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" 31*c7f13d42SJosef Bacik #include "fs.h" 325d4f98a2SYan Zheng 335d4f98a2SYan Zheng /* 340c891389SQu Wenruo * Relocation overview 350c891389SQu Wenruo * 360c891389SQu Wenruo * [What does relocation do] 370c891389SQu Wenruo * 380c891389SQu Wenruo * The objective of relocation is to relocate all extents of the target block 390c891389SQu Wenruo * group to other block groups. 400c891389SQu Wenruo * This is utilized by resize (shrink only), profile converting, compacting 410c891389SQu Wenruo * space, or balance routine to spread chunks over devices. 420c891389SQu Wenruo * 430c891389SQu Wenruo * Before | After 440c891389SQu Wenruo * ------------------------------------------------------------------ 450c891389SQu Wenruo * BG A: 10 data extents | BG A: deleted 460c891389SQu Wenruo * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) 470c891389SQu Wenruo * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) 480c891389SQu Wenruo * 490c891389SQu Wenruo * [How does relocation work] 500c891389SQu Wenruo * 510c891389SQu Wenruo * 1. Mark the target block group read-only 520c891389SQu Wenruo * New extents won't be allocated from the target block group. 530c891389SQu Wenruo * 540c891389SQu Wenruo * 2.1 Record each extent in the target block group 550c891389SQu Wenruo * To build a proper map of extents to be relocated. 560c891389SQu Wenruo * 570c891389SQu Wenruo * 2.2 Build data reloc tree and reloc trees 580c891389SQu Wenruo * Data reloc tree will contain an inode, recording all newly relocated 590c891389SQu Wenruo * data extents. 600c891389SQu Wenruo * There will be only one data reloc tree for one data block group. 610c891389SQu Wenruo * 620c891389SQu Wenruo * Reloc tree will be a special snapshot of its source tree, containing 630c891389SQu Wenruo * relocated tree blocks. 640c891389SQu Wenruo * Each tree referring to a tree block in target block group will get its 650c891389SQu Wenruo * reloc tree built. 660c891389SQu Wenruo * 670c891389SQu Wenruo * 2.3 Swap source tree with its corresponding reloc tree 680c891389SQu Wenruo * Each involved tree only refers to new extents after swap. 690c891389SQu Wenruo * 700c891389SQu Wenruo * 3. Cleanup reloc trees and data reloc tree. 710c891389SQu Wenruo * As old extents in the target block group are still referenced by reloc 720c891389SQu Wenruo * trees, we need to clean them up before really freeing the target block 730c891389SQu Wenruo * group. 740c891389SQu Wenruo * 750c891389SQu Wenruo * The main complexity is in steps 2.2 and 2.3. 760c891389SQu Wenruo * 770c891389SQu Wenruo * The entry point of relocation is relocate_block_group() function. 780c891389SQu Wenruo */ 790c891389SQu Wenruo 800647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 812a979612SQu Wenruo /* 825d4f98a2SYan Zheng * map address of tree root to tree 835d4f98a2SYan Zheng */ 845d4f98a2SYan Zheng struct mapping_node { 85e9a28dc5SQu Wenruo struct { 865d4f98a2SYan Zheng struct rb_node rb_node; 875d4f98a2SYan Zheng u64 bytenr; 88e9a28dc5SQu Wenruo }; /* Use rb_simle_node for search/insert */ 895d4f98a2SYan Zheng void *data; 905d4f98a2SYan Zheng }; 915d4f98a2SYan Zheng 925d4f98a2SYan Zheng struct mapping_tree { 935d4f98a2SYan Zheng struct rb_root rb_root; 945d4f98a2SYan Zheng spinlock_t lock; 955d4f98a2SYan Zheng }; 965d4f98a2SYan Zheng 975d4f98a2SYan Zheng /* 985d4f98a2SYan Zheng * present a tree block to process 995d4f98a2SYan Zheng */ 1005d4f98a2SYan Zheng struct tree_block { 101e9a28dc5SQu Wenruo struct { 1025d4f98a2SYan Zheng struct rb_node rb_node; 1035d4f98a2SYan Zheng u64 bytenr; 104e9a28dc5SQu Wenruo }; /* Use rb_simple_node for search/insert */ 105f7ba2d37SJosef Bacik u64 owner; 1065d4f98a2SYan Zheng struct btrfs_key key; 1075d4f98a2SYan Zheng unsigned int level:8; 1085d4f98a2SYan Zheng unsigned int key_ready:1; 1095d4f98a2SYan Zheng }; 1105d4f98a2SYan Zheng 1110257bb82SYan, Zheng #define MAX_EXTENTS 128 1120257bb82SYan, Zheng 1130257bb82SYan, Zheng struct file_extent_cluster { 1140257bb82SYan, Zheng u64 start; 1150257bb82SYan, Zheng u64 end; 1160257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1170257bb82SYan, Zheng unsigned int nr; 1180257bb82SYan, Zheng }; 1190257bb82SYan, Zheng 1205d4f98a2SYan Zheng struct reloc_control { 1215d4f98a2SYan Zheng /* block group to relocate */ 12232da5386SDavid Sterba struct btrfs_block_group *block_group; 1235d4f98a2SYan Zheng /* extent tree */ 1245d4f98a2SYan Zheng struct btrfs_root *extent_root; 1255d4f98a2SYan Zheng /* inode for moving data */ 1265d4f98a2SYan Zheng struct inode *data_inode; 1273fd0a558SYan, Zheng 1283fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1293fd0a558SYan, Zheng 130a26195a5SQu Wenruo struct btrfs_backref_cache backref_cache; 1313fd0a558SYan, Zheng 1323fd0a558SYan, Zheng struct file_extent_cluster cluster; 1335d4f98a2SYan Zheng /* tree blocks have been processed */ 1345d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1355d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1365d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1375d4f98a2SYan Zheng /* list of reloc trees */ 1385d4f98a2SYan Zheng struct list_head reloc_roots; 139d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 140d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 1413fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1423fd0a558SYan, Zheng u64 merging_rsv_size; 1433fd0a558SYan, Zheng /* size of relocated tree nodes */ 1443fd0a558SYan, Zheng u64 nodes_relocated; 1450647bf56SWang Shilong /* reserved size for block group relocation*/ 1460647bf56SWang Shilong u64 reserved_bytes; 1473fd0a558SYan, Zheng 1485d4f98a2SYan Zheng u64 search_start; 1495d4f98a2SYan Zheng u64 extents_found; 1503fd0a558SYan, Zheng 1513fd0a558SYan, Zheng unsigned int stage:8; 1523fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1533fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1545d4f98a2SYan Zheng unsigned int found_file_extent:1; 1555d4f98a2SYan Zheng }; 1565d4f98a2SYan Zheng 1575d4f98a2SYan Zheng /* stages of data relocation */ 1585d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1595d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1605d4f98a2SYan Zheng 1619569cc20SQu Wenruo static void mark_block_processed(struct reloc_control *rc, 162a26195a5SQu Wenruo struct btrfs_backref_node *node) 1639569cc20SQu Wenruo { 1649569cc20SQu Wenruo u32 blocksize; 1659569cc20SQu Wenruo 1669569cc20SQu Wenruo if (node->level == 0 || 1679569cc20SQu Wenruo in_range(node->bytenr, rc->block_group->start, 1689569cc20SQu Wenruo rc->block_group->length)) { 1699569cc20SQu Wenruo blocksize = rc->extent_root->fs_info->nodesize; 1709569cc20SQu Wenruo set_extent_bits(&rc->processed_blocks, node->bytenr, 1719569cc20SQu Wenruo node->bytenr + blocksize - 1, EXTENT_DIRTY); 1729569cc20SQu Wenruo } 1739569cc20SQu Wenruo node->processed = 1; 1749569cc20SQu Wenruo } 1759569cc20SQu Wenruo 1765d4f98a2SYan Zheng 1775d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1785d4f98a2SYan Zheng { 1796bef4d31SEric Paris tree->rb_root = RB_ROOT; 1805d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1815d4f98a2SYan Zheng } 1825d4f98a2SYan Zheng 1835d4f98a2SYan Zheng /* 1845d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 1855d4f98a2SYan Zheng */ 186a26195a5SQu Wenruo static struct btrfs_backref_node *walk_up_backref( 187a26195a5SQu Wenruo struct btrfs_backref_node *node, 188a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 1895d4f98a2SYan Zheng { 190a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 1915d4f98a2SYan Zheng int idx = *index; 1925d4f98a2SYan Zheng 1935d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 1945d4f98a2SYan Zheng edge = list_entry(node->upper.next, 195a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 1965d4f98a2SYan Zheng edges[idx++] = edge; 1975d4f98a2SYan Zheng node = edge->node[UPPER]; 1985d4f98a2SYan Zheng } 1993fd0a558SYan, Zheng BUG_ON(node->detached); 2005d4f98a2SYan Zheng *index = idx; 2015d4f98a2SYan Zheng return node; 2025d4f98a2SYan Zheng } 2035d4f98a2SYan Zheng 2045d4f98a2SYan Zheng /* 2055d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 2065d4f98a2SYan Zheng */ 207a26195a5SQu Wenruo static struct btrfs_backref_node *walk_down_backref( 208a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 2095d4f98a2SYan Zheng { 210a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 211a26195a5SQu Wenruo struct btrfs_backref_node *lower; 2125d4f98a2SYan Zheng int idx = *index; 2135d4f98a2SYan Zheng 2145d4f98a2SYan Zheng while (idx > 0) { 2155d4f98a2SYan Zheng edge = edges[idx - 1]; 2165d4f98a2SYan Zheng lower = edge->node[LOWER]; 2175d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 2185d4f98a2SYan Zheng idx--; 2195d4f98a2SYan Zheng continue; 2205d4f98a2SYan Zheng } 2215d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 222a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 2235d4f98a2SYan Zheng edges[idx - 1] = edge; 2245d4f98a2SYan Zheng *index = idx; 2255d4f98a2SYan Zheng return edge->node[UPPER]; 2265d4f98a2SYan Zheng } 2275d4f98a2SYan Zheng *index = 0; 2285d4f98a2SYan Zheng return NULL; 2295d4f98a2SYan Zheng } 2305d4f98a2SYan Zheng 231a26195a5SQu Wenruo static void update_backref_node(struct btrfs_backref_cache *cache, 232a26195a5SQu Wenruo struct btrfs_backref_node *node, u64 bytenr) 2333fd0a558SYan, Zheng { 2343fd0a558SYan, Zheng struct rb_node *rb_node; 2353fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 2363fd0a558SYan, Zheng node->bytenr = bytenr; 237e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node); 23843c04fb1SJeff Mahoney if (rb_node) 239982c92cbSQu Wenruo btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST); 2403fd0a558SYan, Zheng } 2413fd0a558SYan, Zheng 2423fd0a558SYan, Zheng /* 2433fd0a558SYan, Zheng * update backref cache after a transaction commit 2443fd0a558SYan, Zheng */ 2453fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 246a26195a5SQu Wenruo struct btrfs_backref_cache *cache) 2473fd0a558SYan, Zheng { 248a26195a5SQu Wenruo struct btrfs_backref_node *node; 2493fd0a558SYan, Zheng int level = 0; 2503fd0a558SYan, Zheng 2513fd0a558SYan, Zheng if (cache->last_trans == 0) { 2523fd0a558SYan, Zheng cache->last_trans = trans->transid; 2533fd0a558SYan, Zheng return 0; 2543fd0a558SYan, Zheng } 2553fd0a558SYan, Zheng 2563fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 2573fd0a558SYan, Zheng return 0; 2583fd0a558SYan, Zheng 2593fd0a558SYan, Zheng /* 2603fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 2613fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 2623fd0a558SYan, Zheng * so the detached nodes are no longer useful. 2633fd0a558SYan, Zheng */ 2643fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2653fd0a558SYan, Zheng node = list_entry(cache->detached.next, 266a26195a5SQu Wenruo struct btrfs_backref_node, list); 267023acb07SQu Wenruo btrfs_backref_cleanup_node(cache, node); 2683fd0a558SYan, Zheng } 2693fd0a558SYan, Zheng 2703fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 2713fd0a558SYan, Zheng node = list_entry(cache->changed.next, 272a26195a5SQu Wenruo struct btrfs_backref_node, list); 2733fd0a558SYan, Zheng list_del_init(&node->list); 2743fd0a558SYan, Zheng BUG_ON(node->pending); 2753fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2763fd0a558SYan, Zheng } 2773fd0a558SYan, Zheng 2783fd0a558SYan, Zheng /* 2793fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 2803fd0a558SYan, Zheng * errors during processing the pending nodes. 2813fd0a558SYan, Zheng */ 2823fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 2833fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 2843fd0a558SYan, Zheng BUG_ON(!node->pending); 2853fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 2863fd0a558SYan, Zheng continue; 2873fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2883fd0a558SYan, Zheng } 2893fd0a558SYan, Zheng } 2903fd0a558SYan, Zheng 2913fd0a558SYan, Zheng cache->last_trans = 0; 2923fd0a558SYan, Zheng return 1; 2933fd0a558SYan, Zheng } 2943fd0a558SYan, Zheng 2956282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 2966282675eSQu Wenruo { 2976282675eSQu Wenruo /* 2986282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 2996282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 3006282675eSQu Wenruo * trying to access reloc_root 3016282675eSQu Wenruo */ 3026282675eSQu Wenruo smp_rmb(); 3036282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 3046282675eSQu Wenruo return true; 3056282675eSQu Wenruo return false; 3066282675eSQu Wenruo } 3076282675eSQu Wenruo 3086282675eSQu Wenruo /* 3096282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 3106282675eSQu Wenruo * 3116282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 3126282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 31355465730SQu Wenruo * from no reloc root. But btrfs_should_ignore_reloc_root() below is a 31455465730SQu Wenruo * special case. 3156282675eSQu Wenruo */ 3166282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 3176282675eSQu Wenruo { 3186282675eSQu Wenruo if (reloc_root_is_dead(root)) 3196282675eSQu Wenruo return false; 3206282675eSQu Wenruo if (!root->reloc_root) 3216282675eSQu Wenruo return false; 3226282675eSQu Wenruo return true; 3236282675eSQu Wenruo } 324f2a97a9dSDavid Sterba 32555465730SQu Wenruo int btrfs_should_ignore_reloc_root(struct btrfs_root *root) 3263fd0a558SYan, Zheng { 3273fd0a558SYan, Zheng struct btrfs_root *reloc_root; 3283fd0a558SYan, Zheng 32992a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 3303fd0a558SYan, Zheng return 0; 3313fd0a558SYan, Zheng 3326282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 3336282675eSQu Wenruo if (reloc_root_is_dead(root)) 3346282675eSQu Wenruo return 1; 3356282675eSQu Wenruo 3363fd0a558SYan, Zheng reloc_root = root->reloc_root; 3373fd0a558SYan, Zheng if (!reloc_root) 3383fd0a558SYan, Zheng return 0; 3393fd0a558SYan, Zheng 3404d4225fcSJosef Bacik if (btrfs_header_generation(reloc_root->commit_root) == 3414d4225fcSJosef Bacik root->fs_info->running_transaction->transid) 3423fd0a558SYan, Zheng return 0; 3433fd0a558SYan, Zheng /* 3443fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 3453fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 3463fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 3473fd0a558SYan, Zheng * relocation. 3483fd0a558SYan, Zheng */ 3493fd0a558SYan, Zheng return 1; 3503fd0a558SYan, Zheng } 35155465730SQu Wenruo 3525d4f98a2SYan Zheng /* 3535d4f98a2SYan Zheng * find reloc tree by address of tree root 3545d4f98a2SYan Zheng */ 3552433bea5SQu Wenruo struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) 3565d4f98a2SYan Zheng { 3572433bea5SQu Wenruo struct reloc_control *rc = fs_info->reloc_ctl; 3585d4f98a2SYan Zheng struct rb_node *rb_node; 3595d4f98a2SYan Zheng struct mapping_node *node; 3605d4f98a2SYan Zheng struct btrfs_root *root = NULL; 3615d4f98a2SYan Zheng 3622433bea5SQu Wenruo ASSERT(rc); 3635d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 364e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr); 3655d4f98a2SYan Zheng if (rb_node) { 3665d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 3670d031dc4SYu Zhe root = node->data; 3685d4f98a2SYan Zheng } 3695d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 37000246528SJosef Bacik return btrfs_grab_root(root); 3715d4f98a2SYan Zheng } 3725d4f98a2SYan Zheng 3735d4f98a2SYan Zheng /* 37429db137bSQu Wenruo * For useless nodes, do two major clean ups: 37529db137bSQu Wenruo * 37629db137bSQu Wenruo * - Cleanup the children edges and nodes 37729db137bSQu Wenruo * If child node is also orphan (no parent) during cleanup, then the child 37829db137bSQu Wenruo * node will also be cleaned up. 37929db137bSQu Wenruo * 38029db137bSQu Wenruo * - Freeing up leaves (level 0), keeps nodes detached 38129db137bSQu Wenruo * For nodes, the node is still cached as "detached" 38229db137bSQu Wenruo * 38329db137bSQu Wenruo * Return false if @node is not in the @useless_nodes list. 38429db137bSQu Wenruo * Return true if @node is in the @useless_nodes list. 38529db137bSQu Wenruo */ 38629db137bSQu Wenruo static bool handle_useless_nodes(struct reloc_control *rc, 387a26195a5SQu Wenruo struct btrfs_backref_node *node) 38829db137bSQu Wenruo { 389a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 39029db137bSQu Wenruo struct list_head *useless_node = &cache->useless_node; 39129db137bSQu Wenruo bool ret = false; 39229db137bSQu Wenruo 39329db137bSQu Wenruo while (!list_empty(useless_node)) { 394a26195a5SQu Wenruo struct btrfs_backref_node *cur; 39529db137bSQu Wenruo 396a26195a5SQu Wenruo cur = list_first_entry(useless_node, struct btrfs_backref_node, 39729db137bSQu Wenruo list); 39829db137bSQu Wenruo list_del_init(&cur->list); 39929db137bSQu Wenruo 40029db137bSQu Wenruo /* Only tree root nodes can be added to @useless_nodes */ 40129db137bSQu Wenruo ASSERT(list_empty(&cur->upper)); 40229db137bSQu Wenruo 40329db137bSQu Wenruo if (cur == node) 40429db137bSQu Wenruo ret = true; 40529db137bSQu Wenruo 40629db137bSQu Wenruo /* The node is the lowest node */ 40729db137bSQu Wenruo if (cur->lowest) { 40829db137bSQu Wenruo list_del_init(&cur->lower); 40929db137bSQu Wenruo cur->lowest = 0; 41029db137bSQu Wenruo } 41129db137bSQu Wenruo 41229db137bSQu Wenruo /* Cleanup the lower edges */ 41329db137bSQu Wenruo while (!list_empty(&cur->lower)) { 414a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 415a26195a5SQu Wenruo struct btrfs_backref_node *lower; 41629db137bSQu Wenruo 41729db137bSQu Wenruo edge = list_entry(cur->lower.next, 418a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 41929db137bSQu Wenruo list_del(&edge->list[UPPER]); 42029db137bSQu Wenruo list_del(&edge->list[LOWER]); 42129db137bSQu Wenruo lower = edge->node[LOWER]; 422741188d3SQu Wenruo btrfs_backref_free_edge(cache, edge); 42329db137bSQu Wenruo 42429db137bSQu Wenruo /* Child node is also orphan, queue for cleanup */ 42529db137bSQu Wenruo if (list_empty(&lower->upper)) 42629db137bSQu Wenruo list_add(&lower->list, useless_node); 42729db137bSQu Wenruo } 42829db137bSQu Wenruo /* Mark this block processed for relocation */ 42929db137bSQu Wenruo mark_block_processed(rc, cur); 43029db137bSQu Wenruo 43129db137bSQu Wenruo /* 43229db137bSQu Wenruo * Backref nodes for tree leaves are deleted from the cache. 43329db137bSQu Wenruo * Backref nodes for upper level tree blocks are left in the 43429db137bSQu Wenruo * cache to avoid unnecessary backref lookup. 43529db137bSQu Wenruo */ 43629db137bSQu Wenruo if (cur->level > 0) { 43729db137bSQu Wenruo list_add(&cur->list, &cache->detached); 43829db137bSQu Wenruo cur->detached = 1; 43929db137bSQu Wenruo } else { 44029db137bSQu Wenruo rb_erase(&cur->rb_node, &cache->rb_root); 441741188d3SQu Wenruo btrfs_backref_free_node(cache, cur); 44229db137bSQu Wenruo } 44329db137bSQu Wenruo } 44429db137bSQu Wenruo return ret; 44529db137bSQu Wenruo } 44629db137bSQu Wenruo 44729db137bSQu Wenruo /* 448e7d571c7SQu Wenruo * Build backref tree for a given tree block. Root of the backref tree 449e7d571c7SQu Wenruo * corresponds the tree block, leaves of the backref tree correspond roots of 450e7d571c7SQu Wenruo * b-trees that reference the tree block. 451e7d571c7SQu Wenruo * 452e7d571c7SQu Wenruo * The basic idea of this function is check backrefs of a given block to find 453e7d571c7SQu Wenruo * upper level blocks that reference the block, and then check backrefs of 454e7d571c7SQu Wenruo * these upper level blocks recursively. The recursion stops when tree root is 455e7d571c7SQu Wenruo * reached or backrefs for the block is cached. 456e7d571c7SQu Wenruo * 457e7d571c7SQu Wenruo * NOTE: if we find that backrefs for a block are cached, we know backrefs for 458e7d571c7SQu Wenruo * all upper level blocks that directly/indirectly reference the block are also 459e7d571c7SQu Wenruo * cached. 460e7d571c7SQu Wenruo */ 461a26195a5SQu Wenruo static noinline_for_stack struct btrfs_backref_node *build_backref_tree( 462e7d571c7SQu Wenruo struct reloc_control *rc, struct btrfs_key *node_key, 463e7d571c7SQu Wenruo int level, u64 bytenr) 464e7d571c7SQu Wenruo { 465e7d571c7SQu Wenruo struct btrfs_backref_iter *iter; 466a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 467e7d571c7SQu Wenruo /* For searching parent of TREE_BLOCK_REF */ 468e7d571c7SQu Wenruo struct btrfs_path *path; 469a26195a5SQu Wenruo struct btrfs_backref_node *cur; 470a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 471a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 472e7d571c7SQu Wenruo int ret; 473e7d571c7SQu Wenruo int err = 0; 474e7d571c7SQu Wenruo 475e7d571c7SQu Wenruo iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS); 476e7d571c7SQu Wenruo if (!iter) 477e7d571c7SQu Wenruo return ERR_PTR(-ENOMEM); 478e7d571c7SQu Wenruo path = btrfs_alloc_path(); 479e7d571c7SQu Wenruo if (!path) { 480e7d571c7SQu Wenruo err = -ENOMEM; 481e7d571c7SQu Wenruo goto out; 482e7d571c7SQu Wenruo } 483e7d571c7SQu Wenruo 484b1818dabSQu Wenruo node = btrfs_backref_alloc_node(cache, bytenr, level); 485e7d571c7SQu Wenruo if (!node) { 486e7d571c7SQu Wenruo err = -ENOMEM; 487e7d571c7SQu Wenruo goto out; 488e7d571c7SQu Wenruo } 489e7d571c7SQu Wenruo 490e7d571c7SQu Wenruo node->lowest = 1; 491e7d571c7SQu Wenruo cur = node; 492e7d571c7SQu Wenruo 493e7d571c7SQu Wenruo /* Breadth-first search to build backref cache */ 494e7d571c7SQu Wenruo do { 4951b60d2ecSQu Wenruo ret = btrfs_backref_add_tree_node(cache, path, iter, node_key, 4961b60d2ecSQu Wenruo cur); 497e7d571c7SQu Wenruo if (ret < 0) { 498e7d571c7SQu Wenruo err = ret; 499e7d571c7SQu Wenruo goto out; 500e7d571c7SQu Wenruo } 501e7d571c7SQu Wenruo edge = list_first_entry_or_null(&cache->pending_edge, 502a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 503e7d571c7SQu Wenruo /* 504e7d571c7SQu Wenruo * The pending list isn't empty, take the first block to 505e7d571c7SQu Wenruo * process 506e7d571c7SQu Wenruo */ 507e7d571c7SQu Wenruo if (edge) { 5085d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 5095d4f98a2SYan Zheng cur = edge->node[UPPER]; 5105d4f98a2SYan Zheng } 511e7d571c7SQu Wenruo } while (edge); 5125d4f98a2SYan Zheng 5131f872924SQu Wenruo /* Finish the upper linkage of newly added edges/nodes */ 514fc997ed0SQu Wenruo ret = btrfs_backref_finish_upper_links(cache, node); 5151f872924SQu Wenruo if (ret < 0) { 5161f872924SQu Wenruo err = ret; 51775bfb9afSJosef Bacik goto out; 51875bfb9afSJosef Bacik } 51975bfb9afSJosef Bacik 52029db137bSQu Wenruo if (handle_useless_nodes(rc, node)) 5213fd0a558SYan, Zheng node = NULL; 5225d4f98a2SYan Zheng out: 52371f572a9SQu Wenruo btrfs_backref_iter_free(iter); 52471f572a9SQu Wenruo btrfs_free_path(path); 5255d4f98a2SYan Zheng if (err) { 5261b23ea18SQu Wenruo btrfs_backref_error_cleanup(cache, node); 5275d4f98a2SYan Zheng return ERR_PTR(err); 5285d4f98a2SYan Zheng } 52975bfb9afSJosef Bacik ASSERT(!node || !node->detached); 53084780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 53184780289SQu Wenruo list_empty(&cache->pending_edge)); 5325d4f98a2SYan Zheng return node; 5335d4f98a2SYan Zheng } 5345d4f98a2SYan Zheng 5355d4f98a2SYan Zheng /* 5363fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 5373fd0a558SYan, Zheng * the backref node is created by cloning backref node that 5383fd0a558SYan, Zheng * corresponds to root of source tree 5393fd0a558SYan, Zheng */ 5403fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 5413fd0a558SYan, Zheng struct reloc_control *rc, 5423fd0a558SYan, Zheng struct btrfs_root *src, 5433fd0a558SYan, Zheng struct btrfs_root *dest) 5443fd0a558SYan, Zheng { 5453fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 546a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 547a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 548a26195a5SQu Wenruo struct btrfs_backref_node *new_node; 549a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 550a26195a5SQu Wenruo struct btrfs_backref_edge *new_edge; 5513fd0a558SYan, Zheng struct rb_node *rb_node; 5523fd0a558SYan, Zheng 5533fd0a558SYan, Zheng if (cache->last_trans > 0) 5543fd0a558SYan, Zheng update_backref_cache(trans, cache); 5553fd0a558SYan, Zheng 556e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start); 5573fd0a558SYan, Zheng if (rb_node) { 558a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, rb_node); 5593fd0a558SYan, Zheng if (node->detached) 5603fd0a558SYan, Zheng node = NULL; 5613fd0a558SYan, Zheng else 5623fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 5633fd0a558SYan, Zheng } 5643fd0a558SYan, Zheng 5653fd0a558SYan, Zheng if (!node) { 566e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, 5673fd0a558SYan, Zheng reloc_root->commit_root->start); 5683fd0a558SYan, Zheng if (rb_node) { 569a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, 5703fd0a558SYan, Zheng rb_node); 5713fd0a558SYan, Zheng BUG_ON(node->detached); 5723fd0a558SYan, Zheng } 5733fd0a558SYan, Zheng } 5743fd0a558SYan, Zheng 5753fd0a558SYan, Zheng if (!node) 5763fd0a558SYan, Zheng return 0; 5773fd0a558SYan, Zheng 578b1818dabSQu Wenruo new_node = btrfs_backref_alloc_node(cache, dest->node->start, 579b1818dabSQu Wenruo node->level); 5803fd0a558SYan, Zheng if (!new_node) 5813fd0a558SYan, Zheng return -ENOMEM; 5823fd0a558SYan, Zheng 5833fd0a558SYan, Zheng new_node->lowest = node->lowest; 5846848ad64SYan, Zheng new_node->checked = 1; 58500246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 5860b530bc5SJosef Bacik ASSERT(new_node->root); 5873fd0a558SYan, Zheng 5883fd0a558SYan, Zheng if (!node->lowest) { 5893fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 59047254d07SQu Wenruo new_edge = btrfs_backref_alloc_edge(cache); 5913fd0a558SYan, Zheng if (!new_edge) 5923fd0a558SYan, Zheng goto fail; 5933fd0a558SYan, Zheng 594f39911e5SQu Wenruo btrfs_backref_link_edge(new_edge, edge->node[LOWER], 595f39911e5SQu Wenruo new_node, LINK_UPPER); 5963fd0a558SYan, Zheng } 59776b9e23dSMiao Xie } else { 59876b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 5993fd0a558SYan, Zheng } 6003fd0a558SYan, Zheng 601e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr, 6023fd0a558SYan, Zheng &new_node->rb_node); 60343c04fb1SJeff Mahoney if (rb_node) 604982c92cbSQu Wenruo btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST); 6053fd0a558SYan, Zheng 6063fd0a558SYan, Zheng if (!new_node->lowest) { 6073fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 6083fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 6093fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 6103fd0a558SYan, Zheng } 6113fd0a558SYan, Zheng } 6123fd0a558SYan, Zheng return 0; 6133fd0a558SYan, Zheng fail: 6143fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 6153fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 616a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 6173fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 618741188d3SQu Wenruo btrfs_backref_free_edge(cache, new_edge); 6193fd0a558SYan, Zheng } 620741188d3SQu Wenruo btrfs_backref_free_node(cache, new_node); 6213fd0a558SYan, Zheng return -ENOMEM; 6223fd0a558SYan, Zheng } 6233fd0a558SYan, Zheng 6243fd0a558SYan, Zheng /* 6255d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 6265d4f98a2SYan Zheng */ 627ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 6285d4f98a2SYan Zheng { 6290b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6305d4f98a2SYan Zheng struct rb_node *rb_node; 6315d4f98a2SYan Zheng struct mapping_node *node; 6320b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 6335d4f98a2SYan Zheng 6345d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 635ffd7b339SJeff Mahoney if (!node) 636ffd7b339SJeff Mahoney return -ENOMEM; 6375d4f98a2SYan Zheng 638ea287ab1SJosef Bacik node->bytenr = root->commit_root->start; 6395d4f98a2SYan Zheng node->data = root; 6405d4f98a2SYan Zheng 6415d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 642e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 6435d4f98a2SYan Zheng node->bytenr, &node->rb_node); 6445d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 645ffd7b339SJeff Mahoney if (rb_node) { 64657a304cfSJosef Bacik btrfs_err(fs_info, 6475d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 6485d163e0eSJeff Mahoney node->bytenr); 64957a304cfSJosef Bacik return -EEXIST; 650ffd7b339SJeff Mahoney } 6515d4f98a2SYan Zheng 6525d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 6535d4f98a2SYan Zheng return 0; 6545d4f98a2SYan Zheng } 6555d4f98a2SYan Zheng 6565d4f98a2SYan Zheng /* 657c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 6585d4f98a2SYan Zheng * mapping 6595d4f98a2SYan Zheng */ 660c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 6615d4f98a2SYan Zheng { 6620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6635d4f98a2SYan Zheng struct rb_node *rb_node; 6645d4f98a2SYan Zheng struct mapping_node *node = NULL; 6650b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 666f44deb74SJosef Bacik bool put_ref = false; 6675d4f98a2SYan Zheng 66865c6e82bSQu Wenruo if (rc && root->node) { 6695d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 670e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 671ea287ab1SJosef Bacik root->commit_root->start); 672c974c464SWang Shilong if (rb_node) { 673c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 674c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 675ea287ab1SJosef Bacik RB_CLEAR_NODE(&node->rb_node); 676c974c464SWang Shilong } 677c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 678c78a10aeSJosef Bacik ASSERT(!node || (struct btrfs_root *)node->data == root); 679389305b2SQu Wenruo } 680c974c464SWang Shilong 681f44deb74SJosef Bacik /* 682f44deb74SJosef Bacik * We only put the reloc root here if it's on the list. There's a lot 683f44deb74SJosef Bacik * of places where the pattern is to splice the rc->reloc_roots, process 684f44deb74SJosef Bacik * the reloc roots, and then add the reloc root back onto 685f44deb74SJosef Bacik * rc->reloc_roots. If we call __del_reloc_root while it's off of the 686f44deb74SJosef Bacik * list we don't want the reference being dropped, because the guy 687f44deb74SJosef Bacik * messing with the list is in charge of the reference. 688f44deb74SJosef Bacik */ 6890b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 690f44deb74SJosef Bacik if (!list_empty(&root->root_list)) { 691f44deb74SJosef Bacik put_ref = true; 692c974c464SWang Shilong list_del_init(&root->root_list); 693f44deb74SJosef Bacik } 6940b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 695f44deb74SJosef Bacik if (put_ref) 696f44deb74SJosef Bacik btrfs_put_root(root); 697c974c464SWang Shilong kfree(node); 698c974c464SWang Shilong } 699c974c464SWang Shilong 700c974c464SWang Shilong /* 701c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 702c974c464SWang Shilong * mapping 703c974c464SWang Shilong */ 704ea287ab1SJosef Bacik static int __update_reloc_root(struct btrfs_root *root) 705c974c464SWang Shilong { 7060b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 707c974c464SWang Shilong struct rb_node *rb_node; 708c974c464SWang Shilong struct mapping_node *node = NULL; 7090b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 710c974c464SWang Shilong 711c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 712e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 713ea287ab1SJosef Bacik root->commit_root->start); 7145d4f98a2SYan Zheng if (rb_node) { 7155d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 7165d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 7175d4f98a2SYan Zheng } 7185d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 7195d4f98a2SYan Zheng 7208f71f3e0SLiu Bo if (!node) 7218f71f3e0SLiu Bo return 0; 7225d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 7235d4f98a2SYan Zheng 7245d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 725ea287ab1SJosef Bacik node->bytenr = root->node->start; 726e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 7275d4f98a2SYan Zheng node->bytenr, &node->rb_node); 7285d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 72943c04fb1SJeff Mahoney if (rb_node) 730982c92cbSQu Wenruo btrfs_backref_panic(fs_info, node->bytenr, -EEXIST); 7315d4f98a2SYan Zheng return 0; 7325d4f98a2SYan Zheng } 7335d4f98a2SYan Zheng 7343fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 7353fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 7365d4f98a2SYan Zheng { 7370b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 7385d4f98a2SYan Zheng struct btrfs_root *reloc_root; 7395d4f98a2SYan Zheng struct extent_buffer *eb; 7405d4f98a2SYan Zheng struct btrfs_root_item *root_item; 7415d4f98a2SYan Zheng struct btrfs_key root_key; 74284c50ba5SJosef Bacik int ret = 0; 74384c50ba5SJosef Bacik bool must_abort = false; 7445d4f98a2SYan Zheng 7455d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 74684c50ba5SJosef Bacik if (!root_item) 74784c50ba5SJosef Bacik return ERR_PTR(-ENOMEM); 7485d4f98a2SYan Zheng 7495d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 7505d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 7513fd0a558SYan, Zheng root_key.offset = objectid; 7525d4f98a2SYan Zheng 7533fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 754054570a1SFilipe Manana u64 commit_root_gen; 755054570a1SFilipe Manana 7563fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 7575d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 7585d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 75984c50ba5SJosef Bacik if (ret) 76084c50ba5SJosef Bacik goto fail; 76184c50ba5SJosef Bacik 762054570a1SFilipe Manana /* 763054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 764054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 765054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 766054570a1SFilipe Manana * either inside the critical section of a transaction commit 767054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 768054570a1SFilipe Manana * it's created before the transaction commit is started. 769054570a1SFilipe Manana */ 770054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 771054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 7723fd0a558SYan, Zheng } else { 7733fd0a558SYan, Zheng /* 7743fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 7753fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 7763fd0a558SYan, Zheng * modified after it was created have RELOC flag 7773fd0a558SYan, Zheng * set in their headers. so it's OK to not update 7783fd0a558SYan, Zheng * the 'last_snapshot'. 7793fd0a558SYan, Zheng */ 7803fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 7813fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 78284c50ba5SJosef Bacik if (ret) 78384c50ba5SJosef Bacik goto fail; 7843fd0a558SYan, Zheng } 7853fd0a558SYan, Zheng 78684c50ba5SJosef Bacik /* 78784c50ba5SJosef Bacik * We have changed references at this point, we must abort the 78884c50ba5SJosef Bacik * transaction if anything fails. 78984c50ba5SJosef Bacik */ 79084c50ba5SJosef Bacik must_abort = true; 79184c50ba5SJosef Bacik 7925d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 7935d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 7945d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 7955d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 7963fd0a558SYan, Zheng 7973fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 7983fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 7993fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 8003fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 801c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, 0); 8023fd0a558SYan, Zheng } 8035d4f98a2SYan Zheng 8045d4f98a2SYan Zheng btrfs_tree_unlock(eb); 8055d4f98a2SYan Zheng free_extent_buffer(eb); 8065d4f98a2SYan Zheng 8070b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 8085d4f98a2SYan Zheng &root_key, root_item); 80984c50ba5SJosef Bacik if (ret) 81084c50ba5SJosef Bacik goto fail; 81184c50ba5SJosef Bacik 8125d4f98a2SYan Zheng kfree(root_item); 8135d4f98a2SYan Zheng 8143dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 81584c50ba5SJosef Bacik if (IS_ERR(reloc_root)) { 81684c50ba5SJosef Bacik ret = PTR_ERR(reloc_root); 81784c50ba5SJosef Bacik goto abort; 81884c50ba5SJosef Bacik } 81992a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 8205d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 8213fd0a558SYan, Zheng return reloc_root; 82284c50ba5SJosef Bacik fail: 82384c50ba5SJosef Bacik kfree(root_item); 82484c50ba5SJosef Bacik abort: 82584c50ba5SJosef Bacik if (must_abort) 82684c50ba5SJosef Bacik btrfs_abort_transaction(trans, ret); 82784c50ba5SJosef Bacik return ERR_PTR(ret); 8283fd0a558SYan, Zheng } 8293fd0a558SYan, Zheng 8303fd0a558SYan, Zheng /* 8313fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 8323fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 833f44deb74SJosef Bacik * 834f44deb74SJosef Bacik * The reloc_root comes out of here with two references, one for 835f44deb74SJosef Bacik * root->reloc_root, and another for being on the rc->reloc_roots list. 8363fd0a558SYan, Zheng */ 8373fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 8383fd0a558SYan, Zheng struct btrfs_root *root) 8393fd0a558SYan, Zheng { 8400b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8413fd0a558SYan, Zheng struct btrfs_root *reloc_root; 8420b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 84320dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 8443fd0a558SYan, Zheng int clear_rsv = 0; 845ffd7b339SJeff Mahoney int ret; 8463fd0a558SYan, Zheng 847aec7db3bSJosef Bacik if (!rc) 8482abc726aSJosef Bacik return 0; 8492abc726aSJosef Bacik 8501fac4a54SQu Wenruo /* 8511fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 8521fac4a54SQu Wenruo * create/update the dead reloc tree 8531fac4a54SQu Wenruo */ 8546282675eSQu Wenruo if (reloc_root_is_dead(root)) 8551fac4a54SQu Wenruo return 0; 8561fac4a54SQu Wenruo 857aec7db3bSJosef Bacik /* 858aec7db3bSJosef Bacik * This is subtle but important. We do not do 859aec7db3bSJosef Bacik * record_root_in_transaction for reloc roots, instead we record their 860aec7db3bSJosef Bacik * corresponding fs root, and then here we update the last trans for the 861aec7db3bSJosef Bacik * reloc root. This means that we have to do this for the entire life 862aec7db3bSJosef Bacik * of the reloc root, regardless of which stage of the relocation we are 863aec7db3bSJosef Bacik * in. 864aec7db3bSJosef Bacik */ 8653fd0a558SYan, Zheng if (root->reloc_root) { 8663fd0a558SYan, Zheng reloc_root = root->reloc_root; 8673fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 8683fd0a558SYan, Zheng return 0; 8693fd0a558SYan, Zheng } 8703fd0a558SYan, Zheng 871aec7db3bSJosef Bacik /* 872aec7db3bSJosef Bacik * We are merging reloc roots, we do not need new reloc trees. Also 873aec7db3bSJosef Bacik * reloc trees never need their own reloc tree. 874aec7db3bSJosef Bacik */ 875aec7db3bSJosef Bacik if (!rc->create_reloc_tree || 876aec7db3bSJosef Bacik root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 877aec7db3bSJosef Bacik return 0; 878aec7db3bSJosef Bacik 87920dd2cbfSMiao Xie if (!trans->reloc_reserved) { 88020dd2cbfSMiao Xie rsv = trans->block_rsv; 8813fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 8823fd0a558SYan, Zheng clear_rsv = 1; 8833fd0a558SYan, Zheng } 8843fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 8853fd0a558SYan, Zheng if (clear_rsv) 88620dd2cbfSMiao Xie trans->block_rsv = rsv; 88700bb36a0SJosef Bacik if (IS_ERR(reloc_root)) 88800bb36a0SJosef Bacik return PTR_ERR(reloc_root); 8895d4f98a2SYan Zheng 890ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 89157a304cfSJosef Bacik ASSERT(ret != -EEXIST); 89200bb36a0SJosef Bacik if (ret) { 89300bb36a0SJosef Bacik /* Pairs with create_reloc_root */ 89400bb36a0SJosef Bacik btrfs_put_root(reloc_root); 89500bb36a0SJosef Bacik return ret; 89600bb36a0SJosef Bacik } 897f44deb74SJosef Bacik root->reloc_root = btrfs_grab_root(reloc_root); 8985d4f98a2SYan Zheng return 0; 8995d4f98a2SYan Zheng } 9005d4f98a2SYan Zheng 9015d4f98a2SYan Zheng /* 9025d4f98a2SYan Zheng * update root item of reloc tree 9035d4f98a2SYan Zheng */ 9045d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 9055d4f98a2SYan Zheng struct btrfs_root *root) 9065d4f98a2SYan Zheng { 9070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 9085d4f98a2SYan Zheng struct btrfs_root *reloc_root; 9095d4f98a2SYan Zheng struct btrfs_root_item *root_item; 9105d4f98a2SYan Zheng int ret; 9115d4f98a2SYan Zheng 9126282675eSQu Wenruo if (!have_reloc_root(root)) 913592fbcd5SJosef Bacik return 0; 9145d4f98a2SYan Zheng 9155d4f98a2SYan Zheng reloc_root = root->reloc_root; 9165d4f98a2SYan Zheng root_item = &reloc_root->root_item; 9175d4f98a2SYan Zheng 918f44deb74SJosef Bacik /* 919f44deb74SJosef Bacik * We are probably ok here, but __del_reloc_root() will drop its ref of 920f44deb74SJosef Bacik * the root. We have the ref for root->reloc_root, but just in case 921f44deb74SJosef Bacik * hold it while we update the reloc root. 922f44deb74SJosef Bacik */ 923f44deb74SJosef Bacik btrfs_grab_root(reloc_root); 924f44deb74SJosef Bacik 925d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 9260b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 9273fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 928d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 9296282675eSQu Wenruo /* 9306282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 9316282675eSQu Wenruo * have_reloc_root will not touch it from now on. 9326282675eSQu Wenruo */ 9336282675eSQu Wenruo smp_wmb(); 934c974c464SWang Shilong __del_reloc_root(reloc_root); 9355d4f98a2SYan Zheng } 9365d4f98a2SYan Zheng 9375d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 938ea287ab1SJosef Bacik __update_reloc_root(reloc_root); 9395d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 9405d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 9415d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 9425d4f98a2SYan Zheng } 9435d4f98a2SYan Zheng 9440b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 9455d4f98a2SYan Zheng &reloc_root->root_key, root_item); 946f44deb74SJosef Bacik btrfs_put_root(reloc_root); 947592fbcd5SJosef Bacik return ret; 9485d4f98a2SYan Zheng } 9495d4f98a2SYan Zheng 9505d4f98a2SYan Zheng /* 9515d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 9525d4f98a2SYan Zheng * in a subvolume 9535d4f98a2SYan Zheng */ 9545d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 9555d4f98a2SYan Zheng { 9565d4f98a2SYan Zheng struct rb_node *node; 9575d4f98a2SYan Zheng struct rb_node *prev; 9585d4f98a2SYan Zheng struct btrfs_inode *entry; 9595d4f98a2SYan Zheng struct inode *inode; 9605d4f98a2SYan Zheng 9615d4f98a2SYan Zheng spin_lock(&root->inode_lock); 9625d4f98a2SYan Zheng again: 9635d4f98a2SYan Zheng node = root->inode_tree.rb_node; 9645d4f98a2SYan Zheng prev = NULL; 9655d4f98a2SYan Zheng while (node) { 9665d4f98a2SYan Zheng prev = node; 9675d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9685d4f98a2SYan Zheng 9694a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 9705d4f98a2SYan Zheng node = node->rb_left; 9714a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 9725d4f98a2SYan Zheng node = node->rb_right; 9735d4f98a2SYan Zheng else 9745d4f98a2SYan Zheng break; 9755d4f98a2SYan Zheng } 9765d4f98a2SYan Zheng if (!node) { 9775d4f98a2SYan Zheng while (prev) { 9785d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 9794a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 9805d4f98a2SYan Zheng node = prev; 9815d4f98a2SYan Zheng break; 9825d4f98a2SYan Zheng } 9835d4f98a2SYan Zheng prev = rb_next(prev); 9845d4f98a2SYan Zheng } 9855d4f98a2SYan Zheng } 9865d4f98a2SYan Zheng while (node) { 9875d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9885d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 9895d4f98a2SYan Zheng if (inode) { 9905d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 9915d4f98a2SYan Zheng return inode; 9925d4f98a2SYan Zheng } 9935d4f98a2SYan Zheng 9944a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 9955d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 9965d4f98a2SYan Zheng goto again; 9975d4f98a2SYan Zheng 9985d4f98a2SYan Zheng node = rb_next(node); 9995d4f98a2SYan Zheng } 10005d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 10015d4f98a2SYan Zheng return NULL; 10025d4f98a2SYan Zheng } 10035d4f98a2SYan Zheng 10045d4f98a2SYan Zheng /* 10055d4f98a2SYan Zheng * get new location of data 10065d4f98a2SYan Zheng */ 10075d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 10085d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 10095d4f98a2SYan Zheng { 10105d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 10115d4f98a2SYan Zheng struct btrfs_path *path; 10125d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10135d4f98a2SYan Zheng struct extent_buffer *leaf; 10145d4f98a2SYan Zheng int ret; 10155d4f98a2SYan Zheng 10165d4f98a2SYan Zheng path = btrfs_alloc_path(); 10175d4f98a2SYan Zheng if (!path) 10185d4f98a2SYan Zheng return -ENOMEM; 10195d4f98a2SYan Zheng 10205d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1021f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1022f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 10235d4f98a2SYan Zheng if (ret < 0) 10245d4f98a2SYan Zheng goto out; 10255d4f98a2SYan Zheng if (ret > 0) { 10265d4f98a2SYan Zheng ret = -ENOENT; 10275d4f98a2SYan Zheng goto out; 10285d4f98a2SYan Zheng } 10295d4f98a2SYan Zheng 10305d4f98a2SYan Zheng leaf = path->nodes[0]; 10315d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 10325d4f98a2SYan Zheng struct btrfs_file_extent_item); 10335d4f98a2SYan Zheng 10345d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 10355d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 10365d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 10375d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 10385d4f98a2SYan Zheng 10395d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 104083d4cfd4SJosef Bacik ret = -EINVAL; 10415d4f98a2SYan Zheng goto out; 10425d4f98a2SYan Zheng } 10435d4f98a2SYan Zheng 10445d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 10455d4f98a2SYan Zheng ret = 0; 10465d4f98a2SYan Zheng out: 10475d4f98a2SYan Zheng btrfs_free_path(path); 10485d4f98a2SYan Zheng return ret; 10495d4f98a2SYan Zheng } 10505d4f98a2SYan Zheng 10515d4f98a2SYan Zheng /* 10525d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 10535d4f98a2SYan Zheng * the new locations. 10545d4f98a2SYan Zheng */ 10553fd0a558SYan, Zheng static noinline_for_stack 10563fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 10575d4f98a2SYan Zheng struct reloc_control *rc, 10585d4f98a2SYan Zheng struct btrfs_root *root, 10593fd0a558SYan, Zheng struct extent_buffer *leaf) 10605d4f98a2SYan Zheng { 10610b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 10625d4f98a2SYan Zheng struct btrfs_key key; 10635d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10645d4f98a2SYan Zheng struct inode *inode = NULL; 10655d4f98a2SYan Zheng u64 parent; 10665d4f98a2SYan Zheng u64 bytenr; 10673fd0a558SYan, Zheng u64 new_bytenr = 0; 10685d4f98a2SYan Zheng u64 num_bytes; 10695d4f98a2SYan Zheng u64 end; 10705d4f98a2SYan Zheng u32 nritems; 10715d4f98a2SYan Zheng u32 i; 107283d4cfd4SJosef Bacik int ret = 0; 10735d4f98a2SYan Zheng int first = 1; 10745d4f98a2SYan Zheng int dirty = 0; 10755d4f98a2SYan Zheng 10765d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 10775d4f98a2SYan Zheng return 0; 10785d4f98a2SYan Zheng 10795d4f98a2SYan Zheng /* reloc trees always use full backref */ 10805d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 10815d4f98a2SYan Zheng parent = leaf->start; 10825d4f98a2SYan Zheng else 10835d4f98a2SYan Zheng parent = 0; 10845d4f98a2SYan Zheng 10855d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 10865d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 108782fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 108882fa113fSQu Wenruo 10895d4f98a2SYan Zheng cond_resched(); 10905d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 10915d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 10925d4f98a2SYan Zheng continue; 10935d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 10945d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 10955d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 10965d4f98a2SYan Zheng continue; 10975d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 10985d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 10995d4f98a2SYan Zheng if (bytenr == 0) 11005d4f98a2SYan Zheng continue; 11019569cc20SQu Wenruo if (!in_range(bytenr, rc->block_group->start, 11029569cc20SQu Wenruo rc->block_group->length)) 11035d4f98a2SYan Zheng continue; 11045d4f98a2SYan Zheng 11055d4f98a2SYan Zheng /* 1106fb12489bSMatthew Wilcox (Oracle) * if we are modifying block in fs tree, wait for read_folio 11075d4f98a2SYan Zheng * to complete and drop the extent cache 11085d4f98a2SYan Zheng */ 11095d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 11105d4f98a2SYan Zheng if (first) { 11115d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11125d4f98a2SYan Zheng first = 0; 11134a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 11143fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 11155d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11165d4f98a2SYan Zheng } 11174a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 11189c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 11199c5c9604SJosef Bacik 11205d4f98a2SYan Zheng end = key.offset + 11215d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 11225d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 11230b246afaSJeff Mahoney fs_info->sectorsize)); 11240b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 11255d4f98a2SYan Zheng end--; 11265d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 11279c5c9604SJosef Bacik key.offset, end, 11289c5c9604SJosef Bacik &cached_state); 11295d4f98a2SYan Zheng if (!ret) 11305d4f98a2SYan Zheng continue; 11315d4f98a2SYan Zheng 11324c0c8cfcSFilipe Manana btrfs_drop_extent_map_range(BTRFS_I(inode), 11334c0c8cfcSFilipe Manana key.offset, end, true); 11345d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 11359c5c9604SJosef Bacik key.offset, end, &cached_state); 11365d4f98a2SYan Zheng } 11375d4f98a2SYan Zheng } 11385d4f98a2SYan Zheng 11395d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 11405d4f98a2SYan Zheng bytenr, num_bytes); 114183d4cfd4SJosef Bacik if (ret) { 114283d4cfd4SJosef Bacik /* 114383d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 114483d4cfd4SJosef Bacik * in the file extent yet. 114583d4cfd4SJosef Bacik */ 114683d4cfd4SJosef Bacik break; 11473fd0a558SYan, Zheng } 11485d4f98a2SYan Zheng 11495d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 11505d4f98a2SYan Zheng dirty = 1; 11515d4f98a2SYan Zheng 11525d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 115382fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 115482fa113fSQu Wenruo num_bytes, parent); 115582fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1156f42c5da6SNikolay Borisov key.objectid, key.offset, 1157f42c5da6SNikolay Borisov root->root_key.objectid, false); 115882fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 115983d4cfd4SJosef Bacik if (ret) { 116066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 116183d4cfd4SJosef Bacik break; 116283d4cfd4SJosef Bacik } 11635d4f98a2SYan Zheng 1164ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1165ffd4bb2aSQu Wenruo num_bytes, parent); 1166ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1167f42c5da6SNikolay Borisov key.objectid, key.offset, 1168f42c5da6SNikolay Borisov root->root_key.objectid, false); 1169ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 117083d4cfd4SJosef Bacik if (ret) { 117166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 117283d4cfd4SJosef Bacik break; 117383d4cfd4SJosef Bacik } 11745d4f98a2SYan Zheng } 11755d4f98a2SYan Zheng if (dirty) 11765d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 11773fd0a558SYan, Zheng if (inode) 11783fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 117983d4cfd4SJosef Bacik return ret; 11805d4f98a2SYan Zheng } 11815d4f98a2SYan Zheng 11825d4f98a2SYan Zheng static noinline_for_stack 11835d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 11845d4f98a2SYan Zheng struct btrfs_path *path, int level) 11855d4f98a2SYan Zheng { 11865d4f98a2SYan Zheng struct btrfs_disk_key key1; 11875d4f98a2SYan Zheng struct btrfs_disk_key key2; 11885d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 11895d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 11905d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 11915d4f98a2SYan Zheng } 11925d4f98a2SYan Zheng 11935d4f98a2SYan Zheng /* 11945d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 11955d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 11965d4f98a2SYan Zheng * reloc tree was create can be replaced. 11975d4f98a2SYan Zheng * 11985d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 11995d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 12005d4f98a2SYan Zheng * errors, a negative error number is returned. 12015d4f98a2SYan Zheng */ 12023fd0a558SYan, Zheng static noinline_for_stack 12033d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 12045d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 12055d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 12065d4f98a2SYan Zheng int lowest_level, int max_level) 12075d4f98a2SYan Zheng { 12080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 12095d4f98a2SYan Zheng struct extent_buffer *eb; 12105d4f98a2SYan Zheng struct extent_buffer *parent; 121182fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 12125d4f98a2SYan Zheng struct btrfs_key key; 12135d4f98a2SYan Zheng u64 old_bytenr; 12145d4f98a2SYan Zheng u64 new_bytenr; 12155d4f98a2SYan Zheng u64 old_ptr_gen; 12165d4f98a2SYan Zheng u64 new_ptr_gen; 12175d4f98a2SYan Zheng u64 last_snapshot; 12185d4f98a2SYan Zheng u32 blocksize; 12193fd0a558SYan, Zheng int cow = 0; 12205d4f98a2SYan Zheng int level; 12215d4f98a2SYan Zheng int ret; 12225d4f98a2SYan Zheng int slot; 12235d4f98a2SYan Zheng 12247a9213a9SJosef Bacik ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 12257a9213a9SJosef Bacik ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 12265d4f98a2SYan Zheng 12275d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 12283fd0a558SYan, Zheng again: 12295d4f98a2SYan Zheng slot = path->slots[lowest_level]; 12305d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 12315d4f98a2SYan Zheng 12325d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 12335d4f98a2SYan Zheng level = btrfs_header_level(eb); 12345d4f98a2SYan Zheng 12355d4f98a2SYan Zheng if (level < lowest_level) { 12365d4f98a2SYan Zheng btrfs_tree_unlock(eb); 12375d4f98a2SYan Zheng free_extent_buffer(eb); 12385d4f98a2SYan Zheng return 0; 12395d4f98a2SYan Zheng } 12405d4f98a2SYan Zheng 12413fd0a558SYan, Zheng if (cow) { 12429631e4ccSJosef Bacik ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb, 12439631e4ccSJosef Bacik BTRFS_NESTING_COW); 124445b87c5dSJosef Bacik if (ret) { 124545b87c5dSJosef Bacik btrfs_tree_unlock(eb); 124645b87c5dSJosef Bacik free_extent_buffer(eb); 124745b87c5dSJosef Bacik return ret; 124845b87c5dSJosef Bacik } 12493fd0a558SYan, Zheng } 12505d4f98a2SYan Zheng 12515d4f98a2SYan Zheng if (next_key) { 12525d4f98a2SYan Zheng next_key->objectid = (u64)-1; 12535d4f98a2SYan Zheng next_key->type = (u8)-1; 12545d4f98a2SYan Zheng next_key->offset = (u64)-1; 12555d4f98a2SYan Zheng } 12565d4f98a2SYan Zheng 12575d4f98a2SYan Zheng parent = eb; 12585d4f98a2SYan Zheng while (1) { 12595d4f98a2SYan Zheng level = btrfs_header_level(parent); 12607a9213a9SJosef Bacik ASSERT(level >= lowest_level); 12615d4f98a2SYan Zheng 1262e3b83361SQu Wenruo ret = btrfs_bin_search(parent, &key, &slot); 1263cbca7d59SFilipe Manana if (ret < 0) 1264cbca7d59SFilipe Manana break; 12655d4f98a2SYan Zheng if (ret && slot > 0) 12665d4f98a2SYan Zheng slot--; 12675d4f98a2SYan Zheng 12685d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 12695d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 12705d4f98a2SYan Zheng 12715d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 12720b246afaSJeff Mahoney blocksize = fs_info->nodesize; 12735d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 12745d4f98a2SYan Zheng 12755d4f98a2SYan Zheng if (level <= max_level) { 12765d4f98a2SYan Zheng eb = path->nodes[level]; 12775d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 12785d4f98a2SYan Zheng path->slots[level]); 12795d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 12805d4f98a2SYan Zheng path->slots[level]); 12815d4f98a2SYan Zheng } else { 12825d4f98a2SYan Zheng new_bytenr = 0; 12835d4f98a2SYan Zheng new_ptr_gen = 0; 12845d4f98a2SYan Zheng } 12855d4f98a2SYan Zheng 1286fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 12875d4f98a2SYan Zheng ret = level; 12885d4f98a2SYan Zheng break; 12895d4f98a2SYan Zheng } 12905d4f98a2SYan Zheng 12915d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 12925d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 12933fd0a558SYan, Zheng if (level <= lowest_level) { 12945d4f98a2SYan Zheng ret = 0; 12955d4f98a2SYan Zheng break; 12965d4f98a2SYan Zheng } 12975d4f98a2SYan Zheng 12986b3426beSJosef Bacik eb = btrfs_read_node_slot(parent, slot); 129964c043deSLiu Bo if (IS_ERR(eb)) { 130064c043deSLiu Bo ret = PTR_ERR(eb); 1301264813acSLiu Bo break; 1302416bc658SJosef Bacik } 13035d4f98a2SYan Zheng btrfs_tree_lock(eb); 13043fd0a558SYan, Zheng if (cow) { 13055d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 13069631e4ccSJosef Bacik slot, &eb, 13079631e4ccSJosef Bacik BTRFS_NESTING_COW); 130845b87c5dSJosef Bacik if (ret) { 130945b87c5dSJosef Bacik btrfs_tree_unlock(eb); 131045b87c5dSJosef Bacik free_extent_buffer(eb); 131145b87c5dSJosef Bacik break; 131245b87c5dSJosef Bacik } 13135d4f98a2SYan Zheng } 13145d4f98a2SYan Zheng 13155d4f98a2SYan Zheng btrfs_tree_unlock(parent); 13165d4f98a2SYan Zheng free_extent_buffer(parent); 13175d4f98a2SYan Zheng 13185d4f98a2SYan Zheng parent = eb; 13195d4f98a2SYan Zheng continue; 13205d4f98a2SYan Zheng } 13215d4f98a2SYan Zheng 13223fd0a558SYan, Zheng if (!cow) { 13233fd0a558SYan, Zheng btrfs_tree_unlock(parent); 13243fd0a558SYan, Zheng free_extent_buffer(parent); 13253fd0a558SYan, Zheng cow = 1; 13263fd0a558SYan, Zheng goto again; 13273fd0a558SYan, Zheng } 13283fd0a558SYan, Zheng 13295d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 13305d4f98a2SYan Zheng path->slots[level]); 1331b3b4aa74SDavid Sterba btrfs_release_path(path); 13325d4f98a2SYan Zheng 13335d4f98a2SYan Zheng path->lowest_level = level; 1334b40130b2SJosef Bacik set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); 13355d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 1336b40130b2SJosef Bacik clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); 13375d4f98a2SYan Zheng path->lowest_level = 0; 13380e9873e2SJosef Bacik if (ret) { 13390e9873e2SJosef Bacik if (ret > 0) 13400e9873e2SJosef Bacik ret = -ENOENT; 13410e9873e2SJosef Bacik break; 13420e9873e2SJosef Bacik } 13435d4f98a2SYan Zheng 13445d4f98a2SYan Zheng /* 1345824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1346824d8dffSQu Wenruo * 1347824d8dffSQu Wenruo * We must trace both trees. 1348824d8dffSQu Wenruo * 1) Tree reloc subtree 1349824d8dffSQu Wenruo * If not traced, we will leak data numbers 1350824d8dffSQu Wenruo * 2) Fs subtree 1351824d8dffSQu Wenruo * If not traced, we will double count old data 1352f616f5cdSQu Wenruo * 1353f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1354f616f5cdSQu Wenruo * the swapped tree blocks. 1355f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1356f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1357824d8dffSQu Wenruo */ 1358370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1359370a11b8SQu Wenruo rc->block_group, parent, slot, 1360370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1361370a11b8SQu Wenruo last_snapshot); 1362370a11b8SQu Wenruo if (ret < 0) 1363370a11b8SQu Wenruo break; 1364824d8dffSQu Wenruo /* 13655d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 13665d4f98a2SYan Zheng */ 13675d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 13685d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 13695d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 13705d4f98a2SYan Zheng 13715d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 13725d4f98a2SYan Zheng path->slots[level], old_bytenr); 13735d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 13745d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 13755d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 13765d4f98a2SYan Zheng 137782fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 137882fa113fSQu Wenruo blocksize, path->nodes[level]->start); 1379f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1380f42c5da6SNikolay Borisov 0, true); 138182fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1382253e258cSJosef Bacik if (ret) { 1383253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1384253e258cSJosef Bacik break; 1385253e258cSJosef Bacik } 138682fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 138782fa113fSQu Wenruo blocksize, 0); 1388f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0, 1389f42c5da6SNikolay Borisov true); 139082fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1391253e258cSJosef Bacik if (ret) { 1392253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1393253e258cSJosef Bacik break; 1394253e258cSJosef Bacik } 13955d4f98a2SYan Zheng 1396ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1397ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1398f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1399f42c5da6SNikolay Borisov 0, true); 1400ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1401253e258cSJosef Bacik if (ret) { 1402253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1403253e258cSJosef Bacik break; 1404253e258cSJosef Bacik } 14055d4f98a2SYan Zheng 1406ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1407ffd4bb2aSQu Wenruo blocksize, 0); 1408f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 1409f42c5da6SNikolay Borisov 0, true); 1410ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1411253e258cSJosef Bacik if (ret) { 1412253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1413253e258cSJosef Bacik break; 1414253e258cSJosef Bacik } 14155d4f98a2SYan Zheng 14165d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 14175d4f98a2SYan Zheng 14185d4f98a2SYan Zheng ret = level; 14195d4f98a2SYan Zheng break; 14205d4f98a2SYan Zheng } 14215d4f98a2SYan Zheng btrfs_tree_unlock(parent); 14225d4f98a2SYan Zheng free_extent_buffer(parent); 14235d4f98a2SYan Zheng return ret; 14245d4f98a2SYan Zheng } 14255d4f98a2SYan Zheng 14265d4f98a2SYan Zheng /* 14275d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 14285d4f98a2SYan Zheng */ 14295d4f98a2SYan Zheng static noinline_for_stack 14305d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14315d4f98a2SYan Zheng int *level) 14325d4f98a2SYan Zheng { 14335d4f98a2SYan Zheng struct extent_buffer *eb; 14345d4f98a2SYan Zheng int i; 14355d4f98a2SYan Zheng u64 last_snapshot; 14365d4f98a2SYan Zheng u32 nritems; 14375d4f98a2SYan Zheng 14385d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14395d4f98a2SYan Zheng 14405d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 14415d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14425d4f98a2SYan Zheng path->nodes[i] = NULL; 14435d4f98a2SYan Zheng } 14445d4f98a2SYan Zheng 14455d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 14465d4f98a2SYan Zheng eb = path->nodes[i]; 14475d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14485d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 14495d4f98a2SYan Zheng path->slots[i]++; 14505d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 14515d4f98a2SYan Zheng last_snapshot) 14525d4f98a2SYan Zheng continue; 14535d4f98a2SYan Zheng 14545d4f98a2SYan Zheng *level = i; 14555d4f98a2SYan Zheng return 0; 14565d4f98a2SYan Zheng } 14575d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14585d4f98a2SYan Zheng path->nodes[i] = NULL; 14595d4f98a2SYan Zheng } 14605d4f98a2SYan Zheng return 1; 14615d4f98a2SYan Zheng } 14625d4f98a2SYan Zheng 14635d4f98a2SYan Zheng /* 14645d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 14655d4f98a2SYan Zheng */ 14665d4f98a2SYan Zheng static noinline_for_stack 14675d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14685d4f98a2SYan Zheng int *level) 14695d4f98a2SYan Zheng { 14705d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 14715d4f98a2SYan Zheng int i; 14725d4f98a2SYan Zheng u64 ptr_gen = 0; 14735d4f98a2SYan Zheng u64 last_snapshot; 14745d4f98a2SYan Zheng u32 nritems; 14755d4f98a2SYan Zheng 14765d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14775d4f98a2SYan Zheng 14785d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 14795d4f98a2SYan Zheng eb = path->nodes[i]; 14805d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14815d4f98a2SYan Zheng while (path->slots[i] < nritems) { 14825d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 14835d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 14845d4f98a2SYan Zheng break; 14855d4f98a2SYan Zheng path->slots[i]++; 14865d4f98a2SYan Zheng } 14875d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 14885d4f98a2SYan Zheng if (i == *level) 14895d4f98a2SYan Zheng break; 14905d4f98a2SYan Zheng *level = i + 1; 14915d4f98a2SYan Zheng return 0; 14925d4f98a2SYan Zheng } 14935d4f98a2SYan Zheng if (i == 1) { 14945d4f98a2SYan Zheng *level = i; 14955d4f98a2SYan Zheng return 0; 14965d4f98a2SYan Zheng } 14975d4f98a2SYan Zheng 14988ef385bbSJosef Bacik eb = btrfs_read_node_slot(eb, path->slots[i]); 14998ef385bbSJosef Bacik if (IS_ERR(eb)) 150064c043deSLiu Bo return PTR_ERR(eb); 15015d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 15025d4f98a2SYan Zheng path->nodes[i - 1] = eb; 15035d4f98a2SYan Zheng path->slots[i - 1] = 0; 15045d4f98a2SYan Zheng } 15055d4f98a2SYan Zheng return 1; 15065d4f98a2SYan Zheng } 15075d4f98a2SYan Zheng 15085d4f98a2SYan Zheng /* 15095d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 15105d4f98a2SYan Zheng * [min_key, max_key) 15115d4f98a2SYan Zheng */ 15125d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 15135d4f98a2SYan Zheng struct btrfs_key *min_key, 15145d4f98a2SYan Zheng struct btrfs_key *max_key) 15155d4f98a2SYan Zheng { 15160b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15175d4f98a2SYan Zheng struct inode *inode = NULL; 15185d4f98a2SYan Zheng u64 objectid; 15195d4f98a2SYan Zheng u64 start, end; 152033345d01SLi Zefan u64 ino; 15215d4f98a2SYan Zheng 15225d4f98a2SYan Zheng objectid = min_key->objectid; 15235d4f98a2SYan Zheng while (1) { 15249c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 15259c5c9604SJosef Bacik 15265d4f98a2SYan Zheng cond_resched(); 15275d4f98a2SYan Zheng iput(inode); 15285d4f98a2SYan Zheng 15295d4f98a2SYan Zheng if (objectid > max_key->objectid) 15305d4f98a2SYan Zheng break; 15315d4f98a2SYan Zheng 15325d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 15335d4f98a2SYan Zheng if (!inode) 15345d4f98a2SYan Zheng break; 15354a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 15365d4f98a2SYan Zheng 153733345d01SLi Zefan if (ino > max_key->objectid) { 15385d4f98a2SYan Zheng iput(inode); 15395d4f98a2SYan Zheng break; 15405d4f98a2SYan Zheng } 15415d4f98a2SYan Zheng 154233345d01SLi Zefan objectid = ino + 1; 15435d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 15445d4f98a2SYan Zheng continue; 15455d4f98a2SYan Zheng 154633345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 15475d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 15485d4f98a2SYan Zheng continue; 15495d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 15505d4f98a2SYan Zheng start = 0; 15515d4f98a2SYan Zheng else { 15525d4f98a2SYan Zheng start = min_key->offset; 15530b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 15545d4f98a2SYan Zheng } 15555d4f98a2SYan Zheng } else { 15565d4f98a2SYan Zheng start = 0; 15575d4f98a2SYan Zheng } 15585d4f98a2SYan Zheng 155933345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 15605d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 15615d4f98a2SYan Zheng continue; 15625d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 15635d4f98a2SYan Zheng end = (u64)-1; 15645d4f98a2SYan Zheng } else { 15655d4f98a2SYan Zheng if (max_key->offset == 0) 15665d4f98a2SYan Zheng continue; 15675d4f98a2SYan Zheng end = max_key->offset; 15680b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 15695d4f98a2SYan Zheng end--; 15705d4f98a2SYan Zheng } 15715d4f98a2SYan Zheng } else { 15725d4f98a2SYan Zheng end = (u64)-1; 15735d4f98a2SYan Zheng } 15745d4f98a2SYan Zheng 1575fb12489bSMatthew Wilcox (Oracle) /* the lock_extent waits for read_folio to complete */ 15769c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 15774c0c8cfcSFilipe Manana btrfs_drop_extent_map_range(BTRFS_I(inode), start, end, true); 15789c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 15795d4f98a2SYan Zheng } 15805d4f98a2SYan Zheng return 0; 15815d4f98a2SYan Zheng } 15825d4f98a2SYan Zheng 15835d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 15845d4f98a2SYan Zheng struct btrfs_key *key) 15855d4f98a2SYan Zheng 15865d4f98a2SYan Zheng { 15875d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 15885d4f98a2SYan Zheng if (!path->nodes[level]) 15895d4f98a2SYan Zheng break; 15905d4f98a2SYan Zheng if (path->slots[level] + 1 < 15915d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 15925d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 15935d4f98a2SYan Zheng path->slots[level] + 1); 15945d4f98a2SYan Zheng return 0; 15955d4f98a2SYan Zheng } 15965d4f98a2SYan Zheng level++; 15975d4f98a2SYan Zheng } 15985d4f98a2SYan Zheng return 1; 15995d4f98a2SYan Zheng } 16005d4f98a2SYan Zheng 16015d4f98a2SYan Zheng /* 1602d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 1603d2311e69SQu Wenruo */ 1604ac54da6cSJosef Bacik static int insert_dirty_subvol(struct btrfs_trans_handle *trans, 1605d2311e69SQu Wenruo struct reloc_control *rc, 1606d2311e69SQu Wenruo struct btrfs_root *root) 1607d2311e69SQu Wenruo { 1608d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1609d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 16107934133fSJosef Bacik int ret; 1611d2311e69SQu Wenruo 1612d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 1613d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 1614d2311e69SQu Wenruo ASSERT(reloc_root); 1615d2311e69SQu Wenruo 1616d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 1617d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 1618d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 1619c8422684SDavid Sterba btrfs_set_root_drop_level(reloc_root_item, 0); 1620d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 16217934133fSJosef Bacik ret = btrfs_update_reloc_root(trans, root); 16227934133fSJosef Bacik if (ret) 16237934133fSJosef Bacik return ret; 1624d2311e69SQu Wenruo 1625d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 162600246528SJosef Bacik btrfs_grab_root(root); 1627d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 1628d2311e69SQu Wenruo } 1629ac54da6cSJosef Bacik 1630ac54da6cSJosef Bacik return 0; 1631d2311e69SQu Wenruo } 1632d2311e69SQu Wenruo 1633d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 1634d2311e69SQu Wenruo { 1635d2311e69SQu Wenruo struct btrfs_root *root; 1636d2311e69SQu Wenruo struct btrfs_root *next; 1637d2311e69SQu Wenruo int ret = 0; 163830d40577SQu Wenruo int ret2; 1639d2311e69SQu Wenruo 1640d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 1641d2311e69SQu Wenruo reloc_dirty_list) { 164230d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 164330d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 1644d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1645d2311e69SQu Wenruo 1646d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 1647d2311e69SQu Wenruo root->reloc_root = NULL; 16486282675eSQu Wenruo /* 16496282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 16506282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 16516282675eSQu Wenruo */ 16526282675eSQu Wenruo smp_wmb(); 16531fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 1654f28de8d8SJosef Bacik if (reloc_root) { 1655f44deb74SJosef Bacik /* 1656f44deb74SJosef Bacik * btrfs_drop_snapshot drops our ref we hold for 1657f44deb74SJosef Bacik * ->reloc_root. If it fails however we must 1658f44deb74SJosef Bacik * drop the ref ourselves. 1659f44deb74SJosef Bacik */ 1660f28de8d8SJosef Bacik ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); 1661f44deb74SJosef Bacik if (ret2 < 0) { 1662f44deb74SJosef Bacik btrfs_put_root(reloc_root); 1663f44deb74SJosef Bacik if (!ret) 1664f28de8d8SJosef Bacik ret = ret2; 1665f28de8d8SJosef Bacik } 1666f44deb74SJosef Bacik } 166700246528SJosef Bacik btrfs_put_root(root); 166830d40577SQu Wenruo } else { 166930d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 16700078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(root, 0, 1); 1671f44deb74SJosef Bacik if (ret2 < 0) { 1672f44deb74SJosef Bacik btrfs_put_root(root); 1673f44deb74SJosef Bacik if (!ret) 167430d40577SQu Wenruo ret = ret2; 167530d40577SQu Wenruo } 1676d2311e69SQu Wenruo } 1677f44deb74SJosef Bacik } 1678d2311e69SQu Wenruo return ret; 1679d2311e69SQu Wenruo } 1680d2311e69SQu Wenruo 1681d2311e69SQu Wenruo /* 16825d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 16835d4f98a2SYan Zheng * fs tree. 16845d4f98a2SYan Zheng */ 16855d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 16865d4f98a2SYan Zheng struct btrfs_root *root) 16875d4f98a2SYan Zheng { 16880b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 16895d4f98a2SYan Zheng struct btrfs_key key; 16905d4f98a2SYan Zheng struct btrfs_key next_key; 16919e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 16925d4f98a2SYan Zheng struct btrfs_root *reloc_root; 16935d4f98a2SYan Zheng struct btrfs_root_item *root_item; 16945d4f98a2SYan Zheng struct btrfs_path *path; 16953fd0a558SYan, Zheng struct extent_buffer *leaf; 1696fca3a45dSJosef Bacik int reserve_level; 16975d4f98a2SYan Zheng int level; 16985d4f98a2SYan Zheng int max_level; 16995d4f98a2SYan Zheng int replaced = 0; 1700c6a592f2SNikolay Borisov int ret = 0; 17013fd0a558SYan, Zheng u32 min_reserved; 17025d4f98a2SYan Zheng 17035d4f98a2SYan Zheng path = btrfs_alloc_path(); 17045d4f98a2SYan Zheng if (!path) 17055d4f98a2SYan Zheng return -ENOMEM; 1706e4058b54SDavid Sterba path->reada = READA_FORWARD; 17075d4f98a2SYan Zheng 17085d4f98a2SYan Zheng reloc_root = root->reloc_root; 17095d4f98a2SYan Zheng root_item = &reloc_root->root_item; 17105d4f98a2SYan Zheng 17115d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 17125d4f98a2SYan Zheng level = btrfs_root_level(root_item); 171367439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 17145d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 17155d4f98a2SYan Zheng path->slots[level] = 0; 17165d4f98a2SYan Zheng } else { 17175d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 17185d4f98a2SYan Zheng 1719c8422684SDavid Sterba level = btrfs_root_drop_level(root_item); 17205d4f98a2SYan Zheng BUG_ON(level == 0); 17215d4f98a2SYan Zheng path->lowest_level = level; 17225d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 172333c66f43SYan Zheng path->lowest_level = 0; 17245d4f98a2SYan Zheng if (ret < 0) { 17255d4f98a2SYan Zheng btrfs_free_path(path); 17265d4f98a2SYan Zheng return ret; 17275d4f98a2SYan Zheng } 17285d4f98a2SYan Zheng 17295d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 17305d4f98a2SYan Zheng path->slots[level]); 17315d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 17325d4f98a2SYan Zheng 17335d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 17345d4f98a2SYan Zheng } 17355d4f98a2SYan Zheng 173644d354abSQu Wenruo /* 173744d354abSQu Wenruo * In merge_reloc_root(), we modify the upper level pointer to swap the 173844d354abSQu Wenruo * tree blocks between reloc tree and subvolume tree. Thus for tree 173944d354abSQu Wenruo * block COW, we COW at most from level 1 to root level for each tree. 174044d354abSQu Wenruo * 174144d354abSQu Wenruo * Thus the needed metadata size is at most root_level * nodesize, 174244d354abSQu Wenruo * and * 2 since we have two trees to COW. 174344d354abSQu Wenruo */ 1744fca3a45dSJosef Bacik reserve_level = max_t(int, 1, btrfs_root_level(root_item)); 1745fca3a45dSJosef Bacik min_reserved = fs_info->nodesize * reserve_level * 2; 17465d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 17475d4f98a2SYan Zheng 17485d4f98a2SYan Zheng while (1) { 17499270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 17509270501cSJosef Bacik min_reserved, 175144d354abSQu Wenruo BTRFS_RESERVE_FLUSH_LIMIT); 1752c6a592f2SNikolay Borisov if (ret) 17539e6a0c52SJosef Bacik goto out; 17549e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 17559e6a0c52SJosef Bacik if (IS_ERR(trans)) { 1756c6a592f2SNikolay Borisov ret = PTR_ERR(trans); 17579e6a0c52SJosef Bacik trans = NULL; 17589e6a0c52SJosef Bacik goto out; 17599e6a0c52SJosef Bacik } 17602abc726aSJosef Bacik 17612abc726aSJosef Bacik /* 17622abc726aSJosef Bacik * At this point we no longer have a reloc_control, so we can't 17632abc726aSJosef Bacik * depend on btrfs_init_reloc_root to update our last_trans. 17642abc726aSJosef Bacik * 17652abc726aSJosef Bacik * But that's ok, we started the trans handle on our 17662abc726aSJosef Bacik * corresponding fs_root, which means it's been added to the 17672abc726aSJosef Bacik * dirty list. At commit time we'll still call 17682abc726aSJosef Bacik * btrfs_update_reloc_root() and update our root item 17692abc726aSJosef Bacik * appropriately. 17702abc726aSJosef Bacik */ 17712abc726aSJosef Bacik reloc_root->last_trans = trans->transid; 17729e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 17733fd0a558SYan, Zheng 17743fd0a558SYan, Zheng replaced = 0; 17755d4f98a2SYan Zheng max_level = level; 17765d4f98a2SYan Zheng 17775d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 1778c6a592f2SNikolay Borisov if (ret < 0) 17795d4f98a2SYan Zheng goto out; 17805d4f98a2SYan Zheng if (ret > 0) 17815d4f98a2SYan Zheng break; 17825d4f98a2SYan Zheng 17835d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 17845d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 17855d4f98a2SYan Zheng ret = 0; 17865d4f98a2SYan Zheng } else { 17873d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 17883fd0a558SYan, Zheng &next_key, level, max_level); 17895d4f98a2SYan Zheng } 1790c6a592f2SNikolay Borisov if (ret < 0) 17915d4f98a2SYan Zheng goto out; 17925d4f98a2SYan Zheng if (ret > 0) { 17935d4f98a2SYan Zheng level = ret; 17945d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 17955d4f98a2SYan Zheng path->slots[level]); 17965d4f98a2SYan Zheng replaced = 1; 17975d4f98a2SYan Zheng } 17985d4f98a2SYan Zheng 17995d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 18005d4f98a2SYan Zheng if (ret > 0) 18015d4f98a2SYan Zheng break; 18025d4f98a2SYan Zheng 18035d4f98a2SYan Zheng BUG_ON(level == 0); 18045d4f98a2SYan Zheng /* 18055d4f98a2SYan Zheng * save the merging progress in the drop_progress. 18065d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 18075d4f98a2SYan Zheng */ 18085d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 18095d4f98a2SYan Zheng path->slots[level]); 1810c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, level); 18115d4f98a2SYan Zheng 18123a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18139e6a0c52SJosef Bacik trans = NULL; 18145d4f98a2SYan Zheng 18152ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18165d4f98a2SYan Zheng 18175d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18185d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18195d4f98a2SYan Zheng } 18205d4f98a2SYan Zheng 18215d4f98a2SYan Zheng /* 18225d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 18235d4f98a2SYan Zheng * relocated and the block is tree root. 18245d4f98a2SYan Zheng */ 18255d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 18269631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf, 18279631e4ccSJosef Bacik BTRFS_NESTING_COW); 18285d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 18295d4f98a2SYan Zheng free_extent_buffer(leaf); 18305d4f98a2SYan Zheng out: 18315d4f98a2SYan Zheng btrfs_free_path(path); 18325d4f98a2SYan Zheng 1833ac54da6cSJosef Bacik if (ret == 0) { 1834ac54da6cSJosef Bacik ret = insert_dirty_subvol(trans, rc, root); 1835ac54da6cSJosef Bacik if (ret) 1836ac54da6cSJosef Bacik btrfs_abort_transaction(trans, ret); 1837ac54da6cSJosef Bacik } 18385d4f98a2SYan Zheng 18399e6a0c52SJosef Bacik if (trans) 18403a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18415d4f98a2SYan Zheng 18422ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18435d4f98a2SYan Zheng 18445d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18455d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18465d4f98a2SYan Zheng 1847c6a592f2SNikolay Borisov return ret; 18485d4f98a2SYan Zheng } 18495d4f98a2SYan Zheng 18503fd0a558SYan, Zheng static noinline_for_stack 18513fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 18525d4f98a2SYan Zheng { 18533fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 18540b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 18553fd0a558SYan, Zheng struct btrfs_root *reloc_root; 18565d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 18573fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 18583fd0a558SYan, Zheng u64 num_bytes = 0; 18593fd0a558SYan, Zheng int ret; 18603fd0a558SYan, Zheng 18610b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 18620b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 18633fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 18640b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 18657585717fSChris Mason 18663fd0a558SYan, Zheng again: 18673fd0a558SYan, Zheng if (!err) { 18683fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 18699270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes, 187008e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 18713fd0a558SYan, Zheng if (ret) 18723fd0a558SYan, Zheng err = ret; 18733fd0a558SYan, Zheng } 18743fd0a558SYan, Zheng 18757a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 18763612b495STsutomu Itoh if (IS_ERR(trans)) { 18773612b495STsutomu Itoh if (!err) 18782ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 187963f018beSNikolay Borisov num_bytes, NULL); 18803612b495STsutomu Itoh return PTR_ERR(trans); 18813612b495STsutomu Itoh } 18823fd0a558SYan, Zheng 18833fd0a558SYan, Zheng if (!err) { 18843fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 18853a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18862ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 188763f018beSNikolay Borisov num_bytes, NULL); 18883fd0a558SYan, Zheng goto again; 18893fd0a558SYan, Zheng } 18903fd0a558SYan, Zheng } 18913fd0a558SYan, Zheng 18923fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 18933fd0a558SYan, Zheng 18943fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 18953fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 18963fd0a558SYan, Zheng struct btrfs_root, root_list); 18973fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 18983fd0a558SYan, Zheng 1899a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1900a820feb5SDavid Sterba false); 1901e0b085b0SJosef Bacik if (IS_ERR(root)) { 1902e0b085b0SJosef Bacik /* 1903e0b085b0SJosef Bacik * Even if we have an error we need this reloc root 1904e0b085b0SJosef Bacik * back on our list so we can clean up properly. 1905e0b085b0SJosef Bacik */ 1906e0b085b0SJosef Bacik list_add(&reloc_root->root_list, &reloc_roots); 1907e0b085b0SJosef Bacik btrfs_abort_transaction(trans, (int)PTR_ERR(root)); 1908e0b085b0SJosef Bacik if (!err) 1909e0b085b0SJosef Bacik err = PTR_ERR(root); 1910e0b085b0SJosef Bacik break; 1911e0b085b0SJosef Bacik } 1912e0b085b0SJosef Bacik ASSERT(root->reloc_root == reloc_root); 19133fd0a558SYan, Zheng 19143fd0a558SYan, Zheng /* 19153fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 19163fd0a558SYan, Zheng * knows it should resumes merging 19173fd0a558SYan, Zheng */ 19183fd0a558SYan, Zheng if (!err) 19193fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 1920bbae13f8SJosef Bacik ret = btrfs_update_reloc_root(trans, root); 19213fd0a558SYan, Zheng 1922bbae13f8SJosef Bacik /* 1923bbae13f8SJosef Bacik * Even if we have an error we need this reloc root back on our 1924bbae13f8SJosef Bacik * list so we can clean up properly. 1925bbae13f8SJosef Bacik */ 19263fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 192700246528SJosef Bacik btrfs_put_root(root); 1928bbae13f8SJosef Bacik 1929bbae13f8SJosef Bacik if (ret) { 1930bbae13f8SJosef Bacik btrfs_abort_transaction(trans, ret); 1931bbae13f8SJosef Bacik if (!err) 1932bbae13f8SJosef Bacik err = ret; 1933bbae13f8SJosef Bacik break; 1934bbae13f8SJosef Bacik } 19353fd0a558SYan, Zheng } 19363fd0a558SYan, Zheng 19373fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 19383fd0a558SYan, Zheng 19393fd0a558SYan, Zheng if (!err) 1940fb686c68SJosef Bacik err = btrfs_commit_transaction(trans); 19413fd0a558SYan, Zheng else 19423a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19433fd0a558SYan, Zheng return err; 19443fd0a558SYan, Zheng } 19453fd0a558SYan, Zheng 19463fd0a558SYan, Zheng static noinline_for_stack 1947aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 1948aca1bba6SLiu Bo { 1949a7571232SNikolay Borisov struct btrfs_root *reloc_root, *tmp; 1950aca1bba6SLiu Bo 1951a7571232SNikolay Borisov list_for_each_entry_safe(reloc_root, tmp, list, root_list) 1952bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 1953aca1bba6SLiu Bo } 1954aca1bba6SLiu Bo 1955aca1bba6SLiu Bo static noinline_for_stack 195694404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 19573fd0a558SYan, Zheng { 19580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 19595d4f98a2SYan Zheng struct btrfs_root *root; 19605d4f98a2SYan Zheng struct btrfs_root *reloc_root; 19613fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 19623fd0a558SYan, Zheng int found = 0; 1963aca1bba6SLiu Bo int ret = 0; 19643fd0a558SYan, Zheng again: 19653fd0a558SYan, Zheng root = rc->extent_root; 19667585717fSChris Mason 19677585717fSChris Mason /* 19687585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 19697585717fSChris Mason * we have to make sure nobody is in the middle of 19707585717fSChris Mason * adding their roots to the list while we are 19717585717fSChris Mason * doing this splice 19727585717fSChris Mason */ 19730b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 19743fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 19750b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 19765d4f98a2SYan Zheng 19773fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 19783fd0a558SYan, Zheng found = 1; 19793fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 19803fd0a558SYan, Zheng struct btrfs_root, root_list); 19815d4f98a2SYan Zheng 1982a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1983a820feb5SDavid Sterba false); 19845d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 198524213fa4SJosef Bacik if (IS_ERR(root)) { 198624213fa4SJosef Bacik /* 198724213fa4SJosef Bacik * For recovery we read the fs roots on mount, 198824213fa4SJosef Bacik * and if we didn't find the root then we marked 198924213fa4SJosef Bacik * the reloc root as a garbage root. For normal 199024213fa4SJosef Bacik * relocation obviously the root should exist in 199124213fa4SJosef Bacik * memory. However there's no reason we can't 199224213fa4SJosef Bacik * handle the error properly here just in case. 199324213fa4SJosef Bacik */ 199424213fa4SJosef Bacik ASSERT(0); 199524213fa4SJosef Bacik ret = PTR_ERR(root); 199624213fa4SJosef Bacik goto out; 199724213fa4SJosef Bacik } 199824213fa4SJosef Bacik if (root->reloc_root != reloc_root) { 199924213fa4SJosef Bacik /* 200024213fa4SJosef Bacik * This is actually impossible without something 200124213fa4SJosef Bacik * going really wrong (like weird race condition 200224213fa4SJosef Bacik * or cosmic rays). 200324213fa4SJosef Bacik */ 200424213fa4SJosef Bacik ASSERT(0); 200524213fa4SJosef Bacik ret = -EINVAL; 200624213fa4SJosef Bacik goto out; 200724213fa4SJosef Bacik } 20083fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 200900246528SJosef Bacik btrfs_put_root(root); 2010b37b39cdSJosef Bacik if (ret) { 201125e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 201225e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 201325e293c2SWang Shilong &reloc_roots); 2014aca1bba6SLiu Bo goto out; 2015b37b39cdSJosef Bacik } 20163fd0a558SYan, Zheng } else { 201751415b6cSQu Wenruo if (!IS_ERR(root)) { 201851415b6cSQu Wenruo if (root->reloc_root == reloc_root) { 201951415b6cSQu Wenruo root->reloc_root = NULL; 202051415b6cSQu Wenruo btrfs_put_root(reloc_root); 202151415b6cSQu Wenruo } 20221dae7e0eSQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, 20231dae7e0eSQu Wenruo &root->state); 202451415b6cSQu Wenruo btrfs_put_root(root); 202551415b6cSQu Wenruo } 202651415b6cSQu Wenruo 20273fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 202830d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 202930d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 203030d40577SQu Wenruo &rc->dirty_subvol_roots); 20313fd0a558SYan, Zheng } 20325d4f98a2SYan Zheng } 20335d4f98a2SYan Zheng 20343fd0a558SYan, Zheng if (found) { 20353fd0a558SYan, Zheng found = 0; 20363fd0a558SYan, Zheng goto again; 20375d4f98a2SYan Zheng } 2038aca1bba6SLiu Bo out: 2039aca1bba6SLiu Bo if (ret) { 20400b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2041aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2042467bb1d2SWang Shilong 2043467bb1d2SWang Shilong /* new reloc root may be added */ 20440b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2045467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 20460b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2047467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2048aca1bba6SLiu Bo } 2049aca1bba6SLiu Bo 20507b7b7431SJosef Bacik /* 20517b7b7431SJosef Bacik * We used to have 20527b7b7431SJosef Bacik * 20537b7b7431SJosef Bacik * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 20547b7b7431SJosef Bacik * 20557b7b7431SJosef Bacik * here, but it's wrong. If we fail to start the transaction in 20567b7b7431SJosef Bacik * prepare_to_merge() we will have only 0 ref reloc roots, none of which 20577b7b7431SJosef Bacik * have actually been removed from the reloc_root_tree rb tree. This is 20587b7b7431SJosef Bacik * fine because we're bailing here, and we hold a reference on the root 20597b7b7431SJosef Bacik * for the list that holds it, so these roots will be cleaned up when we 20607b7b7431SJosef Bacik * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root 20617b7b7431SJosef Bacik * will be cleaned up on unmount. 20627b7b7431SJosef Bacik * 20637b7b7431SJosef Bacik * The remaining nodes will be cleaned up by free_reloc_control. 20647b7b7431SJosef Bacik */ 20655d4f98a2SYan Zheng } 20665d4f98a2SYan Zheng 20675d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 20685d4f98a2SYan Zheng { 20695d4f98a2SYan Zheng struct tree_block *block; 20705d4f98a2SYan Zheng struct rb_node *rb_node; 20715d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 20725d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 20735d4f98a2SYan Zheng rb_erase(rb_node, blocks); 20745d4f98a2SYan Zheng kfree(block); 20755d4f98a2SYan Zheng } 20765d4f98a2SYan Zheng } 20775d4f98a2SYan Zheng 20785d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 20795d4f98a2SYan Zheng struct btrfs_root *reloc_root) 20805d4f98a2SYan Zheng { 20810b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 20825d4f98a2SYan Zheng struct btrfs_root *root; 2083442b1ac5SJosef Bacik int ret; 20845d4f98a2SYan Zheng 20855d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 20865d4f98a2SYan Zheng return 0; 20875d4f98a2SYan Zheng 2088a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); 2089404bccbcSJosef Bacik 2090404bccbcSJosef Bacik /* 2091404bccbcSJosef Bacik * This should succeed, since we can't have a reloc root without having 2092404bccbcSJosef Bacik * already looked up the actual root and created the reloc root for this 2093404bccbcSJosef Bacik * root. 2094404bccbcSJosef Bacik * 2095404bccbcSJosef Bacik * However if there's some sort of corruption where we have a ref to a 2096404bccbcSJosef Bacik * reloc root without a corresponding root this could return ENOENT. 2097404bccbcSJosef Bacik */ 2098404bccbcSJosef Bacik if (IS_ERR(root)) { 2099404bccbcSJosef Bacik ASSERT(0); 2100404bccbcSJosef Bacik return PTR_ERR(root); 2101404bccbcSJosef Bacik } 2102404bccbcSJosef Bacik if (root->reloc_root != reloc_root) { 2103404bccbcSJosef Bacik ASSERT(0); 2104404bccbcSJosef Bacik btrfs_err(fs_info, 2105404bccbcSJosef Bacik "root %llu has two reloc roots associated with it", 2106404bccbcSJosef Bacik reloc_root->root_key.offset); 2107404bccbcSJosef Bacik btrfs_put_root(root); 2108404bccbcSJosef Bacik return -EUCLEAN; 2109404bccbcSJosef Bacik } 2110442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 211100246528SJosef Bacik btrfs_put_root(root); 21125d4f98a2SYan Zheng 2113442b1ac5SJosef Bacik return ret; 21145d4f98a2SYan Zheng } 21155d4f98a2SYan Zheng 21163fd0a558SYan, Zheng static noinline_for_stack 21173fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 21183fd0a558SYan, Zheng struct reloc_control *rc, 2119a26195a5SQu Wenruo struct btrfs_backref_node *node, 2120a26195a5SQu Wenruo struct btrfs_backref_edge *edges[]) 21215d4f98a2SYan Zheng { 2122a26195a5SQu Wenruo struct btrfs_backref_node *next; 21235d4f98a2SYan Zheng struct btrfs_root *root; 21243fd0a558SYan, Zheng int index = 0; 212592de551bSJosef Bacik int ret; 21263fd0a558SYan, Zheng 21275d4f98a2SYan Zheng next = node; 21285d4f98a2SYan Zheng while (1) { 21295d4f98a2SYan Zheng cond_resched(); 21305d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 21315d4f98a2SYan Zheng root = next->root; 21328ee66afeSJosef Bacik 21338ee66afeSJosef Bacik /* 21348ee66afeSJosef Bacik * If there is no root, then our references for this block are 21358ee66afeSJosef Bacik * incomplete, as we should be able to walk all the way up to a 21368ee66afeSJosef Bacik * block that is owned by a root. 21378ee66afeSJosef Bacik * 21388ee66afeSJosef Bacik * This path is only for SHAREABLE roots, so if we come upon a 21398ee66afeSJosef Bacik * non-SHAREABLE root then we have backrefs that resolve 21408ee66afeSJosef Bacik * improperly. 21418ee66afeSJosef Bacik * 21428ee66afeSJosef Bacik * Both of these cases indicate file system corruption, or a bug 21438ee66afeSJosef Bacik * in the backref walking code. 21448ee66afeSJosef Bacik */ 21458ee66afeSJosef Bacik if (!root) { 21468ee66afeSJosef Bacik ASSERT(0); 21478ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21488ee66afeSJosef Bacik "bytenr %llu doesn't have a backref path ending in a root", 21498ee66afeSJosef Bacik node->bytenr); 21508ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21518ee66afeSJosef Bacik } 21528ee66afeSJosef Bacik if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 21538ee66afeSJosef Bacik ASSERT(0); 21548ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21558ee66afeSJosef Bacik "bytenr %llu has multiple refs with one ending in a non-shareable root", 21568ee66afeSJosef Bacik node->bytenr); 21578ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21588ee66afeSJosef Bacik } 21595d4f98a2SYan Zheng 21605d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 216192de551bSJosef Bacik ret = record_reloc_root_in_trans(trans, root); 216292de551bSJosef Bacik if (ret) 216392de551bSJosef Bacik return ERR_PTR(ret); 21645d4f98a2SYan Zheng break; 21655d4f98a2SYan Zheng } 21665d4f98a2SYan Zheng 216792de551bSJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 216892de551bSJosef Bacik if (ret) 216992de551bSJosef Bacik return ERR_PTR(ret); 21703fd0a558SYan, Zheng root = root->reloc_root; 21713fd0a558SYan, Zheng 217239200e59SJosef Bacik /* 217339200e59SJosef Bacik * We could have raced with another thread which failed, so 217439200e59SJosef Bacik * root->reloc_root may not be set, return ENOENT in this case. 217539200e59SJosef Bacik */ 217639200e59SJosef Bacik if (!root) 217739200e59SJosef Bacik return ERR_PTR(-ENOENT); 217839200e59SJosef Bacik 21793fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 21808ee66afeSJosef Bacik /* 21818ee66afeSJosef Bacik * We just created the reloc root, so we shouldn't have 21828ee66afeSJosef Bacik * ->new_bytenr set and this shouldn't be in the changed 21838ee66afeSJosef Bacik * list. If it is then we have multiple roots pointing 21848ee66afeSJosef Bacik * at the same bytenr which indicates corruption, or 21858ee66afeSJosef Bacik * we've made a mistake in the backref walking code. 21868ee66afeSJosef Bacik */ 21878ee66afeSJosef Bacik ASSERT(next->new_bytenr == 0); 21888ee66afeSJosef Bacik ASSERT(list_empty(&next->list)); 21898ee66afeSJosef Bacik if (next->new_bytenr || !list_empty(&next->list)) { 21908ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21918ee66afeSJosef Bacik "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu", 21928ee66afeSJosef Bacik node->bytenr, next->bytenr); 21938ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21948ee66afeSJosef Bacik } 21958ee66afeSJosef Bacik 21963fd0a558SYan, Zheng next->new_bytenr = root->node->start; 219700246528SJosef Bacik btrfs_put_root(next->root); 219800246528SJosef Bacik next->root = btrfs_grab_root(root); 21990b530bc5SJosef Bacik ASSERT(next->root); 22003fd0a558SYan, Zheng list_add_tail(&next->list, 22013fd0a558SYan, Zheng &rc->backref_cache.changed); 22029569cc20SQu Wenruo mark_block_processed(rc, next); 22035d4f98a2SYan Zheng break; 22045d4f98a2SYan Zheng } 22055d4f98a2SYan Zheng 22063fd0a558SYan, Zheng WARN_ON(1); 22075d4f98a2SYan Zheng root = NULL; 22085d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 22095d4f98a2SYan Zheng if (!next || next->level <= node->level) 22105d4f98a2SYan Zheng break; 22115d4f98a2SYan Zheng } 2212cbdc2ebcSJosef Bacik if (!root) { 2213cbdc2ebcSJosef Bacik /* 2214cbdc2ebcSJosef Bacik * This can happen if there's fs corruption or if there's a bug 2215cbdc2ebcSJosef Bacik * in the backref lookup code. 2216cbdc2ebcSJosef Bacik */ 2217cbdc2ebcSJosef Bacik ASSERT(0); 2218cbdc2ebcSJosef Bacik return ERR_PTR(-ENOENT); 2219cbdc2ebcSJosef Bacik } 22205d4f98a2SYan Zheng 22213fd0a558SYan, Zheng next = node; 22223fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 22233fd0a558SYan, Zheng while (1) { 22243fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 22253fd0a558SYan, Zheng if (--index < 0) 22263fd0a558SYan, Zheng break; 22273fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 22283fd0a558SYan, Zheng } 22295d4f98a2SYan Zheng return root; 22305d4f98a2SYan Zheng } 22315d4f98a2SYan Zheng 22323fd0a558SYan, Zheng /* 223392a7cc42SQu Wenruo * Select a tree root for relocation. 223492a7cc42SQu Wenruo * 223592a7cc42SQu Wenruo * Return NULL if the block is not shareable. We should use do_relocation() in 223692a7cc42SQu Wenruo * this case. 223792a7cc42SQu Wenruo * 223892a7cc42SQu Wenruo * Return a tree root pointer if the block is shareable. 223992a7cc42SQu Wenruo * Return -ENOENT if the block is root of reloc tree. 22403fd0a558SYan, Zheng */ 22415d4f98a2SYan Zheng static noinline_for_stack 2242a26195a5SQu Wenruo struct btrfs_root *select_one_root(struct btrfs_backref_node *node) 22435d4f98a2SYan Zheng { 2244a26195a5SQu Wenruo struct btrfs_backref_node *next; 22453fd0a558SYan, Zheng struct btrfs_root *root; 22463fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 2247a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22483fd0a558SYan, Zheng int index = 0; 22493fd0a558SYan, Zheng 22503fd0a558SYan, Zheng next = node; 22513fd0a558SYan, Zheng while (1) { 22523fd0a558SYan, Zheng cond_resched(); 22533fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 22543fd0a558SYan, Zheng root = next->root; 22558717cf44SJosef Bacik 22568717cf44SJosef Bacik /* 22578717cf44SJosef Bacik * This can occur if we have incomplete extent refs leading all 22588717cf44SJosef Bacik * the way up a particular path, in this case return -EUCLEAN. 22598717cf44SJosef Bacik */ 22608717cf44SJosef Bacik if (!root) 22618717cf44SJosef Bacik return ERR_PTR(-EUCLEAN); 22623fd0a558SYan, Zheng 226392a7cc42SQu Wenruo /* No other choice for non-shareable tree */ 226492a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 22653fd0a558SYan, Zheng return root; 22663fd0a558SYan, Zheng 22673fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 22683fd0a558SYan, Zheng fs_root = root; 22693fd0a558SYan, Zheng 22703fd0a558SYan, Zheng if (next != node) 22713fd0a558SYan, Zheng return NULL; 22723fd0a558SYan, Zheng 22733fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 22743fd0a558SYan, Zheng if (!next || next->level <= node->level) 22753fd0a558SYan, Zheng break; 22763fd0a558SYan, Zheng } 22773fd0a558SYan, Zheng 22783fd0a558SYan, Zheng if (!fs_root) 22793fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 22803fd0a558SYan, Zheng return fs_root; 22815d4f98a2SYan Zheng } 22825d4f98a2SYan Zheng 22835d4f98a2SYan Zheng static noinline_for_stack 22843fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 2285a26195a5SQu Wenruo struct btrfs_backref_node *node, int reserve) 22865d4f98a2SYan Zheng { 22870b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2288a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2289a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2290a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22913fd0a558SYan, Zheng u64 num_bytes = 0; 22923fd0a558SYan, Zheng int index = 0; 22935d4f98a2SYan Zheng 22943fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 22953fd0a558SYan, Zheng 22963fd0a558SYan, Zheng while (next) { 22973fd0a558SYan, Zheng cond_resched(); 22985d4f98a2SYan Zheng while (1) { 22993fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 23005d4f98a2SYan Zheng break; 23015d4f98a2SYan Zheng 23020b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 23033fd0a558SYan, Zheng 23043fd0a558SYan, Zheng if (list_empty(&next->upper)) 23053fd0a558SYan, Zheng break; 23063fd0a558SYan, Zheng 23073fd0a558SYan, Zheng edge = list_entry(next->upper.next, 2308a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 23093fd0a558SYan, Zheng edges[index++] = edge; 23103fd0a558SYan, Zheng next = edge->node[UPPER]; 23115d4f98a2SYan Zheng } 23123fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 23133fd0a558SYan, Zheng } 23143fd0a558SYan, Zheng return num_bytes; 23153fd0a558SYan, Zheng } 23163fd0a558SYan, Zheng 23173fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 23183fd0a558SYan, Zheng struct reloc_control *rc, 2319a26195a5SQu Wenruo struct btrfs_backref_node *node) 23203fd0a558SYan, Zheng { 23213fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2322da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23233fd0a558SYan, Zheng u64 num_bytes; 23243fd0a558SYan, Zheng int ret; 23250647bf56SWang Shilong u64 tmp; 23263fd0a558SYan, Zheng 23273fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 23283fd0a558SYan, Zheng 23293fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 23300647bf56SWang Shilong rc->reserved_bytes += num_bytes; 23318ca17f0fSJosef Bacik 23328ca17f0fSJosef Bacik /* 23338ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 23348ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 23358ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 23368ca17f0fSJosef Bacik */ 23379270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes, 23388ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 23393fd0a558SYan, Zheng if (ret) { 2340da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 23410647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 23420647bf56SWang Shilong tmp <<= 1; 23430647bf56SWang Shilong /* 23440647bf56SWang Shilong * only one thread can access block_rsv at this point, 23450647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 23460647bf56SWang Shilong * we expand more reservation size here to allow enough 234752042d8eSAndrea Gelmini * space for relocation and we will return earlier in 23480647bf56SWang Shilong * enospc case. 23490647bf56SWang Shilong */ 2350da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 23510647bf56SWang Shilong RELOCATION_RESERVED_NODES; 23528ca17f0fSJosef Bacik return -EAGAIN; 23533fd0a558SYan, Zheng } 23543fd0a558SYan, Zheng 23553fd0a558SYan, Zheng return 0; 23563fd0a558SYan, Zheng } 23573fd0a558SYan, Zheng 23585d4f98a2SYan Zheng /* 23595d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 23605d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 23615d4f98a2SYan Zheng * 23625d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 23635d4f98a2SYan Zheng * in that case this function just updates pointers. 23645d4f98a2SYan Zheng */ 23655d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 23663fd0a558SYan, Zheng struct reloc_control *rc, 2367a26195a5SQu Wenruo struct btrfs_backref_node *node, 23685d4f98a2SYan Zheng struct btrfs_key *key, 23695d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 23705d4f98a2SYan Zheng { 2371a26195a5SQu Wenruo struct btrfs_backref_node *upper; 2372a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2373a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 23745d4f98a2SYan Zheng struct btrfs_root *root; 23755d4f98a2SYan Zheng struct extent_buffer *eb; 23765d4f98a2SYan Zheng u32 blocksize; 23775d4f98a2SYan Zheng u64 bytenr; 23785d4f98a2SYan Zheng int slot; 23798df01fddSNikolay Borisov int ret = 0; 23805d4f98a2SYan Zheng 2381ffe30dd8SJosef Bacik /* 2382ffe30dd8SJosef Bacik * If we are lowest then this is the first time we're processing this 2383ffe30dd8SJosef Bacik * block, and thus shouldn't have an eb associated with it yet. 2384ffe30dd8SJosef Bacik */ 2385ffe30dd8SJosef Bacik ASSERT(!lowest || !node->eb); 23865d4f98a2SYan Zheng 23875d4f98a2SYan Zheng path->lowest_level = node->level + 1; 23883fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 23895d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 239082fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2391581c1760SQu Wenruo 23925d4f98a2SYan Zheng cond_resched(); 23935d4f98a2SYan Zheng 23945d4f98a2SYan Zheng upper = edge->node[UPPER]; 2395dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 2396cbdc2ebcSJosef Bacik if (IS_ERR(root)) { 2397cbdc2ebcSJosef Bacik ret = PTR_ERR(root); 2398cbdc2ebcSJosef Bacik goto next; 2399cbdc2ebcSJosef Bacik } 24005d4f98a2SYan Zheng 24013fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 24023fd0a558SYan, Zheng if (!lowest) { 2403e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 24048df01fddSNikolay Borisov if (ret < 0) 2405cbca7d59SFilipe Manana goto next; 24063fd0a558SYan, Zheng BUG_ON(ret); 24073fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24083fd0a558SYan, Zheng if (node->eb->start == bytenr) 24093fd0a558SYan, Zheng goto next; 24103fd0a558SYan, Zheng } 2411b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 24123fd0a558SYan, Zheng } 24135d4f98a2SYan Zheng 24145d4f98a2SYan Zheng if (!upper->eb) { 24155d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 24163561b9dbSLiu Bo if (ret) { 24178df01fddSNikolay Borisov if (ret > 0) 24188df01fddSNikolay Borisov ret = -ENOENT; 24193561b9dbSLiu Bo 24203561b9dbSLiu Bo btrfs_release_path(path); 24215d4f98a2SYan Zheng break; 24225d4f98a2SYan Zheng } 24235d4f98a2SYan Zheng 24243fd0a558SYan, Zheng if (!upper->eb) { 24253fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 24263fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 24273fd0a558SYan, Zheng } else { 24283fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 24293fd0a558SYan, Zheng } 24303fd0a558SYan, Zheng 24313fd0a558SYan, Zheng upper->locked = 1; 24323fd0a558SYan, Zheng path->locks[upper->level] = 0; 24333fd0a558SYan, Zheng 24345d4f98a2SYan Zheng slot = path->slots[upper->level]; 2435b3b4aa74SDavid Sterba btrfs_release_path(path); 24365d4f98a2SYan Zheng } else { 2437e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 24388df01fddSNikolay Borisov if (ret < 0) 2439cbca7d59SFilipe Manana goto next; 24405d4f98a2SYan Zheng BUG_ON(ret); 24415d4f98a2SYan Zheng } 24425d4f98a2SYan Zheng 24435d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24443fd0a558SYan, Zheng if (lowest) { 24454547f4d8SLiu Bo if (bytenr != node->bytenr) { 24464547f4d8SLiu Bo btrfs_err(root->fs_info, 24474547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 24484547f4d8SLiu Bo bytenr, node->bytenr, slot, 24494547f4d8SLiu Bo upper->eb->start); 24508df01fddSNikolay Borisov ret = -EIO; 24514547f4d8SLiu Bo goto next; 24524547f4d8SLiu Bo } 24535d4f98a2SYan Zheng } else { 24543fd0a558SYan, Zheng if (node->eb->start == bytenr) 24553fd0a558SYan, Zheng goto next; 24565d4f98a2SYan Zheng } 24575d4f98a2SYan Zheng 2458da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 2459c9752536SJosef Bacik eb = btrfs_read_node_slot(upper->eb, slot); 246064c043deSLiu Bo if (IS_ERR(eb)) { 24618df01fddSNikolay Borisov ret = PTR_ERR(eb); 246264c043deSLiu Bo goto next; 246397d9a8a4STsutomu Itoh } 24645d4f98a2SYan Zheng btrfs_tree_lock(eb); 24655d4f98a2SYan Zheng 24665d4f98a2SYan Zheng if (!node->eb) { 24675d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 24689631e4ccSJosef Bacik slot, &eb, BTRFS_NESTING_COW); 24693fd0a558SYan, Zheng btrfs_tree_unlock(eb); 24703fd0a558SYan, Zheng free_extent_buffer(eb); 24718df01fddSNikolay Borisov if (ret < 0) 24723fd0a558SYan, Zheng goto next; 2473ffe30dd8SJosef Bacik /* 2474ffe30dd8SJosef Bacik * We've just COWed this block, it should have updated 2475ffe30dd8SJosef Bacik * the correct backref node entry. 2476ffe30dd8SJosef Bacik */ 2477ffe30dd8SJosef Bacik ASSERT(node->eb == eb); 24785d4f98a2SYan Zheng } else { 24795d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 24805d4f98a2SYan Zheng node->eb->start); 24815d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 24825d4f98a2SYan Zheng trans->transid); 24835d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 24845d4f98a2SYan Zheng 248582fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 24865d4f98a2SYan Zheng node->eb->start, blocksize, 248782fa113fSQu Wenruo upper->eb->start); 248882fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 2489f42c5da6SNikolay Borisov btrfs_header_owner(upper->eb), 2490f42c5da6SNikolay Borisov root->root_key.objectid, false); 249182fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 2492eb6b7fb4SJosef Bacik if (!ret) 2493eb6b7fb4SJosef Bacik ret = btrfs_drop_subtree(trans, root, eb, 2494eb6b7fb4SJosef Bacik upper->eb); 2495eb6b7fb4SJosef Bacik if (ret) 2496eb6b7fb4SJosef Bacik btrfs_abort_transaction(trans, ret); 24975d4f98a2SYan Zheng } 24983fd0a558SYan, Zheng next: 24993fd0a558SYan, Zheng if (!upper->pending) 2500b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 25013fd0a558SYan, Zheng else 2502b0fe7078SQu Wenruo btrfs_backref_unlock_node_buffer(upper); 25038df01fddSNikolay Borisov if (ret) 25043fd0a558SYan, Zheng break; 25055d4f98a2SYan Zheng } 25063fd0a558SYan, Zheng 25078df01fddSNikolay Borisov if (!ret && node->pending) { 2508b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 25093fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 25103fd0a558SYan, Zheng node->pending = 0; 25115d4f98a2SYan Zheng } 25123fd0a558SYan, Zheng 25135d4f98a2SYan Zheng path->lowest_level = 0; 2514ffe30dd8SJosef Bacik 2515ffe30dd8SJosef Bacik /* 2516ffe30dd8SJosef Bacik * We should have allocated all of our space in the block rsv and thus 2517ffe30dd8SJosef Bacik * shouldn't ENOSPC. 2518ffe30dd8SJosef Bacik */ 2519ffe30dd8SJosef Bacik ASSERT(ret != -ENOSPC); 25208df01fddSNikolay Borisov return ret; 25215d4f98a2SYan Zheng } 25225d4f98a2SYan Zheng 25235d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 25243fd0a558SYan, Zheng struct reloc_control *rc, 2525a26195a5SQu Wenruo struct btrfs_backref_node *node, 25265d4f98a2SYan Zheng struct btrfs_path *path) 25275d4f98a2SYan Zheng { 25285d4f98a2SYan Zheng struct btrfs_key key; 25295d4f98a2SYan Zheng 25305d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 25313fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 25325d4f98a2SYan Zheng } 25335d4f98a2SYan Zheng 25345d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 25353fd0a558SYan, Zheng struct reloc_control *rc, 25363fd0a558SYan, Zheng struct btrfs_path *path, int err) 25375d4f98a2SYan Zheng { 25383fd0a558SYan, Zheng LIST_HEAD(list); 2539a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 2540a26195a5SQu Wenruo struct btrfs_backref_node *node; 25415d4f98a2SYan Zheng int level; 25425d4f98a2SYan Zheng int ret; 25435d4f98a2SYan Zheng 25445d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 25455d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 25465d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 2547a26195a5SQu Wenruo struct btrfs_backref_node, list); 25483fd0a558SYan, Zheng list_move_tail(&node->list, &list); 25493fd0a558SYan, Zheng BUG_ON(!node->pending); 25505d4f98a2SYan Zheng 25513fd0a558SYan, Zheng if (!err) { 25523fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 25535d4f98a2SYan Zheng if (ret < 0) 25545d4f98a2SYan Zheng err = ret; 25555d4f98a2SYan Zheng } 25565d4f98a2SYan Zheng } 25573fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 25583fd0a558SYan, Zheng } 25595d4f98a2SYan Zheng return err; 25605d4f98a2SYan Zheng } 25615d4f98a2SYan Zheng 25625d4f98a2SYan Zheng /* 25635d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 25645d4f98a2SYan Zheng * as processed. 25655d4f98a2SYan Zheng */ 25665d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 2567a26195a5SQu Wenruo struct btrfs_backref_node *node) 25685d4f98a2SYan Zheng { 2569a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2570a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2571a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25725d4f98a2SYan Zheng int index = 0; 25735d4f98a2SYan Zheng 25745d4f98a2SYan Zheng while (next) { 25755d4f98a2SYan Zheng cond_resched(); 25765d4f98a2SYan Zheng while (1) { 25775d4f98a2SYan Zheng if (next->processed) 25785d4f98a2SYan Zheng break; 25795d4f98a2SYan Zheng 25809569cc20SQu Wenruo mark_block_processed(rc, next); 25815d4f98a2SYan Zheng 25825d4f98a2SYan Zheng if (list_empty(&next->upper)) 25835d4f98a2SYan Zheng break; 25845d4f98a2SYan Zheng 25855d4f98a2SYan Zheng edge = list_entry(next->upper.next, 2586a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 25875d4f98a2SYan Zheng edges[index++] = edge; 25885d4f98a2SYan Zheng next = edge->node[UPPER]; 25895d4f98a2SYan Zheng } 25905d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25915d4f98a2SYan Zheng } 25925d4f98a2SYan Zheng } 25935d4f98a2SYan Zheng 25947476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 25955d4f98a2SYan Zheng { 2596da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 25977476dfdaSDavid Sterba 25985d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 25999655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 26005d4f98a2SYan Zheng return 1; 26015d4f98a2SYan Zheng return 0; 26025d4f98a2SYan Zheng } 26035d4f98a2SYan Zheng 26042ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 26055d4f98a2SYan Zheng struct tree_block *block) 26065d4f98a2SYan Zheng { 26075d4f98a2SYan Zheng struct extent_buffer *eb; 26085d4f98a2SYan Zheng 2609f7ba2d37SJosef Bacik eb = read_tree_block(fs_info, block->bytenr, block->owner, 2610f7ba2d37SJosef Bacik block->key.offset, block->level, NULL); 26114eb150d6SQu Wenruo if (IS_ERR(eb)) 261264c043deSLiu Bo return PTR_ERR(eb); 26134eb150d6SQu Wenruo if (!extent_buffer_uptodate(eb)) { 2614416bc658SJosef Bacik free_extent_buffer(eb); 2615416bc658SJosef Bacik return -EIO; 2616416bc658SJosef Bacik } 26175d4f98a2SYan Zheng if (block->level == 0) 26185d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 26195d4f98a2SYan Zheng else 26205d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 26215d4f98a2SYan Zheng free_extent_buffer(eb); 26225d4f98a2SYan Zheng block->key_ready = 1; 26235d4f98a2SYan Zheng return 0; 26245d4f98a2SYan Zheng } 26255d4f98a2SYan Zheng 26265d4f98a2SYan Zheng /* 26275d4f98a2SYan Zheng * helper function to relocate a tree block 26285d4f98a2SYan Zheng */ 26295d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 26305d4f98a2SYan Zheng struct reloc_control *rc, 2631a26195a5SQu Wenruo struct btrfs_backref_node *node, 26325d4f98a2SYan Zheng struct btrfs_key *key, 26335d4f98a2SYan Zheng struct btrfs_path *path) 26345d4f98a2SYan Zheng { 26355d4f98a2SYan Zheng struct btrfs_root *root; 26363fd0a558SYan, Zheng int ret = 0; 26375d4f98a2SYan Zheng 26383fd0a558SYan, Zheng if (!node) 26395d4f98a2SYan Zheng return 0; 26403fd0a558SYan, Zheng 26415f6b2e5cSJosef Bacik /* 26425f6b2e5cSJosef Bacik * If we fail here we want to drop our backref_node because we are going 26435f6b2e5cSJosef Bacik * to start over and regenerate the tree for it. 26445f6b2e5cSJosef Bacik */ 26455f6b2e5cSJosef Bacik ret = reserve_metadata_space(trans, rc, node); 26465f6b2e5cSJosef Bacik if (ret) 26475f6b2e5cSJosef Bacik goto out; 26485f6b2e5cSJosef Bacik 26493fd0a558SYan, Zheng BUG_ON(node->processed); 2650147d256eSZhaolei root = select_one_root(node); 26518717cf44SJosef Bacik if (IS_ERR(root)) { 26528717cf44SJosef Bacik ret = PTR_ERR(root); 26538717cf44SJosef Bacik 26548717cf44SJosef Bacik /* See explanation in select_one_root for the -EUCLEAN case. */ 26558717cf44SJosef Bacik ASSERT(ret == -ENOENT); 26568717cf44SJosef Bacik if (ret == -ENOENT) { 26578717cf44SJosef Bacik ret = 0; 26583fd0a558SYan, Zheng update_processed_blocks(rc, node); 26598717cf44SJosef Bacik } 26603fd0a558SYan, Zheng goto out; 26615d4f98a2SYan Zheng } 26625d4f98a2SYan Zheng 26633fd0a558SYan, Zheng if (root) { 266492a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 26651c7bfa15SJosef Bacik /* 26661c7bfa15SJosef Bacik * This block was the root block of a root, and this is 26671c7bfa15SJosef Bacik * the first time we're processing the block and thus it 26681c7bfa15SJosef Bacik * should not have had the ->new_bytenr modified and 26691c7bfa15SJosef Bacik * should have not been included on the changed list. 26701c7bfa15SJosef Bacik * 26711c7bfa15SJosef Bacik * However in the case of corruption we could have 26721c7bfa15SJosef Bacik * multiple refs pointing to the same block improperly, 26731c7bfa15SJosef Bacik * and thus we would trip over these checks. ASSERT() 26741c7bfa15SJosef Bacik * for the developer case, because it could indicate a 26751c7bfa15SJosef Bacik * bug in the backref code, however error out for a 26761c7bfa15SJosef Bacik * normal user in the case of corruption. 26771c7bfa15SJosef Bacik */ 26781c7bfa15SJosef Bacik ASSERT(node->new_bytenr == 0); 26791c7bfa15SJosef Bacik ASSERT(list_empty(&node->list)); 26801c7bfa15SJosef Bacik if (node->new_bytenr || !list_empty(&node->list)) { 26811c7bfa15SJosef Bacik btrfs_err(root->fs_info, 26821c7bfa15SJosef Bacik "bytenr %llu has improper references to it", 26831c7bfa15SJosef Bacik node->bytenr); 26841c7bfa15SJosef Bacik ret = -EUCLEAN; 26851c7bfa15SJosef Bacik goto out; 26861c7bfa15SJosef Bacik } 2687d18c7bd9SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 2688d18c7bd9SJosef Bacik if (ret) 2689d18c7bd9SJosef Bacik goto out; 269039200e59SJosef Bacik /* 269139200e59SJosef Bacik * Another thread could have failed, need to check if we 269239200e59SJosef Bacik * have reloc_root actually set. 269339200e59SJosef Bacik */ 269439200e59SJosef Bacik if (!root->reloc_root) { 269539200e59SJosef Bacik ret = -ENOENT; 269639200e59SJosef Bacik goto out; 269739200e59SJosef Bacik } 26983fd0a558SYan, Zheng root = root->reloc_root; 26993fd0a558SYan, Zheng node->new_bytenr = root->node->start; 270000246528SJosef Bacik btrfs_put_root(node->root); 270100246528SJosef Bacik node->root = btrfs_grab_root(root); 27020b530bc5SJosef Bacik ASSERT(node->root); 27033fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 27043fd0a558SYan, Zheng } else { 27055d4f98a2SYan Zheng path->lowest_level = node->level; 27062bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 27072bb2e00eSFilipe Manana btrfs_reserve_chunk_metadata(trans, false); 27085d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2709b3b4aa74SDavid Sterba btrfs_release_path(path); 27102bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 27112bb2e00eSFilipe Manana btrfs_trans_release_chunk_metadata(trans); 27123fd0a558SYan, Zheng if (ret > 0) 27135d4f98a2SYan Zheng ret = 0; 27143fd0a558SYan, Zheng } 27153fd0a558SYan, Zheng if (!ret) 27163fd0a558SYan, Zheng update_processed_blocks(rc, node); 27173fd0a558SYan, Zheng } else { 27183fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 27193fd0a558SYan, Zheng } 27205d4f98a2SYan Zheng out: 27210647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 2722023acb07SQu Wenruo btrfs_backref_cleanup_node(&rc->backref_cache, node); 27235d4f98a2SYan Zheng return ret; 27245d4f98a2SYan Zheng } 27255d4f98a2SYan Zheng 27265d4f98a2SYan Zheng /* 27275d4f98a2SYan Zheng * relocate a list of blocks 27285d4f98a2SYan Zheng */ 27295d4f98a2SYan Zheng static noinline_for_stack 27305d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 27315d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 27325d4f98a2SYan Zheng { 27332ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2734a26195a5SQu Wenruo struct btrfs_backref_node *node; 27355d4f98a2SYan Zheng struct btrfs_path *path; 27365d4f98a2SYan Zheng struct tree_block *block; 273798ff7b94SQu Wenruo struct tree_block *next; 27385d4f98a2SYan Zheng int ret; 27395d4f98a2SYan Zheng int err = 0; 27405d4f98a2SYan Zheng 27415d4f98a2SYan Zheng path = btrfs_alloc_path(); 2742e1a12670SLiu Bo if (!path) { 2743e1a12670SLiu Bo err = -ENOMEM; 274434c2b290SDavid Sterba goto out_free_blocks; 2745e1a12670SLiu Bo } 27465d4f98a2SYan Zheng 274798ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 274898ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27495d4f98a2SYan Zheng if (!block->key_ready) 2750f7ba2d37SJosef Bacik btrfs_readahead_tree_block(fs_info, block->bytenr, 2751f7ba2d37SJosef Bacik block->owner, 0, 27523fbaf258SJosef Bacik block->level); 27535d4f98a2SYan Zheng } 27545d4f98a2SYan Zheng 275598ff7b94SQu Wenruo /* Get first keys */ 275698ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 275734c2b290SDavid Sterba if (!block->key_ready) { 27582ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 275934c2b290SDavid Sterba if (err) 276034c2b290SDavid Sterba goto out_free_path; 276134c2b290SDavid Sterba } 27625d4f98a2SYan Zheng } 27635d4f98a2SYan Zheng 276498ff7b94SQu Wenruo /* Do tree relocation */ 276598ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27663fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 27675d4f98a2SYan Zheng block->level, block->bytenr); 27685d4f98a2SYan Zheng if (IS_ERR(node)) { 27695d4f98a2SYan Zheng err = PTR_ERR(node); 27705d4f98a2SYan Zheng goto out; 27715d4f98a2SYan Zheng } 27725d4f98a2SYan Zheng 27735d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 27745d4f98a2SYan Zheng path); 27755d4f98a2SYan Zheng if (ret < 0) { 27765d4f98a2SYan Zheng err = ret; 277750dbbb71SJosef Bacik break; 27785d4f98a2SYan Zheng } 27795d4f98a2SYan Zheng } 27805d4f98a2SYan Zheng out: 27813fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 27825d4f98a2SYan Zheng 278334c2b290SDavid Sterba out_free_path: 27845d4f98a2SYan Zheng btrfs_free_path(path); 278534c2b290SDavid Sterba out_free_blocks: 2786e1a12670SLiu Bo free_block_list(blocks); 27875d4f98a2SYan Zheng return err; 27885d4f98a2SYan Zheng } 27895d4f98a2SYan Zheng 2790056d9becSNikolay Borisov static noinline_for_stack int prealloc_file_extent_cluster( 2791056d9becSNikolay Borisov struct btrfs_inode *inode, 2792efa56464SYan, Zheng struct file_extent_cluster *cluster) 2793efa56464SYan, Zheng { 2794efa56464SYan, Zheng u64 alloc_hint = 0; 2795efa56464SYan, Zheng u64 start; 2796efa56464SYan, Zheng u64 end; 2797056d9becSNikolay Borisov u64 offset = inode->index_cnt; 2798efa56464SYan, Zheng u64 num_bytes; 27994e9d0d01SNikolay Borisov int nr; 2800efa56464SYan, Zheng int ret = 0; 28019d9ea1e6SQu Wenruo u64 i_size = i_size_read(&inode->vfs_inode); 2802dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 2803dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 2804214e61d0SNikolay Borisov u64 cur_offset = prealloc_start; 2805efa56464SYan, Zheng 28069d9ea1e6SQu Wenruo /* 28079d9ea1e6SQu Wenruo * For subpage case, previous i_size may not be aligned to PAGE_SIZE. 28089d9ea1e6SQu Wenruo * This means the range [i_size, PAGE_END + 1) is filled with zeros by 28099d9ea1e6SQu Wenruo * btrfs_do_readpage() call of previously relocated file cluster. 28109d9ea1e6SQu Wenruo * 28119d9ea1e6SQu Wenruo * If the current cluster starts in the above range, btrfs_do_readpage() 28129d9ea1e6SQu Wenruo * will skip the read, and relocate_one_page() will later writeback 28139d9ea1e6SQu Wenruo * the padding zeros as new data, causing data corruption. 28149d9ea1e6SQu Wenruo * 28159d9ea1e6SQu Wenruo * Here we have to manually invalidate the range (i_size, PAGE_END + 1). 28169d9ea1e6SQu Wenruo */ 28179d9ea1e6SQu Wenruo if (!IS_ALIGNED(i_size, PAGE_SIZE)) { 28189d9ea1e6SQu Wenruo struct address_space *mapping = inode->vfs_inode.i_mapping; 28199d9ea1e6SQu Wenruo struct btrfs_fs_info *fs_info = inode->root->fs_info; 28209d9ea1e6SQu Wenruo const u32 sectorsize = fs_info->sectorsize; 28219d9ea1e6SQu Wenruo struct page *page; 28229d9ea1e6SQu Wenruo 28239d9ea1e6SQu Wenruo ASSERT(sectorsize < PAGE_SIZE); 28249d9ea1e6SQu Wenruo ASSERT(IS_ALIGNED(i_size, sectorsize)); 28259d9ea1e6SQu Wenruo 28269d9ea1e6SQu Wenruo /* 28279d9ea1e6SQu Wenruo * Subpage can't handle page with DIRTY but without UPTODATE 28289d9ea1e6SQu Wenruo * bit as it can lead to the following deadlock: 28299d9ea1e6SQu Wenruo * 2830fb12489bSMatthew Wilcox (Oracle) * btrfs_read_folio() 28319d9ea1e6SQu Wenruo * | Page already *locked* 28329d9ea1e6SQu Wenruo * |- btrfs_lock_and_flush_ordered_range() 28339d9ea1e6SQu Wenruo * |- btrfs_start_ordered_extent() 28349d9ea1e6SQu Wenruo * |- extent_write_cache_pages() 28359d9ea1e6SQu Wenruo * |- lock_page() 28369d9ea1e6SQu Wenruo * We try to lock the page we already hold. 28379d9ea1e6SQu Wenruo * 28389d9ea1e6SQu Wenruo * Here we just writeback the whole data reloc inode, so that 28399d9ea1e6SQu Wenruo * we will be ensured to have no dirty range in the page, and 28409d9ea1e6SQu Wenruo * are safe to clear the uptodate bits. 28419d9ea1e6SQu Wenruo * 28429d9ea1e6SQu Wenruo * This shouldn't cause too much overhead, as we need to write 28439d9ea1e6SQu Wenruo * the data back anyway. 28449d9ea1e6SQu Wenruo */ 28459d9ea1e6SQu Wenruo ret = filemap_write_and_wait(mapping); 28469d9ea1e6SQu Wenruo if (ret < 0) 28479d9ea1e6SQu Wenruo return ret; 28489d9ea1e6SQu Wenruo 28499d9ea1e6SQu Wenruo clear_extent_bits(&inode->io_tree, i_size, 28509d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - 1, 28519d9ea1e6SQu Wenruo EXTENT_UPTODATE); 28529d9ea1e6SQu Wenruo page = find_lock_page(mapping, i_size >> PAGE_SHIFT); 28539d9ea1e6SQu Wenruo /* 28549d9ea1e6SQu Wenruo * If page is freed we don't need to do anything then, as we 28559d9ea1e6SQu Wenruo * will re-read the whole page anyway. 28569d9ea1e6SQu Wenruo */ 28579d9ea1e6SQu Wenruo if (page) { 28589d9ea1e6SQu Wenruo btrfs_subpage_clear_uptodate(fs_info, page, i_size, 28599d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - i_size); 28609d9ea1e6SQu Wenruo unlock_page(page); 28619d9ea1e6SQu Wenruo put_page(page); 28629d9ea1e6SQu Wenruo } 28639d9ea1e6SQu Wenruo } 28649d9ea1e6SQu Wenruo 2865efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 2866056d9becSNikolay Borisov ret = btrfs_alloc_data_chunk_ondemand(inode, 2867dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 2868efa56464SYan, Zheng if (ret) 2869214e61d0SNikolay Borisov return ret; 2870efa56464SYan, Zheng 287164708539SJosef Bacik btrfs_inode_lock(&inode->vfs_inode, 0); 28724e9d0d01SNikolay Borisov for (nr = 0; nr < cluster->nr; nr++) { 28739c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 28749c5c9604SJosef Bacik 2875efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 2876efa56464SYan, Zheng if (nr + 1 < cluster->nr) 2877efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 2878efa56464SYan, Zheng else 2879efa56464SYan, Zheng end = cluster->end - offset; 2880efa56464SYan, Zheng 28819c5c9604SJosef Bacik lock_extent(&inode->io_tree, start, end, &cached_state); 2882efa56464SYan, Zheng num_bytes = end + 1 - start; 2883056d9becSNikolay Borisov ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start, 2884efa56464SYan, Zheng num_bytes, num_bytes, 2885efa56464SYan, Zheng end + 1, &alloc_hint); 288618513091SWang Xiaoguang cur_offset = end + 1; 28879c5c9604SJosef Bacik unlock_extent(&inode->io_tree, start, end, &cached_state); 2888efa56464SYan, Zheng if (ret) 2889efa56464SYan, Zheng break; 2890efa56464SYan, Zheng } 289164708539SJosef Bacik btrfs_inode_unlock(&inode->vfs_inode, 0); 2892214e61d0SNikolay Borisov 289318513091SWang Xiaoguang if (cur_offset < prealloc_end) 2894056d9becSNikolay Borisov btrfs_free_reserved_data_space_noquota(inode->root->fs_info, 2895a89ef455SFilipe Manana prealloc_end + 1 - cur_offset); 2896efa56464SYan, Zheng return ret; 2897efa56464SYan, Zheng } 2898efa56464SYan, Zheng 28994b01c44fSJohannes Thumshirn static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode, 29004b01c44fSJohannes Thumshirn u64 start, u64 end, u64 block_start) 29015d4f98a2SYan Zheng { 29025d4f98a2SYan Zheng struct extent_map *em; 29039c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 29040257bb82SYan, Zheng int ret = 0; 29055d4f98a2SYan Zheng 2906172ddd60SDavid Sterba em = alloc_extent_map(); 29070257bb82SYan, Zheng if (!em) 29080257bb82SYan, Zheng return -ENOMEM; 29090257bb82SYan, Zheng 29105d4f98a2SYan Zheng em->start = start; 29110257bb82SYan, Zheng em->len = end + 1 - start; 29120257bb82SYan, Zheng em->block_len = em->len; 29130257bb82SYan, Zheng em->block_start = block_start; 29145d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 29155d4f98a2SYan Zheng 29169c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 2917a1ba4c08SFilipe Manana ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, false); 29189c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 2919a1ba4c08SFilipe Manana free_extent_map(em); 2920a1ba4c08SFilipe Manana 29210257bb82SYan, Zheng return ret; 29220257bb82SYan, Zheng } 29235d4f98a2SYan Zheng 2924726a3421SQu Wenruo /* 2925907d2710SDavid Sterba * Allow error injection to test balance/relocation cancellation 2926726a3421SQu Wenruo */ 29271fec12a5SJosef Bacik noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 2928726a3421SQu Wenruo { 29295cb502f4SQu Wenruo return atomic_read(&fs_info->balance_cancel_req) || 2930907d2710SDavid Sterba atomic_read(&fs_info->reloc_cancel_req) || 29315cb502f4SQu Wenruo fatal_signal_pending(current); 2932726a3421SQu Wenruo } 2933726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 2934726a3421SQu Wenruo 2935c2832898SQu Wenruo static u64 get_cluster_boundary_end(struct file_extent_cluster *cluster, 2936c2832898SQu Wenruo int cluster_nr) 2937c2832898SQu Wenruo { 2938c2832898SQu Wenruo /* Last extent, use cluster end directly */ 2939c2832898SQu Wenruo if (cluster_nr >= cluster->nr - 1) 2940c2832898SQu Wenruo return cluster->end; 2941c2832898SQu Wenruo 2942c2832898SQu Wenruo /* Use next boundary start*/ 2943c2832898SQu Wenruo return cluster->boundary[cluster_nr + 1] - 1; 2944c2832898SQu Wenruo } 2945c2832898SQu Wenruo 2946f47960f4SQu Wenruo static int relocate_one_page(struct inode *inode, struct file_ra_state *ra, 2947f47960f4SQu Wenruo struct file_extent_cluster *cluster, 2948f47960f4SQu Wenruo int *cluster_nr, unsigned long page_index) 2949f47960f4SQu Wenruo { 2950f47960f4SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 2951f47960f4SQu Wenruo u64 offset = BTRFS_I(inode)->index_cnt; 2952f47960f4SQu Wenruo const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT; 2953f47960f4SQu Wenruo gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 2954f47960f4SQu Wenruo struct page *page; 2955f47960f4SQu Wenruo u64 page_start; 2956f47960f4SQu Wenruo u64 page_end; 2957c2832898SQu Wenruo u64 cur; 2958f47960f4SQu Wenruo int ret; 2959f47960f4SQu Wenruo 2960f47960f4SQu Wenruo ASSERT(page_index <= last_index); 2961f47960f4SQu Wenruo page = find_lock_page(inode->i_mapping, page_index); 2962f47960f4SQu Wenruo if (!page) { 2963f47960f4SQu Wenruo page_cache_sync_readahead(inode->i_mapping, ra, NULL, 2964f47960f4SQu Wenruo page_index, last_index + 1 - page_index); 2965f47960f4SQu Wenruo page = find_or_create_page(inode->i_mapping, page_index, mask); 2966c2832898SQu Wenruo if (!page) 2967c2832898SQu Wenruo return -ENOMEM; 2968f47960f4SQu Wenruo } 2969f47960f4SQu Wenruo ret = set_page_extent_mapped(page); 2970f47960f4SQu Wenruo if (ret < 0) 2971f47960f4SQu Wenruo goto release_page; 2972f47960f4SQu Wenruo 2973f47960f4SQu Wenruo if (PageReadahead(page)) 29742ebdd1dfSMatthew Wilcox (Oracle) page_cache_async_readahead(inode->i_mapping, ra, NULL, 29752ebdd1dfSMatthew Wilcox (Oracle) page_folio(page), page_index, 29762ebdd1dfSMatthew Wilcox (Oracle) last_index + 1 - page_index); 2977f47960f4SQu Wenruo 2978f47960f4SQu Wenruo if (!PageUptodate(page)) { 2979fb12489bSMatthew Wilcox (Oracle) btrfs_read_folio(NULL, page_folio(page)); 2980f47960f4SQu Wenruo lock_page(page); 2981f47960f4SQu Wenruo if (!PageUptodate(page)) { 2982f47960f4SQu Wenruo ret = -EIO; 2983f47960f4SQu Wenruo goto release_page; 2984f47960f4SQu Wenruo } 2985f47960f4SQu Wenruo } 2986f47960f4SQu Wenruo 2987f47960f4SQu Wenruo page_start = page_offset(page); 2988f47960f4SQu Wenruo page_end = page_start + PAGE_SIZE - 1; 2989f47960f4SQu Wenruo 2990c2832898SQu Wenruo /* 2991c2832898SQu Wenruo * Start from the cluster, as for subpage case, the cluster can start 2992c2832898SQu Wenruo * inside the page. 2993c2832898SQu Wenruo */ 2994c2832898SQu Wenruo cur = max(page_start, cluster->boundary[*cluster_nr] - offset); 2995c2832898SQu Wenruo while (cur <= page_end) { 29969c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 2997c2832898SQu Wenruo u64 extent_start = cluster->boundary[*cluster_nr] - offset; 2998c2832898SQu Wenruo u64 extent_end = get_cluster_boundary_end(cluster, 2999c2832898SQu Wenruo *cluster_nr) - offset; 3000c2832898SQu Wenruo u64 clamped_start = max(page_start, extent_start); 3001c2832898SQu Wenruo u64 clamped_end = min(page_end, extent_end); 3002c2832898SQu Wenruo u32 clamped_len = clamped_end + 1 - clamped_start; 3003f47960f4SQu Wenruo 3004c2832898SQu Wenruo /* Reserve metadata for this range */ 3005c2832898SQu Wenruo ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 3006d4135134SFilipe Manana clamped_len, clamped_len, 3007d4135134SFilipe Manana false); 3008c2832898SQu Wenruo if (ret) 3009f47960f4SQu Wenruo goto release_page; 3010f47960f4SQu Wenruo 3011c2832898SQu Wenruo /* Mark the range delalloc and dirty for later writeback */ 30129c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, 30139c5c9604SJosef Bacik &cached_state); 3014c2832898SQu Wenruo ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start, 30159c5c9604SJosef Bacik clamped_end, 0, &cached_state); 3016c2832898SQu Wenruo if (ret) { 30179c5c9604SJosef Bacik clear_extent_bit(&BTRFS_I(inode)->io_tree, 3018c2832898SQu Wenruo clamped_start, clamped_end, 30199c5c9604SJosef Bacik EXTENT_LOCKED | EXTENT_BOUNDARY, 30209c5c9604SJosef Bacik &cached_state); 3021c2832898SQu Wenruo btrfs_delalloc_release_metadata(BTRFS_I(inode), 3022c2832898SQu Wenruo clamped_len, true); 3023c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), 3024c2832898SQu Wenruo clamped_len); 3025c2832898SQu Wenruo goto release_page; 3026f47960f4SQu Wenruo } 3027c2832898SQu Wenruo btrfs_page_set_dirty(fs_info, page, clamped_start, clamped_len); 3028f47960f4SQu Wenruo 3029c2832898SQu Wenruo /* 3030c2832898SQu Wenruo * Set the boundary if it's inside the page. 3031c2832898SQu Wenruo * Data relocation requires the destination extents to have the 3032c2832898SQu Wenruo * same size as the source. 3033c2832898SQu Wenruo * EXTENT_BOUNDARY bit prevents current extent from being merged 3034c2832898SQu Wenruo * with previous extent. 3035c2832898SQu Wenruo */ 3036c2832898SQu Wenruo if (in_range(cluster->boundary[*cluster_nr] - offset, 3037c2832898SQu Wenruo page_start, PAGE_SIZE)) { 3038c2832898SQu Wenruo u64 boundary_start = cluster->boundary[*cluster_nr] - 3039c2832898SQu Wenruo offset; 3040c2832898SQu Wenruo u64 boundary_end = boundary_start + 3041c2832898SQu Wenruo fs_info->sectorsize - 1; 3042c2832898SQu Wenruo 3043c2832898SQu Wenruo set_extent_bits(&BTRFS_I(inode)->io_tree, 3044c2832898SQu Wenruo boundary_start, boundary_end, 3045c2832898SQu Wenruo EXTENT_BOUNDARY); 3046c2832898SQu Wenruo } 30479c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, 30489c5c9604SJosef Bacik &cached_state); 3049c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len); 3050c2832898SQu Wenruo cur += clamped_len; 3051c2832898SQu Wenruo 3052c2832898SQu Wenruo /* Crossed extent end, go to next extent */ 3053c2832898SQu Wenruo if (cur >= extent_end) { 3054c2832898SQu Wenruo (*cluster_nr)++; 3055c2832898SQu Wenruo /* Just finished the last extent of the cluster, exit. */ 3056c2832898SQu Wenruo if (*cluster_nr >= cluster->nr) 3057c2832898SQu Wenruo break; 3058c2832898SQu Wenruo } 3059c2832898SQu Wenruo } 3060f47960f4SQu Wenruo unlock_page(page); 3061f47960f4SQu Wenruo put_page(page); 3062f47960f4SQu Wenruo 3063f47960f4SQu Wenruo balance_dirty_pages_ratelimited(inode->i_mapping); 3064f47960f4SQu Wenruo btrfs_throttle(fs_info); 3065f47960f4SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) 3066f47960f4SQu Wenruo ret = -ECANCELED; 3067f47960f4SQu Wenruo return ret; 3068f47960f4SQu Wenruo 3069f47960f4SQu Wenruo release_page: 3070f47960f4SQu Wenruo unlock_page(page); 3071f47960f4SQu Wenruo put_page(page); 3072f47960f4SQu Wenruo return ret; 3073f47960f4SQu Wenruo } 3074f47960f4SQu Wenruo 30750257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 30760257bb82SYan, Zheng struct file_extent_cluster *cluster) 30770257bb82SYan, Zheng { 30780257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 30790257bb82SYan, Zheng unsigned long index; 30800257bb82SYan, Zheng unsigned long last_index; 30810257bb82SYan, Zheng struct file_ra_state *ra; 3082f47960f4SQu Wenruo int cluster_nr = 0; 30830257bb82SYan, Zheng int ret = 0; 30840257bb82SYan, Zheng 30850257bb82SYan, Zheng if (!cluster->nr) 30860257bb82SYan, Zheng return 0; 30870257bb82SYan, Zheng 30880257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 30890257bb82SYan, Zheng if (!ra) 30900257bb82SYan, Zheng return -ENOMEM; 30910257bb82SYan, Zheng 3092056d9becSNikolay Borisov ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster); 30930257bb82SYan, Zheng if (ret) 3094efa56464SYan, Zheng goto out; 30950257bb82SYan, Zheng 30960257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 30970257bb82SYan, Zheng 30984b01c44fSJohannes Thumshirn ret = setup_relocation_extent_mapping(inode, cluster->start - offset, 3099efa56464SYan, Zheng cluster->end - offset, cluster->start); 3100efa56464SYan, Zheng if (ret) 3101efa56464SYan, Zheng goto out; 3102efa56464SYan, Zheng 310309cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 3104f47960f4SQu Wenruo for (index = (cluster->start - offset) >> PAGE_SHIFT; 3105f47960f4SQu Wenruo index <= last_index && !ret; index++) 3106f47960f4SQu Wenruo ret = relocate_one_page(inode, ra, cluster, &cluster_nr, index); 3107f47960f4SQu Wenruo if (ret == 0) 3108f47960f4SQu Wenruo WARN_ON(cluster_nr != cluster->nr); 3109efa56464SYan, Zheng out: 31100257bb82SYan, Zheng kfree(ra); 31110257bb82SYan, Zheng return ret; 31120257bb82SYan, Zheng } 31130257bb82SYan, Zheng 31140257bb82SYan, Zheng static noinline_for_stack 31150257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 31160257bb82SYan, Zheng struct file_extent_cluster *cluster) 31170257bb82SYan, Zheng { 31180257bb82SYan, Zheng int ret; 31190257bb82SYan, Zheng 31200257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 31210257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31220257bb82SYan, Zheng if (ret) 31230257bb82SYan, Zheng return ret; 31240257bb82SYan, Zheng cluster->nr = 0; 31250257bb82SYan, Zheng } 31260257bb82SYan, Zheng 31270257bb82SYan, Zheng if (!cluster->nr) 31280257bb82SYan, Zheng cluster->start = extent_key->objectid; 31290257bb82SYan, Zheng else 31300257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 31310257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 31320257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 31330257bb82SYan, Zheng cluster->nr++; 31340257bb82SYan, Zheng 31350257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 31360257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31370257bb82SYan, Zheng if (ret) 31380257bb82SYan, Zheng return ret; 31390257bb82SYan, Zheng cluster->nr = 0; 31400257bb82SYan, Zheng } 31410257bb82SYan, Zheng return 0; 31425d4f98a2SYan Zheng } 31435d4f98a2SYan Zheng 31445d4f98a2SYan Zheng /* 31455d4f98a2SYan Zheng * helper to add a tree block to the list. 31465d4f98a2SYan Zheng * the major work is getting the generation and level of the block 31475d4f98a2SYan Zheng */ 31485d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 31495d4f98a2SYan Zheng struct btrfs_key *extent_key, 31505d4f98a2SYan Zheng struct btrfs_path *path, 31515d4f98a2SYan Zheng struct rb_root *blocks) 31525d4f98a2SYan Zheng { 31535d4f98a2SYan Zheng struct extent_buffer *eb; 31545d4f98a2SYan Zheng struct btrfs_extent_item *ei; 31555d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 31565d4f98a2SYan Zheng struct tree_block *block; 31575d4f98a2SYan Zheng struct rb_node *rb_node; 31585d4f98a2SYan Zheng u32 item_size; 31595d4f98a2SYan Zheng int level = -1; 31607fdf4b60SWang Shilong u64 generation; 3161f7ba2d37SJosef Bacik u64 owner = 0; 31625d4f98a2SYan Zheng 31635d4f98a2SYan Zheng eb = path->nodes[0]; 31643212fa14SJosef Bacik item_size = btrfs_item_size(eb, path->slots[0]); 31655d4f98a2SYan Zheng 31663173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 31673173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 3168f7ba2d37SJosef Bacik unsigned long ptr = 0, end; 3169f7ba2d37SJosef Bacik 31705d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 31715d4f98a2SYan Zheng struct btrfs_extent_item); 3172f7ba2d37SJosef Bacik end = (unsigned long)ei + item_size; 31733173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 31745d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 31755d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 3176f7ba2d37SJosef Bacik ptr = (unsigned long)(bi + 1); 31775d4f98a2SYan Zheng } else { 31783173a18fSJosef Bacik level = (int)extent_key->offset; 3179f7ba2d37SJosef Bacik ptr = (unsigned long)(ei + 1); 31803173a18fSJosef Bacik } 31813173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 3182f7ba2d37SJosef Bacik 3183f7ba2d37SJosef Bacik /* 3184f7ba2d37SJosef Bacik * We're reading random blocks without knowing their owner ahead 3185f7ba2d37SJosef Bacik * of time. This is ok most of the time, as all reloc roots and 3186f7ba2d37SJosef Bacik * fs roots have the same lock type. However normal trees do 3187f7ba2d37SJosef Bacik * not, and the only way to know ahead of time is to read the 3188f7ba2d37SJosef Bacik * inline ref offset. We know it's an fs root if 3189f7ba2d37SJosef Bacik * 3190f7ba2d37SJosef Bacik * 1. There's more than one ref. 3191f7ba2d37SJosef Bacik * 2. There's a SHARED_DATA_REF_KEY set. 3192f7ba2d37SJosef Bacik * 3. FULL_BACKREF is set on the flags. 3193f7ba2d37SJosef Bacik * 3194f7ba2d37SJosef Bacik * Otherwise it's safe to assume that the ref offset == the 3195f7ba2d37SJosef Bacik * owner of this block, so we can use that when calling 3196f7ba2d37SJosef Bacik * read_tree_block. 3197f7ba2d37SJosef Bacik */ 3198f7ba2d37SJosef Bacik if (btrfs_extent_refs(eb, ei) == 1 && 3199f7ba2d37SJosef Bacik !(btrfs_extent_flags(eb, ei) & 3200f7ba2d37SJosef Bacik BTRFS_BLOCK_FLAG_FULL_BACKREF) && 3201f7ba2d37SJosef Bacik ptr < end) { 3202f7ba2d37SJosef Bacik struct btrfs_extent_inline_ref *iref; 3203f7ba2d37SJosef Bacik int type; 3204f7ba2d37SJosef Bacik 3205f7ba2d37SJosef Bacik iref = (struct btrfs_extent_inline_ref *)ptr; 3206f7ba2d37SJosef Bacik type = btrfs_get_extent_inline_ref_type(eb, iref, 3207f7ba2d37SJosef Bacik BTRFS_REF_TYPE_BLOCK); 3208f7ba2d37SJosef Bacik if (type == BTRFS_REF_TYPE_INVALID) 3209f7ba2d37SJosef Bacik return -EINVAL; 3210f7ba2d37SJosef Bacik if (type == BTRFS_TREE_BLOCK_REF_KEY) 3211f7ba2d37SJosef Bacik owner = btrfs_extent_inline_ref_offset(eb, iref); 3212f7ba2d37SJosef Bacik } 32136d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3214ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3215ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3216ba3c2b19SNikolay Borisov return -EINVAL; 32173173a18fSJosef Bacik } else { 32185d4f98a2SYan Zheng BUG(); 32195d4f98a2SYan Zheng } 32205d4f98a2SYan Zheng 3221b3b4aa74SDavid Sterba btrfs_release_path(path); 32225d4f98a2SYan Zheng 32235d4f98a2SYan Zheng BUG_ON(level == -1); 32245d4f98a2SYan Zheng 32255d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 32265d4f98a2SYan Zheng if (!block) 32275d4f98a2SYan Zheng return -ENOMEM; 32285d4f98a2SYan Zheng 32295d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3230da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 32315d4f98a2SYan Zheng block->key.offset = generation; 32325d4f98a2SYan Zheng block->level = level; 32335d4f98a2SYan Zheng block->key_ready = 0; 3234f7ba2d37SJosef Bacik block->owner = owner; 32355d4f98a2SYan Zheng 3236e9a28dc5SQu Wenruo rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node); 323743c04fb1SJeff Mahoney if (rb_node) 3238982c92cbSQu Wenruo btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr, 3239982c92cbSQu Wenruo -EEXIST); 32405d4f98a2SYan Zheng 32415d4f98a2SYan Zheng return 0; 32425d4f98a2SYan Zheng } 32435d4f98a2SYan Zheng 32445d4f98a2SYan Zheng /* 32455d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 32465d4f98a2SYan Zheng */ 32475d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 32485d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 32495d4f98a2SYan Zheng struct rb_root *blocks) 32505d4f98a2SYan Zheng { 32510b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 32525d4f98a2SYan Zheng struct btrfs_path *path; 32535d4f98a2SYan Zheng struct btrfs_key key; 32545d4f98a2SYan Zheng int ret; 32550b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 32565d4f98a2SYan Zheng 32577476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 32585d4f98a2SYan Zheng return 0; 32595d4f98a2SYan Zheng 3260e9a28dc5SQu Wenruo if (rb_simple_search(blocks, bytenr)) 32615d4f98a2SYan Zheng return 0; 32625d4f98a2SYan Zheng 32635d4f98a2SYan Zheng path = btrfs_alloc_path(); 32645d4f98a2SYan Zheng if (!path) 32655d4f98a2SYan Zheng return -ENOMEM; 3266aee68ee5SJosef Bacik again: 32675d4f98a2SYan Zheng key.objectid = bytenr; 3268aee68ee5SJosef Bacik if (skinny) { 3269aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3270aee68ee5SJosef Bacik key.offset = (u64)-1; 3271aee68ee5SJosef Bacik } else { 32725d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 32735d4f98a2SYan Zheng key.offset = blocksize; 3274aee68ee5SJosef Bacik } 32755d4f98a2SYan Zheng 32765d4f98a2SYan Zheng path->search_commit_root = 1; 32775d4f98a2SYan Zheng path->skip_locking = 1; 32785d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 32795d4f98a2SYan Zheng if (ret < 0) 32805d4f98a2SYan Zheng goto out; 32815d4f98a2SYan Zheng 3282aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3283aee68ee5SJosef Bacik if (path->slots[0]) { 3284aee68ee5SJosef Bacik path->slots[0]--; 3285aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3286aee68ee5SJosef Bacik path->slots[0]); 32873173a18fSJosef Bacik if (key.objectid == bytenr && 3288aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3289aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3290aee68ee5SJosef Bacik key.offset == blocksize))) 32913173a18fSJosef Bacik ret = 0; 32923173a18fSJosef Bacik } 3293aee68ee5SJosef Bacik 3294aee68ee5SJosef Bacik if (ret) { 3295aee68ee5SJosef Bacik skinny = false; 3296aee68ee5SJosef Bacik btrfs_release_path(path); 3297aee68ee5SJosef Bacik goto again; 3298aee68ee5SJosef Bacik } 3299aee68ee5SJosef Bacik } 3300cdccee99SLiu Bo if (ret) { 3301cdccee99SLiu Bo ASSERT(ret == 1); 3302cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3303cdccee99SLiu Bo btrfs_err(fs_info, 3304cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3305cdccee99SLiu Bo bytenr); 3306cdccee99SLiu Bo WARN_ON(1); 3307cdccee99SLiu Bo ret = -EINVAL; 3308cdccee99SLiu Bo goto out; 3309cdccee99SLiu Bo } 33103173a18fSJosef Bacik 33115d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 33125d4f98a2SYan Zheng out: 33135d4f98a2SYan Zheng btrfs_free_path(path); 33145d4f98a2SYan Zheng return ret; 33155d4f98a2SYan Zheng } 33165d4f98a2SYan Zheng 33170af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 331832da5386SDavid Sterba struct btrfs_block_group *block_group, 33191bbc621eSChris Mason struct inode *inode, 33201bbc621eSChris Mason u64 ino) 33210af3d00bSJosef Bacik { 33220af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 33230af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 33240af3d00bSJosef Bacik int ret = 0; 33250af3d00bSJosef Bacik 33260af3d00bSJosef Bacik if (inode) 33270af3d00bSJosef Bacik goto truncate; 33280af3d00bSJosef Bacik 33290202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, ino, root); 33302e19f1f9SAl Viro if (IS_ERR(inode)) 33310af3d00bSJosef Bacik return -ENOENT; 33320af3d00bSJosef Bacik 33330af3d00bSJosef Bacik truncate: 33342ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 33357b61cd92SMiao Xie &fs_info->global_block_rsv); 33367b61cd92SMiao Xie if (ret) 33377b61cd92SMiao Xie goto out; 33387b61cd92SMiao Xie 33397a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 33400af3d00bSJosef Bacik if (IS_ERR(trans)) { 33413612b495STsutomu Itoh ret = PTR_ERR(trans); 33420af3d00bSJosef Bacik goto out; 33430af3d00bSJosef Bacik } 33440af3d00bSJosef Bacik 334577ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 33460af3d00bSJosef Bacik 33473a45bb20SJeff Mahoney btrfs_end_transaction(trans); 33482ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 33490af3d00bSJosef Bacik out: 33500af3d00bSJosef Bacik iput(inode); 33510af3d00bSJosef Bacik return ret; 33520af3d00bSJosef Bacik } 33530af3d00bSJosef Bacik 33545d4f98a2SYan Zheng /* 335519b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 335619b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 33575d4f98a2SYan Zheng */ 335819b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 335919b546d7SQu Wenruo struct btrfs_block_group *block_group, 336019b546d7SQu Wenruo u64 data_bytenr) 33615d4f98a2SYan Zheng { 336219b546d7SQu Wenruo u64 space_cache_ino; 336319b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 33645d4f98a2SYan Zheng struct btrfs_key key; 336519b546d7SQu Wenruo bool found = false; 336619b546d7SQu Wenruo int i; 33675d4f98a2SYan Zheng int ret; 33685d4f98a2SYan Zheng 336919b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 337019b546d7SQu Wenruo return 0; 33715d4f98a2SYan Zheng 337219b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 337350e31ef4SQu Wenruo u8 type; 337450e31ef4SQu Wenruo 337519b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 337619b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 337719b546d7SQu Wenruo continue; 337819b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 337950e31ef4SQu Wenruo type = btrfs_file_extent_type(leaf, ei); 338050e31ef4SQu Wenruo 338150e31ef4SQu Wenruo if ((type == BTRFS_FILE_EXTENT_REG || 338250e31ef4SQu Wenruo type == BTRFS_FILE_EXTENT_PREALLOC) && 338319b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 338419b546d7SQu Wenruo found = true; 338519b546d7SQu Wenruo space_cache_ino = key.objectid; 338619b546d7SQu Wenruo break; 338719b546d7SQu Wenruo } 338819b546d7SQu Wenruo } 338919b546d7SQu Wenruo if (!found) 339019b546d7SQu Wenruo return -ENOENT; 339119b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 339219b546d7SQu Wenruo space_cache_ino); 33930af3d00bSJosef Bacik return ret; 33945d4f98a2SYan Zheng } 33955d4f98a2SYan Zheng 33965d4f98a2SYan Zheng /* 33972c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 33985d4f98a2SYan Zheng */ 33995d4f98a2SYan Zheng static noinline_for_stack 34005d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 34015d4f98a2SYan Zheng struct btrfs_key *extent_key, 34025d4f98a2SYan Zheng struct btrfs_path *path, 34035d4f98a2SYan Zheng struct rb_root *blocks) 34045d4f98a2SYan Zheng { 340519b546d7SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 340619b546d7SQu Wenruo struct ulist *leaves = NULL; 340719b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 340819b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 340919b546d7SQu Wenruo const u32 blocksize = fs_info->nodesize; 3410647f63bdSFilipe David Borba Manana int ret = 0; 34115d4f98a2SYan Zheng 3412b3b4aa74SDavid Sterba btrfs_release_path(path); 341319b546d7SQu Wenruo ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid, 341419b546d7SQu Wenruo 0, &leaves, NULL, true); 341519b546d7SQu Wenruo if (ret < 0) 341619b546d7SQu Wenruo return ret; 341719b546d7SQu Wenruo 341819b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 341919b546d7SQu Wenruo while ((ref_node = ulist_next(leaves, &leaf_uiter))) { 342019b546d7SQu Wenruo struct extent_buffer *eb; 342119b546d7SQu Wenruo 34221b7ec85eSJosef Bacik eb = read_tree_block(fs_info, ref_node->val, 0, 0, 0, NULL); 342319b546d7SQu Wenruo if (IS_ERR(eb)) { 342419b546d7SQu Wenruo ret = PTR_ERR(eb); 342519b546d7SQu Wenruo break; 342619b546d7SQu Wenruo } 342719b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 342819b546d7SQu Wenruo extent_key->objectid); 342919b546d7SQu Wenruo free_extent_buffer(eb); 343019b546d7SQu Wenruo if (ret < 0) 343119b546d7SQu Wenruo break; 343219b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 343319b546d7SQu Wenruo if (ret < 0) 343419b546d7SQu Wenruo break; 343519b546d7SQu Wenruo } 343619b546d7SQu Wenruo if (ret < 0) 34375d4f98a2SYan Zheng free_block_list(blocks); 343819b546d7SQu Wenruo ulist_free(leaves); 343919b546d7SQu Wenruo return ret; 34405d4f98a2SYan Zheng } 34415d4f98a2SYan Zheng 34425d4f98a2SYan Zheng /* 34432c016dc2SLiu Bo * helper to find next unprocessed extent 34445d4f98a2SYan Zheng */ 34455d4f98a2SYan Zheng static noinline_for_stack 3446147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 34473fd0a558SYan, Zheng struct btrfs_key *extent_key) 34485d4f98a2SYan Zheng { 34490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34505d4f98a2SYan Zheng struct btrfs_key key; 34515d4f98a2SYan Zheng struct extent_buffer *leaf; 34525d4f98a2SYan Zheng u64 start, end, last; 34535d4f98a2SYan Zheng int ret; 34545d4f98a2SYan Zheng 3455b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 34565d4f98a2SYan Zheng while (1) { 34575d4f98a2SYan Zheng cond_resched(); 34585d4f98a2SYan Zheng if (rc->search_start >= last) { 34595d4f98a2SYan Zheng ret = 1; 34605d4f98a2SYan Zheng break; 34615d4f98a2SYan Zheng } 34625d4f98a2SYan Zheng 34635d4f98a2SYan Zheng key.objectid = rc->search_start; 34645d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34655d4f98a2SYan Zheng key.offset = 0; 34665d4f98a2SYan Zheng 34675d4f98a2SYan Zheng path->search_commit_root = 1; 34685d4f98a2SYan Zheng path->skip_locking = 1; 34695d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 34705d4f98a2SYan Zheng 0, 0); 34715d4f98a2SYan Zheng if (ret < 0) 34725d4f98a2SYan Zheng break; 34735d4f98a2SYan Zheng next: 34745d4f98a2SYan Zheng leaf = path->nodes[0]; 34755d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 34765d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 34775d4f98a2SYan Zheng if (ret != 0) 34785d4f98a2SYan Zheng break; 34795d4f98a2SYan Zheng leaf = path->nodes[0]; 34805d4f98a2SYan Zheng } 34815d4f98a2SYan Zheng 34825d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 34835d4f98a2SYan Zheng if (key.objectid >= last) { 34845d4f98a2SYan Zheng ret = 1; 34855d4f98a2SYan Zheng break; 34865d4f98a2SYan Zheng } 34875d4f98a2SYan Zheng 34883173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 34893173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 34903173a18fSJosef Bacik path->slots[0]++; 34913173a18fSJosef Bacik goto next; 34923173a18fSJosef Bacik } 34933173a18fSJosef Bacik 34943173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 34955d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 34965d4f98a2SYan Zheng path->slots[0]++; 34975d4f98a2SYan Zheng goto next; 34985d4f98a2SYan Zheng } 34995d4f98a2SYan Zheng 35003173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 35010b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 35023173a18fSJosef Bacik rc->search_start) { 35033173a18fSJosef Bacik path->slots[0]++; 35043173a18fSJosef Bacik goto next; 35053173a18fSJosef Bacik } 35063173a18fSJosef Bacik 35075d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 35085d4f98a2SYan Zheng key.objectid, &start, &end, 3509e6138876SJosef Bacik EXTENT_DIRTY, NULL); 35105d4f98a2SYan Zheng 35115d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3512b3b4aa74SDavid Sterba btrfs_release_path(path); 35135d4f98a2SYan Zheng rc->search_start = end + 1; 35145d4f98a2SYan Zheng } else { 35153173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 35165d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 35173173a18fSJosef Bacik else 35183173a18fSJosef Bacik rc->search_start = key.objectid + 35190b246afaSJeff Mahoney fs_info->nodesize; 35203fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 35215d4f98a2SYan Zheng return 0; 35225d4f98a2SYan Zheng } 35235d4f98a2SYan Zheng } 3524b3b4aa74SDavid Sterba btrfs_release_path(path); 35255d4f98a2SYan Zheng return ret; 35265d4f98a2SYan Zheng } 35275d4f98a2SYan Zheng 35285d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 35295d4f98a2SYan Zheng { 35305d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35317585717fSChris Mason 35327585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35335d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 35347585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35355d4f98a2SYan Zheng } 35365d4f98a2SYan Zheng 35375d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 35385d4f98a2SYan Zheng { 35395d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35407585717fSChris Mason 35417585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35425d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 35437585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35445d4f98a2SYan Zheng } 35455d4f98a2SYan Zheng 35463fd0a558SYan, Zheng static noinline_for_stack 35473fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 35483fd0a558SYan, Zheng { 35493fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3550ac2fabacSJosef Bacik int ret; 35513fd0a558SYan, Zheng 35522ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 355366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 35543fd0a558SYan, Zheng if (!rc->block_rsv) 35553fd0a558SYan, Zheng return -ENOMEM; 35563fd0a558SYan, Zheng 35573fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3558b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 35593fd0a558SYan, Zheng rc->extents_found = 0; 35603fd0a558SYan, Zheng rc->nodes_relocated = 0; 35613fd0a558SYan, Zheng rc->merging_rsv_size = 0; 35620647bf56SWang Shilong rc->reserved_bytes = 0; 3563da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 35640647bf56SWang Shilong RELOCATION_RESERVED_NODES; 35659270501cSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root->fs_info, 3566ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3567ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3568ac2fabacSJosef Bacik if (ret) 3569ac2fabacSJosef Bacik return ret; 35703fd0a558SYan, Zheng 35713fd0a558SYan, Zheng rc->create_reloc_tree = 1; 35723fd0a558SYan, Zheng set_reloc_control(rc); 35733fd0a558SYan, Zheng 35747a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 357528818947SLiu Bo if (IS_ERR(trans)) { 357628818947SLiu Bo unset_reloc_control(rc); 357728818947SLiu Bo /* 357828818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 357928818947SLiu Bo * cleanup. And callers are responsible to free the above 358028818947SLiu Bo * block rsv. 358128818947SLiu Bo */ 358228818947SLiu Bo return PTR_ERR(trans); 358328818947SLiu Bo } 358485f02d6cSZixuan Fu 358585f02d6cSZixuan Fu ret = btrfs_commit_transaction(trans); 358685f02d6cSZixuan Fu if (ret) 358785f02d6cSZixuan Fu unset_reloc_control(rc); 358885f02d6cSZixuan Fu 358985f02d6cSZixuan Fu return ret; 35903fd0a558SYan, Zheng } 359176dda93cSYan, Zheng 35925d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 35935d4f98a2SYan Zheng { 35942ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35955d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 35965d4f98a2SYan Zheng struct btrfs_key key; 35975d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 35985d4f98a2SYan Zheng struct btrfs_path *path; 35995d4f98a2SYan Zheng struct btrfs_extent_item *ei; 36005d4f98a2SYan Zheng u64 flags; 36015d4f98a2SYan Zheng int ret; 36025d4f98a2SYan Zheng int err = 0; 3603c87f08caSChris Mason int progress = 0; 36045d4f98a2SYan Zheng 36055d4f98a2SYan Zheng path = btrfs_alloc_path(); 36063fd0a558SYan, Zheng if (!path) 36075d4f98a2SYan Zheng return -ENOMEM; 3608e4058b54SDavid Sterba path->reada = READA_FORWARD; 36093fd0a558SYan, Zheng 36103fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 36113fd0a558SYan, Zheng if (ret) { 36123fd0a558SYan, Zheng err = ret; 36133fd0a558SYan, Zheng goto out_free; 36142423fdfbSJiri Slaby } 36155d4f98a2SYan Zheng 36165d4f98a2SYan Zheng while (1) { 36170647bf56SWang Shilong rc->reserved_bytes = 0; 36189270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 36199270501cSJosef Bacik rc->block_rsv->size, 36200647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 36210647bf56SWang Shilong if (ret) { 36220647bf56SWang Shilong err = ret; 36230647bf56SWang Shilong break; 36240647bf56SWang Shilong } 3625c87f08caSChris Mason progress++; 3626a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 36270f788c58SLiu Bo if (IS_ERR(trans)) { 36280f788c58SLiu Bo err = PTR_ERR(trans); 36290f788c58SLiu Bo trans = NULL; 36300f788c58SLiu Bo break; 36310f788c58SLiu Bo } 3632c87f08caSChris Mason restart: 36333fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 36343a45bb20SJeff Mahoney btrfs_end_transaction(trans); 363542a657f5SPan Bian trans = NULL; 36363fd0a558SYan, Zheng continue; 36373fd0a558SYan, Zheng } 36383fd0a558SYan, Zheng 3639147d256eSZhaolei ret = find_next_extent(rc, path, &key); 36405d4f98a2SYan Zheng if (ret < 0) 36415d4f98a2SYan Zheng err = ret; 36425d4f98a2SYan Zheng if (ret != 0) 36435d4f98a2SYan Zheng break; 36445d4f98a2SYan Zheng 36455d4f98a2SYan Zheng rc->extents_found++; 36465d4f98a2SYan Zheng 36475d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 36485d4f98a2SYan Zheng struct btrfs_extent_item); 36495d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 36505d4f98a2SYan Zheng 36515d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 36525d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 36535d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 36545d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36555d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 36565d4f98a2SYan Zheng } else { 3657b3b4aa74SDavid Sterba btrfs_release_path(path); 36585d4f98a2SYan Zheng ret = 0; 36595d4f98a2SYan Zheng } 36605d4f98a2SYan Zheng if (ret < 0) { 36613fd0a558SYan, Zheng err = ret; 36625d4f98a2SYan Zheng break; 36635d4f98a2SYan Zheng } 36645d4f98a2SYan Zheng 36655d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 36665d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 36675d4f98a2SYan Zheng if (ret < 0) { 36683fd0a558SYan, Zheng if (ret != -EAGAIN) { 36695d4f98a2SYan Zheng err = ret; 36705d4f98a2SYan Zheng break; 36715d4f98a2SYan Zheng } 36723fd0a558SYan, Zheng rc->extents_found--; 36733fd0a558SYan, Zheng rc->search_start = key.objectid; 36743fd0a558SYan, Zheng } 36755d4f98a2SYan Zheng } 36765d4f98a2SYan Zheng 36773a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 36782ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36793fd0a558SYan, Zheng trans = NULL; 36805d4f98a2SYan Zheng 36815d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 36825d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36835d4f98a2SYan Zheng rc->found_file_extent = 1; 36840257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 36853fd0a558SYan, Zheng &key, &rc->cluster); 36865d4f98a2SYan Zheng if (ret < 0) { 36875d4f98a2SYan Zheng err = ret; 36885d4f98a2SYan Zheng break; 36895d4f98a2SYan Zheng } 36905d4f98a2SYan Zheng } 3691f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 3692f31ea088SQu Wenruo err = -ECANCELED; 3693f31ea088SQu Wenruo break; 3694f31ea088SQu Wenruo } 36955d4f98a2SYan Zheng } 3696c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 369743a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 36989689457bSShilong Wang if (ret == 1) { 3699c87f08caSChris Mason err = 0; 3700c87f08caSChris Mason progress = 0; 3701c87f08caSChris Mason goto restart; 3702c87f08caSChris Mason } 3703c87f08caSChris Mason } 37043fd0a558SYan, Zheng 3705b3b4aa74SDavid Sterba btrfs_release_path(path); 370691166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 37075d4f98a2SYan Zheng 37085d4f98a2SYan Zheng if (trans) { 37093a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 37102ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 37115d4f98a2SYan Zheng } 37125d4f98a2SYan Zheng 37130257bb82SYan, Zheng if (!err) { 37143fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 37153fd0a558SYan, Zheng &rc->cluster); 37160257bb82SYan, Zheng if (ret < 0) 37170257bb82SYan, Zheng err = ret; 37180257bb82SYan, Zheng } 37190257bb82SYan, Zheng 37203fd0a558SYan, Zheng rc->create_reloc_tree = 0; 37213fd0a558SYan, Zheng set_reloc_control(rc); 37220257bb82SYan, Zheng 372313fe1bdbSQu Wenruo btrfs_backref_release_cache(&rc->backref_cache); 372463f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37255d4f98a2SYan Zheng 37267f913c7cSQu Wenruo /* 37277f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 37287f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 37297f913c7cSQu Wenruo * 37307f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 37317f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 37327f913c7cSQu Wenruo * merge_reloc_roots() 37337f913c7cSQu Wenruo */ 37343fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 37355d4f98a2SYan Zheng 37365d4f98a2SYan Zheng merge_reloc_roots(rc); 37375d4f98a2SYan Zheng 37383fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 37395d4f98a2SYan Zheng unset_reloc_control(rc); 374063f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37415d4f98a2SYan Zheng 37425d4f98a2SYan Zheng /* get rid of pinned extents */ 37437a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 374462b99540SQu Wenruo if (IS_ERR(trans)) { 37453612b495STsutomu Itoh err = PTR_ERR(trans); 374662b99540SQu Wenruo goto out_free; 374762b99540SQu Wenruo } 3748fb686c68SJosef Bacik ret = btrfs_commit_transaction(trans); 3749fb686c68SJosef Bacik if (ret && !err) 3750fb686c68SJosef Bacik err = ret; 37516217b0faSJosef Bacik out_free: 3752d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 3753d2311e69SQu Wenruo if (ret < 0 && !err) 3754d2311e69SQu Wenruo err = ret; 37552ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 37563fd0a558SYan, Zheng btrfs_free_path(path); 37575d4f98a2SYan Zheng return err; 37585d4f98a2SYan Zheng } 37595d4f98a2SYan Zheng 37605d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 37610257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 37625d4f98a2SYan Zheng { 37635d4f98a2SYan Zheng struct btrfs_path *path; 37645d4f98a2SYan Zheng struct btrfs_inode_item *item; 37655d4f98a2SYan Zheng struct extent_buffer *leaf; 37665d4f98a2SYan Zheng int ret; 37675d4f98a2SYan Zheng 37685d4f98a2SYan Zheng path = btrfs_alloc_path(); 37695d4f98a2SYan Zheng if (!path) 37705d4f98a2SYan Zheng return -ENOMEM; 37715d4f98a2SYan Zheng 37725d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 37735d4f98a2SYan Zheng if (ret) 37745d4f98a2SYan Zheng goto out; 37755d4f98a2SYan Zheng 37765d4f98a2SYan Zheng leaf = path->nodes[0]; 37775d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 3778b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 37795d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 37800257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 37815d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 3782960a3166SJohannes Thumshirn btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 3783960a3166SJohannes Thumshirn BTRFS_INODE_PREALLOC); 37845d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 37855d4f98a2SYan Zheng out: 37865d4f98a2SYan Zheng btrfs_free_path(path); 37875d4f98a2SYan Zheng return ret; 37885d4f98a2SYan Zheng } 37895d4f98a2SYan Zheng 3790790c1b8cSJosef Bacik static void delete_orphan_inode(struct btrfs_trans_handle *trans, 3791790c1b8cSJosef Bacik struct btrfs_root *root, u64 objectid) 3792790c1b8cSJosef Bacik { 3793790c1b8cSJosef Bacik struct btrfs_path *path; 3794790c1b8cSJosef Bacik struct btrfs_key key; 3795790c1b8cSJosef Bacik int ret = 0; 3796790c1b8cSJosef Bacik 3797790c1b8cSJosef Bacik path = btrfs_alloc_path(); 3798790c1b8cSJosef Bacik if (!path) { 3799790c1b8cSJosef Bacik ret = -ENOMEM; 3800790c1b8cSJosef Bacik goto out; 3801790c1b8cSJosef Bacik } 3802790c1b8cSJosef Bacik 3803790c1b8cSJosef Bacik key.objectid = objectid; 3804790c1b8cSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 3805790c1b8cSJosef Bacik key.offset = 0; 3806790c1b8cSJosef Bacik ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3807790c1b8cSJosef Bacik if (ret) { 3808790c1b8cSJosef Bacik if (ret > 0) 3809790c1b8cSJosef Bacik ret = -ENOENT; 3810790c1b8cSJosef Bacik goto out; 3811790c1b8cSJosef Bacik } 3812790c1b8cSJosef Bacik ret = btrfs_del_item(trans, root, path); 3813790c1b8cSJosef Bacik out: 3814790c1b8cSJosef Bacik if (ret) 3815790c1b8cSJosef Bacik btrfs_abort_transaction(trans, ret); 3816790c1b8cSJosef Bacik btrfs_free_path(path); 3817790c1b8cSJosef Bacik } 3818790c1b8cSJosef Bacik 38195d4f98a2SYan Zheng /* 38205d4f98a2SYan Zheng * helper to create inode for data relocation. 38215d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 38225d4f98a2SYan Zheng */ 38233fd0a558SYan, Zheng static noinline_for_stack 38243fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 382532da5386SDavid Sterba struct btrfs_block_group *group) 38265d4f98a2SYan Zheng { 38275d4f98a2SYan Zheng struct inode *inode = NULL; 38285d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 38295d4f98a2SYan Zheng struct btrfs_root *root; 38304624900dSZhaolei u64 objectid; 38315d4f98a2SYan Zheng int err = 0; 38325d4f98a2SYan Zheng 3833aeb935a4SQu Wenruo root = btrfs_grab_root(fs_info->data_reloc_root); 3834a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 383576deacf0SJosef Bacik if (IS_ERR(trans)) { 383600246528SJosef Bacik btrfs_put_root(root); 38373fd0a558SYan, Zheng return ERR_CAST(trans); 383876deacf0SJosef Bacik } 38395d4f98a2SYan Zheng 3840543068a2SNikolay Borisov err = btrfs_get_free_objectid(root, &objectid); 38415d4f98a2SYan Zheng if (err) 38425d4f98a2SYan Zheng goto out; 38435d4f98a2SYan Zheng 38440257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 3845790c1b8cSJosef Bacik if (err) 3846790c1b8cSJosef Bacik goto out; 38475d4f98a2SYan Zheng 38480202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, objectid, root); 3849790c1b8cSJosef Bacik if (IS_ERR(inode)) { 3850790c1b8cSJosef Bacik delete_orphan_inode(trans, root, objectid); 3851790c1b8cSJosef Bacik err = PTR_ERR(inode); 3852790c1b8cSJosef Bacik inode = NULL; 3853790c1b8cSJosef Bacik goto out; 3854790c1b8cSJosef Bacik } 3855b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 38565d4f98a2SYan Zheng 385773f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 38585d4f98a2SYan Zheng out: 385900246528SJosef Bacik btrfs_put_root(root); 38603a45bb20SJeff Mahoney btrfs_end_transaction(trans); 38612ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 38625d4f98a2SYan Zheng if (err) { 38635d4f98a2SYan Zheng iput(inode); 38645d4f98a2SYan Zheng inode = ERR_PTR(err); 38655d4f98a2SYan Zheng } 38665d4f98a2SYan Zheng return inode; 38675d4f98a2SYan Zheng } 38685d4f98a2SYan Zheng 3869907d2710SDavid Sterba /* 3870907d2710SDavid Sterba * Mark start of chunk relocation that is cancellable. Check if the cancellation 3871907d2710SDavid Sterba * has been requested meanwhile and don't start in that case. 3872907d2710SDavid Sterba * 3873907d2710SDavid Sterba * Return: 3874907d2710SDavid Sterba * 0 success 3875907d2710SDavid Sterba * -EINPROGRESS operation is already in progress, that's probably a bug 3876907d2710SDavid Sterba * -ECANCELED cancellation request was set before the operation started 3877907d2710SDavid Sterba */ 3878907d2710SDavid Sterba static int reloc_chunk_start(struct btrfs_fs_info *fs_info) 3879907d2710SDavid Sterba { 3880907d2710SDavid Sterba if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) { 3881907d2710SDavid Sterba /* This should not happen */ 3882907d2710SDavid Sterba btrfs_err(fs_info, "reloc already running, cannot start"); 3883907d2710SDavid Sterba return -EINPROGRESS; 3884907d2710SDavid Sterba } 3885907d2710SDavid Sterba 3886907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) { 3887907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled on start"); 3888907d2710SDavid Sterba /* 3889907d2710SDavid Sterba * On cancel, clear all requests but let the caller mark 3890907d2710SDavid Sterba * the end after cleanup operations. 3891907d2710SDavid Sterba */ 3892907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3893907d2710SDavid Sterba return -ECANCELED; 3894907d2710SDavid Sterba } 3895907d2710SDavid Sterba return 0; 3896907d2710SDavid Sterba } 3897907d2710SDavid Sterba 3898907d2710SDavid Sterba /* 3899907d2710SDavid Sterba * Mark end of chunk relocation that is cancellable and wake any waiters. 3900907d2710SDavid Sterba */ 3901907d2710SDavid Sterba static void reloc_chunk_end(struct btrfs_fs_info *fs_info) 3902907d2710SDavid Sterba { 3903907d2710SDavid Sterba /* Requested after start, clear bit first so any waiters can continue */ 3904907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) 3905907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled during operation"); 3906907d2710SDavid Sterba clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags); 3907907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3908907d2710SDavid Sterba } 3909907d2710SDavid Sterba 3910c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 39113fd0a558SYan, Zheng { 39123fd0a558SYan, Zheng struct reloc_control *rc; 39133fd0a558SYan, Zheng 39143fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 39153fd0a558SYan, Zheng if (!rc) 39163fd0a558SYan, Zheng return NULL; 39173fd0a558SYan, Zheng 39183fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 3919d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 3920584fb121SQu Wenruo btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1); 39213fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 392243eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 392343eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 39243fd0a558SYan, Zheng return rc; 39253fd0a558SYan, Zheng } 39263fd0a558SYan, Zheng 39271a0afa0eSJosef Bacik static void free_reloc_control(struct reloc_control *rc) 39281a0afa0eSJosef Bacik { 39291a0afa0eSJosef Bacik struct mapping_node *node, *tmp; 39301a0afa0eSJosef Bacik 39311a0afa0eSJosef Bacik free_reloc_roots(&rc->reloc_roots); 39321a0afa0eSJosef Bacik rbtree_postorder_for_each_entry_safe(node, tmp, 39331a0afa0eSJosef Bacik &rc->reloc_root_tree.rb_root, rb_node) 39341a0afa0eSJosef Bacik kfree(node); 39351a0afa0eSJosef Bacik 39361a0afa0eSJosef Bacik kfree(rc); 39371a0afa0eSJosef Bacik } 39381a0afa0eSJosef Bacik 39395d4f98a2SYan Zheng /* 3940ebce0e01SAdam Borowski * Print the block group being relocated 3941ebce0e01SAdam Borowski */ 3942ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 394332da5386SDavid Sterba struct btrfs_block_group *block_group) 3944ebce0e01SAdam Borowski { 3945f89e09cfSAnand Jain char buf[128] = {'\0'}; 3946ebce0e01SAdam Borowski 3947f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 3948ebce0e01SAdam Borowski 3949ebce0e01SAdam Borowski btrfs_info(fs_info, 3950ebce0e01SAdam Borowski "relocating block group %llu flags %s", 3951b3470b5dSDavid Sterba block_group->start, buf); 3952ebce0e01SAdam Borowski } 3953ebce0e01SAdam Borowski 3954430640e3SQu Wenruo static const char *stage_to_string(int stage) 3955430640e3SQu Wenruo { 3956430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 3957430640e3SQu Wenruo return "move data extents"; 3958430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 3959430640e3SQu Wenruo return "update data pointers"; 3960430640e3SQu Wenruo return "unknown"; 3961430640e3SQu Wenruo } 3962430640e3SQu Wenruo 3963ebce0e01SAdam Borowski /* 39645d4f98a2SYan Zheng * function to relocate all extents in a block group. 39655d4f98a2SYan Zheng */ 39666bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 39675d4f98a2SYan Zheng { 396832da5386SDavid Sterba struct btrfs_block_group *bg; 396929cbcf40SJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start); 39705d4f98a2SYan Zheng struct reloc_control *rc; 39710af3d00bSJosef Bacik struct inode *inode; 39720af3d00bSJosef Bacik struct btrfs_path *path; 39735d4f98a2SYan Zheng int ret; 3974f0486c68SYan, Zheng int rw = 0; 39755d4f98a2SYan Zheng int err = 0; 39765d4f98a2SYan Zheng 3977b4be6aefSJosef Bacik /* 3978b4be6aefSJosef Bacik * This only gets set if we had a half-deleted snapshot on mount. We 3979b4be6aefSJosef Bacik * cannot allow relocation to start while we're still trying to clean up 3980b4be6aefSJosef Bacik * these pending deletions. 3981b4be6aefSJosef Bacik */ 3982b4be6aefSJosef Bacik ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE); 3983b4be6aefSJosef Bacik if (ret) 3984b4be6aefSJosef Bacik return ret; 3985b4be6aefSJosef Bacik 3986b4be6aefSJosef Bacik /* We may have been woken up by close_ctree, so bail if we're closing. */ 3987b4be6aefSJosef Bacik if (btrfs_fs_closing(fs_info)) 3988b4be6aefSJosef Bacik return -EINTR; 3989b4be6aefSJosef Bacik 3990eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 3991eede2bf3SOmar Sandoval if (!bg) 3992eede2bf3SOmar Sandoval return -ENOENT; 3993eede2bf3SOmar Sandoval 39940320b353SNaohiro Aota /* 39950320b353SNaohiro Aota * Relocation of a data block group creates ordered extents. Without 39960320b353SNaohiro Aota * sb_start_write(), we can freeze the filesystem while unfinished 39970320b353SNaohiro Aota * ordered extents are left. Such ordered extents can cause a deadlock 39980320b353SNaohiro Aota * e.g. when syncfs() is waiting for their completion but they can't 39990320b353SNaohiro Aota * finish because they block when joining a transaction, due to the 40000320b353SNaohiro Aota * fact that the freeze locks are being held in write mode. 40010320b353SNaohiro Aota */ 40020320b353SNaohiro Aota if (bg->flags & BTRFS_BLOCK_GROUP_DATA) 40030320b353SNaohiro Aota ASSERT(sb_write_started(fs_info->sb)); 40040320b353SNaohiro Aota 4005eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4006eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4007eede2bf3SOmar Sandoval return -ETXTBSY; 4008eede2bf3SOmar Sandoval } 4009eede2bf3SOmar Sandoval 4010c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4011eede2bf3SOmar Sandoval if (!rc) { 4012eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 40135d4f98a2SYan Zheng return -ENOMEM; 4014eede2bf3SOmar Sandoval } 40155d4f98a2SYan Zheng 4016907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4017907d2710SDavid Sterba if (ret < 0) { 4018907d2710SDavid Sterba err = ret; 4019907d2710SDavid Sterba goto out_put_bg; 4020907d2710SDavid Sterba } 4021907d2710SDavid Sterba 4022f0486c68SYan, Zheng rc->extent_root = extent_root; 4023eede2bf3SOmar Sandoval rc->block_group = bg; 40245d4f98a2SYan Zheng 4025b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4026f0486c68SYan, Zheng if (ret) { 4027f0486c68SYan, Zheng err = ret; 4028f0486c68SYan, Zheng goto out; 4029f0486c68SYan, Zheng } 4030f0486c68SYan, Zheng rw = 1; 4031f0486c68SYan, Zheng 40320af3d00bSJosef Bacik path = btrfs_alloc_path(); 40330af3d00bSJosef Bacik if (!path) { 40340af3d00bSJosef Bacik err = -ENOMEM; 40350af3d00bSJosef Bacik goto out; 40360af3d00bSJosef Bacik } 40370af3d00bSJosef Bacik 40387949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 40390af3d00bSJosef Bacik btrfs_free_path(path); 40400af3d00bSJosef Bacik 40410af3d00bSJosef Bacik if (!IS_ERR(inode)) 40421bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 40430af3d00bSJosef Bacik else 40440af3d00bSJosef Bacik ret = PTR_ERR(inode); 40450af3d00bSJosef Bacik 40460af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 40470af3d00bSJosef Bacik err = ret; 40480af3d00bSJosef Bacik goto out; 40490af3d00bSJosef Bacik } 40500af3d00bSJosef Bacik 40515d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 40525d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 40535d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 40545d4f98a2SYan Zheng rc->data_inode = NULL; 40555d4f98a2SYan Zheng goto out; 40565d4f98a2SYan Zheng } 40575d4f98a2SYan Zheng 40580b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 40595d4f98a2SYan Zheng 40609cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4061f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 40626374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4063b3470b5dSDavid Sterba rc->block_group->start, 4064b3470b5dSDavid Sterba rc->block_group->length); 40655d4f98a2SYan Zheng 40667ae9bd18SNaohiro Aota ret = btrfs_zone_finish(rc->block_group); 40677ae9bd18SNaohiro Aota WARN_ON(ret && ret != -EAGAIN); 40687ae9bd18SNaohiro Aota 40695d4f98a2SYan Zheng while (1) { 4070430640e3SQu Wenruo int finishes_stage; 4071430640e3SQu Wenruo 407276dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 40735d4f98a2SYan Zheng ret = relocate_block_group(rc); 407476dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4075ff612ba7SJosef Bacik if (ret < 0) 40765d4f98a2SYan Zheng err = ret; 4077ff612ba7SJosef Bacik 4078430640e3SQu Wenruo finishes_stage = rc->stage; 4079ff612ba7SJosef Bacik /* 4080ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4081ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4082ff612ba7SJosef Bacik * different block group we could end up hitting the 4083ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4084ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4085ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4086ff612ba7SJosef Bacik * out of the loop if we hit an error. 4087ff612ba7SJosef Bacik */ 4088ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4089ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4090ff612ba7SJosef Bacik (u64)-1); 4091ff612ba7SJosef Bacik if (ret) 4092ff612ba7SJosef Bacik err = ret; 4093ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4094ff612ba7SJosef Bacik 0, -1); 4095ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 40965d4f98a2SYan Zheng } 40975d4f98a2SYan Zheng 4098ff612ba7SJosef Bacik if (err < 0) 4099ff612ba7SJosef Bacik goto out; 4100ff612ba7SJosef Bacik 41015d4f98a2SYan Zheng if (rc->extents_found == 0) 41025d4f98a2SYan Zheng break; 41035d4f98a2SYan Zheng 4104430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4105430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 41065d4f98a2SYan Zheng } 41075d4f98a2SYan Zheng 41085d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 41095d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4110bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 41115d4f98a2SYan Zheng out: 4112f0486c68SYan, Zheng if (err && rw) 41132ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 41145d4f98a2SYan Zheng iput(rc->data_inode); 4115907d2710SDavid Sterba out_put_bg: 4116907d2710SDavid Sterba btrfs_put_block_group(bg); 4117907d2710SDavid Sterba reloc_chunk_end(fs_info); 41181a0afa0eSJosef Bacik free_reloc_control(rc); 41195d4f98a2SYan Zheng return err; 41205d4f98a2SYan Zheng } 41215d4f98a2SYan Zheng 412276dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 412376dda93cSYan, Zheng { 41240b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 412576dda93cSYan, Zheng struct btrfs_trans_handle *trans; 412679787eaaSJeff Mahoney int ret, err; 412776dda93cSYan, Zheng 41280b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 412979787eaaSJeff Mahoney if (IS_ERR(trans)) 413079787eaaSJeff Mahoney return PTR_ERR(trans); 413176dda93cSYan, Zheng 413276dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 413376dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 4134c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 413576dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 41360b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 413776dda93cSYan, Zheng &root->root_key, &root->root_item); 413876dda93cSYan, Zheng 41393a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 414079787eaaSJeff Mahoney if (err) 414179787eaaSJeff Mahoney return err; 414279787eaaSJeff Mahoney return ret; 414376dda93cSYan, Zheng } 414476dda93cSYan, Zheng 41455d4f98a2SYan Zheng /* 41465d4f98a2SYan Zheng * recover relocation interrupted by system crash. 41475d4f98a2SYan Zheng * 41485d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 41495d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 41505d4f98a2SYan Zheng */ 41517eefae6bSJosef Bacik int btrfs_recover_relocation(struct btrfs_fs_info *fs_info) 41525d4f98a2SYan Zheng { 41535d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 41545d4f98a2SYan Zheng struct btrfs_key key; 41555d4f98a2SYan Zheng struct btrfs_root *fs_root; 41565d4f98a2SYan Zheng struct btrfs_root *reloc_root; 41575d4f98a2SYan Zheng struct btrfs_path *path; 41585d4f98a2SYan Zheng struct extent_buffer *leaf; 41595d4f98a2SYan Zheng struct reloc_control *rc = NULL; 41605d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41615d4f98a2SYan Zheng int ret; 41625d4f98a2SYan Zheng int err = 0; 41635d4f98a2SYan Zheng 41645d4f98a2SYan Zheng path = btrfs_alloc_path(); 41655d4f98a2SYan Zheng if (!path) 41665d4f98a2SYan Zheng return -ENOMEM; 4167e4058b54SDavid Sterba path->reada = READA_BACK; 41685d4f98a2SYan Zheng 41695d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 41705d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 41715d4f98a2SYan Zheng key.offset = (u64)-1; 41725d4f98a2SYan Zheng 41735d4f98a2SYan Zheng while (1) { 41740b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 41755d4f98a2SYan Zheng path, 0, 0); 41765d4f98a2SYan Zheng if (ret < 0) { 41775d4f98a2SYan Zheng err = ret; 41785d4f98a2SYan Zheng goto out; 41795d4f98a2SYan Zheng } 41805d4f98a2SYan Zheng if (ret > 0) { 41815d4f98a2SYan Zheng if (path->slots[0] == 0) 41825d4f98a2SYan Zheng break; 41835d4f98a2SYan Zheng path->slots[0]--; 41845d4f98a2SYan Zheng } 41855d4f98a2SYan Zheng leaf = path->nodes[0]; 41865d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4187b3b4aa74SDavid Sterba btrfs_release_path(path); 41885d4f98a2SYan Zheng 41895d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 41905d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 41915d4f98a2SYan Zheng break; 41925d4f98a2SYan Zheng 41937eefae6bSJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key); 41945d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 41955d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 41965d4f98a2SYan Zheng goto out; 41975d4f98a2SYan Zheng } 41985d4f98a2SYan Zheng 419992a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 42005d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 42015d4f98a2SYan Zheng 42025d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 4203a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, 4204a820feb5SDavid Sterba reloc_root->root_key.offset, false); 42055d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 420676dda93cSYan, Zheng ret = PTR_ERR(fs_root); 420776dda93cSYan, Zheng if (ret != -ENOENT) { 420876dda93cSYan, Zheng err = ret; 42095d4f98a2SYan Zheng goto out; 42105d4f98a2SYan Zheng } 421179787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 421279787eaaSJeff Mahoney if (ret < 0) { 421379787eaaSJeff Mahoney err = ret; 421479787eaaSJeff Mahoney goto out; 421579787eaaSJeff Mahoney } 4216932fd26dSJosef Bacik } else { 421700246528SJosef Bacik btrfs_put_root(fs_root); 421876dda93cSYan, Zheng } 42195d4f98a2SYan Zheng } 42205d4f98a2SYan Zheng 42215d4f98a2SYan Zheng if (key.offset == 0) 42225d4f98a2SYan Zheng break; 42235d4f98a2SYan Zheng 42245d4f98a2SYan Zheng key.offset--; 42255d4f98a2SYan Zheng } 4226b3b4aa74SDavid Sterba btrfs_release_path(path); 42275d4f98a2SYan Zheng 42285d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 42295d4f98a2SYan Zheng goto out; 42305d4f98a2SYan Zheng 4231c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 42325d4f98a2SYan Zheng if (!rc) { 42335d4f98a2SYan Zheng err = -ENOMEM; 42345d4f98a2SYan Zheng goto out; 42355d4f98a2SYan Zheng } 42365d4f98a2SYan Zheng 4237907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4238907d2710SDavid Sterba if (ret < 0) { 4239907d2710SDavid Sterba err = ret; 4240907d2710SDavid Sterba goto out_end; 4241907d2710SDavid Sterba } 4242907d2710SDavid Sterba 424329cbcf40SJosef Bacik rc->extent_root = btrfs_extent_root(fs_info, 0); 42445d4f98a2SYan Zheng 42455d4f98a2SYan Zheng set_reloc_control(rc); 42465d4f98a2SYan Zheng 42477a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 42483612b495STsutomu Itoh if (IS_ERR(trans)) { 42493612b495STsutomu Itoh err = PTR_ERR(trans); 4250fb2d83eeSJosef Bacik goto out_unset; 42513612b495STsutomu Itoh } 42523fd0a558SYan, Zheng 42533fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 42543fd0a558SYan, Zheng 42555d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 42565d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 42575d4f98a2SYan Zheng struct btrfs_root, root_list); 42585d4f98a2SYan Zheng list_del(&reloc_root->root_list); 42595d4f98a2SYan Zheng 42605d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 42615d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 42625d4f98a2SYan Zheng &rc->reloc_roots); 42635d4f98a2SYan Zheng continue; 42645d4f98a2SYan Zheng } 42655d4f98a2SYan Zheng 4266a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 4267a820feb5SDavid Sterba false); 426879787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 426979787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4270ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42711402d17dSXiyu Yang btrfs_end_transaction(trans); 4272fb2d83eeSJosef Bacik goto out_unset; 427379787eaaSJeff Mahoney } 42745d4f98a2SYan Zheng 4275ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 427657a304cfSJosef Bacik ASSERT(err != -EEXIST); 42773c925863SJosef Bacik if (err) { 42783c925863SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42793c925863SJosef Bacik btrfs_put_root(fs_root); 42803c925863SJosef Bacik btrfs_end_transaction(trans); 42813c925863SJosef Bacik goto out_unset; 42823c925863SJosef Bacik } 4283f44deb74SJosef Bacik fs_root->reloc_root = btrfs_grab_root(reloc_root); 428400246528SJosef Bacik btrfs_put_root(fs_root); 42855d4f98a2SYan Zheng } 42865d4f98a2SYan Zheng 42873a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 428879787eaaSJeff Mahoney if (err) 4289fb2d83eeSJosef Bacik goto out_unset; 42905d4f98a2SYan Zheng 42915d4f98a2SYan Zheng merge_reloc_roots(rc); 42925d4f98a2SYan Zheng 42935d4f98a2SYan Zheng unset_reloc_control(rc); 42945d4f98a2SYan Zheng 42957a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 429662b99540SQu Wenruo if (IS_ERR(trans)) { 42973612b495STsutomu Itoh err = PTR_ERR(trans); 42986217b0faSJosef Bacik goto out_clean; 429962b99540SQu Wenruo } 43003a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 43016217b0faSJosef Bacik out_clean: 4302d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4303d2311e69SQu Wenruo if (ret < 0 && !err) 4304d2311e69SQu Wenruo err = ret; 4305fb2d83eeSJosef Bacik out_unset: 4306fb2d83eeSJosef Bacik unset_reloc_control(rc); 4307907d2710SDavid Sterba out_end: 4308907d2710SDavid Sterba reloc_chunk_end(fs_info); 43091a0afa0eSJosef Bacik free_reloc_control(rc); 43103612b495STsutomu Itoh out: 4311aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4312aca1bba6SLiu Bo 43135d4f98a2SYan Zheng btrfs_free_path(path); 43145d4f98a2SYan Zheng 43155d4f98a2SYan Zheng if (err == 0) { 43165d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 4317aeb935a4SQu Wenruo fs_root = btrfs_grab_root(fs_info->data_reloc_root); 4318aeb935a4SQu Wenruo ASSERT(fs_root); 431966b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 432000246528SJosef Bacik btrfs_put_root(fs_root); 4321932fd26dSJosef Bacik } 43225d4f98a2SYan Zheng return err; 43235d4f98a2SYan Zheng } 43245d4f98a2SYan Zheng 43255d4f98a2SYan Zheng /* 43265d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 43275d4f98a2SYan Zheng * 43285d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 43295d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 43305d4f98a2SYan Zheng */ 43317bfa9535SNikolay Borisov int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len) 43325d4f98a2SYan Zheng { 43337bfa9535SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 4334fc28b25eSJosef Bacik struct btrfs_root *csum_root; 43355d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 43365d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 43375d4f98a2SYan Zheng int ret; 43385d4f98a2SYan Zheng u64 disk_bytenr; 43394577b014SJosef Bacik u64 new_bytenr; 43405d4f98a2SYan Zheng LIST_HEAD(list); 43415d4f98a2SYan Zheng 43427bfa9535SNikolay Borisov ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4343bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 43445d4f98a2SYan Zheng 43457bfa9535SNikolay Borisov disk_bytenr = file_pos + inode->index_cnt; 4346fc28b25eSJosef Bacik csum_root = btrfs_csum_root(fs_info, disk_bytenr); 4347fc28b25eSJosef Bacik ret = btrfs_lookup_csums_range(csum_root, disk_bytenr, 434826ce9114SJosef Bacik disk_bytenr + len - 1, &list, 0, false); 434979787eaaSJeff Mahoney if (ret) 435079787eaaSJeff Mahoney goto out; 43515d4f98a2SYan Zheng 43525d4f98a2SYan Zheng while (!list_empty(&list)) { 43535d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 43545d4f98a2SYan Zheng list_del_init(&sums->list); 43555d4f98a2SYan Zheng 43564577b014SJosef Bacik /* 43574577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 43584577b014SJosef Bacik * We need to do this because we will read in entire prealloc 43594577b014SJosef Bacik * extents but we may have written to say the middle of the 43604577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 43614577b014SJosef Bacik * the right disk offset. 43624577b014SJosef Bacik * 43634577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 43644577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 43654577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 43664577b014SJosef Bacik * disk length. 43674577b014SJosef Bacik */ 4368bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 43694577b014SJosef Bacik sums->bytenr = new_bytenr; 43705d4f98a2SYan Zheng 4371f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 43725d4f98a2SYan Zheng } 437379787eaaSJeff Mahoney out: 43745d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4375411fc6bcSAndi Kleen return ret; 43765d4f98a2SYan Zheng } 43773fd0a558SYan, Zheng 437883d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 43793fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 43803fd0a558SYan, Zheng struct extent_buffer *cow) 43813fd0a558SYan, Zheng { 43820b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 43833fd0a558SYan, Zheng struct reloc_control *rc; 4384a26195a5SQu Wenruo struct btrfs_backref_node *node; 43853fd0a558SYan, Zheng int first_cow = 0; 43863fd0a558SYan, Zheng int level; 438783d4cfd4SJosef Bacik int ret = 0; 43883fd0a558SYan, Zheng 43890b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 43903fd0a558SYan, Zheng if (!rc) 439183d4cfd4SJosef Bacik return 0; 43923fd0a558SYan, Zheng 439337f00a6dSJohannes Thumshirn BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root)); 43943fd0a558SYan, Zheng 43953fd0a558SYan, Zheng level = btrfs_header_level(buf); 43963fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 43973fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 43983fd0a558SYan, Zheng first_cow = 1; 43993fd0a558SYan, Zheng 44003fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 44013fd0a558SYan, Zheng rc->create_reloc_tree) { 44023fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 44033fd0a558SYan, Zheng 44043fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 44053fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 44063fd0a558SYan, Zheng node->new_bytenr != buf->start); 44073fd0a558SYan, Zheng 4408b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 440967439dadSDavid Sterba atomic_inc(&cow->refs); 44103fd0a558SYan, Zheng node->eb = cow; 44113fd0a558SYan, Zheng node->new_bytenr = cow->start; 44123fd0a558SYan, Zheng 44133fd0a558SYan, Zheng if (!node->pending) { 44143fd0a558SYan, Zheng list_move_tail(&node->list, 44153fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 44163fd0a558SYan, Zheng node->pending = 1; 44173fd0a558SYan, Zheng } 44183fd0a558SYan, Zheng 44193fd0a558SYan, Zheng if (first_cow) 44209569cc20SQu Wenruo mark_block_processed(rc, node); 44213fd0a558SYan, Zheng 44223fd0a558SYan, Zheng if (first_cow && level > 0) 44233fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 44243fd0a558SYan, Zheng } 44253fd0a558SYan, Zheng 442683d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 44273fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 442883d4cfd4SJosef Bacik return ret; 44293fd0a558SYan, Zheng } 44303fd0a558SYan, Zheng 44313fd0a558SYan, Zheng /* 44323fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 443301327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 44343fd0a558SYan, Zheng */ 4435147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 44363fd0a558SYan, Zheng u64 *bytes_to_reserve) 44373fd0a558SYan, Zheng { 443810995c04SQu Wenruo struct btrfs_root *root = pending->root; 443910995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44403fd0a558SYan, Zheng 44416282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 44423fd0a558SYan, Zheng return; 44433fd0a558SYan, Zheng 44443fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 44453fd0a558SYan, Zheng return; 44463fd0a558SYan, Zheng 44473fd0a558SYan, Zheng root = root->reloc_root; 44483fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 44493fd0a558SYan, Zheng /* 44503fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 44513fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 44523fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 44533fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 44543fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 44553fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 44563fd0a558SYan, Zheng * tree may use more space than it frees. so we need 44573fd0a558SYan, Zheng * reserve extra space. 44583fd0a558SYan, Zheng */ 44593fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 44603fd0a558SYan, Zheng } 44613fd0a558SYan, Zheng 44623fd0a558SYan, Zheng /* 44633fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 44643fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 4465f44deb74SJosef Bacik * 4466f44deb74SJosef Bacik * This is similar to btrfs_init_reloc_root(), we come out of here with two 4467f44deb74SJosef Bacik * references held on the reloc_root, one for root->reloc_root and one for 4468f44deb74SJosef Bacik * rc->reloc_roots. 44693fd0a558SYan, Zheng */ 447049b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 44713fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 44723fd0a558SYan, Zheng { 44733fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 44743fd0a558SYan, Zheng struct btrfs_root *reloc_root; 44753fd0a558SYan, Zheng struct btrfs_root *new_root; 447610995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44773fd0a558SYan, Zheng int ret; 44783fd0a558SYan, Zheng 44796282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 448049b25e05SJeff Mahoney return 0; 44813fd0a558SYan, Zheng 44823fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 44833fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 44843fd0a558SYan, Zheng 44853fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 44863fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 44873fd0a558SYan, Zheng rc->block_rsv, 44883a584174SLu Fengqi rc->nodes_relocated, true); 448949b25e05SJeff Mahoney if (ret) 449049b25e05SJeff Mahoney return ret; 44913fd0a558SYan, Zheng } 44923fd0a558SYan, Zheng 44933fd0a558SYan, Zheng new_root = pending->snap; 44943fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 44953fd0a558SYan, Zheng new_root->root_key.objectid); 449649b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 449749b25e05SJeff Mahoney return PTR_ERR(reloc_root); 44983fd0a558SYan, Zheng 4499ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 450057a304cfSJosef Bacik ASSERT(ret != -EEXIST); 45013c925863SJosef Bacik if (ret) { 45023c925863SJosef Bacik /* Pairs with create_reloc_root */ 45033c925863SJosef Bacik btrfs_put_root(reloc_root); 45043c925863SJosef Bacik return ret; 45053c925863SJosef Bacik } 4506f44deb74SJosef Bacik new_root->reloc_root = btrfs_grab_root(reloc_root); 45073fd0a558SYan, Zheng 450849b25e05SJeff Mahoney if (rc->create_reloc_tree) 45093fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 451049b25e05SJeff Mahoney return ret; 45113fd0a558SYan, Zheng } 4512