1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 25d4f98a2SYan Zheng /* 35d4f98a2SYan Zheng * Copyright (C) 2009 Oracle. All rights reserved. 45d4f98a2SYan Zheng */ 55d4f98a2SYan Zheng 65d4f98a2SYan Zheng #include <linux/sched.h> 75d4f98a2SYan Zheng #include <linux/pagemap.h> 85d4f98a2SYan Zheng #include <linux/writeback.h> 95d4f98a2SYan Zheng #include <linux/blkdev.h> 105d4f98a2SYan Zheng #include <linux/rbtree.h> 115a0e3ad6STejun Heo #include <linux/slab.h> 12726a3421SQu Wenruo #include <linux/error-injection.h> 135d4f98a2SYan Zheng #include "ctree.h" 145d4f98a2SYan Zheng #include "disk-io.h" 155d4f98a2SYan Zheng #include "transaction.h" 165d4f98a2SYan Zheng #include "volumes.h" 175d4f98a2SYan Zheng #include "locking.h" 185d4f98a2SYan Zheng #include "btrfs_inode.h" 195d4f98a2SYan Zheng #include "async-thread.h" 200af3d00bSJosef Bacik #include "free-space-cache.h" 2162b99540SQu Wenruo #include "qgroup.h" 22cdccee99SLiu Bo #include "print-tree.h" 2386736342SJosef Bacik #include "delalloc-space.h" 24aac0023cSJosef Bacik #include "block-group.h" 2519b546d7SQu Wenruo #include "backref.h" 26e9a28dc5SQu Wenruo #include "misc.h" 27c2832898SQu Wenruo #include "subpage.h" 287ae9bd18SNaohiro Aota #include "zoned.h" 2926c2c454SJosef Bacik #include "inode-item.h" 30f1e5c618SJosef Bacik #include "space-info.h" 31c7f13d42SJosef Bacik #include "fs.h" 3207e81dc9SJosef Bacik #include "accessors.h" 33a0231804SJosef Bacik #include "extent-tree.h" 3445c40c8fSJosef Bacik #include "root-tree.h" 35*7c8ede16SJosef Bacik #include "file-item.h" 365d4f98a2SYan Zheng 375d4f98a2SYan Zheng /* 380c891389SQu Wenruo * Relocation overview 390c891389SQu Wenruo * 400c891389SQu Wenruo * [What does relocation do] 410c891389SQu Wenruo * 420c891389SQu Wenruo * The objective of relocation is to relocate all extents of the target block 430c891389SQu Wenruo * group to other block groups. 440c891389SQu Wenruo * This is utilized by resize (shrink only), profile converting, compacting 450c891389SQu Wenruo * space, or balance routine to spread chunks over devices. 460c891389SQu Wenruo * 470c891389SQu Wenruo * Before | After 480c891389SQu Wenruo * ------------------------------------------------------------------ 490c891389SQu Wenruo * BG A: 10 data extents | BG A: deleted 500c891389SQu Wenruo * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) 510c891389SQu Wenruo * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) 520c891389SQu Wenruo * 530c891389SQu Wenruo * [How does relocation work] 540c891389SQu Wenruo * 550c891389SQu Wenruo * 1. Mark the target block group read-only 560c891389SQu Wenruo * New extents won't be allocated from the target block group. 570c891389SQu Wenruo * 580c891389SQu Wenruo * 2.1 Record each extent in the target block group 590c891389SQu Wenruo * To build a proper map of extents to be relocated. 600c891389SQu Wenruo * 610c891389SQu Wenruo * 2.2 Build data reloc tree and reloc trees 620c891389SQu Wenruo * Data reloc tree will contain an inode, recording all newly relocated 630c891389SQu Wenruo * data extents. 640c891389SQu Wenruo * There will be only one data reloc tree for one data block group. 650c891389SQu Wenruo * 660c891389SQu Wenruo * Reloc tree will be a special snapshot of its source tree, containing 670c891389SQu Wenruo * relocated tree blocks. 680c891389SQu Wenruo * Each tree referring to a tree block in target block group will get its 690c891389SQu Wenruo * reloc tree built. 700c891389SQu Wenruo * 710c891389SQu Wenruo * 2.3 Swap source tree with its corresponding reloc tree 720c891389SQu Wenruo * Each involved tree only refers to new extents after swap. 730c891389SQu Wenruo * 740c891389SQu Wenruo * 3. Cleanup reloc trees and data reloc tree. 750c891389SQu Wenruo * As old extents in the target block group are still referenced by reloc 760c891389SQu Wenruo * trees, we need to clean them up before really freeing the target block 770c891389SQu Wenruo * group. 780c891389SQu Wenruo * 790c891389SQu Wenruo * The main complexity is in steps 2.2 and 2.3. 800c891389SQu Wenruo * 810c891389SQu Wenruo * The entry point of relocation is relocate_block_group() function. 820c891389SQu Wenruo */ 830c891389SQu Wenruo 840647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 852a979612SQu Wenruo /* 865d4f98a2SYan Zheng * map address of tree root to tree 875d4f98a2SYan Zheng */ 885d4f98a2SYan Zheng struct mapping_node { 89e9a28dc5SQu Wenruo struct { 905d4f98a2SYan Zheng struct rb_node rb_node; 915d4f98a2SYan Zheng u64 bytenr; 92e9a28dc5SQu Wenruo }; /* Use rb_simle_node for search/insert */ 935d4f98a2SYan Zheng void *data; 945d4f98a2SYan Zheng }; 955d4f98a2SYan Zheng 965d4f98a2SYan Zheng struct mapping_tree { 975d4f98a2SYan Zheng struct rb_root rb_root; 985d4f98a2SYan Zheng spinlock_t lock; 995d4f98a2SYan Zheng }; 1005d4f98a2SYan Zheng 1015d4f98a2SYan Zheng /* 1025d4f98a2SYan Zheng * present a tree block to process 1035d4f98a2SYan Zheng */ 1045d4f98a2SYan Zheng struct tree_block { 105e9a28dc5SQu Wenruo struct { 1065d4f98a2SYan Zheng struct rb_node rb_node; 1075d4f98a2SYan Zheng u64 bytenr; 108e9a28dc5SQu Wenruo }; /* Use rb_simple_node for search/insert */ 109f7ba2d37SJosef Bacik u64 owner; 1105d4f98a2SYan Zheng struct btrfs_key key; 1115d4f98a2SYan Zheng unsigned int level:8; 1125d4f98a2SYan Zheng unsigned int key_ready:1; 1135d4f98a2SYan Zheng }; 1145d4f98a2SYan Zheng 1150257bb82SYan, Zheng #define MAX_EXTENTS 128 1160257bb82SYan, Zheng 1170257bb82SYan, Zheng struct file_extent_cluster { 1180257bb82SYan, Zheng u64 start; 1190257bb82SYan, Zheng u64 end; 1200257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1210257bb82SYan, Zheng unsigned int nr; 1220257bb82SYan, Zheng }; 1230257bb82SYan, Zheng 1245d4f98a2SYan Zheng struct reloc_control { 1255d4f98a2SYan Zheng /* block group to relocate */ 12632da5386SDavid Sterba struct btrfs_block_group *block_group; 1275d4f98a2SYan Zheng /* extent tree */ 1285d4f98a2SYan Zheng struct btrfs_root *extent_root; 1295d4f98a2SYan Zheng /* inode for moving data */ 1305d4f98a2SYan Zheng struct inode *data_inode; 1313fd0a558SYan, Zheng 1323fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1333fd0a558SYan, Zheng 134a26195a5SQu Wenruo struct btrfs_backref_cache backref_cache; 1353fd0a558SYan, Zheng 1363fd0a558SYan, Zheng struct file_extent_cluster cluster; 1375d4f98a2SYan Zheng /* tree blocks have been processed */ 1385d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1395d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1405d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1415d4f98a2SYan Zheng /* list of reloc trees */ 1425d4f98a2SYan Zheng struct list_head reloc_roots; 143d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 144d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 1453fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1463fd0a558SYan, Zheng u64 merging_rsv_size; 1473fd0a558SYan, Zheng /* size of relocated tree nodes */ 1483fd0a558SYan, Zheng u64 nodes_relocated; 1490647bf56SWang Shilong /* reserved size for block group relocation*/ 1500647bf56SWang Shilong u64 reserved_bytes; 1513fd0a558SYan, Zheng 1525d4f98a2SYan Zheng u64 search_start; 1535d4f98a2SYan Zheng u64 extents_found; 1543fd0a558SYan, Zheng 1553fd0a558SYan, Zheng unsigned int stage:8; 1563fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1573fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1585d4f98a2SYan Zheng unsigned int found_file_extent:1; 1595d4f98a2SYan Zheng }; 1605d4f98a2SYan Zheng 1615d4f98a2SYan Zheng /* stages of data relocation */ 1625d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1635d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1645d4f98a2SYan Zheng 1659569cc20SQu Wenruo static void mark_block_processed(struct reloc_control *rc, 166a26195a5SQu Wenruo struct btrfs_backref_node *node) 1679569cc20SQu Wenruo { 1689569cc20SQu Wenruo u32 blocksize; 1699569cc20SQu Wenruo 1709569cc20SQu Wenruo if (node->level == 0 || 1719569cc20SQu Wenruo in_range(node->bytenr, rc->block_group->start, 1729569cc20SQu Wenruo rc->block_group->length)) { 1739569cc20SQu Wenruo blocksize = rc->extent_root->fs_info->nodesize; 1749569cc20SQu Wenruo set_extent_bits(&rc->processed_blocks, node->bytenr, 1759569cc20SQu Wenruo node->bytenr + blocksize - 1, EXTENT_DIRTY); 1769569cc20SQu Wenruo } 1779569cc20SQu Wenruo node->processed = 1; 1789569cc20SQu Wenruo } 1799569cc20SQu Wenruo 1805d4f98a2SYan Zheng 1815d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1825d4f98a2SYan Zheng { 1836bef4d31SEric Paris tree->rb_root = RB_ROOT; 1845d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1855d4f98a2SYan Zheng } 1865d4f98a2SYan Zheng 1875d4f98a2SYan Zheng /* 1885d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 1895d4f98a2SYan Zheng */ 190a26195a5SQu Wenruo static struct btrfs_backref_node *walk_up_backref( 191a26195a5SQu Wenruo struct btrfs_backref_node *node, 192a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 1935d4f98a2SYan Zheng { 194a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 1955d4f98a2SYan Zheng int idx = *index; 1965d4f98a2SYan Zheng 1975d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 1985d4f98a2SYan Zheng edge = list_entry(node->upper.next, 199a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 2005d4f98a2SYan Zheng edges[idx++] = edge; 2015d4f98a2SYan Zheng node = edge->node[UPPER]; 2025d4f98a2SYan Zheng } 2033fd0a558SYan, Zheng BUG_ON(node->detached); 2045d4f98a2SYan Zheng *index = idx; 2055d4f98a2SYan Zheng return node; 2065d4f98a2SYan Zheng } 2075d4f98a2SYan Zheng 2085d4f98a2SYan Zheng /* 2095d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 2105d4f98a2SYan Zheng */ 211a26195a5SQu Wenruo static struct btrfs_backref_node *walk_down_backref( 212a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 2135d4f98a2SYan Zheng { 214a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 215a26195a5SQu Wenruo struct btrfs_backref_node *lower; 2165d4f98a2SYan Zheng int idx = *index; 2175d4f98a2SYan Zheng 2185d4f98a2SYan Zheng while (idx > 0) { 2195d4f98a2SYan Zheng edge = edges[idx - 1]; 2205d4f98a2SYan Zheng lower = edge->node[LOWER]; 2215d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 2225d4f98a2SYan Zheng idx--; 2235d4f98a2SYan Zheng continue; 2245d4f98a2SYan Zheng } 2255d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 226a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 2275d4f98a2SYan Zheng edges[idx - 1] = edge; 2285d4f98a2SYan Zheng *index = idx; 2295d4f98a2SYan Zheng return edge->node[UPPER]; 2305d4f98a2SYan Zheng } 2315d4f98a2SYan Zheng *index = 0; 2325d4f98a2SYan Zheng return NULL; 2335d4f98a2SYan Zheng } 2345d4f98a2SYan Zheng 235a26195a5SQu Wenruo static void update_backref_node(struct btrfs_backref_cache *cache, 236a26195a5SQu Wenruo struct btrfs_backref_node *node, u64 bytenr) 2373fd0a558SYan, Zheng { 2383fd0a558SYan, Zheng struct rb_node *rb_node; 2393fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 2403fd0a558SYan, Zheng node->bytenr = bytenr; 241e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node); 24243c04fb1SJeff Mahoney if (rb_node) 243982c92cbSQu Wenruo btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST); 2443fd0a558SYan, Zheng } 2453fd0a558SYan, Zheng 2463fd0a558SYan, Zheng /* 2473fd0a558SYan, Zheng * update backref cache after a transaction commit 2483fd0a558SYan, Zheng */ 2493fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 250a26195a5SQu Wenruo struct btrfs_backref_cache *cache) 2513fd0a558SYan, Zheng { 252a26195a5SQu Wenruo struct btrfs_backref_node *node; 2533fd0a558SYan, Zheng int level = 0; 2543fd0a558SYan, Zheng 2553fd0a558SYan, Zheng if (cache->last_trans == 0) { 2563fd0a558SYan, Zheng cache->last_trans = trans->transid; 2573fd0a558SYan, Zheng return 0; 2583fd0a558SYan, Zheng } 2593fd0a558SYan, Zheng 2603fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 2613fd0a558SYan, Zheng return 0; 2623fd0a558SYan, Zheng 2633fd0a558SYan, Zheng /* 2643fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 2653fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 2663fd0a558SYan, Zheng * so the detached nodes are no longer useful. 2673fd0a558SYan, Zheng */ 2683fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2693fd0a558SYan, Zheng node = list_entry(cache->detached.next, 270a26195a5SQu Wenruo struct btrfs_backref_node, list); 271023acb07SQu Wenruo btrfs_backref_cleanup_node(cache, node); 2723fd0a558SYan, Zheng } 2733fd0a558SYan, Zheng 2743fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 2753fd0a558SYan, Zheng node = list_entry(cache->changed.next, 276a26195a5SQu Wenruo struct btrfs_backref_node, list); 2773fd0a558SYan, Zheng list_del_init(&node->list); 2783fd0a558SYan, Zheng BUG_ON(node->pending); 2793fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2803fd0a558SYan, Zheng } 2813fd0a558SYan, Zheng 2823fd0a558SYan, Zheng /* 2833fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 2843fd0a558SYan, Zheng * errors during processing the pending nodes. 2853fd0a558SYan, Zheng */ 2863fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 2873fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 2883fd0a558SYan, Zheng BUG_ON(!node->pending); 2893fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 2903fd0a558SYan, Zheng continue; 2913fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2923fd0a558SYan, Zheng } 2933fd0a558SYan, Zheng } 2943fd0a558SYan, Zheng 2953fd0a558SYan, Zheng cache->last_trans = 0; 2963fd0a558SYan, Zheng return 1; 2973fd0a558SYan, Zheng } 2983fd0a558SYan, Zheng 2996282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 3006282675eSQu Wenruo { 3016282675eSQu Wenruo /* 3026282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 3036282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 3046282675eSQu Wenruo * trying to access reloc_root 3056282675eSQu Wenruo */ 3066282675eSQu Wenruo smp_rmb(); 3076282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 3086282675eSQu Wenruo return true; 3096282675eSQu Wenruo return false; 3106282675eSQu Wenruo } 3116282675eSQu Wenruo 3126282675eSQu Wenruo /* 3136282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 3146282675eSQu Wenruo * 3156282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 3166282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 31755465730SQu Wenruo * from no reloc root. But btrfs_should_ignore_reloc_root() below is a 31855465730SQu Wenruo * special case. 3196282675eSQu Wenruo */ 3206282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 3216282675eSQu Wenruo { 3226282675eSQu Wenruo if (reloc_root_is_dead(root)) 3236282675eSQu Wenruo return false; 3246282675eSQu Wenruo if (!root->reloc_root) 3256282675eSQu Wenruo return false; 3266282675eSQu Wenruo return true; 3276282675eSQu Wenruo } 328f2a97a9dSDavid Sterba 32955465730SQu Wenruo int btrfs_should_ignore_reloc_root(struct btrfs_root *root) 3303fd0a558SYan, Zheng { 3313fd0a558SYan, Zheng struct btrfs_root *reloc_root; 3323fd0a558SYan, Zheng 33392a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 3343fd0a558SYan, Zheng return 0; 3353fd0a558SYan, Zheng 3366282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 3376282675eSQu Wenruo if (reloc_root_is_dead(root)) 3386282675eSQu Wenruo return 1; 3396282675eSQu Wenruo 3403fd0a558SYan, Zheng reloc_root = root->reloc_root; 3413fd0a558SYan, Zheng if (!reloc_root) 3423fd0a558SYan, Zheng return 0; 3433fd0a558SYan, Zheng 3444d4225fcSJosef Bacik if (btrfs_header_generation(reloc_root->commit_root) == 3454d4225fcSJosef Bacik root->fs_info->running_transaction->transid) 3463fd0a558SYan, Zheng return 0; 3473fd0a558SYan, Zheng /* 3483fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 3493fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 3503fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 3513fd0a558SYan, Zheng * relocation. 3523fd0a558SYan, Zheng */ 3533fd0a558SYan, Zheng return 1; 3543fd0a558SYan, Zheng } 35555465730SQu Wenruo 3565d4f98a2SYan Zheng /* 3575d4f98a2SYan Zheng * find reloc tree by address of tree root 3585d4f98a2SYan Zheng */ 3592433bea5SQu Wenruo struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) 3605d4f98a2SYan Zheng { 3612433bea5SQu Wenruo struct reloc_control *rc = fs_info->reloc_ctl; 3625d4f98a2SYan Zheng struct rb_node *rb_node; 3635d4f98a2SYan Zheng struct mapping_node *node; 3645d4f98a2SYan Zheng struct btrfs_root *root = NULL; 3655d4f98a2SYan Zheng 3662433bea5SQu Wenruo ASSERT(rc); 3675d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 368e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr); 3695d4f98a2SYan Zheng if (rb_node) { 3705d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 3710d031dc4SYu Zhe root = node->data; 3725d4f98a2SYan Zheng } 3735d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 37400246528SJosef Bacik return btrfs_grab_root(root); 3755d4f98a2SYan Zheng } 3765d4f98a2SYan Zheng 3775d4f98a2SYan Zheng /* 37829db137bSQu Wenruo * For useless nodes, do two major clean ups: 37929db137bSQu Wenruo * 38029db137bSQu Wenruo * - Cleanup the children edges and nodes 38129db137bSQu Wenruo * If child node is also orphan (no parent) during cleanup, then the child 38229db137bSQu Wenruo * node will also be cleaned up. 38329db137bSQu Wenruo * 38429db137bSQu Wenruo * - Freeing up leaves (level 0), keeps nodes detached 38529db137bSQu Wenruo * For nodes, the node is still cached as "detached" 38629db137bSQu Wenruo * 38729db137bSQu Wenruo * Return false if @node is not in the @useless_nodes list. 38829db137bSQu Wenruo * Return true if @node is in the @useless_nodes list. 38929db137bSQu Wenruo */ 39029db137bSQu Wenruo static bool handle_useless_nodes(struct reloc_control *rc, 391a26195a5SQu Wenruo struct btrfs_backref_node *node) 39229db137bSQu Wenruo { 393a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 39429db137bSQu Wenruo struct list_head *useless_node = &cache->useless_node; 39529db137bSQu Wenruo bool ret = false; 39629db137bSQu Wenruo 39729db137bSQu Wenruo while (!list_empty(useless_node)) { 398a26195a5SQu Wenruo struct btrfs_backref_node *cur; 39929db137bSQu Wenruo 400a26195a5SQu Wenruo cur = list_first_entry(useless_node, struct btrfs_backref_node, 40129db137bSQu Wenruo list); 40229db137bSQu Wenruo list_del_init(&cur->list); 40329db137bSQu Wenruo 40429db137bSQu Wenruo /* Only tree root nodes can be added to @useless_nodes */ 40529db137bSQu Wenruo ASSERT(list_empty(&cur->upper)); 40629db137bSQu Wenruo 40729db137bSQu Wenruo if (cur == node) 40829db137bSQu Wenruo ret = true; 40929db137bSQu Wenruo 41029db137bSQu Wenruo /* The node is the lowest node */ 41129db137bSQu Wenruo if (cur->lowest) { 41229db137bSQu Wenruo list_del_init(&cur->lower); 41329db137bSQu Wenruo cur->lowest = 0; 41429db137bSQu Wenruo } 41529db137bSQu Wenruo 41629db137bSQu Wenruo /* Cleanup the lower edges */ 41729db137bSQu Wenruo while (!list_empty(&cur->lower)) { 418a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 419a26195a5SQu Wenruo struct btrfs_backref_node *lower; 42029db137bSQu Wenruo 42129db137bSQu Wenruo edge = list_entry(cur->lower.next, 422a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 42329db137bSQu Wenruo list_del(&edge->list[UPPER]); 42429db137bSQu Wenruo list_del(&edge->list[LOWER]); 42529db137bSQu Wenruo lower = edge->node[LOWER]; 426741188d3SQu Wenruo btrfs_backref_free_edge(cache, edge); 42729db137bSQu Wenruo 42829db137bSQu Wenruo /* Child node is also orphan, queue for cleanup */ 42929db137bSQu Wenruo if (list_empty(&lower->upper)) 43029db137bSQu Wenruo list_add(&lower->list, useless_node); 43129db137bSQu Wenruo } 43229db137bSQu Wenruo /* Mark this block processed for relocation */ 43329db137bSQu Wenruo mark_block_processed(rc, cur); 43429db137bSQu Wenruo 43529db137bSQu Wenruo /* 43629db137bSQu Wenruo * Backref nodes for tree leaves are deleted from the cache. 43729db137bSQu Wenruo * Backref nodes for upper level tree blocks are left in the 43829db137bSQu Wenruo * cache to avoid unnecessary backref lookup. 43929db137bSQu Wenruo */ 44029db137bSQu Wenruo if (cur->level > 0) { 44129db137bSQu Wenruo list_add(&cur->list, &cache->detached); 44229db137bSQu Wenruo cur->detached = 1; 44329db137bSQu Wenruo } else { 44429db137bSQu Wenruo rb_erase(&cur->rb_node, &cache->rb_root); 445741188d3SQu Wenruo btrfs_backref_free_node(cache, cur); 44629db137bSQu Wenruo } 44729db137bSQu Wenruo } 44829db137bSQu Wenruo return ret; 44929db137bSQu Wenruo } 45029db137bSQu Wenruo 45129db137bSQu Wenruo /* 452e7d571c7SQu Wenruo * Build backref tree for a given tree block. Root of the backref tree 453e7d571c7SQu Wenruo * corresponds the tree block, leaves of the backref tree correspond roots of 454e7d571c7SQu Wenruo * b-trees that reference the tree block. 455e7d571c7SQu Wenruo * 456e7d571c7SQu Wenruo * The basic idea of this function is check backrefs of a given block to find 457e7d571c7SQu Wenruo * upper level blocks that reference the block, and then check backrefs of 458e7d571c7SQu Wenruo * these upper level blocks recursively. The recursion stops when tree root is 459e7d571c7SQu Wenruo * reached or backrefs for the block is cached. 460e7d571c7SQu Wenruo * 461e7d571c7SQu Wenruo * NOTE: if we find that backrefs for a block are cached, we know backrefs for 462e7d571c7SQu Wenruo * all upper level blocks that directly/indirectly reference the block are also 463e7d571c7SQu Wenruo * cached. 464e7d571c7SQu Wenruo */ 465a26195a5SQu Wenruo static noinline_for_stack struct btrfs_backref_node *build_backref_tree( 466e7d571c7SQu Wenruo struct reloc_control *rc, struct btrfs_key *node_key, 467e7d571c7SQu Wenruo int level, u64 bytenr) 468e7d571c7SQu Wenruo { 469e7d571c7SQu Wenruo struct btrfs_backref_iter *iter; 470a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 471e7d571c7SQu Wenruo /* For searching parent of TREE_BLOCK_REF */ 472e7d571c7SQu Wenruo struct btrfs_path *path; 473a26195a5SQu Wenruo struct btrfs_backref_node *cur; 474a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 475a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 476e7d571c7SQu Wenruo int ret; 477e7d571c7SQu Wenruo int err = 0; 478e7d571c7SQu Wenruo 479d68194b2SDavid Sterba iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info); 480e7d571c7SQu Wenruo if (!iter) 481e7d571c7SQu Wenruo return ERR_PTR(-ENOMEM); 482e7d571c7SQu Wenruo path = btrfs_alloc_path(); 483e7d571c7SQu Wenruo if (!path) { 484e7d571c7SQu Wenruo err = -ENOMEM; 485e7d571c7SQu Wenruo goto out; 486e7d571c7SQu Wenruo } 487e7d571c7SQu Wenruo 488b1818dabSQu Wenruo node = btrfs_backref_alloc_node(cache, bytenr, level); 489e7d571c7SQu Wenruo if (!node) { 490e7d571c7SQu Wenruo err = -ENOMEM; 491e7d571c7SQu Wenruo goto out; 492e7d571c7SQu Wenruo } 493e7d571c7SQu Wenruo 494e7d571c7SQu Wenruo node->lowest = 1; 495e7d571c7SQu Wenruo cur = node; 496e7d571c7SQu Wenruo 497e7d571c7SQu Wenruo /* Breadth-first search to build backref cache */ 498e7d571c7SQu Wenruo do { 4991b60d2ecSQu Wenruo ret = btrfs_backref_add_tree_node(cache, path, iter, node_key, 5001b60d2ecSQu Wenruo cur); 501e7d571c7SQu Wenruo if (ret < 0) { 502e7d571c7SQu Wenruo err = ret; 503e7d571c7SQu Wenruo goto out; 504e7d571c7SQu Wenruo } 505e7d571c7SQu Wenruo edge = list_first_entry_or_null(&cache->pending_edge, 506a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 507e7d571c7SQu Wenruo /* 508e7d571c7SQu Wenruo * The pending list isn't empty, take the first block to 509e7d571c7SQu Wenruo * process 510e7d571c7SQu Wenruo */ 511e7d571c7SQu Wenruo if (edge) { 5125d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 5135d4f98a2SYan Zheng cur = edge->node[UPPER]; 5145d4f98a2SYan Zheng } 515e7d571c7SQu Wenruo } while (edge); 5165d4f98a2SYan Zheng 5171f872924SQu Wenruo /* Finish the upper linkage of newly added edges/nodes */ 518fc997ed0SQu Wenruo ret = btrfs_backref_finish_upper_links(cache, node); 5191f872924SQu Wenruo if (ret < 0) { 5201f872924SQu Wenruo err = ret; 52175bfb9afSJosef Bacik goto out; 52275bfb9afSJosef Bacik } 52375bfb9afSJosef Bacik 52429db137bSQu Wenruo if (handle_useless_nodes(rc, node)) 5253fd0a558SYan, Zheng node = NULL; 5265d4f98a2SYan Zheng out: 52771f572a9SQu Wenruo btrfs_backref_iter_free(iter); 52871f572a9SQu Wenruo btrfs_free_path(path); 5295d4f98a2SYan Zheng if (err) { 5301b23ea18SQu Wenruo btrfs_backref_error_cleanup(cache, node); 5315d4f98a2SYan Zheng return ERR_PTR(err); 5325d4f98a2SYan Zheng } 53375bfb9afSJosef Bacik ASSERT(!node || !node->detached); 53484780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 53584780289SQu Wenruo list_empty(&cache->pending_edge)); 5365d4f98a2SYan Zheng return node; 5375d4f98a2SYan Zheng } 5385d4f98a2SYan Zheng 5395d4f98a2SYan Zheng /* 5403fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 5413fd0a558SYan, Zheng * the backref node is created by cloning backref node that 5423fd0a558SYan, Zheng * corresponds to root of source tree 5433fd0a558SYan, Zheng */ 5443fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 5453fd0a558SYan, Zheng struct reloc_control *rc, 5463fd0a558SYan, Zheng struct btrfs_root *src, 5473fd0a558SYan, Zheng struct btrfs_root *dest) 5483fd0a558SYan, Zheng { 5493fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 550a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 551a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 552a26195a5SQu Wenruo struct btrfs_backref_node *new_node; 553a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 554a26195a5SQu Wenruo struct btrfs_backref_edge *new_edge; 5553fd0a558SYan, Zheng struct rb_node *rb_node; 5563fd0a558SYan, Zheng 5573fd0a558SYan, Zheng if (cache->last_trans > 0) 5583fd0a558SYan, Zheng update_backref_cache(trans, cache); 5593fd0a558SYan, Zheng 560e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start); 5613fd0a558SYan, Zheng if (rb_node) { 562a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, rb_node); 5633fd0a558SYan, Zheng if (node->detached) 5643fd0a558SYan, Zheng node = NULL; 5653fd0a558SYan, Zheng else 5663fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 5673fd0a558SYan, Zheng } 5683fd0a558SYan, Zheng 5693fd0a558SYan, Zheng if (!node) { 570e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, 5713fd0a558SYan, Zheng reloc_root->commit_root->start); 5723fd0a558SYan, Zheng if (rb_node) { 573a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, 5743fd0a558SYan, Zheng rb_node); 5753fd0a558SYan, Zheng BUG_ON(node->detached); 5763fd0a558SYan, Zheng } 5773fd0a558SYan, Zheng } 5783fd0a558SYan, Zheng 5793fd0a558SYan, Zheng if (!node) 5803fd0a558SYan, Zheng return 0; 5813fd0a558SYan, Zheng 582b1818dabSQu Wenruo new_node = btrfs_backref_alloc_node(cache, dest->node->start, 583b1818dabSQu Wenruo node->level); 5843fd0a558SYan, Zheng if (!new_node) 5853fd0a558SYan, Zheng return -ENOMEM; 5863fd0a558SYan, Zheng 5873fd0a558SYan, Zheng new_node->lowest = node->lowest; 5886848ad64SYan, Zheng new_node->checked = 1; 58900246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 5900b530bc5SJosef Bacik ASSERT(new_node->root); 5913fd0a558SYan, Zheng 5923fd0a558SYan, Zheng if (!node->lowest) { 5933fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 59447254d07SQu Wenruo new_edge = btrfs_backref_alloc_edge(cache); 5953fd0a558SYan, Zheng if (!new_edge) 5963fd0a558SYan, Zheng goto fail; 5973fd0a558SYan, Zheng 598f39911e5SQu Wenruo btrfs_backref_link_edge(new_edge, edge->node[LOWER], 599f39911e5SQu Wenruo new_node, LINK_UPPER); 6003fd0a558SYan, Zheng } 60176b9e23dSMiao Xie } else { 60276b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 6033fd0a558SYan, Zheng } 6043fd0a558SYan, Zheng 605e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr, 6063fd0a558SYan, Zheng &new_node->rb_node); 60743c04fb1SJeff Mahoney if (rb_node) 608982c92cbSQu Wenruo btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST); 6093fd0a558SYan, Zheng 6103fd0a558SYan, Zheng if (!new_node->lowest) { 6113fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 6123fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 6133fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 6143fd0a558SYan, Zheng } 6153fd0a558SYan, Zheng } 6163fd0a558SYan, Zheng return 0; 6173fd0a558SYan, Zheng fail: 6183fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 6193fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 620a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 6213fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 622741188d3SQu Wenruo btrfs_backref_free_edge(cache, new_edge); 6233fd0a558SYan, Zheng } 624741188d3SQu Wenruo btrfs_backref_free_node(cache, new_node); 6253fd0a558SYan, Zheng return -ENOMEM; 6263fd0a558SYan, Zheng } 6273fd0a558SYan, Zheng 6283fd0a558SYan, Zheng /* 6295d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 6305d4f98a2SYan Zheng */ 631ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 6325d4f98a2SYan Zheng { 6330b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6345d4f98a2SYan Zheng struct rb_node *rb_node; 6355d4f98a2SYan Zheng struct mapping_node *node; 6360b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 6375d4f98a2SYan Zheng 6385d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 639ffd7b339SJeff Mahoney if (!node) 640ffd7b339SJeff Mahoney return -ENOMEM; 6415d4f98a2SYan Zheng 642ea287ab1SJosef Bacik node->bytenr = root->commit_root->start; 6435d4f98a2SYan Zheng node->data = root; 6445d4f98a2SYan Zheng 6455d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 646e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 6475d4f98a2SYan Zheng node->bytenr, &node->rb_node); 6485d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 649ffd7b339SJeff Mahoney if (rb_node) { 65057a304cfSJosef Bacik btrfs_err(fs_info, 6515d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 6525d163e0eSJeff Mahoney node->bytenr); 65357a304cfSJosef Bacik return -EEXIST; 654ffd7b339SJeff Mahoney } 6555d4f98a2SYan Zheng 6565d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 6575d4f98a2SYan Zheng return 0; 6585d4f98a2SYan Zheng } 6595d4f98a2SYan Zheng 6605d4f98a2SYan Zheng /* 661c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 6625d4f98a2SYan Zheng * mapping 6635d4f98a2SYan Zheng */ 664c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 6655d4f98a2SYan Zheng { 6660b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6675d4f98a2SYan Zheng struct rb_node *rb_node; 6685d4f98a2SYan Zheng struct mapping_node *node = NULL; 6690b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 670f44deb74SJosef Bacik bool put_ref = false; 6715d4f98a2SYan Zheng 67265c6e82bSQu Wenruo if (rc && root->node) { 6735d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 674e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 675ea287ab1SJosef Bacik root->commit_root->start); 676c974c464SWang Shilong if (rb_node) { 677c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 678c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 679ea287ab1SJosef Bacik RB_CLEAR_NODE(&node->rb_node); 680c974c464SWang Shilong } 681c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 682c78a10aeSJosef Bacik ASSERT(!node || (struct btrfs_root *)node->data == root); 683389305b2SQu Wenruo } 684c974c464SWang Shilong 685f44deb74SJosef Bacik /* 686f44deb74SJosef Bacik * We only put the reloc root here if it's on the list. There's a lot 687f44deb74SJosef Bacik * of places where the pattern is to splice the rc->reloc_roots, process 688f44deb74SJosef Bacik * the reloc roots, and then add the reloc root back onto 689f44deb74SJosef Bacik * rc->reloc_roots. If we call __del_reloc_root while it's off of the 690f44deb74SJosef Bacik * list we don't want the reference being dropped, because the guy 691f44deb74SJosef Bacik * messing with the list is in charge of the reference. 692f44deb74SJosef Bacik */ 6930b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 694f44deb74SJosef Bacik if (!list_empty(&root->root_list)) { 695f44deb74SJosef Bacik put_ref = true; 696c974c464SWang Shilong list_del_init(&root->root_list); 697f44deb74SJosef Bacik } 6980b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 699f44deb74SJosef Bacik if (put_ref) 700f44deb74SJosef Bacik btrfs_put_root(root); 701c974c464SWang Shilong kfree(node); 702c974c464SWang Shilong } 703c974c464SWang Shilong 704c974c464SWang Shilong /* 705c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 706c974c464SWang Shilong * mapping 707c974c464SWang Shilong */ 708ea287ab1SJosef Bacik static int __update_reloc_root(struct btrfs_root *root) 709c974c464SWang Shilong { 7100b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 711c974c464SWang Shilong struct rb_node *rb_node; 712c974c464SWang Shilong struct mapping_node *node = NULL; 7130b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 714c974c464SWang Shilong 715c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 716e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 717ea287ab1SJosef Bacik root->commit_root->start); 7185d4f98a2SYan Zheng if (rb_node) { 7195d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 7205d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 7215d4f98a2SYan Zheng } 7225d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 7235d4f98a2SYan Zheng 7248f71f3e0SLiu Bo if (!node) 7258f71f3e0SLiu Bo return 0; 7265d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 7275d4f98a2SYan Zheng 7285d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 729ea287ab1SJosef Bacik node->bytenr = root->node->start; 730e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 7315d4f98a2SYan Zheng node->bytenr, &node->rb_node); 7325d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 73343c04fb1SJeff Mahoney if (rb_node) 734982c92cbSQu Wenruo btrfs_backref_panic(fs_info, node->bytenr, -EEXIST); 7355d4f98a2SYan Zheng return 0; 7365d4f98a2SYan Zheng } 7375d4f98a2SYan Zheng 7383fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 7393fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 7405d4f98a2SYan Zheng { 7410b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 7425d4f98a2SYan Zheng struct btrfs_root *reloc_root; 7435d4f98a2SYan Zheng struct extent_buffer *eb; 7445d4f98a2SYan Zheng struct btrfs_root_item *root_item; 7455d4f98a2SYan Zheng struct btrfs_key root_key; 74684c50ba5SJosef Bacik int ret = 0; 74784c50ba5SJosef Bacik bool must_abort = false; 7485d4f98a2SYan Zheng 7495d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 75084c50ba5SJosef Bacik if (!root_item) 75184c50ba5SJosef Bacik return ERR_PTR(-ENOMEM); 7525d4f98a2SYan Zheng 7535d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 7545d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 7553fd0a558SYan, Zheng root_key.offset = objectid; 7565d4f98a2SYan Zheng 7573fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 758054570a1SFilipe Manana u64 commit_root_gen; 759054570a1SFilipe Manana 7603fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 7615d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 7625d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 76384c50ba5SJosef Bacik if (ret) 76484c50ba5SJosef Bacik goto fail; 76584c50ba5SJosef Bacik 766054570a1SFilipe Manana /* 767054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 768054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 769054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 770054570a1SFilipe Manana * either inside the critical section of a transaction commit 771054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 772054570a1SFilipe Manana * it's created before the transaction commit is started. 773054570a1SFilipe Manana */ 774054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 775054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 7763fd0a558SYan, Zheng } else { 7773fd0a558SYan, Zheng /* 7783fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 7793fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 7803fd0a558SYan, Zheng * modified after it was created have RELOC flag 7813fd0a558SYan, Zheng * set in their headers. so it's OK to not update 7823fd0a558SYan, Zheng * the 'last_snapshot'. 7833fd0a558SYan, Zheng */ 7843fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 7853fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 78684c50ba5SJosef Bacik if (ret) 78784c50ba5SJosef Bacik goto fail; 7883fd0a558SYan, Zheng } 7893fd0a558SYan, Zheng 79084c50ba5SJosef Bacik /* 79184c50ba5SJosef Bacik * We have changed references at this point, we must abort the 79284c50ba5SJosef Bacik * transaction if anything fails. 79384c50ba5SJosef Bacik */ 79484c50ba5SJosef Bacik must_abort = true; 79584c50ba5SJosef Bacik 7965d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 7975d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 7985d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 7995d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 8003fd0a558SYan, Zheng 8013fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 8023fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 8033fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 8043fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 805c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, 0); 8063fd0a558SYan, Zheng } 8075d4f98a2SYan Zheng 8085d4f98a2SYan Zheng btrfs_tree_unlock(eb); 8095d4f98a2SYan Zheng free_extent_buffer(eb); 8105d4f98a2SYan Zheng 8110b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 8125d4f98a2SYan Zheng &root_key, root_item); 81384c50ba5SJosef Bacik if (ret) 81484c50ba5SJosef Bacik goto fail; 81584c50ba5SJosef Bacik 8165d4f98a2SYan Zheng kfree(root_item); 8175d4f98a2SYan Zheng 8183dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 81984c50ba5SJosef Bacik if (IS_ERR(reloc_root)) { 82084c50ba5SJosef Bacik ret = PTR_ERR(reloc_root); 82184c50ba5SJosef Bacik goto abort; 82284c50ba5SJosef Bacik } 82392a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 8245d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 8253fd0a558SYan, Zheng return reloc_root; 82684c50ba5SJosef Bacik fail: 82784c50ba5SJosef Bacik kfree(root_item); 82884c50ba5SJosef Bacik abort: 82984c50ba5SJosef Bacik if (must_abort) 83084c50ba5SJosef Bacik btrfs_abort_transaction(trans, ret); 83184c50ba5SJosef Bacik return ERR_PTR(ret); 8323fd0a558SYan, Zheng } 8333fd0a558SYan, Zheng 8343fd0a558SYan, Zheng /* 8353fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 8363fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 837f44deb74SJosef Bacik * 838f44deb74SJosef Bacik * The reloc_root comes out of here with two references, one for 839f44deb74SJosef Bacik * root->reloc_root, and another for being on the rc->reloc_roots list. 8403fd0a558SYan, Zheng */ 8413fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 8423fd0a558SYan, Zheng struct btrfs_root *root) 8433fd0a558SYan, Zheng { 8440b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8453fd0a558SYan, Zheng struct btrfs_root *reloc_root; 8460b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 84720dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 8483fd0a558SYan, Zheng int clear_rsv = 0; 849ffd7b339SJeff Mahoney int ret; 8503fd0a558SYan, Zheng 851aec7db3bSJosef Bacik if (!rc) 8522abc726aSJosef Bacik return 0; 8532abc726aSJosef Bacik 8541fac4a54SQu Wenruo /* 8551fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 8561fac4a54SQu Wenruo * create/update the dead reloc tree 8571fac4a54SQu Wenruo */ 8586282675eSQu Wenruo if (reloc_root_is_dead(root)) 8591fac4a54SQu Wenruo return 0; 8601fac4a54SQu Wenruo 861aec7db3bSJosef Bacik /* 862aec7db3bSJosef Bacik * This is subtle but important. We do not do 863aec7db3bSJosef Bacik * record_root_in_transaction for reloc roots, instead we record their 864aec7db3bSJosef Bacik * corresponding fs root, and then here we update the last trans for the 865aec7db3bSJosef Bacik * reloc root. This means that we have to do this for the entire life 866aec7db3bSJosef Bacik * of the reloc root, regardless of which stage of the relocation we are 867aec7db3bSJosef Bacik * in. 868aec7db3bSJosef Bacik */ 8693fd0a558SYan, Zheng if (root->reloc_root) { 8703fd0a558SYan, Zheng reloc_root = root->reloc_root; 8713fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 8723fd0a558SYan, Zheng return 0; 8733fd0a558SYan, Zheng } 8743fd0a558SYan, Zheng 875aec7db3bSJosef Bacik /* 876aec7db3bSJosef Bacik * We are merging reloc roots, we do not need new reloc trees. Also 877aec7db3bSJosef Bacik * reloc trees never need their own reloc tree. 878aec7db3bSJosef Bacik */ 879aec7db3bSJosef Bacik if (!rc->create_reloc_tree || 880aec7db3bSJosef Bacik root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 881aec7db3bSJosef Bacik return 0; 882aec7db3bSJosef Bacik 88320dd2cbfSMiao Xie if (!trans->reloc_reserved) { 88420dd2cbfSMiao Xie rsv = trans->block_rsv; 8853fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 8863fd0a558SYan, Zheng clear_rsv = 1; 8873fd0a558SYan, Zheng } 8883fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 8893fd0a558SYan, Zheng if (clear_rsv) 89020dd2cbfSMiao Xie trans->block_rsv = rsv; 89100bb36a0SJosef Bacik if (IS_ERR(reloc_root)) 89200bb36a0SJosef Bacik return PTR_ERR(reloc_root); 8935d4f98a2SYan Zheng 894ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 89557a304cfSJosef Bacik ASSERT(ret != -EEXIST); 89600bb36a0SJosef Bacik if (ret) { 89700bb36a0SJosef Bacik /* Pairs with create_reloc_root */ 89800bb36a0SJosef Bacik btrfs_put_root(reloc_root); 89900bb36a0SJosef Bacik return ret; 90000bb36a0SJosef Bacik } 901f44deb74SJosef Bacik root->reloc_root = btrfs_grab_root(reloc_root); 9025d4f98a2SYan Zheng return 0; 9035d4f98a2SYan Zheng } 9045d4f98a2SYan Zheng 9055d4f98a2SYan Zheng /* 9065d4f98a2SYan Zheng * update root item of reloc tree 9075d4f98a2SYan Zheng */ 9085d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 9095d4f98a2SYan Zheng struct btrfs_root *root) 9105d4f98a2SYan Zheng { 9110b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 9125d4f98a2SYan Zheng struct btrfs_root *reloc_root; 9135d4f98a2SYan Zheng struct btrfs_root_item *root_item; 9145d4f98a2SYan Zheng int ret; 9155d4f98a2SYan Zheng 9166282675eSQu Wenruo if (!have_reloc_root(root)) 917592fbcd5SJosef Bacik return 0; 9185d4f98a2SYan Zheng 9195d4f98a2SYan Zheng reloc_root = root->reloc_root; 9205d4f98a2SYan Zheng root_item = &reloc_root->root_item; 9215d4f98a2SYan Zheng 922f44deb74SJosef Bacik /* 923f44deb74SJosef Bacik * We are probably ok here, but __del_reloc_root() will drop its ref of 924f44deb74SJosef Bacik * the root. We have the ref for root->reloc_root, but just in case 925f44deb74SJosef Bacik * hold it while we update the reloc root. 926f44deb74SJosef Bacik */ 927f44deb74SJosef Bacik btrfs_grab_root(reloc_root); 928f44deb74SJosef Bacik 929d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 9300b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 9313fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 932d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 9336282675eSQu Wenruo /* 9346282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 9356282675eSQu Wenruo * have_reloc_root will not touch it from now on. 9366282675eSQu Wenruo */ 9376282675eSQu Wenruo smp_wmb(); 938c974c464SWang Shilong __del_reloc_root(reloc_root); 9395d4f98a2SYan Zheng } 9405d4f98a2SYan Zheng 9415d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 942ea287ab1SJosef Bacik __update_reloc_root(reloc_root); 9435d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 9445d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 9455d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 9465d4f98a2SYan Zheng } 9475d4f98a2SYan Zheng 9480b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 9495d4f98a2SYan Zheng &reloc_root->root_key, root_item); 950f44deb74SJosef Bacik btrfs_put_root(reloc_root); 951592fbcd5SJosef Bacik return ret; 9525d4f98a2SYan Zheng } 9535d4f98a2SYan Zheng 9545d4f98a2SYan Zheng /* 9555d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 9565d4f98a2SYan Zheng * in a subvolume 9575d4f98a2SYan Zheng */ 9585d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 9595d4f98a2SYan Zheng { 9605d4f98a2SYan Zheng struct rb_node *node; 9615d4f98a2SYan Zheng struct rb_node *prev; 9625d4f98a2SYan Zheng struct btrfs_inode *entry; 9635d4f98a2SYan Zheng struct inode *inode; 9645d4f98a2SYan Zheng 9655d4f98a2SYan Zheng spin_lock(&root->inode_lock); 9665d4f98a2SYan Zheng again: 9675d4f98a2SYan Zheng node = root->inode_tree.rb_node; 9685d4f98a2SYan Zheng prev = NULL; 9695d4f98a2SYan Zheng while (node) { 9705d4f98a2SYan Zheng prev = node; 9715d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9725d4f98a2SYan Zheng 9734a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 9745d4f98a2SYan Zheng node = node->rb_left; 9754a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 9765d4f98a2SYan Zheng node = node->rb_right; 9775d4f98a2SYan Zheng else 9785d4f98a2SYan Zheng break; 9795d4f98a2SYan Zheng } 9805d4f98a2SYan Zheng if (!node) { 9815d4f98a2SYan Zheng while (prev) { 9825d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 9834a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 9845d4f98a2SYan Zheng node = prev; 9855d4f98a2SYan Zheng break; 9865d4f98a2SYan Zheng } 9875d4f98a2SYan Zheng prev = rb_next(prev); 9885d4f98a2SYan Zheng } 9895d4f98a2SYan Zheng } 9905d4f98a2SYan Zheng while (node) { 9915d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9925d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 9935d4f98a2SYan Zheng if (inode) { 9945d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 9955d4f98a2SYan Zheng return inode; 9965d4f98a2SYan Zheng } 9975d4f98a2SYan Zheng 9984a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 9995d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 10005d4f98a2SYan Zheng goto again; 10015d4f98a2SYan Zheng 10025d4f98a2SYan Zheng node = rb_next(node); 10035d4f98a2SYan Zheng } 10045d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 10055d4f98a2SYan Zheng return NULL; 10065d4f98a2SYan Zheng } 10075d4f98a2SYan Zheng 10085d4f98a2SYan Zheng /* 10095d4f98a2SYan Zheng * get new location of data 10105d4f98a2SYan Zheng */ 10115d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 10125d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 10135d4f98a2SYan Zheng { 10145d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 10155d4f98a2SYan Zheng struct btrfs_path *path; 10165d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10175d4f98a2SYan Zheng struct extent_buffer *leaf; 10185d4f98a2SYan Zheng int ret; 10195d4f98a2SYan Zheng 10205d4f98a2SYan Zheng path = btrfs_alloc_path(); 10215d4f98a2SYan Zheng if (!path) 10225d4f98a2SYan Zheng return -ENOMEM; 10235d4f98a2SYan Zheng 10245d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1025f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1026f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 10275d4f98a2SYan Zheng if (ret < 0) 10285d4f98a2SYan Zheng goto out; 10295d4f98a2SYan Zheng if (ret > 0) { 10305d4f98a2SYan Zheng ret = -ENOENT; 10315d4f98a2SYan Zheng goto out; 10325d4f98a2SYan Zheng } 10335d4f98a2SYan Zheng 10345d4f98a2SYan Zheng leaf = path->nodes[0]; 10355d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 10365d4f98a2SYan Zheng struct btrfs_file_extent_item); 10375d4f98a2SYan Zheng 10385d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 10395d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 10405d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 10415d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 10425d4f98a2SYan Zheng 10435d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 104483d4cfd4SJosef Bacik ret = -EINVAL; 10455d4f98a2SYan Zheng goto out; 10465d4f98a2SYan Zheng } 10475d4f98a2SYan Zheng 10485d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 10495d4f98a2SYan Zheng ret = 0; 10505d4f98a2SYan Zheng out: 10515d4f98a2SYan Zheng btrfs_free_path(path); 10525d4f98a2SYan Zheng return ret; 10535d4f98a2SYan Zheng } 10545d4f98a2SYan Zheng 10555d4f98a2SYan Zheng /* 10565d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 10575d4f98a2SYan Zheng * the new locations. 10585d4f98a2SYan Zheng */ 10593fd0a558SYan, Zheng static noinline_for_stack 10603fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 10615d4f98a2SYan Zheng struct reloc_control *rc, 10625d4f98a2SYan Zheng struct btrfs_root *root, 10633fd0a558SYan, Zheng struct extent_buffer *leaf) 10645d4f98a2SYan Zheng { 10650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 10665d4f98a2SYan Zheng struct btrfs_key key; 10675d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10685d4f98a2SYan Zheng struct inode *inode = NULL; 10695d4f98a2SYan Zheng u64 parent; 10705d4f98a2SYan Zheng u64 bytenr; 10713fd0a558SYan, Zheng u64 new_bytenr = 0; 10725d4f98a2SYan Zheng u64 num_bytes; 10735d4f98a2SYan Zheng u64 end; 10745d4f98a2SYan Zheng u32 nritems; 10755d4f98a2SYan Zheng u32 i; 107683d4cfd4SJosef Bacik int ret = 0; 10775d4f98a2SYan Zheng int first = 1; 10785d4f98a2SYan Zheng int dirty = 0; 10795d4f98a2SYan Zheng 10805d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 10815d4f98a2SYan Zheng return 0; 10825d4f98a2SYan Zheng 10835d4f98a2SYan Zheng /* reloc trees always use full backref */ 10845d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 10855d4f98a2SYan Zheng parent = leaf->start; 10865d4f98a2SYan Zheng else 10875d4f98a2SYan Zheng parent = 0; 10885d4f98a2SYan Zheng 10895d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 10905d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 109182fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 109282fa113fSQu Wenruo 10935d4f98a2SYan Zheng cond_resched(); 10945d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 10955d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 10965d4f98a2SYan Zheng continue; 10975d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 10985d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 10995d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 11005d4f98a2SYan Zheng continue; 11015d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 11025d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 11035d4f98a2SYan Zheng if (bytenr == 0) 11045d4f98a2SYan Zheng continue; 11059569cc20SQu Wenruo if (!in_range(bytenr, rc->block_group->start, 11069569cc20SQu Wenruo rc->block_group->length)) 11075d4f98a2SYan Zheng continue; 11085d4f98a2SYan Zheng 11095d4f98a2SYan Zheng /* 1110fb12489bSMatthew Wilcox (Oracle) * if we are modifying block in fs tree, wait for read_folio 11115d4f98a2SYan Zheng * to complete and drop the extent cache 11125d4f98a2SYan Zheng */ 11135d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 11145d4f98a2SYan Zheng if (first) { 11155d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11165d4f98a2SYan Zheng first = 0; 11174a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 11183fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 11195d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11205d4f98a2SYan Zheng } 11214a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 11229c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 11239c5c9604SJosef Bacik 11245d4f98a2SYan Zheng end = key.offset + 11255d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 11265d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 11270b246afaSJeff Mahoney fs_info->sectorsize)); 11280b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 11295d4f98a2SYan Zheng end--; 11305d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 11319c5c9604SJosef Bacik key.offset, end, 11329c5c9604SJosef Bacik &cached_state); 11335d4f98a2SYan Zheng if (!ret) 11345d4f98a2SYan Zheng continue; 11355d4f98a2SYan Zheng 11364c0c8cfcSFilipe Manana btrfs_drop_extent_map_range(BTRFS_I(inode), 11374c0c8cfcSFilipe Manana key.offset, end, true); 11385d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 11399c5c9604SJosef Bacik key.offset, end, &cached_state); 11405d4f98a2SYan Zheng } 11415d4f98a2SYan Zheng } 11425d4f98a2SYan Zheng 11435d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 11445d4f98a2SYan Zheng bytenr, num_bytes); 114583d4cfd4SJosef Bacik if (ret) { 114683d4cfd4SJosef Bacik /* 114783d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 114883d4cfd4SJosef Bacik * in the file extent yet. 114983d4cfd4SJosef Bacik */ 115083d4cfd4SJosef Bacik break; 11513fd0a558SYan, Zheng } 11525d4f98a2SYan Zheng 11535d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 11545d4f98a2SYan Zheng dirty = 1; 11555d4f98a2SYan Zheng 11565d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 115782fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 115882fa113fSQu Wenruo num_bytes, parent); 115982fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1160f42c5da6SNikolay Borisov key.objectid, key.offset, 1161f42c5da6SNikolay Borisov root->root_key.objectid, false); 116282fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 116383d4cfd4SJosef Bacik if (ret) { 116466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 116583d4cfd4SJosef Bacik break; 116683d4cfd4SJosef Bacik } 11675d4f98a2SYan Zheng 1168ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1169ffd4bb2aSQu Wenruo num_bytes, parent); 1170ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1171f42c5da6SNikolay Borisov key.objectid, key.offset, 1172f42c5da6SNikolay Borisov root->root_key.objectid, false); 1173ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 117483d4cfd4SJosef Bacik if (ret) { 117566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 117683d4cfd4SJosef Bacik break; 117783d4cfd4SJosef Bacik } 11785d4f98a2SYan Zheng } 11795d4f98a2SYan Zheng if (dirty) 11805d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 11813fd0a558SYan, Zheng if (inode) 11823fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 118383d4cfd4SJosef Bacik return ret; 11845d4f98a2SYan Zheng } 11855d4f98a2SYan Zheng 11865d4f98a2SYan Zheng static noinline_for_stack 11875d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 11885d4f98a2SYan Zheng struct btrfs_path *path, int level) 11895d4f98a2SYan Zheng { 11905d4f98a2SYan Zheng struct btrfs_disk_key key1; 11915d4f98a2SYan Zheng struct btrfs_disk_key key2; 11925d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 11935d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 11945d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 11955d4f98a2SYan Zheng } 11965d4f98a2SYan Zheng 11975d4f98a2SYan Zheng /* 11985d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 11995d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 12005d4f98a2SYan Zheng * reloc tree was create can be replaced. 12015d4f98a2SYan Zheng * 12025d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 12035d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 12045d4f98a2SYan Zheng * errors, a negative error number is returned. 12055d4f98a2SYan Zheng */ 12063fd0a558SYan, Zheng static noinline_for_stack 12073d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 12085d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 12095d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 12105d4f98a2SYan Zheng int lowest_level, int max_level) 12115d4f98a2SYan Zheng { 12120b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 12135d4f98a2SYan Zheng struct extent_buffer *eb; 12145d4f98a2SYan Zheng struct extent_buffer *parent; 121582fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 12165d4f98a2SYan Zheng struct btrfs_key key; 12175d4f98a2SYan Zheng u64 old_bytenr; 12185d4f98a2SYan Zheng u64 new_bytenr; 12195d4f98a2SYan Zheng u64 old_ptr_gen; 12205d4f98a2SYan Zheng u64 new_ptr_gen; 12215d4f98a2SYan Zheng u64 last_snapshot; 12225d4f98a2SYan Zheng u32 blocksize; 12233fd0a558SYan, Zheng int cow = 0; 12245d4f98a2SYan Zheng int level; 12255d4f98a2SYan Zheng int ret; 12265d4f98a2SYan Zheng int slot; 12275d4f98a2SYan Zheng 12287a9213a9SJosef Bacik ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 12297a9213a9SJosef Bacik ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 12305d4f98a2SYan Zheng 12315d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 12323fd0a558SYan, Zheng again: 12335d4f98a2SYan Zheng slot = path->slots[lowest_level]; 12345d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 12355d4f98a2SYan Zheng 12365d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 12375d4f98a2SYan Zheng level = btrfs_header_level(eb); 12385d4f98a2SYan Zheng 12395d4f98a2SYan Zheng if (level < lowest_level) { 12405d4f98a2SYan Zheng btrfs_tree_unlock(eb); 12415d4f98a2SYan Zheng free_extent_buffer(eb); 12425d4f98a2SYan Zheng return 0; 12435d4f98a2SYan Zheng } 12445d4f98a2SYan Zheng 12453fd0a558SYan, Zheng if (cow) { 12469631e4ccSJosef Bacik ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb, 12479631e4ccSJosef Bacik BTRFS_NESTING_COW); 124845b87c5dSJosef Bacik if (ret) { 124945b87c5dSJosef Bacik btrfs_tree_unlock(eb); 125045b87c5dSJosef Bacik free_extent_buffer(eb); 125145b87c5dSJosef Bacik return ret; 125245b87c5dSJosef Bacik } 12533fd0a558SYan, Zheng } 12545d4f98a2SYan Zheng 12555d4f98a2SYan Zheng if (next_key) { 12565d4f98a2SYan Zheng next_key->objectid = (u64)-1; 12575d4f98a2SYan Zheng next_key->type = (u8)-1; 12585d4f98a2SYan Zheng next_key->offset = (u64)-1; 12595d4f98a2SYan Zheng } 12605d4f98a2SYan Zheng 12615d4f98a2SYan Zheng parent = eb; 12625d4f98a2SYan Zheng while (1) { 12635d4f98a2SYan Zheng level = btrfs_header_level(parent); 12647a9213a9SJosef Bacik ASSERT(level >= lowest_level); 12655d4f98a2SYan Zheng 1266e3b83361SQu Wenruo ret = btrfs_bin_search(parent, &key, &slot); 1267cbca7d59SFilipe Manana if (ret < 0) 1268cbca7d59SFilipe Manana break; 12695d4f98a2SYan Zheng if (ret && slot > 0) 12705d4f98a2SYan Zheng slot--; 12715d4f98a2SYan Zheng 12725d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 12735d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 12745d4f98a2SYan Zheng 12755d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 12760b246afaSJeff Mahoney blocksize = fs_info->nodesize; 12775d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 12785d4f98a2SYan Zheng 12795d4f98a2SYan Zheng if (level <= max_level) { 12805d4f98a2SYan Zheng eb = path->nodes[level]; 12815d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 12825d4f98a2SYan Zheng path->slots[level]); 12835d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 12845d4f98a2SYan Zheng path->slots[level]); 12855d4f98a2SYan Zheng } else { 12865d4f98a2SYan Zheng new_bytenr = 0; 12875d4f98a2SYan Zheng new_ptr_gen = 0; 12885d4f98a2SYan Zheng } 12895d4f98a2SYan Zheng 1290fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 12915d4f98a2SYan Zheng ret = level; 12925d4f98a2SYan Zheng break; 12935d4f98a2SYan Zheng } 12945d4f98a2SYan Zheng 12955d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 12965d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 12973fd0a558SYan, Zheng if (level <= lowest_level) { 12985d4f98a2SYan Zheng ret = 0; 12995d4f98a2SYan Zheng break; 13005d4f98a2SYan Zheng } 13015d4f98a2SYan Zheng 13026b3426beSJosef Bacik eb = btrfs_read_node_slot(parent, slot); 130364c043deSLiu Bo if (IS_ERR(eb)) { 130464c043deSLiu Bo ret = PTR_ERR(eb); 1305264813acSLiu Bo break; 1306416bc658SJosef Bacik } 13075d4f98a2SYan Zheng btrfs_tree_lock(eb); 13083fd0a558SYan, Zheng if (cow) { 13095d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 13109631e4ccSJosef Bacik slot, &eb, 13119631e4ccSJosef Bacik BTRFS_NESTING_COW); 131245b87c5dSJosef Bacik if (ret) { 131345b87c5dSJosef Bacik btrfs_tree_unlock(eb); 131445b87c5dSJosef Bacik free_extent_buffer(eb); 131545b87c5dSJosef Bacik break; 131645b87c5dSJosef Bacik } 13175d4f98a2SYan Zheng } 13185d4f98a2SYan Zheng 13195d4f98a2SYan Zheng btrfs_tree_unlock(parent); 13205d4f98a2SYan Zheng free_extent_buffer(parent); 13215d4f98a2SYan Zheng 13225d4f98a2SYan Zheng parent = eb; 13235d4f98a2SYan Zheng continue; 13245d4f98a2SYan Zheng } 13255d4f98a2SYan Zheng 13263fd0a558SYan, Zheng if (!cow) { 13273fd0a558SYan, Zheng btrfs_tree_unlock(parent); 13283fd0a558SYan, Zheng free_extent_buffer(parent); 13293fd0a558SYan, Zheng cow = 1; 13303fd0a558SYan, Zheng goto again; 13313fd0a558SYan, Zheng } 13323fd0a558SYan, Zheng 13335d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 13345d4f98a2SYan Zheng path->slots[level]); 1335b3b4aa74SDavid Sterba btrfs_release_path(path); 13365d4f98a2SYan Zheng 13375d4f98a2SYan Zheng path->lowest_level = level; 1338b40130b2SJosef Bacik set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); 13395d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 1340b40130b2SJosef Bacik clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); 13415d4f98a2SYan Zheng path->lowest_level = 0; 13420e9873e2SJosef Bacik if (ret) { 13430e9873e2SJosef Bacik if (ret > 0) 13440e9873e2SJosef Bacik ret = -ENOENT; 13450e9873e2SJosef Bacik break; 13460e9873e2SJosef Bacik } 13475d4f98a2SYan Zheng 13485d4f98a2SYan Zheng /* 1349824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1350824d8dffSQu Wenruo * 1351824d8dffSQu Wenruo * We must trace both trees. 1352824d8dffSQu Wenruo * 1) Tree reloc subtree 1353824d8dffSQu Wenruo * If not traced, we will leak data numbers 1354824d8dffSQu Wenruo * 2) Fs subtree 1355824d8dffSQu Wenruo * If not traced, we will double count old data 1356f616f5cdSQu Wenruo * 1357f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1358f616f5cdSQu Wenruo * the swapped tree blocks. 1359f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1360f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1361824d8dffSQu Wenruo */ 1362370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1363370a11b8SQu Wenruo rc->block_group, parent, slot, 1364370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1365370a11b8SQu Wenruo last_snapshot); 1366370a11b8SQu Wenruo if (ret < 0) 1367370a11b8SQu Wenruo break; 1368824d8dffSQu Wenruo /* 13695d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 13705d4f98a2SYan Zheng */ 13715d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 13725d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 13735d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 13745d4f98a2SYan Zheng 13755d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 13765d4f98a2SYan Zheng path->slots[level], old_bytenr); 13775d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 13785d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 13795d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 13805d4f98a2SYan Zheng 138182fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 138282fa113fSQu Wenruo blocksize, path->nodes[level]->start); 1383f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1384f42c5da6SNikolay Borisov 0, true); 138582fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1386253e258cSJosef Bacik if (ret) { 1387253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1388253e258cSJosef Bacik break; 1389253e258cSJosef Bacik } 139082fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 139182fa113fSQu Wenruo blocksize, 0); 1392f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0, 1393f42c5da6SNikolay Borisov true); 139482fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1395253e258cSJosef Bacik if (ret) { 1396253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1397253e258cSJosef Bacik break; 1398253e258cSJosef Bacik } 13995d4f98a2SYan Zheng 1400ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1401ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1402f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1403f42c5da6SNikolay Borisov 0, true); 1404ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1405253e258cSJosef Bacik if (ret) { 1406253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1407253e258cSJosef Bacik break; 1408253e258cSJosef Bacik } 14095d4f98a2SYan Zheng 1410ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1411ffd4bb2aSQu Wenruo blocksize, 0); 1412f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 1413f42c5da6SNikolay Borisov 0, true); 1414ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1415253e258cSJosef Bacik if (ret) { 1416253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1417253e258cSJosef Bacik break; 1418253e258cSJosef Bacik } 14195d4f98a2SYan Zheng 14205d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 14215d4f98a2SYan Zheng 14225d4f98a2SYan Zheng ret = level; 14235d4f98a2SYan Zheng break; 14245d4f98a2SYan Zheng } 14255d4f98a2SYan Zheng btrfs_tree_unlock(parent); 14265d4f98a2SYan Zheng free_extent_buffer(parent); 14275d4f98a2SYan Zheng return ret; 14285d4f98a2SYan Zheng } 14295d4f98a2SYan Zheng 14305d4f98a2SYan Zheng /* 14315d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 14325d4f98a2SYan Zheng */ 14335d4f98a2SYan Zheng static noinline_for_stack 14345d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14355d4f98a2SYan Zheng int *level) 14365d4f98a2SYan Zheng { 14375d4f98a2SYan Zheng struct extent_buffer *eb; 14385d4f98a2SYan Zheng int i; 14395d4f98a2SYan Zheng u64 last_snapshot; 14405d4f98a2SYan Zheng u32 nritems; 14415d4f98a2SYan Zheng 14425d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14435d4f98a2SYan Zheng 14445d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 14455d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14465d4f98a2SYan Zheng path->nodes[i] = NULL; 14475d4f98a2SYan Zheng } 14485d4f98a2SYan Zheng 14495d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 14505d4f98a2SYan Zheng eb = path->nodes[i]; 14515d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14525d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 14535d4f98a2SYan Zheng path->slots[i]++; 14545d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 14555d4f98a2SYan Zheng last_snapshot) 14565d4f98a2SYan Zheng continue; 14575d4f98a2SYan Zheng 14585d4f98a2SYan Zheng *level = i; 14595d4f98a2SYan Zheng return 0; 14605d4f98a2SYan Zheng } 14615d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14625d4f98a2SYan Zheng path->nodes[i] = NULL; 14635d4f98a2SYan Zheng } 14645d4f98a2SYan Zheng return 1; 14655d4f98a2SYan Zheng } 14665d4f98a2SYan Zheng 14675d4f98a2SYan Zheng /* 14685d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 14695d4f98a2SYan Zheng */ 14705d4f98a2SYan Zheng static noinline_for_stack 14715d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14725d4f98a2SYan Zheng int *level) 14735d4f98a2SYan Zheng { 14745d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 14755d4f98a2SYan Zheng int i; 14765d4f98a2SYan Zheng u64 ptr_gen = 0; 14775d4f98a2SYan Zheng u64 last_snapshot; 14785d4f98a2SYan Zheng u32 nritems; 14795d4f98a2SYan Zheng 14805d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14815d4f98a2SYan Zheng 14825d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 14835d4f98a2SYan Zheng eb = path->nodes[i]; 14845d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14855d4f98a2SYan Zheng while (path->slots[i] < nritems) { 14865d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 14875d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 14885d4f98a2SYan Zheng break; 14895d4f98a2SYan Zheng path->slots[i]++; 14905d4f98a2SYan Zheng } 14915d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 14925d4f98a2SYan Zheng if (i == *level) 14935d4f98a2SYan Zheng break; 14945d4f98a2SYan Zheng *level = i + 1; 14955d4f98a2SYan Zheng return 0; 14965d4f98a2SYan Zheng } 14975d4f98a2SYan Zheng if (i == 1) { 14985d4f98a2SYan Zheng *level = i; 14995d4f98a2SYan Zheng return 0; 15005d4f98a2SYan Zheng } 15015d4f98a2SYan Zheng 15028ef385bbSJosef Bacik eb = btrfs_read_node_slot(eb, path->slots[i]); 15038ef385bbSJosef Bacik if (IS_ERR(eb)) 150464c043deSLiu Bo return PTR_ERR(eb); 15055d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 15065d4f98a2SYan Zheng path->nodes[i - 1] = eb; 15075d4f98a2SYan Zheng path->slots[i - 1] = 0; 15085d4f98a2SYan Zheng } 15095d4f98a2SYan Zheng return 1; 15105d4f98a2SYan Zheng } 15115d4f98a2SYan Zheng 15125d4f98a2SYan Zheng /* 15135d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 15145d4f98a2SYan Zheng * [min_key, max_key) 15155d4f98a2SYan Zheng */ 15165d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 15175d4f98a2SYan Zheng struct btrfs_key *min_key, 15185d4f98a2SYan Zheng struct btrfs_key *max_key) 15195d4f98a2SYan Zheng { 15200b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15215d4f98a2SYan Zheng struct inode *inode = NULL; 15225d4f98a2SYan Zheng u64 objectid; 15235d4f98a2SYan Zheng u64 start, end; 152433345d01SLi Zefan u64 ino; 15255d4f98a2SYan Zheng 15265d4f98a2SYan Zheng objectid = min_key->objectid; 15275d4f98a2SYan Zheng while (1) { 15289c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 15299c5c9604SJosef Bacik 15305d4f98a2SYan Zheng cond_resched(); 15315d4f98a2SYan Zheng iput(inode); 15325d4f98a2SYan Zheng 15335d4f98a2SYan Zheng if (objectid > max_key->objectid) 15345d4f98a2SYan Zheng break; 15355d4f98a2SYan Zheng 15365d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 15375d4f98a2SYan Zheng if (!inode) 15385d4f98a2SYan Zheng break; 15394a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 15405d4f98a2SYan Zheng 154133345d01SLi Zefan if (ino > max_key->objectid) { 15425d4f98a2SYan Zheng iput(inode); 15435d4f98a2SYan Zheng break; 15445d4f98a2SYan Zheng } 15455d4f98a2SYan Zheng 154633345d01SLi Zefan objectid = ino + 1; 15475d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 15485d4f98a2SYan Zheng continue; 15495d4f98a2SYan Zheng 155033345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 15515d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 15525d4f98a2SYan Zheng continue; 15535d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 15545d4f98a2SYan Zheng start = 0; 15555d4f98a2SYan Zheng else { 15565d4f98a2SYan Zheng start = min_key->offset; 15570b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 15585d4f98a2SYan Zheng } 15595d4f98a2SYan Zheng } else { 15605d4f98a2SYan Zheng start = 0; 15615d4f98a2SYan Zheng } 15625d4f98a2SYan Zheng 156333345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 15645d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 15655d4f98a2SYan Zheng continue; 15665d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 15675d4f98a2SYan Zheng end = (u64)-1; 15685d4f98a2SYan Zheng } else { 15695d4f98a2SYan Zheng if (max_key->offset == 0) 15705d4f98a2SYan Zheng continue; 15715d4f98a2SYan Zheng end = max_key->offset; 15720b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 15735d4f98a2SYan Zheng end--; 15745d4f98a2SYan Zheng } 15755d4f98a2SYan Zheng } else { 15765d4f98a2SYan Zheng end = (u64)-1; 15775d4f98a2SYan Zheng } 15785d4f98a2SYan Zheng 1579fb12489bSMatthew Wilcox (Oracle) /* the lock_extent waits for read_folio to complete */ 15809c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 15814c0c8cfcSFilipe Manana btrfs_drop_extent_map_range(BTRFS_I(inode), start, end, true); 15829c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 15835d4f98a2SYan Zheng } 15845d4f98a2SYan Zheng return 0; 15855d4f98a2SYan Zheng } 15865d4f98a2SYan Zheng 15875d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 15885d4f98a2SYan Zheng struct btrfs_key *key) 15895d4f98a2SYan Zheng 15905d4f98a2SYan Zheng { 15915d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 15925d4f98a2SYan Zheng if (!path->nodes[level]) 15935d4f98a2SYan Zheng break; 15945d4f98a2SYan Zheng if (path->slots[level] + 1 < 15955d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 15965d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 15975d4f98a2SYan Zheng path->slots[level] + 1); 15985d4f98a2SYan Zheng return 0; 15995d4f98a2SYan Zheng } 16005d4f98a2SYan Zheng level++; 16015d4f98a2SYan Zheng } 16025d4f98a2SYan Zheng return 1; 16035d4f98a2SYan Zheng } 16045d4f98a2SYan Zheng 16055d4f98a2SYan Zheng /* 1606d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 1607d2311e69SQu Wenruo */ 1608ac54da6cSJosef Bacik static int insert_dirty_subvol(struct btrfs_trans_handle *trans, 1609d2311e69SQu Wenruo struct reloc_control *rc, 1610d2311e69SQu Wenruo struct btrfs_root *root) 1611d2311e69SQu Wenruo { 1612d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1613d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 16147934133fSJosef Bacik int ret; 1615d2311e69SQu Wenruo 1616d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 1617d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 1618d2311e69SQu Wenruo ASSERT(reloc_root); 1619d2311e69SQu Wenruo 1620d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 1621d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 1622d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 1623c8422684SDavid Sterba btrfs_set_root_drop_level(reloc_root_item, 0); 1624d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 16257934133fSJosef Bacik ret = btrfs_update_reloc_root(trans, root); 16267934133fSJosef Bacik if (ret) 16277934133fSJosef Bacik return ret; 1628d2311e69SQu Wenruo 1629d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 163000246528SJosef Bacik btrfs_grab_root(root); 1631d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 1632d2311e69SQu Wenruo } 1633ac54da6cSJosef Bacik 1634ac54da6cSJosef Bacik return 0; 1635d2311e69SQu Wenruo } 1636d2311e69SQu Wenruo 1637d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 1638d2311e69SQu Wenruo { 1639d2311e69SQu Wenruo struct btrfs_root *root; 1640d2311e69SQu Wenruo struct btrfs_root *next; 1641d2311e69SQu Wenruo int ret = 0; 164230d40577SQu Wenruo int ret2; 1643d2311e69SQu Wenruo 1644d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 1645d2311e69SQu Wenruo reloc_dirty_list) { 164630d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 164730d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 1648d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1649d2311e69SQu Wenruo 1650d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 1651d2311e69SQu Wenruo root->reloc_root = NULL; 16526282675eSQu Wenruo /* 16536282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 16546282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 16556282675eSQu Wenruo */ 16566282675eSQu Wenruo smp_wmb(); 16571fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 1658f28de8d8SJosef Bacik if (reloc_root) { 1659f44deb74SJosef Bacik /* 1660f44deb74SJosef Bacik * btrfs_drop_snapshot drops our ref we hold for 1661f44deb74SJosef Bacik * ->reloc_root. If it fails however we must 1662f44deb74SJosef Bacik * drop the ref ourselves. 1663f44deb74SJosef Bacik */ 1664f28de8d8SJosef Bacik ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); 1665f44deb74SJosef Bacik if (ret2 < 0) { 1666f44deb74SJosef Bacik btrfs_put_root(reloc_root); 1667f44deb74SJosef Bacik if (!ret) 1668f28de8d8SJosef Bacik ret = ret2; 1669f28de8d8SJosef Bacik } 1670f44deb74SJosef Bacik } 167100246528SJosef Bacik btrfs_put_root(root); 167230d40577SQu Wenruo } else { 167330d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 16740078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(root, 0, 1); 1675f44deb74SJosef Bacik if (ret2 < 0) { 1676f44deb74SJosef Bacik btrfs_put_root(root); 1677f44deb74SJosef Bacik if (!ret) 167830d40577SQu Wenruo ret = ret2; 167930d40577SQu Wenruo } 1680d2311e69SQu Wenruo } 1681f44deb74SJosef Bacik } 1682d2311e69SQu Wenruo return ret; 1683d2311e69SQu Wenruo } 1684d2311e69SQu Wenruo 1685d2311e69SQu Wenruo /* 16865d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 16875d4f98a2SYan Zheng * fs tree. 16885d4f98a2SYan Zheng */ 16895d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 16905d4f98a2SYan Zheng struct btrfs_root *root) 16915d4f98a2SYan Zheng { 16920b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 16935d4f98a2SYan Zheng struct btrfs_key key; 16945d4f98a2SYan Zheng struct btrfs_key next_key; 16959e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 16965d4f98a2SYan Zheng struct btrfs_root *reloc_root; 16975d4f98a2SYan Zheng struct btrfs_root_item *root_item; 16985d4f98a2SYan Zheng struct btrfs_path *path; 16993fd0a558SYan, Zheng struct extent_buffer *leaf; 1700fca3a45dSJosef Bacik int reserve_level; 17015d4f98a2SYan Zheng int level; 17025d4f98a2SYan Zheng int max_level; 17035d4f98a2SYan Zheng int replaced = 0; 1704c6a592f2SNikolay Borisov int ret = 0; 17053fd0a558SYan, Zheng u32 min_reserved; 17065d4f98a2SYan Zheng 17075d4f98a2SYan Zheng path = btrfs_alloc_path(); 17085d4f98a2SYan Zheng if (!path) 17095d4f98a2SYan Zheng return -ENOMEM; 1710e4058b54SDavid Sterba path->reada = READA_FORWARD; 17115d4f98a2SYan Zheng 17125d4f98a2SYan Zheng reloc_root = root->reloc_root; 17135d4f98a2SYan Zheng root_item = &reloc_root->root_item; 17145d4f98a2SYan Zheng 17155d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 17165d4f98a2SYan Zheng level = btrfs_root_level(root_item); 171767439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 17185d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 17195d4f98a2SYan Zheng path->slots[level] = 0; 17205d4f98a2SYan Zheng } else { 17215d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 17225d4f98a2SYan Zheng 1723c8422684SDavid Sterba level = btrfs_root_drop_level(root_item); 17245d4f98a2SYan Zheng BUG_ON(level == 0); 17255d4f98a2SYan Zheng path->lowest_level = level; 17265d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 172733c66f43SYan Zheng path->lowest_level = 0; 17285d4f98a2SYan Zheng if (ret < 0) { 17295d4f98a2SYan Zheng btrfs_free_path(path); 17305d4f98a2SYan Zheng return ret; 17315d4f98a2SYan Zheng } 17325d4f98a2SYan Zheng 17335d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 17345d4f98a2SYan Zheng path->slots[level]); 17355d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 17365d4f98a2SYan Zheng 17375d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 17385d4f98a2SYan Zheng } 17395d4f98a2SYan Zheng 174044d354abSQu Wenruo /* 174144d354abSQu Wenruo * In merge_reloc_root(), we modify the upper level pointer to swap the 174244d354abSQu Wenruo * tree blocks between reloc tree and subvolume tree. Thus for tree 174344d354abSQu Wenruo * block COW, we COW at most from level 1 to root level for each tree. 174444d354abSQu Wenruo * 174544d354abSQu Wenruo * Thus the needed metadata size is at most root_level * nodesize, 174644d354abSQu Wenruo * and * 2 since we have two trees to COW. 174744d354abSQu Wenruo */ 1748fca3a45dSJosef Bacik reserve_level = max_t(int, 1, btrfs_root_level(root_item)); 1749fca3a45dSJosef Bacik min_reserved = fs_info->nodesize * reserve_level * 2; 17505d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 17515d4f98a2SYan Zheng 17525d4f98a2SYan Zheng while (1) { 17539270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 17549270501cSJosef Bacik min_reserved, 175544d354abSQu Wenruo BTRFS_RESERVE_FLUSH_LIMIT); 1756c6a592f2SNikolay Borisov if (ret) 17579e6a0c52SJosef Bacik goto out; 17589e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 17599e6a0c52SJosef Bacik if (IS_ERR(trans)) { 1760c6a592f2SNikolay Borisov ret = PTR_ERR(trans); 17619e6a0c52SJosef Bacik trans = NULL; 17629e6a0c52SJosef Bacik goto out; 17639e6a0c52SJosef Bacik } 17642abc726aSJosef Bacik 17652abc726aSJosef Bacik /* 17662abc726aSJosef Bacik * At this point we no longer have a reloc_control, so we can't 17672abc726aSJosef Bacik * depend on btrfs_init_reloc_root to update our last_trans. 17682abc726aSJosef Bacik * 17692abc726aSJosef Bacik * But that's ok, we started the trans handle on our 17702abc726aSJosef Bacik * corresponding fs_root, which means it's been added to the 17712abc726aSJosef Bacik * dirty list. At commit time we'll still call 17722abc726aSJosef Bacik * btrfs_update_reloc_root() and update our root item 17732abc726aSJosef Bacik * appropriately. 17742abc726aSJosef Bacik */ 17752abc726aSJosef Bacik reloc_root->last_trans = trans->transid; 17769e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 17773fd0a558SYan, Zheng 17783fd0a558SYan, Zheng replaced = 0; 17795d4f98a2SYan Zheng max_level = level; 17805d4f98a2SYan Zheng 17815d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 1782c6a592f2SNikolay Borisov if (ret < 0) 17835d4f98a2SYan Zheng goto out; 17845d4f98a2SYan Zheng if (ret > 0) 17855d4f98a2SYan Zheng break; 17865d4f98a2SYan Zheng 17875d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 17885d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 17895d4f98a2SYan Zheng ret = 0; 17905d4f98a2SYan Zheng } else { 17913d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 17923fd0a558SYan, Zheng &next_key, level, max_level); 17935d4f98a2SYan Zheng } 1794c6a592f2SNikolay Borisov if (ret < 0) 17955d4f98a2SYan Zheng goto out; 17965d4f98a2SYan Zheng if (ret > 0) { 17975d4f98a2SYan Zheng level = ret; 17985d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 17995d4f98a2SYan Zheng path->slots[level]); 18005d4f98a2SYan Zheng replaced = 1; 18015d4f98a2SYan Zheng } 18025d4f98a2SYan Zheng 18035d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 18045d4f98a2SYan Zheng if (ret > 0) 18055d4f98a2SYan Zheng break; 18065d4f98a2SYan Zheng 18075d4f98a2SYan Zheng BUG_ON(level == 0); 18085d4f98a2SYan Zheng /* 18095d4f98a2SYan Zheng * save the merging progress in the drop_progress. 18105d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 18115d4f98a2SYan Zheng */ 18125d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 18135d4f98a2SYan Zheng path->slots[level]); 1814c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, level); 18155d4f98a2SYan Zheng 18163a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18179e6a0c52SJosef Bacik trans = NULL; 18185d4f98a2SYan Zheng 18192ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18205d4f98a2SYan Zheng 18215d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18225d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18235d4f98a2SYan Zheng } 18245d4f98a2SYan Zheng 18255d4f98a2SYan Zheng /* 18265d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 18275d4f98a2SYan Zheng * relocated and the block is tree root. 18285d4f98a2SYan Zheng */ 18295d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 18309631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf, 18319631e4ccSJosef Bacik BTRFS_NESTING_COW); 18325d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 18335d4f98a2SYan Zheng free_extent_buffer(leaf); 18345d4f98a2SYan Zheng out: 18355d4f98a2SYan Zheng btrfs_free_path(path); 18365d4f98a2SYan Zheng 1837ac54da6cSJosef Bacik if (ret == 0) { 1838ac54da6cSJosef Bacik ret = insert_dirty_subvol(trans, rc, root); 1839ac54da6cSJosef Bacik if (ret) 1840ac54da6cSJosef Bacik btrfs_abort_transaction(trans, ret); 1841ac54da6cSJosef Bacik } 18425d4f98a2SYan Zheng 18439e6a0c52SJosef Bacik if (trans) 18443a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18455d4f98a2SYan Zheng 18462ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18475d4f98a2SYan Zheng 18485d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18495d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18505d4f98a2SYan Zheng 1851c6a592f2SNikolay Borisov return ret; 18525d4f98a2SYan Zheng } 18535d4f98a2SYan Zheng 18543fd0a558SYan, Zheng static noinline_for_stack 18553fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 18565d4f98a2SYan Zheng { 18573fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 18580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 18593fd0a558SYan, Zheng struct btrfs_root *reloc_root; 18605d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 18613fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 18623fd0a558SYan, Zheng u64 num_bytes = 0; 18633fd0a558SYan, Zheng int ret; 18643fd0a558SYan, Zheng 18650b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 18660b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 18673fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 18680b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 18697585717fSChris Mason 18703fd0a558SYan, Zheng again: 18713fd0a558SYan, Zheng if (!err) { 18723fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 18739270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes, 187408e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 18753fd0a558SYan, Zheng if (ret) 18763fd0a558SYan, Zheng err = ret; 18773fd0a558SYan, Zheng } 18783fd0a558SYan, Zheng 18797a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 18803612b495STsutomu Itoh if (IS_ERR(trans)) { 18813612b495STsutomu Itoh if (!err) 18822ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 188363f018beSNikolay Borisov num_bytes, NULL); 18843612b495STsutomu Itoh return PTR_ERR(trans); 18853612b495STsutomu Itoh } 18863fd0a558SYan, Zheng 18873fd0a558SYan, Zheng if (!err) { 18883fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 18893a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18902ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 189163f018beSNikolay Borisov num_bytes, NULL); 18923fd0a558SYan, Zheng goto again; 18933fd0a558SYan, Zheng } 18943fd0a558SYan, Zheng } 18953fd0a558SYan, Zheng 18963fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 18973fd0a558SYan, Zheng 18983fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 18993fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 19003fd0a558SYan, Zheng struct btrfs_root, root_list); 19013fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 19023fd0a558SYan, Zheng 1903a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1904a820feb5SDavid Sterba false); 1905e0b085b0SJosef Bacik if (IS_ERR(root)) { 1906e0b085b0SJosef Bacik /* 1907e0b085b0SJosef Bacik * Even if we have an error we need this reloc root 1908e0b085b0SJosef Bacik * back on our list so we can clean up properly. 1909e0b085b0SJosef Bacik */ 1910e0b085b0SJosef Bacik list_add(&reloc_root->root_list, &reloc_roots); 1911e0b085b0SJosef Bacik btrfs_abort_transaction(trans, (int)PTR_ERR(root)); 1912e0b085b0SJosef Bacik if (!err) 1913e0b085b0SJosef Bacik err = PTR_ERR(root); 1914e0b085b0SJosef Bacik break; 1915e0b085b0SJosef Bacik } 1916e0b085b0SJosef Bacik ASSERT(root->reloc_root == reloc_root); 19173fd0a558SYan, Zheng 19183fd0a558SYan, Zheng /* 19193fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 19203fd0a558SYan, Zheng * knows it should resumes merging 19213fd0a558SYan, Zheng */ 19223fd0a558SYan, Zheng if (!err) 19233fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 1924bbae13f8SJosef Bacik ret = btrfs_update_reloc_root(trans, root); 19253fd0a558SYan, Zheng 1926bbae13f8SJosef Bacik /* 1927bbae13f8SJosef Bacik * Even if we have an error we need this reloc root back on our 1928bbae13f8SJosef Bacik * list so we can clean up properly. 1929bbae13f8SJosef Bacik */ 19303fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 193100246528SJosef Bacik btrfs_put_root(root); 1932bbae13f8SJosef Bacik 1933bbae13f8SJosef Bacik if (ret) { 1934bbae13f8SJosef Bacik btrfs_abort_transaction(trans, ret); 1935bbae13f8SJosef Bacik if (!err) 1936bbae13f8SJosef Bacik err = ret; 1937bbae13f8SJosef Bacik break; 1938bbae13f8SJosef Bacik } 19393fd0a558SYan, Zheng } 19403fd0a558SYan, Zheng 19413fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 19423fd0a558SYan, Zheng 19433fd0a558SYan, Zheng if (!err) 1944fb686c68SJosef Bacik err = btrfs_commit_transaction(trans); 19453fd0a558SYan, Zheng else 19463a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19473fd0a558SYan, Zheng return err; 19483fd0a558SYan, Zheng } 19493fd0a558SYan, Zheng 19503fd0a558SYan, Zheng static noinline_for_stack 1951aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 1952aca1bba6SLiu Bo { 1953a7571232SNikolay Borisov struct btrfs_root *reloc_root, *tmp; 1954aca1bba6SLiu Bo 1955a7571232SNikolay Borisov list_for_each_entry_safe(reloc_root, tmp, list, root_list) 1956bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 1957aca1bba6SLiu Bo } 1958aca1bba6SLiu Bo 1959aca1bba6SLiu Bo static noinline_for_stack 196094404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 19613fd0a558SYan, Zheng { 19620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 19635d4f98a2SYan Zheng struct btrfs_root *root; 19645d4f98a2SYan Zheng struct btrfs_root *reloc_root; 19653fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 19663fd0a558SYan, Zheng int found = 0; 1967aca1bba6SLiu Bo int ret = 0; 19683fd0a558SYan, Zheng again: 19693fd0a558SYan, Zheng root = rc->extent_root; 19707585717fSChris Mason 19717585717fSChris Mason /* 19727585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 19737585717fSChris Mason * we have to make sure nobody is in the middle of 19747585717fSChris Mason * adding their roots to the list while we are 19757585717fSChris Mason * doing this splice 19767585717fSChris Mason */ 19770b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 19783fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 19790b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 19805d4f98a2SYan Zheng 19813fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 19823fd0a558SYan, Zheng found = 1; 19833fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 19843fd0a558SYan, Zheng struct btrfs_root, root_list); 19855d4f98a2SYan Zheng 1986a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1987a820feb5SDavid Sterba false); 19885d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 198924213fa4SJosef Bacik if (IS_ERR(root)) { 199024213fa4SJosef Bacik /* 199124213fa4SJosef Bacik * For recovery we read the fs roots on mount, 199224213fa4SJosef Bacik * and if we didn't find the root then we marked 199324213fa4SJosef Bacik * the reloc root as a garbage root. For normal 199424213fa4SJosef Bacik * relocation obviously the root should exist in 199524213fa4SJosef Bacik * memory. However there's no reason we can't 199624213fa4SJosef Bacik * handle the error properly here just in case. 199724213fa4SJosef Bacik */ 199824213fa4SJosef Bacik ASSERT(0); 199924213fa4SJosef Bacik ret = PTR_ERR(root); 200024213fa4SJosef Bacik goto out; 200124213fa4SJosef Bacik } 200224213fa4SJosef Bacik if (root->reloc_root != reloc_root) { 200324213fa4SJosef Bacik /* 200424213fa4SJosef Bacik * This is actually impossible without something 200524213fa4SJosef Bacik * going really wrong (like weird race condition 200624213fa4SJosef Bacik * or cosmic rays). 200724213fa4SJosef Bacik */ 200824213fa4SJosef Bacik ASSERT(0); 200924213fa4SJosef Bacik ret = -EINVAL; 201024213fa4SJosef Bacik goto out; 201124213fa4SJosef Bacik } 20123fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 201300246528SJosef Bacik btrfs_put_root(root); 2014b37b39cdSJosef Bacik if (ret) { 201525e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 201625e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 201725e293c2SWang Shilong &reloc_roots); 2018aca1bba6SLiu Bo goto out; 2019b37b39cdSJosef Bacik } 20203fd0a558SYan, Zheng } else { 202151415b6cSQu Wenruo if (!IS_ERR(root)) { 202251415b6cSQu Wenruo if (root->reloc_root == reloc_root) { 202351415b6cSQu Wenruo root->reloc_root = NULL; 202451415b6cSQu Wenruo btrfs_put_root(reloc_root); 202551415b6cSQu Wenruo } 20261dae7e0eSQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, 20271dae7e0eSQu Wenruo &root->state); 202851415b6cSQu Wenruo btrfs_put_root(root); 202951415b6cSQu Wenruo } 203051415b6cSQu Wenruo 20313fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 203230d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 203330d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 203430d40577SQu Wenruo &rc->dirty_subvol_roots); 20353fd0a558SYan, Zheng } 20365d4f98a2SYan Zheng } 20375d4f98a2SYan Zheng 20383fd0a558SYan, Zheng if (found) { 20393fd0a558SYan, Zheng found = 0; 20403fd0a558SYan, Zheng goto again; 20415d4f98a2SYan Zheng } 2042aca1bba6SLiu Bo out: 2043aca1bba6SLiu Bo if (ret) { 20440b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2045aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2046467bb1d2SWang Shilong 2047467bb1d2SWang Shilong /* new reloc root may be added */ 20480b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2049467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 20500b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2051467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2052aca1bba6SLiu Bo } 2053aca1bba6SLiu Bo 20547b7b7431SJosef Bacik /* 20557b7b7431SJosef Bacik * We used to have 20567b7b7431SJosef Bacik * 20577b7b7431SJosef Bacik * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 20587b7b7431SJosef Bacik * 20597b7b7431SJosef Bacik * here, but it's wrong. If we fail to start the transaction in 20607b7b7431SJosef Bacik * prepare_to_merge() we will have only 0 ref reloc roots, none of which 20617b7b7431SJosef Bacik * have actually been removed from the reloc_root_tree rb tree. This is 20627b7b7431SJosef Bacik * fine because we're bailing here, and we hold a reference on the root 20637b7b7431SJosef Bacik * for the list that holds it, so these roots will be cleaned up when we 20647b7b7431SJosef Bacik * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root 20657b7b7431SJosef Bacik * will be cleaned up on unmount. 20667b7b7431SJosef Bacik * 20677b7b7431SJosef Bacik * The remaining nodes will be cleaned up by free_reloc_control. 20687b7b7431SJosef Bacik */ 20695d4f98a2SYan Zheng } 20705d4f98a2SYan Zheng 20715d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 20725d4f98a2SYan Zheng { 20735d4f98a2SYan Zheng struct tree_block *block; 20745d4f98a2SYan Zheng struct rb_node *rb_node; 20755d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 20765d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 20775d4f98a2SYan Zheng rb_erase(rb_node, blocks); 20785d4f98a2SYan Zheng kfree(block); 20795d4f98a2SYan Zheng } 20805d4f98a2SYan Zheng } 20815d4f98a2SYan Zheng 20825d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 20835d4f98a2SYan Zheng struct btrfs_root *reloc_root) 20845d4f98a2SYan Zheng { 20850b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 20865d4f98a2SYan Zheng struct btrfs_root *root; 2087442b1ac5SJosef Bacik int ret; 20885d4f98a2SYan Zheng 20895d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 20905d4f98a2SYan Zheng return 0; 20915d4f98a2SYan Zheng 2092a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); 2093404bccbcSJosef Bacik 2094404bccbcSJosef Bacik /* 2095404bccbcSJosef Bacik * This should succeed, since we can't have a reloc root without having 2096404bccbcSJosef Bacik * already looked up the actual root and created the reloc root for this 2097404bccbcSJosef Bacik * root. 2098404bccbcSJosef Bacik * 2099404bccbcSJosef Bacik * However if there's some sort of corruption where we have a ref to a 2100404bccbcSJosef Bacik * reloc root without a corresponding root this could return ENOENT. 2101404bccbcSJosef Bacik */ 2102404bccbcSJosef Bacik if (IS_ERR(root)) { 2103404bccbcSJosef Bacik ASSERT(0); 2104404bccbcSJosef Bacik return PTR_ERR(root); 2105404bccbcSJosef Bacik } 2106404bccbcSJosef Bacik if (root->reloc_root != reloc_root) { 2107404bccbcSJosef Bacik ASSERT(0); 2108404bccbcSJosef Bacik btrfs_err(fs_info, 2109404bccbcSJosef Bacik "root %llu has two reloc roots associated with it", 2110404bccbcSJosef Bacik reloc_root->root_key.offset); 2111404bccbcSJosef Bacik btrfs_put_root(root); 2112404bccbcSJosef Bacik return -EUCLEAN; 2113404bccbcSJosef Bacik } 2114442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 211500246528SJosef Bacik btrfs_put_root(root); 21165d4f98a2SYan Zheng 2117442b1ac5SJosef Bacik return ret; 21185d4f98a2SYan Zheng } 21195d4f98a2SYan Zheng 21203fd0a558SYan, Zheng static noinline_for_stack 21213fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 21223fd0a558SYan, Zheng struct reloc_control *rc, 2123a26195a5SQu Wenruo struct btrfs_backref_node *node, 2124a26195a5SQu Wenruo struct btrfs_backref_edge *edges[]) 21255d4f98a2SYan Zheng { 2126a26195a5SQu Wenruo struct btrfs_backref_node *next; 21275d4f98a2SYan Zheng struct btrfs_root *root; 21283fd0a558SYan, Zheng int index = 0; 212992de551bSJosef Bacik int ret; 21303fd0a558SYan, Zheng 21315d4f98a2SYan Zheng next = node; 21325d4f98a2SYan Zheng while (1) { 21335d4f98a2SYan Zheng cond_resched(); 21345d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 21355d4f98a2SYan Zheng root = next->root; 21368ee66afeSJosef Bacik 21378ee66afeSJosef Bacik /* 21388ee66afeSJosef Bacik * If there is no root, then our references for this block are 21398ee66afeSJosef Bacik * incomplete, as we should be able to walk all the way up to a 21408ee66afeSJosef Bacik * block that is owned by a root. 21418ee66afeSJosef Bacik * 21428ee66afeSJosef Bacik * This path is only for SHAREABLE roots, so if we come upon a 21438ee66afeSJosef Bacik * non-SHAREABLE root then we have backrefs that resolve 21448ee66afeSJosef Bacik * improperly. 21458ee66afeSJosef Bacik * 21468ee66afeSJosef Bacik * Both of these cases indicate file system corruption, or a bug 21478ee66afeSJosef Bacik * in the backref walking code. 21488ee66afeSJosef Bacik */ 21498ee66afeSJosef Bacik if (!root) { 21508ee66afeSJosef Bacik ASSERT(0); 21518ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21528ee66afeSJosef Bacik "bytenr %llu doesn't have a backref path ending in a root", 21538ee66afeSJosef Bacik node->bytenr); 21548ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21558ee66afeSJosef Bacik } 21568ee66afeSJosef Bacik if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 21578ee66afeSJosef Bacik ASSERT(0); 21588ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21598ee66afeSJosef Bacik "bytenr %llu has multiple refs with one ending in a non-shareable root", 21608ee66afeSJosef Bacik node->bytenr); 21618ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21628ee66afeSJosef Bacik } 21635d4f98a2SYan Zheng 21645d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 216592de551bSJosef Bacik ret = record_reloc_root_in_trans(trans, root); 216692de551bSJosef Bacik if (ret) 216792de551bSJosef Bacik return ERR_PTR(ret); 21685d4f98a2SYan Zheng break; 21695d4f98a2SYan Zheng } 21705d4f98a2SYan Zheng 217192de551bSJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 217292de551bSJosef Bacik if (ret) 217392de551bSJosef Bacik return ERR_PTR(ret); 21743fd0a558SYan, Zheng root = root->reloc_root; 21753fd0a558SYan, Zheng 217639200e59SJosef Bacik /* 217739200e59SJosef Bacik * We could have raced with another thread which failed, so 217839200e59SJosef Bacik * root->reloc_root may not be set, return ENOENT in this case. 217939200e59SJosef Bacik */ 218039200e59SJosef Bacik if (!root) 218139200e59SJosef Bacik return ERR_PTR(-ENOENT); 218239200e59SJosef Bacik 21833fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 21848ee66afeSJosef Bacik /* 21858ee66afeSJosef Bacik * We just created the reloc root, so we shouldn't have 21868ee66afeSJosef Bacik * ->new_bytenr set and this shouldn't be in the changed 21878ee66afeSJosef Bacik * list. If it is then we have multiple roots pointing 21888ee66afeSJosef Bacik * at the same bytenr which indicates corruption, or 21898ee66afeSJosef Bacik * we've made a mistake in the backref walking code. 21908ee66afeSJosef Bacik */ 21918ee66afeSJosef Bacik ASSERT(next->new_bytenr == 0); 21928ee66afeSJosef Bacik ASSERT(list_empty(&next->list)); 21938ee66afeSJosef Bacik if (next->new_bytenr || !list_empty(&next->list)) { 21948ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21958ee66afeSJosef Bacik "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu", 21968ee66afeSJosef Bacik node->bytenr, next->bytenr); 21978ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21988ee66afeSJosef Bacik } 21998ee66afeSJosef Bacik 22003fd0a558SYan, Zheng next->new_bytenr = root->node->start; 220100246528SJosef Bacik btrfs_put_root(next->root); 220200246528SJosef Bacik next->root = btrfs_grab_root(root); 22030b530bc5SJosef Bacik ASSERT(next->root); 22043fd0a558SYan, Zheng list_add_tail(&next->list, 22053fd0a558SYan, Zheng &rc->backref_cache.changed); 22069569cc20SQu Wenruo mark_block_processed(rc, next); 22075d4f98a2SYan Zheng break; 22085d4f98a2SYan Zheng } 22095d4f98a2SYan Zheng 22103fd0a558SYan, Zheng WARN_ON(1); 22115d4f98a2SYan Zheng root = NULL; 22125d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 22135d4f98a2SYan Zheng if (!next || next->level <= node->level) 22145d4f98a2SYan Zheng break; 22155d4f98a2SYan Zheng } 2216cbdc2ebcSJosef Bacik if (!root) { 2217cbdc2ebcSJosef Bacik /* 2218cbdc2ebcSJosef Bacik * This can happen if there's fs corruption or if there's a bug 2219cbdc2ebcSJosef Bacik * in the backref lookup code. 2220cbdc2ebcSJosef Bacik */ 2221cbdc2ebcSJosef Bacik ASSERT(0); 2222cbdc2ebcSJosef Bacik return ERR_PTR(-ENOENT); 2223cbdc2ebcSJosef Bacik } 22245d4f98a2SYan Zheng 22253fd0a558SYan, Zheng next = node; 22263fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 22273fd0a558SYan, Zheng while (1) { 22283fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 22293fd0a558SYan, Zheng if (--index < 0) 22303fd0a558SYan, Zheng break; 22313fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 22323fd0a558SYan, Zheng } 22335d4f98a2SYan Zheng return root; 22345d4f98a2SYan Zheng } 22355d4f98a2SYan Zheng 22363fd0a558SYan, Zheng /* 223792a7cc42SQu Wenruo * Select a tree root for relocation. 223892a7cc42SQu Wenruo * 223992a7cc42SQu Wenruo * Return NULL if the block is not shareable. We should use do_relocation() in 224092a7cc42SQu Wenruo * this case. 224192a7cc42SQu Wenruo * 224292a7cc42SQu Wenruo * Return a tree root pointer if the block is shareable. 224392a7cc42SQu Wenruo * Return -ENOENT if the block is root of reloc tree. 22443fd0a558SYan, Zheng */ 22455d4f98a2SYan Zheng static noinline_for_stack 2246a26195a5SQu Wenruo struct btrfs_root *select_one_root(struct btrfs_backref_node *node) 22475d4f98a2SYan Zheng { 2248a26195a5SQu Wenruo struct btrfs_backref_node *next; 22493fd0a558SYan, Zheng struct btrfs_root *root; 22503fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 2251a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22523fd0a558SYan, Zheng int index = 0; 22533fd0a558SYan, Zheng 22543fd0a558SYan, Zheng next = node; 22553fd0a558SYan, Zheng while (1) { 22563fd0a558SYan, Zheng cond_resched(); 22573fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 22583fd0a558SYan, Zheng root = next->root; 22598717cf44SJosef Bacik 22608717cf44SJosef Bacik /* 22618717cf44SJosef Bacik * This can occur if we have incomplete extent refs leading all 22628717cf44SJosef Bacik * the way up a particular path, in this case return -EUCLEAN. 22638717cf44SJosef Bacik */ 22648717cf44SJosef Bacik if (!root) 22658717cf44SJosef Bacik return ERR_PTR(-EUCLEAN); 22663fd0a558SYan, Zheng 226792a7cc42SQu Wenruo /* No other choice for non-shareable tree */ 226892a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 22693fd0a558SYan, Zheng return root; 22703fd0a558SYan, Zheng 22713fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 22723fd0a558SYan, Zheng fs_root = root; 22733fd0a558SYan, Zheng 22743fd0a558SYan, Zheng if (next != node) 22753fd0a558SYan, Zheng return NULL; 22763fd0a558SYan, Zheng 22773fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 22783fd0a558SYan, Zheng if (!next || next->level <= node->level) 22793fd0a558SYan, Zheng break; 22803fd0a558SYan, Zheng } 22813fd0a558SYan, Zheng 22823fd0a558SYan, Zheng if (!fs_root) 22833fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 22843fd0a558SYan, Zheng return fs_root; 22855d4f98a2SYan Zheng } 22865d4f98a2SYan Zheng 22875d4f98a2SYan Zheng static noinline_for_stack 22883fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 2289a26195a5SQu Wenruo struct btrfs_backref_node *node, int reserve) 22905d4f98a2SYan Zheng { 22910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2292a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2293a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2294a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22953fd0a558SYan, Zheng u64 num_bytes = 0; 22963fd0a558SYan, Zheng int index = 0; 22975d4f98a2SYan Zheng 22983fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 22993fd0a558SYan, Zheng 23003fd0a558SYan, Zheng while (next) { 23013fd0a558SYan, Zheng cond_resched(); 23025d4f98a2SYan Zheng while (1) { 23033fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 23045d4f98a2SYan Zheng break; 23055d4f98a2SYan Zheng 23060b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 23073fd0a558SYan, Zheng 23083fd0a558SYan, Zheng if (list_empty(&next->upper)) 23093fd0a558SYan, Zheng break; 23103fd0a558SYan, Zheng 23113fd0a558SYan, Zheng edge = list_entry(next->upper.next, 2312a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 23133fd0a558SYan, Zheng edges[index++] = edge; 23143fd0a558SYan, Zheng next = edge->node[UPPER]; 23155d4f98a2SYan Zheng } 23163fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 23173fd0a558SYan, Zheng } 23183fd0a558SYan, Zheng return num_bytes; 23193fd0a558SYan, Zheng } 23203fd0a558SYan, Zheng 23213fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 23223fd0a558SYan, Zheng struct reloc_control *rc, 2323a26195a5SQu Wenruo struct btrfs_backref_node *node) 23243fd0a558SYan, Zheng { 23253fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2326da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23273fd0a558SYan, Zheng u64 num_bytes; 23283fd0a558SYan, Zheng int ret; 23290647bf56SWang Shilong u64 tmp; 23303fd0a558SYan, Zheng 23313fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 23323fd0a558SYan, Zheng 23333fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 23340647bf56SWang Shilong rc->reserved_bytes += num_bytes; 23358ca17f0fSJosef Bacik 23368ca17f0fSJosef Bacik /* 23378ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 23388ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 23398ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 23408ca17f0fSJosef Bacik */ 23419270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes, 23428ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 23433fd0a558SYan, Zheng if (ret) { 2344da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 23450647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 23460647bf56SWang Shilong tmp <<= 1; 23470647bf56SWang Shilong /* 23480647bf56SWang Shilong * only one thread can access block_rsv at this point, 23490647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 23500647bf56SWang Shilong * we expand more reservation size here to allow enough 235152042d8eSAndrea Gelmini * space for relocation and we will return earlier in 23520647bf56SWang Shilong * enospc case. 23530647bf56SWang Shilong */ 2354da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 23550647bf56SWang Shilong RELOCATION_RESERVED_NODES; 23568ca17f0fSJosef Bacik return -EAGAIN; 23573fd0a558SYan, Zheng } 23583fd0a558SYan, Zheng 23593fd0a558SYan, Zheng return 0; 23603fd0a558SYan, Zheng } 23613fd0a558SYan, Zheng 23625d4f98a2SYan Zheng /* 23635d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 23645d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 23655d4f98a2SYan Zheng * 23665d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 23675d4f98a2SYan Zheng * in that case this function just updates pointers. 23685d4f98a2SYan Zheng */ 23695d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 23703fd0a558SYan, Zheng struct reloc_control *rc, 2371a26195a5SQu Wenruo struct btrfs_backref_node *node, 23725d4f98a2SYan Zheng struct btrfs_key *key, 23735d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 23745d4f98a2SYan Zheng { 2375a26195a5SQu Wenruo struct btrfs_backref_node *upper; 2376a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2377a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 23785d4f98a2SYan Zheng struct btrfs_root *root; 23795d4f98a2SYan Zheng struct extent_buffer *eb; 23805d4f98a2SYan Zheng u32 blocksize; 23815d4f98a2SYan Zheng u64 bytenr; 23825d4f98a2SYan Zheng int slot; 23838df01fddSNikolay Borisov int ret = 0; 23845d4f98a2SYan Zheng 2385ffe30dd8SJosef Bacik /* 2386ffe30dd8SJosef Bacik * If we are lowest then this is the first time we're processing this 2387ffe30dd8SJosef Bacik * block, and thus shouldn't have an eb associated with it yet. 2388ffe30dd8SJosef Bacik */ 2389ffe30dd8SJosef Bacik ASSERT(!lowest || !node->eb); 23905d4f98a2SYan Zheng 23915d4f98a2SYan Zheng path->lowest_level = node->level + 1; 23923fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 23935d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 239482fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2395581c1760SQu Wenruo 23965d4f98a2SYan Zheng cond_resched(); 23975d4f98a2SYan Zheng 23985d4f98a2SYan Zheng upper = edge->node[UPPER]; 2399dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 2400cbdc2ebcSJosef Bacik if (IS_ERR(root)) { 2401cbdc2ebcSJosef Bacik ret = PTR_ERR(root); 2402cbdc2ebcSJosef Bacik goto next; 2403cbdc2ebcSJosef Bacik } 24045d4f98a2SYan Zheng 24053fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 24063fd0a558SYan, Zheng if (!lowest) { 2407e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 24088df01fddSNikolay Borisov if (ret < 0) 2409cbca7d59SFilipe Manana goto next; 24103fd0a558SYan, Zheng BUG_ON(ret); 24113fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24123fd0a558SYan, Zheng if (node->eb->start == bytenr) 24133fd0a558SYan, Zheng goto next; 24143fd0a558SYan, Zheng } 2415b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 24163fd0a558SYan, Zheng } 24175d4f98a2SYan Zheng 24185d4f98a2SYan Zheng if (!upper->eb) { 24195d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 24203561b9dbSLiu Bo if (ret) { 24218df01fddSNikolay Borisov if (ret > 0) 24228df01fddSNikolay Borisov ret = -ENOENT; 24233561b9dbSLiu Bo 24243561b9dbSLiu Bo btrfs_release_path(path); 24255d4f98a2SYan Zheng break; 24265d4f98a2SYan Zheng } 24275d4f98a2SYan Zheng 24283fd0a558SYan, Zheng if (!upper->eb) { 24293fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 24303fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 24313fd0a558SYan, Zheng } else { 24323fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 24333fd0a558SYan, Zheng } 24343fd0a558SYan, Zheng 24353fd0a558SYan, Zheng upper->locked = 1; 24363fd0a558SYan, Zheng path->locks[upper->level] = 0; 24373fd0a558SYan, Zheng 24385d4f98a2SYan Zheng slot = path->slots[upper->level]; 2439b3b4aa74SDavid Sterba btrfs_release_path(path); 24405d4f98a2SYan Zheng } else { 2441e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 24428df01fddSNikolay Borisov if (ret < 0) 2443cbca7d59SFilipe Manana goto next; 24445d4f98a2SYan Zheng BUG_ON(ret); 24455d4f98a2SYan Zheng } 24465d4f98a2SYan Zheng 24475d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24483fd0a558SYan, Zheng if (lowest) { 24494547f4d8SLiu Bo if (bytenr != node->bytenr) { 24504547f4d8SLiu Bo btrfs_err(root->fs_info, 24514547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 24524547f4d8SLiu Bo bytenr, node->bytenr, slot, 24534547f4d8SLiu Bo upper->eb->start); 24548df01fddSNikolay Borisov ret = -EIO; 24554547f4d8SLiu Bo goto next; 24564547f4d8SLiu Bo } 24575d4f98a2SYan Zheng } else { 24583fd0a558SYan, Zheng if (node->eb->start == bytenr) 24593fd0a558SYan, Zheng goto next; 24605d4f98a2SYan Zheng } 24615d4f98a2SYan Zheng 2462da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 2463c9752536SJosef Bacik eb = btrfs_read_node_slot(upper->eb, slot); 246464c043deSLiu Bo if (IS_ERR(eb)) { 24658df01fddSNikolay Borisov ret = PTR_ERR(eb); 246664c043deSLiu Bo goto next; 246797d9a8a4STsutomu Itoh } 24685d4f98a2SYan Zheng btrfs_tree_lock(eb); 24695d4f98a2SYan Zheng 24705d4f98a2SYan Zheng if (!node->eb) { 24715d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 24729631e4ccSJosef Bacik slot, &eb, BTRFS_NESTING_COW); 24733fd0a558SYan, Zheng btrfs_tree_unlock(eb); 24743fd0a558SYan, Zheng free_extent_buffer(eb); 24758df01fddSNikolay Borisov if (ret < 0) 24763fd0a558SYan, Zheng goto next; 2477ffe30dd8SJosef Bacik /* 2478ffe30dd8SJosef Bacik * We've just COWed this block, it should have updated 2479ffe30dd8SJosef Bacik * the correct backref node entry. 2480ffe30dd8SJosef Bacik */ 2481ffe30dd8SJosef Bacik ASSERT(node->eb == eb); 24825d4f98a2SYan Zheng } else { 24835d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 24845d4f98a2SYan Zheng node->eb->start); 24855d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 24865d4f98a2SYan Zheng trans->transid); 24875d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 24885d4f98a2SYan Zheng 248982fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 24905d4f98a2SYan Zheng node->eb->start, blocksize, 249182fa113fSQu Wenruo upper->eb->start); 249282fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 2493f42c5da6SNikolay Borisov btrfs_header_owner(upper->eb), 2494f42c5da6SNikolay Borisov root->root_key.objectid, false); 249582fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 2496eb6b7fb4SJosef Bacik if (!ret) 2497eb6b7fb4SJosef Bacik ret = btrfs_drop_subtree(trans, root, eb, 2498eb6b7fb4SJosef Bacik upper->eb); 2499eb6b7fb4SJosef Bacik if (ret) 2500eb6b7fb4SJosef Bacik btrfs_abort_transaction(trans, ret); 25015d4f98a2SYan Zheng } 25023fd0a558SYan, Zheng next: 25033fd0a558SYan, Zheng if (!upper->pending) 2504b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 25053fd0a558SYan, Zheng else 2506b0fe7078SQu Wenruo btrfs_backref_unlock_node_buffer(upper); 25078df01fddSNikolay Borisov if (ret) 25083fd0a558SYan, Zheng break; 25095d4f98a2SYan Zheng } 25103fd0a558SYan, Zheng 25118df01fddSNikolay Borisov if (!ret && node->pending) { 2512b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 25133fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 25143fd0a558SYan, Zheng node->pending = 0; 25155d4f98a2SYan Zheng } 25163fd0a558SYan, Zheng 25175d4f98a2SYan Zheng path->lowest_level = 0; 2518ffe30dd8SJosef Bacik 2519ffe30dd8SJosef Bacik /* 2520ffe30dd8SJosef Bacik * We should have allocated all of our space in the block rsv and thus 2521ffe30dd8SJosef Bacik * shouldn't ENOSPC. 2522ffe30dd8SJosef Bacik */ 2523ffe30dd8SJosef Bacik ASSERT(ret != -ENOSPC); 25248df01fddSNikolay Borisov return ret; 25255d4f98a2SYan Zheng } 25265d4f98a2SYan Zheng 25275d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 25283fd0a558SYan, Zheng struct reloc_control *rc, 2529a26195a5SQu Wenruo struct btrfs_backref_node *node, 25305d4f98a2SYan Zheng struct btrfs_path *path) 25315d4f98a2SYan Zheng { 25325d4f98a2SYan Zheng struct btrfs_key key; 25335d4f98a2SYan Zheng 25345d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 25353fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 25365d4f98a2SYan Zheng } 25375d4f98a2SYan Zheng 25385d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 25393fd0a558SYan, Zheng struct reloc_control *rc, 25403fd0a558SYan, Zheng struct btrfs_path *path, int err) 25415d4f98a2SYan Zheng { 25423fd0a558SYan, Zheng LIST_HEAD(list); 2543a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 2544a26195a5SQu Wenruo struct btrfs_backref_node *node; 25455d4f98a2SYan Zheng int level; 25465d4f98a2SYan Zheng int ret; 25475d4f98a2SYan Zheng 25485d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 25495d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 25505d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 2551a26195a5SQu Wenruo struct btrfs_backref_node, list); 25523fd0a558SYan, Zheng list_move_tail(&node->list, &list); 25533fd0a558SYan, Zheng BUG_ON(!node->pending); 25545d4f98a2SYan Zheng 25553fd0a558SYan, Zheng if (!err) { 25563fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 25575d4f98a2SYan Zheng if (ret < 0) 25585d4f98a2SYan Zheng err = ret; 25595d4f98a2SYan Zheng } 25605d4f98a2SYan Zheng } 25613fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 25623fd0a558SYan, Zheng } 25635d4f98a2SYan Zheng return err; 25645d4f98a2SYan Zheng } 25655d4f98a2SYan Zheng 25665d4f98a2SYan Zheng /* 25675d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 25685d4f98a2SYan Zheng * as processed. 25695d4f98a2SYan Zheng */ 25705d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 2571a26195a5SQu Wenruo struct btrfs_backref_node *node) 25725d4f98a2SYan Zheng { 2573a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2574a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2575a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25765d4f98a2SYan Zheng int index = 0; 25775d4f98a2SYan Zheng 25785d4f98a2SYan Zheng while (next) { 25795d4f98a2SYan Zheng cond_resched(); 25805d4f98a2SYan Zheng while (1) { 25815d4f98a2SYan Zheng if (next->processed) 25825d4f98a2SYan Zheng break; 25835d4f98a2SYan Zheng 25849569cc20SQu Wenruo mark_block_processed(rc, next); 25855d4f98a2SYan Zheng 25865d4f98a2SYan Zheng if (list_empty(&next->upper)) 25875d4f98a2SYan Zheng break; 25885d4f98a2SYan Zheng 25895d4f98a2SYan Zheng edge = list_entry(next->upper.next, 2590a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 25915d4f98a2SYan Zheng edges[index++] = edge; 25925d4f98a2SYan Zheng next = edge->node[UPPER]; 25935d4f98a2SYan Zheng } 25945d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25955d4f98a2SYan Zheng } 25965d4f98a2SYan Zheng } 25975d4f98a2SYan Zheng 25987476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 25995d4f98a2SYan Zheng { 2600da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 26017476dfdaSDavid Sterba 26025d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 26039655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 26045d4f98a2SYan Zheng return 1; 26055d4f98a2SYan Zheng return 0; 26065d4f98a2SYan Zheng } 26075d4f98a2SYan Zheng 26082ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 26095d4f98a2SYan Zheng struct tree_block *block) 26105d4f98a2SYan Zheng { 26115d4f98a2SYan Zheng struct extent_buffer *eb; 26125d4f98a2SYan Zheng 2613f7ba2d37SJosef Bacik eb = read_tree_block(fs_info, block->bytenr, block->owner, 2614f7ba2d37SJosef Bacik block->key.offset, block->level, NULL); 26154eb150d6SQu Wenruo if (IS_ERR(eb)) 261664c043deSLiu Bo return PTR_ERR(eb); 26174eb150d6SQu Wenruo if (!extent_buffer_uptodate(eb)) { 2618416bc658SJosef Bacik free_extent_buffer(eb); 2619416bc658SJosef Bacik return -EIO; 2620416bc658SJosef Bacik } 26215d4f98a2SYan Zheng if (block->level == 0) 26225d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 26235d4f98a2SYan Zheng else 26245d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 26255d4f98a2SYan Zheng free_extent_buffer(eb); 26265d4f98a2SYan Zheng block->key_ready = 1; 26275d4f98a2SYan Zheng return 0; 26285d4f98a2SYan Zheng } 26295d4f98a2SYan Zheng 26305d4f98a2SYan Zheng /* 26315d4f98a2SYan Zheng * helper function to relocate a tree block 26325d4f98a2SYan Zheng */ 26335d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 26345d4f98a2SYan Zheng struct reloc_control *rc, 2635a26195a5SQu Wenruo struct btrfs_backref_node *node, 26365d4f98a2SYan Zheng struct btrfs_key *key, 26375d4f98a2SYan Zheng struct btrfs_path *path) 26385d4f98a2SYan Zheng { 26395d4f98a2SYan Zheng struct btrfs_root *root; 26403fd0a558SYan, Zheng int ret = 0; 26415d4f98a2SYan Zheng 26423fd0a558SYan, Zheng if (!node) 26435d4f98a2SYan Zheng return 0; 26443fd0a558SYan, Zheng 26455f6b2e5cSJosef Bacik /* 26465f6b2e5cSJosef Bacik * If we fail here we want to drop our backref_node because we are going 26475f6b2e5cSJosef Bacik * to start over and regenerate the tree for it. 26485f6b2e5cSJosef Bacik */ 26495f6b2e5cSJosef Bacik ret = reserve_metadata_space(trans, rc, node); 26505f6b2e5cSJosef Bacik if (ret) 26515f6b2e5cSJosef Bacik goto out; 26525f6b2e5cSJosef Bacik 26533fd0a558SYan, Zheng BUG_ON(node->processed); 2654147d256eSZhaolei root = select_one_root(node); 26558717cf44SJosef Bacik if (IS_ERR(root)) { 26568717cf44SJosef Bacik ret = PTR_ERR(root); 26578717cf44SJosef Bacik 26588717cf44SJosef Bacik /* See explanation in select_one_root for the -EUCLEAN case. */ 26598717cf44SJosef Bacik ASSERT(ret == -ENOENT); 26608717cf44SJosef Bacik if (ret == -ENOENT) { 26618717cf44SJosef Bacik ret = 0; 26623fd0a558SYan, Zheng update_processed_blocks(rc, node); 26638717cf44SJosef Bacik } 26643fd0a558SYan, Zheng goto out; 26655d4f98a2SYan Zheng } 26665d4f98a2SYan Zheng 26673fd0a558SYan, Zheng if (root) { 266892a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 26691c7bfa15SJosef Bacik /* 26701c7bfa15SJosef Bacik * This block was the root block of a root, and this is 26711c7bfa15SJosef Bacik * the first time we're processing the block and thus it 26721c7bfa15SJosef Bacik * should not have had the ->new_bytenr modified and 26731c7bfa15SJosef Bacik * should have not been included on the changed list. 26741c7bfa15SJosef Bacik * 26751c7bfa15SJosef Bacik * However in the case of corruption we could have 26761c7bfa15SJosef Bacik * multiple refs pointing to the same block improperly, 26771c7bfa15SJosef Bacik * and thus we would trip over these checks. ASSERT() 26781c7bfa15SJosef Bacik * for the developer case, because it could indicate a 26791c7bfa15SJosef Bacik * bug in the backref code, however error out for a 26801c7bfa15SJosef Bacik * normal user in the case of corruption. 26811c7bfa15SJosef Bacik */ 26821c7bfa15SJosef Bacik ASSERT(node->new_bytenr == 0); 26831c7bfa15SJosef Bacik ASSERT(list_empty(&node->list)); 26841c7bfa15SJosef Bacik if (node->new_bytenr || !list_empty(&node->list)) { 26851c7bfa15SJosef Bacik btrfs_err(root->fs_info, 26861c7bfa15SJosef Bacik "bytenr %llu has improper references to it", 26871c7bfa15SJosef Bacik node->bytenr); 26881c7bfa15SJosef Bacik ret = -EUCLEAN; 26891c7bfa15SJosef Bacik goto out; 26901c7bfa15SJosef Bacik } 2691d18c7bd9SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 2692d18c7bd9SJosef Bacik if (ret) 2693d18c7bd9SJosef Bacik goto out; 269439200e59SJosef Bacik /* 269539200e59SJosef Bacik * Another thread could have failed, need to check if we 269639200e59SJosef Bacik * have reloc_root actually set. 269739200e59SJosef Bacik */ 269839200e59SJosef Bacik if (!root->reloc_root) { 269939200e59SJosef Bacik ret = -ENOENT; 270039200e59SJosef Bacik goto out; 270139200e59SJosef Bacik } 27023fd0a558SYan, Zheng root = root->reloc_root; 27033fd0a558SYan, Zheng node->new_bytenr = root->node->start; 270400246528SJosef Bacik btrfs_put_root(node->root); 270500246528SJosef Bacik node->root = btrfs_grab_root(root); 27060b530bc5SJosef Bacik ASSERT(node->root); 27073fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 27083fd0a558SYan, Zheng } else { 27095d4f98a2SYan Zheng path->lowest_level = node->level; 27102bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 27112bb2e00eSFilipe Manana btrfs_reserve_chunk_metadata(trans, false); 27125d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2713b3b4aa74SDavid Sterba btrfs_release_path(path); 27142bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 27152bb2e00eSFilipe Manana btrfs_trans_release_chunk_metadata(trans); 27163fd0a558SYan, Zheng if (ret > 0) 27175d4f98a2SYan Zheng ret = 0; 27183fd0a558SYan, Zheng } 27193fd0a558SYan, Zheng if (!ret) 27203fd0a558SYan, Zheng update_processed_blocks(rc, node); 27213fd0a558SYan, Zheng } else { 27223fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 27233fd0a558SYan, Zheng } 27245d4f98a2SYan Zheng out: 27250647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 2726023acb07SQu Wenruo btrfs_backref_cleanup_node(&rc->backref_cache, node); 27275d4f98a2SYan Zheng return ret; 27285d4f98a2SYan Zheng } 27295d4f98a2SYan Zheng 27305d4f98a2SYan Zheng /* 27315d4f98a2SYan Zheng * relocate a list of blocks 27325d4f98a2SYan Zheng */ 27335d4f98a2SYan Zheng static noinline_for_stack 27345d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 27355d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 27365d4f98a2SYan Zheng { 27372ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2738a26195a5SQu Wenruo struct btrfs_backref_node *node; 27395d4f98a2SYan Zheng struct btrfs_path *path; 27405d4f98a2SYan Zheng struct tree_block *block; 274198ff7b94SQu Wenruo struct tree_block *next; 27425d4f98a2SYan Zheng int ret; 27435d4f98a2SYan Zheng int err = 0; 27445d4f98a2SYan Zheng 27455d4f98a2SYan Zheng path = btrfs_alloc_path(); 2746e1a12670SLiu Bo if (!path) { 2747e1a12670SLiu Bo err = -ENOMEM; 274834c2b290SDavid Sterba goto out_free_blocks; 2749e1a12670SLiu Bo } 27505d4f98a2SYan Zheng 275198ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 275298ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27535d4f98a2SYan Zheng if (!block->key_ready) 2754f7ba2d37SJosef Bacik btrfs_readahead_tree_block(fs_info, block->bytenr, 2755f7ba2d37SJosef Bacik block->owner, 0, 27563fbaf258SJosef Bacik block->level); 27575d4f98a2SYan Zheng } 27585d4f98a2SYan Zheng 275998ff7b94SQu Wenruo /* Get first keys */ 276098ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 276134c2b290SDavid Sterba if (!block->key_ready) { 27622ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 276334c2b290SDavid Sterba if (err) 276434c2b290SDavid Sterba goto out_free_path; 276534c2b290SDavid Sterba } 27665d4f98a2SYan Zheng } 27675d4f98a2SYan Zheng 276898ff7b94SQu Wenruo /* Do tree relocation */ 276998ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27703fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 27715d4f98a2SYan Zheng block->level, block->bytenr); 27725d4f98a2SYan Zheng if (IS_ERR(node)) { 27735d4f98a2SYan Zheng err = PTR_ERR(node); 27745d4f98a2SYan Zheng goto out; 27755d4f98a2SYan Zheng } 27765d4f98a2SYan Zheng 27775d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 27785d4f98a2SYan Zheng path); 27795d4f98a2SYan Zheng if (ret < 0) { 27805d4f98a2SYan Zheng err = ret; 278150dbbb71SJosef Bacik break; 27825d4f98a2SYan Zheng } 27835d4f98a2SYan Zheng } 27845d4f98a2SYan Zheng out: 27853fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 27865d4f98a2SYan Zheng 278734c2b290SDavid Sterba out_free_path: 27885d4f98a2SYan Zheng btrfs_free_path(path); 278934c2b290SDavid Sterba out_free_blocks: 2790e1a12670SLiu Bo free_block_list(blocks); 27915d4f98a2SYan Zheng return err; 27925d4f98a2SYan Zheng } 27935d4f98a2SYan Zheng 2794056d9becSNikolay Borisov static noinline_for_stack int prealloc_file_extent_cluster( 2795056d9becSNikolay Borisov struct btrfs_inode *inode, 2796efa56464SYan, Zheng struct file_extent_cluster *cluster) 2797efa56464SYan, Zheng { 2798efa56464SYan, Zheng u64 alloc_hint = 0; 2799efa56464SYan, Zheng u64 start; 2800efa56464SYan, Zheng u64 end; 2801056d9becSNikolay Borisov u64 offset = inode->index_cnt; 2802efa56464SYan, Zheng u64 num_bytes; 28034e9d0d01SNikolay Borisov int nr; 2804efa56464SYan, Zheng int ret = 0; 28059d9ea1e6SQu Wenruo u64 i_size = i_size_read(&inode->vfs_inode); 2806dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 2807dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 2808214e61d0SNikolay Borisov u64 cur_offset = prealloc_start; 2809efa56464SYan, Zheng 28109d9ea1e6SQu Wenruo /* 28119d9ea1e6SQu Wenruo * For subpage case, previous i_size may not be aligned to PAGE_SIZE. 28129d9ea1e6SQu Wenruo * This means the range [i_size, PAGE_END + 1) is filled with zeros by 28139d9ea1e6SQu Wenruo * btrfs_do_readpage() call of previously relocated file cluster. 28149d9ea1e6SQu Wenruo * 28159d9ea1e6SQu Wenruo * If the current cluster starts in the above range, btrfs_do_readpage() 28169d9ea1e6SQu Wenruo * will skip the read, and relocate_one_page() will later writeback 28179d9ea1e6SQu Wenruo * the padding zeros as new data, causing data corruption. 28189d9ea1e6SQu Wenruo * 28199d9ea1e6SQu Wenruo * Here we have to manually invalidate the range (i_size, PAGE_END + 1). 28209d9ea1e6SQu Wenruo */ 28219d9ea1e6SQu Wenruo if (!IS_ALIGNED(i_size, PAGE_SIZE)) { 28229d9ea1e6SQu Wenruo struct address_space *mapping = inode->vfs_inode.i_mapping; 28239d9ea1e6SQu Wenruo struct btrfs_fs_info *fs_info = inode->root->fs_info; 28249d9ea1e6SQu Wenruo const u32 sectorsize = fs_info->sectorsize; 28259d9ea1e6SQu Wenruo struct page *page; 28269d9ea1e6SQu Wenruo 28279d9ea1e6SQu Wenruo ASSERT(sectorsize < PAGE_SIZE); 28289d9ea1e6SQu Wenruo ASSERT(IS_ALIGNED(i_size, sectorsize)); 28299d9ea1e6SQu Wenruo 28309d9ea1e6SQu Wenruo /* 28319d9ea1e6SQu Wenruo * Subpage can't handle page with DIRTY but without UPTODATE 28329d9ea1e6SQu Wenruo * bit as it can lead to the following deadlock: 28339d9ea1e6SQu Wenruo * 2834fb12489bSMatthew Wilcox (Oracle) * btrfs_read_folio() 28359d9ea1e6SQu Wenruo * | Page already *locked* 28369d9ea1e6SQu Wenruo * |- btrfs_lock_and_flush_ordered_range() 28379d9ea1e6SQu Wenruo * |- btrfs_start_ordered_extent() 28389d9ea1e6SQu Wenruo * |- extent_write_cache_pages() 28399d9ea1e6SQu Wenruo * |- lock_page() 28409d9ea1e6SQu Wenruo * We try to lock the page we already hold. 28419d9ea1e6SQu Wenruo * 28429d9ea1e6SQu Wenruo * Here we just writeback the whole data reloc inode, so that 28439d9ea1e6SQu Wenruo * we will be ensured to have no dirty range in the page, and 28449d9ea1e6SQu Wenruo * are safe to clear the uptodate bits. 28459d9ea1e6SQu Wenruo * 28469d9ea1e6SQu Wenruo * This shouldn't cause too much overhead, as we need to write 28479d9ea1e6SQu Wenruo * the data back anyway. 28489d9ea1e6SQu Wenruo */ 28499d9ea1e6SQu Wenruo ret = filemap_write_and_wait(mapping); 28509d9ea1e6SQu Wenruo if (ret < 0) 28519d9ea1e6SQu Wenruo return ret; 28529d9ea1e6SQu Wenruo 28539d9ea1e6SQu Wenruo clear_extent_bits(&inode->io_tree, i_size, 28549d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - 1, 28559d9ea1e6SQu Wenruo EXTENT_UPTODATE); 28569d9ea1e6SQu Wenruo page = find_lock_page(mapping, i_size >> PAGE_SHIFT); 28579d9ea1e6SQu Wenruo /* 28589d9ea1e6SQu Wenruo * If page is freed we don't need to do anything then, as we 28599d9ea1e6SQu Wenruo * will re-read the whole page anyway. 28609d9ea1e6SQu Wenruo */ 28619d9ea1e6SQu Wenruo if (page) { 28629d9ea1e6SQu Wenruo btrfs_subpage_clear_uptodate(fs_info, page, i_size, 28639d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - i_size); 28649d9ea1e6SQu Wenruo unlock_page(page); 28659d9ea1e6SQu Wenruo put_page(page); 28669d9ea1e6SQu Wenruo } 28679d9ea1e6SQu Wenruo } 28689d9ea1e6SQu Wenruo 2869efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 2870056d9becSNikolay Borisov ret = btrfs_alloc_data_chunk_ondemand(inode, 2871dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 2872efa56464SYan, Zheng if (ret) 2873214e61d0SNikolay Borisov return ret; 2874efa56464SYan, Zheng 287564708539SJosef Bacik btrfs_inode_lock(&inode->vfs_inode, 0); 28764e9d0d01SNikolay Borisov for (nr = 0; nr < cluster->nr; nr++) { 28779c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 28789c5c9604SJosef Bacik 2879efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 2880efa56464SYan, Zheng if (nr + 1 < cluster->nr) 2881efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 2882efa56464SYan, Zheng else 2883efa56464SYan, Zheng end = cluster->end - offset; 2884efa56464SYan, Zheng 28859c5c9604SJosef Bacik lock_extent(&inode->io_tree, start, end, &cached_state); 2886efa56464SYan, Zheng num_bytes = end + 1 - start; 2887056d9becSNikolay Borisov ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start, 2888efa56464SYan, Zheng num_bytes, num_bytes, 2889efa56464SYan, Zheng end + 1, &alloc_hint); 289018513091SWang Xiaoguang cur_offset = end + 1; 28919c5c9604SJosef Bacik unlock_extent(&inode->io_tree, start, end, &cached_state); 2892efa56464SYan, Zheng if (ret) 2893efa56464SYan, Zheng break; 2894efa56464SYan, Zheng } 289564708539SJosef Bacik btrfs_inode_unlock(&inode->vfs_inode, 0); 2896214e61d0SNikolay Borisov 289718513091SWang Xiaoguang if (cur_offset < prealloc_end) 2898056d9becSNikolay Borisov btrfs_free_reserved_data_space_noquota(inode->root->fs_info, 2899a89ef455SFilipe Manana prealloc_end + 1 - cur_offset); 2900efa56464SYan, Zheng return ret; 2901efa56464SYan, Zheng } 2902efa56464SYan, Zheng 29034b01c44fSJohannes Thumshirn static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode, 29044b01c44fSJohannes Thumshirn u64 start, u64 end, u64 block_start) 29055d4f98a2SYan Zheng { 29065d4f98a2SYan Zheng struct extent_map *em; 29079c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 29080257bb82SYan, Zheng int ret = 0; 29095d4f98a2SYan Zheng 2910172ddd60SDavid Sterba em = alloc_extent_map(); 29110257bb82SYan, Zheng if (!em) 29120257bb82SYan, Zheng return -ENOMEM; 29130257bb82SYan, Zheng 29145d4f98a2SYan Zheng em->start = start; 29150257bb82SYan, Zheng em->len = end + 1 - start; 29160257bb82SYan, Zheng em->block_len = em->len; 29170257bb82SYan, Zheng em->block_start = block_start; 29185d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 29195d4f98a2SYan Zheng 29209c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 2921a1ba4c08SFilipe Manana ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, false); 29229c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 2923a1ba4c08SFilipe Manana free_extent_map(em); 2924a1ba4c08SFilipe Manana 29250257bb82SYan, Zheng return ret; 29260257bb82SYan, Zheng } 29275d4f98a2SYan Zheng 2928726a3421SQu Wenruo /* 2929907d2710SDavid Sterba * Allow error injection to test balance/relocation cancellation 2930726a3421SQu Wenruo */ 29311fec12a5SJosef Bacik noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 2932726a3421SQu Wenruo { 29335cb502f4SQu Wenruo return atomic_read(&fs_info->balance_cancel_req) || 2934907d2710SDavid Sterba atomic_read(&fs_info->reloc_cancel_req) || 29355cb502f4SQu Wenruo fatal_signal_pending(current); 2936726a3421SQu Wenruo } 2937726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 2938726a3421SQu Wenruo 2939c2832898SQu Wenruo static u64 get_cluster_boundary_end(struct file_extent_cluster *cluster, 2940c2832898SQu Wenruo int cluster_nr) 2941c2832898SQu Wenruo { 2942c2832898SQu Wenruo /* Last extent, use cluster end directly */ 2943c2832898SQu Wenruo if (cluster_nr >= cluster->nr - 1) 2944c2832898SQu Wenruo return cluster->end; 2945c2832898SQu Wenruo 2946c2832898SQu Wenruo /* Use next boundary start*/ 2947c2832898SQu Wenruo return cluster->boundary[cluster_nr + 1] - 1; 2948c2832898SQu Wenruo } 2949c2832898SQu Wenruo 2950f47960f4SQu Wenruo static int relocate_one_page(struct inode *inode, struct file_ra_state *ra, 2951f47960f4SQu Wenruo struct file_extent_cluster *cluster, 2952f47960f4SQu Wenruo int *cluster_nr, unsigned long page_index) 2953f47960f4SQu Wenruo { 2954f47960f4SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 2955f47960f4SQu Wenruo u64 offset = BTRFS_I(inode)->index_cnt; 2956f47960f4SQu Wenruo const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT; 2957f47960f4SQu Wenruo gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 2958f47960f4SQu Wenruo struct page *page; 2959f47960f4SQu Wenruo u64 page_start; 2960f47960f4SQu Wenruo u64 page_end; 2961c2832898SQu Wenruo u64 cur; 2962f47960f4SQu Wenruo int ret; 2963f47960f4SQu Wenruo 2964f47960f4SQu Wenruo ASSERT(page_index <= last_index); 2965f47960f4SQu Wenruo page = find_lock_page(inode->i_mapping, page_index); 2966f47960f4SQu Wenruo if (!page) { 2967f47960f4SQu Wenruo page_cache_sync_readahead(inode->i_mapping, ra, NULL, 2968f47960f4SQu Wenruo page_index, last_index + 1 - page_index); 2969f47960f4SQu Wenruo page = find_or_create_page(inode->i_mapping, page_index, mask); 2970c2832898SQu Wenruo if (!page) 2971c2832898SQu Wenruo return -ENOMEM; 2972f47960f4SQu Wenruo } 2973f47960f4SQu Wenruo ret = set_page_extent_mapped(page); 2974f47960f4SQu Wenruo if (ret < 0) 2975f47960f4SQu Wenruo goto release_page; 2976f47960f4SQu Wenruo 2977f47960f4SQu Wenruo if (PageReadahead(page)) 29782ebdd1dfSMatthew Wilcox (Oracle) page_cache_async_readahead(inode->i_mapping, ra, NULL, 29792ebdd1dfSMatthew Wilcox (Oracle) page_folio(page), page_index, 29802ebdd1dfSMatthew Wilcox (Oracle) last_index + 1 - page_index); 2981f47960f4SQu Wenruo 2982f47960f4SQu Wenruo if (!PageUptodate(page)) { 2983fb12489bSMatthew Wilcox (Oracle) btrfs_read_folio(NULL, page_folio(page)); 2984f47960f4SQu Wenruo lock_page(page); 2985f47960f4SQu Wenruo if (!PageUptodate(page)) { 2986f47960f4SQu Wenruo ret = -EIO; 2987f47960f4SQu Wenruo goto release_page; 2988f47960f4SQu Wenruo } 2989f47960f4SQu Wenruo } 2990f47960f4SQu Wenruo 2991f47960f4SQu Wenruo page_start = page_offset(page); 2992f47960f4SQu Wenruo page_end = page_start + PAGE_SIZE - 1; 2993f47960f4SQu Wenruo 2994c2832898SQu Wenruo /* 2995c2832898SQu Wenruo * Start from the cluster, as for subpage case, the cluster can start 2996c2832898SQu Wenruo * inside the page. 2997c2832898SQu Wenruo */ 2998c2832898SQu Wenruo cur = max(page_start, cluster->boundary[*cluster_nr] - offset); 2999c2832898SQu Wenruo while (cur <= page_end) { 30009c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 3001c2832898SQu Wenruo u64 extent_start = cluster->boundary[*cluster_nr] - offset; 3002c2832898SQu Wenruo u64 extent_end = get_cluster_boundary_end(cluster, 3003c2832898SQu Wenruo *cluster_nr) - offset; 3004c2832898SQu Wenruo u64 clamped_start = max(page_start, extent_start); 3005c2832898SQu Wenruo u64 clamped_end = min(page_end, extent_end); 3006c2832898SQu Wenruo u32 clamped_len = clamped_end + 1 - clamped_start; 3007f47960f4SQu Wenruo 3008c2832898SQu Wenruo /* Reserve metadata for this range */ 3009c2832898SQu Wenruo ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 3010d4135134SFilipe Manana clamped_len, clamped_len, 3011d4135134SFilipe Manana false); 3012c2832898SQu Wenruo if (ret) 3013f47960f4SQu Wenruo goto release_page; 3014f47960f4SQu Wenruo 3015c2832898SQu Wenruo /* Mark the range delalloc and dirty for later writeback */ 30169c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, 30179c5c9604SJosef Bacik &cached_state); 3018c2832898SQu Wenruo ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start, 30199c5c9604SJosef Bacik clamped_end, 0, &cached_state); 3020c2832898SQu Wenruo if (ret) { 30219c5c9604SJosef Bacik clear_extent_bit(&BTRFS_I(inode)->io_tree, 3022c2832898SQu Wenruo clamped_start, clamped_end, 30239c5c9604SJosef Bacik EXTENT_LOCKED | EXTENT_BOUNDARY, 30249c5c9604SJosef Bacik &cached_state); 3025c2832898SQu Wenruo btrfs_delalloc_release_metadata(BTRFS_I(inode), 3026c2832898SQu Wenruo clamped_len, true); 3027c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), 3028c2832898SQu Wenruo clamped_len); 3029c2832898SQu Wenruo goto release_page; 3030f47960f4SQu Wenruo } 3031c2832898SQu Wenruo btrfs_page_set_dirty(fs_info, page, clamped_start, clamped_len); 3032f47960f4SQu Wenruo 3033c2832898SQu Wenruo /* 3034c2832898SQu Wenruo * Set the boundary if it's inside the page. 3035c2832898SQu Wenruo * Data relocation requires the destination extents to have the 3036c2832898SQu Wenruo * same size as the source. 3037c2832898SQu Wenruo * EXTENT_BOUNDARY bit prevents current extent from being merged 3038c2832898SQu Wenruo * with previous extent. 3039c2832898SQu Wenruo */ 3040c2832898SQu Wenruo if (in_range(cluster->boundary[*cluster_nr] - offset, 3041c2832898SQu Wenruo page_start, PAGE_SIZE)) { 3042c2832898SQu Wenruo u64 boundary_start = cluster->boundary[*cluster_nr] - 3043c2832898SQu Wenruo offset; 3044c2832898SQu Wenruo u64 boundary_end = boundary_start + 3045c2832898SQu Wenruo fs_info->sectorsize - 1; 3046c2832898SQu Wenruo 3047c2832898SQu Wenruo set_extent_bits(&BTRFS_I(inode)->io_tree, 3048c2832898SQu Wenruo boundary_start, boundary_end, 3049c2832898SQu Wenruo EXTENT_BOUNDARY); 3050c2832898SQu Wenruo } 30519c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, 30529c5c9604SJosef Bacik &cached_state); 3053c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len); 3054c2832898SQu Wenruo cur += clamped_len; 3055c2832898SQu Wenruo 3056c2832898SQu Wenruo /* Crossed extent end, go to next extent */ 3057c2832898SQu Wenruo if (cur >= extent_end) { 3058c2832898SQu Wenruo (*cluster_nr)++; 3059c2832898SQu Wenruo /* Just finished the last extent of the cluster, exit. */ 3060c2832898SQu Wenruo if (*cluster_nr >= cluster->nr) 3061c2832898SQu Wenruo break; 3062c2832898SQu Wenruo } 3063c2832898SQu Wenruo } 3064f47960f4SQu Wenruo unlock_page(page); 3065f47960f4SQu Wenruo put_page(page); 3066f47960f4SQu Wenruo 3067f47960f4SQu Wenruo balance_dirty_pages_ratelimited(inode->i_mapping); 3068f47960f4SQu Wenruo btrfs_throttle(fs_info); 3069f47960f4SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) 3070f47960f4SQu Wenruo ret = -ECANCELED; 3071f47960f4SQu Wenruo return ret; 3072f47960f4SQu Wenruo 3073f47960f4SQu Wenruo release_page: 3074f47960f4SQu Wenruo unlock_page(page); 3075f47960f4SQu Wenruo put_page(page); 3076f47960f4SQu Wenruo return ret; 3077f47960f4SQu Wenruo } 3078f47960f4SQu Wenruo 30790257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 30800257bb82SYan, Zheng struct file_extent_cluster *cluster) 30810257bb82SYan, Zheng { 30820257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 30830257bb82SYan, Zheng unsigned long index; 30840257bb82SYan, Zheng unsigned long last_index; 30850257bb82SYan, Zheng struct file_ra_state *ra; 3086f47960f4SQu Wenruo int cluster_nr = 0; 30870257bb82SYan, Zheng int ret = 0; 30880257bb82SYan, Zheng 30890257bb82SYan, Zheng if (!cluster->nr) 30900257bb82SYan, Zheng return 0; 30910257bb82SYan, Zheng 30920257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 30930257bb82SYan, Zheng if (!ra) 30940257bb82SYan, Zheng return -ENOMEM; 30950257bb82SYan, Zheng 3096056d9becSNikolay Borisov ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster); 30970257bb82SYan, Zheng if (ret) 3098efa56464SYan, Zheng goto out; 30990257bb82SYan, Zheng 31000257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 31010257bb82SYan, Zheng 31024b01c44fSJohannes Thumshirn ret = setup_relocation_extent_mapping(inode, cluster->start - offset, 3103efa56464SYan, Zheng cluster->end - offset, cluster->start); 3104efa56464SYan, Zheng if (ret) 3105efa56464SYan, Zheng goto out; 3106efa56464SYan, Zheng 310709cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 3108f47960f4SQu Wenruo for (index = (cluster->start - offset) >> PAGE_SHIFT; 3109f47960f4SQu Wenruo index <= last_index && !ret; index++) 3110f47960f4SQu Wenruo ret = relocate_one_page(inode, ra, cluster, &cluster_nr, index); 3111f47960f4SQu Wenruo if (ret == 0) 3112f47960f4SQu Wenruo WARN_ON(cluster_nr != cluster->nr); 3113efa56464SYan, Zheng out: 31140257bb82SYan, Zheng kfree(ra); 31150257bb82SYan, Zheng return ret; 31160257bb82SYan, Zheng } 31170257bb82SYan, Zheng 31180257bb82SYan, Zheng static noinline_for_stack 31190257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 31200257bb82SYan, Zheng struct file_extent_cluster *cluster) 31210257bb82SYan, Zheng { 31220257bb82SYan, Zheng int ret; 31230257bb82SYan, Zheng 31240257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 31250257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31260257bb82SYan, Zheng if (ret) 31270257bb82SYan, Zheng return ret; 31280257bb82SYan, Zheng cluster->nr = 0; 31290257bb82SYan, Zheng } 31300257bb82SYan, Zheng 31310257bb82SYan, Zheng if (!cluster->nr) 31320257bb82SYan, Zheng cluster->start = extent_key->objectid; 31330257bb82SYan, Zheng else 31340257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 31350257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 31360257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 31370257bb82SYan, Zheng cluster->nr++; 31380257bb82SYan, Zheng 31390257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 31400257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31410257bb82SYan, Zheng if (ret) 31420257bb82SYan, Zheng return ret; 31430257bb82SYan, Zheng cluster->nr = 0; 31440257bb82SYan, Zheng } 31450257bb82SYan, Zheng return 0; 31465d4f98a2SYan Zheng } 31475d4f98a2SYan Zheng 31485d4f98a2SYan Zheng /* 31495d4f98a2SYan Zheng * helper to add a tree block to the list. 31505d4f98a2SYan Zheng * the major work is getting the generation and level of the block 31515d4f98a2SYan Zheng */ 31525d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 31535d4f98a2SYan Zheng struct btrfs_key *extent_key, 31545d4f98a2SYan Zheng struct btrfs_path *path, 31555d4f98a2SYan Zheng struct rb_root *blocks) 31565d4f98a2SYan Zheng { 31575d4f98a2SYan Zheng struct extent_buffer *eb; 31585d4f98a2SYan Zheng struct btrfs_extent_item *ei; 31595d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 31605d4f98a2SYan Zheng struct tree_block *block; 31615d4f98a2SYan Zheng struct rb_node *rb_node; 31625d4f98a2SYan Zheng u32 item_size; 31635d4f98a2SYan Zheng int level = -1; 31647fdf4b60SWang Shilong u64 generation; 3165f7ba2d37SJosef Bacik u64 owner = 0; 31665d4f98a2SYan Zheng 31675d4f98a2SYan Zheng eb = path->nodes[0]; 31683212fa14SJosef Bacik item_size = btrfs_item_size(eb, path->slots[0]); 31695d4f98a2SYan Zheng 31703173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 31713173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 3172f7ba2d37SJosef Bacik unsigned long ptr = 0, end; 3173f7ba2d37SJosef Bacik 31745d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 31755d4f98a2SYan Zheng struct btrfs_extent_item); 3176f7ba2d37SJosef Bacik end = (unsigned long)ei + item_size; 31773173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 31785d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 31795d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 3180f7ba2d37SJosef Bacik ptr = (unsigned long)(bi + 1); 31815d4f98a2SYan Zheng } else { 31823173a18fSJosef Bacik level = (int)extent_key->offset; 3183f7ba2d37SJosef Bacik ptr = (unsigned long)(ei + 1); 31843173a18fSJosef Bacik } 31853173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 3186f7ba2d37SJosef Bacik 3187f7ba2d37SJosef Bacik /* 3188f7ba2d37SJosef Bacik * We're reading random blocks without knowing their owner ahead 3189f7ba2d37SJosef Bacik * of time. This is ok most of the time, as all reloc roots and 3190f7ba2d37SJosef Bacik * fs roots have the same lock type. However normal trees do 3191f7ba2d37SJosef Bacik * not, and the only way to know ahead of time is to read the 3192f7ba2d37SJosef Bacik * inline ref offset. We know it's an fs root if 3193f7ba2d37SJosef Bacik * 3194f7ba2d37SJosef Bacik * 1. There's more than one ref. 3195f7ba2d37SJosef Bacik * 2. There's a SHARED_DATA_REF_KEY set. 3196f7ba2d37SJosef Bacik * 3. FULL_BACKREF is set on the flags. 3197f7ba2d37SJosef Bacik * 3198f7ba2d37SJosef Bacik * Otherwise it's safe to assume that the ref offset == the 3199f7ba2d37SJosef Bacik * owner of this block, so we can use that when calling 3200f7ba2d37SJosef Bacik * read_tree_block. 3201f7ba2d37SJosef Bacik */ 3202f7ba2d37SJosef Bacik if (btrfs_extent_refs(eb, ei) == 1 && 3203f7ba2d37SJosef Bacik !(btrfs_extent_flags(eb, ei) & 3204f7ba2d37SJosef Bacik BTRFS_BLOCK_FLAG_FULL_BACKREF) && 3205f7ba2d37SJosef Bacik ptr < end) { 3206f7ba2d37SJosef Bacik struct btrfs_extent_inline_ref *iref; 3207f7ba2d37SJosef Bacik int type; 3208f7ba2d37SJosef Bacik 3209f7ba2d37SJosef Bacik iref = (struct btrfs_extent_inline_ref *)ptr; 3210f7ba2d37SJosef Bacik type = btrfs_get_extent_inline_ref_type(eb, iref, 3211f7ba2d37SJosef Bacik BTRFS_REF_TYPE_BLOCK); 3212f7ba2d37SJosef Bacik if (type == BTRFS_REF_TYPE_INVALID) 3213f7ba2d37SJosef Bacik return -EINVAL; 3214f7ba2d37SJosef Bacik if (type == BTRFS_TREE_BLOCK_REF_KEY) 3215f7ba2d37SJosef Bacik owner = btrfs_extent_inline_ref_offset(eb, iref); 3216f7ba2d37SJosef Bacik } 32176d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3218ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3219ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3220ba3c2b19SNikolay Borisov return -EINVAL; 32213173a18fSJosef Bacik } else { 32225d4f98a2SYan Zheng BUG(); 32235d4f98a2SYan Zheng } 32245d4f98a2SYan Zheng 3225b3b4aa74SDavid Sterba btrfs_release_path(path); 32265d4f98a2SYan Zheng 32275d4f98a2SYan Zheng BUG_ON(level == -1); 32285d4f98a2SYan Zheng 32295d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 32305d4f98a2SYan Zheng if (!block) 32315d4f98a2SYan Zheng return -ENOMEM; 32325d4f98a2SYan Zheng 32335d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3234da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 32355d4f98a2SYan Zheng block->key.offset = generation; 32365d4f98a2SYan Zheng block->level = level; 32375d4f98a2SYan Zheng block->key_ready = 0; 3238f7ba2d37SJosef Bacik block->owner = owner; 32395d4f98a2SYan Zheng 3240e9a28dc5SQu Wenruo rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node); 324143c04fb1SJeff Mahoney if (rb_node) 3242982c92cbSQu Wenruo btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr, 3243982c92cbSQu Wenruo -EEXIST); 32445d4f98a2SYan Zheng 32455d4f98a2SYan Zheng return 0; 32465d4f98a2SYan Zheng } 32475d4f98a2SYan Zheng 32485d4f98a2SYan Zheng /* 32495d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 32505d4f98a2SYan Zheng */ 32515d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 32525d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 32535d4f98a2SYan Zheng struct rb_root *blocks) 32545d4f98a2SYan Zheng { 32550b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 32565d4f98a2SYan Zheng struct btrfs_path *path; 32575d4f98a2SYan Zheng struct btrfs_key key; 32585d4f98a2SYan Zheng int ret; 32590b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 32605d4f98a2SYan Zheng 32617476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 32625d4f98a2SYan Zheng return 0; 32635d4f98a2SYan Zheng 3264e9a28dc5SQu Wenruo if (rb_simple_search(blocks, bytenr)) 32655d4f98a2SYan Zheng return 0; 32665d4f98a2SYan Zheng 32675d4f98a2SYan Zheng path = btrfs_alloc_path(); 32685d4f98a2SYan Zheng if (!path) 32695d4f98a2SYan Zheng return -ENOMEM; 3270aee68ee5SJosef Bacik again: 32715d4f98a2SYan Zheng key.objectid = bytenr; 3272aee68ee5SJosef Bacik if (skinny) { 3273aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3274aee68ee5SJosef Bacik key.offset = (u64)-1; 3275aee68ee5SJosef Bacik } else { 32765d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 32775d4f98a2SYan Zheng key.offset = blocksize; 3278aee68ee5SJosef Bacik } 32795d4f98a2SYan Zheng 32805d4f98a2SYan Zheng path->search_commit_root = 1; 32815d4f98a2SYan Zheng path->skip_locking = 1; 32825d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 32835d4f98a2SYan Zheng if (ret < 0) 32845d4f98a2SYan Zheng goto out; 32855d4f98a2SYan Zheng 3286aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3287aee68ee5SJosef Bacik if (path->slots[0]) { 3288aee68ee5SJosef Bacik path->slots[0]--; 3289aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3290aee68ee5SJosef Bacik path->slots[0]); 32913173a18fSJosef Bacik if (key.objectid == bytenr && 3292aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3293aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3294aee68ee5SJosef Bacik key.offset == blocksize))) 32953173a18fSJosef Bacik ret = 0; 32963173a18fSJosef Bacik } 3297aee68ee5SJosef Bacik 3298aee68ee5SJosef Bacik if (ret) { 3299aee68ee5SJosef Bacik skinny = false; 3300aee68ee5SJosef Bacik btrfs_release_path(path); 3301aee68ee5SJosef Bacik goto again; 3302aee68ee5SJosef Bacik } 3303aee68ee5SJosef Bacik } 3304cdccee99SLiu Bo if (ret) { 3305cdccee99SLiu Bo ASSERT(ret == 1); 3306cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3307cdccee99SLiu Bo btrfs_err(fs_info, 3308cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3309cdccee99SLiu Bo bytenr); 3310cdccee99SLiu Bo WARN_ON(1); 3311cdccee99SLiu Bo ret = -EINVAL; 3312cdccee99SLiu Bo goto out; 3313cdccee99SLiu Bo } 33143173a18fSJosef Bacik 33155d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 33165d4f98a2SYan Zheng out: 33175d4f98a2SYan Zheng btrfs_free_path(path); 33185d4f98a2SYan Zheng return ret; 33195d4f98a2SYan Zheng } 33205d4f98a2SYan Zheng 33210af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 332232da5386SDavid Sterba struct btrfs_block_group *block_group, 33231bbc621eSChris Mason struct inode *inode, 33241bbc621eSChris Mason u64 ino) 33250af3d00bSJosef Bacik { 33260af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 33270af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 33280af3d00bSJosef Bacik int ret = 0; 33290af3d00bSJosef Bacik 33300af3d00bSJosef Bacik if (inode) 33310af3d00bSJosef Bacik goto truncate; 33320af3d00bSJosef Bacik 33330202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, ino, root); 33342e19f1f9SAl Viro if (IS_ERR(inode)) 33350af3d00bSJosef Bacik return -ENOENT; 33360af3d00bSJosef Bacik 33370af3d00bSJosef Bacik truncate: 33382ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 33397b61cd92SMiao Xie &fs_info->global_block_rsv); 33407b61cd92SMiao Xie if (ret) 33417b61cd92SMiao Xie goto out; 33427b61cd92SMiao Xie 33437a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 33440af3d00bSJosef Bacik if (IS_ERR(trans)) { 33453612b495STsutomu Itoh ret = PTR_ERR(trans); 33460af3d00bSJosef Bacik goto out; 33470af3d00bSJosef Bacik } 33480af3d00bSJosef Bacik 334977ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 33500af3d00bSJosef Bacik 33513a45bb20SJeff Mahoney btrfs_end_transaction(trans); 33522ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 33530af3d00bSJosef Bacik out: 33540af3d00bSJosef Bacik iput(inode); 33550af3d00bSJosef Bacik return ret; 33560af3d00bSJosef Bacik } 33570af3d00bSJosef Bacik 33585d4f98a2SYan Zheng /* 335919b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 336019b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 33615d4f98a2SYan Zheng */ 336219b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 336319b546d7SQu Wenruo struct btrfs_block_group *block_group, 336419b546d7SQu Wenruo u64 data_bytenr) 33655d4f98a2SYan Zheng { 336619b546d7SQu Wenruo u64 space_cache_ino; 336719b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 33685d4f98a2SYan Zheng struct btrfs_key key; 336919b546d7SQu Wenruo bool found = false; 337019b546d7SQu Wenruo int i; 33715d4f98a2SYan Zheng int ret; 33725d4f98a2SYan Zheng 337319b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 337419b546d7SQu Wenruo return 0; 33755d4f98a2SYan Zheng 337619b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 337750e31ef4SQu Wenruo u8 type; 337850e31ef4SQu Wenruo 337919b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 338019b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 338119b546d7SQu Wenruo continue; 338219b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 338350e31ef4SQu Wenruo type = btrfs_file_extent_type(leaf, ei); 338450e31ef4SQu Wenruo 338550e31ef4SQu Wenruo if ((type == BTRFS_FILE_EXTENT_REG || 338650e31ef4SQu Wenruo type == BTRFS_FILE_EXTENT_PREALLOC) && 338719b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 338819b546d7SQu Wenruo found = true; 338919b546d7SQu Wenruo space_cache_ino = key.objectid; 339019b546d7SQu Wenruo break; 339119b546d7SQu Wenruo } 339219b546d7SQu Wenruo } 339319b546d7SQu Wenruo if (!found) 339419b546d7SQu Wenruo return -ENOENT; 339519b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 339619b546d7SQu Wenruo space_cache_ino); 33970af3d00bSJosef Bacik return ret; 33985d4f98a2SYan Zheng } 33995d4f98a2SYan Zheng 34005d4f98a2SYan Zheng /* 34012c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 34025d4f98a2SYan Zheng */ 34035d4f98a2SYan Zheng static noinline_for_stack 34045d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 34055d4f98a2SYan Zheng struct btrfs_key *extent_key, 34065d4f98a2SYan Zheng struct btrfs_path *path, 34075d4f98a2SYan Zheng struct rb_root *blocks) 34085d4f98a2SYan Zheng { 340919b546d7SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 341019b546d7SQu Wenruo struct ulist *leaves = NULL; 341119b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 341219b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 341319b546d7SQu Wenruo const u32 blocksize = fs_info->nodesize; 3414647f63bdSFilipe David Borba Manana int ret = 0; 34155d4f98a2SYan Zheng 3416b3b4aa74SDavid Sterba btrfs_release_path(path); 341719b546d7SQu Wenruo ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid, 341819b546d7SQu Wenruo 0, &leaves, NULL, true); 341919b546d7SQu Wenruo if (ret < 0) 342019b546d7SQu Wenruo return ret; 342119b546d7SQu Wenruo 342219b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 342319b546d7SQu Wenruo while ((ref_node = ulist_next(leaves, &leaf_uiter))) { 342419b546d7SQu Wenruo struct extent_buffer *eb; 342519b546d7SQu Wenruo 34261b7ec85eSJosef Bacik eb = read_tree_block(fs_info, ref_node->val, 0, 0, 0, NULL); 342719b546d7SQu Wenruo if (IS_ERR(eb)) { 342819b546d7SQu Wenruo ret = PTR_ERR(eb); 342919b546d7SQu Wenruo break; 343019b546d7SQu Wenruo } 343119b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 343219b546d7SQu Wenruo extent_key->objectid); 343319b546d7SQu Wenruo free_extent_buffer(eb); 343419b546d7SQu Wenruo if (ret < 0) 343519b546d7SQu Wenruo break; 343619b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 343719b546d7SQu Wenruo if (ret < 0) 343819b546d7SQu Wenruo break; 343919b546d7SQu Wenruo } 344019b546d7SQu Wenruo if (ret < 0) 34415d4f98a2SYan Zheng free_block_list(blocks); 344219b546d7SQu Wenruo ulist_free(leaves); 344319b546d7SQu Wenruo return ret; 34445d4f98a2SYan Zheng } 34455d4f98a2SYan Zheng 34465d4f98a2SYan Zheng /* 34472c016dc2SLiu Bo * helper to find next unprocessed extent 34485d4f98a2SYan Zheng */ 34495d4f98a2SYan Zheng static noinline_for_stack 3450147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 34513fd0a558SYan, Zheng struct btrfs_key *extent_key) 34525d4f98a2SYan Zheng { 34530b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34545d4f98a2SYan Zheng struct btrfs_key key; 34555d4f98a2SYan Zheng struct extent_buffer *leaf; 34565d4f98a2SYan Zheng u64 start, end, last; 34575d4f98a2SYan Zheng int ret; 34585d4f98a2SYan Zheng 3459b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 34605d4f98a2SYan Zheng while (1) { 34615d4f98a2SYan Zheng cond_resched(); 34625d4f98a2SYan Zheng if (rc->search_start >= last) { 34635d4f98a2SYan Zheng ret = 1; 34645d4f98a2SYan Zheng break; 34655d4f98a2SYan Zheng } 34665d4f98a2SYan Zheng 34675d4f98a2SYan Zheng key.objectid = rc->search_start; 34685d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34695d4f98a2SYan Zheng key.offset = 0; 34705d4f98a2SYan Zheng 34715d4f98a2SYan Zheng path->search_commit_root = 1; 34725d4f98a2SYan Zheng path->skip_locking = 1; 34735d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 34745d4f98a2SYan Zheng 0, 0); 34755d4f98a2SYan Zheng if (ret < 0) 34765d4f98a2SYan Zheng break; 34775d4f98a2SYan Zheng next: 34785d4f98a2SYan Zheng leaf = path->nodes[0]; 34795d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 34805d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 34815d4f98a2SYan Zheng if (ret != 0) 34825d4f98a2SYan Zheng break; 34835d4f98a2SYan Zheng leaf = path->nodes[0]; 34845d4f98a2SYan Zheng } 34855d4f98a2SYan Zheng 34865d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 34875d4f98a2SYan Zheng if (key.objectid >= last) { 34885d4f98a2SYan Zheng ret = 1; 34895d4f98a2SYan Zheng break; 34905d4f98a2SYan Zheng } 34915d4f98a2SYan Zheng 34923173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 34933173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 34943173a18fSJosef Bacik path->slots[0]++; 34953173a18fSJosef Bacik goto next; 34963173a18fSJosef Bacik } 34973173a18fSJosef Bacik 34983173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 34995d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 35005d4f98a2SYan Zheng path->slots[0]++; 35015d4f98a2SYan Zheng goto next; 35025d4f98a2SYan Zheng } 35035d4f98a2SYan Zheng 35043173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 35050b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 35063173a18fSJosef Bacik rc->search_start) { 35073173a18fSJosef Bacik path->slots[0]++; 35083173a18fSJosef Bacik goto next; 35093173a18fSJosef Bacik } 35103173a18fSJosef Bacik 35115d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 35125d4f98a2SYan Zheng key.objectid, &start, &end, 3513e6138876SJosef Bacik EXTENT_DIRTY, NULL); 35145d4f98a2SYan Zheng 35155d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3516b3b4aa74SDavid Sterba btrfs_release_path(path); 35175d4f98a2SYan Zheng rc->search_start = end + 1; 35185d4f98a2SYan Zheng } else { 35193173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 35205d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 35213173a18fSJosef Bacik else 35223173a18fSJosef Bacik rc->search_start = key.objectid + 35230b246afaSJeff Mahoney fs_info->nodesize; 35243fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 35255d4f98a2SYan Zheng return 0; 35265d4f98a2SYan Zheng } 35275d4f98a2SYan Zheng } 3528b3b4aa74SDavid Sterba btrfs_release_path(path); 35295d4f98a2SYan Zheng return ret; 35305d4f98a2SYan Zheng } 35315d4f98a2SYan Zheng 35325d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 35335d4f98a2SYan Zheng { 35345d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35357585717fSChris Mason 35367585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35375d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 35387585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35395d4f98a2SYan Zheng } 35405d4f98a2SYan Zheng 35415d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 35425d4f98a2SYan Zheng { 35435d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35447585717fSChris Mason 35457585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35465d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 35477585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35485d4f98a2SYan Zheng } 35495d4f98a2SYan Zheng 35503fd0a558SYan, Zheng static noinline_for_stack 35513fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 35523fd0a558SYan, Zheng { 35533fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3554ac2fabacSJosef Bacik int ret; 35553fd0a558SYan, Zheng 35562ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 355766d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 35583fd0a558SYan, Zheng if (!rc->block_rsv) 35593fd0a558SYan, Zheng return -ENOMEM; 35603fd0a558SYan, Zheng 35613fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3562b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 35633fd0a558SYan, Zheng rc->extents_found = 0; 35643fd0a558SYan, Zheng rc->nodes_relocated = 0; 35653fd0a558SYan, Zheng rc->merging_rsv_size = 0; 35660647bf56SWang Shilong rc->reserved_bytes = 0; 3567da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 35680647bf56SWang Shilong RELOCATION_RESERVED_NODES; 35699270501cSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root->fs_info, 3570ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3571ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3572ac2fabacSJosef Bacik if (ret) 3573ac2fabacSJosef Bacik return ret; 35743fd0a558SYan, Zheng 35753fd0a558SYan, Zheng rc->create_reloc_tree = 1; 35763fd0a558SYan, Zheng set_reloc_control(rc); 35773fd0a558SYan, Zheng 35787a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 357928818947SLiu Bo if (IS_ERR(trans)) { 358028818947SLiu Bo unset_reloc_control(rc); 358128818947SLiu Bo /* 358228818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 358328818947SLiu Bo * cleanup. And callers are responsible to free the above 358428818947SLiu Bo * block rsv. 358528818947SLiu Bo */ 358628818947SLiu Bo return PTR_ERR(trans); 358728818947SLiu Bo } 358885f02d6cSZixuan Fu 358985f02d6cSZixuan Fu ret = btrfs_commit_transaction(trans); 359085f02d6cSZixuan Fu if (ret) 359185f02d6cSZixuan Fu unset_reloc_control(rc); 359285f02d6cSZixuan Fu 359385f02d6cSZixuan Fu return ret; 35943fd0a558SYan, Zheng } 359576dda93cSYan, Zheng 35965d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 35975d4f98a2SYan Zheng { 35982ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35995d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 36005d4f98a2SYan Zheng struct btrfs_key key; 36015d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 36025d4f98a2SYan Zheng struct btrfs_path *path; 36035d4f98a2SYan Zheng struct btrfs_extent_item *ei; 36045d4f98a2SYan Zheng u64 flags; 36055d4f98a2SYan Zheng int ret; 36065d4f98a2SYan Zheng int err = 0; 3607c87f08caSChris Mason int progress = 0; 36085d4f98a2SYan Zheng 36095d4f98a2SYan Zheng path = btrfs_alloc_path(); 36103fd0a558SYan, Zheng if (!path) 36115d4f98a2SYan Zheng return -ENOMEM; 3612e4058b54SDavid Sterba path->reada = READA_FORWARD; 36133fd0a558SYan, Zheng 36143fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 36153fd0a558SYan, Zheng if (ret) { 36163fd0a558SYan, Zheng err = ret; 36173fd0a558SYan, Zheng goto out_free; 36182423fdfbSJiri Slaby } 36195d4f98a2SYan Zheng 36205d4f98a2SYan Zheng while (1) { 36210647bf56SWang Shilong rc->reserved_bytes = 0; 36229270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 36239270501cSJosef Bacik rc->block_rsv->size, 36240647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 36250647bf56SWang Shilong if (ret) { 36260647bf56SWang Shilong err = ret; 36270647bf56SWang Shilong break; 36280647bf56SWang Shilong } 3629c87f08caSChris Mason progress++; 3630a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 36310f788c58SLiu Bo if (IS_ERR(trans)) { 36320f788c58SLiu Bo err = PTR_ERR(trans); 36330f788c58SLiu Bo trans = NULL; 36340f788c58SLiu Bo break; 36350f788c58SLiu Bo } 3636c87f08caSChris Mason restart: 36373fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 36383a45bb20SJeff Mahoney btrfs_end_transaction(trans); 363942a657f5SPan Bian trans = NULL; 36403fd0a558SYan, Zheng continue; 36413fd0a558SYan, Zheng } 36423fd0a558SYan, Zheng 3643147d256eSZhaolei ret = find_next_extent(rc, path, &key); 36445d4f98a2SYan Zheng if (ret < 0) 36455d4f98a2SYan Zheng err = ret; 36465d4f98a2SYan Zheng if (ret != 0) 36475d4f98a2SYan Zheng break; 36485d4f98a2SYan Zheng 36495d4f98a2SYan Zheng rc->extents_found++; 36505d4f98a2SYan Zheng 36515d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 36525d4f98a2SYan Zheng struct btrfs_extent_item); 36535d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 36545d4f98a2SYan Zheng 36555d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 36565d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 36575d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 36585d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36595d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 36605d4f98a2SYan Zheng } else { 3661b3b4aa74SDavid Sterba btrfs_release_path(path); 36625d4f98a2SYan Zheng ret = 0; 36635d4f98a2SYan Zheng } 36645d4f98a2SYan Zheng if (ret < 0) { 36653fd0a558SYan, Zheng err = ret; 36665d4f98a2SYan Zheng break; 36675d4f98a2SYan Zheng } 36685d4f98a2SYan Zheng 36695d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 36705d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 36715d4f98a2SYan Zheng if (ret < 0) { 36723fd0a558SYan, Zheng if (ret != -EAGAIN) { 36735d4f98a2SYan Zheng err = ret; 36745d4f98a2SYan Zheng break; 36755d4f98a2SYan Zheng } 36763fd0a558SYan, Zheng rc->extents_found--; 36773fd0a558SYan, Zheng rc->search_start = key.objectid; 36783fd0a558SYan, Zheng } 36795d4f98a2SYan Zheng } 36805d4f98a2SYan Zheng 36813a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 36822ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36833fd0a558SYan, Zheng trans = NULL; 36845d4f98a2SYan Zheng 36855d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 36865d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36875d4f98a2SYan Zheng rc->found_file_extent = 1; 36880257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 36893fd0a558SYan, Zheng &key, &rc->cluster); 36905d4f98a2SYan Zheng if (ret < 0) { 36915d4f98a2SYan Zheng err = ret; 36925d4f98a2SYan Zheng break; 36935d4f98a2SYan Zheng } 36945d4f98a2SYan Zheng } 3695f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 3696f31ea088SQu Wenruo err = -ECANCELED; 3697f31ea088SQu Wenruo break; 3698f31ea088SQu Wenruo } 36995d4f98a2SYan Zheng } 3700c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 370143a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 37029689457bSShilong Wang if (ret == 1) { 3703c87f08caSChris Mason err = 0; 3704c87f08caSChris Mason progress = 0; 3705c87f08caSChris Mason goto restart; 3706c87f08caSChris Mason } 3707c87f08caSChris Mason } 37083fd0a558SYan, Zheng 3709b3b4aa74SDavid Sterba btrfs_release_path(path); 371091166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 37115d4f98a2SYan Zheng 37125d4f98a2SYan Zheng if (trans) { 37133a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 37142ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 37155d4f98a2SYan Zheng } 37165d4f98a2SYan Zheng 37170257bb82SYan, Zheng if (!err) { 37183fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 37193fd0a558SYan, Zheng &rc->cluster); 37200257bb82SYan, Zheng if (ret < 0) 37210257bb82SYan, Zheng err = ret; 37220257bb82SYan, Zheng } 37230257bb82SYan, Zheng 37243fd0a558SYan, Zheng rc->create_reloc_tree = 0; 37253fd0a558SYan, Zheng set_reloc_control(rc); 37260257bb82SYan, Zheng 372713fe1bdbSQu Wenruo btrfs_backref_release_cache(&rc->backref_cache); 372863f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37295d4f98a2SYan Zheng 37307f913c7cSQu Wenruo /* 37317f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 37327f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 37337f913c7cSQu Wenruo * 37347f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 37357f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 37367f913c7cSQu Wenruo * merge_reloc_roots() 37377f913c7cSQu Wenruo */ 37383fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 37395d4f98a2SYan Zheng 37405d4f98a2SYan Zheng merge_reloc_roots(rc); 37415d4f98a2SYan Zheng 37423fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 37435d4f98a2SYan Zheng unset_reloc_control(rc); 374463f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37455d4f98a2SYan Zheng 37465d4f98a2SYan Zheng /* get rid of pinned extents */ 37477a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 374862b99540SQu Wenruo if (IS_ERR(trans)) { 37493612b495STsutomu Itoh err = PTR_ERR(trans); 375062b99540SQu Wenruo goto out_free; 375162b99540SQu Wenruo } 3752fb686c68SJosef Bacik ret = btrfs_commit_transaction(trans); 3753fb686c68SJosef Bacik if (ret && !err) 3754fb686c68SJosef Bacik err = ret; 37556217b0faSJosef Bacik out_free: 3756d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 3757d2311e69SQu Wenruo if (ret < 0 && !err) 3758d2311e69SQu Wenruo err = ret; 37592ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 37603fd0a558SYan, Zheng btrfs_free_path(path); 37615d4f98a2SYan Zheng return err; 37625d4f98a2SYan Zheng } 37635d4f98a2SYan Zheng 37645d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 37650257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 37665d4f98a2SYan Zheng { 37675d4f98a2SYan Zheng struct btrfs_path *path; 37685d4f98a2SYan Zheng struct btrfs_inode_item *item; 37695d4f98a2SYan Zheng struct extent_buffer *leaf; 37705d4f98a2SYan Zheng int ret; 37715d4f98a2SYan Zheng 37725d4f98a2SYan Zheng path = btrfs_alloc_path(); 37735d4f98a2SYan Zheng if (!path) 37745d4f98a2SYan Zheng return -ENOMEM; 37755d4f98a2SYan Zheng 37765d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 37775d4f98a2SYan Zheng if (ret) 37785d4f98a2SYan Zheng goto out; 37795d4f98a2SYan Zheng 37805d4f98a2SYan Zheng leaf = path->nodes[0]; 37815d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 3782b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 37835d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 37840257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 37855d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 3786960a3166SJohannes Thumshirn btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 3787960a3166SJohannes Thumshirn BTRFS_INODE_PREALLOC); 37885d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 37895d4f98a2SYan Zheng out: 37905d4f98a2SYan Zheng btrfs_free_path(path); 37915d4f98a2SYan Zheng return ret; 37925d4f98a2SYan Zheng } 37935d4f98a2SYan Zheng 3794790c1b8cSJosef Bacik static void delete_orphan_inode(struct btrfs_trans_handle *trans, 3795790c1b8cSJosef Bacik struct btrfs_root *root, u64 objectid) 3796790c1b8cSJosef Bacik { 3797790c1b8cSJosef Bacik struct btrfs_path *path; 3798790c1b8cSJosef Bacik struct btrfs_key key; 3799790c1b8cSJosef Bacik int ret = 0; 3800790c1b8cSJosef Bacik 3801790c1b8cSJosef Bacik path = btrfs_alloc_path(); 3802790c1b8cSJosef Bacik if (!path) { 3803790c1b8cSJosef Bacik ret = -ENOMEM; 3804790c1b8cSJosef Bacik goto out; 3805790c1b8cSJosef Bacik } 3806790c1b8cSJosef Bacik 3807790c1b8cSJosef Bacik key.objectid = objectid; 3808790c1b8cSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 3809790c1b8cSJosef Bacik key.offset = 0; 3810790c1b8cSJosef Bacik ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3811790c1b8cSJosef Bacik if (ret) { 3812790c1b8cSJosef Bacik if (ret > 0) 3813790c1b8cSJosef Bacik ret = -ENOENT; 3814790c1b8cSJosef Bacik goto out; 3815790c1b8cSJosef Bacik } 3816790c1b8cSJosef Bacik ret = btrfs_del_item(trans, root, path); 3817790c1b8cSJosef Bacik out: 3818790c1b8cSJosef Bacik if (ret) 3819790c1b8cSJosef Bacik btrfs_abort_transaction(trans, ret); 3820790c1b8cSJosef Bacik btrfs_free_path(path); 3821790c1b8cSJosef Bacik } 3822790c1b8cSJosef Bacik 38235d4f98a2SYan Zheng /* 38245d4f98a2SYan Zheng * helper to create inode for data relocation. 38255d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 38265d4f98a2SYan Zheng */ 38273fd0a558SYan, Zheng static noinline_for_stack 38283fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 382932da5386SDavid Sterba struct btrfs_block_group *group) 38305d4f98a2SYan Zheng { 38315d4f98a2SYan Zheng struct inode *inode = NULL; 38325d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 38335d4f98a2SYan Zheng struct btrfs_root *root; 38344624900dSZhaolei u64 objectid; 38355d4f98a2SYan Zheng int err = 0; 38365d4f98a2SYan Zheng 3837aeb935a4SQu Wenruo root = btrfs_grab_root(fs_info->data_reloc_root); 3838a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 383976deacf0SJosef Bacik if (IS_ERR(trans)) { 384000246528SJosef Bacik btrfs_put_root(root); 38413fd0a558SYan, Zheng return ERR_CAST(trans); 384276deacf0SJosef Bacik } 38435d4f98a2SYan Zheng 3844543068a2SNikolay Borisov err = btrfs_get_free_objectid(root, &objectid); 38455d4f98a2SYan Zheng if (err) 38465d4f98a2SYan Zheng goto out; 38475d4f98a2SYan Zheng 38480257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 3849790c1b8cSJosef Bacik if (err) 3850790c1b8cSJosef Bacik goto out; 38515d4f98a2SYan Zheng 38520202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, objectid, root); 3853790c1b8cSJosef Bacik if (IS_ERR(inode)) { 3854790c1b8cSJosef Bacik delete_orphan_inode(trans, root, objectid); 3855790c1b8cSJosef Bacik err = PTR_ERR(inode); 3856790c1b8cSJosef Bacik inode = NULL; 3857790c1b8cSJosef Bacik goto out; 3858790c1b8cSJosef Bacik } 3859b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 38605d4f98a2SYan Zheng 386173f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 38625d4f98a2SYan Zheng out: 386300246528SJosef Bacik btrfs_put_root(root); 38643a45bb20SJeff Mahoney btrfs_end_transaction(trans); 38652ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 38665d4f98a2SYan Zheng if (err) { 38675d4f98a2SYan Zheng iput(inode); 38685d4f98a2SYan Zheng inode = ERR_PTR(err); 38695d4f98a2SYan Zheng } 38705d4f98a2SYan Zheng return inode; 38715d4f98a2SYan Zheng } 38725d4f98a2SYan Zheng 3873907d2710SDavid Sterba /* 3874907d2710SDavid Sterba * Mark start of chunk relocation that is cancellable. Check if the cancellation 3875907d2710SDavid Sterba * has been requested meanwhile and don't start in that case. 3876907d2710SDavid Sterba * 3877907d2710SDavid Sterba * Return: 3878907d2710SDavid Sterba * 0 success 3879907d2710SDavid Sterba * -EINPROGRESS operation is already in progress, that's probably a bug 3880907d2710SDavid Sterba * -ECANCELED cancellation request was set before the operation started 3881907d2710SDavid Sterba */ 3882907d2710SDavid Sterba static int reloc_chunk_start(struct btrfs_fs_info *fs_info) 3883907d2710SDavid Sterba { 3884907d2710SDavid Sterba if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) { 3885907d2710SDavid Sterba /* This should not happen */ 3886907d2710SDavid Sterba btrfs_err(fs_info, "reloc already running, cannot start"); 3887907d2710SDavid Sterba return -EINPROGRESS; 3888907d2710SDavid Sterba } 3889907d2710SDavid Sterba 3890907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) { 3891907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled on start"); 3892907d2710SDavid Sterba /* 3893907d2710SDavid Sterba * On cancel, clear all requests but let the caller mark 3894907d2710SDavid Sterba * the end after cleanup operations. 3895907d2710SDavid Sterba */ 3896907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3897907d2710SDavid Sterba return -ECANCELED; 3898907d2710SDavid Sterba } 3899907d2710SDavid Sterba return 0; 3900907d2710SDavid Sterba } 3901907d2710SDavid Sterba 3902907d2710SDavid Sterba /* 3903907d2710SDavid Sterba * Mark end of chunk relocation that is cancellable and wake any waiters. 3904907d2710SDavid Sterba */ 3905907d2710SDavid Sterba static void reloc_chunk_end(struct btrfs_fs_info *fs_info) 3906907d2710SDavid Sterba { 3907907d2710SDavid Sterba /* Requested after start, clear bit first so any waiters can continue */ 3908907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) 3909907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled during operation"); 3910907d2710SDavid Sterba clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags); 3911907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3912907d2710SDavid Sterba } 3913907d2710SDavid Sterba 3914c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 39153fd0a558SYan, Zheng { 39163fd0a558SYan, Zheng struct reloc_control *rc; 39173fd0a558SYan, Zheng 39183fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 39193fd0a558SYan, Zheng if (!rc) 39203fd0a558SYan, Zheng return NULL; 39213fd0a558SYan, Zheng 39223fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 3923d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 3924584fb121SQu Wenruo btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1); 39253fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 392643eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 392743eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 39283fd0a558SYan, Zheng return rc; 39293fd0a558SYan, Zheng } 39303fd0a558SYan, Zheng 39311a0afa0eSJosef Bacik static void free_reloc_control(struct reloc_control *rc) 39321a0afa0eSJosef Bacik { 39331a0afa0eSJosef Bacik struct mapping_node *node, *tmp; 39341a0afa0eSJosef Bacik 39351a0afa0eSJosef Bacik free_reloc_roots(&rc->reloc_roots); 39361a0afa0eSJosef Bacik rbtree_postorder_for_each_entry_safe(node, tmp, 39371a0afa0eSJosef Bacik &rc->reloc_root_tree.rb_root, rb_node) 39381a0afa0eSJosef Bacik kfree(node); 39391a0afa0eSJosef Bacik 39401a0afa0eSJosef Bacik kfree(rc); 39411a0afa0eSJosef Bacik } 39421a0afa0eSJosef Bacik 39435d4f98a2SYan Zheng /* 3944ebce0e01SAdam Borowski * Print the block group being relocated 3945ebce0e01SAdam Borowski */ 3946ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 394732da5386SDavid Sterba struct btrfs_block_group *block_group) 3948ebce0e01SAdam Borowski { 3949f89e09cfSAnand Jain char buf[128] = {'\0'}; 3950ebce0e01SAdam Borowski 3951f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 3952ebce0e01SAdam Borowski 3953ebce0e01SAdam Borowski btrfs_info(fs_info, 3954ebce0e01SAdam Borowski "relocating block group %llu flags %s", 3955b3470b5dSDavid Sterba block_group->start, buf); 3956ebce0e01SAdam Borowski } 3957ebce0e01SAdam Borowski 3958430640e3SQu Wenruo static const char *stage_to_string(int stage) 3959430640e3SQu Wenruo { 3960430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 3961430640e3SQu Wenruo return "move data extents"; 3962430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 3963430640e3SQu Wenruo return "update data pointers"; 3964430640e3SQu Wenruo return "unknown"; 3965430640e3SQu Wenruo } 3966430640e3SQu Wenruo 3967ebce0e01SAdam Borowski /* 39685d4f98a2SYan Zheng * function to relocate all extents in a block group. 39695d4f98a2SYan Zheng */ 39706bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 39715d4f98a2SYan Zheng { 397232da5386SDavid Sterba struct btrfs_block_group *bg; 397329cbcf40SJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start); 39745d4f98a2SYan Zheng struct reloc_control *rc; 39750af3d00bSJosef Bacik struct inode *inode; 39760af3d00bSJosef Bacik struct btrfs_path *path; 39775d4f98a2SYan Zheng int ret; 3978f0486c68SYan, Zheng int rw = 0; 39795d4f98a2SYan Zheng int err = 0; 39805d4f98a2SYan Zheng 3981b4be6aefSJosef Bacik /* 3982b4be6aefSJosef Bacik * This only gets set if we had a half-deleted snapshot on mount. We 3983b4be6aefSJosef Bacik * cannot allow relocation to start while we're still trying to clean up 3984b4be6aefSJosef Bacik * these pending deletions. 3985b4be6aefSJosef Bacik */ 3986b4be6aefSJosef Bacik ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE); 3987b4be6aefSJosef Bacik if (ret) 3988b4be6aefSJosef Bacik return ret; 3989b4be6aefSJosef Bacik 3990b4be6aefSJosef Bacik /* We may have been woken up by close_ctree, so bail if we're closing. */ 3991b4be6aefSJosef Bacik if (btrfs_fs_closing(fs_info)) 3992b4be6aefSJosef Bacik return -EINTR; 3993b4be6aefSJosef Bacik 3994eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 3995eede2bf3SOmar Sandoval if (!bg) 3996eede2bf3SOmar Sandoval return -ENOENT; 3997eede2bf3SOmar Sandoval 39980320b353SNaohiro Aota /* 39990320b353SNaohiro Aota * Relocation of a data block group creates ordered extents. Without 40000320b353SNaohiro Aota * sb_start_write(), we can freeze the filesystem while unfinished 40010320b353SNaohiro Aota * ordered extents are left. Such ordered extents can cause a deadlock 40020320b353SNaohiro Aota * e.g. when syncfs() is waiting for their completion but they can't 40030320b353SNaohiro Aota * finish because they block when joining a transaction, due to the 40040320b353SNaohiro Aota * fact that the freeze locks are being held in write mode. 40050320b353SNaohiro Aota */ 40060320b353SNaohiro Aota if (bg->flags & BTRFS_BLOCK_GROUP_DATA) 40070320b353SNaohiro Aota ASSERT(sb_write_started(fs_info->sb)); 40080320b353SNaohiro Aota 4009eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4010eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4011eede2bf3SOmar Sandoval return -ETXTBSY; 4012eede2bf3SOmar Sandoval } 4013eede2bf3SOmar Sandoval 4014c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4015eede2bf3SOmar Sandoval if (!rc) { 4016eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 40175d4f98a2SYan Zheng return -ENOMEM; 4018eede2bf3SOmar Sandoval } 40195d4f98a2SYan Zheng 4020907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4021907d2710SDavid Sterba if (ret < 0) { 4022907d2710SDavid Sterba err = ret; 4023907d2710SDavid Sterba goto out_put_bg; 4024907d2710SDavid Sterba } 4025907d2710SDavid Sterba 4026f0486c68SYan, Zheng rc->extent_root = extent_root; 4027eede2bf3SOmar Sandoval rc->block_group = bg; 40285d4f98a2SYan Zheng 4029b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4030f0486c68SYan, Zheng if (ret) { 4031f0486c68SYan, Zheng err = ret; 4032f0486c68SYan, Zheng goto out; 4033f0486c68SYan, Zheng } 4034f0486c68SYan, Zheng rw = 1; 4035f0486c68SYan, Zheng 40360af3d00bSJosef Bacik path = btrfs_alloc_path(); 40370af3d00bSJosef Bacik if (!path) { 40380af3d00bSJosef Bacik err = -ENOMEM; 40390af3d00bSJosef Bacik goto out; 40400af3d00bSJosef Bacik } 40410af3d00bSJosef Bacik 40427949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 40430af3d00bSJosef Bacik btrfs_free_path(path); 40440af3d00bSJosef Bacik 40450af3d00bSJosef Bacik if (!IS_ERR(inode)) 40461bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 40470af3d00bSJosef Bacik else 40480af3d00bSJosef Bacik ret = PTR_ERR(inode); 40490af3d00bSJosef Bacik 40500af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 40510af3d00bSJosef Bacik err = ret; 40520af3d00bSJosef Bacik goto out; 40530af3d00bSJosef Bacik } 40540af3d00bSJosef Bacik 40555d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 40565d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 40575d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 40585d4f98a2SYan Zheng rc->data_inode = NULL; 40595d4f98a2SYan Zheng goto out; 40605d4f98a2SYan Zheng } 40615d4f98a2SYan Zheng 40620b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 40635d4f98a2SYan Zheng 40649cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4065f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 40666374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4067b3470b5dSDavid Sterba rc->block_group->start, 4068b3470b5dSDavid Sterba rc->block_group->length); 40695d4f98a2SYan Zheng 40707ae9bd18SNaohiro Aota ret = btrfs_zone_finish(rc->block_group); 40717ae9bd18SNaohiro Aota WARN_ON(ret && ret != -EAGAIN); 40727ae9bd18SNaohiro Aota 40735d4f98a2SYan Zheng while (1) { 4074430640e3SQu Wenruo int finishes_stage; 4075430640e3SQu Wenruo 407676dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 40775d4f98a2SYan Zheng ret = relocate_block_group(rc); 407876dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4079ff612ba7SJosef Bacik if (ret < 0) 40805d4f98a2SYan Zheng err = ret; 4081ff612ba7SJosef Bacik 4082430640e3SQu Wenruo finishes_stage = rc->stage; 4083ff612ba7SJosef Bacik /* 4084ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4085ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4086ff612ba7SJosef Bacik * different block group we could end up hitting the 4087ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4088ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4089ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4090ff612ba7SJosef Bacik * out of the loop if we hit an error. 4091ff612ba7SJosef Bacik */ 4092ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4093ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4094ff612ba7SJosef Bacik (u64)-1); 4095ff612ba7SJosef Bacik if (ret) 4096ff612ba7SJosef Bacik err = ret; 4097ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4098ff612ba7SJosef Bacik 0, -1); 4099ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 41005d4f98a2SYan Zheng } 41015d4f98a2SYan Zheng 4102ff612ba7SJosef Bacik if (err < 0) 4103ff612ba7SJosef Bacik goto out; 4104ff612ba7SJosef Bacik 41055d4f98a2SYan Zheng if (rc->extents_found == 0) 41065d4f98a2SYan Zheng break; 41075d4f98a2SYan Zheng 4108430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4109430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 41105d4f98a2SYan Zheng } 41115d4f98a2SYan Zheng 41125d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 41135d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4114bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 41155d4f98a2SYan Zheng out: 4116f0486c68SYan, Zheng if (err && rw) 41172ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 41185d4f98a2SYan Zheng iput(rc->data_inode); 4119907d2710SDavid Sterba out_put_bg: 4120907d2710SDavid Sterba btrfs_put_block_group(bg); 4121907d2710SDavid Sterba reloc_chunk_end(fs_info); 41221a0afa0eSJosef Bacik free_reloc_control(rc); 41235d4f98a2SYan Zheng return err; 41245d4f98a2SYan Zheng } 41255d4f98a2SYan Zheng 412676dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 412776dda93cSYan, Zheng { 41280b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 412976dda93cSYan, Zheng struct btrfs_trans_handle *trans; 413079787eaaSJeff Mahoney int ret, err; 413176dda93cSYan, Zheng 41320b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 413379787eaaSJeff Mahoney if (IS_ERR(trans)) 413479787eaaSJeff Mahoney return PTR_ERR(trans); 413576dda93cSYan, Zheng 413676dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 413776dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 4138c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 413976dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 41400b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 414176dda93cSYan, Zheng &root->root_key, &root->root_item); 414276dda93cSYan, Zheng 41433a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 414479787eaaSJeff Mahoney if (err) 414579787eaaSJeff Mahoney return err; 414679787eaaSJeff Mahoney return ret; 414776dda93cSYan, Zheng } 414876dda93cSYan, Zheng 41495d4f98a2SYan Zheng /* 41505d4f98a2SYan Zheng * recover relocation interrupted by system crash. 41515d4f98a2SYan Zheng * 41525d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 41535d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 41545d4f98a2SYan Zheng */ 41557eefae6bSJosef Bacik int btrfs_recover_relocation(struct btrfs_fs_info *fs_info) 41565d4f98a2SYan Zheng { 41575d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 41585d4f98a2SYan Zheng struct btrfs_key key; 41595d4f98a2SYan Zheng struct btrfs_root *fs_root; 41605d4f98a2SYan Zheng struct btrfs_root *reloc_root; 41615d4f98a2SYan Zheng struct btrfs_path *path; 41625d4f98a2SYan Zheng struct extent_buffer *leaf; 41635d4f98a2SYan Zheng struct reloc_control *rc = NULL; 41645d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41655d4f98a2SYan Zheng int ret; 41665d4f98a2SYan Zheng int err = 0; 41675d4f98a2SYan Zheng 41685d4f98a2SYan Zheng path = btrfs_alloc_path(); 41695d4f98a2SYan Zheng if (!path) 41705d4f98a2SYan Zheng return -ENOMEM; 4171e4058b54SDavid Sterba path->reada = READA_BACK; 41725d4f98a2SYan Zheng 41735d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 41745d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 41755d4f98a2SYan Zheng key.offset = (u64)-1; 41765d4f98a2SYan Zheng 41775d4f98a2SYan Zheng while (1) { 41780b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 41795d4f98a2SYan Zheng path, 0, 0); 41805d4f98a2SYan Zheng if (ret < 0) { 41815d4f98a2SYan Zheng err = ret; 41825d4f98a2SYan Zheng goto out; 41835d4f98a2SYan Zheng } 41845d4f98a2SYan Zheng if (ret > 0) { 41855d4f98a2SYan Zheng if (path->slots[0] == 0) 41865d4f98a2SYan Zheng break; 41875d4f98a2SYan Zheng path->slots[0]--; 41885d4f98a2SYan Zheng } 41895d4f98a2SYan Zheng leaf = path->nodes[0]; 41905d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4191b3b4aa74SDavid Sterba btrfs_release_path(path); 41925d4f98a2SYan Zheng 41935d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 41945d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 41955d4f98a2SYan Zheng break; 41965d4f98a2SYan Zheng 41977eefae6bSJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key); 41985d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 41995d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 42005d4f98a2SYan Zheng goto out; 42015d4f98a2SYan Zheng } 42025d4f98a2SYan Zheng 420392a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 42045d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 42055d4f98a2SYan Zheng 42065d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 4207a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, 4208a820feb5SDavid Sterba reloc_root->root_key.offset, false); 42095d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 421076dda93cSYan, Zheng ret = PTR_ERR(fs_root); 421176dda93cSYan, Zheng if (ret != -ENOENT) { 421276dda93cSYan, Zheng err = ret; 42135d4f98a2SYan Zheng goto out; 42145d4f98a2SYan Zheng } 421579787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 421679787eaaSJeff Mahoney if (ret < 0) { 421779787eaaSJeff Mahoney err = ret; 421879787eaaSJeff Mahoney goto out; 421979787eaaSJeff Mahoney } 4220932fd26dSJosef Bacik } else { 422100246528SJosef Bacik btrfs_put_root(fs_root); 422276dda93cSYan, Zheng } 42235d4f98a2SYan Zheng } 42245d4f98a2SYan Zheng 42255d4f98a2SYan Zheng if (key.offset == 0) 42265d4f98a2SYan Zheng break; 42275d4f98a2SYan Zheng 42285d4f98a2SYan Zheng key.offset--; 42295d4f98a2SYan Zheng } 4230b3b4aa74SDavid Sterba btrfs_release_path(path); 42315d4f98a2SYan Zheng 42325d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 42335d4f98a2SYan Zheng goto out; 42345d4f98a2SYan Zheng 4235c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 42365d4f98a2SYan Zheng if (!rc) { 42375d4f98a2SYan Zheng err = -ENOMEM; 42385d4f98a2SYan Zheng goto out; 42395d4f98a2SYan Zheng } 42405d4f98a2SYan Zheng 4241907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4242907d2710SDavid Sterba if (ret < 0) { 4243907d2710SDavid Sterba err = ret; 4244907d2710SDavid Sterba goto out_end; 4245907d2710SDavid Sterba } 4246907d2710SDavid Sterba 424729cbcf40SJosef Bacik rc->extent_root = btrfs_extent_root(fs_info, 0); 42485d4f98a2SYan Zheng 42495d4f98a2SYan Zheng set_reloc_control(rc); 42505d4f98a2SYan Zheng 42517a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 42523612b495STsutomu Itoh if (IS_ERR(trans)) { 42533612b495STsutomu Itoh err = PTR_ERR(trans); 4254fb2d83eeSJosef Bacik goto out_unset; 42553612b495STsutomu Itoh } 42563fd0a558SYan, Zheng 42573fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 42583fd0a558SYan, Zheng 42595d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 42605d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 42615d4f98a2SYan Zheng struct btrfs_root, root_list); 42625d4f98a2SYan Zheng list_del(&reloc_root->root_list); 42635d4f98a2SYan Zheng 42645d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 42655d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 42665d4f98a2SYan Zheng &rc->reloc_roots); 42675d4f98a2SYan Zheng continue; 42685d4f98a2SYan Zheng } 42695d4f98a2SYan Zheng 4270a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 4271a820feb5SDavid Sterba false); 427279787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 427379787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4274ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42751402d17dSXiyu Yang btrfs_end_transaction(trans); 4276fb2d83eeSJosef Bacik goto out_unset; 427779787eaaSJeff Mahoney } 42785d4f98a2SYan Zheng 4279ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 428057a304cfSJosef Bacik ASSERT(err != -EEXIST); 42813c925863SJosef Bacik if (err) { 42823c925863SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42833c925863SJosef Bacik btrfs_put_root(fs_root); 42843c925863SJosef Bacik btrfs_end_transaction(trans); 42853c925863SJosef Bacik goto out_unset; 42863c925863SJosef Bacik } 4287f44deb74SJosef Bacik fs_root->reloc_root = btrfs_grab_root(reloc_root); 428800246528SJosef Bacik btrfs_put_root(fs_root); 42895d4f98a2SYan Zheng } 42905d4f98a2SYan Zheng 42913a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 429279787eaaSJeff Mahoney if (err) 4293fb2d83eeSJosef Bacik goto out_unset; 42945d4f98a2SYan Zheng 42955d4f98a2SYan Zheng merge_reloc_roots(rc); 42965d4f98a2SYan Zheng 42975d4f98a2SYan Zheng unset_reloc_control(rc); 42985d4f98a2SYan Zheng 42997a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 430062b99540SQu Wenruo if (IS_ERR(trans)) { 43013612b495STsutomu Itoh err = PTR_ERR(trans); 43026217b0faSJosef Bacik goto out_clean; 430362b99540SQu Wenruo } 43043a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 43056217b0faSJosef Bacik out_clean: 4306d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4307d2311e69SQu Wenruo if (ret < 0 && !err) 4308d2311e69SQu Wenruo err = ret; 4309fb2d83eeSJosef Bacik out_unset: 4310fb2d83eeSJosef Bacik unset_reloc_control(rc); 4311907d2710SDavid Sterba out_end: 4312907d2710SDavid Sterba reloc_chunk_end(fs_info); 43131a0afa0eSJosef Bacik free_reloc_control(rc); 43143612b495STsutomu Itoh out: 4315aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4316aca1bba6SLiu Bo 43175d4f98a2SYan Zheng btrfs_free_path(path); 43185d4f98a2SYan Zheng 43195d4f98a2SYan Zheng if (err == 0) { 43205d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 4321aeb935a4SQu Wenruo fs_root = btrfs_grab_root(fs_info->data_reloc_root); 4322aeb935a4SQu Wenruo ASSERT(fs_root); 432366b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 432400246528SJosef Bacik btrfs_put_root(fs_root); 4325932fd26dSJosef Bacik } 43265d4f98a2SYan Zheng return err; 43275d4f98a2SYan Zheng } 43285d4f98a2SYan Zheng 43295d4f98a2SYan Zheng /* 43305d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 43315d4f98a2SYan Zheng * 43325d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 43335d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 43345d4f98a2SYan Zheng */ 43357bfa9535SNikolay Borisov int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len) 43365d4f98a2SYan Zheng { 43377bfa9535SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 4338fc28b25eSJosef Bacik struct btrfs_root *csum_root; 43395d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 43405d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 43415d4f98a2SYan Zheng int ret; 43425d4f98a2SYan Zheng u64 disk_bytenr; 43434577b014SJosef Bacik u64 new_bytenr; 43445d4f98a2SYan Zheng LIST_HEAD(list); 43455d4f98a2SYan Zheng 43467bfa9535SNikolay Borisov ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4347bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 43485d4f98a2SYan Zheng 43497bfa9535SNikolay Borisov disk_bytenr = file_pos + inode->index_cnt; 4350fc28b25eSJosef Bacik csum_root = btrfs_csum_root(fs_info, disk_bytenr); 4351fc28b25eSJosef Bacik ret = btrfs_lookup_csums_range(csum_root, disk_bytenr, 435226ce9114SJosef Bacik disk_bytenr + len - 1, &list, 0, false); 435379787eaaSJeff Mahoney if (ret) 435479787eaaSJeff Mahoney goto out; 43555d4f98a2SYan Zheng 43565d4f98a2SYan Zheng while (!list_empty(&list)) { 43575d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 43585d4f98a2SYan Zheng list_del_init(&sums->list); 43595d4f98a2SYan Zheng 43604577b014SJosef Bacik /* 43614577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 43624577b014SJosef Bacik * We need to do this because we will read in entire prealloc 43634577b014SJosef Bacik * extents but we may have written to say the middle of the 43644577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 43654577b014SJosef Bacik * the right disk offset. 43664577b014SJosef Bacik * 43674577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 43684577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 43694577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 43704577b014SJosef Bacik * disk length. 43714577b014SJosef Bacik */ 4372bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 43734577b014SJosef Bacik sums->bytenr = new_bytenr; 43745d4f98a2SYan Zheng 4375f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 43765d4f98a2SYan Zheng } 437779787eaaSJeff Mahoney out: 43785d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4379411fc6bcSAndi Kleen return ret; 43805d4f98a2SYan Zheng } 43813fd0a558SYan, Zheng 438283d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 43833fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 43843fd0a558SYan, Zheng struct extent_buffer *cow) 43853fd0a558SYan, Zheng { 43860b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 43873fd0a558SYan, Zheng struct reloc_control *rc; 4388a26195a5SQu Wenruo struct btrfs_backref_node *node; 43893fd0a558SYan, Zheng int first_cow = 0; 43903fd0a558SYan, Zheng int level; 439183d4cfd4SJosef Bacik int ret = 0; 43923fd0a558SYan, Zheng 43930b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 43943fd0a558SYan, Zheng if (!rc) 439583d4cfd4SJosef Bacik return 0; 43963fd0a558SYan, Zheng 439737f00a6dSJohannes Thumshirn BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root)); 43983fd0a558SYan, Zheng 43993fd0a558SYan, Zheng level = btrfs_header_level(buf); 44003fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 44013fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 44023fd0a558SYan, Zheng first_cow = 1; 44033fd0a558SYan, Zheng 44043fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 44053fd0a558SYan, Zheng rc->create_reloc_tree) { 44063fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 44073fd0a558SYan, Zheng 44083fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 44093fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 44103fd0a558SYan, Zheng node->new_bytenr != buf->start); 44113fd0a558SYan, Zheng 4412b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 441367439dadSDavid Sterba atomic_inc(&cow->refs); 44143fd0a558SYan, Zheng node->eb = cow; 44153fd0a558SYan, Zheng node->new_bytenr = cow->start; 44163fd0a558SYan, Zheng 44173fd0a558SYan, Zheng if (!node->pending) { 44183fd0a558SYan, Zheng list_move_tail(&node->list, 44193fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 44203fd0a558SYan, Zheng node->pending = 1; 44213fd0a558SYan, Zheng } 44223fd0a558SYan, Zheng 44233fd0a558SYan, Zheng if (first_cow) 44249569cc20SQu Wenruo mark_block_processed(rc, node); 44253fd0a558SYan, Zheng 44263fd0a558SYan, Zheng if (first_cow && level > 0) 44273fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 44283fd0a558SYan, Zheng } 44293fd0a558SYan, Zheng 443083d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 44313fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 443283d4cfd4SJosef Bacik return ret; 44333fd0a558SYan, Zheng } 44343fd0a558SYan, Zheng 44353fd0a558SYan, Zheng /* 44363fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 443701327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 44383fd0a558SYan, Zheng */ 4439147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 44403fd0a558SYan, Zheng u64 *bytes_to_reserve) 44413fd0a558SYan, Zheng { 444210995c04SQu Wenruo struct btrfs_root *root = pending->root; 444310995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44443fd0a558SYan, Zheng 44456282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 44463fd0a558SYan, Zheng return; 44473fd0a558SYan, Zheng 44483fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 44493fd0a558SYan, Zheng return; 44503fd0a558SYan, Zheng 44513fd0a558SYan, Zheng root = root->reloc_root; 44523fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 44533fd0a558SYan, Zheng /* 44543fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 44553fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 44563fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 44573fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 44583fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 44593fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 44603fd0a558SYan, Zheng * tree may use more space than it frees. so we need 44613fd0a558SYan, Zheng * reserve extra space. 44623fd0a558SYan, Zheng */ 44633fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 44643fd0a558SYan, Zheng } 44653fd0a558SYan, Zheng 44663fd0a558SYan, Zheng /* 44673fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 44683fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 4469f44deb74SJosef Bacik * 4470f44deb74SJosef Bacik * This is similar to btrfs_init_reloc_root(), we come out of here with two 4471f44deb74SJosef Bacik * references held on the reloc_root, one for root->reloc_root and one for 4472f44deb74SJosef Bacik * rc->reloc_roots. 44733fd0a558SYan, Zheng */ 447449b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 44753fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 44763fd0a558SYan, Zheng { 44773fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 44783fd0a558SYan, Zheng struct btrfs_root *reloc_root; 44793fd0a558SYan, Zheng struct btrfs_root *new_root; 448010995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44813fd0a558SYan, Zheng int ret; 44823fd0a558SYan, Zheng 44836282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 448449b25e05SJeff Mahoney return 0; 44853fd0a558SYan, Zheng 44863fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 44873fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 44883fd0a558SYan, Zheng 44893fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 44903fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 44913fd0a558SYan, Zheng rc->block_rsv, 44923a584174SLu Fengqi rc->nodes_relocated, true); 449349b25e05SJeff Mahoney if (ret) 449449b25e05SJeff Mahoney return ret; 44953fd0a558SYan, Zheng } 44963fd0a558SYan, Zheng 44973fd0a558SYan, Zheng new_root = pending->snap; 44983fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 44993fd0a558SYan, Zheng new_root->root_key.objectid); 450049b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 450149b25e05SJeff Mahoney return PTR_ERR(reloc_root); 45023fd0a558SYan, Zheng 4503ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 450457a304cfSJosef Bacik ASSERT(ret != -EEXIST); 45053c925863SJosef Bacik if (ret) { 45063c925863SJosef Bacik /* Pairs with create_reloc_root */ 45073c925863SJosef Bacik btrfs_put_root(reloc_root); 45083c925863SJosef Bacik return ret; 45093c925863SJosef Bacik } 4510f44deb74SJosef Bacik new_root->reloc_root = btrfs_grab_root(reloc_root); 45113fd0a558SYan, Zheng 451249b25e05SJeff Mahoney if (rc->create_reloc_tree) 45133fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 451449b25e05SJeff Mahoney return ret; 45153fd0a558SYan, Zheng } 4516