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" 295d4f98a2SYan Zheng 305d4f98a2SYan Zheng /* 310c891389SQu Wenruo * Relocation overview 320c891389SQu Wenruo * 330c891389SQu Wenruo * [What does relocation do] 340c891389SQu Wenruo * 350c891389SQu Wenruo * The objective of relocation is to relocate all extents of the target block 360c891389SQu Wenruo * group to other block groups. 370c891389SQu Wenruo * This is utilized by resize (shrink only), profile converting, compacting 380c891389SQu Wenruo * space, or balance routine to spread chunks over devices. 390c891389SQu Wenruo * 400c891389SQu Wenruo * Before | After 410c891389SQu Wenruo * ------------------------------------------------------------------ 420c891389SQu Wenruo * BG A: 10 data extents | BG A: deleted 430c891389SQu Wenruo * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) 440c891389SQu Wenruo * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) 450c891389SQu Wenruo * 460c891389SQu Wenruo * [How does relocation work] 470c891389SQu Wenruo * 480c891389SQu Wenruo * 1. Mark the target block group read-only 490c891389SQu Wenruo * New extents won't be allocated from the target block group. 500c891389SQu Wenruo * 510c891389SQu Wenruo * 2.1 Record each extent in the target block group 520c891389SQu Wenruo * To build a proper map of extents to be relocated. 530c891389SQu Wenruo * 540c891389SQu Wenruo * 2.2 Build data reloc tree and reloc trees 550c891389SQu Wenruo * Data reloc tree will contain an inode, recording all newly relocated 560c891389SQu Wenruo * data extents. 570c891389SQu Wenruo * There will be only one data reloc tree for one data block group. 580c891389SQu Wenruo * 590c891389SQu Wenruo * Reloc tree will be a special snapshot of its source tree, containing 600c891389SQu Wenruo * relocated tree blocks. 610c891389SQu Wenruo * Each tree referring to a tree block in target block group will get its 620c891389SQu Wenruo * reloc tree built. 630c891389SQu Wenruo * 640c891389SQu Wenruo * 2.3 Swap source tree with its corresponding reloc tree 650c891389SQu Wenruo * Each involved tree only refers to new extents after swap. 660c891389SQu Wenruo * 670c891389SQu Wenruo * 3. Cleanup reloc trees and data reloc tree. 680c891389SQu Wenruo * As old extents in the target block group are still referenced by reloc 690c891389SQu Wenruo * trees, we need to clean them up before really freeing the target block 700c891389SQu Wenruo * group. 710c891389SQu Wenruo * 720c891389SQu Wenruo * The main complexity is in steps 2.2 and 2.3. 730c891389SQu Wenruo * 740c891389SQu Wenruo * The entry point of relocation is relocate_block_group() function. 750c891389SQu Wenruo */ 760c891389SQu Wenruo 770647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 782a979612SQu Wenruo /* 795d4f98a2SYan Zheng * map address of tree root to tree 805d4f98a2SYan Zheng */ 815d4f98a2SYan Zheng struct mapping_node { 82e9a28dc5SQu Wenruo struct { 835d4f98a2SYan Zheng struct rb_node rb_node; 845d4f98a2SYan Zheng u64 bytenr; 85e9a28dc5SQu Wenruo }; /* Use rb_simle_node for search/insert */ 865d4f98a2SYan Zheng void *data; 875d4f98a2SYan Zheng }; 885d4f98a2SYan Zheng 895d4f98a2SYan Zheng struct mapping_tree { 905d4f98a2SYan Zheng struct rb_root rb_root; 915d4f98a2SYan Zheng spinlock_t lock; 925d4f98a2SYan Zheng }; 935d4f98a2SYan Zheng 945d4f98a2SYan Zheng /* 955d4f98a2SYan Zheng * present a tree block to process 965d4f98a2SYan Zheng */ 975d4f98a2SYan Zheng struct tree_block { 98e9a28dc5SQu Wenruo struct { 995d4f98a2SYan Zheng struct rb_node rb_node; 1005d4f98a2SYan Zheng u64 bytenr; 101e9a28dc5SQu Wenruo }; /* Use rb_simple_node for search/insert */ 102f7ba2d37SJosef Bacik u64 owner; 1035d4f98a2SYan Zheng struct btrfs_key key; 1045d4f98a2SYan Zheng unsigned int level:8; 1055d4f98a2SYan Zheng unsigned int key_ready:1; 1065d4f98a2SYan Zheng }; 1075d4f98a2SYan Zheng 1080257bb82SYan, Zheng #define MAX_EXTENTS 128 1090257bb82SYan, Zheng 1100257bb82SYan, Zheng struct file_extent_cluster { 1110257bb82SYan, Zheng u64 start; 1120257bb82SYan, Zheng u64 end; 1130257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1140257bb82SYan, Zheng unsigned int nr; 1150257bb82SYan, Zheng }; 1160257bb82SYan, Zheng 1175d4f98a2SYan Zheng struct reloc_control { 1185d4f98a2SYan Zheng /* block group to relocate */ 11932da5386SDavid Sterba struct btrfs_block_group *block_group; 1205d4f98a2SYan Zheng /* extent tree */ 1215d4f98a2SYan Zheng struct btrfs_root *extent_root; 1225d4f98a2SYan Zheng /* inode for moving data */ 1235d4f98a2SYan Zheng struct inode *data_inode; 1243fd0a558SYan, Zheng 1253fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1263fd0a558SYan, Zheng 127a26195a5SQu Wenruo struct btrfs_backref_cache backref_cache; 1283fd0a558SYan, Zheng 1293fd0a558SYan, Zheng struct file_extent_cluster cluster; 1305d4f98a2SYan Zheng /* tree blocks have been processed */ 1315d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1325d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1335d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1345d4f98a2SYan Zheng /* list of reloc trees */ 1355d4f98a2SYan Zheng struct list_head reloc_roots; 136d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 137d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 1383fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1393fd0a558SYan, Zheng u64 merging_rsv_size; 1403fd0a558SYan, Zheng /* size of relocated tree nodes */ 1413fd0a558SYan, Zheng u64 nodes_relocated; 1420647bf56SWang Shilong /* reserved size for block group relocation*/ 1430647bf56SWang Shilong u64 reserved_bytes; 1443fd0a558SYan, Zheng 1455d4f98a2SYan Zheng u64 search_start; 1465d4f98a2SYan Zheng u64 extents_found; 1473fd0a558SYan, Zheng 1483fd0a558SYan, Zheng unsigned int stage:8; 1493fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1503fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1515d4f98a2SYan Zheng unsigned int found_file_extent:1; 1525d4f98a2SYan Zheng }; 1535d4f98a2SYan Zheng 1545d4f98a2SYan Zheng /* stages of data relocation */ 1555d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1565d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1575d4f98a2SYan Zheng 1589569cc20SQu Wenruo static void mark_block_processed(struct reloc_control *rc, 159a26195a5SQu Wenruo struct btrfs_backref_node *node) 1609569cc20SQu Wenruo { 1619569cc20SQu Wenruo u32 blocksize; 1629569cc20SQu Wenruo 1639569cc20SQu Wenruo if (node->level == 0 || 1649569cc20SQu Wenruo in_range(node->bytenr, rc->block_group->start, 1659569cc20SQu Wenruo rc->block_group->length)) { 1669569cc20SQu Wenruo blocksize = rc->extent_root->fs_info->nodesize; 1679569cc20SQu Wenruo set_extent_bits(&rc->processed_blocks, node->bytenr, 1689569cc20SQu Wenruo node->bytenr + blocksize - 1, EXTENT_DIRTY); 1699569cc20SQu Wenruo } 1709569cc20SQu Wenruo node->processed = 1; 1719569cc20SQu Wenruo } 1729569cc20SQu Wenruo 1735d4f98a2SYan Zheng 1745d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1755d4f98a2SYan Zheng { 1766bef4d31SEric Paris tree->rb_root = RB_ROOT; 1775d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1785d4f98a2SYan Zheng } 1795d4f98a2SYan Zheng 1805d4f98a2SYan Zheng /* 1815d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 1825d4f98a2SYan Zheng */ 183a26195a5SQu Wenruo static struct btrfs_backref_node *walk_up_backref( 184a26195a5SQu Wenruo struct btrfs_backref_node *node, 185a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 1865d4f98a2SYan Zheng { 187a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 1885d4f98a2SYan Zheng int idx = *index; 1895d4f98a2SYan Zheng 1905d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 1915d4f98a2SYan Zheng edge = list_entry(node->upper.next, 192a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 1935d4f98a2SYan Zheng edges[idx++] = edge; 1945d4f98a2SYan Zheng node = edge->node[UPPER]; 1955d4f98a2SYan Zheng } 1963fd0a558SYan, Zheng BUG_ON(node->detached); 1975d4f98a2SYan Zheng *index = idx; 1985d4f98a2SYan Zheng return node; 1995d4f98a2SYan Zheng } 2005d4f98a2SYan Zheng 2015d4f98a2SYan Zheng /* 2025d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 2035d4f98a2SYan Zheng */ 204a26195a5SQu Wenruo static struct btrfs_backref_node *walk_down_backref( 205a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 2065d4f98a2SYan Zheng { 207a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 208a26195a5SQu Wenruo struct btrfs_backref_node *lower; 2095d4f98a2SYan Zheng int idx = *index; 2105d4f98a2SYan Zheng 2115d4f98a2SYan Zheng while (idx > 0) { 2125d4f98a2SYan Zheng edge = edges[idx - 1]; 2135d4f98a2SYan Zheng lower = edge->node[LOWER]; 2145d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 2155d4f98a2SYan Zheng idx--; 2165d4f98a2SYan Zheng continue; 2175d4f98a2SYan Zheng } 2185d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 219a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 2205d4f98a2SYan Zheng edges[idx - 1] = edge; 2215d4f98a2SYan Zheng *index = idx; 2225d4f98a2SYan Zheng return edge->node[UPPER]; 2235d4f98a2SYan Zheng } 2245d4f98a2SYan Zheng *index = 0; 2255d4f98a2SYan Zheng return NULL; 2265d4f98a2SYan Zheng } 2275d4f98a2SYan Zheng 228a26195a5SQu Wenruo static void update_backref_node(struct btrfs_backref_cache *cache, 229a26195a5SQu Wenruo struct btrfs_backref_node *node, u64 bytenr) 2303fd0a558SYan, Zheng { 2313fd0a558SYan, Zheng struct rb_node *rb_node; 2323fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 2333fd0a558SYan, Zheng node->bytenr = bytenr; 234e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node); 23543c04fb1SJeff Mahoney if (rb_node) 236982c92cbSQu Wenruo btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST); 2373fd0a558SYan, Zheng } 2383fd0a558SYan, Zheng 2393fd0a558SYan, Zheng /* 2403fd0a558SYan, Zheng * update backref cache after a transaction commit 2413fd0a558SYan, Zheng */ 2423fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 243a26195a5SQu Wenruo struct btrfs_backref_cache *cache) 2443fd0a558SYan, Zheng { 245a26195a5SQu Wenruo struct btrfs_backref_node *node; 2463fd0a558SYan, Zheng int level = 0; 2473fd0a558SYan, Zheng 2483fd0a558SYan, Zheng if (cache->last_trans == 0) { 2493fd0a558SYan, Zheng cache->last_trans = trans->transid; 2503fd0a558SYan, Zheng return 0; 2513fd0a558SYan, Zheng } 2523fd0a558SYan, Zheng 2533fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 2543fd0a558SYan, Zheng return 0; 2553fd0a558SYan, Zheng 2563fd0a558SYan, Zheng /* 2573fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 2583fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 2593fd0a558SYan, Zheng * so the detached nodes are no longer useful. 2603fd0a558SYan, Zheng */ 2613fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2623fd0a558SYan, Zheng node = list_entry(cache->detached.next, 263a26195a5SQu Wenruo struct btrfs_backref_node, list); 264023acb07SQu Wenruo btrfs_backref_cleanup_node(cache, node); 2653fd0a558SYan, Zheng } 2663fd0a558SYan, Zheng 2673fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 2683fd0a558SYan, Zheng node = list_entry(cache->changed.next, 269a26195a5SQu Wenruo struct btrfs_backref_node, list); 2703fd0a558SYan, Zheng list_del_init(&node->list); 2713fd0a558SYan, Zheng BUG_ON(node->pending); 2723fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2733fd0a558SYan, Zheng } 2743fd0a558SYan, Zheng 2753fd0a558SYan, Zheng /* 2763fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 2773fd0a558SYan, Zheng * errors during processing the pending nodes. 2783fd0a558SYan, Zheng */ 2793fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 2803fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 2813fd0a558SYan, Zheng BUG_ON(!node->pending); 2823fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 2833fd0a558SYan, Zheng continue; 2843fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2853fd0a558SYan, Zheng } 2863fd0a558SYan, Zheng } 2873fd0a558SYan, Zheng 2883fd0a558SYan, Zheng cache->last_trans = 0; 2893fd0a558SYan, Zheng return 1; 2903fd0a558SYan, Zheng } 2913fd0a558SYan, Zheng 2926282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 2936282675eSQu Wenruo { 2946282675eSQu Wenruo /* 2956282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 2966282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 2976282675eSQu Wenruo * trying to access reloc_root 2986282675eSQu Wenruo */ 2996282675eSQu Wenruo smp_rmb(); 3006282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 3016282675eSQu Wenruo return true; 3026282675eSQu Wenruo return false; 3036282675eSQu Wenruo } 3046282675eSQu Wenruo 3056282675eSQu Wenruo /* 3066282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 3076282675eSQu Wenruo * 3086282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 3096282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 31055465730SQu Wenruo * from no reloc root. But btrfs_should_ignore_reloc_root() below is a 31155465730SQu Wenruo * special case. 3126282675eSQu Wenruo */ 3136282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 3146282675eSQu Wenruo { 3156282675eSQu Wenruo if (reloc_root_is_dead(root)) 3166282675eSQu Wenruo return false; 3176282675eSQu Wenruo if (!root->reloc_root) 3186282675eSQu Wenruo return false; 3196282675eSQu Wenruo return true; 3206282675eSQu Wenruo } 321f2a97a9dSDavid Sterba 32255465730SQu Wenruo int btrfs_should_ignore_reloc_root(struct btrfs_root *root) 3233fd0a558SYan, Zheng { 3243fd0a558SYan, Zheng struct btrfs_root *reloc_root; 3253fd0a558SYan, Zheng 32692a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 3273fd0a558SYan, Zheng return 0; 3283fd0a558SYan, Zheng 3296282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 3306282675eSQu Wenruo if (reloc_root_is_dead(root)) 3316282675eSQu Wenruo return 1; 3326282675eSQu Wenruo 3333fd0a558SYan, Zheng reloc_root = root->reloc_root; 3343fd0a558SYan, Zheng if (!reloc_root) 3353fd0a558SYan, Zheng return 0; 3363fd0a558SYan, Zheng 3374d4225fcSJosef Bacik if (btrfs_header_generation(reloc_root->commit_root) == 3384d4225fcSJosef Bacik root->fs_info->running_transaction->transid) 3393fd0a558SYan, Zheng return 0; 3403fd0a558SYan, Zheng /* 3413fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 3423fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 3433fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 3443fd0a558SYan, Zheng * relocation. 3453fd0a558SYan, Zheng */ 3463fd0a558SYan, Zheng return 1; 3473fd0a558SYan, Zheng } 34855465730SQu Wenruo 3495d4f98a2SYan Zheng /* 3505d4f98a2SYan Zheng * find reloc tree by address of tree root 3515d4f98a2SYan Zheng */ 3522433bea5SQu Wenruo struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) 3535d4f98a2SYan Zheng { 3542433bea5SQu Wenruo struct reloc_control *rc = fs_info->reloc_ctl; 3555d4f98a2SYan Zheng struct rb_node *rb_node; 3565d4f98a2SYan Zheng struct mapping_node *node; 3575d4f98a2SYan Zheng struct btrfs_root *root = NULL; 3585d4f98a2SYan Zheng 3592433bea5SQu Wenruo ASSERT(rc); 3605d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 361e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr); 3625d4f98a2SYan Zheng if (rb_node) { 3635d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 3645d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 3655d4f98a2SYan Zheng } 3665d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 36700246528SJosef Bacik return btrfs_grab_root(root); 3685d4f98a2SYan Zheng } 3695d4f98a2SYan Zheng 3705d4f98a2SYan Zheng /* 37129db137bSQu Wenruo * For useless nodes, do two major clean ups: 37229db137bSQu Wenruo * 37329db137bSQu Wenruo * - Cleanup the children edges and nodes 37429db137bSQu Wenruo * If child node is also orphan (no parent) during cleanup, then the child 37529db137bSQu Wenruo * node will also be cleaned up. 37629db137bSQu Wenruo * 37729db137bSQu Wenruo * - Freeing up leaves (level 0), keeps nodes detached 37829db137bSQu Wenruo * For nodes, the node is still cached as "detached" 37929db137bSQu Wenruo * 38029db137bSQu Wenruo * Return false if @node is not in the @useless_nodes list. 38129db137bSQu Wenruo * Return true if @node is in the @useless_nodes list. 38229db137bSQu Wenruo */ 38329db137bSQu Wenruo static bool handle_useless_nodes(struct reloc_control *rc, 384a26195a5SQu Wenruo struct btrfs_backref_node *node) 38529db137bSQu Wenruo { 386a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 38729db137bSQu Wenruo struct list_head *useless_node = &cache->useless_node; 38829db137bSQu Wenruo bool ret = false; 38929db137bSQu Wenruo 39029db137bSQu Wenruo while (!list_empty(useless_node)) { 391a26195a5SQu Wenruo struct btrfs_backref_node *cur; 39229db137bSQu Wenruo 393a26195a5SQu Wenruo cur = list_first_entry(useless_node, struct btrfs_backref_node, 39429db137bSQu Wenruo list); 39529db137bSQu Wenruo list_del_init(&cur->list); 39629db137bSQu Wenruo 39729db137bSQu Wenruo /* Only tree root nodes can be added to @useless_nodes */ 39829db137bSQu Wenruo ASSERT(list_empty(&cur->upper)); 39929db137bSQu Wenruo 40029db137bSQu Wenruo if (cur == node) 40129db137bSQu Wenruo ret = true; 40229db137bSQu Wenruo 40329db137bSQu Wenruo /* The node is the lowest node */ 40429db137bSQu Wenruo if (cur->lowest) { 40529db137bSQu Wenruo list_del_init(&cur->lower); 40629db137bSQu Wenruo cur->lowest = 0; 40729db137bSQu Wenruo } 40829db137bSQu Wenruo 40929db137bSQu Wenruo /* Cleanup the lower edges */ 41029db137bSQu Wenruo while (!list_empty(&cur->lower)) { 411a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 412a26195a5SQu Wenruo struct btrfs_backref_node *lower; 41329db137bSQu Wenruo 41429db137bSQu Wenruo edge = list_entry(cur->lower.next, 415a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 41629db137bSQu Wenruo list_del(&edge->list[UPPER]); 41729db137bSQu Wenruo list_del(&edge->list[LOWER]); 41829db137bSQu Wenruo lower = edge->node[LOWER]; 419741188d3SQu Wenruo btrfs_backref_free_edge(cache, edge); 42029db137bSQu Wenruo 42129db137bSQu Wenruo /* Child node is also orphan, queue for cleanup */ 42229db137bSQu Wenruo if (list_empty(&lower->upper)) 42329db137bSQu Wenruo list_add(&lower->list, useless_node); 42429db137bSQu Wenruo } 42529db137bSQu Wenruo /* Mark this block processed for relocation */ 42629db137bSQu Wenruo mark_block_processed(rc, cur); 42729db137bSQu Wenruo 42829db137bSQu Wenruo /* 42929db137bSQu Wenruo * Backref nodes for tree leaves are deleted from the cache. 43029db137bSQu Wenruo * Backref nodes for upper level tree blocks are left in the 43129db137bSQu Wenruo * cache to avoid unnecessary backref lookup. 43229db137bSQu Wenruo */ 43329db137bSQu Wenruo if (cur->level > 0) { 43429db137bSQu Wenruo list_add(&cur->list, &cache->detached); 43529db137bSQu Wenruo cur->detached = 1; 43629db137bSQu Wenruo } else { 43729db137bSQu Wenruo rb_erase(&cur->rb_node, &cache->rb_root); 438741188d3SQu Wenruo btrfs_backref_free_node(cache, cur); 43929db137bSQu Wenruo } 44029db137bSQu Wenruo } 44129db137bSQu Wenruo return ret; 44229db137bSQu Wenruo } 44329db137bSQu Wenruo 44429db137bSQu Wenruo /* 445e7d571c7SQu Wenruo * Build backref tree for a given tree block. Root of the backref tree 446e7d571c7SQu Wenruo * corresponds the tree block, leaves of the backref tree correspond roots of 447e7d571c7SQu Wenruo * b-trees that reference the tree block. 448e7d571c7SQu Wenruo * 449e7d571c7SQu Wenruo * The basic idea of this function is check backrefs of a given block to find 450e7d571c7SQu Wenruo * upper level blocks that reference the block, and then check backrefs of 451e7d571c7SQu Wenruo * these upper level blocks recursively. The recursion stops when tree root is 452e7d571c7SQu Wenruo * reached or backrefs for the block is cached. 453e7d571c7SQu Wenruo * 454e7d571c7SQu Wenruo * NOTE: if we find that backrefs for a block are cached, we know backrefs for 455e7d571c7SQu Wenruo * all upper level blocks that directly/indirectly reference the block are also 456e7d571c7SQu Wenruo * cached. 457e7d571c7SQu Wenruo */ 458a26195a5SQu Wenruo static noinline_for_stack struct btrfs_backref_node *build_backref_tree( 459e7d571c7SQu Wenruo struct reloc_control *rc, struct btrfs_key *node_key, 460e7d571c7SQu Wenruo int level, u64 bytenr) 461e7d571c7SQu Wenruo { 462e7d571c7SQu Wenruo struct btrfs_backref_iter *iter; 463a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 464e7d571c7SQu Wenruo /* For searching parent of TREE_BLOCK_REF */ 465e7d571c7SQu Wenruo struct btrfs_path *path; 466a26195a5SQu Wenruo struct btrfs_backref_node *cur; 467a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 468a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 469e7d571c7SQu Wenruo int ret; 470e7d571c7SQu Wenruo int err = 0; 471e7d571c7SQu Wenruo 472e7d571c7SQu Wenruo iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS); 473e7d571c7SQu Wenruo if (!iter) 474e7d571c7SQu Wenruo return ERR_PTR(-ENOMEM); 475e7d571c7SQu Wenruo path = btrfs_alloc_path(); 476e7d571c7SQu Wenruo if (!path) { 477e7d571c7SQu Wenruo err = -ENOMEM; 478e7d571c7SQu Wenruo goto out; 479e7d571c7SQu Wenruo } 480e7d571c7SQu Wenruo 481b1818dabSQu Wenruo node = btrfs_backref_alloc_node(cache, bytenr, level); 482e7d571c7SQu Wenruo if (!node) { 483e7d571c7SQu Wenruo err = -ENOMEM; 484e7d571c7SQu Wenruo goto out; 485e7d571c7SQu Wenruo } 486e7d571c7SQu Wenruo 487e7d571c7SQu Wenruo node->lowest = 1; 488e7d571c7SQu Wenruo cur = node; 489e7d571c7SQu Wenruo 490e7d571c7SQu Wenruo /* Breadth-first search to build backref cache */ 491e7d571c7SQu Wenruo do { 4921b60d2ecSQu Wenruo ret = btrfs_backref_add_tree_node(cache, path, iter, node_key, 4931b60d2ecSQu Wenruo cur); 494e7d571c7SQu Wenruo if (ret < 0) { 495e7d571c7SQu Wenruo err = ret; 496e7d571c7SQu Wenruo goto out; 497e7d571c7SQu Wenruo } 498e7d571c7SQu Wenruo edge = list_first_entry_or_null(&cache->pending_edge, 499a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 500e7d571c7SQu Wenruo /* 501e7d571c7SQu Wenruo * The pending list isn't empty, take the first block to 502e7d571c7SQu Wenruo * process 503e7d571c7SQu Wenruo */ 504e7d571c7SQu Wenruo if (edge) { 5055d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 5065d4f98a2SYan Zheng cur = edge->node[UPPER]; 5075d4f98a2SYan Zheng } 508e7d571c7SQu Wenruo } while (edge); 5095d4f98a2SYan Zheng 5101f872924SQu Wenruo /* Finish the upper linkage of newly added edges/nodes */ 511fc997ed0SQu Wenruo ret = btrfs_backref_finish_upper_links(cache, node); 5121f872924SQu Wenruo if (ret < 0) { 5131f872924SQu Wenruo err = ret; 51475bfb9afSJosef Bacik goto out; 51575bfb9afSJosef Bacik } 51675bfb9afSJosef Bacik 51729db137bSQu Wenruo if (handle_useless_nodes(rc, node)) 5183fd0a558SYan, Zheng node = NULL; 5195d4f98a2SYan Zheng out: 52071f572a9SQu Wenruo btrfs_backref_iter_free(iter); 52171f572a9SQu Wenruo btrfs_free_path(path); 5225d4f98a2SYan Zheng if (err) { 5231b23ea18SQu Wenruo btrfs_backref_error_cleanup(cache, node); 5245d4f98a2SYan Zheng return ERR_PTR(err); 5255d4f98a2SYan Zheng } 52675bfb9afSJosef Bacik ASSERT(!node || !node->detached); 52784780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 52884780289SQu Wenruo list_empty(&cache->pending_edge)); 5295d4f98a2SYan Zheng return node; 5305d4f98a2SYan Zheng } 5315d4f98a2SYan Zheng 5325d4f98a2SYan Zheng /* 5333fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 5343fd0a558SYan, Zheng * the backref node is created by cloning backref node that 5353fd0a558SYan, Zheng * corresponds to root of source tree 5363fd0a558SYan, Zheng */ 5373fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 5383fd0a558SYan, Zheng struct reloc_control *rc, 5393fd0a558SYan, Zheng struct btrfs_root *src, 5403fd0a558SYan, Zheng struct btrfs_root *dest) 5413fd0a558SYan, Zheng { 5423fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 543a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 544a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 545a26195a5SQu Wenruo struct btrfs_backref_node *new_node; 546a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 547a26195a5SQu Wenruo struct btrfs_backref_edge *new_edge; 5483fd0a558SYan, Zheng struct rb_node *rb_node; 5493fd0a558SYan, Zheng 5503fd0a558SYan, Zheng if (cache->last_trans > 0) 5513fd0a558SYan, Zheng update_backref_cache(trans, cache); 5523fd0a558SYan, Zheng 553e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start); 5543fd0a558SYan, Zheng if (rb_node) { 555a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, rb_node); 5563fd0a558SYan, Zheng if (node->detached) 5573fd0a558SYan, Zheng node = NULL; 5583fd0a558SYan, Zheng else 5593fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 5603fd0a558SYan, Zheng } 5613fd0a558SYan, Zheng 5623fd0a558SYan, Zheng if (!node) { 563e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, 5643fd0a558SYan, Zheng reloc_root->commit_root->start); 5653fd0a558SYan, Zheng if (rb_node) { 566a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, 5673fd0a558SYan, Zheng rb_node); 5683fd0a558SYan, Zheng BUG_ON(node->detached); 5693fd0a558SYan, Zheng } 5703fd0a558SYan, Zheng } 5713fd0a558SYan, Zheng 5723fd0a558SYan, Zheng if (!node) 5733fd0a558SYan, Zheng return 0; 5743fd0a558SYan, Zheng 575b1818dabSQu Wenruo new_node = btrfs_backref_alloc_node(cache, dest->node->start, 576b1818dabSQu Wenruo node->level); 5773fd0a558SYan, Zheng if (!new_node) 5783fd0a558SYan, Zheng return -ENOMEM; 5793fd0a558SYan, Zheng 5803fd0a558SYan, Zheng new_node->lowest = node->lowest; 5816848ad64SYan, Zheng new_node->checked = 1; 58200246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 5830b530bc5SJosef Bacik ASSERT(new_node->root); 5843fd0a558SYan, Zheng 5853fd0a558SYan, Zheng if (!node->lowest) { 5863fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 58747254d07SQu Wenruo new_edge = btrfs_backref_alloc_edge(cache); 5883fd0a558SYan, Zheng if (!new_edge) 5893fd0a558SYan, Zheng goto fail; 5903fd0a558SYan, Zheng 591f39911e5SQu Wenruo btrfs_backref_link_edge(new_edge, edge->node[LOWER], 592f39911e5SQu Wenruo new_node, LINK_UPPER); 5933fd0a558SYan, Zheng } 59476b9e23dSMiao Xie } else { 59576b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 5963fd0a558SYan, Zheng } 5973fd0a558SYan, Zheng 598e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr, 5993fd0a558SYan, Zheng &new_node->rb_node); 60043c04fb1SJeff Mahoney if (rb_node) 601982c92cbSQu Wenruo btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST); 6023fd0a558SYan, Zheng 6033fd0a558SYan, Zheng if (!new_node->lowest) { 6043fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 6053fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 6063fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 6073fd0a558SYan, Zheng } 6083fd0a558SYan, Zheng } 6093fd0a558SYan, Zheng return 0; 6103fd0a558SYan, Zheng fail: 6113fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 6123fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 613a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 6143fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 615741188d3SQu Wenruo btrfs_backref_free_edge(cache, new_edge); 6163fd0a558SYan, Zheng } 617741188d3SQu Wenruo btrfs_backref_free_node(cache, new_node); 6183fd0a558SYan, Zheng return -ENOMEM; 6193fd0a558SYan, Zheng } 6203fd0a558SYan, Zheng 6213fd0a558SYan, Zheng /* 6225d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 6235d4f98a2SYan Zheng */ 624ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 6255d4f98a2SYan Zheng { 6260b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6275d4f98a2SYan Zheng struct rb_node *rb_node; 6285d4f98a2SYan Zheng struct mapping_node *node; 6290b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 6305d4f98a2SYan Zheng 6315d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 632ffd7b339SJeff Mahoney if (!node) 633ffd7b339SJeff Mahoney return -ENOMEM; 6345d4f98a2SYan Zheng 635ea287ab1SJosef Bacik node->bytenr = root->commit_root->start; 6365d4f98a2SYan Zheng node->data = root; 6375d4f98a2SYan Zheng 6385d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 639e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 6405d4f98a2SYan Zheng node->bytenr, &node->rb_node); 6415d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 642ffd7b339SJeff Mahoney if (rb_node) { 64357a304cfSJosef Bacik btrfs_err(fs_info, 6445d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 6455d163e0eSJeff Mahoney node->bytenr); 64657a304cfSJosef Bacik return -EEXIST; 647ffd7b339SJeff Mahoney } 6485d4f98a2SYan Zheng 6495d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 6505d4f98a2SYan Zheng return 0; 6515d4f98a2SYan Zheng } 6525d4f98a2SYan Zheng 6535d4f98a2SYan Zheng /* 654c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 6555d4f98a2SYan Zheng * mapping 6565d4f98a2SYan Zheng */ 657c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 6585d4f98a2SYan Zheng { 6590b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6605d4f98a2SYan Zheng struct rb_node *rb_node; 6615d4f98a2SYan Zheng struct mapping_node *node = NULL; 6620b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 663f44deb74SJosef Bacik bool put_ref = false; 6645d4f98a2SYan Zheng 66565c6e82bSQu Wenruo if (rc && root->node) { 6665d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 667e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 668ea287ab1SJosef Bacik root->commit_root->start); 669c974c464SWang Shilong if (rb_node) { 670c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 671c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 672ea287ab1SJosef Bacik RB_CLEAR_NODE(&node->rb_node); 673c974c464SWang Shilong } 674c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 675c78a10aeSJosef Bacik ASSERT(!node || (struct btrfs_root *)node->data == root); 676389305b2SQu Wenruo } 677c974c464SWang Shilong 678f44deb74SJosef Bacik /* 679f44deb74SJosef Bacik * We only put the reloc root here if it's on the list. There's a lot 680f44deb74SJosef Bacik * of places where the pattern is to splice the rc->reloc_roots, process 681f44deb74SJosef Bacik * the reloc roots, and then add the reloc root back onto 682f44deb74SJosef Bacik * rc->reloc_roots. If we call __del_reloc_root while it's off of the 683f44deb74SJosef Bacik * list we don't want the reference being dropped, because the guy 684f44deb74SJosef Bacik * messing with the list is in charge of the reference. 685f44deb74SJosef Bacik */ 6860b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 687f44deb74SJosef Bacik if (!list_empty(&root->root_list)) { 688f44deb74SJosef Bacik put_ref = true; 689c974c464SWang Shilong list_del_init(&root->root_list); 690f44deb74SJosef Bacik } 6910b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 692f44deb74SJosef Bacik if (put_ref) 693f44deb74SJosef Bacik btrfs_put_root(root); 694c974c464SWang Shilong kfree(node); 695c974c464SWang Shilong } 696c974c464SWang Shilong 697c974c464SWang Shilong /* 698c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 699c974c464SWang Shilong * mapping 700c974c464SWang Shilong */ 701ea287ab1SJosef Bacik static int __update_reloc_root(struct btrfs_root *root) 702c974c464SWang Shilong { 7030b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 704c974c464SWang Shilong struct rb_node *rb_node; 705c974c464SWang Shilong struct mapping_node *node = NULL; 7060b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 707c974c464SWang Shilong 708c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 709e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 710ea287ab1SJosef Bacik root->commit_root->start); 7115d4f98a2SYan Zheng if (rb_node) { 7125d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 7135d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 7145d4f98a2SYan Zheng } 7155d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 7165d4f98a2SYan Zheng 7178f71f3e0SLiu Bo if (!node) 7188f71f3e0SLiu Bo return 0; 7195d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 7205d4f98a2SYan Zheng 7215d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 722ea287ab1SJosef Bacik node->bytenr = root->node->start; 723e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 7245d4f98a2SYan Zheng node->bytenr, &node->rb_node); 7255d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 72643c04fb1SJeff Mahoney if (rb_node) 727982c92cbSQu Wenruo btrfs_backref_panic(fs_info, node->bytenr, -EEXIST); 7285d4f98a2SYan Zheng return 0; 7295d4f98a2SYan Zheng } 7305d4f98a2SYan Zheng 7313fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 7323fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 7335d4f98a2SYan Zheng { 7340b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 7355d4f98a2SYan Zheng struct btrfs_root *reloc_root; 7365d4f98a2SYan Zheng struct extent_buffer *eb; 7375d4f98a2SYan Zheng struct btrfs_root_item *root_item; 7385d4f98a2SYan Zheng struct btrfs_key root_key; 73984c50ba5SJosef Bacik int ret = 0; 74084c50ba5SJosef Bacik bool must_abort = false; 7415d4f98a2SYan Zheng 7425d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 74384c50ba5SJosef Bacik if (!root_item) 74484c50ba5SJosef Bacik return ERR_PTR(-ENOMEM); 7455d4f98a2SYan Zheng 7465d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 7475d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 7483fd0a558SYan, Zheng root_key.offset = objectid; 7495d4f98a2SYan Zheng 7503fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 751054570a1SFilipe Manana u64 commit_root_gen; 752054570a1SFilipe Manana 7533fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 7545d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 7555d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 75684c50ba5SJosef Bacik if (ret) 75784c50ba5SJosef Bacik goto fail; 75884c50ba5SJosef Bacik 759054570a1SFilipe Manana /* 760054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 761054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 762054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 763054570a1SFilipe Manana * either inside the critical section of a transaction commit 764054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 765054570a1SFilipe Manana * it's created before the transaction commit is started. 766054570a1SFilipe Manana */ 767054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 768054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 7693fd0a558SYan, Zheng } else { 7703fd0a558SYan, Zheng /* 7713fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 7723fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 7733fd0a558SYan, Zheng * modified after it was created have RELOC flag 7743fd0a558SYan, Zheng * set in their headers. so it's OK to not update 7753fd0a558SYan, Zheng * the 'last_snapshot'. 7763fd0a558SYan, Zheng */ 7773fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 7783fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 77984c50ba5SJosef Bacik if (ret) 78084c50ba5SJosef Bacik goto fail; 7813fd0a558SYan, Zheng } 7823fd0a558SYan, Zheng 78384c50ba5SJosef Bacik /* 78484c50ba5SJosef Bacik * We have changed references at this point, we must abort the 78584c50ba5SJosef Bacik * transaction if anything fails. 78684c50ba5SJosef Bacik */ 78784c50ba5SJosef Bacik must_abort = true; 78884c50ba5SJosef Bacik 7895d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 7905d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 7915d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 7925d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 7933fd0a558SYan, Zheng 7943fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 7953fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 7963fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 7973fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 798c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, 0); 7993fd0a558SYan, Zheng } 8005d4f98a2SYan Zheng 8015d4f98a2SYan Zheng btrfs_tree_unlock(eb); 8025d4f98a2SYan Zheng free_extent_buffer(eb); 8035d4f98a2SYan Zheng 8040b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 8055d4f98a2SYan Zheng &root_key, root_item); 80684c50ba5SJosef Bacik if (ret) 80784c50ba5SJosef Bacik goto fail; 80884c50ba5SJosef Bacik 8095d4f98a2SYan Zheng kfree(root_item); 8105d4f98a2SYan Zheng 8113dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 81284c50ba5SJosef Bacik if (IS_ERR(reloc_root)) { 81384c50ba5SJosef Bacik ret = PTR_ERR(reloc_root); 81484c50ba5SJosef Bacik goto abort; 81584c50ba5SJosef Bacik } 81692a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 8175d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 8183fd0a558SYan, Zheng return reloc_root; 81984c50ba5SJosef Bacik fail: 82084c50ba5SJosef Bacik kfree(root_item); 82184c50ba5SJosef Bacik abort: 82284c50ba5SJosef Bacik if (must_abort) 82384c50ba5SJosef Bacik btrfs_abort_transaction(trans, ret); 82484c50ba5SJosef Bacik return ERR_PTR(ret); 8253fd0a558SYan, Zheng } 8263fd0a558SYan, Zheng 8273fd0a558SYan, Zheng /* 8283fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 8293fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 830f44deb74SJosef Bacik * 831f44deb74SJosef Bacik * The reloc_root comes out of here with two references, one for 832f44deb74SJosef Bacik * root->reloc_root, and another for being on the rc->reloc_roots list. 8333fd0a558SYan, Zheng */ 8343fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 8353fd0a558SYan, Zheng struct btrfs_root *root) 8363fd0a558SYan, Zheng { 8370b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8383fd0a558SYan, Zheng struct btrfs_root *reloc_root; 8390b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 84020dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 8413fd0a558SYan, Zheng int clear_rsv = 0; 842ffd7b339SJeff Mahoney int ret; 8433fd0a558SYan, Zheng 844aec7db3bSJosef Bacik if (!rc) 8452abc726aSJosef Bacik return 0; 8462abc726aSJosef Bacik 8471fac4a54SQu Wenruo /* 8481fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 8491fac4a54SQu Wenruo * create/update the dead reloc tree 8501fac4a54SQu Wenruo */ 8516282675eSQu Wenruo if (reloc_root_is_dead(root)) 8521fac4a54SQu Wenruo return 0; 8531fac4a54SQu Wenruo 854aec7db3bSJosef Bacik /* 855aec7db3bSJosef Bacik * This is subtle but important. We do not do 856aec7db3bSJosef Bacik * record_root_in_transaction for reloc roots, instead we record their 857aec7db3bSJosef Bacik * corresponding fs root, and then here we update the last trans for the 858aec7db3bSJosef Bacik * reloc root. This means that we have to do this for the entire life 859aec7db3bSJosef Bacik * of the reloc root, regardless of which stage of the relocation we are 860aec7db3bSJosef Bacik * in. 861aec7db3bSJosef Bacik */ 8623fd0a558SYan, Zheng if (root->reloc_root) { 8633fd0a558SYan, Zheng reloc_root = root->reloc_root; 8643fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 8653fd0a558SYan, Zheng return 0; 8663fd0a558SYan, Zheng } 8673fd0a558SYan, Zheng 868aec7db3bSJosef Bacik /* 869aec7db3bSJosef Bacik * We are merging reloc roots, we do not need new reloc trees. Also 870aec7db3bSJosef Bacik * reloc trees never need their own reloc tree. 871aec7db3bSJosef Bacik */ 872aec7db3bSJosef Bacik if (!rc->create_reloc_tree || 873aec7db3bSJosef Bacik root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 874aec7db3bSJosef Bacik return 0; 875aec7db3bSJosef Bacik 87620dd2cbfSMiao Xie if (!trans->reloc_reserved) { 87720dd2cbfSMiao Xie rsv = trans->block_rsv; 8783fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 8793fd0a558SYan, Zheng clear_rsv = 1; 8803fd0a558SYan, Zheng } 8813fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 8823fd0a558SYan, Zheng if (clear_rsv) 88320dd2cbfSMiao Xie trans->block_rsv = rsv; 88400bb36a0SJosef Bacik if (IS_ERR(reloc_root)) 88500bb36a0SJosef Bacik return PTR_ERR(reloc_root); 8865d4f98a2SYan Zheng 887ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 88857a304cfSJosef Bacik ASSERT(ret != -EEXIST); 88900bb36a0SJosef Bacik if (ret) { 89000bb36a0SJosef Bacik /* Pairs with create_reloc_root */ 89100bb36a0SJosef Bacik btrfs_put_root(reloc_root); 89200bb36a0SJosef Bacik return ret; 89300bb36a0SJosef Bacik } 894f44deb74SJosef Bacik root->reloc_root = btrfs_grab_root(reloc_root); 8955d4f98a2SYan Zheng return 0; 8965d4f98a2SYan Zheng } 8975d4f98a2SYan Zheng 8985d4f98a2SYan Zheng /* 8995d4f98a2SYan Zheng * update root item of reloc tree 9005d4f98a2SYan Zheng */ 9015d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 9025d4f98a2SYan Zheng struct btrfs_root *root) 9035d4f98a2SYan Zheng { 9040b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 9055d4f98a2SYan Zheng struct btrfs_root *reloc_root; 9065d4f98a2SYan Zheng struct btrfs_root_item *root_item; 9075d4f98a2SYan Zheng int ret; 9085d4f98a2SYan Zheng 9096282675eSQu Wenruo if (!have_reloc_root(root)) 910592fbcd5SJosef Bacik return 0; 9115d4f98a2SYan Zheng 9125d4f98a2SYan Zheng reloc_root = root->reloc_root; 9135d4f98a2SYan Zheng root_item = &reloc_root->root_item; 9145d4f98a2SYan Zheng 915f44deb74SJosef Bacik /* 916f44deb74SJosef Bacik * We are probably ok here, but __del_reloc_root() will drop its ref of 917f44deb74SJosef Bacik * the root. We have the ref for root->reloc_root, but just in case 918f44deb74SJosef Bacik * hold it while we update the reloc root. 919f44deb74SJosef Bacik */ 920f44deb74SJosef Bacik btrfs_grab_root(reloc_root); 921f44deb74SJosef Bacik 922d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 9230b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 9243fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 925d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 9266282675eSQu Wenruo /* 9276282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 9286282675eSQu Wenruo * have_reloc_root will not touch it from now on. 9296282675eSQu Wenruo */ 9306282675eSQu Wenruo smp_wmb(); 931c974c464SWang Shilong __del_reloc_root(reloc_root); 9325d4f98a2SYan Zheng } 9335d4f98a2SYan Zheng 9345d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 935ea287ab1SJosef Bacik __update_reloc_root(reloc_root); 9365d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 9375d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 9385d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 9395d4f98a2SYan Zheng } 9405d4f98a2SYan Zheng 9410b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 9425d4f98a2SYan Zheng &reloc_root->root_key, root_item); 943f44deb74SJosef Bacik btrfs_put_root(reloc_root); 944592fbcd5SJosef Bacik return ret; 9455d4f98a2SYan Zheng } 9465d4f98a2SYan Zheng 9475d4f98a2SYan Zheng /* 9485d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 9495d4f98a2SYan Zheng * in a subvolume 9505d4f98a2SYan Zheng */ 9515d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 9525d4f98a2SYan Zheng { 9535d4f98a2SYan Zheng struct rb_node *node; 9545d4f98a2SYan Zheng struct rb_node *prev; 9555d4f98a2SYan Zheng struct btrfs_inode *entry; 9565d4f98a2SYan Zheng struct inode *inode; 9575d4f98a2SYan Zheng 9585d4f98a2SYan Zheng spin_lock(&root->inode_lock); 9595d4f98a2SYan Zheng again: 9605d4f98a2SYan Zheng node = root->inode_tree.rb_node; 9615d4f98a2SYan Zheng prev = NULL; 9625d4f98a2SYan Zheng while (node) { 9635d4f98a2SYan Zheng prev = node; 9645d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9655d4f98a2SYan Zheng 9664a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 9675d4f98a2SYan Zheng node = node->rb_left; 9684a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 9695d4f98a2SYan Zheng node = node->rb_right; 9705d4f98a2SYan Zheng else 9715d4f98a2SYan Zheng break; 9725d4f98a2SYan Zheng } 9735d4f98a2SYan Zheng if (!node) { 9745d4f98a2SYan Zheng while (prev) { 9755d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 9764a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 9775d4f98a2SYan Zheng node = prev; 9785d4f98a2SYan Zheng break; 9795d4f98a2SYan Zheng } 9805d4f98a2SYan Zheng prev = rb_next(prev); 9815d4f98a2SYan Zheng } 9825d4f98a2SYan Zheng } 9835d4f98a2SYan Zheng while (node) { 9845d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9855d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 9865d4f98a2SYan Zheng if (inode) { 9875d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 9885d4f98a2SYan Zheng return inode; 9895d4f98a2SYan Zheng } 9905d4f98a2SYan Zheng 9914a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 9925d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 9935d4f98a2SYan Zheng goto again; 9945d4f98a2SYan Zheng 9955d4f98a2SYan Zheng node = rb_next(node); 9965d4f98a2SYan Zheng } 9975d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 9985d4f98a2SYan Zheng return NULL; 9995d4f98a2SYan Zheng } 10005d4f98a2SYan Zheng 10015d4f98a2SYan Zheng /* 10025d4f98a2SYan Zheng * get new location of data 10035d4f98a2SYan Zheng */ 10045d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 10055d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 10065d4f98a2SYan Zheng { 10075d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 10085d4f98a2SYan Zheng struct btrfs_path *path; 10095d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10105d4f98a2SYan Zheng struct extent_buffer *leaf; 10115d4f98a2SYan Zheng int ret; 10125d4f98a2SYan Zheng 10135d4f98a2SYan Zheng path = btrfs_alloc_path(); 10145d4f98a2SYan Zheng if (!path) 10155d4f98a2SYan Zheng return -ENOMEM; 10165d4f98a2SYan Zheng 10175d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1018f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1019f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 10205d4f98a2SYan Zheng if (ret < 0) 10215d4f98a2SYan Zheng goto out; 10225d4f98a2SYan Zheng if (ret > 0) { 10235d4f98a2SYan Zheng ret = -ENOENT; 10245d4f98a2SYan Zheng goto out; 10255d4f98a2SYan Zheng } 10265d4f98a2SYan Zheng 10275d4f98a2SYan Zheng leaf = path->nodes[0]; 10285d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 10295d4f98a2SYan Zheng struct btrfs_file_extent_item); 10305d4f98a2SYan Zheng 10315d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 10325d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 10335d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 10345d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 10355d4f98a2SYan Zheng 10365d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 103783d4cfd4SJosef Bacik ret = -EINVAL; 10385d4f98a2SYan Zheng goto out; 10395d4f98a2SYan Zheng } 10405d4f98a2SYan Zheng 10415d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 10425d4f98a2SYan Zheng ret = 0; 10435d4f98a2SYan Zheng out: 10445d4f98a2SYan Zheng btrfs_free_path(path); 10455d4f98a2SYan Zheng return ret; 10465d4f98a2SYan Zheng } 10475d4f98a2SYan Zheng 10485d4f98a2SYan Zheng /* 10495d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 10505d4f98a2SYan Zheng * the new locations. 10515d4f98a2SYan Zheng */ 10523fd0a558SYan, Zheng static noinline_for_stack 10533fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 10545d4f98a2SYan Zheng struct reloc_control *rc, 10555d4f98a2SYan Zheng struct btrfs_root *root, 10563fd0a558SYan, Zheng struct extent_buffer *leaf) 10575d4f98a2SYan Zheng { 10580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 10595d4f98a2SYan Zheng struct btrfs_key key; 10605d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10615d4f98a2SYan Zheng struct inode *inode = NULL; 10625d4f98a2SYan Zheng u64 parent; 10635d4f98a2SYan Zheng u64 bytenr; 10643fd0a558SYan, Zheng u64 new_bytenr = 0; 10655d4f98a2SYan Zheng u64 num_bytes; 10665d4f98a2SYan Zheng u64 end; 10675d4f98a2SYan Zheng u32 nritems; 10685d4f98a2SYan Zheng u32 i; 106983d4cfd4SJosef Bacik int ret = 0; 10705d4f98a2SYan Zheng int first = 1; 10715d4f98a2SYan Zheng int dirty = 0; 10725d4f98a2SYan Zheng 10735d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 10745d4f98a2SYan Zheng return 0; 10755d4f98a2SYan Zheng 10765d4f98a2SYan Zheng /* reloc trees always use full backref */ 10775d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 10785d4f98a2SYan Zheng parent = leaf->start; 10795d4f98a2SYan Zheng else 10805d4f98a2SYan Zheng parent = 0; 10815d4f98a2SYan Zheng 10825d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 10835d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 108482fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 108582fa113fSQu Wenruo 10865d4f98a2SYan Zheng cond_resched(); 10875d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 10885d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 10895d4f98a2SYan Zheng continue; 10905d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 10915d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 10925d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 10935d4f98a2SYan Zheng continue; 10945d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 10955d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 10965d4f98a2SYan Zheng if (bytenr == 0) 10975d4f98a2SYan Zheng continue; 10989569cc20SQu Wenruo if (!in_range(bytenr, rc->block_group->start, 10999569cc20SQu Wenruo rc->block_group->length)) 11005d4f98a2SYan Zheng continue; 11015d4f98a2SYan Zheng 11025d4f98a2SYan Zheng /* 11035d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 11045d4f98a2SYan Zheng * to complete and drop the extent cache 11055d4f98a2SYan Zheng */ 11065d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 11075d4f98a2SYan Zheng if (first) { 11085d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11095d4f98a2SYan Zheng first = 0; 11104a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 11113fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 11125d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11135d4f98a2SYan Zheng } 11144a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 11155d4f98a2SYan Zheng end = key.offset + 11165d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 11175d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 11180b246afaSJeff Mahoney fs_info->sectorsize)); 11190b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 11205d4f98a2SYan Zheng end--; 11215d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1122d0082371SJeff Mahoney key.offset, end); 11235d4f98a2SYan Zheng if (!ret) 11245d4f98a2SYan Zheng continue; 11255d4f98a2SYan Zheng 1126dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1127dcdbc059SNikolay Borisov key.offset, end, 1); 11285d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1129d0082371SJeff Mahoney key.offset, end); 11305d4f98a2SYan Zheng } 11315d4f98a2SYan Zheng } 11325d4f98a2SYan Zheng 11335d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 11345d4f98a2SYan Zheng bytenr, num_bytes); 113583d4cfd4SJosef Bacik if (ret) { 113683d4cfd4SJosef Bacik /* 113783d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 113883d4cfd4SJosef Bacik * in the file extent yet. 113983d4cfd4SJosef Bacik */ 114083d4cfd4SJosef Bacik break; 11413fd0a558SYan, Zheng } 11425d4f98a2SYan Zheng 11435d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 11445d4f98a2SYan Zheng dirty = 1; 11455d4f98a2SYan Zheng 11465d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 114782fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 114882fa113fSQu Wenruo num_bytes, parent); 114982fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 115082fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1151b06c4bf5SFilipe Manana key.objectid, key.offset); 115282fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 115383d4cfd4SJosef Bacik if (ret) { 115466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 115583d4cfd4SJosef Bacik break; 115683d4cfd4SJosef Bacik } 11575d4f98a2SYan Zheng 1158ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1159ffd4bb2aSQu Wenruo num_bytes, parent); 1160ffd4bb2aSQu Wenruo ref.real_root = root->root_key.objectid; 1161ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1162b06c4bf5SFilipe Manana key.objectid, key.offset); 1163ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 116483d4cfd4SJosef Bacik if (ret) { 116566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 116683d4cfd4SJosef Bacik break; 116783d4cfd4SJosef Bacik } 11685d4f98a2SYan Zheng } 11695d4f98a2SYan Zheng if (dirty) 11705d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 11713fd0a558SYan, Zheng if (inode) 11723fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 117383d4cfd4SJosef Bacik return ret; 11745d4f98a2SYan Zheng } 11755d4f98a2SYan Zheng 11765d4f98a2SYan Zheng static noinline_for_stack 11775d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 11785d4f98a2SYan Zheng struct btrfs_path *path, int level) 11795d4f98a2SYan Zheng { 11805d4f98a2SYan Zheng struct btrfs_disk_key key1; 11815d4f98a2SYan Zheng struct btrfs_disk_key key2; 11825d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 11835d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 11845d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 11855d4f98a2SYan Zheng } 11865d4f98a2SYan Zheng 11875d4f98a2SYan Zheng /* 11885d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 11895d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 11905d4f98a2SYan Zheng * reloc tree was create can be replaced. 11915d4f98a2SYan Zheng * 11925d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 11935d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 11945d4f98a2SYan Zheng * errors, a negative error number is returned. 11955d4f98a2SYan Zheng */ 11963fd0a558SYan, Zheng static noinline_for_stack 11973d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 11985d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 11995d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 12005d4f98a2SYan Zheng int lowest_level, int max_level) 12015d4f98a2SYan Zheng { 12020b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 12035d4f98a2SYan Zheng struct extent_buffer *eb; 12045d4f98a2SYan Zheng struct extent_buffer *parent; 120582fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 12065d4f98a2SYan Zheng struct btrfs_key key; 12075d4f98a2SYan Zheng u64 old_bytenr; 12085d4f98a2SYan Zheng u64 new_bytenr; 12095d4f98a2SYan Zheng u64 old_ptr_gen; 12105d4f98a2SYan Zheng u64 new_ptr_gen; 12115d4f98a2SYan Zheng u64 last_snapshot; 12125d4f98a2SYan Zheng u32 blocksize; 12133fd0a558SYan, Zheng int cow = 0; 12145d4f98a2SYan Zheng int level; 12155d4f98a2SYan Zheng int ret; 12165d4f98a2SYan Zheng int slot; 12175d4f98a2SYan Zheng 12187a9213a9SJosef Bacik ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 12197a9213a9SJosef Bacik ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 12205d4f98a2SYan Zheng 12215d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 12223fd0a558SYan, Zheng again: 12235d4f98a2SYan Zheng slot = path->slots[lowest_level]; 12245d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 12255d4f98a2SYan Zheng 12265d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 12275d4f98a2SYan Zheng level = btrfs_header_level(eb); 12285d4f98a2SYan Zheng 12295d4f98a2SYan Zheng if (level < lowest_level) { 12305d4f98a2SYan Zheng btrfs_tree_unlock(eb); 12315d4f98a2SYan Zheng free_extent_buffer(eb); 12325d4f98a2SYan Zheng return 0; 12335d4f98a2SYan Zheng } 12345d4f98a2SYan Zheng 12353fd0a558SYan, Zheng if (cow) { 12369631e4ccSJosef Bacik ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb, 12379631e4ccSJosef Bacik BTRFS_NESTING_COW); 123845b87c5dSJosef Bacik if (ret) { 123945b87c5dSJosef Bacik btrfs_tree_unlock(eb); 124045b87c5dSJosef Bacik free_extent_buffer(eb); 124145b87c5dSJosef Bacik return ret; 124245b87c5dSJosef Bacik } 12433fd0a558SYan, Zheng } 12445d4f98a2SYan Zheng 12455d4f98a2SYan Zheng if (next_key) { 12465d4f98a2SYan Zheng next_key->objectid = (u64)-1; 12475d4f98a2SYan Zheng next_key->type = (u8)-1; 12485d4f98a2SYan Zheng next_key->offset = (u64)-1; 12495d4f98a2SYan Zheng } 12505d4f98a2SYan Zheng 12515d4f98a2SYan Zheng parent = eb; 12525d4f98a2SYan Zheng while (1) { 12535d4f98a2SYan Zheng level = btrfs_header_level(parent); 12547a9213a9SJosef Bacik ASSERT(level >= lowest_level); 12555d4f98a2SYan Zheng 1256e3b83361SQu Wenruo ret = btrfs_bin_search(parent, &key, &slot); 1257cbca7d59SFilipe Manana if (ret < 0) 1258cbca7d59SFilipe Manana break; 12595d4f98a2SYan Zheng if (ret && slot > 0) 12605d4f98a2SYan Zheng slot--; 12615d4f98a2SYan Zheng 12625d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 12635d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 12645d4f98a2SYan Zheng 12655d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 12660b246afaSJeff Mahoney blocksize = fs_info->nodesize; 12675d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 12685d4f98a2SYan Zheng 12695d4f98a2SYan Zheng if (level <= max_level) { 12705d4f98a2SYan Zheng eb = path->nodes[level]; 12715d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 12725d4f98a2SYan Zheng path->slots[level]); 12735d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 12745d4f98a2SYan Zheng path->slots[level]); 12755d4f98a2SYan Zheng } else { 12765d4f98a2SYan Zheng new_bytenr = 0; 12775d4f98a2SYan Zheng new_ptr_gen = 0; 12785d4f98a2SYan Zheng } 12795d4f98a2SYan Zheng 1280fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 12815d4f98a2SYan Zheng ret = level; 12825d4f98a2SYan Zheng break; 12835d4f98a2SYan Zheng } 12845d4f98a2SYan Zheng 12855d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 12865d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 12873fd0a558SYan, Zheng if (level <= lowest_level) { 12885d4f98a2SYan Zheng ret = 0; 12895d4f98a2SYan Zheng break; 12905d4f98a2SYan Zheng } 12915d4f98a2SYan Zheng 12926b3426beSJosef Bacik eb = btrfs_read_node_slot(parent, slot); 129364c043deSLiu Bo if (IS_ERR(eb)) { 129464c043deSLiu Bo ret = PTR_ERR(eb); 1295264813acSLiu Bo break; 1296416bc658SJosef Bacik } 12975d4f98a2SYan Zheng btrfs_tree_lock(eb); 12983fd0a558SYan, Zheng if (cow) { 12995d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 13009631e4ccSJosef Bacik slot, &eb, 13019631e4ccSJosef Bacik BTRFS_NESTING_COW); 130245b87c5dSJosef Bacik if (ret) { 130345b87c5dSJosef Bacik btrfs_tree_unlock(eb); 130445b87c5dSJosef Bacik free_extent_buffer(eb); 130545b87c5dSJosef Bacik break; 130645b87c5dSJosef Bacik } 13075d4f98a2SYan Zheng } 13085d4f98a2SYan Zheng 13095d4f98a2SYan Zheng btrfs_tree_unlock(parent); 13105d4f98a2SYan Zheng free_extent_buffer(parent); 13115d4f98a2SYan Zheng 13125d4f98a2SYan Zheng parent = eb; 13135d4f98a2SYan Zheng continue; 13145d4f98a2SYan Zheng } 13155d4f98a2SYan Zheng 13163fd0a558SYan, Zheng if (!cow) { 13173fd0a558SYan, Zheng btrfs_tree_unlock(parent); 13183fd0a558SYan, Zheng free_extent_buffer(parent); 13193fd0a558SYan, Zheng cow = 1; 13203fd0a558SYan, Zheng goto again; 13213fd0a558SYan, Zheng } 13223fd0a558SYan, Zheng 13235d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 13245d4f98a2SYan Zheng path->slots[level]); 1325b3b4aa74SDavid Sterba btrfs_release_path(path); 13265d4f98a2SYan Zheng 13275d4f98a2SYan Zheng path->lowest_level = level; 13285d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 13295d4f98a2SYan Zheng path->lowest_level = 0; 13300e9873e2SJosef Bacik if (ret) { 13310e9873e2SJosef Bacik if (ret > 0) 13320e9873e2SJosef Bacik ret = -ENOENT; 13330e9873e2SJosef Bacik break; 13340e9873e2SJosef Bacik } 13355d4f98a2SYan Zheng 13365d4f98a2SYan Zheng /* 1337824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1338824d8dffSQu Wenruo * 1339824d8dffSQu Wenruo * We must trace both trees. 1340824d8dffSQu Wenruo * 1) Tree reloc subtree 1341824d8dffSQu Wenruo * If not traced, we will leak data numbers 1342824d8dffSQu Wenruo * 2) Fs subtree 1343824d8dffSQu Wenruo * If not traced, we will double count old data 1344f616f5cdSQu Wenruo * 1345f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1346f616f5cdSQu Wenruo * the swapped tree blocks. 1347f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1348f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1349824d8dffSQu Wenruo */ 1350370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1351370a11b8SQu Wenruo rc->block_group, parent, slot, 1352370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1353370a11b8SQu Wenruo last_snapshot); 1354370a11b8SQu Wenruo if (ret < 0) 1355370a11b8SQu Wenruo break; 1356824d8dffSQu Wenruo /* 13575d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 13585d4f98a2SYan Zheng */ 13595d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 13605d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 13615d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 13625d4f98a2SYan Zheng 13635d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 13645d4f98a2SYan Zheng path->slots[level], old_bytenr); 13655d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 13665d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 13675d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 13685d4f98a2SYan Zheng 136982fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 137082fa113fSQu Wenruo blocksize, path->nodes[level]->start); 137182fa113fSQu Wenruo ref.skip_qgroup = true; 137282fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 137382fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1374253e258cSJosef Bacik if (ret) { 1375253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1376253e258cSJosef Bacik break; 1377253e258cSJosef Bacik } 137882fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 137982fa113fSQu Wenruo blocksize, 0); 138082fa113fSQu Wenruo ref.skip_qgroup = true; 138182fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 138282fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1383253e258cSJosef Bacik if (ret) { 1384253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1385253e258cSJosef Bacik break; 1386253e258cSJosef Bacik } 13875d4f98a2SYan Zheng 1388ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1389ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1390ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 1391ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1392ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1393253e258cSJosef Bacik if (ret) { 1394253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1395253e258cSJosef Bacik break; 1396253e258cSJosef Bacik } 13975d4f98a2SYan Zheng 1398ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1399ffd4bb2aSQu Wenruo blocksize, 0); 1400ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 1401ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1402ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1403253e258cSJosef Bacik if (ret) { 1404253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1405253e258cSJosef Bacik break; 1406253e258cSJosef Bacik } 14075d4f98a2SYan Zheng 14085d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 14095d4f98a2SYan Zheng 14105d4f98a2SYan Zheng ret = level; 14115d4f98a2SYan Zheng break; 14125d4f98a2SYan Zheng } 14135d4f98a2SYan Zheng btrfs_tree_unlock(parent); 14145d4f98a2SYan Zheng free_extent_buffer(parent); 14155d4f98a2SYan Zheng return ret; 14165d4f98a2SYan Zheng } 14175d4f98a2SYan Zheng 14185d4f98a2SYan Zheng /* 14195d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 14205d4f98a2SYan Zheng */ 14215d4f98a2SYan Zheng static noinline_for_stack 14225d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14235d4f98a2SYan Zheng int *level) 14245d4f98a2SYan Zheng { 14255d4f98a2SYan Zheng struct extent_buffer *eb; 14265d4f98a2SYan Zheng int i; 14275d4f98a2SYan Zheng u64 last_snapshot; 14285d4f98a2SYan Zheng u32 nritems; 14295d4f98a2SYan Zheng 14305d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14315d4f98a2SYan Zheng 14325d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 14335d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14345d4f98a2SYan Zheng path->nodes[i] = NULL; 14355d4f98a2SYan Zheng } 14365d4f98a2SYan Zheng 14375d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 14385d4f98a2SYan Zheng eb = path->nodes[i]; 14395d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14405d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 14415d4f98a2SYan Zheng path->slots[i]++; 14425d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 14435d4f98a2SYan Zheng last_snapshot) 14445d4f98a2SYan Zheng continue; 14455d4f98a2SYan Zheng 14465d4f98a2SYan Zheng *level = i; 14475d4f98a2SYan Zheng return 0; 14485d4f98a2SYan Zheng } 14495d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14505d4f98a2SYan Zheng path->nodes[i] = NULL; 14515d4f98a2SYan Zheng } 14525d4f98a2SYan Zheng return 1; 14535d4f98a2SYan Zheng } 14545d4f98a2SYan Zheng 14555d4f98a2SYan Zheng /* 14565d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 14575d4f98a2SYan Zheng */ 14585d4f98a2SYan Zheng static noinline_for_stack 14595d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14605d4f98a2SYan Zheng int *level) 14615d4f98a2SYan Zheng { 14625d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 14635d4f98a2SYan Zheng int i; 14645d4f98a2SYan Zheng u64 ptr_gen = 0; 14655d4f98a2SYan Zheng u64 last_snapshot; 14665d4f98a2SYan Zheng u32 nritems; 14675d4f98a2SYan Zheng 14685d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14695d4f98a2SYan Zheng 14705d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 14715d4f98a2SYan Zheng eb = path->nodes[i]; 14725d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14735d4f98a2SYan Zheng while (path->slots[i] < nritems) { 14745d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 14755d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 14765d4f98a2SYan Zheng break; 14775d4f98a2SYan Zheng path->slots[i]++; 14785d4f98a2SYan Zheng } 14795d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 14805d4f98a2SYan Zheng if (i == *level) 14815d4f98a2SYan Zheng break; 14825d4f98a2SYan Zheng *level = i + 1; 14835d4f98a2SYan Zheng return 0; 14845d4f98a2SYan Zheng } 14855d4f98a2SYan Zheng if (i == 1) { 14865d4f98a2SYan Zheng *level = i; 14875d4f98a2SYan Zheng return 0; 14885d4f98a2SYan Zheng } 14895d4f98a2SYan Zheng 14908ef385bbSJosef Bacik eb = btrfs_read_node_slot(eb, path->slots[i]); 14918ef385bbSJosef Bacik if (IS_ERR(eb)) 149264c043deSLiu Bo return PTR_ERR(eb); 14935d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 14945d4f98a2SYan Zheng path->nodes[i - 1] = eb; 14955d4f98a2SYan Zheng path->slots[i - 1] = 0; 14965d4f98a2SYan Zheng } 14975d4f98a2SYan Zheng return 1; 14985d4f98a2SYan Zheng } 14995d4f98a2SYan Zheng 15005d4f98a2SYan Zheng /* 15015d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 15025d4f98a2SYan Zheng * [min_key, max_key) 15035d4f98a2SYan Zheng */ 15045d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 15055d4f98a2SYan Zheng struct btrfs_key *min_key, 15065d4f98a2SYan Zheng struct btrfs_key *max_key) 15075d4f98a2SYan Zheng { 15080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15095d4f98a2SYan Zheng struct inode *inode = NULL; 15105d4f98a2SYan Zheng u64 objectid; 15115d4f98a2SYan Zheng u64 start, end; 151233345d01SLi Zefan u64 ino; 15135d4f98a2SYan Zheng 15145d4f98a2SYan Zheng objectid = min_key->objectid; 15155d4f98a2SYan Zheng while (1) { 15165d4f98a2SYan Zheng cond_resched(); 15175d4f98a2SYan Zheng iput(inode); 15185d4f98a2SYan Zheng 15195d4f98a2SYan Zheng if (objectid > max_key->objectid) 15205d4f98a2SYan Zheng break; 15215d4f98a2SYan Zheng 15225d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 15235d4f98a2SYan Zheng if (!inode) 15245d4f98a2SYan Zheng break; 15254a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 15265d4f98a2SYan Zheng 152733345d01SLi Zefan if (ino > max_key->objectid) { 15285d4f98a2SYan Zheng iput(inode); 15295d4f98a2SYan Zheng break; 15305d4f98a2SYan Zheng } 15315d4f98a2SYan Zheng 153233345d01SLi Zefan objectid = ino + 1; 15335d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 15345d4f98a2SYan Zheng continue; 15355d4f98a2SYan Zheng 153633345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 15375d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 15385d4f98a2SYan Zheng continue; 15395d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 15405d4f98a2SYan Zheng start = 0; 15415d4f98a2SYan Zheng else { 15425d4f98a2SYan Zheng start = min_key->offset; 15430b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 15445d4f98a2SYan Zheng } 15455d4f98a2SYan Zheng } else { 15465d4f98a2SYan Zheng start = 0; 15475d4f98a2SYan Zheng } 15485d4f98a2SYan Zheng 154933345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 15505d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 15515d4f98a2SYan Zheng continue; 15525d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 15535d4f98a2SYan Zheng end = (u64)-1; 15545d4f98a2SYan Zheng } else { 15555d4f98a2SYan Zheng if (max_key->offset == 0) 15565d4f98a2SYan Zheng continue; 15575d4f98a2SYan Zheng end = max_key->offset; 15580b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 15595d4f98a2SYan Zheng end--; 15605d4f98a2SYan Zheng } 15615d4f98a2SYan Zheng } else { 15625d4f98a2SYan Zheng end = (u64)-1; 15635d4f98a2SYan Zheng } 15645d4f98a2SYan Zheng 15655d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 1566d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 1567dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 1568d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 15695d4f98a2SYan Zheng } 15705d4f98a2SYan Zheng return 0; 15715d4f98a2SYan Zheng } 15725d4f98a2SYan Zheng 15735d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 15745d4f98a2SYan Zheng struct btrfs_key *key) 15755d4f98a2SYan Zheng 15765d4f98a2SYan Zheng { 15775d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 15785d4f98a2SYan Zheng if (!path->nodes[level]) 15795d4f98a2SYan Zheng break; 15805d4f98a2SYan Zheng if (path->slots[level] + 1 < 15815d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 15825d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 15835d4f98a2SYan Zheng path->slots[level] + 1); 15845d4f98a2SYan Zheng return 0; 15855d4f98a2SYan Zheng } 15865d4f98a2SYan Zheng level++; 15875d4f98a2SYan Zheng } 15885d4f98a2SYan Zheng return 1; 15895d4f98a2SYan Zheng } 15905d4f98a2SYan Zheng 15915d4f98a2SYan Zheng /* 1592d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 1593d2311e69SQu Wenruo */ 1594ac54da6cSJosef Bacik static int insert_dirty_subvol(struct btrfs_trans_handle *trans, 1595d2311e69SQu Wenruo struct reloc_control *rc, 1596d2311e69SQu Wenruo struct btrfs_root *root) 1597d2311e69SQu Wenruo { 1598d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1599d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 16007934133fSJosef Bacik int ret; 1601d2311e69SQu Wenruo 1602d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 1603d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 1604d2311e69SQu Wenruo ASSERT(reloc_root); 1605d2311e69SQu Wenruo 1606d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 1607d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 1608d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 1609c8422684SDavid Sterba btrfs_set_root_drop_level(reloc_root_item, 0); 1610d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 16117934133fSJosef Bacik ret = btrfs_update_reloc_root(trans, root); 16127934133fSJosef Bacik if (ret) 16137934133fSJosef Bacik return ret; 1614d2311e69SQu Wenruo 1615d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 161600246528SJosef Bacik btrfs_grab_root(root); 1617d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 1618d2311e69SQu Wenruo } 1619ac54da6cSJosef Bacik 1620ac54da6cSJosef Bacik return 0; 1621d2311e69SQu Wenruo } 1622d2311e69SQu Wenruo 1623d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 1624d2311e69SQu Wenruo { 1625d2311e69SQu Wenruo struct btrfs_root *root; 1626d2311e69SQu Wenruo struct btrfs_root *next; 1627d2311e69SQu Wenruo int ret = 0; 162830d40577SQu Wenruo int ret2; 1629d2311e69SQu Wenruo 1630d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 1631d2311e69SQu Wenruo reloc_dirty_list) { 163230d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 163330d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 1634d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1635d2311e69SQu Wenruo 1636d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 1637d2311e69SQu Wenruo root->reloc_root = NULL; 16386282675eSQu Wenruo /* 16396282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 16406282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 16416282675eSQu Wenruo */ 16426282675eSQu Wenruo smp_wmb(); 16431fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 1644f28de8d8SJosef Bacik if (reloc_root) { 1645f44deb74SJosef Bacik /* 1646f44deb74SJosef Bacik * btrfs_drop_snapshot drops our ref we hold for 1647f44deb74SJosef Bacik * ->reloc_root. If it fails however we must 1648f44deb74SJosef Bacik * drop the ref ourselves. 1649f44deb74SJosef Bacik */ 1650f28de8d8SJosef Bacik ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); 1651f44deb74SJosef Bacik if (ret2 < 0) { 1652f44deb74SJosef Bacik btrfs_put_root(reloc_root); 1653f44deb74SJosef Bacik if (!ret) 1654f28de8d8SJosef Bacik ret = ret2; 1655f28de8d8SJosef Bacik } 1656f44deb74SJosef Bacik } 165700246528SJosef Bacik btrfs_put_root(root); 165830d40577SQu Wenruo } else { 165930d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 16600078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(root, 0, 1); 1661f44deb74SJosef Bacik if (ret2 < 0) { 1662f44deb74SJosef Bacik btrfs_put_root(root); 1663f44deb74SJosef Bacik if (!ret) 166430d40577SQu Wenruo ret = ret2; 166530d40577SQu Wenruo } 1666d2311e69SQu Wenruo } 1667f44deb74SJosef Bacik } 1668d2311e69SQu Wenruo return ret; 1669d2311e69SQu Wenruo } 1670d2311e69SQu Wenruo 1671d2311e69SQu Wenruo /* 16725d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 16735d4f98a2SYan Zheng * fs tree. 16745d4f98a2SYan Zheng */ 16755d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 16765d4f98a2SYan Zheng struct btrfs_root *root) 16775d4f98a2SYan Zheng { 16780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 16795d4f98a2SYan Zheng struct btrfs_key key; 16805d4f98a2SYan Zheng struct btrfs_key next_key; 16819e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 16825d4f98a2SYan Zheng struct btrfs_root *reloc_root; 16835d4f98a2SYan Zheng struct btrfs_root_item *root_item; 16845d4f98a2SYan Zheng struct btrfs_path *path; 16853fd0a558SYan, Zheng struct extent_buffer *leaf; 1686fca3a45dSJosef Bacik int reserve_level; 16875d4f98a2SYan Zheng int level; 16885d4f98a2SYan Zheng int max_level; 16895d4f98a2SYan Zheng int replaced = 0; 1690c6a592f2SNikolay Borisov int ret = 0; 16913fd0a558SYan, Zheng u32 min_reserved; 16925d4f98a2SYan Zheng 16935d4f98a2SYan Zheng path = btrfs_alloc_path(); 16945d4f98a2SYan Zheng if (!path) 16955d4f98a2SYan Zheng return -ENOMEM; 1696e4058b54SDavid Sterba path->reada = READA_FORWARD; 16975d4f98a2SYan Zheng 16985d4f98a2SYan Zheng reloc_root = root->reloc_root; 16995d4f98a2SYan Zheng root_item = &reloc_root->root_item; 17005d4f98a2SYan Zheng 17015d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 17025d4f98a2SYan Zheng level = btrfs_root_level(root_item); 170367439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 17045d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 17055d4f98a2SYan Zheng path->slots[level] = 0; 17065d4f98a2SYan Zheng } else { 17075d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 17085d4f98a2SYan Zheng 1709c8422684SDavid Sterba level = btrfs_root_drop_level(root_item); 17105d4f98a2SYan Zheng BUG_ON(level == 0); 17115d4f98a2SYan Zheng path->lowest_level = level; 17125d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 171333c66f43SYan Zheng path->lowest_level = 0; 17145d4f98a2SYan Zheng if (ret < 0) { 17155d4f98a2SYan Zheng btrfs_free_path(path); 17165d4f98a2SYan Zheng return ret; 17175d4f98a2SYan Zheng } 17185d4f98a2SYan Zheng 17195d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 17205d4f98a2SYan Zheng path->slots[level]); 17215d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 17225d4f98a2SYan Zheng 17235d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 17245d4f98a2SYan Zheng } 17255d4f98a2SYan Zheng 172644d354abSQu Wenruo /* 172744d354abSQu Wenruo * In merge_reloc_root(), we modify the upper level pointer to swap the 172844d354abSQu Wenruo * tree blocks between reloc tree and subvolume tree. Thus for tree 172944d354abSQu Wenruo * block COW, we COW at most from level 1 to root level for each tree. 173044d354abSQu Wenruo * 173144d354abSQu Wenruo * Thus the needed metadata size is at most root_level * nodesize, 173244d354abSQu Wenruo * and * 2 since we have two trees to COW. 173344d354abSQu Wenruo */ 1734fca3a45dSJosef Bacik reserve_level = max_t(int, 1, btrfs_root_level(root_item)); 1735fca3a45dSJosef Bacik min_reserved = fs_info->nodesize * reserve_level * 2; 17365d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 17375d4f98a2SYan Zheng 17385d4f98a2SYan Zheng while (1) { 173908e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 174044d354abSQu Wenruo BTRFS_RESERVE_FLUSH_LIMIT); 1741c6a592f2SNikolay Borisov if (ret) 17429e6a0c52SJosef Bacik goto out; 17439e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 17449e6a0c52SJosef Bacik if (IS_ERR(trans)) { 1745c6a592f2SNikolay Borisov ret = PTR_ERR(trans); 17469e6a0c52SJosef Bacik trans = NULL; 17479e6a0c52SJosef Bacik goto out; 17489e6a0c52SJosef Bacik } 17492abc726aSJosef Bacik 17502abc726aSJosef Bacik /* 17512abc726aSJosef Bacik * At this point we no longer have a reloc_control, so we can't 17522abc726aSJosef Bacik * depend on btrfs_init_reloc_root to update our last_trans. 17532abc726aSJosef Bacik * 17542abc726aSJosef Bacik * But that's ok, we started the trans handle on our 17552abc726aSJosef Bacik * corresponding fs_root, which means it's been added to the 17562abc726aSJosef Bacik * dirty list. At commit time we'll still call 17572abc726aSJosef Bacik * btrfs_update_reloc_root() and update our root item 17582abc726aSJosef Bacik * appropriately. 17592abc726aSJosef Bacik */ 17602abc726aSJosef Bacik reloc_root->last_trans = trans->transid; 17619e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 17623fd0a558SYan, Zheng 17633fd0a558SYan, Zheng replaced = 0; 17645d4f98a2SYan Zheng max_level = level; 17655d4f98a2SYan Zheng 17665d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 1767c6a592f2SNikolay Borisov if (ret < 0) 17685d4f98a2SYan Zheng goto out; 17695d4f98a2SYan Zheng if (ret > 0) 17705d4f98a2SYan Zheng break; 17715d4f98a2SYan Zheng 17725d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 17735d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 17745d4f98a2SYan Zheng ret = 0; 17755d4f98a2SYan Zheng } else { 17763d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 17773fd0a558SYan, Zheng &next_key, level, max_level); 17785d4f98a2SYan Zheng } 1779c6a592f2SNikolay Borisov if (ret < 0) 17805d4f98a2SYan Zheng goto out; 17815d4f98a2SYan Zheng if (ret > 0) { 17825d4f98a2SYan Zheng level = ret; 17835d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 17845d4f98a2SYan Zheng path->slots[level]); 17855d4f98a2SYan Zheng replaced = 1; 17865d4f98a2SYan Zheng } 17875d4f98a2SYan Zheng 17885d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 17895d4f98a2SYan Zheng if (ret > 0) 17905d4f98a2SYan Zheng break; 17915d4f98a2SYan Zheng 17925d4f98a2SYan Zheng BUG_ON(level == 0); 17935d4f98a2SYan Zheng /* 17945d4f98a2SYan Zheng * save the merging progress in the drop_progress. 17955d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 17965d4f98a2SYan Zheng */ 17975d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 17985d4f98a2SYan Zheng path->slots[level]); 1799c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, level); 18005d4f98a2SYan Zheng 18013a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18029e6a0c52SJosef Bacik trans = NULL; 18035d4f98a2SYan Zheng 18042ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18055d4f98a2SYan Zheng 18065d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18075d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18085d4f98a2SYan Zheng } 18095d4f98a2SYan Zheng 18105d4f98a2SYan Zheng /* 18115d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 18125d4f98a2SYan Zheng * relocated and the block is tree root. 18135d4f98a2SYan Zheng */ 18145d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 18159631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf, 18169631e4ccSJosef Bacik BTRFS_NESTING_COW); 18175d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 18185d4f98a2SYan Zheng free_extent_buffer(leaf); 18195d4f98a2SYan Zheng out: 18205d4f98a2SYan Zheng btrfs_free_path(path); 18215d4f98a2SYan Zheng 1822ac54da6cSJosef Bacik if (ret == 0) { 1823ac54da6cSJosef Bacik ret = insert_dirty_subvol(trans, rc, root); 1824ac54da6cSJosef Bacik if (ret) 1825ac54da6cSJosef Bacik btrfs_abort_transaction(trans, ret); 1826ac54da6cSJosef Bacik } 18275d4f98a2SYan Zheng 18289e6a0c52SJosef Bacik if (trans) 18293a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18305d4f98a2SYan Zheng 18312ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18325d4f98a2SYan Zheng 18335d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18345d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18355d4f98a2SYan Zheng 1836c6a592f2SNikolay Borisov return ret; 18375d4f98a2SYan Zheng } 18385d4f98a2SYan Zheng 18393fd0a558SYan, Zheng static noinline_for_stack 18403fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 18415d4f98a2SYan Zheng { 18423fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 18430b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 18443fd0a558SYan, Zheng struct btrfs_root *reloc_root; 18455d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 18463fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 18473fd0a558SYan, Zheng u64 num_bytes = 0; 18483fd0a558SYan, Zheng int ret; 18493fd0a558SYan, Zheng 18500b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 18510b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 18523fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 18530b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 18547585717fSChris Mason 18553fd0a558SYan, Zheng again: 18563fd0a558SYan, Zheng if (!err) { 18573fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 185808e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 185908e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 18603fd0a558SYan, Zheng if (ret) 18613fd0a558SYan, Zheng err = ret; 18623fd0a558SYan, Zheng } 18633fd0a558SYan, Zheng 18647a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 18653612b495STsutomu Itoh if (IS_ERR(trans)) { 18663612b495STsutomu Itoh if (!err) 18672ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 186863f018beSNikolay Borisov num_bytes, NULL); 18693612b495STsutomu Itoh return PTR_ERR(trans); 18703612b495STsutomu Itoh } 18713fd0a558SYan, Zheng 18723fd0a558SYan, Zheng if (!err) { 18733fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 18743a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18752ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 187663f018beSNikolay Borisov num_bytes, NULL); 18773fd0a558SYan, Zheng goto again; 18783fd0a558SYan, Zheng } 18793fd0a558SYan, Zheng } 18803fd0a558SYan, Zheng 18813fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 18823fd0a558SYan, Zheng 18833fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 18843fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 18853fd0a558SYan, Zheng struct btrfs_root, root_list); 18863fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 18873fd0a558SYan, Zheng 1888a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1889a820feb5SDavid Sterba false); 1890e0b085b0SJosef Bacik if (IS_ERR(root)) { 1891e0b085b0SJosef Bacik /* 1892e0b085b0SJosef Bacik * Even if we have an error we need this reloc root 1893e0b085b0SJosef Bacik * back on our list so we can clean up properly. 1894e0b085b0SJosef Bacik */ 1895e0b085b0SJosef Bacik list_add(&reloc_root->root_list, &reloc_roots); 1896e0b085b0SJosef Bacik btrfs_abort_transaction(trans, (int)PTR_ERR(root)); 1897e0b085b0SJosef Bacik if (!err) 1898e0b085b0SJosef Bacik err = PTR_ERR(root); 1899e0b085b0SJosef Bacik break; 1900e0b085b0SJosef Bacik } 1901e0b085b0SJosef Bacik ASSERT(root->reloc_root == reloc_root); 19023fd0a558SYan, Zheng 19033fd0a558SYan, Zheng /* 19043fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 19053fd0a558SYan, Zheng * knows it should resumes merging 19063fd0a558SYan, Zheng */ 19073fd0a558SYan, Zheng if (!err) 19083fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 1909bbae13f8SJosef Bacik ret = btrfs_update_reloc_root(trans, root); 19103fd0a558SYan, Zheng 1911bbae13f8SJosef Bacik /* 1912bbae13f8SJosef Bacik * Even if we have an error we need this reloc root back on our 1913bbae13f8SJosef Bacik * list so we can clean up properly. 1914bbae13f8SJosef Bacik */ 19153fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 191600246528SJosef Bacik btrfs_put_root(root); 1917bbae13f8SJosef Bacik 1918bbae13f8SJosef Bacik if (ret) { 1919bbae13f8SJosef Bacik btrfs_abort_transaction(trans, ret); 1920bbae13f8SJosef Bacik if (!err) 1921bbae13f8SJosef Bacik err = ret; 1922bbae13f8SJosef Bacik break; 1923bbae13f8SJosef Bacik } 19243fd0a558SYan, Zheng } 19253fd0a558SYan, Zheng 19263fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 19273fd0a558SYan, Zheng 19283fd0a558SYan, Zheng if (!err) 1929fb686c68SJosef Bacik err = btrfs_commit_transaction(trans); 19303fd0a558SYan, Zheng else 19313a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19323fd0a558SYan, Zheng return err; 19333fd0a558SYan, Zheng } 19343fd0a558SYan, Zheng 19353fd0a558SYan, Zheng static noinline_for_stack 1936aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 1937aca1bba6SLiu Bo { 1938a7571232SNikolay Borisov struct btrfs_root *reloc_root, *tmp; 1939aca1bba6SLiu Bo 1940a7571232SNikolay Borisov list_for_each_entry_safe(reloc_root, tmp, list, root_list) 1941bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 1942aca1bba6SLiu Bo } 1943aca1bba6SLiu Bo 1944aca1bba6SLiu Bo static noinline_for_stack 194594404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 19463fd0a558SYan, Zheng { 19470b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 19485d4f98a2SYan Zheng struct btrfs_root *root; 19495d4f98a2SYan Zheng struct btrfs_root *reloc_root; 19503fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 19513fd0a558SYan, Zheng int found = 0; 1952aca1bba6SLiu Bo int ret = 0; 19533fd0a558SYan, Zheng again: 19543fd0a558SYan, Zheng root = rc->extent_root; 19557585717fSChris Mason 19567585717fSChris Mason /* 19577585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 19587585717fSChris Mason * we have to make sure nobody is in the middle of 19597585717fSChris Mason * adding their roots to the list while we are 19607585717fSChris Mason * doing this splice 19617585717fSChris Mason */ 19620b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 19633fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 19640b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 19655d4f98a2SYan Zheng 19663fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 19673fd0a558SYan, Zheng found = 1; 19683fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 19693fd0a558SYan, Zheng struct btrfs_root, root_list); 19705d4f98a2SYan Zheng 1971a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1972a820feb5SDavid Sterba false); 19735d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 197424213fa4SJosef Bacik if (IS_ERR(root)) { 197524213fa4SJosef Bacik /* 197624213fa4SJosef Bacik * For recovery we read the fs roots on mount, 197724213fa4SJosef Bacik * and if we didn't find the root then we marked 197824213fa4SJosef Bacik * the reloc root as a garbage root. For normal 197924213fa4SJosef Bacik * relocation obviously the root should exist in 198024213fa4SJosef Bacik * memory. However there's no reason we can't 198124213fa4SJosef Bacik * handle the error properly here just in case. 198224213fa4SJosef Bacik */ 198324213fa4SJosef Bacik ASSERT(0); 198424213fa4SJosef Bacik ret = PTR_ERR(root); 198524213fa4SJosef Bacik goto out; 198624213fa4SJosef Bacik } 198724213fa4SJosef Bacik if (root->reloc_root != reloc_root) { 198824213fa4SJosef Bacik /* 198924213fa4SJosef Bacik * This is actually impossible without something 199024213fa4SJosef Bacik * going really wrong (like weird race condition 199124213fa4SJosef Bacik * or cosmic rays). 199224213fa4SJosef Bacik */ 199324213fa4SJosef Bacik ASSERT(0); 199424213fa4SJosef Bacik ret = -EINVAL; 199524213fa4SJosef Bacik goto out; 199624213fa4SJosef Bacik } 19973fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 199800246528SJosef Bacik btrfs_put_root(root); 1999b37b39cdSJosef Bacik if (ret) { 200025e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 200125e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 200225e293c2SWang Shilong &reloc_roots); 2003aca1bba6SLiu Bo goto out; 2004b37b39cdSJosef Bacik } 20053fd0a558SYan, Zheng } else { 200651415b6cSQu Wenruo if (!IS_ERR(root)) { 200751415b6cSQu Wenruo if (root->reloc_root == reloc_root) { 200851415b6cSQu Wenruo root->reloc_root = NULL; 200951415b6cSQu Wenruo btrfs_put_root(reloc_root); 201051415b6cSQu Wenruo } 20111dae7e0eSQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, 20121dae7e0eSQu Wenruo &root->state); 201351415b6cSQu Wenruo btrfs_put_root(root); 201451415b6cSQu Wenruo } 201551415b6cSQu Wenruo 20163fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 201730d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 201830d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 201930d40577SQu Wenruo &rc->dirty_subvol_roots); 20203fd0a558SYan, Zheng } 20215d4f98a2SYan Zheng } 20225d4f98a2SYan Zheng 20233fd0a558SYan, Zheng if (found) { 20243fd0a558SYan, Zheng found = 0; 20253fd0a558SYan, Zheng goto again; 20265d4f98a2SYan Zheng } 2027aca1bba6SLiu Bo out: 2028aca1bba6SLiu Bo if (ret) { 20290b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2030aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2031467bb1d2SWang Shilong 2032467bb1d2SWang Shilong /* new reloc root may be added */ 20330b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2034467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 20350b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2036467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2037aca1bba6SLiu Bo } 2038aca1bba6SLiu Bo 20397b7b7431SJosef Bacik /* 20407b7b7431SJosef Bacik * We used to have 20417b7b7431SJosef Bacik * 20427b7b7431SJosef Bacik * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 20437b7b7431SJosef Bacik * 20447b7b7431SJosef Bacik * here, but it's wrong. If we fail to start the transaction in 20457b7b7431SJosef Bacik * prepare_to_merge() we will have only 0 ref reloc roots, none of which 20467b7b7431SJosef Bacik * have actually been removed from the reloc_root_tree rb tree. This is 20477b7b7431SJosef Bacik * fine because we're bailing here, and we hold a reference on the root 20487b7b7431SJosef Bacik * for the list that holds it, so these roots will be cleaned up when we 20497b7b7431SJosef Bacik * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root 20507b7b7431SJosef Bacik * will be cleaned up on unmount. 20517b7b7431SJosef Bacik * 20527b7b7431SJosef Bacik * The remaining nodes will be cleaned up by free_reloc_control. 20537b7b7431SJosef Bacik */ 20545d4f98a2SYan Zheng } 20555d4f98a2SYan Zheng 20565d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 20575d4f98a2SYan Zheng { 20585d4f98a2SYan Zheng struct tree_block *block; 20595d4f98a2SYan Zheng struct rb_node *rb_node; 20605d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 20615d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 20625d4f98a2SYan Zheng rb_erase(rb_node, blocks); 20635d4f98a2SYan Zheng kfree(block); 20645d4f98a2SYan Zheng } 20655d4f98a2SYan Zheng } 20665d4f98a2SYan Zheng 20675d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 20685d4f98a2SYan Zheng struct btrfs_root *reloc_root) 20695d4f98a2SYan Zheng { 20700b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 20715d4f98a2SYan Zheng struct btrfs_root *root; 2072442b1ac5SJosef Bacik int ret; 20735d4f98a2SYan Zheng 20745d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 20755d4f98a2SYan Zheng return 0; 20765d4f98a2SYan Zheng 2077a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); 2078404bccbcSJosef Bacik 2079404bccbcSJosef Bacik /* 2080404bccbcSJosef Bacik * This should succeed, since we can't have a reloc root without having 2081404bccbcSJosef Bacik * already looked up the actual root and created the reloc root for this 2082404bccbcSJosef Bacik * root. 2083404bccbcSJosef Bacik * 2084404bccbcSJosef Bacik * However if there's some sort of corruption where we have a ref to a 2085404bccbcSJosef Bacik * reloc root without a corresponding root this could return ENOENT. 2086404bccbcSJosef Bacik */ 2087404bccbcSJosef Bacik if (IS_ERR(root)) { 2088404bccbcSJosef Bacik ASSERT(0); 2089404bccbcSJosef Bacik return PTR_ERR(root); 2090404bccbcSJosef Bacik } 2091404bccbcSJosef Bacik if (root->reloc_root != reloc_root) { 2092404bccbcSJosef Bacik ASSERT(0); 2093404bccbcSJosef Bacik btrfs_err(fs_info, 2094404bccbcSJosef Bacik "root %llu has two reloc roots associated with it", 2095404bccbcSJosef Bacik reloc_root->root_key.offset); 2096404bccbcSJosef Bacik btrfs_put_root(root); 2097404bccbcSJosef Bacik return -EUCLEAN; 2098404bccbcSJosef Bacik } 2099442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 210000246528SJosef Bacik btrfs_put_root(root); 21015d4f98a2SYan Zheng 2102442b1ac5SJosef Bacik return ret; 21035d4f98a2SYan Zheng } 21045d4f98a2SYan Zheng 21053fd0a558SYan, Zheng static noinline_for_stack 21063fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 21073fd0a558SYan, Zheng struct reloc_control *rc, 2108a26195a5SQu Wenruo struct btrfs_backref_node *node, 2109a26195a5SQu Wenruo struct btrfs_backref_edge *edges[]) 21105d4f98a2SYan Zheng { 2111a26195a5SQu Wenruo struct btrfs_backref_node *next; 21125d4f98a2SYan Zheng struct btrfs_root *root; 21133fd0a558SYan, Zheng int index = 0; 211492de551bSJosef Bacik int ret; 21153fd0a558SYan, Zheng 21165d4f98a2SYan Zheng next = node; 21175d4f98a2SYan Zheng while (1) { 21185d4f98a2SYan Zheng cond_resched(); 21195d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 21205d4f98a2SYan Zheng root = next->root; 21218ee66afeSJosef Bacik 21228ee66afeSJosef Bacik /* 21238ee66afeSJosef Bacik * If there is no root, then our references for this block are 21248ee66afeSJosef Bacik * incomplete, as we should be able to walk all the way up to a 21258ee66afeSJosef Bacik * block that is owned by a root. 21268ee66afeSJosef Bacik * 21278ee66afeSJosef Bacik * This path is only for SHAREABLE roots, so if we come upon a 21288ee66afeSJosef Bacik * non-SHAREABLE root then we have backrefs that resolve 21298ee66afeSJosef Bacik * improperly. 21308ee66afeSJosef Bacik * 21318ee66afeSJosef Bacik * Both of these cases indicate file system corruption, or a bug 21328ee66afeSJosef Bacik * in the backref walking code. 21338ee66afeSJosef Bacik */ 21348ee66afeSJosef Bacik if (!root) { 21358ee66afeSJosef Bacik ASSERT(0); 21368ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21378ee66afeSJosef Bacik "bytenr %llu doesn't have a backref path ending in a root", 21388ee66afeSJosef Bacik node->bytenr); 21398ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21408ee66afeSJosef Bacik } 21418ee66afeSJosef Bacik if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 21428ee66afeSJosef Bacik ASSERT(0); 21438ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21448ee66afeSJosef Bacik "bytenr %llu has multiple refs with one ending in a non-shareable root", 21458ee66afeSJosef Bacik node->bytenr); 21468ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21478ee66afeSJosef Bacik } 21485d4f98a2SYan Zheng 21495d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 215092de551bSJosef Bacik ret = record_reloc_root_in_trans(trans, root); 215192de551bSJosef Bacik if (ret) 215292de551bSJosef Bacik return ERR_PTR(ret); 21535d4f98a2SYan Zheng break; 21545d4f98a2SYan Zheng } 21555d4f98a2SYan Zheng 215692de551bSJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 215792de551bSJosef Bacik if (ret) 215892de551bSJosef Bacik return ERR_PTR(ret); 21593fd0a558SYan, Zheng root = root->reloc_root; 21603fd0a558SYan, Zheng 216139200e59SJosef Bacik /* 216239200e59SJosef Bacik * We could have raced with another thread which failed, so 216339200e59SJosef Bacik * root->reloc_root may not be set, return ENOENT in this case. 216439200e59SJosef Bacik */ 216539200e59SJosef Bacik if (!root) 216639200e59SJosef Bacik return ERR_PTR(-ENOENT); 216739200e59SJosef Bacik 21683fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 21698ee66afeSJosef Bacik /* 21708ee66afeSJosef Bacik * We just created the reloc root, so we shouldn't have 21718ee66afeSJosef Bacik * ->new_bytenr set and this shouldn't be in the changed 21728ee66afeSJosef Bacik * list. If it is then we have multiple roots pointing 21738ee66afeSJosef Bacik * at the same bytenr which indicates corruption, or 21748ee66afeSJosef Bacik * we've made a mistake in the backref walking code. 21758ee66afeSJosef Bacik */ 21768ee66afeSJosef Bacik ASSERT(next->new_bytenr == 0); 21778ee66afeSJosef Bacik ASSERT(list_empty(&next->list)); 21788ee66afeSJosef Bacik if (next->new_bytenr || !list_empty(&next->list)) { 21798ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21808ee66afeSJosef Bacik "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu", 21818ee66afeSJosef Bacik node->bytenr, next->bytenr); 21828ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21838ee66afeSJosef Bacik } 21848ee66afeSJosef Bacik 21853fd0a558SYan, Zheng next->new_bytenr = root->node->start; 218600246528SJosef Bacik btrfs_put_root(next->root); 218700246528SJosef Bacik next->root = btrfs_grab_root(root); 21880b530bc5SJosef Bacik ASSERT(next->root); 21893fd0a558SYan, Zheng list_add_tail(&next->list, 21903fd0a558SYan, Zheng &rc->backref_cache.changed); 21919569cc20SQu Wenruo mark_block_processed(rc, next); 21925d4f98a2SYan Zheng break; 21935d4f98a2SYan Zheng } 21945d4f98a2SYan Zheng 21953fd0a558SYan, Zheng WARN_ON(1); 21965d4f98a2SYan Zheng root = NULL; 21975d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 21985d4f98a2SYan Zheng if (!next || next->level <= node->level) 21995d4f98a2SYan Zheng break; 22005d4f98a2SYan Zheng } 2201cbdc2ebcSJosef Bacik if (!root) { 2202cbdc2ebcSJosef Bacik /* 2203cbdc2ebcSJosef Bacik * This can happen if there's fs corruption or if there's a bug 2204cbdc2ebcSJosef Bacik * in the backref lookup code. 2205cbdc2ebcSJosef Bacik */ 2206cbdc2ebcSJosef Bacik ASSERT(0); 2207cbdc2ebcSJosef Bacik return ERR_PTR(-ENOENT); 2208cbdc2ebcSJosef Bacik } 22095d4f98a2SYan Zheng 22103fd0a558SYan, Zheng next = node; 22113fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 22123fd0a558SYan, Zheng while (1) { 22133fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 22143fd0a558SYan, Zheng if (--index < 0) 22153fd0a558SYan, Zheng break; 22163fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 22173fd0a558SYan, Zheng } 22185d4f98a2SYan Zheng return root; 22195d4f98a2SYan Zheng } 22205d4f98a2SYan Zheng 22213fd0a558SYan, Zheng /* 222292a7cc42SQu Wenruo * Select a tree root for relocation. 222392a7cc42SQu Wenruo * 222492a7cc42SQu Wenruo * Return NULL if the block is not shareable. We should use do_relocation() in 222592a7cc42SQu Wenruo * this case. 222692a7cc42SQu Wenruo * 222792a7cc42SQu Wenruo * Return a tree root pointer if the block is shareable. 222892a7cc42SQu Wenruo * Return -ENOENT if the block is root of reloc tree. 22293fd0a558SYan, Zheng */ 22305d4f98a2SYan Zheng static noinline_for_stack 2231a26195a5SQu Wenruo struct btrfs_root *select_one_root(struct btrfs_backref_node *node) 22325d4f98a2SYan Zheng { 2233a26195a5SQu Wenruo struct btrfs_backref_node *next; 22343fd0a558SYan, Zheng struct btrfs_root *root; 22353fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 2236a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22373fd0a558SYan, Zheng int index = 0; 22383fd0a558SYan, Zheng 22393fd0a558SYan, Zheng next = node; 22403fd0a558SYan, Zheng while (1) { 22413fd0a558SYan, Zheng cond_resched(); 22423fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 22433fd0a558SYan, Zheng root = next->root; 22448717cf44SJosef Bacik 22458717cf44SJosef Bacik /* 22468717cf44SJosef Bacik * This can occur if we have incomplete extent refs leading all 22478717cf44SJosef Bacik * the way up a particular path, in this case return -EUCLEAN. 22488717cf44SJosef Bacik */ 22498717cf44SJosef Bacik if (!root) 22508717cf44SJosef Bacik return ERR_PTR(-EUCLEAN); 22513fd0a558SYan, Zheng 225292a7cc42SQu Wenruo /* No other choice for non-shareable tree */ 225392a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 22543fd0a558SYan, Zheng return root; 22553fd0a558SYan, Zheng 22563fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 22573fd0a558SYan, Zheng fs_root = root; 22583fd0a558SYan, Zheng 22593fd0a558SYan, Zheng if (next != node) 22603fd0a558SYan, Zheng return NULL; 22613fd0a558SYan, Zheng 22623fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 22633fd0a558SYan, Zheng if (!next || next->level <= node->level) 22643fd0a558SYan, Zheng break; 22653fd0a558SYan, Zheng } 22663fd0a558SYan, Zheng 22673fd0a558SYan, Zheng if (!fs_root) 22683fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 22693fd0a558SYan, Zheng return fs_root; 22705d4f98a2SYan Zheng } 22715d4f98a2SYan Zheng 22725d4f98a2SYan Zheng static noinline_for_stack 22733fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 2274a26195a5SQu Wenruo struct btrfs_backref_node *node, int reserve) 22755d4f98a2SYan Zheng { 22760b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2277a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2278a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2279a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22803fd0a558SYan, Zheng u64 num_bytes = 0; 22813fd0a558SYan, Zheng int index = 0; 22825d4f98a2SYan Zheng 22833fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 22843fd0a558SYan, Zheng 22853fd0a558SYan, Zheng while (next) { 22863fd0a558SYan, Zheng cond_resched(); 22875d4f98a2SYan Zheng while (1) { 22883fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 22895d4f98a2SYan Zheng break; 22905d4f98a2SYan Zheng 22910b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 22923fd0a558SYan, Zheng 22933fd0a558SYan, Zheng if (list_empty(&next->upper)) 22943fd0a558SYan, Zheng break; 22953fd0a558SYan, Zheng 22963fd0a558SYan, Zheng edge = list_entry(next->upper.next, 2297a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 22983fd0a558SYan, Zheng edges[index++] = edge; 22993fd0a558SYan, Zheng next = edge->node[UPPER]; 23005d4f98a2SYan Zheng } 23013fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 23023fd0a558SYan, Zheng } 23033fd0a558SYan, Zheng return num_bytes; 23043fd0a558SYan, Zheng } 23053fd0a558SYan, Zheng 23063fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 23073fd0a558SYan, Zheng struct reloc_control *rc, 2308a26195a5SQu Wenruo struct btrfs_backref_node *node) 23093fd0a558SYan, Zheng { 23103fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2311da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23123fd0a558SYan, Zheng u64 num_bytes; 23133fd0a558SYan, Zheng int ret; 23140647bf56SWang Shilong u64 tmp; 23153fd0a558SYan, Zheng 23163fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 23173fd0a558SYan, Zheng 23183fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 23190647bf56SWang Shilong rc->reserved_bytes += num_bytes; 23208ca17f0fSJosef Bacik 23218ca17f0fSJosef Bacik /* 23228ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 23238ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 23248ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 23258ca17f0fSJosef Bacik */ 23260647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 23278ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 23283fd0a558SYan, Zheng if (ret) { 2329da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 23300647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 23310647bf56SWang Shilong tmp <<= 1; 23320647bf56SWang Shilong /* 23330647bf56SWang Shilong * only one thread can access block_rsv at this point, 23340647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 23350647bf56SWang Shilong * we expand more reservation size here to allow enough 233652042d8eSAndrea Gelmini * space for relocation and we will return earlier in 23370647bf56SWang Shilong * enospc case. 23380647bf56SWang Shilong */ 2339da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 23400647bf56SWang Shilong RELOCATION_RESERVED_NODES; 23418ca17f0fSJosef Bacik return -EAGAIN; 23423fd0a558SYan, Zheng } 23433fd0a558SYan, Zheng 23443fd0a558SYan, Zheng return 0; 23453fd0a558SYan, Zheng } 23463fd0a558SYan, Zheng 23475d4f98a2SYan Zheng /* 23485d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 23495d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 23505d4f98a2SYan Zheng * 23515d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 23525d4f98a2SYan Zheng * in that case this function just updates pointers. 23535d4f98a2SYan Zheng */ 23545d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 23553fd0a558SYan, Zheng struct reloc_control *rc, 2356a26195a5SQu Wenruo struct btrfs_backref_node *node, 23575d4f98a2SYan Zheng struct btrfs_key *key, 23585d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 23595d4f98a2SYan Zheng { 2360a26195a5SQu Wenruo struct btrfs_backref_node *upper; 2361a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2362a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 23635d4f98a2SYan Zheng struct btrfs_root *root; 23645d4f98a2SYan Zheng struct extent_buffer *eb; 23655d4f98a2SYan Zheng u32 blocksize; 23665d4f98a2SYan Zheng u64 bytenr; 23675d4f98a2SYan Zheng int slot; 23688df01fddSNikolay Borisov int ret = 0; 23695d4f98a2SYan Zheng 2370ffe30dd8SJosef Bacik /* 2371ffe30dd8SJosef Bacik * If we are lowest then this is the first time we're processing this 2372ffe30dd8SJosef Bacik * block, and thus shouldn't have an eb associated with it yet. 2373ffe30dd8SJosef Bacik */ 2374ffe30dd8SJosef Bacik ASSERT(!lowest || !node->eb); 23755d4f98a2SYan Zheng 23765d4f98a2SYan Zheng path->lowest_level = node->level + 1; 23773fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 23785d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 237982fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2380581c1760SQu Wenruo 23815d4f98a2SYan Zheng cond_resched(); 23825d4f98a2SYan Zheng 23835d4f98a2SYan Zheng upper = edge->node[UPPER]; 2384dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 2385cbdc2ebcSJosef Bacik if (IS_ERR(root)) { 2386cbdc2ebcSJosef Bacik ret = PTR_ERR(root); 2387cbdc2ebcSJosef Bacik goto next; 2388cbdc2ebcSJosef Bacik } 23895d4f98a2SYan Zheng 23903fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 23913fd0a558SYan, Zheng if (!lowest) { 2392e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 23938df01fddSNikolay Borisov if (ret < 0) 2394cbca7d59SFilipe Manana goto next; 23953fd0a558SYan, Zheng BUG_ON(ret); 23963fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 23973fd0a558SYan, Zheng if (node->eb->start == bytenr) 23983fd0a558SYan, Zheng goto next; 23993fd0a558SYan, Zheng } 2400b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 24013fd0a558SYan, Zheng } 24025d4f98a2SYan Zheng 24035d4f98a2SYan Zheng if (!upper->eb) { 24045d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 24053561b9dbSLiu Bo if (ret) { 24068df01fddSNikolay Borisov if (ret > 0) 24078df01fddSNikolay Borisov ret = -ENOENT; 24083561b9dbSLiu Bo 24093561b9dbSLiu Bo btrfs_release_path(path); 24105d4f98a2SYan Zheng break; 24115d4f98a2SYan Zheng } 24125d4f98a2SYan Zheng 24133fd0a558SYan, Zheng if (!upper->eb) { 24143fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 24153fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 24163fd0a558SYan, Zheng } else { 24173fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 24183fd0a558SYan, Zheng } 24193fd0a558SYan, Zheng 24203fd0a558SYan, Zheng upper->locked = 1; 24213fd0a558SYan, Zheng path->locks[upper->level] = 0; 24223fd0a558SYan, Zheng 24235d4f98a2SYan Zheng slot = path->slots[upper->level]; 2424b3b4aa74SDavid Sterba btrfs_release_path(path); 24255d4f98a2SYan Zheng } else { 2426e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 24278df01fddSNikolay Borisov if (ret < 0) 2428cbca7d59SFilipe Manana goto next; 24295d4f98a2SYan Zheng BUG_ON(ret); 24305d4f98a2SYan Zheng } 24315d4f98a2SYan Zheng 24325d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24333fd0a558SYan, Zheng if (lowest) { 24344547f4d8SLiu Bo if (bytenr != node->bytenr) { 24354547f4d8SLiu Bo btrfs_err(root->fs_info, 24364547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 24374547f4d8SLiu Bo bytenr, node->bytenr, slot, 24384547f4d8SLiu Bo upper->eb->start); 24398df01fddSNikolay Borisov ret = -EIO; 24404547f4d8SLiu Bo goto next; 24414547f4d8SLiu Bo } 24425d4f98a2SYan Zheng } else { 24433fd0a558SYan, Zheng if (node->eb->start == bytenr) 24443fd0a558SYan, Zheng goto next; 24455d4f98a2SYan Zheng } 24465d4f98a2SYan Zheng 2447da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 2448c9752536SJosef Bacik eb = btrfs_read_node_slot(upper->eb, slot); 244964c043deSLiu Bo if (IS_ERR(eb)) { 24508df01fddSNikolay Borisov ret = PTR_ERR(eb); 245164c043deSLiu Bo goto next; 245297d9a8a4STsutomu Itoh } 24535d4f98a2SYan Zheng btrfs_tree_lock(eb); 24545d4f98a2SYan Zheng 24555d4f98a2SYan Zheng if (!node->eb) { 24565d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 24579631e4ccSJosef Bacik slot, &eb, BTRFS_NESTING_COW); 24583fd0a558SYan, Zheng btrfs_tree_unlock(eb); 24593fd0a558SYan, Zheng free_extent_buffer(eb); 24608df01fddSNikolay Borisov if (ret < 0) 24613fd0a558SYan, Zheng goto next; 2462ffe30dd8SJosef Bacik /* 2463ffe30dd8SJosef Bacik * We've just COWed this block, it should have updated 2464ffe30dd8SJosef Bacik * the correct backref node entry. 2465ffe30dd8SJosef Bacik */ 2466ffe30dd8SJosef Bacik ASSERT(node->eb == eb); 24675d4f98a2SYan Zheng } else { 24685d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 24695d4f98a2SYan Zheng node->eb->start); 24705d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 24715d4f98a2SYan Zheng trans->transid); 24725d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 24735d4f98a2SYan Zheng 247482fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 24755d4f98a2SYan Zheng node->eb->start, blocksize, 247682fa113fSQu Wenruo upper->eb->start); 247782fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 247882fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 247982fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 248082fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 2481eb6b7fb4SJosef Bacik if (!ret) 2482eb6b7fb4SJosef Bacik ret = btrfs_drop_subtree(trans, root, eb, 2483eb6b7fb4SJosef Bacik upper->eb); 2484eb6b7fb4SJosef Bacik if (ret) 2485eb6b7fb4SJosef Bacik btrfs_abort_transaction(trans, ret); 24865d4f98a2SYan Zheng } 24873fd0a558SYan, Zheng next: 24883fd0a558SYan, Zheng if (!upper->pending) 2489b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 24903fd0a558SYan, Zheng else 2491b0fe7078SQu Wenruo btrfs_backref_unlock_node_buffer(upper); 24928df01fddSNikolay Borisov if (ret) 24933fd0a558SYan, Zheng break; 24945d4f98a2SYan Zheng } 24953fd0a558SYan, Zheng 24968df01fddSNikolay Borisov if (!ret && node->pending) { 2497b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 24983fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 24993fd0a558SYan, Zheng node->pending = 0; 25005d4f98a2SYan Zheng } 25013fd0a558SYan, Zheng 25025d4f98a2SYan Zheng path->lowest_level = 0; 2503ffe30dd8SJosef Bacik 2504ffe30dd8SJosef Bacik /* 2505ffe30dd8SJosef Bacik * We should have allocated all of our space in the block rsv and thus 2506ffe30dd8SJosef Bacik * shouldn't ENOSPC. 2507ffe30dd8SJosef Bacik */ 2508ffe30dd8SJosef Bacik ASSERT(ret != -ENOSPC); 25098df01fddSNikolay Borisov return ret; 25105d4f98a2SYan Zheng } 25115d4f98a2SYan Zheng 25125d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 25133fd0a558SYan, Zheng struct reloc_control *rc, 2514a26195a5SQu Wenruo struct btrfs_backref_node *node, 25155d4f98a2SYan Zheng struct btrfs_path *path) 25165d4f98a2SYan Zheng { 25175d4f98a2SYan Zheng struct btrfs_key key; 25185d4f98a2SYan Zheng 25195d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 25203fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 25215d4f98a2SYan Zheng } 25225d4f98a2SYan Zheng 25235d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 25243fd0a558SYan, Zheng struct reloc_control *rc, 25253fd0a558SYan, Zheng struct btrfs_path *path, int err) 25265d4f98a2SYan Zheng { 25273fd0a558SYan, Zheng LIST_HEAD(list); 2528a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 2529a26195a5SQu Wenruo struct btrfs_backref_node *node; 25305d4f98a2SYan Zheng int level; 25315d4f98a2SYan Zheng int ret; 25325d4f98a2SYan Zheng 25335d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 25345d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 25355d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 2536a26195a5SQu Wenruo struct btrfs_backref_node, list); 25373fd0a558SYan, Zheng list_move_tail(&node->list, &list); 25383fd0a558SYan, Zheng BUG_ON(!node->pending); 25395d4f98a2SYan Zheng 25403fd0a558SYan, Zheng if (!err) { 25413fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 25425d4f98a2SYan Zheng if (ret < 0) 25435d4f98a2SYan Zheng err = ret; 25445d4f98a2SYan Zheng } 25455d4f98a2SYan Zheng } 25463fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 25473fd0a558SYan, Zheng } 25485d4f98a2SYan Zheng return err; 25495d4f98a2SYan Zheng } 25505d4f98a2SYan Zheng 25515d4f98a2SYan Zheng /* 25525d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 25535d4f98a2SYan Zheng * as processed. 25545d4f98a2SYan Zheng */ 25555d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 2556a26195a5SQu Wenruo struct btrfs_backref_node *node) 25575d4f98a2SYan Zheng { 2558a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2559a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2560a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25615d4f98a2SYan Zheng int index = 0; 25625d4f98a2SYan Zheng 25635d4f98a2SYan Zheng while (next) { 25645d4f98a2SYan Zheng cond_resched(); 25655d4f98a2SYan Zheng while (1) { 25665d4f98a2SYan Zheng if (next->processed) 25675d4f98a2SYan Zheng break; 25685d4f98a2SYan Zheng 25699569cc20SQu Wenruo mark_block_processed(rc, next); 25705d4f98a2SYan Zheng 25715d4f98a2SYan Zheng if (list_empty(&next->upper)) 25725d4f98a2SYan Zheng break; 25735d4f98a2SYan Zheng 25745d4f98a2SYan Zheng edge = list_entry(next->upper.next, 2575a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 25765d4f98a2SYan Zheng edges[index++] = edge; 25775d4f98a2SYan Zheng next = edge->node[UPPER]; 25785d4f98a2SYan Zheng } 25795d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25805d4f98a2SYan Zheng } 25815d4f98a2SYan Zheng } 25825d4f98a2SYan Zheng 25837476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 25845d4f98a2SYan Zheng { 2585da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 25867476dfdaSDavid Sterba 25875d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 25889655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 25895d4f98a2SYan Zheng return 1; 25905d4f98a2SYan Zheng return 0; 25915d4f98a2SYan Zheng } 25925d4f98a2SYan Zheng 25932ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 25945d4f98a2SYan Zheng struct tree_block *block) 25955d4f98a2SYan Zheng { 25965d4f98a2SYan Zheng struct extent_buffer *eb; 25975d4f98a2SYan Zheng 2598f7ba2d37SJosef Bacik eb = read_tree_block(fs_info, block->bytenr, block->owner, 2599f7ba2d37SJosef Bacik block->key.offset, block->level, NULL); 260064c043deSLiu Bo if (IS_ERR(eb)) { 260164c043deSLiu Bo return PTR_ERR(eb); 260264c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2603416bc658SJosef Bacik free_extent_buffer(eb); 2604416bc658SJosef Bacik return -EIO; 2605416bc658SJosef Bacik } 26065d4f98a2SYan Zheng if (block->level == 0) 26075d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 26085d4f98a2SYan Zheng else 26095d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 26105d4f98a2SYan Zheng free_extent_buffer(eb); 26115d4f98a2SYan Zheng block->key_ready = 1; 26125d4f98a2SYan Zheng return 0; 26135d4f98a2SYan Zheng } 26145d4f98a2SYan Zheng 26155d4f98a2SYan Zheng /* 26165d4f98a2SYan Zheng * helper function to relocate a tree block 26175d4f98a2SYan Zheng */ 26185d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 26195d4f98a2SYan Zheng struct reloc_control *rc, 2620a26195a5SQu Wenruo struct btrfs_backref_node *node, 26215d4f98a2SYan Zheng struct btrfs_key *key, 26225d4f98a2SYan Zheng struct btrfs_path *path) 26235d4f98a2SYan Zheng { 26245d4f98a2SYan Zheng struct btrfs_root *root; 26253fd0a558SYan, Zheng int ret = 0; 26265d4f98a2SYan Zheng 26273fd0a558SYan, Zheng if (!node) 26285d4f98a2SYan Zheng return 0; 26293fd0a558SYan, Zheng 26305f6b2e5cSJosef Bacik /* 26315f6b2e5cSJosef Bacik * If we fail here we want to drop our backref_node because we are going 26325f6b2e5cSJosef Bacik * to start over and regenerate the tree for it. 26335f6b2e5cSJosef Bacik */ 26345f6b2e5cSJosef Bacik ret = reserve_metadata_space(trans, rc, node); 26355f6b2e5cSJosef Bacik if (ret) 26365f6b2e5cSJosef Bacik goto out; 26375f6b2e5cSJosef Bacik 26383fd0a558SYan, Zheng BUG_ON(node->processed); 2639147d256eSZhaolei root = select_one_root(node); 26408717cf44SJosef Bacik if (IS_ERR(root)) { 26418717cf44SJosef Bacik ret = PTR_ERR(root); 26428717cf44SJosef Bacik 26438717cf44SJosef Bacik /* See explanation in select_one_root for the -EUCLEAN case. */ 26448717cf44SJosef Bacik ASSERT(ret == -ENOENT); 26458717cf44SJosef Bacik if (ret == -ENOENT) { 26468717cf44SJosef Bacik ret = 0; 26473fd0a558SYan, Zheng update_processed_blocks(rc, node); 26488717cf44SJosef Bacik } 26493fd0a558SYan, Zheng goto out; 26505d4f98a2SYan Zheng } 26515d4f98a2SYan Zheng 26523fd0a558SYan, Zheng if (root) { 265392a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 26541c7bfa15SJosef Bacik /* 26551c7bfa15SJosef Bacik * This block was the root block of a root, and this is 26561c7bfa15SJosef Bacik * the first time we're processing the block and thus it 26571c7bfa15SJosef Bacik * should not have had the ->new_bytenr modified and 26581c7bfa15SJosef Bacik * should have not been included on the changed list. 26591c7bfa15SJosef Bacik * 26601c7bfa15SJosef Bacik * However in the case of corruption we could have 26611c7bfa15SJosef Bacik * multiple refs pointing to the same block improperly, 26621c7bfa15SJosef Bacik * and thus we would trip over these checks. ASSERT() 26631c7bfa15SJosef Bacik * for the developer case, because it could indicate a 26641c7bfa15SJosef Bacik * bug in the backref code, however error out for a 26651c7bfa15SJosef Bacik * normal user in the case of corruption. 26661c7bfa15SJosef Bacik */ 26671c7bfa15SJosef Bacik ASSERT(node->new_bytenr == 0); 26681c7bfa15SJosef Bacik ASSERT(list_empty(&node->list)); 26691c7bfa15SJosef Bacik if (node->new_bytenr || !list_empty(&node->list)) { 26701c7bfa15SJosef Bacik btrfs_err(root->fs_info, 26711c7bfa15SJosef Bacik "bytenr %llu has improper references to it", 26721c7bfa15SJosef Bacik node->bytenr); 26731c7bfa15SJosef Bacik ret = -EUCLEAN; 26741c7bfa15SJosef Bacik goto out; 26751c7bfa15SJosef Bacik } 2676d18c7bd9SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 2677d18c7bd9SJosef Bacik if (ret) 2678d18c7bd9SJosef Bacik goto out; 267939200e59SJosef Bacik /* 268039200e59SJosef Bacik * Another thread could have failed, need to check if we 268139200e59SJosef Bacik * have reloc_root actually set. 268239200e59SJosef Bacik */ 268339200e59SJosef Bacik if (!root->reloc_root) { 268439200e59SJosef Bacik ret = -ENOENT; 268539200e59SJosef Bacik goto out; 268639200e59SJosef Bacik } 26873fd0a558SYan, Zheng root = root->reloc_root; 26883fd0a558SYan, Zheng node->new_bytenr = root->node->start; 268900246528SJosef Bacik btrfs_put_root(node->root); 269000246528SJosef Bacik node->root = btrfs_grab_root(root); 26910b530bc5SJosef Bacik ASSERT(node->root); 26923fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 26933fd0a558SYan, Zheng } else { 26945d4f98a2SYan Zheng path->lowest_level = node->level; 26955d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2696b3b4aa74SDavid Sterba btrfs_release_path(path); 26973fd0a558SYan, Zheng if (ret > 0) 26985d4f98a2SYan Zheng ret = 0; 26993fd0a558SYan, Zheng } 27003fd0a558SYan, Zheng if (!ret) 27013fd0a558SYan, Zheng update_processed_blocks(rc, node); 27023fd0a558SYan, Zheng } else { 27033fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 27043fd0a558SYan, Zheng } 27055d4f98a2SYan Zheng out: 27060647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 2707023acb07SQu Wenruo btrfs_backref_cleanup_node(&rc->backref_cache, node); 27085d4f98a2SYan Zheng return ret; 27095d4f98a2SYan Zheng } 27105d4f98a2SYan Zheng 27115d4f98a2SYan Zheng /* 27125d4f98a2SYan Zheng * relocate a list of blocks 27135d4f98a2SYan Zheng */ 27145d4f98a2SYan Zheng static noinline_for_stack 27155d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 27165d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 27175d4f98a2SYan Zheng { 27182ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2719a26195a5SQu Wenruo struct btrfs_backref_node *node; 27205d4f98a2SYan Zheng struct btrfs_path *path; 27215d4f98a2SYan Zheng struct tree_block *block; 272298ff7b94SQu Wenruo struct tree_block *next; 27235d4f98a2SYan Zheng int ret; 27245d4f98a2SYan Zheng int err = 0; 27255d4f98a2SYan Zheng 27265d4f98a2SYan Zheng path = btrfs_alloc_path(); 2727e1a12670SLiu Bo if (!path) { 2728e1a12670SLiu Bo err = -ENOMEM; 272934c2b290SDavid Sterba goto out_free_blocks; 2730e1a12670SLiu Bo } 27315d4f98a2SYan Zheng 273298ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 273398ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27345d4f98a2SYan Zheng if (!block->key_ready) 2735f7ba2d37SJosef Bacik btrfs_readahead_tree_block(fs_info, block->bytenr, 2736f7ba2d37SJosef Bacik block->owner, 0, 27373fbaf258SJosef Bacik block->level); 27385d4f98a2SYan Zheng } 27395d4f98a2SYan Zheng 274098ff7b94SQu Wenruo /* Get first keys */ 274198ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 274234c2b290SDavid Sterba if (!block->key_ready) { 27432ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 274434c2b290SDavid Sterba if (err) 274534c2b290SDavid Sterba goto out_free_path; 274634c2b290SDavid Sterba } 27475d4f98a2SYan Zheng } 27485d4f98a2SYan Zheng 274998ff7b94SQu Wenruo /* Do tree relocation */ 275098ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27513fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 27525d4f98a2SYan Zheng block->level, block->bytenr); 27535d4f98a2SYan Zheng if (IS_ERR(node)) { 27545d4f98a2SYan Zheng err = PTR_ERR(node); 27555d4f98a2SYan Zheng goto out; 27565d4f98a2SYan Zheng } 27575d4f98a2SYan Zheng 27585d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 27595d4f98a2SYan Zheng path); 27605d4f98a2SYan Zheng if (ret < 0) { 27615d4f98a2SYan Zheng err = ret; 276250dbbb71SJosef Bacik break; 27635d4f98a2SYan Zheng } 27645d4f98a2SYan Zheng } 27655d4f98a2SYan Zheng out: 27663fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 27675d4f98a2SYan Zheng 276834c2b290SDavid Sterba out_free_path: 27695d4f98a2SYan Zheng btrfs_free_path(path); 277034c2b290SDavid Sterba out_free_blocks: 2771e1a12670SLiu Bo free_block_list(blocks); 27725d4f98a2SYan Zheng return err; 27735d4f98a2SYan Zheng } 27745d4f98a2SYan Zheng 2775056d9becSNikolay Borisov static noinline_for_stack int prealloc_file_extent_cluster( 2776056d9becSNikolay Borisov struct btrfs_inode *inode, 2777efa56464SYan, Zheng struct file_extent_cluster *cluster) 2778efa56464SYan, Zheng { 2779efa56464SYan, Zheng u64 alloc_hint = 0; 2780efa56464SYan, Zheng u64 start; 2781efa56464SYan, Zheng u64 end; 2782056d9becSNikolay Borisov u64 offset = inode->index_cnt; 2783efa56464SYan, Zheng u64 num_bytes; 27844e9d0d01SNikolay Borisov int nr; 2785efa56464SYan, Zheng int ret = 0; 27869d9ea1e6SQu Wenruo u64 i_size = i_size_read(&inode->vfs_inode); 2787dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 2788dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 2789214e61d0SNikolay Borisov u64 cur_offset = prealloc_start; 2790efa56464SYan, Zheng 27919d9ea1e6SQu Wenruo /* 27929d9ea1e6SQu Wenruo * For subpage case, previous i_size may not be aligned to PAGE_SIZE. 27939d9ea1e6SQu Wenruo * This means the range [i_size, PAGE_END + 1) is filled with zeros by 27949d9ea1e6SQu Wenruo * btrfs_do_readpage() call of previously relocated file cluster. 27959d9ea1e6SQu Wenruo * 27969d9ea1e6SQu Wenruo * If the current cluster starts in the above range, btrfs_do_readpage() 27979d9ea1e6SQu Wenruo * will skip the read, and relocate_one_page() will later writeback 27989d9ea1e6SQu Wenruo * the padding zeros as new data, causing data corruption. 27999d9ea1e6SQu Wenruo * 28009d9ea1e6SQu Wenruo * Here we have to manually invalidate the range (i_size, PAGE_END + 1). 28019d9ea1e6SQu Wenruo */ 28029d9ea1e6SQu Wenruo if (!IS_ALIGNED(i_size, PAGE_SIZE)) { 28039d9ea1e6SQu Wenruo struct address_space *mapping = inode->vfs_inode.i_mapping; 28049d9ea1e6SQu Wenruo struct btrfs_fs_info *fs_info = inode->root->fs_info; 28059d9ea1e6SQu Wenruo const u32 sectorsize = fs_info->sectorsize; 28069d9ea1e6SQu Wenruo struct page *page; 28079d9ea1e6SQu Wenruo 28089d9ea1e6SQu Wenruo ASSERT(sectorsize < PAGE_SIZE); 28099d9ea1e6SQu Wenruo ASSERT(IS_ALIGNED(i_size, sectorsize)); 28109d9ea1e6SQu Wenruo 28119d9ea1e6SQu Wenruo /* 28129d9ea1e6SQu Wenruo * Subpage can't handle page with DIRTY but without UPTODATE 28139d9ea1e6SQu Wenruo * bit as it can lead to the following deadlock: 28149d9ea1e6SQu Wenruo * 28159d9ea1e6SQu Wenruo * btrfs_readpage() 28169d9ea1e6SQu Wenruo * | Page already *locked* 28179d9ea1e6SQu Wenruo * |- btrfs_lock_and_flush_ordered_range() 28189d9ea1e6SQu Wenruo * |- btrfs_start_ordered_extent() 28199d9ea1e6SQu Wenruo * |- extent_write_cache_pages() 28209d9ea1e6SQu Wenruo * |- lock_page() 28219d9ea1e6SQu Wenruo * We try to lock the page we already hold. 28229d9ea1e6SQu Wenruo * 28239d9ea1e6SQu Wenruo * Here we just writeback the whole data reloc inode, so that 28249d9ea1e6SQu Wenruo * we will be ensured to have no dirty range in the page, and 28259d9ea1e6SQu Wenruo * are safe to clear the uptodate bits. 28269d9ea1e6SQu Wenruo * 28279d9ea1e6SQu Wenruo * This shouldn't cause too much overhead, as we need to write 28289d9ea1e6SQu Wenruo * the data back anyway. 28299d9ea1e6SQu Wenruo */ 28309d9ea1e6SQu Wenruo ret = filemap_write_and_wait(mapping); 28319d9ea1e6SQu Wenruo if (ret < 0) 28329d9ea1e6SQu Wenruo return ret; 28339d9ea1e6SQu Wenruo 28349d9ea1e6SQu Wenruo clear_extent_bits(&inode->io_tree, i_size, 28359d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - 1, 28369d9ea1e6SQu Wenruo EXTENT_UPTODATE); 28379d9ea1e6SQu Wenruo page = find_lock_page(mapping, i_size >> PAGE_SHIFT); 28389d9ea1e6SQu Wenruo /* 28399d9ea1e6SQu Wenruo * If page is freed we don't need to do anything then, as we 28409d9ea1e6SQu Wenruo * will re-read the whole page anyway. 28419d9ea1e6SQu Wenruo */ 28429d9ea1e6SQu Wenruo if (page) { 28439d9ea1e6SQu Wenruo btrfs_subpage_clear_uptodate(fs_info, page, i_size, 28449d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - i_size); 28459d9ea1e6SQu Wenruo unlock_page(page); 28469d9ea1e6SQu Wenruo put_page(page); 28479d9ea1e6SQu Wenruo } 28489d9ea1e6SQu Wenruo } 28499d9ea1e6SQu Wenruo 2850efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 2851056d9becSNikolay Borisov ret = btrfs_alloc_data_chunk_ondemand(inode, 2852dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 2853efa56464SYan, Zheng if (ret) 2854214e61d0SNikolay Borisov return ret; 2855efa56464SYan, Zheng 285664708539SJosef Bacik btrfs_inode_lock(&inode->vfs_inode, 0); 28574e9d0d01SNikolay Borisov for (nr = 0; nr < cluster->nr; nr++) { 2858efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 2859efa56464SYan, Zheng if (nr + 1 < cluster->nr) 2860efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 2861efa56464SYan, Zheng else 2862efa56464SYan, Zheng end = cluster->end - offset; 2863efa56464SYan, Zheng 2864056d9becSNikolay Borisov lock_extent(&inode->io_tree, start, end); 2865efa56464SYan, Zheng num_bytes = end + 1 - start; 2866056d9becSNikolay Borisov ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start, 2867efa56464SYan, Zheng num_bytes, num_bytes, 2868efa56464SYan, Zheng end + 1, &alloc_hint); 286918513091SWang Xiaoguang cur_offset = end + 1; 2870056d9becSNikolay Borisov unlock_extent(&inode->io_tree, start, end); 2871efa56464SYan, Zheng if (ret) 2872efa56464SYan, Zheng break; 2873efa56464SYan, Zheng } 287464708539SJosef Bacik btrfs_inode_unlock(&inode->vfs_inode, 0); 2875214e61d0SNikolay Borisov 287618513091SWang Xiaoguang if (cur_offset < prealloc_end) 2877056d9becSNikolay Borisov btrfs_free_reserved_data_space_noquota(inode->root->fs_info, 2878a89ef455SFilipe Manana prealloc_end + 1 - cur_offset); 2879efa56464SYan, Zheng return ret; 2880efa56464SYan, Zheng } 2881efa56464SYan, Zheng 2882efa56464SYan, Zheng static noinline_for_stack 28830257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 28840257bb82SYan, Zheng u64 block_start) 28855d4f98a2SYan Zheng { 28865d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 28875d4f98a2SYan Zheng struct extent_map *em; 28880257bb82SYan, Zheng int ret = 0; 28895d4f98a2SYan Zheng 2890172ddd60SDavid Sterba em = alloc_extent_map(); 28910257bb82SYan, Zheng if (!em) 28920257bb82SYan, Zheng return -ENOMEM; 28930257bb82SYan, Zheng 28945d4f98a2SYan Zheng em->start = start; 28950257bb82SYan, Zheng em->len = end + 1 - start; 28960257bb82SYan, Zheng em->block_len = em->len; 28970257bb82SYan, Zheng em->block_start = block_start; 28985d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 28995d4f98a2SYan Zheng 2900d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 29015d4f98a2SYan Zheng while (1) { 2902890871beSChris Mason write_lock(&em_tree->lock); 290309a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2904890871beSChris Mason write_unlock(&em_tree->lock); 29055d4f98a2SYan Zheng if (ret != -EEXIST) { 29065d4f98a2SYan Zheng free_extent_map(em); 29075d4f98a2SYan Zheng break; 29085d4f98a2SYan Zheng } 2909dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 29105d4f98a2SYan Zheng } 2911d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 29120257bb82SYan, Zheng return ret; 29130257bb82SYan, Zheng } 29145d4f98a2SYan Zheng 2915726a3421SQu Wenruo /* 2916907d2710SDavid Sterba * Allow error injection to test balance/relocation cancellation 2917726a3421SQu Wenruo */ 29181fec12a5SJosef Bacik noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 2919726a3421SQu Wenruo { 29205cb502f4SQu Wenruo return atomic_read(&fs_info->balance_cancel_req) || 2921907d2710SDavid Sterba atomic_read(&fs_info->reloc_cancel_req) || 29225cb502f4SQu Wenruo fatal_signal_pending(current); 2923726a3421SQu Wenruo } 2924726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 2925726a3421SQu Wenruo 2926c2832898SQu Wenruo static u64 get_cluster_boundary_end(struct file_extent_cluster *cluster, 2927c2832898SQu Wenruo int cluster_nr) 2928c2832898SQu Wenruo { 2929c2832898SQu Wenruo /* Last extent, use cluster end directly */ 2930c2832898SQu Wenruo if (cluster_nr >= cluster->nr - 1) 2931c2832898SQu Wenruo return cluster->end; 2932c2832898SQu Wenruo 2933c2832898SQu Wenruo /* Use next boundary start*/ 2934c2832898SQu Wenruo return cluster->boundary[cluster_nr + 1] - 1; 2935c2832898SQu Wenruo } 2936c2832898SQu Wenruo 2937f47960f4SQu Wenruo static int relocate_one_page(struct inode *inode, struct file_ra_state *ra, 2938f47960f4SQu Wenruo struct file_extent_cluster *cluster, 2939f47960f4SQu Wenruo int *cluster_nr, unsigned long page_index) 2940f47960f4SQu Wenruo { 2941f47960f4SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 2942f47960f4SQu Wenruo u64 offset = BTRFS_I(inode)->index_cnt; 2943f47960f4SQu Wenruo const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT; 2944f47960f4SQu Wenruo gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 2945f47960f4SQu Wenruo struct page *page; 2946f47960f4SQu Wenruo u64 page_start; 2947f47960f4SQu Wenruo u64 page_end; 2948c2832898SQu Wenruo u64 cur; 2949f47960f4SQu Wenruo int ret; 2950f47960f4SQu Wenruo 2951f47960f4SQu Wenruo ASSERT(page_index <= last_index); 2952f47960f4SQu Wenruo page = find_lock_page(inode->i_mapping, page_index); 2953f47960f4SQu Wenruo if (!page) { 2954f47960f4SQu Wenruo page_cache_sync_readahead(inode->i_mapping, ra, NULL, 2955f47960f4SQu Wenruo page_index, last_index + 1 - page_index); 2956f47960f4SQu Wenruo page = find_or_create_page(inode->i_mapping, page_index, mask); 2957c2832898SQu Wenruo if (!page) 2958c2832898SQu Wenruo return -ENOMEM; 2959f47960f4SQu Wenruo } 2960f47960f4SQu Wenruo ret = set_page_extent_mapped(page); 2961f47960f4SQu Wenruo if (ret < 0) 2962f47960f4SQu Wenruo goto release_page; 2963f47960f4SQu Wenruo 2964f47960f4SQu Wenruo if (PageReadahead(page)) 2965f47960f4SQu Wenruo page_cache_async_readahead(inode->i_mapping, ra, NULL, page, 2966f47960f4SQu Wenruo page_index, last_index + 1 - page_index); 2967f47960f4SQu Wenruo 2968f47960f4SQu Wenruo if (!PageUptodate(page)) { 2969f47960f4SQu Wenruo btrfs_readpage(NULL, page); 2970f47960f4SQu Wenruo lock_page(page); 2971f47960f4SQu Wenruo if (!PageUptodate(page)) { 2972f47960f4SQu Wenruo ret = -EIO; 2973f47960f4SQu Wenruo goto release_page; 2974f47960f4SQu Wenruo } 2975f47960f4SQu Wenruo } 2976f47960f4SQu Wenruo 2977f47960f4SQu Wenruo page_start = page_offset(page); 2978f47960f4SQu Wenruo page_end = page_start + PAGE_SIZE - 1; 2979f47960f4SQu Wenruo 2980c2832898SQu Wenruo /* 2981c2832898SQu Wenruo * Start from the cluster, as for subpage case, the cluster can start 2982c2832898SQu Wenruo * inside the page. 2983c2832898SQu Wenruo */ 2984c2832898SQu Wenruo cur = max(page_start, cluster->boundary[*cluster_nr] - offset); 2985c2832898SQu Wenruo while (cur <= page_end) { 2986c2832898SQu Wenruo u64 extent_start = cluster->boundary[*cluster_nr] - offset; 2987c2832898SQu Wenruo u64 extent_end = get_cluster_boundary_end(cluster, 2988c2832898SQu Wenruo *cluster_nr) - offset; 2989c2832898SQu Wenruo u64 clamped_start = max(page_start, extent_start); 2990c2832898SQu Wenruo u64 clamped_end = min(page_end, extent_end); 2991c2832898SQu Wenruo u32 clamped_len = clamped_end + 1 - clamped_start; 2992f47960f4SQu Wenruo 2993c2832898SQu Wenruo /* Reserve metadata for this range */ 2994c2832898SQu Wenruo ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 2995c2832898SQu Wenruo clamped_len); 2996c2832898SQu Wenruo if (ret) 2997f47960f4SQu Wenruo goto release_page; 2998f47960f4SQu Wenruo 2999c2832898SQu Wenruo /* Mark the range delalloc and dirty for later writeback */ 3000c2832898SQu Wenruo lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end); 3001c2832898SQu Wenruo ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start, 3002c2832898SQu Wenruo clamped_end, 0, NULL); 3003c2832898SQu Wenruo if (ret) { 3004c2832898SQu Wenruo clear_extent_bits(&BTRFS_I(inode)->io_tree, 3005c2832898SQu Wenruo clamped_start, clamped_end, 3006c2832898SQu Wenruo EXTENT_LOCKED | EXTENT_BOUNDARY); 3007c2832898SQu Wenruo btrfs_delalloc_release_metadata(BTRFS_I(inode), 3008c2832898SQu Wenruo clamped_len, true); 3009c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), 3010c2832898SQu Wenruo clamped_len); 3011c2832898SQu Wenruo goto release_page; 3012f47960f4SQu Wenruo } 3013c2832898SQu Wenruo btrfs_page_set_dirty(fs_info, page, clamped_start, clamped_len); 3014f47960f4SQu Wenruo 3015c2832898SQu Wenruo /* 3016c2832898SQu Wenruo * Set the boundary if it's inside the page. 3017c2832898SQu Wenruo * Data relocation requires the destination extents to have the 3018c2832898SQu Wenruo * same size as the source. 3019c2832898SQu Wenruo * EXTENT_BOUNDARY bit prevents current extent from being merged 3020c2832898SQu Wenruo * with previous extent. 3021c2832898SQu Wenruo */ 3022c2832898SQu Wenruo if (in_range(cluster->boundary[*cluster_nr] - offset, 3023c2832898SQu Wenruo page_start, PAGE_SIZE)) { 3024c2832898SQu Wenruo u64 boundary_start = cluster->boundary[*cluster_nr] - 3025c2832898SQu Wenruo offset; 3026c2832898SQu Wenruo u64 boundary_end = boundary_start + 3027c2832898SQu Wenruo fs_info->sectorsize - 1; 3028c2832898SQu Wenruo 3029c2832898SQu Wenruo set_extent_bits(&BTRFS_I(inode)->io_tree, 3030c2832898SQu Wenruo boundary_start, boundary_end, 3031c2832898SQu Wenruo EXTENT_BOUNDARY); 3032c2832898SQu Wenruo } 3033c2832898SQu Wenruo unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end); 3034c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len); 3035c2832898SQu Wenruo cur += clamped_len; 3036c2832898SQu Wenruo 3037c2832898SQu Wenruo /* Crossed extent end, go to next extent */ 3038c2832898SQu Wenruo if (cur >= extent_end) { 3039c2832898SQu Wenruo (*cluster_nr)++; 3040c2832898SQu Wenruo /* Just finished the last extent of the cluster, exit. */ 3041c2832898SQu Wenruo if (*cluster_nr >= cluster->nr) 3042c2832898SQu Wenruo break; 3043c2832898SQu Wenruo } 3044c2832898SQu Wenruo } 3045f47960f4SQu Wenruo unlock_page(page); 3046f47960f4SQu Wenruo put_page(page); 3047f47960f4SQu Wenruo 3048f47960f4SQu Wenruo balance_dirty_pages_ratelimited(inode->i_mapping); 3049f47960f4SQu Wenruo btrfs_throttle(fs_info); 3050f47960f4SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) 3051f47960f4SQu Wenruo ret = -ECANCELED; 3052f47960f4SQu Wenruo return ret; 3053f47960f4SQu Wenruo 3054f47960f4SQu Wenruo release_page: 3055f47960f4SQu Wenruo unlock_page(page); 3056f47960f4SQu Wenruo put_page(page); 3057f47960f4SQu Wenruo return ret; 3058f47960f4SQu Wenruo } 3059f47960f4SQu Wenruo 30600257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 30610257bb82SYan, Zheng struct file_extent_cluster *cluster) 30620257bb82SYan, Zheng { 30630257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 30640257bb82SYan, Zheng unsigned long index; 30650257bb82SYan, Zheng unsigned long last_index; 30660257bb82SYan, Zheng struct file_ra_state *ra; 3067f47960f4SQu Wenruo int cluster_nr = 0; 30680257bb82SYan, Zheng int ret = 0; 30690257bb82SYan, Zheng 30700257bb82SYan, Zheng if (!cluster->nr) 30710257bb82SYan, Zheng return 0; 30720257bb82SYan, Zheng 30730257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 30740257bb82SYan, Zheng if (!ra) 30750257bb82SYan, Zheng return -ENOMEM; 30760257bb82SYan, Zheng 3077056d9becSNikolay Borisov ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster); 30780257bb82SYan, Zheng if (ret) 3079efa56464SYan, Zheng goto out; 30800257bb82SYan, Zheng 30810257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 30820257bb82SYan, Zheng 3083efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3084efa56464SYan, Zheng cluster->end - offset, cluster->start); 3085efa56464SYan, Zheng if (ret) 3086efa56464SYan, Zheng goto out; 3087efa56464SYan, Zheng 308809cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 3089f47960f4SQu Wenruo for (index = (cluster->start - offset) >> PAGE_SHIFT; 3090f47960f4SQu Wenruo index <= last_index && !ret; index++) 3091f47960f4SQu Wenruo ret = relocate_one_page(inode, ra, cluster, &cluster_nr, index); 3092f47960f4SQu Wenruo if (ret == 0) 3093f47960f4SQu Wenruo WARN_ON(cluster_nr != cluster->nr); 3094efa56464SYan, Zheng out: 30950257bb82SYan, Zheng kfree(ra); 30960257bb82SYan, Zheng return ret; 30970257bb82SYan, Zheng } 30980257bb82SYan, Zheng 30990257bb82SYan, Zheng static noinline_for_stack 31000257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 31010257bb82SYan, Zheng struct file_extent_cluster *cluster) 31020257bb82SYan, Zheng { 31030257bb82SYan, Zheng int ret; 31040257bb82SYan, Zheng 31050257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 31060257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31070257bb82SYan, Zheng if (ret) 31080257bb82SYan, Zheng return ret; 31090257bb82SYan, Zheng cluster->nr = 0; 31100257bb82SYan, Zheng } 31110257bb82SYan, Zheng 31120257bb82SYan, Zheng if (!cluster->nr) 31130257bb82SYan, Zheng cluster->start = extent_key->objectid; 31140257bb82SYan, Zheng else 31150257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 31160257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 31170257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 31180257bb82SYan, Zheng cluster->nr++; 31190257bb82SYan, Zheng 31200257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 31210257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31220257bb82SYan, Zheng if (ret) 31230257bb82SYan, Zheng return ret; 31240257bb82SYan, Zheng cluster->nr = 0; 31250257bb82SYan, Zheng } 31260257bb82SYan, Zheng return 0; 31275d4f98a2SYan Zheng } 31285d4f98a2SYan Zheng 31295d4f98a2SYan Zheng /* 31305d4f98a2SYan Zheng * helper to add a tree block to the list. 31315d4f98a2SYan Zheng * the major work is getting the generation and level of the block 31325d4f98a2SYan Zheng */ 31335d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 31345d4f98a2SYan Zheng struct btrfs_key *extent_key, 31355d4f98a2SYan Zheng struct btrfs_path *path, 31365d4f98a2SYan Zheng struct rb_root *blocks) 31375d4f98a2SYan Zheng { 31385d4f98a2SYan Zheng struct extent_buffer *eb; 31395d4f98a2SYan Zheng struct btrfs_extent_item *ei; 31405d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 31415d4f98a2SYan Zheng struct tree_block *block; 31425d4f98a2SYan Zheng struct rb_node *rb_node; 31435d4f98a2SYan Zheng u32 item_size; 31445d4f98a2SYan Zheng int level = -1; 31457fdf4b60SWang Shilong u64 generation; 3146f7ba2d37SJosef Bacik u64 owner = 0; 31475d4f98a2SYan Zheng 31485d4f98a2SYan Zheng eb = path->nodes[0]; 31495d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 31505d4f98a2SYan Zheng 31513173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 31523173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 3153f7ba2d37SJosef Bacik unsigned long ptr = 0, end; 3154f7ba2d37SJosef Bacik 31555d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 31565d4f98a2SYan Zheng struct btrfs_extent_item); 3157f7ba2d37SJosef Bacik end = (unsigned long)ei + item_size; 31583173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 31595d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 31605d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 3161f7ba2d37SJosef Bacik ptr = (unsigned long)(bi + 1); 31625d4f98a2SYan Zheng } else { 31633173a18fSJosef Bacik level = (int)extent_key->offset; 3164f7ba2d37SJosef Bacik ptr = (unsigned long)(ei + 1); 31653173a18fSJosef Bacik } 31663173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 3167f7ba2d37SJosef Bacik 3168f7ba2d37SJosef Bacik /* 3169f7ba2d37SJosef Bacik * We're reading random blocks without knowing their owner ahead 3170f7ba2d37SJosef Bacik * of time. This is ok most of the time, as all reloc roots and 3171f7ba2d37SJosef Bacik * fs roots have the same lock type. However normal trees do 3172f7ba2d37SJosef Bacik * not, and the only way to know ahead of time is to read the 3173f7ba2d37SJosef Bacik * inline ref offset. We know it's an fs root if 3174f7ba2d37SJosef Bacik * 3175f7ba2d37SJosef Bacik * 1. There's more than one ref. 3176f7ba2d37SJosef Bacik * 2. There's a SHARED_DATA_REF_KEY set. 3177f7ba2d37SJosef Bacik * 3. FULL_BACKREF is set on the flags. 3178f7ba2d37SJosef Bacik * 3179f7ba2d37SJosef Bacik * Otherwise it's safe to assume that the ref offset == the 3180f7ba2d37SJosef Bacik * owner of this block, so we can use that when calling 3181f7ba2d37SJosef Bacik * read_tree_block. 3182f7ba2d37SJosef Bacik */ 3183f7ba2d37SJosef Bacik if (btrfs_extent_refs(eb, ei) == 1 && 3184f7ba2d37SJosef Bacik !(btrfs_extent_flags(eb, ei) & 3185f7ba2d37SJosef Bacik BTRFS_BLOCK_FLAG_FULL_BACKREF) && 3186f7ba2d37SJosef Bacik ptr < end) { 3187f7ba2d37SJosef Bacik struct btrfs_extent_inline_ref *iref; 3188f7ba2d37SJosef Bacik int type; 3189f7ba2d37SJosef Bacik 3190f7ba2d37SJosef Bacik iref = (struct btrfs_extent_inline_ref *)ptr; 3191f7ba2d37SJosef Bacik type = btrfs_get_extent_inline_ref_type(eb, iref, 3192f7ba2d37SJosef Bacik BTRFS_REF_TYPE_BLOCK); 3193f7ba2d37SJosef Bacik if (type == BTRFS_REF_TYPE_INVALID) 3194f7ba2d37SJosef Bacik return -EINVAL; 3195f7ba2d37SJosef Bacik if (type == BTRFS_TREE_BLOCK_REF_KEY) 3196f7ba2d37SJosef Bacik owner = btrfs_extent_inline_ref_offset(eb, iref); 3197f7ba2d37SJosef Bacik } 31986d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3199ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3200ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3201ba3c2b19SNikolay Borisov return -EINVAL; 32023173a18fSJosef Bacik } else { 32035d4f98a2SYan Zheng BUG(); 32045d4f98a2SYan Zheng } 32055d4f98a2SYan Zheng 3206b3b4aa74SDavid Sterba btrfs_release_path(path); 32075d4f98a2SYan Zheng 32085d4f98a2SYan Zheng BUG_ON(level == -1); 32095d4f98a2SYan Zheng 32105d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 32115d4f98a2SYan Zheng if (!block) 32125d4f98a2SYan Zheng return -ENOMEM; 32135d4f98a2SYan Zheng 32145d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3215da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 32165d4f98a2SYan Zheng block->key.offset = generation; 32175d4f98a2SYan Zheng block->level = level; 32185d4f98a2SYan Zheng block->key_ready = 0; 3219f7ba2d37SJosef Bacik block->owner = owner; 32205d4f98a2SYan Zheng 3221e9a28dc5SQu Wenruo rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node); 322243c04fb1SJeff Mahoney if (rb_node) 3223982c92cbSQu Wenruo btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr, 3224982c92cbSQu Wenruo -EEXIST); 32255d4f98a2SYan Zheng 32265d4f98a2SYan Zheng return 0; 32275d4f98a2SYan Zheng } 32285d4f98a2SYan Zheng 32295d4f98a2SYan Zheng /* 32305d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 32315d4f98a2SYan Zheng */ 32325d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 32335d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 32345d4f98a2SYan Zheng struct rb_root *blocks) 32355d4f98a2SYan Zheng { 32360b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 32375d4f98a2SYan Zheng struct btrfs_path *path; 32385d4f98a2SYan Zheng struct btrfs_key key; 32395d4f98a2SYan Zheng int ret; 32400b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 32415d4f98a2SYan Zheng 32427476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 32435d4f98a2SYan Zheng return 0; 32445d4f98a2SYan Zheng 3245e9a28dc5SQu Wenruo if (rb_simple_search(blocks, bytenr)) 32465d4f98a2SYan Zheng return 0; 32475d4f98a2SYan Zheng 32485d4f98a2SYan Zheng path = btrfs_alloc_path(); 32495d4f98a2SYan Zheng if (!path) 32505d4f98a2SYan Zheng return -ENOMEM; 3251aee68ee5SJosef Bacik again: 32525d4f98a2SYan Zheng key.objectid = bytenr; 3253aee68ee5SJosef Bacik if (skinny) { 3254aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3255aee68ee5SJosef Bacik key.offset = (u64)-1; 3256aee68ee5SJosef Bacik } else { 32575d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 32585d4f98a2SYan Zheng key.offset = blocksize; 3259aee68ee5SJosef Bacik } 32605d4f98a2SYan Zheng 32615d4f98a2SYan Zheng path->search_commit_root = 1; 32625d4f98a2SYan Zheng path->skip_locking = 1; 32635d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 32645d4f98a2SYan Zheng if (ret < 0) 32655d4f98a2SYan Zheng goto out; 32665d4f98a2SYan Zheng 3267aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3268aee68ee5SJosef Bacik if (path->slots[0]) { 3269aee68ee5SJosef Bacik path->slots[0]--; 3270aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3271aee68ee5SJosef Bacik path->slots[0]); 32723173a18fSJosef Bacik if (key.objectid == bytenr && 3273aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3274aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3275aee68ee5SJosef Bacik key.offset == blocksize))) 32763173a18fSJosef Bacik ret = 0; 32773173a18fSJosef Bacik } 3278aee68ee5SJosef Bacik 3279aee68ee5SJosef Bacik if (ret) { 3280aee68ee5SJosef Bacik skinny = false; 3281aee68ee5SJosef Bacik btrfs_release_path(path); 3282aee68ee5SJosef Bacik goto again; 3283aee68ee5SJosef Bacik } 3284aee68ee5SJosef Bacik } 3285cdccee99SLiu Bo if (ret) { 3286cdccee99SLiu Bo ASSERT(ret == 1); 3287cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3288cdccee99SLiu Bo btrfs_err(fs_info, 3289cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3290cdccee99SLiu Bo bytenr); 3291cdccee99SLiu Bo WARN_ON(1); 3292cdccee99SLiu Bo ret = -EINVAL; 3293cdccee99SLiu Bo goto out; 3294cdccee99SLiu Bo } 32953173a18fSJosef Bacik 32965d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 32975d4f98a2SYan Zheng out: 32985d4f98a2SYan Zheng btrfs_free_path(path); 32995d4f98a2SYan Zheng return ret; 33005d4f98a2SYan Zheng } 33015d4f98a2SYan Zheng 33020af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 330332da5386SDavid Sterba struct btrfs_block_group *block_group, 33041bbc621eSChris Mason struct inode *inode, 33051bbc621eSChris Mason u64 ino) 33060af3d00bSJosef Bacik { 33070af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 33080af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 33090af3d00bSJosef Bacik int ret = 0; 33100af3d00bSJosef Bacik 33110af3d00bSJosef Bacik if (inode) 33120af3d00bSJosef Bacik goto truncate; 33130af3d00bSJosef Bacik 33140202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, ino, root); 33152e19f1f9SAl Viro if (IS_ERR(inode)) 33160af3d00bSJosef Bacik return -ENOENT; 33170af3d00bSJosef Bacik 33180af3d00bSJosef Bacik truncate: 33192ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 33207b61cd92SMiao Xie &fs_info->global_block_rsv); 33217b61cd92SMiao Xie if (ret) 33227b61cd92SMiao Xie goto out; 33237b61cd92SMiao Xie 33247a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 33250af3d00bSJosef Bacik if (IS_ERR(trans)) { 33263612b495STsutomu Itoh ret = PTR_ERR(trans); 33270af3d00bSJosef Bacik goto out; 33280af3d00bSJosef Bacik } 33290af3d00bSJosef Bacik 333077ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 33310af3d00bSJosef Bacik 33323a45bb20SJeff Mahoney btrfs_end_transaction(trans); 33332ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 33340af3d00bSJosef Bacik out: 33350af3d00bSJosef Bacik iput(inode); 33360af3d00bSJosef Bacik return ret; 33370af3d00bSJosef Bacik } 33380af3d00bSJosef Bacik 33395d4f98a2SYan Zheng /* 334019b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 334119b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 33425d4f98a2SYan Zheng */ 334319b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 334419b546d7SQu Wenruo struct btrfs_block_group *block_group, 334519b546d7SQu Wenruo u64 data_bytenr) 33465d4f98a2SYan Zheng { 334719b546d7SQu Wenruo u64 space_cache_ino; 334819b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 33495d4f98a2SYan Zheng struct btrfs_key key; 335019b546d7SQu Wenruo bool found = false; 335119b546d7SQu Wenruo int i; 33525d4f98a2SYan Zheng int ret; 33535d4f98a2SYan Zheng 335419b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 335519b546d7SQu Wenruo return 0; 33565d4f98a2SYan Zheng 335719b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 335850e31ef4SQu Wenruo u8 type; 335950e31ef4SQu Wenruo 336019b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 336119b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 336219b546d7SQu Wenruo continue; 336319b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 336450e31ef4SQu Wenruo type = btrfs_file_extent_type(leaf, ei); 336550e31ef4SQu Wenruo 336650e31ef4SQu Wenruo if ((type == BTRFS_FILE_EXTENT_REG || 336750e31ef4SQu Wenruo type == BTRFS_FILE_EXTENT_PREALLOC) && 336819b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 336919b546d7SQu Wenruo found = true; 337019b546d7SQu Wenruo space_cache_ino = key.objectid; 337119b546d7SQu Wenruo break; 337219b546d7SQu Wenruo } 337319b546d7SQu Wenruo } 337419b546d7SQu Wenruo if (!found) 337519b546d7SQu Wenruo return -ENOENT; 337619b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 337719b546d7SQu Wenruo space_cache_ino); 33780af3d00bSJosef Bacik return ret; 33795d4f98a2SYan Zheng } 33805d4f98a2SYan Zheng 33815d4f98a2SYan Zheng /* 33822c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 33835d4f98a2SYan Zheng */ 33845d4f98a2SYan Zheng static noinline_for_stack 33855d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 33865d4f98a2SYan Zheng struct btrfs_key *extent_key, 33875d4f98a2SYan Zheng struct btrfs_path *path, 33885d4f98a2SYan Zheng struct rb_root *blocks) 33895d4f98a2SYan Zheng { 339019b546d7SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 339119b546d7SQu Wenruo struct ulist *leaves = NULL; 339219b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 339319b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 339419b546d7SQu Wenruo const u32 blocksize = fs_info->nodesize; 3395647f63bdSFilipe David Borba Manana int ret = 0; 33965d4f98a2SYan Zheng 3397b3b4aa74SDavid Sterba btrfs_release_path(path); 339819b546d7SQu Wenruo ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid, 339919b546d7SQu Wenruo 0, &leaves, NULL, true); 340019b546d7SQu Wenruo if (ret < 0) 340119b546d7SQu Wenruo return ret; 340219b546d7SQu Wenruo 340319b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 340419b546d7SQu Wenruo while ((ref_node = ulist_next(leaves, &leaf_uiter))) { 340519b546d7SQu Wenruo struct extent_buffer *eb; 340619b546d7SQu Wenruo 34071b7ec85eSJosef Bacik eb = read_tree_block(fs_info, ref_node->val, 0, 0, 0, NULL); 340819b546d7SQu Wenruo if (IS_ERR(eb)) { 340919b546d7SQu Wenruo ret = PTR_ERR(eb); 341019b546d7SQu Wenruo break; 341119b546d7SQu Wenruo } 341219b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 341319b546d7SQu Wenruo extent_key->objectid); 341419b546d7SQu Wenruo free_extent_buffer(eb); 341519b546d7SQu Wenruo if (ret < 0) 341619b546d7SQu Wenruo break; 341719b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 341819b546d7SQu Wenruo if (ret < 0) 341919b546d7SQu Wenruo break; 342019b546d7SQu Wenruo } 342119b546d7SQu Wenruo if (ret < 0) 34225d4f98a2SYan Zheng free_block_list(blocks); 342319b546d7SQu Wenruo ulist_free(leaves); 342419b546d7SQu Wenruo return ret; 34255d4f98a2SYan Zheng } 34265d4f98a2SYan Zheng 34275d4f98a2SYan Zheng /* 34282c016dc2SLiu Bo * helper to find next unprocessed extent 34295d4f98a2SYan Zheng */ 34305d4f98a2SYan Zheng static noinline_for_stack 3431147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 34323fd0a558SYan, Zheng struct btrfs_key *extent_key) 34335d4f98a2SYan Zheng { 34340b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34355d4f98a2SYan Zheng struct btrfs_key key; 34365d4f98a2SYan Zheng struct extent_buffer *leaf; 34375d4f98a2SYan Zheng u64 start, end, last; 34385d4f98a2SYan Zheng int ret; 34395d4f98a2SYan Zheng 3440b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 34415d4f98a2SYan Zheng while (1) { 34425d4f98a2SYan Zheng cond_resched(); 34435d4f98a2SYan Zheng if (rc->search_start >= last) { 34445d4f98a2SYan Zheng ret = 1; 34455d4f98a2SYan Zheng break; 34465d4f98a2SYan Zheng } 34475d4f98a2SYan Zheng 34485d4f98a2SYan Zheng key.objectid = rc->search_start; 34495d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34505d4f98a2SYan Zheng key.offset = 0; 34515d4f98a2SYan Zheng 34525d4f98a2SYan Zheng path->search_commit_root = 1; 34535d4f98a2SYan Zheng path->skip_locking = 1; 34545d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 34555d4f98a2SYan Zheng 0, 0); 34565d4f98a2SYan Zheng if (ret < 0) 34575d4f98a2SYan Zheng break; 34585d4f98a2SYan Zheng next: 34595d4f98a2SYan Zheng leaf = path->nodes[0]; 34605d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 34615d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 34625d4f98a2SYan Zheng if (ret != 0) 34635d4f98a2SYan Zheng break; 34645d4f98a2SYan Zheng leaf = path->nodes[0]; 34655d4f98a2SYan Zheng } 34665d4f98a2SYan Zheng 34675d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 34685d4f98a2SYan Zheng if (key.objectid >= last) { 34695d4f98a2SYan Zheng ret = 1; 34705d4f98a2SYan Zheng break; 34715d4f98a2SYan Zheng } 34725d4f98a2SYan Zheng 34733173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 34743173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 34753173a18fSJosef Bacik path->slots[0]++; 34763173a18fSJosef Bacik goto next; 34773173a18fSJosef Bacik } 34783173a18fSJosef Bacik 34793173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 34805d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 34815d4f98a2SYan Zheng path->slots[0]++; 34825d4f98a2SYan Zheng goto next; 34835d4f98a2SYan Zheng } 34845d4f98a2SYan Zheng 34853173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 34860b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 34873173a18fSJosef Bacik rc->search_start) { 34883173a18fSJosef Bacik path->slots[0]++; 34893173a18fSJosef Bacik goto next; 34903173a18fSJosef Bacik } 34913173a18fSJosef Bacik 34925d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 34935d4f98a2SYan Zheng key.objectid, &start, &end, 3494e6138876SJosef Bacik EXTENT_DIRTY, NULL); 34955d4f98a2SYan Zheng 34965d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3497b3b4aa74SDavid Sterba btrfs_release_path(path); 34985d4f98a2SYan Zheng rc->search_start = end + 1; 34995d4f98a2SYan Zheng } else { 35003173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 35015d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 35023173a18fSJosef Bacik else 35033173a18fSJosef Bacik rc->search_start = key.objectid + 35040b246afaSJeff Mahoney fs_info->nodesize; 35053fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 35065d4f98a2SYan Zheng return 0; 35075d4f98a2SYan Zheng } 35085d4f98a2SYan Zheng } 3509b3b4aa74SDavid Sterba btrfs_release_path(path); 35105d4f98a2SYan Zheng return ret; 35115d4f98a2SYan Zheng } 35125d4f98a2SYan Zheng 35135d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 35145d4f98a2SYan Zheng { 35155d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35167585717fSChris Mason 35177585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35185d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 35197585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35205d4f98a2SYan Zheng } 35215d4f98a2SYan Zheng 35225d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 35235d4f98a2SYan Zheng { 35245d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35257585717fSChris Mason 35267585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35275d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 35287585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35295d4f98a2SYan Zheng } 35305d4f98a2SYan Zheng 35313fd0a558SYan, Zheng static noinline_for_stack 35323fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 35333fd0a558SYan, Zheng { 35343fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3535ac2fabacSJosef Bacik int ret; 35363fd0a558SYan, Zheng 35372ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 353866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 35393fd0a558SYan, Zheng if (!rc->block_rsv) 35403fd0a558SYan, Zheng return -ENOMEM; 35413fd0a558SYan, Zheng 35423fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3543b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 35443fd0a558SYan, Zheng rc->extents_found = 0; 35453fd0a558SYan, Zheng rc->nodes_relocated = 0; 35463fd0a558SYan, Zheng rc->merging_rsv_size = 0; 35470647bf56SWang Shilong rc->reserved_bytes = 0; 3548da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 35490647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3550ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3551ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3552ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3553ac2fabacSJosef Bacik if (ret) 3554ac2fabacSJosef Bacik return ret; 35553fd0a558SYan, Zheng 35563fd0a558SYan, Zheng rc->create_reloc_tree = 1; 35573fd0a558SYan, Zheng set_reloc_control(rc); 35583fd0a558SYan, Zheng 35597a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 356028818947SLiu Bo if (IS_ERR(trans)) { 356128818947SLiu Bo unset_reloc_control(rc); 356228818947SLiu Bo /* 356328818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 356428818947SLiu Bo * cleanup. And callers are responsible to free the above 356528818947SLiu Bo * block rsv. 356628818947SLiu Bo */ 356728818947SLiu Bo return PTR_ERR(trans); 356828818947SLiu Bo } 3569fb686c68SJosef Bacik return btrfs_commit_transaction(trans); 35703fd0a558SYan, Zheng } 357176dda93cSYan, Zheng 35725d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 35735d4f98a2SYan Zheng { 35742ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35755d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 35765d4f98a2SYan Zheng struct btrfs_key key; 35775d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 35785d4f98a2SYan Zheng struct btrfs_path *path; 35795d4f98a2SYan Zheng struct btrfs_extent_item *ei; 35805d4f98a2SYan Zheng u64 flags; 35815d4f98a2SYan Zheng int ret; 35825d4f98a2SYan Zheng int err = 0; 3583c87f08caSChris Mason int progress = 0; 35845d4f98a2SYan Zheng 35855d4f98a2SYan Zheng path = btrfs_alloc_path(); 35863fd0a558SYan, Zheng if (!path) 35875d4f98a2SYan Zheng return -ENOMEM; 3588e4058b54SDavid Sterba path->reada = READA_FORWARD; 35893fd0a558SYan, Zheng 35903fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 35913fd0a558SYan, Zheng if (ret) { 35923fd0a558SYan, Zheng err = ret; 35933fd0a558SYan, Zheng goto out_free; 35942423fdfbSJiri Slaby } 35955d4f98a2SYan Zheng 35965d4f98a2SYan Zheng while (1) { 35970647bf56SWang Shilong rc->reserved_bytes = 0; 35980647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 35990647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 36000647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 36010647bf56SWang Shilong if (ret) { 36020647bf56SWang Shilong err = ret; 36030647bf56SWang Shilong break; 36040647bf56SWang Shilong } 3605c87f08caSChris Mason progress++; 3606a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 36070f788c58SLiu Bo if (IS_ERR(trans)) { 36080f788c58SLiu Bo err = PTR_ERR(trans); 36090f788c58SLiu Bo trans = NULL; 36100f788c58SLiu Bo break; 36110f788c58SLiu Bo } 3612c87f08caSChris Mason restart: 36133fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 36143a45bb20SJeff Mahoney btrfs_end_transaction(trans); 361542a657f5SPan Bian trans = NULL; 36163fd0a558SYan, Zheng continue; 36173fd0a558SYan, Zheng } 36183fd0a558SYan, Zheng 3619147d256eSZhaolei ret = find_next_extent(rc, path, &key); 36205d4f98a2SYan Zheng if (ret < 0) 36215d4f98a2SYan Zheng err = ret; 36225d4f98a2SYan Zheng if (ret != 0) 36235d4f98a2SYan Zheng break; 36245d4f98a2SYan Zheng 36255d4f98a2SYan Zheng rc->extents_found++; 36265d4f98a2SYan Zheng 36275d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 36285d4f98a2SYan Zheng struct btrfs_extent_item); 36295d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 36305d4f98a2SYan Zheng 36315d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 36325d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 36335d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 36345d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36355d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 36365d4f98a2SYan Zheng } else { 3637b3b4aa74SDavid Sterba btrfs_release_path(path); 36385d4f98a2SYan Zheng ret = 0; 36395d4f98a2SYan Zheng } 36405d4f98a2SYan Zheng if (ret < 0) { 36413fd0a558SYan, Zheng err = ret; 36425d4f98a2SYan Zheng break; 36435d4f98a2SYan Zheng } 36445d4f98a2SYan Zheng 36455d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 36465d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 36475d4f98a2SYan Zheng if (ret < 0) { 36483fd0a558SYan, Zheng if (ret != -EAGAIN) { 36495d4f98a2SYan Zheng err = ret; 36505d4f98a2SYan Zheng break; 36515d4f98a2SYan Zheng } 36523fd0a558SYan, Zheng rc->extents_found--; 36533fd0a558SYan, Zheng rc->search_start = key.objectid; 36543fd0a558SYan, Zheng } 36555d4f98a2SYan Zheng } 36565d4f98a2SYan Zheng 36573a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 36582ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36593fd0a558SYan, Zheng trans = NULL; 36605d4f98a2SYan Zheng 36615d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 36625d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36635d4f98a2SYan Zheng rc->found_file_extent = 1; 36640257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 36653fd0a558SYan, Zheng &key, &rc->cluster); 36665d4f98a2SYan Zheng if (ret < 0) { 36675d4f98a2SYan Zheng err = ret; 36685d4f98a2SYan Zheng break; 36695d4f98a2SYan Zheng } 36705d4f98a2SYan Zheng } 3671f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 3672f31ea088SQu Wenruo err = -ECANCELED; 3673f31ea088SQu Wenruo break; 3674f31ea088SQu Wenruo } 36755d4f98a2SYan Zheng } 3676c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 367743a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 36789689457bSShilong Wang if (ret == 1) { 3679c87f08caSChris Mason err = 0; 3680c87f08caSChris Mason progress = 0; 3681c87f08caSChris Mason goto restart; 3682c87f08caSChris Mason } 3683c87f08caSChris Mason } 36843fd0a558SYan, Zheng 3685b3b4aa74SDavid Sterba btrfs_release_path(path); 368691166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 36875d4f98a2SYan Zheng 36885d4f98a2SYan Zheng if (trans) { 36893a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 36902ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36915d4f98a2SYan Zheng } 36925d4f98a2SYan Zheng 36930257bb82SYan, Zheng if (!err) { 36943fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 36953fd0a558SYan, Zheng &rc->cluster); 36960257bb82SYan, Zheng if (ret < 0) 36970257bb82SYan, Zheng err = ret; 36980257bb82SYan, Zheng } 36990257bb82SYan, Zheng 37003fd0a558SYan, Zheng rc->create_reloc_tree = 0; 37013fd0a558SYan, Zheng set_reloc_control(rc); 37020257bb82SYan, Zheng 370313fe1bdbSQu Wenruo btrfs_backref_release_cache(&rc->backref_cache); 370463f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37055d4f98a2SYan Zheng 37067f913c7cSQu Wenruo /* 37077f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 37087f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 37097f913c7cSQu Wenruo * 37107f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 37117f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 37127f913c7cSQu Wenruo * merge_reloc_roots() 37137f913c7cSQu Wenruo */ 37143fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 37155d4f98a2SYan Zheng 37165d4f98a2SYan Zheng merge_reloc_roots(rc); 37175d4f98a2SYan Zheng 37183fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 37195d4f98a2SYan Zheng unset_reloc_control(rc); 372063f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37215d4f98a2SYan Zheng 37225d4f98a2SYan Zheng /* get rid of pinned extents */ 37237a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 372462b99540SQu Wenruo if (IS_ERR(trans)) { 37253612b495STsutomu Itoh err = PTR_ERR(trans); 372662b99540SQu Wenruo goto out_free; 372762b99540SQu Wenruo } 3728fb686c68SJosef Bacik ret = btrfs_commit_transaction(trans); 3729fb686c68SJosef Bacik if (ret && !err) 3730fb686c68SJosef Bacik err = ret; 37316217b0faSJosef Bacik out_free: 3732d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 3733d2311e69SQu Wenruo if (ret < 0 && !err) 3734d2311e69SQu Wenruo err = ret; 37352ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 37363fd0a558SYan, Zheng btrfs_free_path(path); 37375d4f98a2SYan Zheng return err; 37385d4f98a2SYan Zheng } 37395d4f98a2SYan Zheng 37405d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 37410257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 37425d4f98a2SYan Zheng { 37435d4f98a2SYan Zheng struct btrfs_path *path; 37445d4f98a2SYan Zheng struct btrfs_inode_item *item; 37455d4f98a2SYan Zheng struct extent_buffer *leaf; 37465d4f98a2SYan Zheng int ret; 37475d4f98a2SYan Zheng 37485d4f98a2SYan Zheng path = btrfs_alloc_path(); 37495d4f98a2SYan Zheng if (!path) 37505d4f98a2SYan Zheng return -ENOMEM; 37515d4f98a2SYan Zheng 37525d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 37535d4f98a2SYan Zheng if (ret) 37545d4f98a2SYan Zheng goto out; 37555d4f98a2SYan Zheng 37565d4f98a2SYan Zheng leaf = path->nodes[0]; 37575d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 3758b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 37595d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 37600257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 37615d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 3762*960a3166SJohannes Thumshirn btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 3763*960a3166SJohannes Thumshirn BTRFS_INODE_PREALLOC); 37645d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 37655d4f98a2SYan Zheng out: 37665d4f98a2SYan Zheng btrfs_free_path(path); 37675d4f98a2SYan Zheng return ret; 37685d4f98a2SYan Zheng } 37695d4f98a2SYan Zheng 3770790c1b8cSJosef Bacik static void delete_orphan_inode(struct btrfs_trans_handle *trans, 3771790c1b8cSJosef Bacik struct btrfs_root *root, u64 objectid) 3772790c1b8cSJosef Bacik { 3773790c1b8cSJosef Bacik struct btrfs_path *path; 3774790c1b8cSJosef Bacik struct btrfs_key key; 3775790c1b8cSJosef Bacik int ret = 0; 3776790c1b8cSJosef Bacik 3777790c1b8cSJosef Bacik path = btrfs_alloc_path(); 3778790c1b8cSJosef Bacik if (!path) { 3779790c1b8cSJosef Bacik ret = -ENOMEM; 3780790c1b8cSJosef Bacik goto out; 3781790c1b8cSJosef Bacik } 3782790c1b8cSJosef Bacik 3783790c1b8cSJosef Bacik key.objectid = objectid; 3784790c1b8cSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 3785790c1b8cSJosef Bacik key.offset = 0; 3786790c1b8cSJosef Bacik ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3787790c1b8cSJosef Bacik if (ret) { 3788790c1b8cSJosef Bacik if (ret > 0) 3789790c1b8cSJosef Bacik ret = -ENOENT; 3790790c1b8cSJosef Bacik goto out; 3791790c1b8cSJosef Bacik } 3792790c1b8cSJosef Bacik ret = btrfs_del_item(trans, root, path); 3793790c1b8cSJosef Bacik out: 3794790c1b8cSJosef Bacik if (ret) 3795790c1b8cSJosef Bacik btrfs_abort_transaction(trans, ret); 3796790c1b8cSJosef Bacik btrfs_free_path(path); 3797790c1b8cSJosef Bacik } 3798790c1b8cSJosef Bacik 37995d4f98a2SYan Zheng /* 38005d4f98a2SYan Zheng * helper to create inode for data relocation. 38015d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 38025d4f98a2SYan Zheng */ 38033fd0a558SYan, Zheng static noinline_for_stack 38043fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 380532da5386SDavid Sterba struct btrfs_block_group *group) 38065d4f98a2SYan Zheng { 38075d4f98a2SYan Zheng struct inode *inode = NULL; 38085d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 38095d4f98a2SYan Zheng struct btrfs_root *root; 38104624900dSZhaolei u64 objectid; 38115d4f98a2SYan Zheng int err = 0; 38125d4f98a2SYan Zheng 3813aeb935a4SQu Wenruo root = btrfs_grab_root(fs_info->data_reloc_root); 3814a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 381576deacf0SJosef Bacik if (IS_ERR(trans)) { 381600246528SJosef Bacik btrfs_put_root(root); 38173fd0a558SYan, Zheng return ERR_CAST(trans); 381876deacf0SJosef Bacik } 38195d4f98a2SYan Zheng 3820543068a2SNikolay Borisov err = btrfs_get_free_objectid(root, &objectid); 38215d4f98a2SYan Zheng if (err) 38225d4f98a2SYan Zheng goto out; 38235d4f98a2SYan Zheng 38240257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 3825790c1b8cSJosef Bacik if (err) 3826790c1b8cSJosef Bacik goto out; 38275d4f98a2SYan Zheng 38280202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, objectid, root); 3829790c1b8cSJosef Bacik if (IS_ERR(inode)) { 3830790c1b8cSJosef Bacik delete_orphan_inode(trans, root, objectid); 3831790c1b8cSJosef Bacik err = PTR_ERR(inode); 3832790c1b8cSJosef Bacik inode = NULL; 3833790c1b8cSJosef Bacik goto out; 3834790c1b8cSJosef Bacik } 3835b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 38365d4f98a2SYan Zheng 383773f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 38385d4f98a2SYan Zheng out: 383900246528SJosef Bacik btrfs_put_root(root); 38403a45bb20SJeff Mahoney btrfs_end_transaction(trans); 38412ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 38425d4f98a2SYan Zheng if (err) { 38435d4f98a2SYan Zheng if (inode) 38445d4f98a2SYan Zheng iput(inode); 38455d4f98a2SYan Zheng inode = ERR_PTR(err); 38465d4f98a2SYan Zheng } 38475d4f98a2SYan Zheng return inode; 38485d4f98a2SYan Zheng } 38495d4f98a2SYan Zheng 3850907d2710SDavid Sterba /* 3851907d2710SDavid Sterba * Mark start of chunk relocation that is cancellable. Check if the cancellation 3852907d2710SDavid Sterba * has been requested meanwhile and don't start in that case. 3853907d2710SDavid Sterba * 3854907d2710SDavid Sterba * Return: 3855907d2710SDavid Sterba * 0 success 3856907d2710SDavid Sterba * -EINPROGRESS operation is already in progress, that's probably a bug 3857907d2710SDavid Sterba * -ECANCELED cancellation request was set before the operation started 38581cea5cf0SFilipe Manana * -EAGAIN can not start because there are ongoing send operations 3859907d2710SDavid Sterba */ 3860907d2710SDavid Sterba static int reloc_chunk_start(struct btrfs_fs_info *fs_info) 3861907d2710SDavid Sterba { 38621cea5cf0SFilipe Manana spin_lock(&fs_info->send_reloc_lock); 38631cea5cf0SFilipe Manana if (fs_info->send_in_progress) { 38641cea5cf0SFilipe Manana btrfs_warn_rl(fs_info, 38651cea5cf0SFilipe Manana "cannot run relocation while send operations are in progress (%d in progress)", 38661cea5cf0SFilipe Manana fs_info->send_in_progress); 38671cea5cf0SFilipe Manana spin_unlock(&fs_info->send_reloc_lock); 38681cea5cf0SFilipe Manana return -EAGAIN; 38691cea5cf0SFilipe Manana } 3870907d2710SDavid Sterba if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) { 3871907d2710SDavid Sterba /* This should not happen */ 38721cea5cf0SFilipe Manana spin_unlock(&fs_info->send_reloc_lock); 3873907d2710SDavid Sterba btrfs_err(fs_info, "reloc already running, cannot start"); 3874907d2710SDavid Sterba return -EINPROGRESS; 3875907d2710SDavid Sterba } 38761cea5cf0SFilipe Manana spin_unlock(&fs_info->send_reloc_lock); 3877907d2710SDavid Sterba 3878907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) { 3879907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled on start"); 3880907d2710SDavid Sterba /* 3881907d2710SDavid Sterba * On cancel, clear all requests but let the caller mark 3882907d2710SDavid Sterba * the end after cleanup operations. 3883907d2710SDavid Sterba */ 3884907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3885907d2710SDavid Sterba return -ECANCELED; 3886907d2710SDavid Sterba } 3887907d2710SDavid Sterba return 0; 3888907d2710SDavid Sterba } 3889907d2710SDavid Sterba 3890907d2710SDavid Sterba /* 3891907d2710SDavid Sterba * Mark end of chunk relocation that is cancellable and wake any waiters. 3892907d2710SDavid Sterba */ 3893907d2710SDavid Sterba static void reloc_chunk_end(struct btrfs_fs_info *fs_info) 3894907d2710SDavid Sterba { 3895907d2710SDavid Sterba /* Requested after start, clear bit first so any waiters can continue */ 3896907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) 3897907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled during operation"); 38981cea5cf0SFilipe Manana spin_lock(&fs_info->send_reloc_lock); 3899907d2710SDavid Sterba clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags); 39001cea5cf0SFilipe Manana spin_unlock(&fs_info->send_reloc_lock); 3901907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3902907d2710SDavid Sterba } 3903907d2710SDavid Sterba 3904c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 39053fd0a558SYan, Zheng { 39063fd0a558SYan, Zheng struct reloc_control *rc; 39073fd0a558SYan, Zheng 39083fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 39093fd0a558SYan, Zheng if (!rc) 39103fd0a558SYan, Zheng return NULL; 39113fd0a558SYan, Zheng 39123fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 3913d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 3914584fb121SQu Wenruo btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1); 39153fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 391643eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 391743eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 39183fd0a558SYan, Zheng return rc; 39193fd0a558SYan, Zheng } 39203fd0a558SYan, Zheng 39211a0afa0eSJosef Bacik static void free_reloc_control(struct reloc_control *rc) 39221a0afa0eSJosef Bacik { 39231a0afa0eSJosef Bacik struct mapping_node *node, *tmp; 39241a0afa0eSJosef Bacik 39251a0afa0eSJosef Bacik free_reloc_roots(&rc->reloc_roots); 39261a0afa0eSJosef Bacik rbtree_postorder_for_each_entry_safe(node, tmp, 39271a0afa0eSJosef Bacik &rc->reloc_root_tree.rb_root, rb_node) 39281a0afa0eSJosef Bacik kfree(node); 39291a0afa0eSJosef Bacik 39301a0afa0eSJosef Bacik kfree(rc); 39311a0afa0eSJosef Bacik } 39321a0afa0eSJosef Bacik 39335d4f98a2SYan Zheng /* 3934ebce0e01SAdam Borowski * Print the block group being relocated 3935ebce0e01SAdam Borowski */ 3936ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 393732da5386SDavid Sterba struct btrfs_block_group *block_group) 3938ebce0e01SAdam Borowski { 3939f89e09cfSAnand Jain char buf[128] = {'\0'}; 3940ebce0e01SAdam Borowski 3941f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 3942ebce0e01SAdam Borowski 3943ebce0e01SAdam Borowski btrfs_info(fs_info, 3944ebce0e01SAdam Borowski "relocating block group %llu flags %s", 3945b3470b5dSDavid Sterba block_group->start, buf); 3946ebce0e01SAdam Borowski } 3947ebce0e01SAdam Borowski 3948430640e3SQu Wenruo static const char *stage_to_string(int stage) 3949430640e3SQu Wenruo { 3950430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 3951430640e3SQu Wenruo return "move data extents"; 3952430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 3953430640e3SQu Wenruo return "update data pointers"; 3954430640e3SQu Wenruo return "unknown"; 3955430640e3SQu Wenruo } 3956430640e3SQu Wenruo 3957ebce0e01SAdam Borowski /* 39585d4f98a2SYan Zheng * function to relocate all extents in a block group. 39595d4f98a2SYan Zheng */ 39606bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 39615d4f98a2SYan Zheng { 396232da5386SDavid Sterba struct btrfs_block_group *bg; 39636bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 39645d4f98a2SYan Zheng struct reloc_control *rc; 39650af3d00bSJosef Bacik struct inode *inode; 39660af3d00bSJosef Bacik struct btrfs_path *path; 39675d4f98a2SYan Zheng int ret; 3968f0486c68SYan, Zheng int rw = 0; 39695d4f98a2SYan Zheng int err = 0; 39705d4f98a2SYan Zheng 3971eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 3972eede2bf3SOmar Sandoval if (!bg) 3973eede2bf3SOmar Sandoval return -ENOENT; 3974eede2bf3SOmar Sandoval 3975eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 3976eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 3977eede2bf3SOmar Sandoval return -ETXTBSY; 3978eede2bf3SOmar Sandoval } 3979eede2bf3SOmar Sandoval 3980c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 3981eede2bf3SOmar Sandoval if (!rc) { 3982eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 39835d4f98a2SYan Zheng return -ENOMEM; 3984eede2bf3SOmar Sandoval } 39855d4f98a2SYan Zheng 3986907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 3987907d2710SDavid Sterba if (ret < 0) { 3988907d2710SDavid Sterba err = ret; 3989907d2710SDavid Sterba goto out_put_bg; 3990907d2710SDavid Sterba } 3991907d2710SDavid Sterba 3992f0486c68SYan, Zheng rc->extent_root = extent_root; 3993eede2bf3SOmar Sandoval rc->block_group = bg; 39945d4f98a2SYan Zheng 3995b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 3996f0486c68SYan, Zheng if (ret) { 3997f0486c68SYan, Zheng err = ret; 3998f0486c68SYan, Zheng goto out; 3999f0486c68SYan, Zheng } 4000f0486c68SYan, Zheng rw = 1; 4001f0486c68SYan, Zheng 40020af3d00bSJosef Bacik path = btrfs_alloc_path(); 40030af3d00bSJosef Bacik if (!path) { 40040af3d00bSJosef Bacik err = -ENOMEM; 40050af3d00bSJosef Bacik goto out; 40060af3d00bSJosef Bacik } 40070af3d00bSJosef Bacik 40087949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 40090af3d00bSJosef Bacik btrfs_free_path(path); 40100af3d00bSJosef Bacik 40110af3d00bSJosef Bacik if (!IS_ERR(inode)) 40121bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 40130af3d00bSJosef Bacik else 40140af3d00bSJosef Bacik ret = PTR_ERR(inode); 40150af3d00bSJosef Bacik 40160af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 40170af3d00bSJosef Bacik err = ret; 40180af3d00bSJosef Bacik goto out; 40190af3d00bSJosef Bacik } 40200af3d00bSJosef Bacik 40215d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 40225d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 40235d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 40245d4f98a2SYan Zheng rc->data_inode = NULL; 40255d4f98a2SYan Zheng goto out; 40265d4f98a2SYan Zheng } 40275d4f98a2SYan Zheng 40280b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 40295d4f98a2SYan Zheng 40309cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4031f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 40326374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4033b3470b5dSDavid Sterba rc->block_group->start, 4034b3470b5dSDavid Sterba rc->block_group->length); 40355d4f98a2SYan Zheng 40367ae9bd18SNaohiro Aota ret = btrfs_zone_finish(rc->block_group); 40377ae9bd18SNaohiro Aota WARN_ON(ret && ret != -EAGAIN); 40387ae9bd18SNaohiro Aota 40395d4f98a2SYan Zheng while (1) { 4040430640e3SQu Wenruo int finishes_stage; 4041430640e3SQu Wenruo 404276dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 40435d4f98a2SYan Zheng ret = relocate_block_group(rc); 404476dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4045ff612ba7SJosef Bacik if (ret < 0) 40465d4f98a2SYan Zheng err = ret; 4047ff612ba7SJosef Bacik 4048430640e3SQu Wenruo finishes_stage = rc->stage; 4049ff612ba7SJosef Bacik /* 4050ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4051ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4052ff612ba7SJosef Bacik * different block group we could end up hitting the 4053ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4054ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4055ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4056ff612ba7SJosef Bacik * out of the loop if we hit an error. 4057ff612ba7SJosef Bacik */ 4058ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4059ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4060ff612ba7SJosef Bacik (u64)-1); 4061ff612ba7SJosef Bacik if (ret) 4062ff612ba7SJosef Bacik err = ret; 4063ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4064ff612ba7SJosef Bacik 0, -1); 4065ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 40665d4f98a2SYan Zheng } 40675d4f98a2SYan Zheng 4068ff612ba7SJosef Bacik if (err < 0) 4069ff612ba7SJosef Bacik goto out; 4070ff612ba7SJosef Bacik 40715d4f98a2SYan Zheng if (rc->extents_found == 0) 40725d4f98a2SYan Zheng break; 40735d4f98a2SYan Zheng 4074430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4075430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 40765d4f98a2SYan Zheng } 40775d4f98a2SYan Zheng 40785d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 40795d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4080bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 40815d4f98a2SYan Zheng out: 4082f0486c68SYan, Zheng if (err && rw) 40832ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 40845d4f98a2SYan Zheng iput(rc->data_inode); 4085907d2710SDavid Sterba out_put_bg: 4086907d2710SDavid Sterba btrfs_put_block_group(bg); 4087907d2710SDavid Sterba reloc_chunk_end(fs_info); 40881a0afa0eSJosef Bacik free_reloc_control(rc); 40895d4f98a2SYan Zheng return err; 40905d4f98a2SYan Zheng } 40915d4f98a2SYan Zheng 409276dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 409376dda93cSYan, Zheng { 40940b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 409576dda93cSYan, Zheng struct btrfs_trans_handle *trans; 409679787eaaSJeff Mahoney int ret, err; 409776dda93cSYan, Zheng 40980b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 409979787eaaSJeff Mahoney if (IS_ERR(trans)) 410079787eaaSJeff Mahoney return PTR_ERR(trans); 410176dda93cSYan, Zheng 410276dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 410376dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 4104c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 410576dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 41060b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 410776dda93cSYan, Zheng &root->root_key, &root->root_item); 410876dda93cSYan, Zheng 41093a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 411079787eaaSJeff Mahoney if (err) 411179787eaaSJeff Mahoney return err; 411279787eaaSJeff Mahoney return ret; 411376dda93cSYan, Zheng } 411476dda93cSYan, Zheng 41155d4f98a2SYan Zheng /* 41165d4f98a2SYan Zheng * recover relocation interrupted by system crash. 41175d4f98a2SYan Zheng * 41185d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 41195d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 41205d4f98a2SYan Zheng */ 41215d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 41225d4f98a2SYan Zheng { 41230b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 41245d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 41255d4f98a2SYan Zheng struct btrfs_key key; 41265d4f98a2SYan Zheng struct btrfs_root *fs_root; 41275d4f98a2SYan Zheng struct btrfs_root *reloc_root; 41285d4f98a2SYan Zheng struct btrfs_path *path; 41295d4f98a2SYan Zheng struct extent_buffer *leaf; 41305d4f98a2SYan Zheng struct reloc_control *rc = NULL; 41315d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41325d4f98a2SYan Zheng int ret; 41335d4f98a2SYan Zheng int err = 0; 41345d4f98a2SYan Zheng 41355d4f98a2SYan Zheng path = btrfs_alloc_path(); 41365d4f98a2SYan Zheng if (!path) 41375d4f98a2SYan Zheng return -ENOMEM; 4138e4058b54SDavid Sterba path->reada = READA_BACK; 41395d4f98a2SYan Zheng 41405d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 41415d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 41425d4f98a2SYan Zheng key.offset = (u64)-1; 41435d4f98a2SYan Zheng 41445d4f98a2SYan Zheng while (1) { 41450b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 41465d4f98a2SYan Zheng path, 0, 0); 41475d4f98a2SYan Zheng if (ret < 0) { 41485d4f98a2SYan Zheng err = ret; 41495d4f98a2SYan Zheng goto out; 41505d4f98a2SYan Zheng } 41515d4f98a2SYan Zheng if (ret > 0) { 41525d4f98a2SYan Zheng if (path->slots[0] == 0) 41535d4f98a2SYan Zheng break; 41545d4f98a2SYan Zheng path->slots[0]--; 41555d4f98a2SYan Zheng } 41565d4f98a2SYan Zheng leaf = path->nodes[0]; 41575d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4158b3b4aa74SDavid Sterba btrfs_release_path(path); 41595d4f98a2SYan Zheng 41605d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 41615d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 41625d4f98a2SYan Zheng break; 41635d4f98a2SYan Zheng 41643dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(root, &key); 41655d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 41665d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 41675d4f98a2SYan Zheng goto out; 41685d4f98a2SYan Zheng } 41695d4f98a2SYan Zheng 417092a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 41715d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 41725d4f98a2SYan Zheng 41735d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 4174a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, 4175a820feb5SDavid Sterba reloc_root->root_key.offset, false); 41765d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 417776dda93cSYan, Zheng ret = PTR_ERR(fs_root); 417876dda93cSYan, Zheng if (ret != -ENOENT) { 417976dda93cSYan, Zheng err = ret; 41805d4f98a2SYan Zheng goto out; 41815d4f98a2SYan Zheng } 418279787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 418379787eaaSJeff Mahoney if (ret < 0) { 418479787eaaSJeff Mahoney err = ret; 418579787eaaSJeff Mahoney goto out; 418679787eaaSJeff Mahoney } 4187932fd26dSJosef Bacik } else { 418800246528SJosef Bacik btrfs_put_root(fs_root); 418976dda93cSYan, Zheng } 41905d4f98a2SYan Zheng } 41915d4f98a2SYan Zheng 41925d4f98a2SYan Zheng if (key.offset == 0) 41935d4f98a2SYan Zheng break; 41945d4f98a2SYan Zheng 41955d4f98a2SYan Zheng key.offset--; 41965d4f98a2SYan Zheng } 4197b3b4aa74SDavid Sterba btrfs_release_path(path); 41985d4f98a2SYan Zheng 41995d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 42005d4f98a2SYan Zheng goto out; 42015d4f98a2SYan Zheng 4202c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 42035d4f98a2SYan Zheng if (!rc) { 42045d4f98a2SYan Zheng err = -ENOMEM; 42055d4f98a2SYan Zheng goto out; 42065d4f98a2SYan Zheng } 42075d4f98a2SYan Zheng 4208907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4209907d2710SDavid Sterba if (ret < 0) { 4210907d2710SDavid Sterba err = ret; 4211907d2710SDavid Sterba goto out_end; 4212907d2710SDavid Sterba } 4213907d2710SDavid Sterba 42140b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 42155d4f98a2SYan Zheng 42165d4f98a2SYan Zheng set_reloc_control(rc); 42175d4f98a2SYan Zheng 42187a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 42193612b495STsutomu Itoh if (IS_ERR(trans)) { 42203612b495STsutomu Itoh err = PTR_ERR(trans); 4221fb2d83eeSJosef Bacik goto out_unset; 42223612b495STsutomu Itoh } 42233fd0a558SYan, Zheng 42243fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 42253fd0a558SYan, Zheng 42265d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 42275d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 42285d4f98a2SYan Zheng struct btrfs_root, root_list); 42295d4f98a2SYan Zheng list_del(&reloc_root->root_list); 42305d4f98a2SYan Zheng 42315d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 42325d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 42335d4f98a2SYan Zheng &rc->reloc_roots); 42345d4f98a2SYan Zheng continue; 42355d4f98a2SYan Zheng } 42365d4f98a2SYan Zheng 4237a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 4238a820feb5SDavid Sterba false); 423979787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 424079787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4241ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42421402d17dSXiyu Yang btrfs_end_transaction(trans); 4243fb2d83eeSJosef Bacik goto out_unset; 424479787eaaSJeff Mahoney } 42455d4f98a2SYan Zheng 4246ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 424757a304cfSJosef Bacik ASSERT(err != -EEXIST); 42483c925863SJosef Bacik if (err) { 42493c925863SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42503c925863SJosef Bacik btrfs_put_root(fs_root); 42513c925863SJosef Bacik btrfs_end_transaction(trans); 42523c925863SJosef Bacik goto out_unset; 42533c925863SJosef Bacik } 4254f44deb74SJosef Bacik fs_root->reloc_root = btrfs_grab_root(reloc_root); 425500246528SJosef Bacik btrfs_put_root(fs_root); 42565d4f98a2SYan Zheng } 42575d4f98a2SYan Zheng 42583a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 425979787eaaSJeff Mahoney if (err) 4260fb2d83eeSJosef Bacik goto out_unset; 42615d4f98a2SYan Zheng 42625d4f98a2SYan Zheng merge_reloc_roots(rc); 42635d4f98a2SYan Zheng 42645d4f98a2SYan Zheng unset_reloc_control(rc); 42655d4f98a2SYan Zheng 42667a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 426762b99540SQu Wenruo if (IS_ERR(trans)) { 42683612b495STsutomu Itoh err = PTR_ERR(trans); 42696217b0faSJosef Bacik goto out_clean; 427062b99540SQu Wenruo } 42713a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 42726217b0faSJosef Bacik out_clean: 4273d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4274d2311e69SQu Wenruo if (ret < 0 && !err) 4275d2311e69SQu Wenruo err = ret; 4276fb2d83eeSJosef Bacik out_unset: 4277fb2d83eeSJosef Bacik unset_reloc_control(rc); 4278907d2710SDavid Sterba out_end: 4279907d2710SDavid Sterba reloc_chunk_end(fs_info); 42801a0afa0eSJosef Bacik free_reloc_control(rc); 42813612b495STsutomu Itoh out: 4282aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4283aca1bba6SLiu Bo 42845d4f98a2SYan Zheng btrfs_free_path(path); 42855d4f98a2SYan Zheng 42865d4f98a2SYan Zheng if (err == 0) { 42875d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 4288aeb935a4SQu Wenruo fs_root = btrfs_grab_root(fs_info->data_reloc_root); 4289aeb935a4SQu Wenruo ASSERT(fs_root); 429066b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 429100246528SJosef Bacik btrfs_put_root(fs_root); 4292932fd26dSJosef Bacik } 42935d4f98a2SYan Zheng return err; 42945d4f98a2SYan Zheng } 42955d4f98a2SYan Zheng 42965d4f98a2SYan Zheng /* 42975d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 42985d4f98a2SYan Zheng * 42995d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 43005d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 43015d4f98a2SYan Zheng */ 43027bfa9535SNikolay Borisov int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len) 43035d4f98a2SYan Zheng { 43047bfa9535SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 43055d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 43065d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 43075d4f98a2SYan Zheng int ret; 43085d4f98a2SYan Zheng u64 disk_bytenr; 43094577b014SJosef Bacik u64 new_bytenr; 43105d4f98a2SYan Zheng LIST_HEAD(list); 43115d4f98a2SYan Zheng 43127bfa9535SNikolay Borisov ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4313bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 43145d4f98a2SYan Zheng 43157bfa9535SNikolay Borisov disk_bytenr = file_pos + inode->index_cnt; 43160b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4317a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 431879787eaaSJeff Mahoney if (ret) 431979787eaaSJeff Mahoney goto out; 43205d4f98a2SYan Zheng 43215d4f98a2SYan Zheng while (!list_empty(&list)) { 43225d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 43235d4f98a2SYan Zheng list_del_init(&sums->list); 43245d4f98a2SYan Zheng 43254577b014SJosef Bacik /* 43264577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 43274577b014SJosef Bacik * We need to do this because we will read in entire prealloc 43284577b014SJosef Bacik * extents but we may have written to say the middle of the 43294577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 43304577b014SJosef Bacik * the right disk offset. 43314577b014SJosef Bacik * 43324577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 43334577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 43344577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 43354577b014SJosef Bacik * disk length. 43364577b014SJosef Bacik */ 4337bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 43384577b014SJosef Bacik sums->bytenr = new_bytenr; 43395d4f98a2SYan Zheng 4340f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 43415d4f98a2SYan Zheng } 434279787eaaSJeff Mahoney out: 43435d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4344411fc6bcSAndi Kleen return ret; 43455d4f98a2SYan Zheng } 43463fd0a558SYan, Zheng 434783d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 43483fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 43493fd0a558SYan, Zheng struct extent_buffer *cow) 43503fd0a558SYan, Zheng { 43510b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 43523fd0a558SYan, Zheng struct reloc_control *rc; 4353a26195a5SQu Wenruo struct btrfs_backref_node *node; 43543fd0a558SYan, Zheng int first_cow = 0; 43553fd0a558SYan, Zheng int level; 435683d4cfd4SJosef Bacik int ret = 0; 43573fd0a558SYan, Zheng 43580b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 43593fd0a558SYan, Zheng if (!rc) 436083d4cfd4SJosef Bacik return 0; 43613fd0a558SYan, Zheng 436237f00a6dSJohannes Thumshirn BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root)); 43633fd0a558SYan, Zheng 43643fd0a558SYan, Zheng level = btrfs_header_level(buf); 43653fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 43663fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 43673fd0a558SYan, Zheng first_cow = 1; 43683fd0a558SYan, Zheng 43693fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 43703fd0a558SYan, Zheng rc->create_reloc_tree) { 43713fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 43723fd0a558SYan, Zheng 43733fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 43743fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 43753fd0a558SYan, Zheng node->new_bytenr != buf->start); 43763fd0a558SYan, Zheng 4377b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 437867439dadSDavid Sterba atomic_inc(&cow->refs); 43793fd0a558SYan, Zheng node->eb = cow; 43803fd0a558SYan, Zheng node->new_bytenr = cow->start; 43813fd0a558SYan, Zheng 43823fd0a558SYan, Zheng if (!node->pending) { 43833fd0a558SYan, Zheng list_move_tail(&node->list, 43843fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 43853fd0a558SYan, Zheng node->pending = 1; 43863fd0a558SYan, Zheng } 43873fd0a558SYan, Zheng 43883fd0a558SYan, Zheng if (first_cow) 43899569cc20SQu Wenruo mark_block_processed(rc, node); 43903fd0a558SYan, Zheng 43913fd0a558SYan, Zheng if (first_cow && level > 0) 43923fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 43933fd0a558SYan, Zheng } 43943fd0a558SYan, Zheng 439583d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 43963fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 439783d4cfd4SJosef Bacik return ret; 43983fd0a558SYan, Zheng } 43993fd0a558SYan, Zheng 44003fd0a558SYan, Zheng /* 44013fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 440201327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 44033fd0a558SYan, Zheng */ 4404147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 44053fd0a558SYan, Zheng u64 *bytes_to_reserve) 44063fd0a558SYan, Zheng { 440710995c04SQu Wenruo struct btrfs_root *root = pending->root; 440810995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44093fd0a558SYan, Zheng 44106282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 44113fd0a558SYan, Zheng return; 44123fd0a558SYan, Zheng 44133fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 44143fd0a558SYan, Zheng return; 44153fd0a558SYan, Zheng 44163fd0a558SYan, Zheng root = root->reloc_root; 44173fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 44183fd0a558SYan, Zheng /* 44193fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 44203fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 44213fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 44223fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 44233fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 44243fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 44253fd0a558SYan, Zheng * tree may use more space than it frees. so we need 44263fd0a558SYan, Zheng * reserve extra space. 44273fd0a558SYan, Zheng */ 44283fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 44293fd0a558SYan, Zheng } 44303fd0a558SYan, Zheng 44313fd0a558SYan, Zheng /* 44323fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 44333fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 4434f44deb74SJosef Bacik * 4435f44deb74SJosef Bacik * This is similar to btrfs_init_reloc_root(), we come out of here with two 4436f44deb74SJosef Bacik * references held on the reloc_root, one for root->reloc_root and one for 4437f44deb74SJosef Bacik * rc->reloc_roots. 44383fd0a558SYan, Zheng */ 443949b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 44403fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 44413fd0a558SYan, Zheng { 44423fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 44433fd0a558SYan, Zheng struct btrfs_root *reloc_root; 44443fd0a558SYan, Zheng struct btrfs_root *new_root; 444510995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44463fd0a558SYan, Zheng int ret; 44473fd0a558SYan, Zheng 44486282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 444949b25e05SJeff Mahoney return 0; 44503fd0a558SYan, Zheng 44513fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 44523fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 44533fd0a558SYan, Zheng 44543fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 44553fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 44563fd0a558SYan, Zheng rc->block_rsv, 44573a584174SLu Fengqi rc->nodes_relocated, true); 445849b25e05SJeff Mahoney if (ret) 445949b25e05SJeff Mahoney return ret; 44603fd0a558SYan, Zheng } 44613fd0a558SYan, Zheng 44623fd0a558SYan, Zheng new_root = pending->snap; 44633fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 44643fd0a558SYan, Zheng new_root->root_key.objectid); 446549b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 446649b25e05SJeff Mahoney return PTR_ERR(reloc_root); 44673fd0a558SYan, Zheng 4468ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 446957a304cfSJosef Bacik ASSERT(ret != -EEXIST); 44703c925863SJosef Bacik if (ret) { 44713c925863SJosef Bacik /* Pairs with create_reloc_root */ 44723c925863SJosef Bacik btrfs_put_root(reloc_root); 44733c925863SJosef Bacik return ret; 44743c925863SJosef Bacik } 4475f44deb74SJosef Bacik new_root->reloc_root = btrfs_grab_root(reloc_root); 44763fd0a558SYan, Zheng 447749b25e05SJeff Mahoney if (rc->create_reloc_tree) 44783fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 447949b25e05SJeff Mahoney return ret; 44803fd0a558SYan, Zheng } 4481