQu Wenruo new_edge = btrfs_backref_alloc_edge(cache); 5893fd0a558SYan, Zheng if (!new_edge) 5903fd0a558SYan, Zheng goto fail; 5913fd0a558SYan, Zheng 592f39911e5SQu Wenruo btrfs_backref_link_edge(new_edge, edge->node[LOWER], 593f39911e5SQu Wenruo new_node, LINK_UPPER); 5943fd0a558SYan, Zheng } 59576b9e23dSMiao Xie } else { 59676b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 5973fd0a558SYan, Zheng } 5983fd0a558SYan, Zheng 599e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr, 6003fd0a558SYan, Zheng &new_node->rb_node); 60143c04fb1SJeff Mahoney if (rb_node) 602982c92cbSQu Wenruo btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST); 6033fd0a558SYan, Zheng 6043fd0a558SYan, Zheng if (!new_node->lowest) { 6053fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 6063fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 6073fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 6083fd0a558SYan, Zheng } 6093fd0a558SYan, Zheng } 6103fd0a558SYan, Zheng return 0; 6113fd0a558SYan, Zheng fail: 6123fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 6133fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 614a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 6153fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 616741188d3SQu Wenruo btrfs_backref_free_edge(cache, new_edge); 6173fd0a558SYan, Zheng } 618741188d3SQu Wenruo btrfs_backref_free_node(cache, new_node); 6193fd0a558SYan, Zheng return -ENOMEM; 6203fd0a558SYan, Zheng } 6213fd0a558SYan, Zheng 6223fd0a558SYan, Zheng /* 6235d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 6245d4f98a2SYan Zheng */ 625ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 6265d4f98a2SYan Zheng { 6270b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6285d4f98a2SYan Zheng struct rb_node *rb_node; 6295d4f98a2SYan Zheng struct mapping_node *node; 6300b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 6315d4f98a2SYan Zheng 6325d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 633ffd7b339SJeff Mahoney if (!node) 634ffd7b339SJeff Mahoney return -ENOMEM; 6355d4f98a2SYan Zheng 636ea287ab1SJosef Bacik node->bytenr = root->commit_root->start; 6375d4f98a2SYan Zheng node->data = root; 6385d4f98a2SYan Zheng 6395d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 640e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 6415d4f98a2SYan Zheng node->bytenr, &node->rb_node); 6425d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 643ffd7b339SJeff Mahoney if (rb_node) { 64457a304cfSJosef Bacik btrfs_err(fs_info, 6455d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 6465d163e0eSJeff Mahoney node->bytenr); 64757a304cfSJosef Bacik return -EEXIST; 648ffd7b339SJeff Mahoney } 6495d4f98a2SYan Zheng 6505d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 6515d4f98a2SYan Zheng return 0; 6525d4f98a2SYan Zheng } 6535d4f98a2SYan Zheng 6545d4f98a2SYan Zheng /* 655c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 6565d4f98a2SYan Zheng * mapping 6575d4f98a2SYan Zheng */ 658c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 6595d4f98a2SYan Zheng { 6600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6615d4f98a2SYan Zheng struct rb_node *rb_node; 6625d4f98a2SYan Zheng struct mapping_node *node = NULL; 6630b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 664f44deb74SJosef Bacik bool put_ref = false; 6655d4f98a2SYan Zheng 66665c6e82bSQu Wenruo if (rc && root->node) { 6675d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 668e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 669ea287ab1SJosef Bacik root->commit_root->start); 670c974c464SWang Shilong if (rb_node) { 671c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 672c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 673ea287ab1SJosef Bacik RB_CLEAR_NODE(&node->rb_node); 674c974c464SWang Shilong } 675c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 676c78a10aeSJosef Bacik ASSERT(!node || (struct btrfs_root *)node->data == root); 677389305b2SQu Wenruo } 678c974c464SWang Shilong 679f44deb74SJosef Bacik /* 680f44deb74SJosef Bacik * We only put the reloc root here if it's on the list. There's a lot 681f44deb74SJosef Bacik * of places where the pattern is to splice the rc->reloc_roots, process 682f44deb74SJosef Bacik * the reloc roots, and then add the reloc root back onto 683f44deb74SJosef Bacik * rc->reloc_roots. If we call __del_reloc_root while it's off of the 684f44deb74SJosef Bacik * list we don't want the reference being dropped, because the guy 685f44deb74SJosef Bacik * messing with the list is in charge of the reference. 686f44deb74SJosef Bacik */ 6870b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 688f44deb74SJosef Bacik if (!list_empty(&root->root_list)) { 689f44deb74SJosef Bacik put_ref = true; 690c974c464SWang Shilong list_del_init(&root->root_list); 691f44deb74SJosef Bacik } 6920b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 693f44deb74SJosef Bacik if (put_ref) 694f44deb74SJosef Bacik btrfs_put_root(root); 695c974c464SWang Shilong kfree(node); 696c974c464SWang Shilong } 697c974c464SWang Shilong 698c974c464SWang Shilong /* 699c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 700c974c464SWang Shilong * mapping 701c974c464SWang Shilong */ 702ea287ab1SJosef Bacik static int __update_reloc_root(struct btrfs_root *root) 703c974c464SWang Shilong { 7040b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 705c974c464SWang Shilong struct rb_node *rb_node; 706c974c464SWang Shilong struct mapping_node *node = NULL; 7070b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 708c974c464SWang Shilong 709c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 710e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 711ea287ab1SJosef Bacik root->commit_root->start); 7125d4f98a2SYan Zheng if (rb_node) { 7135d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 7145d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 7155d4f98a2SYan Zheng } 7165d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 7175d4f98a2SYan Zheng 7188f71f3e0SLiu Bo if (!node) 7198f71f3e0SLiu Bo return 0; 7205d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 7215d4f98a2SYan Zheng 7225d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 723ea287ab1SJosef Bacik node->bytenr = root->node->start; 724e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 7255d4f98a2SYan Zheng node->bytenr, &node->rb_node); 7265d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 72743c04fb1SJeff Mahoney if (rb_node) 728982c92cbSQu Wenruo btrfs_backref_panic(fs_info, node->bytenr, -EEXIST); 7295d4f98a2SYan Zheng return 0; 7305d4f98a2SYan Zheng } 7315d4f98a2SYan Zheng 7323fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 7333fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 7345d4f98a2SYan Zheng { 7350b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 7365d4f98a2SYan Zheng struct btrfs_root *reloc_root; 7375d4f98a2SYan Zheng struct extent_buffer *eb; 7385d4f98a2SYan Zheng struct btrfs_root_item *root_item; 7395d4f98a2SYan Zheng struct btrfs_key root_key; 74084c50ba5SJosef Bacik int ret = 0; 74184c50ba5SJosef Bacik bool must_abort = false; 7425d4f98a2SYan Zheng 7435d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 74484c50ba5SJosef Bacik if (!root_item) 74584c50ba5SJosef Bacik return ERR_PTR(-ENOMEM); 7465d4f98a2SYan Zheng 7475d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 7485d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 7493fd0a558SYan, Zheng root_key.offset = objectid; 7505d4f98a2SYan Zheng 7513fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 752054570a1SFilipe Manana u64 commit_root_gen; 753054570a1SFilipe Manana 7543fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 7555d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 7565d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 75784c50ba5SJosef Bacik if (ret) 75884c50ba5SJosef Bacik goto fail; 75984c50ba5SJosef Bacik 760054570a1SFilipe Manana /* 761054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 762054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 763054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 764054570a1SFilipe Manana * either inside the critical section of a transaction commit 765054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 766054570a1SFilipe Manana * it's created before the transaction commit is started. 767054570a1SFilipe Manana */ 768054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 769054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 7703fd0a558SYan, Zheng } else { 7713fd0a558SYan, Zheng /* 7723fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 7733fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 7743fd0a558SYan, Zheng * modified after it was created have RELOC flag 7753fd0a558SYan, Zheng * set in their headers. so it's OK to not update 7763fd0a558SYan, Zheng * the 'last_snapshot'. 7773fd0a558SYan, Zheng */ 7783fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 7793fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 78084c50ba5SJosef Bacik if (ret) 78184c50ba5SJosef Bacik goto fail; 7823fd0a558SYan, Zheng } 7833fd0a558SYan, Zheng 78484c50ba5SJosef Bacik /* 78584c50ba5SJosef Bacik * We have changed references at this point, we must abort the 78684c50ba5SJosef Bacik * transaction if anything fails. 78784c50ba5SJosef Bacik */ 78884c50ba5SJosef Bacik must_abort = true; 78984c50ba5SJosef Bacik 7905d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 7915d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 7925d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 7935d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 7943fd0a558SYan, Zheng 7953fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 7963fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 7973fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 7983fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 799c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, 0); 8003fd0a558SYan, Zheng } 8015d4f98a2SYan Zheng 8025d4f98a2SYan Zheng btrfs_tree_unlock(eb); 8035d4f98a2SYan Zheng free_extent_buffer(eb); 8045d4f98a2SYan Zheng 8050b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 8065d4f98a2SYan Zheng &root_key, root_item); 80784c50ba5SJosef Bacik if (ret) 80884c50ba5SJosef Bacik goto fail; 80984c50ba5SJosef Bacik 8105d4f98a2SYan Zheng kfree(root_item); 8115d4f98a2SYan Zheng 8123dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 81384c50ba5SJosef Bacik if (IS_ERR(reloc_root)) { 81484c50ba5SJosef Bacik ret = PTR_ERR(reloc_root); 81584c50ba5SJosef Bacik goto abort; 81684c50ba5SJosef Bacik } 81792a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 8185d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 8193fd0a558SYan, Zheng return reloc_root; 82084c50ba5SJosef Bacik fail: 82184c50ba5SJosef Bacik kfree(root_item); 82284c50ba5SJosef Bacik abort: 82384c50ba5SJosef Bacik if (must_abort) 82484c50ba5SJosef Bacik btrfs_abort_transaction(trans, ret); 82584c50ba5SJosef Bacik return ERR_PTR(ret); 8263fd0a558SYan, Zheng } 8273fd0a558SYan, Zheng 8283fd0a558SYan, Zheng /* 8293fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 8303fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 831f44deb74SJosef Bacik * 832f44deb74SJosef Bacik * The reloc_root comes out of here with two references, one for 833f44deb74SJosef Bacik * root->reloc_root, and another for being on the rc->reloc_roots list. 8343fd0a558SYan, Zheng */ 8353fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 8363fd0a558SYan, Zheng struct btrfs_root *root) 8373fd0a558SYan, Zheng { 8380b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8393fd0a558SYan, Zheng struct btrfs_root *reloc_root; 8400b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 84120dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 8423fd0a558SYan, Zheng int clear_rsv = 0; 843ffd7b339SJeff Mahoney int ret; 8443fd0a558SYan, Zheng 845aec7db3bSJosef Bacik if (!rc) 8462abc726aSJosef Bacik return 0; 8472abc726aSJosef Bacik 8481fac4a54SQu Wenruo /* 8491fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 8501fac4a54SQu Wenruo * create/update the dead reloc tree 8511fac4a54SQu Wenruo */ 8526282675eSQu Wenruo if (reloc_root_is_dead(root)) 8531fac4a54SQu Wenruo return 0; 8541fac4a54SQu Wenruo 855aec7db3bSJosef Bacik /* 856aec7db3bSJosef Bacik * This is subtle but important. We do not do 857aec7db3bSJosef Bacik * record_root_in_transaction for reloc roots, instead we record their 858aec7db3bSJosef Bacik * corresponding fs root, and then here we update the last trans for the 859aec7db3bSJosef Bacik * reloc root. This means that we have to do this for the entire life 860aec7db3bSJosef Bacik * of the reloc root, regardless of which stage of the relocation we are 861aec7db3bSJosef Bacik * in. 862aec7db3bSJosef Bacik */ 8633fd0a558SYan, Zheng if (root->reloc_root) { 8643fd0a558SYan, Zheng reloc_root = root->reloc_root; 8653fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 8663fd0a558SYan, Zheng return 0; 8673fd0a558SYan, Zheng } 8683fd0a558SYan, Zheng 869aec7db3bSJosef Bacik /* 870aec7db3bSJosef Bacik * We are merging reloc roots, we do not need new reloc trees. Also 871aec7db3bSJosef Bacik * reloc trees never need their own reloc tree. 872aec7db3bSJosef Bacik */ 873aec7db3bSJosef Bacik if (!rc->create_reloc_tree || 874aec7db3bSJosef Bacik root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 875aec7db3bSJosef Bacik return 0; 876aec7db3bSJosef Bacik 87720dd2cbfSMiao Xie if (!trans->reloc_reserved) { 87820dd2cbfSMiao Xie rsv = trans->block_rsv; 8793fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 8803fd0a558SYan, Zheng clear_rsv = 1; 8813fd0a558SYan, Zheng } 8823fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 8833fd0a558SYan, Zheng if (clear_rsv) 88420dd2cbfSMiao Xie trans->block_rsv = rsv; 88500bb36a0SJosef Bacik if (IS_ERR(reloc_root)) 88600bb36a0SJosef Bacik return PTR_ERR(reloc_root); 8875d4f98a2SYan Zheng 888ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 88957a304cfSJosef Bacik ASSERT(ret != -EEXIST); 89000bb36a0SJosef Bacik if (ret) { 89100bb36a0SJosef Bacik /* Pairs with create_reloc_root */ 89200bb36a0SJosef Bacik btrfs_put_root(reloc_root); 89300bb36a0SJosef Bacik return ret; 89400bb36a0SJosef Bacik } 895f44deb74SJosef Bacik root->reloc_root = btrfs_grab_root(reloc_root); 8965d4f98a2SYan Zheng return 0; 8975d4f98a2SYan Zheng } 8985d4f98a2SYan Zheng 8995d4f98a2SYan Zheng /* 9005d4f98a2SYan Zheng * update root item of reloc tree 9015d4f98a2SYan Zheng */ 9025d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 9035d4f98a2SYan Zheng struct btrfs_root *root) 9045d4f98a2SYan Zheng { 9050b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 9065d4f98a2SYan Zheng struct btrfs_root *reloc_root; 9075d4f98a2SYan Zheng struct btrfs_root_item *root_item; 9085d4f98a2SYan Zheng int ret; 9095d4f98a2SYan Zheng 9106282675eSQu Wenruo if (!have_reloc_root(root)) 911592fbcd5SJosef Bacik return 0; 9125d4f98a2SYan Zheng 9135d4f98a2SYan Zheng reloc_root = root->reloc_root; 9145d4f98a2SYan Zheng root_item = &reloc_root->root_item; 9155d4f98a2SYan Zheng 916f44deb74SJosef Bacik /* 917f44deb74SJosef Bacik * We are probably ok here, but __del_reloc_root() will drop its ref of 918f44deb74SJosef Bacik * the root. We have the ref for root->reloc_root, but just in case 919f44deb74SJosef Bacik * hold it while we update the reloc root. 920f44deb74SJosef Bacik */ 921f44deb74SJosef Bacik btrfs_grab_root(reloc_root); 922f44deb74SJosef Bacik 923d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 9240b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 9253fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 926d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 9276282675eSQu Wenruo /* 9286282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 9296282675eSQu Wenruo * have_reloc_root will not touch it from now on. 9306282675eSQu Wenruo */ 9316282675eSQu Wenruo smp_wmb(); 932c974c464SWang Shilong __del_reloc_root(reloc_root); 9335d4f98a2SYan Zheng } 9345d4f98a2SYan Zheng 9355d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 936ea287ab1SJosef Bacik __update_reloc_root(reloc_root); 9375d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 9385d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 9395d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 9405d4f98a2SYan Zheng } 9415d4f98a2SYan Zheng 9420b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 9435d4f98a2SYan Zheng &reloc_root->root_key, root_item); 944f44deb74SJosef Bacik btrfs_put_root(reloc_root); 945592fbcd5SJosef Bacik return ret; 9465d4f98a2SYan Zheng } 9475d4f98a2SYan Zheng 9485d4f98a2SYan Zheng /* 9495d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 9505d4f98a2SYan Zheng * in a subvolume 9515d4f98a2SYan Zheng */ 9525d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 9535d4f98a2SYan Zheng { 9545d4f98a2SYan Zheng struct rb_node *node; 9555d4f98a2SYan Zheng struct rb_node *prev; 9565d4f98a2SYan Zheng struct btrfs_inode *entry; 9575d4f98a2SYan Zheng struct inode *inode; 9585d4f98a2SYan Zheng 9595d4f98a2SYan Zheng spin_lock(&root->inode_lock); 9605d4f98a2SYan Zheng again: 9615d4f98a2SYan Zheng node = root->inode_tree.rb_node; 9625d4f98a2SYan Zheng prev = NULL; 9635d4f98a2SYan Zheng while (node) { 9645d4f98a2SYan Zheng prev = node; 9655d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9665d4f98a2SYan Zheng 9674a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 9685d4f98a2SYan Zheng node = node->rb_left; 9694a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 9705d4f98a2SYan Zheng node = node->rb_right; 9715d4f98a2SYan Zheng else 9725d4f98a2SYan Zheng break; 9735d4f98a2SYan Zheng } 9745d4f98a2SYan Zheng if (!node) { 9755d4f98a2SYan Zheng while (prev) { 9765d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 9774a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 9785d4f98a2SYan Zheng node = prev; 9795d4f98a2SYan Zheng break; 9805d4f98a2SYan Zheng } 9815d4f98a2SYan Zheng prev = rb_next(prev); 9825d4f98a2SYan Zheng } 9835d4f98a2SYan Zheng } 9845d4f98a2SYan Zheng while (node) { 9855d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9865d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 9875d4f98a2SYan Zheng if (inode) { 9885d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 9895d4f98a2SYan Zheng return inode; 9905d4f98a2SYan Zheng } 9915d4f98a2SYan Zheng 9924a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 9935d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 9945d4f98a2SYan Zheng goto again; 9955d4f98a2SYan Zheng 9965d4f98a2SYan Zheng node = rb_next(node); 9975d4f98a2SYan Zheng } 9985d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 9995d4f98a2SYan Zheng return NULL; 10005d4f98a2SYan Zheng } 10015d4f98a2SYan Zheng 10025d4f98a2SYan Zheng /* 10035d4f98a2SYan Zheng * get new location of data 10045d4f98a2SYan Zheng */ 10055d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 10065d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 10075d4f98a2SYan Zheng { 10085d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 10095d4f98a2SYan Zheng struct btrfs_path *path; 10105d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10115d4f98a2SYan Zheng struct extent_buffer *leaf; 10125d4f98a2SYan Zheng int ret; 10135d4f98a2SYan Zheng 10145d4f98a2SYan Zheng path = btrfs_alloc_path(); 10155d4f98a2SYan Zheng if (!path) 10165d4f98a2SYan Zheng return -ENOMEM; 10175d4f98a2SYan Zheng 10185d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1019f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1020f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 10215d4f98a2SYan Zheng if (ret < 0) 10225d4f98a2SYan Zheng goto out; 10235d4f98a2SYan Zheng if (ret > 0) { 10245d4f98a2SYan Zheng ret = -ENOENT; 10255d4f98a2SYan Zheng goto out; 10265d4f98a2SYan Zheng } 10275d4f98a2SYan Zheng 10285d4f98a2SYan Zheng leaf = path->nodes[0]; 10295d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 10305d4f98a2SYan Zheng struct btrfs_file_extent_item); 10315d4f98a2SYan Zheng 10325d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 10335d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 10345d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 10355d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 10365d4f98a2SYan Zheng 10375d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 103883d4cfd4SJosef Bacik ret = -EINVAL; 10395d4f98a2SYan Zheng goto out; 10405d4f98a2SYan Zheng } 10415d4f98a2SYan Zheng 10425d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 10435d4f98a2SYan Zheng ret = 0; 10445d4f98a2SYan Zheng out: 10455d4f98a2SYan Zheng btrfs_free_path(path); 10465d4f98a2SYan Zheng return ret; 10475d4f98a2SYan Zheng } 10485d4f98a2SYan Zheng 10495d4f98a2SYan Zheng /* 10505d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 10515d4f98a2SYan Zheng * the new locations. 10525d4f98a2SYan Zheng */ 10533fd0a558SYan, Zheng static noinline_for_stack 10543fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 10555d4f98a2SYan Zheng struct reloc_control *rc, 10565d4f98a2SYan Zheng struct btrfs_root *root, 10573fd0a558SYan, Zheng struct extent_buffer *leaf) 10585d4f98a2SYan Zheng { 10590b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 10605d4f98a2SYan Zheng struct btrfs_key key; 10615d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10625d4f98a2SYan Zheng struct inode *inode = NULL; 10635d4f98a2SYan Zheng u64 parent; 10645d4f98a2SYan Zheng u64 bytenr; 10653fd0a558SYan, Zheng u64 new_bytenr = 0; 10665d4f98a2SYan Zheng u64 num_bytes; 10675d4f98a2SYan Zheng u64 end; 10685d4f98a2SYan Zheng u32 nritems; 10695d4f98a2SYan Zheng u32 i; 107083d4cfd4SJosef Bacik int ret = 0; 10715d4f98a2SYan Zheng int first = 1; 10725d4f98a2SYan Zheng int dirty = 0; 10735d4f98a2SYan Zheng 10745d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 10755d4f98a2SYan Zheng return 0; 10765d4f98a2SYan Zheng 10775d4f98a2SYan Zheng /* reloc trees always use full backref */ 10785d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 10795d4f98a2SYan Zheng parent = leaf->start; 10805d4f98a2SYan Zheng else 10815d4f98a2SYan Zheng parent = 0; 10825d4f98a2SYan Zheng 10835d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 10845d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 108582fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 108682fa113fSQu Wenruo 10875d4f98a2SYan Zheng cond_resched(); 10885d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 10895d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 10905d4f98a2SYan Zheng continue; 10915d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 10925d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 10935d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 10945d4f98a2SYan Zheng continue; 10955d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 10965d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 10975d4f98a2SYan Zheng if (bytenr == 0) 10985d4f98a2SYan Zheng continue; 10999569cc20SQu Wenruo if (!in_range(bytenr, rc->block_group->start, 11009569cc20SQu Wenruo rc->block_group->length)) 11015d4f98a2SYan Zheng continue; 11025d4f98a2SYan Zheng 11035d4f98a2SYan Zheng /* 11045d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 11055d4f98a2SYan Zheng * to complete and drop the extent cache 11065d4f98a2SYan Zheng */ 11075d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 11085d4f98a2SYan Zheng if (first) { 11095d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11105d4f98a2SYan Zheng first = 0; 11114a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 11123fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 11135d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11145d4f98a2SYan Zheng } 11154a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 11165d4f98a2SYan Zheng end = key.offset + 11175d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 11185d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 11190b246afaSJeff Mahoney fs_info->sectorsize)); 11200b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 11215d4f98a2SYan Zheng end--; 11225d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1123d0082371SJeff Mahoney key.offset, end); 11245d4f98a2SYan Zheng if (!ret) 11255d4f98a2SYan Zheng continue; 11265d4f98a2SYan Zheng 1127dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1128dcdbc059SNikolay Borisov key.offset, end, 1); 11295d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1130d0082371SJeff Mahoney key.offset, end); 11315d4f98a2SYan Zheng } 11325d4f98a2SYan Zheng } 11335d4f98a2SYan Zheng 11345d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 11355d4f98a2SYan Zheng bytenr, num_bytes); 113683d4cfd4SJosef Bacik if (ret) { 113783d4cfd4SJosef Bacik /* 113883d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 113983d4cfd4SJosef Bacik * in the file extent yet. 114083d4cfd4SJosef Bacik */ 114183d4cfd4SJosef Bacik break; 11423fd0a558SYan, Zheng } 11435d4f98a2SYan Zheng 11445d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 11455d4f98a2SYan Zheng dirty = 1; 11465d4f98a2SYan Zheng 11475d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 114882fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 114982fa113fSQu Wenruo num_bytes, parent); 115082fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1151f42c5da6SNikolay Borisov key.objectid, key.offset, 1152f42c5da6SNikolay Borisov root->root_key.objectid, false); 115382fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 115483d4cfd4SJosef Bacik if (ret) { 115566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 115683d4cfd4SJosef Bacik break; 115783d4cfd4SJosef Bacik } 11585d4f98a2SYan Zheng 1159ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1160ffd4bb2aSQu Wenruo num_bytes, parent); 1161ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1162f42c5da6SNikolay Borisov key.objectid, key.offset, 1163f42c5da6SNikolay Borisov root->root_key.objectid, false); 1164ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 116583d4cfd4SJosef Bacik if (ret) { 116666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 116783d4cfd4SJosef Bacik break; 116883d4cfd4SJosef Bacik } 11695d4f98a2SYan Zheng } 11705d4f98a2SYan Zheng if (dirty) 11715d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 11723fd0a558SYan, Zheng if (inode) 11733fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 117483d4cfd4SJosef Bacik return ret; 11755d4f98a2SYan Zheng } 11765d4f98a2SYan Zheng 11775d4f98a2SYan Zheng static noinline_for_stack 11785d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 11795d4f98a2SYan Zheng struct btrfs_path *path, int level) 11805d4f98a2SYan Zheng { 11815d4f98a2SYan Zheng struct btrfs_disk_key key1; 11825d4f98a2SYan Zheng struct btrfs_disk_key key2; 11835d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 11845d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 11855d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 11865d4f98a2SYan Zheng } 11875d4f98a2SYan Zheng 11885d4f98a2SYan Zheng /* 11895d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 11905d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 11915d4f98a2SYan Zheng * reloc tree was create can be replaced. 11925d4f98a2SYan Zheng * 11935d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 11945d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 11955d4f98a2SYan Zheng * errors, a negative error number is returned. 11965d4f98a2SYan Zheng */ 11973fd0a558SYan, Zheng static noinline_for_stack 11983d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 11995d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 12005d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 12015d4f98a2SYan Zheng int lowest_level, int max_level) 12025d4f98a2SYan Zheng { 12030b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 12045d4f98a2SYan Zheng struct extent_buffer *eb; 12055d4f98a2SYan Zheng struct extent_buffer *parent; 120682fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 12075d4f98a2SYan Zheng struct btrfs_key key; 12085d4f98a2SYan Zheng u64 old_bytenr; 12095d4f98a2SYan Zheng u64 new_bytenr; 12105d4f98a2SYan Zheng u64 old_ptr_gen; 12115d4f98a2SYan Zheng u64 new_ptr_gen; 12125d4f98a2SYan Zheng u64 last_snapshot; 12135d4f98a2SYan Zheng u32 blocksize; 12143fd0a558SYan, Zheng int cow = 0; 12155d4f98a2SYan Zheng int level; 12165d4f98a2SYan Zheng int ret; 12175d4f98a2SYan Zheng int slot; 12185d4f98a2SYan Zheng 12197a9213a9SJosef Bacik ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 12207a9213a9SJosef Bacik ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 12215d4f98a2SYan Zheng 12225d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 12233fd0a558SYan, Zheng again: 12245d4f98a2SYan Zheng slot = path->slots[lowest_level]; 12255d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 12265d4f98a2SYan Zheng 12275d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 12285d4f98a2SYan Zheng level = btrfs_header_level(eb); 12295d4f98a2SYan Zheng 12305d4f98a2SYan Zheng if (level < lowest_level) { 12315d4f98a2SYan Zheng btrfs_tree_unlock(eb); 12325d4f98a2SYan Zheng free_extent_buffer(eb); 12335d4f98a2SYan Zheng return 0; 12345d4f98a2SYan Zheng } 12355d4f98a2SYan Zheng 12363fd0a558SYan, Zheng if (cow) { 12379631e4ccSJosef Bacik ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb, 12389631e4ccSJosef Bacik BTRFS_NESTING_COW); 123945b87c5dSJosef Bacik if (ret) { 124045b87c5dSJosef Bacik btrfs_tree_unlock(eb); 124145b87c5dSJosef Bacik free_extent_buffer(eb); 124245b87c5dSJosef Bacik return ret; 124345b87c5dSJosef Bacik } 12443fd0a558SYan, Zheng } 12455d4f98a2SYan Zheng 12465d4f98a2SYan Zheng if (next_key) { 12475d4f98a2SYan Zheng next_key->objectid = (u64)-1; 12485d4f98a2SYan Zheng next_key->type = (u8)-1; 12495d4f98a2SYan Zheng next_key->offset = (u64)-1; 12505d4f98a2SYan Zheng } 12515d4f98a2SYan Zheng 12525d4f98a2SYan Zheng parent = eb; 12535d4f98a2SYan Zheng while (1) { 12545d4f98a2SYan Zheng level = btrfs_header_level(parent); 12557a9213a9SJosef Bacik ASSERT(level >= lowest_level); 12565d4f98a2SYan Zheng 1257e3b83361SQu Wenruo ret = btrfs_bin_search(parent, &key, &slot); 1258cbca7d59SFilipe Manana if (ret < 0) 1259cbca7d59SFilipe Manana break; 12605d4f98a2SYan Zheng if (ret && slot > 0) 12615d4f98a2SYan Zheng slot--; 12625d4f98a2SYan Zheng 12635d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 12645d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 12655d4f98a2SYan Zheng 12665d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 12670b246afaSJeff Mahoney blocksize = fs_info->nodesize; 12685d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 12695d4f98a2SYan Zheng 12705d4f98a2SYan Zheng if (level <= max_level) { 12715d4f98a2SYan Zheng eb = path->nodes[level]; 12725d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 12735d4f98a2SYan Zheng path->slots[level]); 12745d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 12755d4f98a2SYan Zheng path->slots[level]); 12765d4f98a2SYan Zheng } else { 12775d4f98a2SYan Zheng new_bytenr = 0; 12785d4f98a2SYan Zheng new_ptr_gen = 0; 12795d4f98a2SYan Zheng } 12805d4f98a2SYan Zheng 1281fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 12825d4f98a2SYan Zheng ret = level; 12835d4f98a2SYan Zheng break; 12845d4f98a2SYan Zheng } 12855d4f98a2SYan Zheng 12865d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 12875d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 12883fd0a558SYan, Zheng if (level <= lowest_level) { 12895d4f98a2SYan Zheng ret = 0; 12905d4f98a2SYan Zheng break; 12915d4f98a2SYan Zheng } 12925d4f98a2SYan Zheng 12936b3426beSJosef Bacik eb = btrfs_read_node_slot(parent, slot); 129464c043deSLiu Bo if (IS_ERR(eb)) { 129564c043deSLiu Bo ret = PTR_ERR(eb); 1296264813acSLiu Bo break; 1297416bc658SJosef Bacik } 12985d4f98a2SYan Zheng btrfs_tree_lock(eb); 12993fd0a558SYan, Zheng if (cow) { 13005d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 13019631e4ccSJosef Bacik slot, &eb, 13029631e4ccSJosef Bacik BTRFS_NESTING_COW); 130345b87c5dSJosef Bacik if (ret) { 130445b87c5dSJosef Bacik btrfs_tree_unlock(eb); 130545b87c5dSJosef Bacik free_extent_buffer(eb); 130645b87c5dSJosef Bacik break; 130745b87c5dSJosef Bacik } 13085d4f98a2SYan Zheng } 13095d4f98a2SYan Zheng 13105d4f98a2SYan Zheng btrfs_tree_unlock(parent); 13115d4f98a2SYan Zheng free_extent_buffer(parent); 13125d4f98a2SYan Zheng 13135d4f98a2SYan Zheng parent = eb; 13145d4f98a2SYan Zheng continue; 13155d4f98a2SYan Zheng } 13165d4f98a2SYan Zheng 13173fd0a558SYan, Zheng if (!cow) { 13183fd0a558SYan, Zheng btrfs_tree_unlock(parent); 13193fd0a558SYan, Zheng free_extent_buffer(parent); 13203fd0a558SYan, Zheng cow = 1; 13213fd0a558SYan, Zheng goto again; 13223fd0a558SYan, Zheng } 13233fd0a558SYan, Zheng 13245d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 13255d4f98a2SYan Zheng path->slots[level]); 1326b3b4aa74SDavid Sterba btrfs_release_path(path); 13275d4f98a2SYan Zheng 13285d4f98a2SYan Zheng path->lowest_level = level; 13295d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 13305d4f98a2SYan Zheng path->lowest_level = 0; 13310e9873e2SJosef Bacik if (ret) { 13320e9873e2SJosef Bacik if (ret > 0) 13330e9873e2SJosef Bacik ret = -ENOENT; 13340e9873e2SJosef Bacik break; 13350e9873e2SJosef Bacik } 13365d4f98a2SYan Zheng 13375d4f98a2SYan Zheng /* 1338824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1339824d8dffSQu Wenruo * 1340824d8dffSQu Wenruo * We must trace both trees. 1341824d8dffSQu Wenruo * 1) Tree reloc subtree 1342824d8dffSQu Wenruo * If not traced, we will leak data numbers 1343824d8dffSQu Wenruo * 2) Fs subtree 1344824d8dffSQu Wenruo * If not traced, we will double count old data 1345f616f5cdSQu Wenruo * 1346f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1347f616f5cdSQu Wenruo * the swapped tree blocks. 1348f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1349f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1350824d8dffSQu Wenruo */ 1351370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1352370a11b8SQu Wenruo rc->block_group, parent, slot, 1353370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1354370a11b8SQu Wenruo last_snapshot); 1355370a11b8SQu Wenruo if (ret < 0) 1356370a11b8SQu Wenruo break; 1357824d8dffSQu Wenruo /* 13585d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 13595d4f98a2SYan Zheng */ 13605d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 13615d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 13625d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 13635d4f98a2SYan Zheng 13645d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 13655d4f98a2SYan Zheng path->slots[level], old_bytenr); 13665d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 13675d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 13685d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 13695d4f98a2SYan Zheng 137082fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 137182fa113fSQu Wenruo blocksize, path->nodes[level]->start); 1372f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1373f42c5da6SNikolay Borisov 0, true); 137482fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1375253e258cSJosef Bacik if (ret) { 1376253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1377253e258cSJosef Bacik break; 1378253e258cSJosef Bacik } 137982fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 138082fa113fSQu Wenruo blocksize, 0); 1381f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0, 1382f42c5da6SNikolay Borisov true); 138382fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1384253e258cSJosef Bacik if (ret) { 1385253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1386253e258cSJosef Bacik break; 1387253e258cSJosef Bacik } 13885d4f98a2SYan Zheng 1389ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1390ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1391f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1392f42c5da6SNikolay Borisov 0, true); 1393ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1394253e258cSJosef Bacik if (ret) { 1395253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1396253e258cSJosef Bacik break; 1397253e258cSJosef Bacik } 13985d4f98a2SYan Zheng 1399ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1400ffd4bb2aSQu Wenruo blocksize, 0); 1401f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 1402f42c5da6SNikolay Borisov 0, true); 1403ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1404253e258cSJosef Bacik if (ret) { 1405253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1406253e258cSJosef Bacik break; 1407253e258cSJosef Bacik } 14085d4f98a2SYan Zheng 14095d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 14105d4f98a2SYan Zheng 14115d4f98a2SYan Zheng ret = level; 14125d4f98a2SYan Zheng break; 14135d4f98a2SYan Zheng } 14145d4f98a2SYan Zheng btrfs_tree_unlock(parent); 14155d4f98a2SYan Zheng free_extent_buffer(parent); 14165d4f98a2SYan Zheng return ret; 14175d4f98a2SYan Zheng } 14185d4f98a2SYan Zheng 14195d4f98a2SYan Zheng /* 14205d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 14215d4f98a2SYan Zheng */ 14225d4f98a2SYan Zheng static noinline_for_stack 14235d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14245d4f98a2SYan Zheng int *level) 14255d4f98a2SYan Zheng { 14265d4f98a2SYan Zheng struct extent_buffer *eb; 14275d4f98a2SYan Zheng int i; 14285d4f98a2SYan Zheng u64 last_snapshot; 14295d4f98a2SYan Zheng u32 nritems; 14305d4f98a2SYan Zheng 14315d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14325d4f98a2SYan Zheng 14335d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 14345d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14355d4f98a2SYan Zheng path->nodes[i] = NULL; 14365d4f98a2SYan Zheng } 14375d4f98a2SYan Zheng 14385d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 14395d4f98a2SYan Zheng eb = path->nodes[i]; 14405d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14415d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 14425d4f98a2SYan Zheng path->slots[i]++; 14435d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 14445d4f98a2SYan Zheng last_snapshot) 14455d4f98a2SYan Zheng continue; 14465d4f98a2SYan Zheng 14475d4f98a2SYan Zheng *level = i; 14485d4f98a2SYan Zheng return 0; 14495d4f98a2SYan Zheng } 14505d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14515d4f98a2SYan Zheng path->nodes[i] = NULL; 14525d4f98a2SYan Zheng } 14535d4f98a2SYan Zheng return 1; 14545d4f98a2SYan Zheng } 14555d4f98a2SYan Zheng 14565d4f98a2SYan Zheng /* 14575d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 14585d4f98a2SYan Zheng */ 14595d4f98a2SYan Zheng static noinline_for_stack 14605d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14615d4f98a2SYan Zheng int *level) 14625d4f98a2SYan Zheng { 14635d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 14645d4f98a2SYan Zheng int i; 14655d4f98a2SYan Zheng u64 ptr_gen = 0; 14665d4f98a2SYan Zheng u64 last_snapshot; 14675d4f98a2SYan Zheng u32 nritems; 14685d4f98a2SYan Zheng 14695d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14705d4f98a2SYan Zheng 14715d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 14725d4f98a2SYan Zheng eb = path->nodes[i]; 14735d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14745d4f98a2SYan Zheng while (path->slots[i] < nritems) { 14755d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 14765d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 14775d4f98a2SYan Zheng break; 14785d4f98a2SYan Zheng path->slots[i]++; 14795d4f98a2SYan Zheng } 14805d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 14815d4f98a2SYan Zheng if (i == *level) 14825d4f98a2SYan Zheng break; 14835d4f98a2SYan Zheng *level = i + 1; 14845d4f98a2SYan Zheng return 0; 14855d4f98a2SYan Zheng } 14865d4f98a2SYan Zheng if (i == 1) { 14875d4f98a2SYan Zheng *level = i; 14885d4f98a2SYan Zheng return 0; 14895d4f98a2SYan Zheng } 14905d4f98a2SYan Zheng 14918ef385bbSJosef Bacik eb = btrfs_read_node_slot(eb, path->slots[i]); 14928ef385bbSJosef Bacik if (IS_ERR(eb)) 149364c043deSLiu Bo return PTR_ERR(eb); 14945d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 14955d4f98a2SYan Zheng path->nodes[i - 1] = eb; 14965d4f98a2SYan Zheng path->slots[i - 1] = 0; 14975d4f98a2SYan Zheng } 14985d4f98a2SYan Zheng return 1; 14995d4f98a2SYan Zheng } 15005d4f98a2SYan Zheng 15015d4f98a2SYan Zheng /* 15025d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 15035d4f98a2SYan Zheng * [min_key, max_key) 15045d4f98a2SYan Zheng */ 15055d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 15065d4f98a2SYan Zheng struct btrfs_key *min_key, 15075d4f98a2SYan Zheng struct btrfs_key *max_key) 15085d4f98a2SYan Zheng { 15090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15105d4f98a2SYan Zheng struct inode *inode = NULL; 15115d4f98a2SYan Zheng u64 objectid; 15125d4f98a2SYan Zheng u64 start, end; 151333345d01SLi Zefan u64 ino; 15145d4f98a2SYan Zheng 15155d4f98a2SYan Zheng objectid = min_key->objectid; 15165d4f98a2SYan Zheng while (1) { 15175d4f98a2SYan Zheng cond_resched(); 15185d4f98a2SYan Zheng iput(inode); 15195d4f98a2SYan Zheng 15205d4f98a2SYan Zheng if (objectid > max_key->objectid) 15215d4f98a2SYan Zheng break; 15225d4f98a2SYan Zheng 15235d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 15245d4f98a2SYan Zheng if (!inode) 15255d4f98a2SYan Zheng break; 15264a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 15275d4f98a2SYan Zheng 152833345d01SLi Zefan if (ino > max_key->objectid) { 15295d4f98a2SYan Zheng iput(inode); 15305d4f98a2SYan Zheng break; 15315d4f98a2SYan Zheng } 15325d4f98a2SYan Zheng 153333345d01SLi Zefan objectid = ino + 1; 15345d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 15355d4f98a2SYan Zheng continue; 15365d4f98a2SYan Zheng 153733345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 15385d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 15395d4f98a2SYan Zheng continue; 15405d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 15415d4f98a2SYan Zheng start = 0; 15425d4f98a2SYan Zheng else { 15435d4f98a2SYan Zheng start = min_key->offset; 15440b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 15455d4f98a2SYan Zheng } 15465d4f98a2SYan Zheng } else { 15475d4f98a2SYan Zheng start = 0; 15485d4f98a2SYan Zheng } 15495d4f98a2SYan Zheng 155033345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 15515d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 15525d4f98a2SYan Zheng continue; 15535d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 15545d4f98a2SYan Zheng end = (u64)-1; 15555d4f98a2SYan Zheng } else { 15565d4f98a2SYan Zheng if (max_key->offset == 0) 15575d4f98a2SYan Zheng continue; 15585d4f98a2SYan Zheng end = max_key->offset; 15590b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 15605d4f98a2SYan Zheng end--; 15615d4f98a2SYan Zheng } 15625d4f98a2SYan Zheng } else { 15635d4f98a2SYan Zheng end = (u64)-1; 15645d4f98a2SYan Zheng } 15655d4f98a2SYan Zheng 15665d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 1567d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 1568dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 1569d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 15705d4f98a2SYan Zheng } 15715d4f98a2SYan Zheng return 0; 15725d4f98a2SYan Zheng } 15735d4f98a2SYan Zheng 15745d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 15755d4f98a2SYan Zheng struct btrfs_key *key) 15765d4f98a2SYan Zheng 15775d4f98a2SYan Zheng { 15785d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 15795d4f98a2SYan Zheng if (!path->nodes[level]) 15805d4f98a2SYan Zheng break; 15815d4f98a2SYan Zheng if (path->slots[level] + 1 < 15825d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 15835d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 15845d4f98a2SYan Zheng path->slots[level] + 1); 15855d4f98a2SYan Zheng return 0; 15865d4f98a2SYan Zheng } 15875d4f98a2SYan Zheng level++; 15885d4f98a2SYan Zheng } 15895d4f98a2SYan Zheng return 1; 15905d4f98a2SYan Zheng } 15915d4f98a2SYan Zheng 15925d4f98a2SYan Zheng /* 1593d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 1594d2311e69SQu Wenruo */ 1595ac54da6cSJosef Bacik static int insert_dirty_subvol(struct btrfs_trans_handle *trans, 1596d2311e69SQu Wenruo struct reloc_control *rc, 1597d2311e69SQu Wenruo struct btrfs_root *root) 1598d2311e69SQu Wenruo { 1599d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1600d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 16017934133fSJosef Bacik int ret; 1602d2311e69SQu Wenruo 1603d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 1604d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 1605d2311e69SQu Wenruo ASSERT(reloc_root); 1606d2311e69SQu Wenruo 1607d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 1608d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 1609d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 1610c8422684SDavid Sterba btrfs_set_root_drop_level(reloc_root_item, 0); 1611d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 16127934133fSJosef Bacik ret = btrfs_update_reloc_root(trans, root); 16137934133fSJosef Bacik if (ret) 16147934133fSJosef Bacik return ret; 1615d2311e69SQu Wenruo 1616d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 161700246528SJosef Bacik btrfs_grab_root(root); 1618d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 1619d2311e69SQu Wenruo } 1620ac54da6cSJosef Bacik 1621ac54da6cSJosef Bacik return 0; 1622d2311e69SQu Wenruo } 1623d2311e69SQu Wenruo 1624d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 1625d2311e69SQu Wenruo { 1626d2311e69SQu Wenruo struct btrfs_root *root; 1627d2311e69SQu Wenruo struct btrfs_root *next; 1628d2311e69SQu Wenruo int ret = 0; 162930d40577SQu Wenruo int ret2; 1630d2311e69SQu Wenruo 1631d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 1632d2311e69SQu Wenruo reloc_dirty_list) { 163330d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 163430d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 1635d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1636d2311e69SQu Wenruo 1637d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 1638d2311e69SQu Wenruo root->reloc_root = NULL; 16396282675eSQu Wenruo /* 16406282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 16416282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 16426282675eSQu Wenruo */ 16436282675eSQu Wenruo smp_wmb(); 16441fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 1645f28de8d8SJosef Bacik if (reloc_root) { 1646f44deb74SJosef Bacik /* 1647f44deb74SJosef Bacik * btrfs_drop_snapshot drops our ref we hold for 1648f44deb74SJosef Bacik * ->reloc_root. If it fails however we must 1649f44deb74SJosef Bacik * drop the ref ourselves. 1650f44deb74SJosef Bacik */ 1651f28de8d8SJosef Bacik ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); 1652f44deb74SJosef Bacik if (ret2 < 0) { 1653f44deb74SJosef Bacik btrfs_put_root(reloc_root); 1654f44deb74SJosef Bacik if (!ret) 1655f28de8d8SJosef Bacik ret = ret2; 1656f28de8d8SJosef Bacik } 1657f44deb74SJosef Bacik } 165800246528SJosef Bacik btrfs_put_root(root); 165930d40577SQu Wenruo } else { 166030d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 16610078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(root, 0, 1); 1662f44deb74SJosef Bacik if (ret2 < 0) { 1663f44deb74SJosef Bacik btrfs_put_root(root); 1664f44deb74SJosef Bacik if (!ret) 166530d40577SQu Wenruo ret = ret2; 166630d40577SQu Wenruo } 1667d2311e69SQu Wenruo } 1668f44deb74SJosef Bacik } 1669d2311e69SQu Wenruo return ret; 1670d2311e69SQu Wenruo } 1671d2311e69SQu Wenruo 1672d2311e69SQu Wenruo /* 16735d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 16745d4f98a2SYan Zheng * fs tree. 16755d4f98a2SYan Zheng */ 16765d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 16775d4f98a2SYan Zheng struct btrfs_root *root) 16785d4f98a2SYan Zheng { 16790b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 16805d4f98a2SYan Zheng struct btrfs_key key; 16815d4f98a2SYan Zheng struct btrfs_key next_key; 16829e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 16835d4f98a2SYan Zheng struct btrfs_root *reloc_root; 16845d4f98a2SYan Zheng struct btrfs_root_item *root_item; 16855d4f98a2SYan Zheng struct btrfs_path *path; 16863fd0a558SYan, Zheng struct extent_buffer *leaf; 1687fca3a45dSJosef Bacik int reserve_level; 16885d4f98a2SYan Zheng int level; 16895d4f98a2SYan Zheng int max_level; 16905d4f98a2SYan Zheng int replaced = 0; 1691c6a592f2SNikolay Borisov int ret = 0; 16923fd0a558SYan, Zheng u32 min_reserved; 16935d4f98a2SYan Zheng 16945d4f98a2SYan Zheng path = btrfs_alloc_path(); 16955d4f98a2SYan Zheng if (!path) 16965d4f98a2SYan Zheng return -ENOMEM; 1697e4058b54SDavid Sterba path->reada = READA_FORWARD; 16985d4f98a2SYan Zheng 16995d4f98a2SYan Zheng reloc_root = root->reloc_root; 17005d4f98a2SYan Zheng root_item = &reloc_root->root_item; 17015d4f98a2SYan Zheng 17025d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 17035d4f98a2SYan Zheng level = btrfs_root_level(root_item); 170467439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 17055d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 17065d4f98a2SYan Zheng path->slots[level] = 0; 17075d4f98a2SYan Zheng } else { 17085d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 17095d4f98a2SYan Zheng 1710c8422684SDavid Sterba level = btrfs_root_drop_level(root_item); 17115d4f98a2SYan Zheng BUG_ON(level == 0); 17125d4f98a2SYan Zheng path->lowest_level = level; 17135d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 171433c66f43SYan Zheng path->lowest_level = 0; 17155d4f98a2SYan Zheng if (ret < 0) { 17165d4f98a2SYan Zheng btrfs_free_path(path); 17175d4f98a2SYan Zheng return ret; 17185d4f98a2SYan Zheng } 17195d4f98a2SYan Zheng 17205d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 17215d4f98a2SYan Zheng path->slots[level]); 17225d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 17235d4f98a2SYan Zheng 17245d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 17255d4f98a2SYan Zheng } 17265d4f98a2SYan Zheng 172744d354abSQu Wenruo /* 172844d354abSQu Wenruo * In merge_reloc_root(), we modify the upper level pointer to swap the 172944d354abSQu Wenruo * tree blocks between reloc tree and subvolume tree. Thus for tree 173044d354abSQu Wenruo * block COW, we COW at most from level 1 to root level for each tree. 173144d354abSQu Wenruo * 173244d354abSQu Wenruo * Thus the needed metadata size is at most root_level * nodesize, 173344d354abSQu Wenruo * and * 2 since we have two trees to COW. 173444d354abSQu Wenruo */ 1735fca3a45dSJosef Bacik reserve_level = max_t(int, 1, btrfs_root_level(root_item)); 1736fca3a45dSJosef Bacik min_reserved = fs_info->nodesize * reserve_level * 2; 17375d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 17385d4f98a2SYan Zheng 17395d4f98a2SYan Zheng while (1) { 17409270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 17419270501cSJosef Bacik min_reserved, 174244d354abSQu Wenruo BTRFS_RESERVE_FLUSH_LIMIT); 1743c6a592f2SNikolay Borisov if (ret) 17449e6a0c52SJosef Bacik goto out; 17459e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 17469e6a0c52SJosef Bacik if (IS_ERR(trans)) { 1747c6a592f2SNikolay Borisov ret = PTR_ERR(trans); 17489e6a0c52SJosef Bacik trans = NULL; 17499e6a0c52SJosef Bacik goto out; 17509e6a0c52SJosef Bacik } 17512abc726aSJosef Bacik 17522abc726aSJosef Bacik /* 17532abc726aSJosef Bacik * At this point we no longer have a reloc_control, so we can't 17542abc726aSJosef Bacik * depend on btrfs_init_reloc_root to update our last_trans. 17552abc726aSJosef Bacik * 17562abc726aSJosef Bacik * But that's ok, we started the trans handle on our 17572abc726aSJosef Bacik * corresponding fs_root, which means it's been added to the 17582abc726aSJosef Bacik * dirty list. At commit time we'll still call 17592abc726aSJosef Bacik * btrfs_update_reloc_root() and update our root item 17602abc726aSJosef Bacik * appropriately. 17612abc726aSJosef Bacik */ 17622abc726aSJosef Bacik reloc_root->last_trans = trans->transid; 17639e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 17643fd0a558SYan, Zheng 17653fd0a558SYan, Zheng replaced = 0; 17665d4f98a2SYan Zheng max_level = level; 17675d4f98a2SYan Zheng 17685d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 1769c6a592f2SNikolay Borisov if (ret < 0) 17705d4f98a2SYan Zheng goto out; 17715d4f98a2SYan Zheng if (ret > 0) 17725d4f98a2SYan Zheng break; 17735d4f98a2SYan Zheng 17745d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 17755d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 17765d4f98a2SYan Zheng ret = 0; 17775d4f98a2SYan Zheng } else { 17783d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 17793fd0a558SYan, Zheng &next_key, level, max_level); 17805d4f98a2SYan Zheng } 1781c6a592f2SNikolay Borisov if (ret < 0) 17825d4f98a2SYan Zheng goto out; 17835d4f98a2SYan Zheng if (ret > 0) { 17845d4f98a2SYan Zheng level = ret; 17855d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 17865d4f98a2SYan Zheng path->slots[level]); 17875d4f98a2SYan Zheng replaced = 1; 17885d4f98a2SYan Zheng } 17895d4f98a2SYan Zheng 17905d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 17915d4f98a2SYan Zheng if (ret > 0) 17925d4f98a2SYan Zheng break; 17935d4f98a2SYan Zheng 17945d4f98a2SYan Zheng BUG_ON(level == 0); 17955d4f98a2SYan Zheng /* 17965d4f98a2SYan Zheng * save the merging progress in the drop_progress. 17975d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 17985d4f98a2SYan Zheng */ 17995d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 18005d4f98a2SYan Zheng path->slots[level]); 1801c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, level); 18025d4f98a2SYan Zheng 18033a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18049e6a0c52SJosef Bacik trans = NULL; 18055d4f98a2SYan Zheng 18062ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18075d4f98a2SYan Zheng 18085d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18095d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18105d4f98a2SYan Zheng } 18115d4f98a2SYan Zheng 18125d4f98a2SYan Zheng /* 18135d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 18145d4f98a2SYan Zheng * relocated and the block is tree root. 18155d4f98a2SYan Zheng */ 18165d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 18179631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf, 18189631e4ccSJosef Bacik BTRFS_NESTING_COW); 18195d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 18205d4f98a2SYan Zheng free_extent_buffer(leaf); 18215d4f98a2SYan Zheng out: 18225d4f98a2SYan Zheng btrfs_free_path(path); 18235d4f98a2SYan Zheng 1824ac54da6cSJosef Bacik if (ret == 0) { 1825ac54da6cSJosef Bacik ret = insert_dirty_subvol(trans, rc, root); 1826ac54da6cSJosef Bacik if (ret) 1827ac54da6cSJosef Bacik btrfs_abort_transaction(trans, ret); 1828ac54da6cSJosef Bacik } 18295d4f98a2SYan Zheng 18309e6a0c52SJosef Bacik if (trans) 18313a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18325d4f98a2SYan Zheng 18332ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18345d4f98a2SYan Zheng 18355d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18365d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18375d4f98a2SYan Zheng 1838c6a592f2SNikolay Borisov return ret; 18395d4f98a2SYan Zheng } 18405d4f98a2SYan Zheng 18413fd0a558SYan, Zheng static noinline_for_stack 18423fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 18435d4f98a2SYan Zheng { 18443fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 18450b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 18463fd0a558SYan, Zheng struct btrfs_root *reloc_root; 18475d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 18483fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 18493fd0a558SYan, Zheng u64 num_bytes = 0; 18503fd0a558SYan, Zheng int ret; 18513fd0a558SYan, Zheng 18520b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 18530b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 18543fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 18550b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 18567585717fSChris Mason 18573fd0a558SYan, Zheng again: 18583fd0a558SYan, Zheng if (!err) { 18593fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 18609270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes, 186108e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 18623fd0a558SYan, Zheng if (ret) 18633fd0a558SYan, Zheng err = ret; 18643fd0a558SYan, Zheng } 18653fd0a558SYan, Zheng 18667a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 18673612b495STsutomu Itoh if (IS_ERR(trans)) { 18683612b495STsutomu Itoh if (!err) 18692ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 187063f018beSNikolay Borisov num_bytes, NULL); 18713612b495STsutomu Itoh return PTR_ERR(trans); 18723612b495STsutomu Itoh } 18733fd0a558SYan, Zheng 18743fd0a558SYan, Zheng if (!err) { 18753fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 18763a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18772ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 187863f018beSNikolay Borisov num_bytes, NULL); 18793fd0a558SYan, Zheng goto again; 18803fd0a558SYan, Zheng } 18813fd0a558SYan, Zheng } 18823fd0a558SYan, Zheng 18833fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 18843fd0a558SYan, Zheng 18853fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 18863fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 18873fd0a558SYan, Zheng struct btrfs_root, root_list); 18883fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 18893fd0a558SYan, Zheng 1890a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1891a820feb5SDavid Sterba false); 1892e0b085b0SJosef Bacik if (IS_ERR(root)) { 1893e0b085b0SJosef Bacik /* 1894e0b085b0SJosef Bacik * Even if we have an error we need this reloc root 1895e0b085b0SJosef Bacik * back on our list so we can clean up properly. 1896e0b085b0SJosef Bacik */ 1897e0b085b0SJosef Bacik list_add(&reloc_root->root_list, &reloc_roots); 1898e0b085b0SJosef Bacik btrfs_abort_transaction(trans, (int)PTR_ERR(root)); 1899e0b085b0SJosef Bacik if (!err) 1900e0b085b0SJosef Bacik err = PTR_ERR(root); 1901e0b085b0SJosef Bacik break; 1902e0b085b0SJosef Bacik } 1903e0b085b0SJosef Bacik ASSERT(root->reloc_root == reloc_root); 19043fd0a558SYan, Zheng 19053fd0a558SYan, Zheng /* 19063fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 19073fd0a558SYan, Zheng * knows it should resumes merging 19083fd0a558SYan, Zheng */ 19093fd0a558SYan, Zheng if (!err) 19103fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 1911bbae13f8SJosef Bacik ret = btrfs_update_reloc_root(trans, root); 19123fd0a558SYan, Zheng 1913bbae13f8SJosef Bacik /* 1914bbae13f8SJosef Bacik * Even if we have an error we need this reloc root back on our 1915bbae13f8SJosef Bacik * list so we can clean up properly. 1916bbae13f8SJosef Bacik */ 19173fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 191800246528SJosef Bacik btrfs_put_root(root); 1919bbae13f8SJosef Bacik 1920bbae13f8SJosef Bacik if (ret) { 1921bbae13f8SJosef Bacik btrfs_abort_transaction(trans, ret); 1922bbae13f8SJosef Bacik if (!err) 1923bbae13f8SJosef Bacik err = ret; 1924bbae13f8SJosef Bacik break; 1925bbae13f8SJosef Bacik } 19263fd0a558SYan, Zheng } 19273fd0a558SYan, Zheng 19283fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 19293fd0a558SYan, Zheng 19303fd0a558SYan, Zheng if (!err) 1931fb686c68SJosef Bacik err = btrfs_commit_transaction(trans); 19323fd0a558SYan, Zheng else 19333a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19343fd0a558SYan, Zheng return err; 19353fd0a558SYan, Zheng } 19363fd0a558SYan, Zheng 19373fd0a558SYan, Zheng static noinline_for_stack 1938aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 1939aca1bba6SLiu Bo { 1940a7571232SNikolay Borisov struct btrfs_root *reloc_root, *tmp; 1941aca1bba6SLiu Bo 1942a7571232SNikolay Borisov list_for_each_entry_safe(reloc_root, tmp, list, root_list) 1943bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 1944aca1bba6SLiu Bo } 1945aca1bba6SLiu Bo 1946aca1bba6SLiu Bo static noinline_for_stack 194794404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 19483fd0a558SYan, Zheng { 19490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 19505d4f98a2SYan Zheng struct btrfs_root *root; 19515d4f98a2SYan Zheng struct btrfs_root *reloc_root; 19523fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 19533fd0a558SYan, Zheng int found = 0; 1954aca1bba6SLiu Bo int ret = 0; 19553fd0a558SYan, Zheng again: 19563fd0a558SYan, Zheng root = rc->extent_root; 19577585717fSChris Mason 19587585717fSChris Mason /* 19597585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 19607585717fSChris Mason * we have to make sure nobody is in the middle of 19617585717fSChris Mason * adding their roots to the list while we are 19627585717fSChris Mason * doing this splice 19637585717fSChris Mason */ 19640b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 19653fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 19660b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 19675d4f98a2SYan Zheng 19683fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 19693fd0a558SYan, Zheng found = 1; 19703fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 19713fd0a558SYan, Zheng struct btrfs_root, root_list); 19725d4f98a2SYan Zheng 1973a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1974a820feb5SDavid Sterba false); 19755d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 197624213fa4SJosef Bacik if (IS_ERR(root)) { 197724213fa4SJosef Bacik /* 197824213fa4SJosef Bacik * For recovery we read the fs roots on mount, 197924213fa4SJosef Bacik * and if we didn't find the root then we marked 198024213fa4SJosef Bacik * the reloc root as a garbage root. For normal 198124213fa4SJosef Bacik * relocation obviously the root should exist in 198224213fa4SJosef Bacik * memory. However there's no reason we can't 198324213fa4SJosef Bacik * handle the error properly here just in case. 198424213fa4SJosef Bacik */ 198524213fa4SJosef Bacik ASSERT(0); 198624213fa4SJosef Bacik ret = PTR_ERR(root); 198724213fa4SJosef Bacik goto out; 198824213fa4SJosef Bacik } 198924213fa4SJosef Bacik if (root->reloc_root != reloc_root) { 199024213fa4SJosef Bacik /* 199124213fa4SJosef Bacik * This is actually impossible without something 199224213fa4SJosef Bacik * going really wrong (like weird race condition 199324213fa4SJosef Bacik * or cosmic rays). 199424213fa4SJosef Bacik */ 199524213fa4SJosef Bacik ASSERT(0); 199624213fa4SJosef Bacik ret = -EINVAL; 199724213fa4SJosef Bacik goto out; 199824213fa4SJosef Bacik } 19993fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 200000246528SJosef Bacik btrfs_put_root(root); 2001b37b39cdSJosef Bacik if (ret) { 200225e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 200325e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 200425e293c2SWang Shilong &reloc_roots); 2005aca1bba6SLiu Bo goto out; 2006b37b39cdSJosef Bacik } 20073fd0a558SYan, Zheng } else { 200851415b6cSQu Wenruo if (!IS_ERR(root)) { 200951415b6cSQu Wenruo if (root->reloc_root == reloc_root) { 201051415b6cSQu Wenruo root->reloc_root = NULL; 201151415b6cSQu Wenruo btrfs_put_root(reloc_root); 201251415b6cSQu Wenruo } 20131dae7e0eSQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, 20141dae7e0eSQu Wenruo &root->state); 201551415b6cSQu Wenruo btrfs_put_root(root); 201651415b6cSQu Wenruo } 201751415b6cSQu Wenruo 20183fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 201930d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 202030d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 202130d40577SQu Wenruo &rc->dirty_subvol_roots); 20223fd0a558SYan, Zheng } 20235d4f98a2SYan Zheng } 20245d4f98a2SYan Zheng 20253fd0a558SYan, Zheng if (found) { 20263fd0a558SYan, Zheng found = 0; 20273fd0a558SYan, Zheng goto again; 20285d4f98a2SYan Zheng } 2029aca1bba6SLiu Bo out: 2030aca1bba6SLiu Bo if (ret) { 20310b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2032aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2033467bb1d2SWang Shilong 2034467bb1d2SWang Shilong /* new reloc root may be added */ 20350b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2036467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 20370b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2038467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2039aca1bba6SLiu Bo } 2040aca1bba6SLiu Bo 20417b7b7431SJosef Bacik /* 20427b7b7431SJosef Bacik * We used to have 20437b7b7431SJosef Bacik * 20447b7b7431SJosef Bacik * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 20457b7b7431SJosef Bacik * 20467b7b7431SJosef Bacik * here, but it's wrong. If we fail to start the transaction in 20477b7b7431SJosef Bacik * prepare_to_merge() we will have only 0 ref reloc roots, none of which 20487b7b7431SJosef Bacik * have actually been removed from the reloc_root_tree rb tree. This is 20497b7b7431SJosef Bacik * fine because we're bailing here, and we hold a reference on the root 20507b7b7431SJosef Bacik * for the list that holds it, so these roots will be cleaned up when we 20517b7b7431SJosef Bacik * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root 20527b7b7431SJosef Bacik * will be cleaned up on unmount. 20537b7b7431SJosef Bacik * 20547b7b7431SJosef Bacik * The remaining nodes will be cleaned up by free_reloc_control. 20557b7b7431SJosef Bacik */ 20565d4f98a2SYan Zheng } 20575d4f98a2SYan Zheng 20585d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 20595d4f98a2SYan Zheng { 20605d4f98a2SYan Zheng struct tree_block *block; 20615d4f98a2SYan Zheng struct rb_node *rb_node; 20625d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 20635d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 20645d4f98a2SYan Zheng rb_erase(rb_node, blocks); 20655d4f98a2SYan Zheng kfree(block); 20665d4f98a2SYan Zheng } 20675d4f98a2SYan Zheng } 20685d4f98a2SYan Zheng 20695d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 20705d4f98a2SYan Zheng struct btrfs_root *reloc_root) 20715d4f98a2SYan Zheng { 20720b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 20735d4f98a2SYan Zheng struct btrfs_root *root; 2074442b1ac5SJosef Bacik int ret; 20755d4f98a2SYan Zheng 20765d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 20775d4f98a2SYan Zheng return 0; 20785d4f98a2SYan Zheng 2079a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); 2080404bccbcSJosef Bacik 2081404bccbcSJosef Bacik /* 2082404bccbcSJosef Bacik * This should succeed, since we can't have a reloc root without having 2083404bccbcSJosef Bacik * already looked up the actual root and created the reloc root for this 2084404bccbcSJosef Bacik * root. 2085404bccbcSJosef Bacik * 2086404bccbcSJosef Bacik * However if there's some sort of corruption where we have a ref to a 2087404bccbcSJosef Bacik * reloc root without a corresponding root this could return ENOENT. 2088404bccbcSJosef Bacik */ 2089404bccbcSJosef Bacik if (IS_ERR(root)) { 2090404bccbcSJosef Bacik ASSERT(0); 2091404bccbcSJosef Bacik return PTR_ERR(root); 2092404bccbcSJosef Bacik } 2093404bccbcSJosef Bacik if (root->reloc_root != reloc_root) { 2094404bccbcSJosef Bacik ASSERT(0); 2095404bccbcSJosef Bacik btrfs_err(fs_info, 2096404bccbcSJosef Bacik "root %llu has two reloc roots associated with it", 2097404bccbcSJosef Bacik reloc_root->root_key.offset); 2098404bccbcSJosef Bacik btrfs_put_root(root); 2099404bccbcSJosef Bacik return -EUCLEAN; 2100404bccbcSJosef Bacik } 2101442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 210200246528SJosef Bacik btrfs_put_root(root); 21035d4f98a2SYan Zheng 2104442b1ac5SJosef Bacik return ret; 21055d4f98a2SYan Zheng } 21065d4f98a2SYan Zheng 21073fd0a558SYan, Zheng static noinline_for_stack 21083fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 21093fd0a558SYan, Zheng struct reloc_control *rc, 2110a26195a5SQu Wenruo struct btrfs_backref_node *node, 2111a26195a5SQu Wenruo struct btrfs_backref_edge *edges[]) 21125d4f98a2SYan Zheng { 2113a26195a5SQu Wenruo struct btrfs_backref_node *next; 21145d4f98a2SYan Zheng struct btrfs_root *root; 21153fd0a558SYan, Zheng int index = 0; 211692de551bSJosef Bacik int ret; 21173fd0a558SYan, Zheng 21185d4f98a2SYan Zheng next = node; 21195d4f98a2SYan Zheng while (1) { 21205d4f98a2SYan Zheng cond_resched(); 21215d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 21225d4f98a2SYan Zheng root = next->root; 21238ee66afeSJosef Bacik 21248ee66afeSJosef Bacik /* 21258ee66afeSJosef Bacik * If there is no root, then our references for this block are 21268ee66afeSJosef Bacik * incomplete, as we should be able to walk all the way up to a 21278ee66afeSJosef Bacik * block that is owned by a root. 21288ee66afeSJosef Bacik * 21298ee66afeSJosef Bacik * This path is only for SHAREABLE roots, so if we come upon a 21308ee66afeSJosef Bacik * non-SHAREABLE root then we have backrefs that resolve 21318ee66afeSJosef Bacik * improperly. 21328ee66afeSJosef Bacik * 21338ee66afeSJosef Bacik * Both of these cases indicate file system corruption, or a bug 21348ee66afeSJosef Bacik * in the backref walking code. 21358ee66afeSJosef Bacik */ 21368ee66afeSJosef Bacik if (!root) { 21378ee66afeSJosef Bacik ASSERT(0); 21388ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21398ee66afeSJosef Bacik "bytenr %llu doesn't have a backref path ending in a root", 21408ee66afeSJosef Bacik node->bytenr); 21418ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21428ee66afeSJosef Bacik } 21438ee66afeSJosef Bacik if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 21448ee66afeSJosef Bacik ASSERT(0); 21458ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21468ee66afeSJosef Bacik "bytenr %llu has multiple refs with one ending in a non-shareable root", 21478ee66afeSJosef Bacik node->bytenr); 21488ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21498ee66afeSJosef Bacik } 21505d4f98a2SYan Zheng 21515d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 215292de551bSJosef Bacik ret = record_reloc_root_in_trans(trans, root); 215392de551bSJosef Bacik if (ret) 215492de551bSJosef Bacik return ERR_PTR(ret); 21555d4f98a2SYan Zheng break; 21565d4f98a2SYan Zheng } 21575d4f98a2SYan Zheng 215892de551bSJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 215992de551bSJosef Bacik if (ret) 216092de551bSJosef Bacik return ERR_PTR(ret); 21613fd0a558SYan, Zheng root = root->reloc_root; 21623fd0a558SYan, Zheng 216339200e59SJosef Bacik /* 216439200e59SJosef Bacik * We could have raced with another thread which failed, so 216539200e59SJosef Bacik * root->reloc_root may not be set, return ENOENT in this case. 216639200e59SJosef Bacik */ 216739200e59SJosef Bacik if (!root) 216839200e59SJosef Bacik return ERR_PTR(-ENOENT); 216939200e59SJosef Bacik 21703fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 21718ee66afeSJosef Bacik /* 21728ee66afeSJosef Bacik * We just created the reloc root, so we shouldn't have 21738ee66afeSJosef Bacik * ->new_bytenr set and this shouldn't be in the changed 21748ee66afeSJosef Bacik * list. If it is then we have multiple roots pointing 21758ee66afeSJosef Bacik * at the same bytenr which indicates corruption, or 21768ee66afeSJosef Bacik * we've made a mistake in the backref walking code. 21778ee66afeSJosef Bacik */ 21788ee66afeSJosef Bacik ASSERT(next->new_bytenr == 0); 21798ee66afeSJosef Bacik ASSERT(list_empty(&next->list)); 21808ee66afeSJosef Bacik if (next->new_bytenr || !list_empty(&next->list)) { 21818ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21828ee66afeSJosef Bacik "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu", 21838ee66afeSJosef Bacik node->bytenr, next->bytenr); 21848ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21858ee66afeSJosef Bacik } 21868ee66afeSJosef Bacik 21873fd0a558SYan, Zheng next->new_bytenr = root->node->start; 218800246528SJosef Bacik btrfs_put_root(next->root); 218900246528SJosef Bacik next->root = btrfs_grab_root(root); 21900b530bc5SJosef Bacik ASSERT(next->root); 21913fd0a558SYan, Zheng list_add_tail(&next->list, 21923fd0a558SYan, Zheng &rc->backref_cache.changed); 21939569cc20SQu Wenruo mark_block_processed(rc, next); 21945d4f98a2SYan Zheng break; 21955d4f98a2SYan Zheng } 21965d4f98a2SYan Zheng 21973fd0a558SYan, Zheng WARN_ON(1); 21985d4f98a2SYan Zheng root = NULL; 21995d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 22005d4f98a2SYan Zheng if (!next || next->level <= node->level) 22015d4f98a2SYan Zheng break; 22025d4f98a2SYan Zheng } 2203cbdc2ebcSJosef Bacik if (!root) { 2204cbdc2ebcSJosef Bacik /* 2205cbdc2ebcSJosef Bacik * This can happen if there's fs corruption or if there's a bug 2206cbdc2ebcSJosef Bacik * in the backref lookup code. 2207cbdc2ebcSJosef Bacik */ 2208cbdc2ebcSJosef Bacik ASSERT(0); 2209cbdc2ebcSJosef Bacik return ERR_PTR(-ENOENT); 2210cbdc2ebcSJosef Bacik } 22115d4f98a2SYan Zheng 22123fd0a558SYan, Zheng next = node; 22133fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 22143fd0a558SYan, Zheng while (1) { 22153fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 22163fd0a558SYan, Zheng if (--index < 0) 22173fd0a558SYan, Zheng break; 22183fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 22193fd0a558SYan, Zheng } 22205d4f98a2SYan Zheng return root; 22215d4f98a2SYan Zheng } 22225d4f98a2SYan Zheng 22233fd0a558SYan, Zheng /* 222492a7cc42SQu Wenruo * Select a tree root for relocation. 222592a7cc42SQu Wenruo * 222692a7cc42SQu Wenruo * Return NULL if the block is not shareable. We should use do_relocation() in 222792a7cc42SQu Wenruo * this case. 222892a7cc42SQu Wenruo * 222992a7cc42SQu Wenruo * Return a tree root pointer if the block is shareable. 223092a7cc42SQu Wenruo * Return -ENOENT if the block is root of reloc tree. 22313fd0a558SYan, Zheng */ 22325d4f98a2SYan Zheng static noinline_for_stack 2233a26195a5SQu Wenruo struct btrfs_root *select_one_root(struct btrfs_backref_node *node) 22345d4f98a2SYan Zheng { 2235a26195a5SQu Wenruo struct btrfs_backref_node *next; 22363fd0a558SYan, Zheng struct btrfs_root *root; 22373fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 2238a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22393fd0a558SYan, Zheng int index = 0; 22403fd0a558SYan, Zheng 22413fd0a558SYan, Zheng next = node; 22423fd0a558SYan, Zheng while (1) { 22433fd0a558SYan, Zheng cond_resched(); 22443fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 22453fd0a558SYan, Zheng root = next->root; 22468717cf44SJosef Bacik 22478717cf44SJosef Bacik /* 22488717cf44SJosef Bacik * This can occur if we have incomplete extent refs leading all 22498717cf44SJosef Bacik * the way up a particular path, in this case return -EUCLEAN. 22508717cf44SJosef Bacik */ 22518717cf44SJosef Bacik if (!root) 22528717cf44SJosef Bacik return ERR_PTR(-EUCLEAN); 22533fd0a558SYan, Zheng 225492a7cc42SQu Wenruo /* No other choice for non-shareable tree */ 225592a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 22563fd0a558SYan, Zheng return root; 22573fd0a558SYan, Zheng 22583fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 22593fd0a558SYan, Zheng fs_root = root; 22603fd0a558SYan, Zheng 22613fd0a558SYan, Zheng if (next != node) 22623fd0a558SYan, Zheng return NULL; 22633fd0a558SYan, Zheng 22643fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 22653fd0a558SYan, Zheng if (!next || next->level <= node->level) 22663fd0a558SYan, Zheng break; 22673fd0a558SYan, Zheng } 22683fd0a558SYan, Zheng 22693fd0a558SYan, Zheng if (!fs_root) 22703fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 22713fd0a558SYan, Zheng return fs_root; 22725d4f98a2SYan Zheng } 22735d4f98a2SYan Zheng 22745d4f98a2SYan Zheng static noinline_for_stack 22753fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 2276a26195a5SQu Wenruo struct btrfs_backref_node *node, int reserve) 22775d4f98a2SYan Zheng { 22780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2279a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2280a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2281a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22823fd0a558SYan, Zheng u64 num_bytes = 0; 22833fd0a558SYan, Zheng int index = 0; 22845d4f98a2SYan Zheng 22853fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 22863fd0a558SYan, Zheng 22873fd0a558SYan, Zheng while (next) { 22883fd0a558SYan, Zheng cond_resched(); 22895d4f98a2SYan Zheng while (1) { 22903fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 22915d4f98a2SYan Zheng break; 22925d4f98a2SYan Zheng 22930b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 22943fd0a558SYan, Zheng 22953fd0a558SYan, Zheng if (list_empty(&next->upper)) 22963fd0a558SYan, Zheng break; 22973fd0a558SYan, Zheng 22983fd0a558SYan, Zheng edge = list_entry(next->upper.next, 2299a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 23003fd0a558SYan, Zheng edges[index++] = edge; 23013fd0a558SYan, Zheng next = edge->node[UPPER]; 23025d4f98a2SYan Zheng } 23033fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 23043fd0a558SYan, Zheng } 23053fd0a558SYan, Zheng return num_bytes; 23063fd0a558SYan, Zheng } 23073fd0a558SYan, Zheng 23083fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 23093fd0a558SYan, Zheng struct reloc_control *rc, 2310a26195a5SQu Wenruo struct btrfs_backref_node *node) 23113fd0a558SYan, Zheng { 23123fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2313da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23143fd0a558SYan, Zheng u64 num_bytes; 23153fd0a558SYan, Zheng int ret; 23160647bf56SWang Shilong u64 tmp; 23173fd0a558SYan, Zheng 23183fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 23193fd0a558SYan, Zheng 23203fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 23210647bf56SWang Shilong rc->reserved_bytes += num_bytes; 23228ca17f0fSJosef Bacik 23238ca17f0fSJosef Bacik /* 23248ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 23258ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 23268ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 23278ca17f0fSJosef Bacik */ 23289270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes, 23298ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 23303fd0a558SYan, Zheng if (ret) { 2331da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 23320647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 23330647bf56SWang Shilong tmp <<= 1; 23340647bf56SWang Shilong /* 23350647bf56SWang Shilong * only one thread can access block_rsv at this point, 23360647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 23370647bf56SWang Shilong * we expand more reservation size here to allow enough 233852042d8eSAndrea Gelmini * space for relocation and we will return earlier in 23390647bf56SWang Shilong * enospc case. 23400647bf56SWang Shilong */ 2341da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 23420647bf56SWang Shilong RELOCATION_RESERVED_NODES; 23438ca17f0fSJosef Bacik return -EAGAIN; 23443fd0a558SYan, Zheng } 23453fd0a558SYan, Zheng 23463fd0a558SYan, Zheng return 0; 23473fd0a558SYan, Zheng } 23483fd0a558SYan, Zheng 23495d4f98a2SYan Zheng /* 23505d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 23515d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 23525d4f98a2SYan Zheng * 23535d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 23545d4f98a2SYan Zheng * in that case this function just updates pointers. 23555d4f98a2SYan Zheng */ 23565d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 23573fd0a558SYan, Zheng struct reloc_control *rc, 2358a26195a5SQu Wenruo struct btrfs_backref_node *node, 23595d4f98a2SYan Zheng struct btrfs_key *key, 23605d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 23615d4f98a2SYan Zheng { 2362a26195a5SQu Wenruo struct btrfs_backref_node *upper; 2363a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2364a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 23655d4f98a2SYan Zheng struct btrfs_root *root; 23665d4f98a2SYan Zheng struct extent_buffer *eb; 23675d4f98a2SYan Zheng u32 blocksize; 23685d4f98a2SYan Zheng u64 bytenr; 23695d4f98a2SYan Zheng int slot; 23708df01fddSNikolay Borisov int ret = 0; 23715d4f98a2SYan Zheng 2372ffe30dd8SJosef Bacik /* 2373ffe30dd8SJosef Bacik * If we are lowest then this is the first time we're processing this 2374ffe30dd8SJosef Bacik * block, and thus shouldn't have an eb associated with it yet. 2375ffe30dd8SJosef Bacik */ 2376ffe30dd8SJosef Bacik ASSERT(!lowest || !node->eb); 23775d4f98a2SYan Zheng 23785d4f98a2SYan Zheng path->lowest_level = node->level + 1; 23793fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 23805d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 238182fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2382581c1760SQu Wenruo 23835d4f98a2SYan Zheng cond_resched(); 23845d4f98a2SYan Zheng 23855d4f98a2SYan Zheng upper = edge->node[UPPER]; 2386dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 2387cbdc2ebcSJosef Bacik if (IS_ERR(root)) { 2388cbdc2ebcSJosef Bacik ret = PTR_ERR(root); 2389cbdc2ebcSJosef Bacik goto next; 2390cbdc2ebcSJosef Bacik } 23915d4f98a2SYan Zheng 23923fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 23933fd0a558SYan, Zheng if (!lowest) { 2394e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 23958df01fddSNikolay Borisov if (ret < 0) 2396cbca7d59SFilipe Manana goto next; 23973fd0a558SYan, Zheng BUG_ON(ret); 23983fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 23993fd0a558SYan, Zheng if (node->eb->start == bytenr) 24003fd0a558SYan, Zheng goto next; 24013fd0a558SYan, Zheng } 2402b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 24033fd0a558SYan, Zheng } 24045d4f98a2SYan Zheng 24055d4f98a2SYan Zheng if (!upper->eb) { 24065d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 24073561b9dbSLiu Bo if (ret) { 24088df01fddSNikolay Borisov if (ret > 0) 24098df01fddSNikolay Borisov ret = -ENOENT; 24103561b9dbSLiu Bo 24113561b9dbSLiu Bo btrfs_release_path(path); 24125d4f98a2SYan Zheng break; 24135d4f98a2SYan Zheng } 24145d4f98a2SYan Zheng 24153fd0a558SYan, Zheng if (!upper->eb) { 24163fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 24173fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 24183fd0a558SYan, Zheng } else { 24193fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 24203fd0a558SYan, Zheng } 24213fd0a558SYan, Zheng 24223fd0a558SYan, Zheng upper->locked = 1; 24233fd0a558SYan, Zheng path->locks[upper->level] = 0; 24243fd0a558SYan, Zheng 24255d4f98a2SYan Zheng slot = path->slots[upper->level]; 2426b3b4aa74SDavid Sterba btrfs_release_path(path); 24275d4f98a2SYan Zheng } else { 2428e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 24298df01fddSNikolay Borisov if (ret < 0) 2430cbca7d59SFilipe Manana goto next; 24315d4f98a2SYan Zheng BUG_ON(ret); 24325d4f98a2SYan Zheng } 24335d4f98a2SYan Zheng 24345d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24353fd0a558SYan, Zheng if (lowest) { 24364547f4d8SLiu Bo if (bytenr != node->bytenr) { 24374547f4d8SLiu Bo btrfs_err(root->fs_info, 24384547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 24394547f4d8SLiu Bo bytenr, node->bytenr, slot, 24404547f4d8SLiu Bo upper->eb->start); 24418df01fddSNikolay Borisov ret = -EIO; 24424547f4d8SLiu Bo goto next; 24434547f4d8SLiu Bo } 24445d4f98a2SYan Zheng } else { 24453fd0a558SYan, Zheng if (node->eb->start == bytenr) 24463fd0a558SYan, Zheng goto next; 24475d4f98a2SYan Zheng } 24485d4f98a2SYan Zheng 2449da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 2450c9752536SJosef Bacik eb = btrfs_read_node_slot(upper->eb, slot); 245164c043deSLiu Bo if (IS_ERR(eb)) { 24528df01fddSNikolay Borisov ret = PTR_ERR(eb); 245364c043deSLiu Bo goto next; 245497d9a8a4STsutomu Itoh } 24555d4f98a2SYan Zheng btrfs_tree_lock(eb); 24565d4f98a2SYan Zheng 24575d4f98a2SYan Zheng if (!node->eb) { 24585d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 24599631e4ccSJosef Bacik slot, &eb, BTRFS_NESTING_COW); 24603fd0a558SYan, Zheng btrfs_tree_unlock(eb); 24613fd0a558SYan, Zheng free_extent_buffer(eb); 24628df01fddSNikolay Borisov if (ret < 0) 24633fd0a558SYan, Zheng goto next; 2464ffe30dd8SJosef Bacik /* 2465ffe30dd8SJosef Bacik * We've just COWed this block, it should have updated 2466ffe30dd8SJosef Bacik * the correct backref node entry. 2467ffe30dd8SJosef Bacik */ 2468ffe30dd8SJosef Bacik ASSERT(node->eb == eb); 24695d4f98a2SYan Zheng } else { 24705d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 24715d4f98a2SYan Zheng node->eb->start); 24725d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 24735d4f98a2SYan Zheng trans->transid); 24745d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 24755d4f98a2SYan Zheng 247682fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 24775d4f98a2SYan Zheng node->eb->start, blocksize, 247882fa113fSQu Wenruo upper->eb->start); 247982fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 2480f42c5da6SNikolay Borisov btrfs_header_owner(upper->eb), 2481f42c5da6SNikolay Borisov root->root_key.objectid, false); 248282fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 2483eb6b7fb4SJosef Bacik if (!ret) 2484eb6b7fb4SJosef Bacik ret = btrfs_drop_subtree(trans, root, eb, 2485eb6b7fb4SJosef Bacik upper->eb); 2486eb6b7fb4SJosef Bacik if (ret) 2487eb6b7fb4SJosef Bacik btrfs_abort_transaction(trans, ret); 24885d4f98a2SYan Zheng } 24893fd0a558SYan, Zheng next: 24903fd0a558SYan, Zheng if (!upper->pending) 2491b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 24923fd0a558SYan, Zheng else 2493b0fe7078SQu Wenruo btrfs_backref_unlock_node_buffer(upper); 24948df01fddSNikolay Borisov if (ret) 24953fd0a558SYan, Zheng break; 24965d4f98a2SYan Zheng } 24973fd0a558SYan, Zheng 24988df01fddSNikolay Borisov if (!ret && node->pending) { 2499b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 25003fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 25013fd0a558SYan, Zheng node->pending = 0; 25025d4f98a2SYan Zheng } 25033fd0a558SYan, Zheng 25045d4f98a2SYan Zheng path->lowest_level = 0; 2505ffe30dd8SJosef Bacik 2506ffe30dd8SJosef Bacik /* 2507ffe30dd8SJosef Bacik * We should have allocated all of our space in the block rsv and thus 2508ffe30dd8SJosef Bacik * shouldn't ENOSPC. 2509ffe30dd8SJosef Bacik */ 2510ffe30dd8SJosef Bacik ASSERT(ret != -ENOSPC); 25118df01fddSNikolay Borisov return ret; 25125d4f98a2SYan Zheng } 25135d4f98a2SYan Zheng 25145d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 25153fd0a558SYan, Zheng struct reloc_control *rc, 2516a26195a5SQu Wenruo struct btrfs_backref_node *node, 25175d4f98a2SYan Zheng struct btrfs_path *path) 25185d4f98a2SYan Zheng { 25195d4f98a2SYan Zheng struct btrfs_key key; 25205d4f98a2SYan Zheng 25215d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 25223fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 25235d4f98a2SYan Zheng } 25245d4f98a2SYan Zheng 25255d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 25263fd0a558SYan, Zheng struct reloc_control *rc, 25273fd0a558SYan, Zheng struct btrfs_path *path, int err) 25285d4f98a2SYan Zheng { 25293fd0a558SYan, Zheng LIST_HEAD(list); 2530a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 2531a26195a5SQu Wenruo struct btrfs_backref_node *node; 25325d4f98a2SYan Zheng int level; 25335d4f98a2SYan Zheng int ret; 25345d4f98a2SYan Zheng 25355d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 25365d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 25375d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 2538a26195a5SQu Wenruo struct btrfs_backref_node, list); 25393fd0a558SYan, Zheng list_move_tail(&node->list, &list); 25403fd0a558SYan, Zheng BUG_ON(!node->pending); 25415d4f98a2SYan Zheng 25423fd0a558SYan, Zheng if (!err) { 25433fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 25445d4f98a2SYan Zheng if (ret < 0) 25455d4f98a2SYan Zheng err = ret; 25465d4f98a2SYan Zheng } 25475d4f98a2SYan Zheng } 25483fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 25493fd0a558SYan, Zheng } 25505d4f98a2SYan Zheng return err; 25515d4f98a2SYan Zheng } 25525d4f98a2SYan Zheng 25535d4f98a2SYan Zheng /* 25545d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 25555d4f98a2SYan Zheng * as processed. 25565d4f98a2SYan Zheng */ 25575d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 2558a26195a5SQu Wenruo struct btrfs_backref_node *node) 25595d4f98a2SYan Zheng { 2560a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2561a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2562a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25635d4f98a2SYan Zheng int index = 0; 25645d4f98a2SYan Zheng 25655d4f98a2SYan Zheng while (next) { 25665d4f98a2SYan Zheng cond_resched(); 25675d4f98a2SYan Zheng while (1) { 25685d4f98a2SYan Zheng if (next->processed) 25695d4f98a2SYan Zheng break; 25705d4f98a2SYan Zheng 25719569cc20SQu Wenruo mark_block_processed(rc, next); 25725d4f98a2SYan Zheng 25735d4f98a2SYan Zheng if (list_empty(&next->upper)) 25745d4f98a2SYan Zheng break; 25755d4f98a2SYan Zheng 25765d4f98a2SYan Zheng edge = list_entry(next->upper.next, 2577a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 25785d4f98a2SYan Zheng edges[index++] = edge; 25795d4f98a2SYan Zheng next = edge->node[UPPER]; 25805d4f98a2SYan Zheng } 25815d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25825d4f98a2SYan Zheng } 25835d4f98a2SYan Zheng } 25845d4f98a2SYan Zheng 25857476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 25865d4f98a2SYan Zheng { 2587da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 25887476dfdaSDavid Sterba 25895d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 25909655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 25915d4f98a2SYan Zheng return 1; 25925d4f98a2SYan Zheng return 0; 25935d4f98a2SYan Zheng } 25945d4f98a2SYan Zheng 25952ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 25965d4f98a2SYan Zheng struct tree_block *block) 25975d4f98a2SYan Zheng { 25985d4f98a2SYan Zheng struct extent_buffer *eb; 25995d4f98a2SYan Zheng 2600f7ba2d37SJosef Bacik eb = read_tree_block(fs_info, block->bytenr, block->owner, 2601f7ba2d37SJosef Bacik block->key.offset, block->level, NULL); 260264c043deSLiu Bo if (IS_ERR(eb)) { 260364c043deSLiu Bo return PTR_ERR(eb); 260464c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2605416bc658SJosef Bacik free_extent_buffer(eb); 2606416bc658SJosef Bacik return -EIO; 2607416bc658SJosef Bacik } 26085d4f98a2SYan Zheng if (block->level == 0) 26095d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 26105d4f98a2SYan Zheng else 26115d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 26125d4f98a2SYan Zheng free_extent_buffer(eb); 26135d4f98a2SYan Zheng block->key_ready = 1; 26145d4f98a2SYan Zheng return 0; 26155d4f98a2SYan Zheng } 26165d4f98a2SYan Zheng 26175d4f98a2SYan Zheng /* 26185d4f98a2SYan Zheng * helper function to relocate a tree block 26195d4f98a2SYan Zheng */ 26205d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 26215d4f98a2SYan Zheng struct reloc_control *rc, 2622a26195a5SQu Wenruo struct btrfs_backref_node *node, 26235d4f98a2SYan Zheng struct btrfs_key *key, 26245d4f98a2SYan Zheng struct btrfs_path *path) 26255d4f98a2SYan Zheng { 26265d4f98a2SYan Zheng struct btrfs_root *root; 26273fd0a558SYan, Zheng int ret = 0; 26285d4f98a2SYan Zheng 26293fd0a558SYan, Zheng if (!node) 26305d4f98a2SYan Zheng return 0; 26313fd0a558SYan, Zheng 26325f6b2e5cSJosef Bacik /* 26335f6b2e5cSJosef Bacik * If we fail here we want to drop our backref_node because we are going 26345f6b2e5cSJosef Bacik * to start over and regenerate the tree for it. 26355f6b2e5cSJosef Bacik */ 26365f6b2e5cSJosef Bacik ret = reserve_metadata_space(trans, rc, node); 26375f6b2e5cSJosef Bacik if (ret) 26385f6b2e5cSJosef Bacik goto out; 26395f6b2e5cSJosef Bacik 26403fd0a558SYan, Zheng BUG_ON(node->processed); 2641147d256eSZhaolei root = select_one_root(node); 26428717cf44SJosef Bacik if (IS_ERR(root)) { 26438717cf44SJosef Bacik ret = PTR_ERR(root); 26448717cf44SJosef Bacik 26458717cf44SJosef Bacik /* See explanation in select_one_root for the -EUCLEAN case. */ 26468717cf44SJosef Bacik ASSERT(ret == -ENOENT); 26478717cf44SJosef Bacik if (ret == -ENOENT) { 26488717cf44SJosef Bacik ret = 0; 26493fd0a558SYan, Zheng update_processed_blocks(rc, node); 26508717cf44SJosef Bacik } 26513fd0a558SYan, Zheng goto out; 26525d4f98a2SYan Zheng } 26535d4f98a2SYan Zheng 26543fd0a558SYan, Zheng if (root) { 265592a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 26561c7bfa15SJosef Bacik /* 26571c7bfa15SJosef Bacik * This block was the root block of a root, and this is 26581c7bfa15SJosef Bacik * the first time we're processing the block and thus it 26591c7bfa15SJosef Bacik * should not have had the ->new_bytenr modified and 26601c7bfa15SJosef Bacik * should have not been included on the changed list. 26611c7bfa15SJosef Bacik * 26621c7bfa15SJosef Bacik * However in the case of corruption we could have 26631c7bfa15SJosef Bacik * multiple refs pointing to the same block improperly, 26641c7bfa15SJosef Bacik * and thus we would trip over these checks. ASSERT() 26651c7bfa15SJosef Bacik * for the developer case, because it could indicate a 26661c7bfa15SJosef Bacik * bug in the backref code, however error out for a 26671c7bfa15SJosef Bacik * normal user in the case of corruption. 26681c7bfa15SJosef Bacik */ 26691c7bfa15SJosef Bacik ASSERT(node->new_bytenr == 0); 26701c7bfa15SJosef Bacik ASSERT(list_empty(&node->list)); 26711c7bfa15SJosef Bacik if (node->new_bytenr || !list_empty(&node->list)) { 26721c7bfa15SJosef Bacik btrfs_err(root->fs_info, 26731c7bfa15SJosef Bacik "bytenr %llu has improper references to it", 26741c7bfa15SJosef Bacik node->bytenr); 26751c7bfa15SJosef Bacik ret = -EUCLEAN; 26761c7bfa15SJosef Bacik goto out; 26771c7bfa15SJosef Bacik } 2678d18c7bd9SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 2679d18c7bd9SJosef Bacik if (ret) 2680d18c7bd9SJosef Bacik goto out; 268139200e59SJosef Bacik /* 268239200e59SJosef Bacik * Another thread could have failed, need to check if we 268339200e59SJosef Bacik * have reloc_root actually set. 268439200e59SJosef Bacik */ 268539200e59SJosef Bacik if (!root->reloc_root) { 268639200e59SJosef Bacik ret = -ENOENT; 268739200e59SJosef Bacik goto out; 268839200e59SJosef Bacik } 26893fd0a558SYan, Zheng root = root->reloc_root; 26903fd0a558SYan, Zheng node->new_bytenr = root->node->start; 269100246528SJosef Bacik btrfs_put_root(node->root); 269200246528SJosef Bacik node->root = btrfs_grab_root(root); 26930b530bc5SJosef Bacik ASSERT(node->root); 26943fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 26953fd0a558SYan, Zheng } else { 26965d4f98a2SYan Zheng path->lowest_level = node->level; 26972bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 26982bb2e00eSFilipe Manana btrfs_reserve_chunk_metadata(trans, false); 26995d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2700b3b4aa74SDavid Sterba btrfs_release_path(path); 27012bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 27022bb2e00eSFilipe Manana btrfs_trans_release_chunk_metadata(trans); 27033fd0a558SYan, Zheng if (ret > 0) 27045d4f98a2SYan Zheng ret = 0; 27053fd0a558SYan, Zheng } 27063fd0a558SYan, Zheng if (!ret) 27073fd0a558SYan, Zheng update_processed_blocks(rc, node); 27083fd0a558SYan, Zheng } else { 27093fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 27103fd0a558SYan, Zheng } 27115d4f98a2SYan Zheng out: 27120647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 2713023acb07SQu Wenruo btrfs_backref_cleanup_node(&rc->backref_cache, node); 27145d4f98a2SYan Zheng return ret; 27155d4f98a2SYan Zheng } 27165d4f98a2SYan Zheng 27175d4f98a2SYan Zheng /* 27185d4f98a2SYan Zheng * relocate a list of blocks 27195d4f98a2SYan Zheng */ 27205d4f98a2SYan Zheng static noinline_for_stack 27215d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 27225d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 27235d4f98a2SYan Zheng { 27242ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2725a26195a5SQu Wenruo struct btrfs_backref_node *node; 27265d4f98a2SYan Zheng struct btrfs_path *path; 27275d4f98a2SYan Zheng struct tree_block *block; 272898ff7b94SQu Wenruo struct tree_block *next; 27295d4f98a2SYan Zheng int ret; 27305d4f98a2SYan Zheng int err = 0; 27315d4f98a2SYan Zheng 27325d4f98a2SYan Zheng path = btrfs_alloc_path(); 2733e1a12670SLiu Bo if (!path) { 2734e1a12670SLiu Bo err = -ENOMEM; 273534c2b290SDavid Sterba goto out_free_blocks; 2736e1a12670SLiu Bo } 27375d4f98a2SYan Zheng 273898ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 273998ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27405d4f98a2SYan Zheng if (!block->key_ready) 2741f7ba2d37SJosef Bacik btrfs_readahead_tree_block(fs_info, block->bytenr, 2742f7ba2d37SJosef Bacik block->owner, 0, 27433fbaf258SJosef Bacik block->level); 27445d4f98a2SYan Zheng } 27455d4f98a2SYan Zheng 274698ff7b94SQu Wenruo /* Get first keys */ 274798ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 274834c2b290SDavid Sterba if (!block->key_ready) { 27492ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 275034c2b290SDavid Sterba if (err) 275134c2b290SDavid Sterba goto out_free_path; 275234c2b290SDavid Sterba } 27535d4f98a2SYan Zheng } 27545d4f98a2SYan Zheng 275598ff7b94SQu Wenruo /* Do tree relocation */ 275698ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27573fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 27585d4f98a2SYan Zheng block->level, block->bytenr); 27595d4f98a2SYan Zheng if (IS_ERR(node)) { 27605d4f98a2SYan Zheng err = PTR_ERR(node); 27615d4f98a2SYan Zheng goto out; 27625d4f98a2SYan Zheng } 27635d4f98a2SYan Zheng 27645d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 27655d4f98a2SYan Zheng path); 27665d4f98a2SYan Zheng if (ret < 0) { 27675d4f98a2SYan Zheng err = ret; 276850dbbb71SJosef Bacik break; 27695d4f98a2SYan Zheng } 27705d4f98a2SYan Zheng } 27715d4f98a2SYan Zheng out: 27723fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 27735d4f98a2SYan Zheng 277434c2b290SDavid Sterba out_free_path: 27755d4f98a2SYan Zheng btrfs_free_path(path); 277634c2b290SDavid Sterba out_free_blocks: 2777e1a12670SLiu Bo free_block_list(blocks); 27785d4f98a2SYan Zheng return err; 27795d4f98a2SYan Zheng } 27805d4f98a2SYan Zheng 2781056d9becSNikolay Borisov static noinline_for_stack int prealloc_file_extent_cluster( 2782056d9becSNikolay Borisov struct btrfs_inode *inode, 2783efa56464SYan, Zheng struct file_extent_cluster *cluster) 2784efa56464SYan, Zheng { 2785efa56464SYan, Zheng u64 alloc_hint = 0; 2786efa56464SYan, Zheng u64 start; 2787efa56464SYan, Zheng u64 end; 2788056d9becSNikolay Borisov u64 offset = inode->index_cnt; 2789efa56464SYan, Zheng u64 num_bytes; 27904e9d0d01SNikolay Borisov int nr; 2791efa56464SYan, Zheng int ret = 0; 27929d9ea1e6SQu Wenruo u64 i_size = i_size_read(&inode->vfs_inode); 2793dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 2794dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 2795214e61d0SNikolay Borisov u64 cur_offset = prealloc_start; 2796efa56464SYan, Zheng 27979d9ea1e6SQu Wenruo /* 27989d9ea1e6SQu Wenruo * For subpage case, previous i_size may not be aligned to PAGE_SIZE. 27999d9ea1e6SQu Wenruo * This means the range [i_size, PAGE_END + 1) is filled with zeros by 28009d9ea1e6SQu Wenruo * btrfs_do_readpage() call of previously relocated file cluster. 28019d9ea1e6SQu Wenruo * 28029d9ea1e6SQu Wenruo * If the current cluster starts in the above range, btrfs_do_readpage() 28039d9ea1e6SQu Wenruo * will skip the read, and relocate_one_page() will later writeback 28049d9ea1e6SQu Wenruo * the padding zeros as new data, causing data corruption. 28059d9ea1e6SQu Wenruo * 28069d9ea1e6SQu Wenruo * Here we have to manually invalidate the range (i_size, PAGE_END + 1). 28079d9ea1e6SQu Wenruo */ 28089d9ea1e6SQu Wenruo if (!IS_ALIGNED(i_size, PAGE_SIZE)) { 28099d9ea1e6SQu Wenruo struct address_space *mapping = inode->vfs_inode.i_mapping; 28109d9ea1e6SQu Wenruo struct btrfs_fs_info *fs_info = inode->root->fs_info; 28119d9ea1e6SQu Wenruo const u32 sectorsize = fs_info->sectorsize; 28129d9ea1e6SQu Wenruo struct page *page; 28139d9ea1e6SQu Wenruo 28149d9ea1e6SQu Wenruo ASSERT(sectorsize < PAGE_SIZE); 28159d9ea1e6SQu Wenruo ASSERT(IS_ALIGNED(i_size, sectorsize)); 28169d9ea1e6SQu Wenruo 28179d9ea1e6SQu Wenruo /* 28189d9ea1e6SQu Wenruo * Subpage can't handle page with DIRTY but without UPTODATE 28199d9ea1e6SQu Wenruo * bit as it can lead to the following deadlock: 28209d9ea1e6SQu Wenruo * 28219d9ea1e6SQu Wenruo * btrfs_readpage() 28229d9ea1e6SQu Wenruo * | Page already *locked* 28239d9ea1e6SQu Wenruo * |- btrfs_lock_and_flush_ordered_range() 28249d9ea1e6SQu Wenruo * |- btrfs_start_ordered_extent() 28259d9ea1e6SQu Wenruo * |- extent_write_cache_pages() 28269d9ea1e6SQu Wenruo * |- lock_page() 28279d9ea1e6SQu Wenruo * We try to lock the page we already hold. 28289d9ea1e6SQu Wenruo * 28299d9ea1e6SQu Wenruo * Here we just writeback the whole data reloc inode, so that 28309d9ea1e6SQu Wenruo * we will be ensured to have no dirty range in the page, and 28319d9ea1e6SQu Wenruo * are safe to clear the uptodate bits. 28329d9ea1e6SQu Wenruo * 28339d9ea1e6SQu Wenruo * This shouldn't cause too much overhead, as we need to write 28349d9ea1e6SQu Wenruo * the data back anyway. 28359d9ea1e6SQu Wenruo */ 28369d9ea1e6SQu Wenruo ret = filemap_write_and_wait(mapping); 28379d9ea1e6SQu Wenruo if (ret < 0) 28389d9ea1e6SQu Wenruo return ret; 28399d9ea1e6SQu Wenruo 28409d9ea1e6SQu Wenruo clear_extent_bits(&inode->io_tree, i_size, 28419d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - 1, 28429d9ea1e6SQu Wenruo EXTENT_UPTODATE); 28439d9ea1e6SQu Wenruo page = find_lock_page(mapping, i_size >> PAGE_SHIFT); 28449d9ea1e6SQu Wenruo /* 28459d9ea1e6SQu Wenruo * If page is freed we don't need to do anything then, as we 28469d9ea1e6SQu Wenruo * will re-read the whole page anyway. 28479d9ea1e6SQu Wenruo */ 28489d9ea1e6SQu Wenruo if (page) { 28499d9ea1e6SQu Wenruo btrfs_subpage_clear_uptodate(fs_info, page, i_size, 28509d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - i_size); 28519d9ea1e6SQu Wenruo unlock_page(page); 28529d9ea1e6SQu Wenruo put_page(page); 28539d9ea1e6SQu Wenruo } 28549d9ea1e6SQu Wenruo } 28559d9ea1e6SQu Wenruo 2856efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 2857056d9becSNikolay Borisov ret = btrfs_alloc_data_chunk_ondemand(inode, 2858dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 2859efa56464SYan, Zheng if (ret) 2860214e61d0SNikolay Borisov return ret; 2861efa56464SYan, Zheng 286264708539SJosef Bacik btrfs_inode_lock(&inode->vfs_inode, 0); 28634e9d0d01SNikolay Borisov for (nr = 0; nr < cluster->nr; nr++) { 2864efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 2865efa56464SYan, Zheng if (nr + 1 < cluster->nr) 2866efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 2867efa56464SYan, Zheng else 2868efa56464SYan, Zheng end = cluster->end - offset; 2869efa56464SYan, Zheng 2870056d9becSNikolay Borisov lock_extent(&inode->io_tree, start, end); 2871efa56464SYan, Zheng num_bytes = end + 1 - start; 2872056d9becSNikolay Borisov ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start, 2873efa56464SYan, Zheng num_bytes, num_bytes, 2874efa56464SYan, Zheng end + 1, &alloc_hint); 287518513091SWang Xiaoguang cur_offset = end + 1; 2876056d9becSNikolay Borisov unlock_extent(&inode->io_tree, start, end); 2877efa56464SYan, Zheng if (ret) 2878efa56464SYan, Zheng break; 2879efa56464SYan, Zheng } 288064708539SJosef Bacik btrfs_inode_unlock(&inode->vfs_inode, 0); 2881214e61d0SNikolay Borisov 288218513091SWang Xiaoguang if (cur_offset < prealloc_end) 2883056d9becSNikolay Borisov btrfs_free_reserved_data_space_noquota(inode->root->fs_info, 2884a89ef455SFilipe Manana prealloc_end + 1 - cur_offset); 2885efa56464SYan, Zheng return ret; 2886efa56464SYan, Zheng } 2887efa56464SYan, Zheng 28884b01c44fSJohannes Thumshirn static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode, 28894b01c44fSJohannes Thumshirn u64 start, u64 end, u64 block_start) 28905d4f98a2SYan Zheng { 28915d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 28925d4f98a2SYan Zheng struct extent_map *em; 28930257bb82SYan, Zheng int ret = 0; 28945d4f98a2SYan Zheng 2895172ddd60SDavid Sterba em = alloc_extent_map(); 28960257bb82SYan, Zheng if (!em) 28970257bb82SYan, Zheng return -ENOMEM; 28980257bb82SYan, Zheng 28995d4f98a2SYan Zheng em->start = start; 29000257bb82SYan, Zheng em->len = end + 1 - start; 29010257bb82SYan, Zheng em->block_len = em->len; 29020257bb82SYan, Zheng em->block_start = block_start; 29035d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 29045d4f98a2SYan Zheng 2905d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 29065d4f98a2SYan Zheng while (1) { 2907890871beSChris Mason write_lock(&em_tree->lock); 290809a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2909890871beSChris Mason write_unlock(&em_tree->lock); 29105d4f98a2SYan Zheng if (ret != -EEXIST) { 29115d4f98a2SYan Zheng free_extent_map(em); 29125d4f98a2SYan Zheng break; 29135d4f98a2SYan Zheng } 2914dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 29155d4f98a2SYan Zheng } 2916d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 29170257bb82SYan, Zheng return ret; 29180257bb82SYan, Zheng } 29195d4f98a2SYan Zheng 2920726a3421SQu Wenruo /* 2921907d2710SDavid Sterba * Allow error injection to test balance/relocation cancellation 2922726a3421SQu Wenruo */ 29231fec12a5SJosef Bacik noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 2924726a3421SQu Wenruo { 29255cb502f4SQu Wenruo return atomic_read(&fs_info->balance_cancel_req) || 2926907d2710SDavid Sterba atomic_read(&fs_info->reloc_cancel_req) || 29275cb502f4SQu Wenruo fatal_signal_pending(current); 2928726a3421SQu Wenruo } 2929726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 2930726a3421SQu Wenruo 2931c2832898SQu Wenruo static u64 get_cluster_boundary_end(struct file_extent_cluster *cluster, 2932c2832898SQu Wenruo int cluster_nr) 2933c2832898SQu Wenruo { 2934c2832898SQu Wenruo /* Last extent, use cluster end directly */ 2935c2832898SQu Wenruo if (cluster_nr >= cluster->nr - 1) 2936c2832898SQu Wenruo return cluster->end; 2937c2832898SQu Wenruo 2938c2832898SQu Wenruo /* Use next boundary start*/ 2939c2832898SQu Wenruo return cluster->boundary[cluster_nr + 1] - 1; 2940c2832898SQu Wenruo } 2941c2832898SQu Wenruo 2942f47960f4SQu Wenruo static int relocate_one_page(struct inode *inode, struct file_ra_state *ra, 2943f47960f4SQu Wenruo struct file_extent_cluster *cluster, 2944f47960f4SQu Wenruo int *cluster_nr, unsigned long page_index) 2945f47960f4SQu Wenruo { 2946f47960f4SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 2947f47960f4SQu Wenruo u64 offset = BTRFS_I(inode)->index_cnt; 2948f47960f4SQu Wenruo const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT; 2949f47960f4SQu Wenruo gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 2950f47960f4SQu Wenruo struct page *page; 2951f47960f4SQu Wenruo u64 page_start; 2952f47960f4SQu Wenruo u64 page_end; 2953c2832898SQu Wenruo u64 cur; 2954f47960f4SQu Wenruo int ret; 2955f47960f4SQu Wenruo 2956f47960f4SQu Wenruo ASSERT(page_index <= last_index); 2957f47960f4SQu Wenruo page = find_lock_page(inode->i_mapping, page_index); 2958f47960f4SQu Wenruo if (!page) { 2959f47960f4SQu Wenruo page_cache_sync_readahead(inode->i_mapping, ra, NULL, 2960f47960f4SQu Wenruo page_index, last_index + 1 - page_index); 2961f47960f4SQu Wenruo page = find_or_create_page(inode->i_mapping, page_index, mask); 2962c2832898SQu Wenruo if (!page) 2963c2832898SQu Wenruo return -ENOMEM; 2964f47960f4SQu Wenruo } 2965f47960f4SQu Wenruo ret = set_page_extent_mapped(page); 2966f47960f4SQu Wenruo if (ret < 0) 2967f47960f4SQu Wenruo goto release_page; 2968f47960f4SQu Wenruo 2969f47960f4SQu Wenruo if (PageReadahead(page)) 2970f47960f4SQu Wenruo page_cache_async_readahead(inode->i_mapping, ra, NULL, page, 2971f47960f4SQu Wenruo page_index, last_index + 1 - page_index); 2972f47960f4SQu Wenruo 2973f47960f4SQu Wenruo if (!PageUptodate(page)) { 2974f47960f4SQu Wenruo btrfs_readpage(NULL, page); 2975f47960f4SQu Wenruo lock_page(page); 2976f47960f4SQu Wenruo if (!PageUptodate(page)) { 2977f47960f4SQu Wenruo ret = -EIO; 2978f47960f4SQu Wenruo goto release_page; 2979f47960f4SQu Wenruo } 2980f47960f4SQu Wenruo } 2981f47960f4SQu Wenruo 2982f47960f4SQu Wenruo page_start = page_offset(page); 2983f47960f4SQu Wenruo page_end = page_start + PAGE_SIZE - 1; 2984f47960f4SQu Wenruo 2985c2832898SQu Wenruo /* 2986c2832898SQu Wenruo * Start from the cluster, as for subpage case, the cluster can start 2987c2832898SQu Wenruo * inside the page. 2988c2832898SQu Wenruo */ 2989c2832898SQu Wenruo cur = max(page_start, cluster->boundary[*cluster_nr] - offset); 2990c2832898SQu Wenruo while (cur <= page_end) { 2991c2832898SQu Wenruo u64 extent_start = cluster->boundary[*cluster_nr] - offset; 2992c2832898SQu Wenruo u64 extent_end = get_cluster_boundary_end(cluster, 2993c2832898SQu Wenruo *cluster_nr) - offset; 2994c2832898SQu Wenruo u64 clamped_start = max(page_start, extent_start); 2995c2832898SQu Wenruo u64 clamped_end = min(page_end, extent_end); 2996c2832898SQu Wenruo u32 clamped_len = clamped_end + 1 - clamped_start; 2997f47960f4SQu Wenruo 2998c2832898SQu Wenruo /* Reserve metadata for this range */ 2999c2832898SQu Wenruo ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 3000c2832898SQu Wenruo clamped_len); 3001c2832898SQu Wenruo if (ret) 3002f47960f4SQu Wenruo goto release_page; 3003f47960f4SQu Wenruo 3004c2832898SQu Wenruo /* Mark the range delalloc and dirty for later writeback */ 3005c2832898SQu Wenruo lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end); 3006c2832898SQu Wenruo ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start, 3007c2832898SQu Wenruo clamped_end, 0, NULL); 3008c2832898SQu Wenruo if (ret) { 3009c2832898SQu Wenruo clear_extent_bits(&BTRFS_I(inode)->io_tree, 3010c2832898SQu Wenruo clamped_start, clamped_end, 3011c2832898SQu Wenruo EXTENT_LOCKED | EXTENT_BOUNDARY); 3012c2832898SQu Wenruo btrfs_delalloc_release_metadata(BTRFS_I(inode), 3013c2832898SQu Wenruo clamped_len, true); 3014c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), 3015c2832898SQu Wenruo clamped_len); 3016c2832898SQu Wenruo goto release_page; 3017f47960f4SQu Wenruo } 3018c2832898SQu Wenruo btrfs_page_set_dirty(fs_info, page, clamped_start, clamped_len); 3019f47960f4SQu Wenruo 3020c2832898SQu Wenruo /* 3021c2832898SQu Wenruo * Set the boundary if it's inside the page. 3022c2832898SQu Wenruo * Data relocation requires the destination extents to have the 3023c2832898SQu Wenruo * same size as the source. 3024c2832898SQu Wenruo * EXTENT_BOUNDARY bit prevents current extent from being merged 3025c2832898SQu Wenruo * with previous extent. 3026c2832898SQu Wenruo */ 3027c2832898SQu Wenruo if (in_range(cluster->boundary[*cluster_nr] - offset, 3028c2832898SQu Wenruo page_start, PAGE_SIZE)) { 3029c2832898SQu Wenruo u64 boundary_start = cluster->boundary[*cluster_nr] - 3030c2832898SQu Wenruo offset; 3031c2832898SQu Wenruo u64 boundary_end = boundary_start + 3032c2832898SQu Wenruo fs_info->sectorsize - 1; 3033c2832898SQu Wenruo 3034c2832898SQu Wenruo set_extent_bits(&BTRFS_I(inode)->io_tree, 3035c2832898SQu Wenruo boundary_start, boundary_end, 3036c2832898SQu Wenruo EXTENT_BOUNDARY); 3037c2832898SQu Wenruo } 3038c2832898SQu Wenruo unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end); 3039c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len); 3040c2832898SQu Wenruo cur += clamped_len; 3041c2832898SQu Wenruo 3042c2832898SQu Wenruo /* Crossed extent end, go to next extent */ 3043c2832898SQu Wenruo if (cur >= extent_end) { 3044c2832898SQu Wenruo (*cluster_nr)++; 3045c2832898SQu Wenruo /* Just finished the last extent of the cluster, exit. */ 3046c2832898SQu Wenruo if (*cluster_nr >= cluster->nr) 3047c2832898SQu Wenruo break; 3048c2832898SQu Wenruo } 3049c2832898SQu Wenruo } 3050f47960f4SQu Wenruo unlock_page(page); 3051f47960f4SQu Wenruo put_page(page); 3052f47960f4SQu Wenruo 3053f47960f4SQu Wenruo balance_dirty_pages_ratelimited(inode->i_mapping); 3054f47960f4SQu Wenruo btrfs_throttle(fs_info); 3055f47960f4SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) 3056f47960f4SQu Wenruo ret = -ECANCELED; 3057f47960f4SQu Wenruo return ret; 3058f47960f4SQu Wenruo 3059f47960f4SQu Wenruo release_page: 3060f47960f4SQu Wenruo unlock_page(page); 3061f47960f4SQu Wenruo put_page(page); 3062f47960f4SQu Wenruo return ret; 3063f47960f4SQu Wenruo } 3064f47960f4SQu Wenruo 30650257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 30660257bb82SYan, Zheng struct file_extent_cluster *cluster) 30670257bb82SYan, Zheng { 30680257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 30690257bb82SYan, Zheng unsigned long index; 30700257bb82SYan, Zheng unsigned long last_index; 30710257bb82SYan, Zheng struct file_ra_state *ra; 3072f47960f4SQu Wenruo int cluster_nr = 0; 30730257bb82SYan, Zheng int ret = 0; 30740257bb82SYan, Zheng 30750257bb82SYan, Zheng if (!cluster->nr) 30760257bb82SYan, Zheng return 0; 30770257bb82SYan, Zheng 30780257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 30790257bb82SYan, Zheng if (!ra) 30800257bb82SYan, Zheng return -ENOMEM; 30810257bb82SYan, Zheng 3082056d9becSNikolay Borisov ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster); 30830257bb82SYan, Zheng if (ret) 3084efa56464SYan, Zheng goto out; 30850257bb82SYan, Zheng 30860257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 30870257bb82SYan, Zheng 30884b01c44fSJohannes Thumshirn ret = setup_relocation_extent_mapping(inode, cluster->start - offset, 3089efa56464SYan, Zheng cluster->end - offset, cluster->start); 3090efa56464SYan, Zheng if (ret) 3091efa56464SYan, Zheng goto out; 3092efa56464SYan, Zheng 309309cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 3094f47960f4SQu Wenruo for (index = (cluster->start - offset) >> PAGE_SHIFT; 3095f47960f4SQu Wenruo index <= last_index && !ret; index++) 3096f47960f4SQu Wenruo ret = relocate_one_page(inode, ra, cluster, &cluster_nr, index); 3097f47960f4SQu Wenruo if (ret == 0) 3098f47960f4SQu Wenruo WARN_ON(cluster_nr != cluster->nr); 3099efa56464SYan, Zheng out: 31000257bb82SYan, Zheng kfree(ra); 31010257bb82SYan, Zheng return ret; 31020257bb82SYan, Zheng } 31030257bb82SYan, Zheng 31040257bb82SYan, Zheng static noinline_for_stack 31050257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 31060257bb82SYan, Zheng struct file_extent_cluster *cluster) 31070257bb82SYan, Zheng { 31080257bb82SYan, Zheng int ret; 31090257bb82SYan, Zheng 31100257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 31110257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31120257bb82SYan, Zheng if (ret) 31130257bb82SYan, Zheng return ret; 31140257bb82SYan, Zheng cluster->nr = 0; 31150257bb82SYan, Zheng } 31160257bb82SYan, Zheng 31170257bb82SYan, Zheng if (!cluster->nr) 31180257bb82SYan, Zheng cluster->start = extent_key->objectid; 31190257bb82SYan, Zheng else 31200257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 31210257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 31220257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 31230257bb82SYan, Zheng cluster->nr++; 31240257bb82SYan, Zheng 31250257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 31260257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31270257bb82SYan, Zheng if (ret) 31280257bb82SYan, Zheng return ret; 31290257bb82SYan, Zheng cluster->nr = 0; 31300257bb82SYan, Zheng } 31310257bb82SYan, Zheng return 0; 31325d4f98a2SYan Zheng } 31335d4f98a2SYan Zheng 31345d4f98a2SYan Zheng /* 31355d4f98a2SYan Zheng * helper to add a tree block to the list. 31365d4f98a2SYan Zheng * the major work is getting the generation and level of the block 31375d4f98a2SYan Zheng */ 31385d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 31395d4f98a2SYan Zheng struct btrfs_key *extent_key, 31405d4f98a2SYan Zheng struct btrfs_path *path, 31415d4f98a2SYan Zheng struct rb_root *blocks) 31425d4f98a2SYan Zheng { 31435d4f98a2SYan Zheng struct extent_buffer *eb; 31445d4f98a2SYan Zheng struct btrfs_extent_item *ei; 31455d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 31465d4f98a2SYan Zheng struct tree_block *block; 31475d4f98a2SYan Zheng struct rb_node *rb_node; 31485d4f98a2SYan Zheng u32 item_size; 31495d4f98a2SYan Zheng int level = -1; 31507fdf4b60SWang Shilong u64 generation; 3151f7ba2d37SJosef Bacik u64 owner = 0; 31525d4f98a2SYan Zheng 31535d4f98a2SYan Zheng eb = path->nodes[0]; 31543212fa14SJosef Bacik item_size = btrfs_item_size(eb, path->slots[0]); 31555d4f98a2SYan Zheng 31563173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 31573173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 3158f7ba2d37SJosef Bacik unsigned long ptr = 0, end; 3159f7ba2d37SJosef Bacik 31605d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 31615d4f98a2SYan Zheng struct btrfs_extent_item); 3162f7ba2d37SJosef Bacik end = (unsigned long)ei + item_size; 31633173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 31645d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 31655d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 3166f7ba2d37SJosef Bacik ptr = (unsigned long)(bi + 1); 31675d4f98a2SYan Zheng } else { 31683173a18fSJosef Bacik level = (int)extent_key->offset; 3169f7ba2d37SJosef Bacik ptr = (unsigned long)(ei + 1); 31703173a18fSJosef Bacik } 31713173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 3172f7ba2d37SJosef Bacik 3173f7ba2d37SJosef Bacik /* 3174f7ba2d37SJosef Bacik * We're reading random blocks without knowing their owner ahead 3175f7ba2d37SJosef Bacik * of time. This is ok most of the time, as all reloc roots and 3176f7ba2d37SJosef Bacik * fs roots have the same lock type. However normal trees do 3177f7ba2d37SJosef Bacik * not, and the only way to know ahead of time is to read the 3178f7ba2d37SJosef Bacik * inline ref offset. We know it's an fs root if 3179f7ba2d37SJosef Bacik * 3180f7ba2d37SJosef Bacik * 1. There's more than one ref. 3181f7ba2d37SJosef Bacik * 2. There's a SHARED_DATA_REF_KEY set. 3182f7ba2d37SJosef Bacik * 3. FULL_BACKREF is set on the flags. 3183f7ba2d37SJosef Bacik * 3184f7ba2d37SJosef Bacik * Otherwise it's safe to assume that the ref offset == the 3185f7ba2d37SJosef Bacik * owner of this block, so we can use that when calling 3186f7ba2d37SJosef Bacik * read_tree_block. 3187f7ba2d37SJosef Bacik */ 3188f7ba2d37SJosef Bacik if (btrfs_extent_refs(eb, ei) == 1 && 3189f7ba2d37SJosef Bacik !(btrfs_extent_flags(eb, ei) & 3190f7ba2d37SJosef Bacik BTRFS_BLOCK_FLAG_FULL_BACKREF) && 3191f7ba2d37SJosef Bacik ptr < end) { 3192f7ba2d37SJosef Bacik struct btrfs_extent_inline_ref *iref; 3193f7ba2d37SJosef Bacik int type; 3194f7ba2d37SJosef Bacik 3195f7ba2d37SJosef Bacik iref = (struct btrfs_extent_inline_ref *)ptr; 3196f7ba2d37SJosef Bacik type = btrfs_get_extent_inline_ref_type(eb, iref, 3197f7ba2d37SJosef Bacik BTRFS_REF_TYPE_BLOCK); 3198f7ba2d37SJosef Bacik if (type == BTRFS_REF_TYPE_INVALID) 3199f7ba2d37SJosef Bacik return -EINVAL; 3200f7ba2d37SJosef Bacik if (type == BTRFS_TREE_BLOCK_REF_KEY) 3201f7ba2d37SJosef Bacik owner = btrfs_extent_inline_ref_offset(eb, iref); 3202f7ba2d37SJosef Bacik } 32036d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3204ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3205ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3206ba3c2b19SNikolay Borisov return -EINVAL; 32073173a18fSJosef Bacik } else { 32085d4f98a2SYan Zheng BUG(); 32095d4f98a2SYan Zheng } 32105d4f98a2SYan Zheng 3211b3b4aa74SDavid Sterba btrfs_release_path(path); 32125d4f98a2SYan Zheng 32135d4f98a2SYan Zheng BUG_ON(level == -1); 32145d4f98a2SYan Zheng 32155d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 32165d4f98a2SYan Zheng if (!block) 32175d4f98a2SYan Zheng return -ENOMEM; 32185d4f98a2SYan Zheng 32195d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3220da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 32215d4f98a2SYan Zheng block->key.offset = generation; 32225d4f98a2SYan Zheng block->level = level; 32235d4f98a2SYan Zheng block->key_ready = 0; 3224f7ba2d37SJosef Bacik block->owner = owner; 32255d4f98a2SYan Zheng 3226e9a28dc5SQu Wenruo rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node); 322743c04fb1SJeff Mahoney if (rb_node) 3228982c92cbSQu Wenruo btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr, 3229982c92cbSQu Wenruo -EEXIST); 32305d4f98a2SYan Zheng 32315d4f98a2SYan Zheng return 0; 32325d4f98a2SYan Zheng } 32335d4f98a2SYan Zheng 32345d4f98a2SYan Zheng /* 32355d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 32365d4f98a2SYan Zheng */ 32375d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 32385d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 32395d4f98a2SYan Zheng struct rb_root *blocks) 32405d4f98a2SYan Zheng { 32410b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 32425d4f98a2SYan Zheng struct btrfs_path *path; 32435d4f98a2SYan Zheng struct btrfs_key key; 32445d4f98a2SYan Zheng int ret; 32450b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 32465d4f98a2SYan Zheng 32477476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 32485d4f98a2SYan Zheng return 0; 32495d4f98a2SYan Zheng 3250e9a28dc5SQu Wenruo if (rb_simple_search(blocks, bytenr)) 32515d4f98a2SYan Zheng return 0; 32525d4f98a2SYan Zheng 32535d4f98a2SYan Zheng path = btrfs_alloc_path(); 32545d4f98a2SYan Zheng if (!path) 32555d4f98a2SYan Zheng return -ENOMEM; 3256aee68ee5SJosef Bacik again: 32575d4f98a2SYan Zheng key.objectid = bytenr; 3258aee68ee5SJosef Bacik if (skinny) { 3259aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3260aee68ee5SJosef Bacik key.offset = (u64)-1; 3261aee68ee5SJosef Bacik } else { 32625d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 32635d4f98a2SYan Zheng key.offset = blocksize; 3264aee68ee5SJosef Bacik } 32655d4f98a2SYan Zheng 32665d4f98a2SYan Zheng path->search_commit_root = 1; 32675d4f98a2SYan Zheng path->skip_locking = 1; 32685d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 32695d4f98a2SYan Zheng if (ret < 0) 32705d4f98a2SYan Zheng goto out; 32715d4f98a2SYan Zheng 3272aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3273aee68ee5SJosef Bacik if (path->slots[0]) { 3274aee68ee5SJosef Bacik path->slots[0]--; 3275aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3276aee68ee5SJosef Bacik path->slots[0]); 32773173a18fSJosef Bacik if (key.objectid == bytenr && 3278aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3279aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3280aee68ee5SJosef Bacik key.offset == blocksize))) 32813173a18fSJosef Bacik ret = 0; 32823173a18fSJosef Bacik } 3283aee68ee5SJosef Bacik 3284aee68ee5SJosef Bacik if (ret) { 3285aee68ee5SJosef Bacik skinny = false; 3286aee68ee5SJosef Bacik btrfs_release_path(path); 3287aee68ee5SJosef Bacik goto again; 3288aee68ee5SJosef Bacik } 3289aee68ee5SJosef Bacik } 3290cdccee99SLiu Bo if (ret) { 3291cdccee99SLiu Bo ASSERT(ret == 1); 3292cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3293cdccee99SLiu Bo btrfs_err(fs_info, 3294cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3295cdccee99SLiu Bo bytenr); 3296cdccee99SLiu Bo WARN_ON(1); 3297cdccee99SLiu Bo ret = -EINVAL; 3298cdccee99SLiu Bo goto out; 3299cdccee99SLiu Bo } 33003173a18fSJosef Bacik 33015d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 33025d4f98a2SYan Zheng out: 33035d4f98a2SYan Zheng btrfs_free_path(path); 33045d4f98a2SYan Zheng return ret; 33055d4f98a2SYan Zheng } 33065d4f98a2SYan Zheng 33070af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 330832da5386SDavid Sterba struct btrfs_block_group *block_group, 33091bbc621eSChris Mason struct inode *inode, 33101bbc621eSChris Mason u64 ino) 33110af3d00bSJosef Bacik { 33120af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 33130af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 33140af3d00bSJosef Bacik int ret = 0; 33150af3d00bSJosef Bacik 33160af3d00bSJosef Bacik if (inode) 33170af3d00bSJosef Bacik goto truncate; 33180af3d00bSJosef Bacik 33190202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, ino, root); 33202e19f1f9SAl Viro if (IS_ERR(inode)) 33210af3d00bSJosef Bacik return -ENOENT; 33220af3d00bSJosef Bacik 33230af3d00bSJosef Bacik truncate: 33242ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 33257b61cd92SMiao Xie &fs_info->global_block_rsv); 33267b61cd92SMiao Xie if (ret) 33277b61cd92SMiao Xie goto out; 33287b61cd92SMiao Xie 33297a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 33300af3d00bSJosef Bacik if (IS_ERR(trans)) { 33313612b495STsutomu Itoh ret = PTR_ERR(trans); 33320af3d00bSJosef Bacik goto out; 33330af3d00bSJosef Bacik } 33340af3d00bSJosef Bacik 333577ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 33360af3d00bSJosef Bacik 33373a45bb20SJeff Mahoney btrfs_end_transaction(trans); 33382ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 33390af3d00bSJosef Bacik out: 33400af3d00bSJosef Bacik iput(inode); 33410af3d00bSJosef Bacik return ret; 33420af3d00bSJosef Bacik } 33430af3d00bSJosef Bacik 33445d4f98a2SYan Zheng /* 334519b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 334619b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 33475d4f98a2SYan Zheng */ 334819b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 334919b546d7SQu Wenruo struct btrfs_block_group *block_group, 335019b546d7SQu Wenruo u64 data_bytenr) 33515d4f98a2SYan Zheng { 335219b546d7SQu Wenruo u64 space_cache_ino; 335319b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 33545d4f98a2SYan Zheng struct btrfs_key key; 335519b546d7SQu Wenruo bool found = false; 335619b546d7SQu Wenruo int i; 33575d4f98a2SYan Zheng int ret; 33585d4f98a2SYan Zheng 335919b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 336019b546d7SQu Wenruo return 0; 33615d4f98a2SYan Zheng 336219b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 336350e31ef4SQu Wenruo u8 type; 336450e31ef4SQu Wenruo 336519b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 336619b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 336719b546d7SQu Wenruo continue; 336819b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 336950e31ef4SQu Wenruo type = btrfs_file_extent_type(leaf, ei); 337050e31ef4SQu Wenruo 337150e31ef4SQu Wenruo if ((type == BTRFS_FILE_EXTENT_REG || 337250e31ef4SQu Wenruo type == BTRFS_FILE_EXTENT_PREALLOC) && 337319b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 337419b546d7SQu Wenruo found = true; 337519b546d7SQu Wenruo space_cache_ino = key.objectid; 337619b546d7SQu Wenruo break; 337719b546d7SQu Wenruo } 337819b546d7SQu Wenruo } 337919b546d7SQu Wenruo if (!found) 338019b546d7SQu Wenruo return -ENOENT; 338119b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 338219b546d7SQu Wenruo space_cache_ino); 33830af3d00bSJosef Bacik return ret; 33845d4f98a2SYan Zheng } 33855d4f98a2SYan Zheng 33865d4f98a2SYan Zheng /* 33872c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 33885d4f98a2SYan Zheng */ 33895d4f98a2SYan Zheng static noinline_for_stack 33905d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 33915d4f98a2SYan Zheng struct btrfs_key *extent_key, 33925d4f98a2SYan Zheng struct btrfs_path *path, 33935d4f98a2SYan Zheng struct rb_root *blocks) 33945d4f98a2SYan Zheng { 339519b546d7SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 339619b546d7SQu Wenruo struct ulist *leaves = NULL; 339719b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 339819b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 339919b546d7SQu Wenruo const u32 blocksize = fs_info->nodesize; 3400647f63bdSFilipe David Borba Manana int ret = 0; 34015d4f98a2SYan Zheng 3402b3b4aa74SDavid Sterba btrfs_release_path(path); 340319b546d7SQu Wenruo ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid, 340419b546d7SQu Wenruo 0, &leaves, NULL, true); 340519b546d7SQu Wenruo if (ret < 0) 340619b546d7SQu Wenruo return ret; 340719b546d7SQu Wenruo 340819b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 340919b546d7SQu Wenruo while ((ref_node = ulist_next(leaves, &leaf_uiter))) { 341019b546d7SQu Wenruo struct extent_buffer *eb; 341119b546d7SQu Wenruo 34121b7ec85eSJosef Bacik eb = read_tree_block(fs_info, ref_node->val, 0, 0, 0, NULL); 341319b546d7SQu Wenruo if (IS_ERR(eb)) { 341419b546d7SQu Wenruo ret = PTR_ERR(eb); 341519b546d7SQu Wenruo break; 341619b546d7SQu Wenruo } 341719b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 341819b546d7SQu Wenruo extent_key->objectid); 341919b546d7SQu Wenruo free_extent_buffer(eb); 342019b546d7SQu Wenruo if (ret < 0) 342119b546d7SQu Wenruo break; 342219b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 342319b546d7SQu Wenruo if (ret < 0) 342419b546d7SQu Wenruo break; 342519b546d7SQu Wenruo } 342619b546d7SQu Wenruo if (ret < 0) 34275d4f98a2SYan Zheng free_block_list(blocks); 342819b546d7SQu Wenruo ulist_free(leaves); 342919b546d7SQu Wenruo return ret; 34305d4f98a2SYan Zheng } 34315d4f98a2SYan Zheng 34325d4f98a2SYan Zheng /* 34332c016dc2SLiu Bo * helper to find next unprocessed extent 34345d4f98a2SYan Zheng */ 34355d4f98a2SYan Zheng static noinline_for_stack 3436147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 34373fd0a558SYan, Zheng struct btrfs_key *extent_key) 34385d4f98a2SYan Zheng { 34390b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34405d4f98a2SYan Zheng struct btrfs_key key; 34415d4f98a2SYan Zheng struct extent_buffer *leaf; 34425d4f98a2SYan Zheng u64 start, end, last; 34435d4f98a2SYan Zheng int ret; 34445d4f98a2SYan Zheng 3445b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 34465d4f98a2SYan Zheng while (1) { 34475d4f98a2SYan Zheng cond_resched(); 34485d4f98a2SYan Zheng if (rc->search_start >= last) { 34495d4f98a2SYan Zheng ret = 1; 34505d4f98a2SYan Zheng break; 34515d4f98a2SYan Zheng } 34525d4f98a2SYan Zheng 34535d4f98a2SYan Zheng key.objectid = rc->search_start; 34545d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34555d4f98a2SYan Zheng key.offset = 0; 34565d4f98a2SYan Zheng 34575d4f98a2SYan Zheng path->search_commit_root = 1; 34585d4f98a2SYan Zheng path->skip_locking = 1; 34595d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 34605d4f98a2SYan Zheng 0, 0); 34615d4f98a2SYan Zheng if (ret < 0) 34625d4f98a2SYan Zheng break; 34635d4f98a2SYan Zheng next: 34645d4f98a2SYan Zheng leaf = path->nodes[0]; 34655d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 34665d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 34675d4f98a2SYan Zheng if (ret != 0) 34685d4f98a2SYan Zheng break; 34695d4f98a2SYan Zheng leaf = path->nodes[0]; 34705d4f98a2SYan Zheng } 34715d4f98a2SYan Zheng 34725d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 34735d4f98a2SYan Zheng if (key.objectid >= last) { 34745d4f98a2SYan Zheng ret = 1; 34755d4f98a2SYan Zheng break; 34765d4f98a2SYan Zheng } 34775d4f98a2SYan Zheng 34783173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 34793173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 34803173a18fSJosef Bacik path->slots[0]++; 34813173a18fSJosef Bacik goto next; 34823173a18fSJosef Bacik } 34833173a18fSJosef Bacik 34843173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 34855d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 34865d4f98a2SYan Zheng path->slots[0]++; 34875d4f98a2SYan Zheng goto next; 34885d4f98a2SYan Zheng } 34895d4f98a2SYan Zheng 34903173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 34910b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 34923173a18fSJosef Bacik rc->search_start) { 34933173a18fSJosef Bacik path->slots[0]++; 34943173a18fSJosef Bacik goto next; 34953173a18fSJosef Bacik } 34963173a18fSJosef Bacik 34975d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 34985d4f98a2SYan Zheng key.objectid, &start, &end, 3499e6138876SJosef Bacik EXTENT_DIRTY, NULL); 35005d4f98a2SYan Zheng 35015d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3502b3b4aa74SDavid Sterba btrfs_release_path(path); 35035d4f98a2SYan Zheng rc->search_start = end + 1; 35045d4f98a2SYan Zheng } else { 35053173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 35065d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 35073173a18fSJosef Bacik else 35083173a18fSJosef Bacik rc->search_start = key.objectid + 35090b246afaSJeff Mahoney fs_info->nodesize; 35103fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 35115d4f98a2SYan Zheng return 0; 35125d4f98a2SYan Zheng } 35135d4f98a2SYan Zheng } 3514b3b4aa74SDavid Sterba btrfs_release_path(path); 35155d4f98a2SYan Zheng return ret; 35165d4f98a2SYan Zheng } 35175d4f98a2SYan Zheng 35185d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 35195d4f98a2SYan Zheng { 35205d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35217585717fSChris Mason 35227585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35235d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 35247585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35255d4f98a2SYan Zheng } 35265d4f98a2SYan Zheng 35275d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 35285d4f98a2SYan Zheng { 35295d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35307585717fSChris Mason 35317585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35325d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 35337585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35345d4f98a2SYan Zheng } 35355d4f98a2SYan Zheng 35363fd0a558SYan, Zheng static noinline_for_stack 35373fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 35383fd0a558SYan, Zheng { 35393fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3540ac2fabacSJosef Bacik int ret; 35413fd0a558SYan, Zheng 35422ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 354366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 35443fd0a558SYan, Zheng if (!rc->block_rsv) 35453fd0a558SYan, Zheng return -ENOMEM; 35463fd0a558SYan, Zheng 35473fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3548b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 35493fd0a558SYan, Zheng rc->extents_found = 0; 35503fd0a558SYan, Zheng rc->nodes_relocated = 0; 35513fd0a558SYan, Zheng rc->merging_rsv_size = 0; 35520647bf56SWang Shilong rc->reserved_bytes = 0; 3553da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 35540647bf56SWang Shilong RELOCATION_RESERVED_NODES; 35559270501cSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root->fs_info, 3556ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3557ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3558ac2fabacSJosef Bacik if (ret) 3559ac2fabacSJosef Bacik return ret; 35603fd0a558SYan, Zheng 35613fd0a558SYan, Zheng rc->create_reloc_tree = 1; 35623fd0a558SYan, Zheng set_reloc_control(rc); 35633fd0a558SYan, Zheng 35647a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 356528818947SLiu Bo if (IS_ERR(trans)) { 356628818947SLiu Bo unset_reloc_control(rc); 356728818947SLiu Bo /* 356828818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 356928818947SLiu Bo * cleanup. And callers are responsible to free the above 357028818947SLiu Bo * block rsv. 357128818947SLiu Bo */ 357228818947SLiu Bo return PTR_ERR(trans); 357328818947SLiu Bo } 3574fb686c68SJosef Bacik return btrfs_commit_transaction(trans); 35753fd0a558SYan, Zheng } 357676dda93cSYan, Zheng 35775d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 35785d4f98a2SYan Zheng { 35792ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35805d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 35815d4f98a2SYan Zheng struct btrfs_key key; 35825d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 35835d4f98a2SYan Zheng struct btrfs_path *path; 35845d4f98a2SYan Zheng struct btrfs_extent_item *ei; 35855d4f98a2SYan Zheng u64 flags; 35865d4f98a2SYan Zheng int ret; 35875d4f98a2SYan Zheng int err = 0; 3588c87f08caSChris Mason int progress = 0; 35895d4f98a2SYan Zheng 35905d4f98a2SYan Zheng path = btrfs_alloc_path(); 35913fd0a558SYan, Zheng if (!path) 35925d4f98a2SYan Zheng return -ENOMEM; 3593e4058b54SDavid Sterba path->reada = READA_FORWARD; 35943fd0a558SYan, Zheng 35953fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 35963fd0a558SYan, Zheng if (ret) { 35973fd0a558SYan, Zheng err = ret; 35983fd0a558SYan, Zheng goto out_free; 35992423fdfbSJiri Slaby } 36005d4f98a2SYan Zheng 36015d4f98a2SYan Zheng while (1) { 36020647bf56SWang Shilong rc->reserved_bytes = 0; 36039270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 36049270501cSJosef Bacik rc->block_rsv->size, 36050647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 36060647bf56SWang Shilong if (ret) { 36070647bf56SWang Shilong err = ret; 36080647bf56SWang Shilong break; 36090647bf56SWang Shilong } 3610c87f08caSChris Mason progress++; 3611a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 36120f788c58SLiu Bo if (IS_ERR(trans)) { 36130f788c58SLiu Bo err = PTR_ERR(trans); 36140f788c58SLiu Bo trans = NULL; 36150f788c58SLiu Bo break; 36160f788c58SLiu Bo } 3617c87f08caSChris Mason restart: 36183fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 36193a45bb20SJeff Mahoney btrfs_end_transaction(trans); 362042a657f5SPan Bian trans = NULL; 36213fd0a558SYan, Zheng continue; 36223fd0a558SYan, Zheng } 36233fd0a558SYan, Zheng 3624147d256eSZhaolei ret = find_next_extent(rc, path, &key); 36255d4f98a2SYan Zheng if (ret < 0) 36265d4f98a2SYan Zheng err = ret; 36275d4f98a2SYan Zheng if (ret != 0) 36285d4f98a2SYan Zheng break; 36295d4f98a2SYan Zheng 36305d4f98a2SYan Zheng rc->extents_found++; 36315d4f98a2SYan Zheng 36325d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 36335d4f98a2SYan Zheng struct btrfs_extent_item); 36345d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 36355d4f98a2SYan Zheng 36365d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 36375d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 36385d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 36395d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36405d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 36415d4f98a2SYan Zheng } else { 3642b3b4aa74SDavid Sterba btrfs_release_path(path); 36435d4f98a2SYan Zheng ret = 0; 36445d4f98a2SYan Zheng } 36455d4f98a2SYan Zheng if (ret < 0) { 36463fd0a558SYan, Zheng err = ret; 36475d4f98a2SYan Zheng break; 36485d4f98a2SYan Zheng } 36495d4f98a2SYan Zheng 36505d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 36515d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 36525d4f98a2SYan Zheng if (ret < 0) { 36533fd0a558SYan, Zheng if (ret != -EAGAIN) { 36545d4f98a2SYan Zheng err = ret; 36555d4f98a2SYan Zheng break; 36565d4f98a2SYan Zheng } 36573fd0a558SYan, Zheng rc->extents_found--; 36583fd0a558SYan, Zheng rc->search_start = key.objectid; 36593fd0a558SYan, Zheng } 36605d4f98a2SYan Zheng } 36615d4f98a2SYan Zheng 36623a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 36632ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36643fd0a558SYan, Zheng trans = NULL; 36655d4f98a2SYan Zheng 36665d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 36675d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36685d4f98a2SYan Zheng rc->found_file_extent = 1; 36690257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 36703fd0a558SYan, Zheng &key, &rc->cluster); 36715d4f98a2SYan Zheng if (ret < 0) { 36725d4f98a2SYan Zheng err = ret; 36735d4f98a2SYan Zheng break; 36745d4f98a2SYan Zheng } 36755d4f98a2SYan Zheng } 3676f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 3677f31ea088SQu Wenruo err = -ECANCELED; 3678f31ea088SQu Wenruo break; 3679f31ea088SQu Wenruo } 36805d4f98a2SYan Zheng } 3681c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 368243a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 36839689457bSShilong Wang if (ret == 1) { 3684c87f08caSChris Mason err = 0; 3685c87f08caSChris Mason progress = 0; 3686c87f08caSChris Mason goto restart; 3687c87f08caSChris Mason } 3688c87f08caSChris Mason } 36893fd0a558SYan, Zheng 3690b3b4aa74SDavid Sterba btrfs_release_path(path); 369191166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 36925d4f98a2SYan Zheng 36935d4f98a2SYan Zheng if (trans) { 36943a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 36952ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36965d4f98a2SYan Zheng } 36975d4f98a2SYan Zheng 36980257bb82SYan, Zheng if (!err) { 36993fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 37003fd0a558SYan, Zheng &rc->cluster); 37010257bb82SYan, Zheng if (ret < 0) 37020257bb82SYan, Zheng err = ret; 37030257bb82SYan, Zheng } 37040257bb82SYan, Zheng 37053fd0a558SYan, Zheng rc->create_reloc_tree = 0; 37063fd0a558SYan, Zheng set_reloc_control(rc); 37070257bb82SYan, Zheng 370813fe1bdbSQu Wenruo btrfs_backref_release_cache(&rc->backref_cache); 370963f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37105d4f98a2SYan Zheng 37117f913c7cSQu Wenruo /* 37127f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 37137f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 37147f913c7cSQu Wenruo * 37157f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 37167f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 37177f913c7cSQu Wenruo * merge_reloc_roots() 37187f913c7cSQu Wenruo */ 37193fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 37205d4f98a2SYan Zheng 37215d4f98a2SYan Zheng merge_reloc_roots(rc); 37225d4f98a2SYan Zheng 37233fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 37245d4f98a2SYan Zheng unset_reloc_control(rc); 372563f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37265d4f98a2SYan Zheng 37275d4f98a2SYan Zheng /* get rid of pinned extents */ 37287a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 372962b99540SQu Wenruo if (IS_ERR(trans)) { 37303612b495STsutomu Itoh err = PTR_ERR(trans); 373162b99540SQu Wenruo goto out_free; 373262b99540SQu Wenruo } 3733fb686c68SJosef Bacik ret = btrfs_commit_transaction(trans); 3734fb686c68SJosef Bacik if (ret && !err) 3735fb686c68SJosef Bacik err = ret; 37366217b0faSJosef Bacik out_free: 3737d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 3738d2311e69SQu Wenruo if (ret < 0 && !err) 3739d2311e69SQu Wenruo err = ret; 37402ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 37413fd0a558SYan, Zheng btrfs_free_path(path); 37425d4f98a2SYan Zheng return err; 37435d4f98a2SYan Zheng } 37445d4f98a2SYan Zheng 37455d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 37460257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 37475d4f98a2SYan Zheng { 37485d4f98a2SYan Zheng struct btrfs_path *path; 37495d4f98a2SYan Zheng struct btrfs_inode_item *item; 37505d4f98a2SYan Zheng struct extent_buffer *leaf; 37515d4f98a2SYan Zheng int ret; 37525d4f98a2SYan Zheng 37535d4f98a2SYan Zheng path = btrfs_alloc_path(); 37545d4f98a2SYan Zheng if (!path) 37555d4f98a2SYan Zheng return -ENOMEM; 37565d4f98a2SYan Zheng 37575d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 37585d4f98a2SYan Zheng if (ret) 37595d4f98a2SYan Zheng goto out; 37605d4f98a2SYan Zheng 37615d4f98a2SYan Zheng leaf = path->nodes[0]; 37625d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 3763b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 37645d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 37650257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 37665d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 3767960a3166SJohannes Thumshirn btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 3768960a3166SJohannes Thumshirn BTRFS_INODE_PREALLOC); 37695d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 37705d4f98a2SYan Zheng out: 37715d4f98a2SYan Zheng btrfs_free_path(path); 37725d4f98a2SYan Zheng return ret; 37735d4f98a2SYan Zheng } 37745d4f98a2SYan Zheng 3775790c1b8cSJosef Bacik static void delete_orphan_inode(struct btrfs_trans_handle *trans, 3776790c1b8cSJosef Bacik struct btrfs_root *root, u64 objectid) 3777790c1b8cSJosef Bacik { 3778790c1b8cSJosef Bacik struct btrfs_path *path; 3779790c1b8cSJosef Bacik struct btrfs_key key; 3780790c1b8cSJosef Bacik int ret = 0; 3781790c1b8cSJosef Bacik 3782790c1b8cSJosef Bacik path = btrfs_alloc_path(); 3783790c1b8cSJosef Bacik if (!path) { 3784790c1b8cSJosef Bacik ret = -ENOMEM; 3785790c1b8cSJosef Bacik goto out; 3786790c1b8cSJosef Bacik } 3787790c1b8cSJosef Bacik 3788790c1b8cSJosef Bacik key.objectid = objectid; 3789790c1b8cSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 3790790c1b8cSJosef Bacik key.offset = 0; 3791790c1b8cSJosef Bacik ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3792790c1b8cSJosef Bacik if (ret) { 3793790c1b8cSJosef Bacik if (ret > 0) 3794790c1b8cSJosef Bacik ret = -ENOENT; 3795790c1b8cSJosef Bacik goto out; 3796790c1b8cSJosef Bacik } 3797790c1b8cSJosef Bacik ret = btrfs_del_item(trans, root, path); 3798790c1b8cSJosef Bacik out: 3799790c1b8cSJosef Bacik if (ret) 3800790c1b8cSJosef Bacik btrfs_abort_transaction(trans, ret); 3801790c1b8cSJosef Bacik btrfs_free_path(path); 3802790c1b8cSJosef Bacik } 3803790c1b8cSJosef Bacik 38045d4f98a2SYan Zheng /* 38055d4f98a2SYan Zheng * helper to create inode for data relocation. 38065d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 38075d4f98a2SYan Zheng */ 38083fd0a558SYan, Zheng static noinline_for_stack 38093fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 381032da5386SDavid Sterba struct btrfs_block_group *group) 38115d4f98a2SYan Zheng { 38125d4f98a2SYan Zheng struct inode *inode = NULL; 38135d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 38145d4f98a2SYan Zheng struct btrfs_root *root; 38154624900dSZhaolei u64 objectid; 38165d4f98a2SYan Zheng int err = 0; 38175d4f98a2SYan Zheng 3818aeb935a4SQu Wenruo root = btrfs_grab_root(fs_info->data_reloc_root); 3819a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 382076deacf0SJosef Bacik if (IS_ERR(trans)) { 382100246528SJosef Bacik btrfs_put_root(root); 38223fd0a558SYan, Zheng return ERR_CAST(trans); 382376deacf0SJosef Bacik } 38245d4f98a2SYan Zheng 3825543068a2SNikolay Borisov err = btrfs_get_free_objectid(root, &objectid); 38265d4f98a2SYan Zheng if (err) 38275d4f98a2SYan Zheng goto out; 38285d4f98a2SYan Zheng 38290257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 3830790c1b8cSJosef Bacik if (err) 3831790c1b8cSJosef Bacik goto out; 38325d4f98a2SYan Zheng 38330202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, objectid, root); 3834790c1b8cSJosef Bacik if (IS_ERR(inode)) { 3835790c1b8cSJosef Bacik delete_orphan_inode(trans, root, objectid); 3836790c1b8cSJosef Bacik err = PTR_ERR(inode); 3837790c1b8cSJosef Bacik inode = NULL; 3838790c1b8cSJosef Bacik goto out; 3839790c1b8cSJosef Bacik } 3840b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 38415d4f98a2SYan Zheng 384273f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 38435d4f98a2SYan Zheng out: 384400246528SJosef Bacik btrfs_put_root(root); 38453a45bb20SJeff Mahoney btrfs_end_transaction(trans); 38462ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 38475d4f98a2SYan Zheng if (err) { 38485d4f98a2SYan Zheng if (inode) 38495d4f98a2SYan Zheng iput(inode); 38505d4f98a2SYan Zheng inode = ERR_PTR(err); 38515d4f98a2SYan Zheng } 38525d4f98a2SYan Zheng return inode; 38535d4f98a2SYan Zheng } 38545d4f98a2SYan Zheng 3855907d2710SDavid Sterba /* 3856907d2710SDavid Sterba * Mark start of chunk relocation that is cancellable. Check if the cancellation 3857907d2710SDavid Sterba * has been requested meanwhile and don't start in that case. 3858907d2710SDavid Sterba * 3859907d2710SDavid Sterba * Return: 3860907d2710SDavid Sterba * 0 success 3861907d2710SDavid Sterba * -EINPROGRESS operation is already in progress, that's probably a bug 3862907d2710SDavid Sterba * -ECANCELED cancellation request was set before the operation started 3863907d2710SDavid Sterba */ 3864907d2710SDavid Sterba static int reloc_chunk_start(struct btrfs_fs_info *fs_info) 3865907d2710SDavid Sterba { 3866907d2710SDavid Sterba if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) { 3867907d2710SDavid Sterba /* This should not happen */ 3868907d2710SDavid Sterba btrfs_err(fs_info, "reloc already running, cannot start"); 3869907d2710SDavid Sterba return -EINPROGRESS; 3870907d2710SDavid Sterba } 3871907d2710SDavid Sterba 3872907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) { 3873907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled on start"); 3874907d2710SDavid Sterba /* 3875907d2710SDavid Sterba * On cancel, clear all requests but let the caller mark 3876907d2710SDavid Sterba * the end after cleanup operations. 3877907d2710SDavid Sterba */ 3878907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3879907d2710SDavid Sterba return -ECANCELED; 3880907d2710SDavid Sterba } 3881907d2710SDavid Sterba return 0; 3882907d2710SDavid Sterba } 3883907d2710SDavid Sterba 3884907d2710SDavid Sterba /* 3885907d2710SDavid Sterba * Mark end of chunk relocation that is cancellable and wake any waiters. 3886907d2710SDavid Sterba */ 3887907d2710SDavid Sterba static void reloc_chunk_end(struct btrfs_fs_info *fs_info) 3888907d2710SDavid Sterba { 3889907d2710SDavid Sterba /* Requested after start, clear bit first so any waiters can continue */ 3890907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) 3891907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled during operation"); 3892907d2710SDavid Sterba clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags); 3893907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3894907d2710SDavid Sterba } 3895907d2710SDavid Sterba 3896c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 38973fd0a558SYan, Zheng { 38983fd0a558SYan, Zheng struct reloc_control *rc; 38993fd0a558SYan, Zheng 39003fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 39013fd0a558SYan, Zheng if (!rc) 39023fd0a558SYan, Zheng return NULL; 39033fd0a558SYan, Zheng 39043fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 3905d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 3906584fb121SQu Wenruo btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1); 39073fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 390843eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 390943eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 39103fd0a558SYan, Zheng return rc; 39113fd0a558SYan, Zheng } 39123fd0a558SYan, Zheng 39131a0afa0eSJosef Bacik static void free_reloc_control(struct reloc_control *rc) 39141a0afa0eSJosef Bacik { 39151a0afa0eSJosef Bacik struct mapping_node *node, *tmp; 39161a0afa0eSJosef Bacik 39171a0afa0eSJosef Bacik free_reloc_roots(&rc->reloc_roots); 39181a0afa0eSJosef Bacik rbtree_postorder_for_each_entry_safe(node, tmp, 39191a0afa0eSJosef Bacik &rc->reloc_root_tree.rb_root, rb_node) 39201a0afa0eSJosef Bacik kfree(node); 39211a0afa0eSJosef Bacik 39221a0afa0eSJosef Bacik kfree(rc); 39231a0afa0eSJosef Bacik } 39241a0afa0eSJosef Bacik 39255d4f98a2SYan Zheng /* 3926ebce0e01SAdam Borowski * Print the block group being relocated 3927ebce0e01SAdam Borowski */ 3928ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 392932da5386SDavid Sterba struct btrfs_block_group *block_group) 3930ebce0e01SAdam Borowski { 3931f89e09cfSAnand Jain char buf[128] = {'\0'}; 3932ebce0e01SAdam Borowski 3933f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 3934ebce0e01SAdam Borowski 3935ebce0e01SAdam Borowski btrfs_info(fs_info, 3936ebce0e01SAdam Borowski "relocating block group %llu flags %s", 3937b3470b5dSDavid Sterba block_group->start, buf); 3938ebce0e01SAdam Borowski } 3939ebce0e01SAdam Borowski 3940430640e3SQu Wenruo static const char *stage_to_string(int stage) 3941430640e3SQu Wenruo { 3942430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 3943430640e3SQu Wenruo return "move data extents"; 3944430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 3945430640e3SQu Wenruo return "update data pointers"; 3946430640e3SQu Wenruo return "unknown"; 3947430640e3SQu Wenruo } 3948430640e3SQu Wenruo 3949ebce0e01SAdam Borowski /* 39505d4f98a2SYan Zheng * function to relocate all extents in a block group. 39515d4f98a2SYan Zheng */ 39526bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 39535d4f98a2SYan Zheng { 395432da5386SDavid Sterba struct btrfs_block_group *bg; 395529cbcf40SJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start); 39565d4f98a2SYan Zheng struct reloc_control *rc; 39570af3d00bSJosef Bacik struct inode *inode; 39580af3d00bSJosef Bacik struct btrfs_path *path; 39595d4f98a2SYan Zheng int ret; 3960f0486c68SYan, Zheng int rw = 0; 39615d4f98a2SYan Zheng int err = 0; 39625d4f98a2SYan Zheng 3963*b4be6aefSJosef Bacik /* 3964*b4be6aefSJosef Bacik * This only gets set if we had a half-deleted snapshot on mount. We 3965*b4be6aefSJosef Bacik * cannot allow relocation to start while we're still trying to clean up 3966*b4be6aefSJosef Bacik * these pending deletions. 3967*b4be6aefSJosef Bacik */ 3968*b4be6aefSJosef Bacik ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE); 3969*b4be6aefSJosef Bacik if (ret) 3970*b4be6aefSJosef Bacik return ret; 3971*b4be6aefSJosef Bacik 3972*b4be6aefSJosef Bacik /* We may have been woken up by close_ctree, so bail if we're closing. */ 3973*b4be6aefSJosef Bacik if (btrfs_fs_closing(fs_info)) 3974*b4be6aefSJosef Bacik return -EINTR; 3975*b4be6aefSJosef Bacik 3976eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 3977eede2bf3SOmar Sandoval if (!bg) 3978eede2bf3SOmar Sandoval return -ENOENT; 3979eede2bf3SOmar Sandoval 3980eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 3981eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 3982eede2bf3SOmar Sandoval return -ETXTBSY; 3983eede2bf3SOmar Sandoval } 3984eede2bf3SOmar Sandoval 3985c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 3986eede2bf3SOmar Sandoval if (!rc) { 3987eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 39885d4f98a2SYan Zheng return -ENOMEM; 3989eede2bf3SOmar Sandoval } 39905d4f98a2SYan Zheng 3991907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 3992907d2710SDavid Sterba if (ret < 0) { 3993907d2710SDavid Sterba err = ret; 3994907d2710SDavid Sterba goto out_put_bg; 3995907d2710SDavid Sterba } 3996907d2710SDavid Sterba 3997f0486c68SYan, Zheng rc->extent_root = extent_root; 3998eede2bf3SOmar Sandoval rc->block_group = bg; 39995d4f98a2SYan Zheng 4000b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4001f0486c68SYan, Zheng if (ret) { 4002f0486c68SYan, Zheng err = ret; 4003f0486c68SYan, Zheng goto out; 4004f0486c68SYan, Zheng } 4005f0486c68SYan, Zheng rw = 1; 4006f0486c68SYan, Zheng 40070af3d00bSJosef Bacik path = btrfs_alloc_path(); 40080af3d00bSJosef Bacik if (!path) { 40090af3d00bSJosef Bacik err = -ENOMEM; 40100af3d00bSJosef Bacik goto out; 40110af3d00bSJosef Bacik } 40120af3d00bSJosef Bacik 40137949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 40140af3d00bSJosef Bacik btrfs_free_path(path); 40150af3d00bSJosef Bacik 40160af3d00bSJosef Bacik if (!IS_ERR(inode)) 40171bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 40180af3d00bSJosef Bacik else 40190af3d00bSJosef Bacik ret = PTR_ERR(inode); 40200af3d00bSJosef Bacik 40210af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 40220af3d00bSJosef Bacik err = ret; 40230af3d00bSJosef Bacik goto out; 40240af3d00bSJosef Bacik } 40250af3d00bSJosef Bacik 40265d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 40275d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 40285d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 40295d4f98a2SYan Zheng rc->data_inode = NULL; 40305d4f98a2SYan Zheng goto out; 40315d4f98a2SYan Zheng } 40325d4f98a2SYan Zheng 40330b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 40345d4f98a2SYan Zheng 40359cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4036f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 40376374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4038b3470b5dSDavid Sterba rc->block_group->start, 4039b3470b5dSDavid Sterba rc->block_group->length); 40405d4f98a2SYan Zheng 40417ae9bd18SNaohiro Aota ret = btrfs_zone_finish(rc->block_group); 40427ae9bd18SNaohiro Aota WARN_ON(ret && ret != -EAGAIN); 40437ae9bd18SNaohiro Aota 40445d4f98a2SYan Zheng while (1) { 4045430640e3SQu Wenruo int finishes_stage; 4046430640e3SQu Wenruo 404776dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 40485d4f98a2SYan Zheng ret = relocate_block_group(rc); 404976dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4050ff612ba7SJosef Bacik if (ret < 0) 40515d4f98a2SYan Zheng err = ret; 4052ff612ba7SJosef Bacik 4053430640e3SQu Wenruo finishes_stage = rc->stage; 4054ff612ba7SJosef Bacik /* 4055ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4056ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4057ff612ba7SJosef Bacik * different block group we could end up hitting the 4058ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4059ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4060ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4061ff612ba7SJosef Bacik * out of the loop if we hit an error. 4062ff612ba7SJosef Bacik */ 4063ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4064ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4065ff612ba7SJosef Bacik (u64)-1); 4066ff612ba7SJosef Bacik if (ret) 4067ff612ba7SJosef Bacik err = ret; 4068ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4069ff612ba7SJosef Bacik 0, -1); 4070ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 40715d4f98a2SYan Zheng } 40725d4f98a2SYan Zheng 4073ff612ba7SJosef Bacik if (err < 0) 4074ff612ba7SJosef Bacik goto out; 4075ff612ba7SJosef Bacik 40765d4f98a2SYan Zheng if (rc->extents_found == 0) 40775d4f98a2SYan Zheng break; 40785d4f98a2SYan Zheng 4079430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4080430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 40815d4f98a2SYan Zheng } 40825d4f98a2SYan Zheng 40835d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 40845d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4085bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 40865d4f98a2SYan Zheng out: 4087f0486c68SYan, Zheng if (err && rw) 40882ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 40895d4f98a2SYan Zheng iput(rc->data_inode); 4090907d2710SDavid Sterba out_put_bg: 4091907d2710SDavid Sterba btrfs_put_block_group(bg); 4092907d2710SDavid Sterba reloc_chunk_end(fs_info); 40931a0afa0eSJosef Bacik free_reloc_control(rc); 40945d4f98a2SYan Zheng return err; 40955d4f98a2SYan Zheng } 40965d4f98a2SYan Zheng 409776dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 409876dda93cSYan, Zheng { 40990b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 410076dda93cSYan, Zheng struct btrfs_trans_handle *trans; 410179787eaaSJeff Mahoney int ret, err; 410276dda93cSYan, Zheng 41030b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 410479787eaaSJeff Mahoney if (IS_ERR(trans)) 410579787eaaSJeff Mahoney return PTR_ERR(trans); 410676dda93cSYan, Zheng 410776dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 410876dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 4109c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 411076dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 41110b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 411276dda93cSYan, Zheng &root->root_key, &root->root_item); 411376dda93cSYan, Zheng 41143a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 411579787eaaSJeff Mahoney if (err) 411679787eaaSJeff Mahoney return err; 411779787eaaSJeff Mahoney return ret; 411876dda93cSYan, Zheng } 411976dda93cSYan, Zheng 41205d4f98a2SYan Zheng /* 41215d4f98a2SYan Zheng * recover relocation interrupted by system crash. 41225d4f98a2SYan Zheng * 41235d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 41245d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 41255d4f98a2SYan Zheng */ 41265d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 41275d4f98a2SYan Zheng { 41280b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 41295d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 41305d4f98a2SYan Zheng struct btrfs_key key; 41315d4f98a2SYan Zheng struct btrfs_root *fs_root; 41325d4f98a2SYan Zheng struct btrfs_root *reloc_root; 41335d4f98a2SYan Zheng struct btrfs_path *path; 41345d4f98a2SYan Zheng struct extent_buffer *leaf; 41355d4f98a2SYan Zheng struct reloc_control *rc = NULL; 41365d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41375d4f98a2SYan Zheng int ret; 41385d4f98a2SYan Zheng int err = 0; 41395d4f98a2SYan Zheng 41405d4f98a2SYan Zheng path = btrfs_alloc_path(); 41415d4f98a2SYan Zheng if (!path) 41425d4f98a2SYan Zheng return -ENOMEM; 4143e4058b54SDavid Sterba path->reada = READA_BACK; 41445d4f98a2SYan Zheng 41455d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 41465d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 41475d4f98a2SYan Zheng key.offset = (u64)-1; 41485d4f98a2SYan Zheng 41495d4f98a2SYan Zheng while (1) { 41500b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 41515d4f98a2SYan Zheng path, 0, 0); 41525d4f98a2SYan Zheng if (ret < 0) { 41535d4f98a2SYan Zheng err = ret; 41545d4f98a2SYan Zheng goto out; 41555d4f98a2SYan Zheng } 41565d4f98a2SYan Zheng if (ret > 0) { 41575d4f98a2SYan Zheng if (path->slots[0] == 0) 41585d4f98a2SYan Zheng break; 41595d4f98a2SYan Zheng path->slots[0]--; 41605d4f98a2SYan Zheng } 41615d4f98a2SYan Zheng leaf = path->nodes[0]; 41625d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4163b3b4aa74SDavid Sterba btrfs_release_path(path); 41645d4f98a2SYan Zheng 41655d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 41665d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 41675d4f98a2SYan Zheng break; 41685d4f98a2SYan Zheng 41693dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(root, &key); 41705d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 41715d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 41725d4f98a2SYan Zheng goto out; 41735d4f98a2SYan Zheng } 41745d4f98a2SYan Zheng 417592a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 41765d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 41775d4f98a2SYan Zheng 41785d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 4179a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, 4180a820feb5SDavid Sterba reloc_root->root_key.offset, false); 41815d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 418276dda93cSYan, Zheng ret = PTR_ERR(fs_root); 418376dda93cSYan, Zheng if (ret != -ENOENT) { 418476dda93cSYan, Zheng err = ret; 41855d4f98a2SYan Zheng goto out; 41865d4f98a2SYan Zheng } 418779787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 418879787eaaSJeff Mahoney if (ret < 0) { 418979787eaaSJeff Mahoney err = ret; 419079787eaaSJeff Mahoney goto out; 419179787eaaSJeff Mahoney } 4192932fd26dSJosef Bacik } else { 419300246528SJosef Bacik btrfs_put_root(fs_root); 419476dda93cSYan, Zheng } 41955d4f98a2SYan Zheng } 41965d4f98a2SYan Zheng 41975d4f98a2SYan Zheng if (key.offset == 0) 41985d4f98a2SYan Zheng break; 41995d4f98a2SYan Zheng 42005d4f98a2SYan Zheng key.offset--; 42015d4f98a2SYan Zheng } 4202b3b4aa74SDavid Sterba btrfs_release_path(path); 42035d4f98a2SYan Zheng 42045d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 42055d4f98a2SYan Zheng goto out; 42065d4f98a2SYan Zheng 4207c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 42085d4f98a2SYan Zheng if (!rc) { 42095d4f98a2SYan Zheng err = -ENOMEM; 42105d4f98a2SYan Zheng goto out; 42115d4f98a2SYan Zheng } 42125d4f98a2SYan Zheng 4213907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4214907d2710SDavid Sterba if (ret < 0) { 4215907d2710SDavid Sterba err = ret; 4216907d2710SDavid Sterba goto out_end; 4217907d2710SDavid Sterba } 4218907d2710SDavid Sterba 421929cbcf40SJosef Bacik rc->extent_root = btrfs_extent_root(fs_info, 0); 42205d4f98a2SYan Zheng 42215d4f98a2SYan Zheng set_reloc_control(rc); 42225d4f98a2SYan Zheng 42237a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 42243612b495STsutomu Itoh if (IS_ERR(trans)) { 42253612b495STsutomu Itoh err = PTR_ERR(trans); 4226fb2d83eeSJosef Bacik goto out_unset; 42273612b495STsutomu Itoh } 42283fd0a558SYan, Zheng 42293fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 42303fd0a558SYan, Zheng 42315d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 42325d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 42335d4f98a2SYan Zheng struct btrfs_root, root_list); 42345d4f98a2SYan Zheng list_del(&reloc_root->root_list); 42355d4f98a2SYan Zheng 42365d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 42375d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 42385d4f98a2SYan Zheng &rc->reloc_roots); 42395d4f98a2SYan Zheng continue; 42405d4f98a2SYan Zheng } 42415d4f98a2SYan Zheng 4242a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 4243a820feb5SDavid Sterba false); 424479787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 424579787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4246ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42471402d17dSXiyu Yang btrfs_end_transaction(trans); 4248fb2d83eeSJosef Bacik goto out_unset; 424979787eaaSJeff Mahoney } 42505d4f98a2SYan Zheng 4251ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 425257a304cfSJosef Bacik ASSERT(err != -EEXIST); 42533c925863SJosef Bacik if (err) { 42543c925863SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42553c925863SJosef Bacik btrfs_put_root(fs_root); 42563c925863SJosef Bacik btrfs_end_transaction(trans); 42573c925863SJosef Bacik goto out_unset; 42583c925863SJosef Bacik } 4259f44deb74SJosef Bacik fs_root->reloc_root = btrfs_grab_root(reloc_root); 426000246528SJosef Bacik btrfs_put_root(fs_root); 42615d4f98a2SYan Zheng } 42625d4f98a2SYan Zheng 42633a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 426479787eaaSJeff Mahoney if (err) 4265fb2d83eeSJosef Bacik goto out_unset; 42665d4f98a2SYan Zheng 42675d4f98a2SYan Zheng merge_reloc_roots(rc); 42685d4f98a2SYan Zheng 42695d4f98a2SYan Zheng unset_reloc_control(rc); 42705d4f98a2SYan Zheng 42717a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 427262b99540SQu Wenruo if (IS_ERR(trans)) { 42733612b495STsutomu Itoh err = PTR_ERR(trans); 42746217b0faSJosef Bacik goto out_clean; 427562b99540SQu Wenruo } 42763a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 42776217b0faSJosef Bacik out_clean: 4278d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4279d2311e69SQu Wenruo if (ret < 0 && !err) 4280d2311e69SQu Wenruo err = ret; 4281fb2d83eeSJosef Bacik out_unset: 4282fb2d83eeSJosef Bacik unset_reloc_control(rc); 4283907d2710SDavid Sterba out_end: 4284907d2710SDavid Sterba reloc_chunk_end(fs_info); 42851a0afa0eSJosef Bacik free_reloc_control(rc); 42863612b495STsutomu Itoh out: 4287aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4288aca1bba6SLiu Bo 42895d4f98a2SYan Zheng btrfs_free_path(path); 42905d4f98a2SYan Zheng 42915d4f98a2SYan Zheng if (err == 0) { 42925d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 4293aeb935a4SQu Wenruo fs_root = btrfs_grab_root(fs_info->data_reloc_root); 4294aeb935a4SQu Wenruo ASSERT(fs_root); 429566b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 429600246528SJosef Bacik btrfs_put_root(fs_root); 4297932fd26dSJosef Bacik } 42985d4f98a2SYan Zheng return err; 42995d4f98a2SYan Zheng } 43005d4f98a2SYan Zheng 43015d4f98a2SYan Zheng /* 43025d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 43035d4f98a2SYan Zheng * 43045d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 43055d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 43065d4f98a2SYan Zheng */ 43077bfa9535SNikolay Borisov int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len) 43085d4f98a2SYan Zheng { 43097bfa9535SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 4310fc28b25eSJosef Bacik struct btrfs_root *csum_root; 43115d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 43125d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 43135d4f98a2SYan Zheng int ret; 43145d4f98a2SYan Zheng u64 disk_bytenr; 43154577b014SJosef Bacik u64 new_bytenr; 43165d4f98a2SYan Zheng LIST_HEAD(list); 43175d4f98a2SYan Zheng 43187bfa9535SNikolay Borisov ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4319bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 43205d4f98a2SYan Zheng 43217bfa9535SNikolay Borisov disk_bytenr = file_pos + inode->index_cnt; 4322fc28b25eSJosef Bacik csum_root = btrfs_csum_root(fs_info, disk_bytenr); 4323fc28b25eSJosef Bacik ret = btrfs_lookup_csums_range(csum_root, disk_bytenr, 4324a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 432579787eaaSJeff Mahoney if (ret) 432679787eaaSJeff Mahoney goto out; 43275d4f98a2SYan Zheng 43285d4f98a2SYan Zheng while (!list_empty(&list)) { 43295d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 43305d4f98a2SYan Zheng list_del_init(&sums->list); 43315d4f98a2SYan Zheng 43324577b014SJosef Bacik /* 43334577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 43344577b014SJosef Bacik * We need to do this because we will read in entire prealloc 43354577b014SJosef Bacik * extents but we may have written to say the middle of the 43364577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 43374577b014SJosef Bacik * the right disk offset. 43384577b014SJosef Bacik * 43394577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 43404577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 43414577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 43424577b014SJosef Bacik * disk length. 43434577b014SJosef Bacik */ 4344bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 43454577b014SJosef Bacik sums->bytenr = new_bytenr; 43465d4f98a2SYan Zheng 4347f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 43485d4f98a2SYan Zheng } 434979787eaaSJeff Mahoney out: 43505d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4351411fc6bcSAndi Kleen return ret; 43525d4f98a2SYan Zheng } 43533fd0a558SYan, Zheng 435483d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 43553fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 43563fd0a558SYan, Zheng struct extent_buffer *cow) 43573fd0a558SYan, Zheng { 43580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 43593fd0a558SYan, Zheng struct reloc_control *rc; 4360a26195a5SQu Wenruo struct btrfs_backref_node *node; 43613fd0a558SYan, Zheng int first_cow = 0; 43623fd0a558SYan, Zheng int level; 436383d4cfd4SJosef Bacik int ret = 0; 43643fd0a558SYan, Zheng 43650b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 43663fd0a558SYan, Zheng if (!rc) 436783d4cfd4SJosef Bacik return 0; 43683fd0a558SYan, Zheng 436937f00a6dSJohannes Thumshirn BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root)); 43703fd0a558SYan, Zheng 43713fd0a558SYan, Zheng level = btrfs_header_level(buf); 43723fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 43733fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 43743fd0a558SYan, Zheng first_cow = 1; 43753fd0a558SYan, Zheng 43763fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 43773fd0a558SYan, Zheng rc->create_reloc_tree) { 43783fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 43793fd0a558SYan, Zheng 43803fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 43813fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 43823fd0a558SYan, Zheng node->new_bytenr != buf->start); 43833fd0a558SYan, Zheng 4384b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 438567439dadSDavid Sterba atomic_inc(&cow->refs); 43863fd0a558SYan, Zheng node->eb = cow; 43873fd0a558SYan, Zheng node->new_bytenr = cow->start; 43883fd0a558SYan, Zheng 43893fd0a558SYan, Zheng if (!node->pending) { 43903fd0a558SYan, Zheng list_move_tail(&node->list, 43913fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 43923fd0a558SYan, Zheng node->pending = 1; 43933fd0a558SYan, Zheng } 43943fd0a558SYan, Zheng 43953fd0a558SYan, Zheng if (first_cow) 43969569cc20SQu Wenruo mark_block_processed(rc, node); 43973fd0a558SYan, Zheng 43983fd0a558SYan, Zheng if (first_cow && level > 0) 43993fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 44003fd0a558SYan, Zheng } 44013fd0a558SYan, Zheng 440283d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 44033fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 440483d4cfd4SJosef Bacik return ret; 44053fd0a558SYan, Zheng } 44063fd0a558SYan, Zheng 44073fd0a558SYan, Zheng /* 44083fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 440901327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 44103fd0a558SYan, Zheng */ 4411147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 44123fd0a558SYan, Zheng u64 *bytes_to_reserve) 44133fd0a558SYan, Zheng { 441410995c04SQu Wenruo struct btrfs_root *root = pending->root; 441510995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44163fd0a558SYan, Zheng 44176282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 44183fd0a558SYan, Zheng return; 44193fd0a558SYan, Zheng 44203fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 44213fd0a558SYan, Zheng return; 44223fd0a558SYan, Zheng 44233fd0a558SYan, Zheng root = root->reloc_root; 44243fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 44253fd0a558SYan, Zheng /* 44263fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 44273fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 44283fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 44293fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 44303fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 44313fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 44323fd0a558SYan, Zheng * tree may use more space than it frees. so we need 44333fd0a558SYan, Zheng * reserve extra space. 44343fd0a558SYan, Zheng */ 44353fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 44363fd0a558SYan, Zheng } 44373fd0a558SYan, Zheng 44383fd0a558SYan, Zheng /* 44393fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 44403fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 4441f44deb74SJosef Bacik * 4442f44deb74SJosef Bacik * This is similar to btrfs_init_reloc_root(), we come out of here with two 4443f44deb74SJosef Bacik * references held on the reloc_root, one for root->reloc_root and one for 4444f44deb74SJosef Bacik * rc->reloc_roots. 44453fd0a558SYan, Zheng */ 444649b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 44473fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 44483fd0a558SYan, Zheng { 44493fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 44503fd0a558SYan, Zheng struct btrfs_root *reloc_root; 44513fd0a558SYan, Zheng struct btrfs_root *new_root; 445210995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44533fd0a558SYan, Zheng int ret; 44543fd0a558SYan, Zheng 44556282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 445649b25e05SJeff Mahoney return 0; 44573fd0a558SYan, Zheng 44583fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 44593fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 44603fd0a558SYan, Zheng 44613fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 44623fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 44633fd0a558SYan, Zheng rc->block_rsv, 44643a584174SLu Fengqi rc->nodes_relocated, true); 446549b25e05SJeff Mahoney if (ret) 446649b25e05SJeff Mahoney return ret; 44673fd0a558SYan, Zheng } 44683fd0a558SYan, Zheng 44693fd0a558SYan, Zheng new_root = pending->snap; 44703fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 44713fd0a558SYan, Zheng new_root->root_key.objectid); 447249b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 447349b25e05SJeff Mahoney return PTR_ERR(reloc_root); 44743fd0a558SYan, Zheng 4475ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 447657a304cfSJosef Bacik ASSERT(ret != -EEXIST); 44773c925863SJosef Bacik if (ret) { 44783c925863SJosef Bacik /* Pairs with create_reloc_root */ 44793c925863SJosef Bacik btrfs_put_root(reloc_root); 44803c925863SJosef Bacik return ret; 44813c925863SJosef Bacik } 4482f44deb74SJosef Bacik new_root->reloc_root = btrfs_grab_root(reloc_root); 44833fd0a558SYan, Zheng 448449b25e05SJeff Mahoney if (rc->create_reloc_tree) 44853fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 448649b25e05SJeff Mahoney return ret; 44873fd0a558SYan, Zheng } 4488