1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 25d4f98a2SYan Zheng /* 35d4f98a2SYan Zheng * Copyright (C) 2009 Oracle. All rights reserved. 45d4f98a2SYan Zheng */ 55d4f98a2SYan Zheng 65d4f98a2SYan Zheng #include <linux/sched.h> 75d4f98a2SYan Zheng #include <linux/pagemap.h> 85d4f98a2SYan Zheng #include <linux/writeback.h> 95d4f98a2SYan Zheng #include <linux/blkdev.h> 105d4f98a2SYan Zheng #include <linux/rbtree.h> 115a0e3ad6STejun Heo #include <linux/slab.h> 12726a3421SQu Wenruo #include <linux/error-injection.h> 135d4f98a2SYan Zheng #include "ctree.h" 145d4f98a2SYan Zheng #include "disk-io.h" 155d4f98a2SYan Zheng #include "transaction.h" 165d4f98a2SYan Zheng #include "volumes.h" 175d4f98a2SYan Zheng #include "locking.h" 185d4f98a2SYan Zheng #include "btrfs_inode.h" 195d4f98a2SYan Zheng #include "async-thread.h" 200af3d00bSJosef Bacik #include "free-space-cache.h" 2162b99540SQu Wenruo #include "qgroup.h" 22cdccee99SLiu Bo #include "print-tree.h" 2386736342SJosef Bacik #include "delalloc-space.h" 24aac0023cSJosef Bacik #include "block-group.h" 2519b546d7SQu Wenruo #include "backref.h" 26e9a28dc5SQu Wenruo #include "misc.h" 27c2832898SQu Wenruo #include "subpage.h" 287ae9bd18SNaohiro Aota #include "zoned.h" 2926c2c454SJosef Bacik #include "inode-item.h" 30f1e5c618SJosef Bacik #include "space-info.h" 31c7f13d42SJosef Bacik #include "fs.h" 3207e81dc9SJosef Bacik #include "accessors.h" 33a0231804SJosef Bacik #include "extent-tree.h" 3445c40c8fSJosef Bacik #include "root-tree.h" 357c8ede16SJosef Bacik #include "file-item.h" 3667707479SJosef Bacik #include "relocation.h" 377f0add25SJosef Bacik #include "super.h" 385d4f98a2SYan Zheng 395d4f98a2SYan Zheng /* 400c891389SQu Wenruo * Relocation overview 410c891389SQu Wenruo * 420c891389SQu Wenruo * [What does relocation do] 430c891389SQu Wenruo * 440c891389SQu Wenruo * The objective of relocation is to relocate all extents of the target block 450c891389SQu Wenruo * group to other block groups. 460c891389SQu Wenruo * This is utilized by resize (shrink only), profile converting, compacting 470c891389SQu Wenruo * space, or balance routine to spread chunks over devices. 480c891389SQu Wenruo * 490c891389SQu Wenruo * Before | After 500c891389SQu Wenruo * ------------------------------------------------------------------ 510c891389SQu Wenruo * BG A: 10 data extents | BG A: deleted 520c891389SQu Wenruo * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) 530c891389SQu Wenruo * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) 540c891389SQu Wenruo * 550c891389SQu Wenruo * [How does relocation work] 560c891389SQu Wenruo * 570c891389SQu Wenruo * 1. Mark the target block group read-only 580c891389SQu Wenruo * New extents won't be allocated from the target block group. 590c891389SQu Wenruo * 600c891389SQu Wenruo * 2.1 Record each extent in the target block group 610c891389SQu Wenruo * To build a proper map of extents to be relocated. 620c891389SQu Wenruo * 630c891389SQu Wenruo * 2.2 Build data reloc tree and reloc trees 640c891389SQu Wenruo * Data reloc tree will contain an inode, recording all newly relocated 650c891389SQu Wenruo * data extents. 660c891389SQu Wenruo * There will be only one data reloc tree for one data block group. 670c891389SQu Wenruo * 680c891389SQu Wenruo * Reloc tree will be a special snapshot of its source tree, containing 690c891389SQu Wenruo * relocated tree blocks. 700c891389SQu Wenruo * Each tree referring to a tree block in target block group will get its 710c891389SQu Wenruo * reloc tree built. 720c891389SQu Wenruo * 730c891389SQu Wenruo * 2.3 Swap source tree with its corresponding reloc tree 740c891389SQu Wenruo * Each involved tree only refers to new extents after swap. 750c891389SQu Wenruo * 760c891389SQu Wenruo * 3. Cleanup reloc trees and data reloc tree. 770c891389SQu Wenruo * As old extents in the target block group are still referenced by reloc 780c891389SQu Wenruo * trees, we need to clean them up before really freeing the target block 790c891389SQu Wenruo * group. 800c891389SQu Wenruo * 810c891389SQu Wenruo * The main complexity is in steps 2.2 and 2.3. 820c891389SQu Wenruo * 830c891389SQu Wenruo * The entry point of relocation is relocate_block_group() function. 840c891389SQu Wenruo */ 850c891389SQu Wenruo 860647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 872a979612SQu Wenruo /* 885d4f98a2SYan Zheng * map address of tree root to tree 895d4f98a2SYan Zheng */ 905d4f98a2SYan Zheng struct mapping_node { 91e9a28dc5SQu Wenruo struct { 925d4f98a2SYan Zheng struct rb_node rb_node; 935d4f98a2SYan Zheng u64 bytenr; 94e9a28dc5SQu Wenruo }; /* Use rb_simle_node for search/insert */ 955d4f98a2SYan Zheng void *data; 965d4f98a2SYan Zheng }; 975d4f98a2SYan Zheng 985d4f98a2SYan Zheng struct mapping_tree { 995d4f98a2SYan Zheng struct rb_root rb_root; 1005d4f98a2SYan Zheng spinlock_t lock; 1015d4f98a2SYan Zheng }; 1025d4f98a2SYan Zheng 1035d4f98a2SYan Zheng /* 1045d4f98a2SYan Zheng * present a tree block to process 1055d4f98a2SYan Zheng */ 1065d4f98a2SYan Zheng struct tree_block { 107e9a28dc5SQu Wenruo struct { 1085d4f98a2SYan Zheng struct rb_node rb_node; 1095d4f98a2SYan Zheng u64 bytenr; 110e9a28dc5SQu Wenruo }; /* Use rb_simple_node for search/insert */ 111f7ba2d37SJosef Bacik u64 owner; 1125d4f98a2SYan Zheng struct btrfs_key key; 1135d4f98a2SYan Zheng unsigned int level:8; 1145d4f98a2SYan Zheng unsigned int key_ready:1; 1155d4f98a2SYan Zheng }; 1165d4f98a2SYan Zheng 1170257bb82SYan, Zheng #define MAX_EXTENTS 128 1180257bb82SYan, Zheng 1190257bb82SYan, Zheng struct file_extent_cluster { 1200257bb82SYan, Zheng u64 start; 1210257bb82SYan, Zheng u64 end; 1220257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1230257bb82SYan, Zheng unsigned int nr; 1240257bb82SYan, Zheng }; 1250257bb82SYan, Zheng 1265d4f98a2SYan Zheng struct reloc_control { 1275d4f98a2SYan Zheng /* block group to relocate */ 12832da5386SDavid Sterba struct btrfs_block_group *block_group; 1295d4f98a2SYan Zheng /* extent tree */ 1305d4f98a2SYan Zheng struct btrfs_root *extent_root; 1315d4f98a2SYan Zheng /* inode for moving data */ 1325d4f98a2SYan Zheng struct inode *data_inode; 1333fd0a558SYan, Zheng 1343fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1353fd0a558SYan, Zheng 136a26195a5SQu Wenruo struct btrfs_backref_cache backref_cache; 1373fd0a558SYan, Zheng 1383fd0a558SYan, Zheng struct file_extent_cluster cluster; 1395d4f98a2SYan Zheng /* tree blocks have been processed */ 1405d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1415d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1425d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1435d4f98a2SYan Zheng /* list of reloc trees */ 1445d4f98a2SYan Zheng struct list_head reloc_roots; 145d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 146d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 1473fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1483fd0a558SYan, Zheng u64 merging_rsv_size; 1493fd0a558SYan, Zheng /* size of relocated tree nodes */ 1503fd0a558SYan, Zheng u64 nodes_relocated; 1510647bf56SWang Shilong /* reserved size for block group relocation*/ 1520647bf56SWang Shilong u64 reserved_bytes; 1533fd0a558SYan, Zheng 1545d4f98a2SYan Zheng u64 search_start; 1555d4f98a2SYan Zheng u64 extents_found; 1563fd0a558SYan, Zheng 1573fd0a558SYan, Zheng unsigned int stage:8; 1583fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1593fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1605d4f98a2SYan Zheng unsigned int found_file_extent:1; 1615d4f98a2SYan Zheng }; 1625d4f98a2SYan Zheng 1635d4f98a2SYan Zheng /* stages of data relocation */ 1645d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1655d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1665d4f98a2SYan Zheng 1679569cc20SQu Wenruo static void mark_block_processed(struct reloc_control *rc, 168a26195a5SQu Wenruo struct btrfs_backref_node *node) 1699569cc20SQu Wenruo { 1709569cc20SQu Wenruo u32 blocksize; 1719569cc20SQu Wenruo 1729569cc20SQu Wenruo if (node->level == 0 || 1739569cc20SQu Wenruo in_range(node->bytenr, rc->block_group->start, 1749569cc20SQu Wenruo rc->block_group->length)) { 1759569cc20SQu Wenruo blocksize = rc->extent_root->fs_info->nodesize; 1769569cc20SQu Wenruo set_extent_bits(&rc->processed_blocks, node->bytenr, 1779569cc20SQu Wenruo node->bytenr + blocksize - 1, EXTENT_DIRTY); 1789569cc20SQu Wenruo } 1799569cc20SQu Wenruo node->processed = 1; 1809569cc20SQu Wenruo } 1819569cc20SQu Wenruo 1825d4f98a2SYan Zheng 1835d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1845d4f98a2SYan Zheng { 1856bef4d31SEric Paris tree->rb_root = RB_ROOT; 1865d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1875d4f98a2SYan Zheng } 1885d4f98a2SYan Zheng 1895d4f98a2SYan Zheng /* 1905d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 1915d4f98a2SYan Zheng */ 192a26195a5SQu Wenruo static struct btrfs_backref_node *walk_up_backref( 193a26195a5SQu Wenruo struct btrfs_backref_node *node, 194a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 1955d4f98a2SYan Zheng { 196a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 1975d4f98a2SYan Zheng int idx = *index; 1985d4f98a2SYan Zheng 1995d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 2005d4f98a2SYan Zheng edge = list_entry(node->upper.next, 201a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 2025d4f98a2SYan Zheng edges[idx++] = edge; 2035d4f98a2SYan Zheng node = edge->node[UPPER]; 2045d4f98a2SYan Zheng } 2053fd0a558SYan, Zheng BUG_ON(node->detached); 2065d4f98a2SYan Zheng *index = idx; 2075d4f98a2SYan Zheng return node; 2085d4f98a2SYan Zheng } 2095d4f98a2SYan Zheng 2105d4f98a2SYan Zheng /* 2115d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 2125d4f98a2SYan Zheng */ 213a26195a5SQu Wenruo static struct btrfs_backref_node *walk_down_backref( 214a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 2155d4f98a2SYan Zheng { 216a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 217a26195a5SQu Wenruo struct btrfs_backref_node *lower; 2185d4f98a2SYan Zheng int idx = *index; 2195d4f98a2SYan Zheng 2205d4f98a2SYan Zheng while (idx > 0) { 2215d4f98a2SYan Zheng edge = edges[idx - 1]; 2225d4f98a2SYan Zheng lower = edge->node[LOWER]; 2235d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 2245d4f98a2SYan Zheng idx--; 2255d4f98a2SYan Zheng continue; 2265d4f98a2SYan Zheng } 2275d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 228a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 2295d4f98a2SYan Zheng edges[idx - 1] = edge; 2305d4f98a2SYan Zheng *index = idx; 2315d4f98a2SYan Zheng return edge->node[UPPER]; 2325d4f98a2SYan Zheng } 2335d4f98a2SYan Zheng *index = 0; 2345d4f98a2SYan Zheng return NULL; 2355d4f98a2SYan Zheng } 2365d4f98a2SYan Zheng 237a26195a5SQu Wenruo static void update_backref_node(struct btrfs_backref_cache *cache, 238a26195a5SQu Wenruo struct btrfs_backref_node *node, u64 bytenr) 2393fd0a558SYan, Zheng { 2403fd0a558SYan, Zheng struct rb_node *rb_node; 2413fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 2423fd0a558SYan, Zheng node->bytenr = bytenr; 243e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node); 24443c04fb1SJeff Mahoney if (rb_node) 245982c92cbSQu Wenruo btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST); 2463fd0a558SYan, Zheng } 2473fd0a558SYan, Zheng 2483fd0a558SYan, Zheng /* 2493fd0a558SYan, Zheng * update backref cache after a transaction commit 2503fd0a558SYan, Zheng */ 2513fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 252a26195a5SQu Wenruo struct btrfs_backref_cache *cache) 2533fd0a558SYan, Zheng { 254a26195a5SQu Wenruo struct btrfs_backref_node *node; 2553fd0a558SYan, Zheng int level = 0; 2563fd0a558SYan, Zheng 2573fd0a558SYan, Zheng if (cache->last_trans == 0) { 2583fd0a558SYan, Zheng cache->last_trans = trans->transid; 2593fd0a558SYan, Zheng return 0; 2603fd0a558SYan, Zheng } 2613fd0a558SYan, Zheng 2623fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 2633fd0a558SYan, Zheng return 0; 2643fd0a558SYan, Zheng 2653fd0a558SYan, Zheng /* 2663fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 2673fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 2683fd0a558SYan, Zheng * so the detached nodes are no longer useful. 2693fd0a558SYan, Zheng */ 2703fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2713fd0a558SYan, Zheng node = list_entry(cache->detached.next, 272a26195a5SQu Wenruo struct btrfs_backref_node, list); 273023acb07SQu Wenruo btrfs_backref_cleanup_node(cache, node); 2743fd0a558SYan, Zheng } 2753fd0a558SYan, Zheng 2763fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 2773fd0a558SYan, Zheng node = list_entry(cache->changed.next, 278a26195a5SQu Wenruo struct btrfs_backref_node, list); 2793fd0a558SYan, Zheng list_del_init(&node->list); 2803fd0a558SYan, Zheng BUG_ON(node->pending); 2813fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2823fd0a558SYan, Zheng } 2833fd0a558SYan, Zheng 2843fd0a558SYan, Zheng /* 2853fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 2863fd0a558SYan, Zheng * errors during processing the pending nodes. 2873fd0a558SYan, Zheng */ 2883fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 2893fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 2903fd0a558SYan, Zheng BUG_ON(!node->pending); 2913fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 2923fd0a558SYan, Zheng continue; 2933fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2943fd0a558SYan, Zheng } 2953fd0a558SYan, Zheng } 2963fd0a558SYan, Zheng 2973fd0a558SYan, Zheng cache->last_trans = 0; 2983fd0a558SYan, Zheng return 1; 2993fd0a558SYan, Zheng } 3003fd0a558SYan, Zheng 3016282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 3026282675eSQu Wenruo { 3036282675eSQu Wenruo /* 3046282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 3056282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 3066282675eSQu Wenruo * trying to access reloc_root 3076282675eSQu Wenruo */ 3086282675eSQu Wenruo smp_rmb(); 3096282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 3106282675eSQu Wenruo return true; 3116282675eSQu Wenruo return false; 3126282675eSQu Wenruo } 3136282675eSQu Wenruo 3146282675eSQu Wenruo /* 3156282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 3166282675eSQu Wenruo * 3176282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 3186282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 31955465730SQu Wenruo * from no reloc root. But btrfs_should_ignore_reloc_root() below is a 32055465730SQu Wenruo * special case. 3216282675eSQu Wenruo */ 3226282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 3236282675eSQu Wenruo { 3246282675eSQu Wenruo if (reloc_root_is_dead(root)) 3256282675eSQu Wenruo return false; 3266282675eSQu Wenruo if (!root->reloc_root) 3276282675eSQu Wenruo return false; 3286282675eSQu Wenruo return true; 3296282675eSQu Wenruo } 330f2a97a9dSDavid Sterba 33155465730SQu Wenruo int btrfs_should_ignore_reloc_root(struct btrfs_root *root) 3323fd0a558SYan, Zheng { 3333fd0a558SYan, Zheng struct btrfs_root *reloc_root; 3343fd0a558SYan, Zheng 33592a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 3363fd0a558SYan, Zheng return 0; 3373fd0a558SYan, Zheng 3386282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 3396282675eSQu Wenruo if (reloc_root_is_dead(root)) 3406282675eSQu Wenruo return 1; 3416282675eSQu Wenruo 3423fd0a558SYan, Zheng reloc_root = root->reloc_root; 3433fd0a558SYan, Zheng if (!reloc_root) 3443fd0a558SYan, Zheng return 0; 3453fd0a558SYan, Zheng 3464d4225fcSJosef Bacik if (btrfs_header_generation(reloc_root->commit_root) == 3474d4225fcSJosef Bacik root->fs_info->running_transaction->transid) 3483fd0a558SYan, Zheng return 0; 3493fd0a558SYan, Zheng /* 3503fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 3513fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 3523fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 3533fd0a558SYan, Zheng * relocation. 3543fd0a558SYan, Zheng */ 3553fd0a558SYan, Zheng return 1; 3563fd0a558SYan, Zheng } 35755465730SQu Wenruo 3585d4f98a2SYan Zheng /* 3595d4f98a2SYan Zheng * find reloc tree by address of tree root 3605d4f98a2SYan Zheng */ 3612433bea5SQu Wenruo struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) 3625d4f98a2SYan Zheng { 3632433bea5SQu Wenruo struct reloc_control *rc = fs_info->reloc_ctl; 3645d4f98a2SYan Zheng struct rb_node *rb_node; 3655d4f98a2SYan Zheng struct mapping_node *node; 3665d4f98a2SYan Zheng struct btrfs_root *root = NULL; 3675d4f98a2SYan Zheng 3682433bea5SQu Wenruo ASSERT(rc); 3695d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 370e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr); 3715d4f98a2SYan Zheng if (rb_node) { 3725d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 3730d031dc4SYu Zhe root = node->data; 3745d4f98a2SYan Zheng } 3755d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 37600246528SJosef Bacik return btrfs_grab_root(root); 3775d4f98a2SYan Zheng } 3785d4f98a2SYan Zheng 3795d4f98a2SYan Zheng /* 38029db137bSQu Wenruo * For useless nodes, do two major clean ups: 38129db137bSQu Wenruo * 38229db137bSQu Wenruo * - Cleanup the children edges and nodes 38329db137bSQu Wenruo * If child node is also orphan (no parent) during cleanup, then the child 38429db137bSQu Wenruo * node will also be cleaned up. 38529db137bSQu Wenruo * 38629db137bSQu Wenruo * - Freeing up leaves (level 0), keeps nodes detached 38729db137bSQu Wenruo * For nodes, the node is still cached as "detached" 38829db137bSQu Wenruo * 38929db137bSQu Wenruo * Return false if @node is not in the @useless_nodes list. 39029db137bSQu Wenruo * Return true if @node is in the @useless_nodes list. 39129db137bSQu Wenruo */ 39229db137bSQu Wenruo static bool handle_useless_nodes(struct reloc_control *rc, 393a26195a5SQu Wenruo struct btrfs_backref_node *node) 39429db137bSQu Wenruo { 395a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 39629db137bSQu Wenruo struct list_head *useless_node = &cache->useless_node; 39729db137bSQu Wenruo bool ret = false; 39829db137bSQu Wenruo 39929db137bSQu Wenruo while (!list_empty(useless_node)) { 400a26195a5SQu Wenruo struct btrfs_backref_node *cur; 40129db137bSQu Wenruo 402a26195a5SQu Wenruo cur = list_first_entry(useless_node, struct btrfs_backref_node, 40329db137bSQu Wenruo list); 40429db137bSQu Wenruo list_del_init(&cur->list); 40529db137bSQu Wenruo 40629db137bSQu Wenruo /* Only tree root nodes can be added to @useless_nodes */ 40729db137bSQu Wenruo ASSERT(list_empty(&cur->upper)); 40829db137bSQu Wenruo 40929db137bSQu Wenruo if (cur == node) 41029db137bSQu Wenruo ret = true; 41129db137bSQu Wenruo 41229db137bSQu Wenruo /* The node is the lowest node */ 41329db137bSQu Wenruo if (cur->lowest) { 41429db137bSQu Wenruo list_del_init(&cur->lower); 41529db137bSQu Wenruo cur->lowest = 0; 41629db137bSQu Wenruo } 41729db137bSQu Wenruo 41829db137bSQu Wenruo /* Cleanup the lower edges */ 41929db137bSQu Wenruo while (!list_empty(&cur->lower)) { 420a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 421a26195a5SQu Wenruo struct btrfs_backref_node *lower; 42229db137bSQu Wenruo 42329db137bSQu Wenruo edge = list_entry(cur->lower.next, 424a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 42529db137bSQu Wenruo list_del(&edge->list[UPPER]); 42629db137bSQu Wenruo list_del(&edge->list[LOWER]); 42729db137bSQu Wenruo lower = edge->node[LOWER]; 428741188d3SQu Wenruo btrfs_backref_free_edge(cache, edge); 42929db137bSQu Wenruo 43029db137bSQu Wenruo /* Child node is also orphan, queue for cleanup */ 43129db137bSQu Wenruo if (list_empty(&lower->upper)) 43229db137bSQu Wenruo list_add(&lower->list, useless_node); 43329db137bSQu Wenruo } 43429db137bSQu Wenruo /* Mark this block processed for relocation */ 43529db137bSQu Wenruo mark_block_processed(rc, cur); 43629db137bSQu Wenruo 43729db137bSQu Wenruo /* 43829db137bSQu Wenruo * Backref nodes for tree leaves are deleted from the cache. 43929db137bSQu Wenruo * Backref nodes for upper level tree blocks are left in the 44029db137bSQu Wenruo * cache to avoid unnecessary backref lookup. 44129db137bSQu Wenruo */ 44229db137bSQu Wenruo if (cur->level > 0) { 44329db137bSQu Wenruo list_add(&cur->list, &cache->detached); 44429db137bSQu Wenruo cur->detached = 1; 44529db137bSQu Wenruo } else { 44629db137bSQu Wenruo rb_erase(&cur->rb_node, &cache->rb_root); 447741188d3SQu Wenruo btrfs_backref_free_node(cache, cur); 44829db137bSQu Wenruo } 44929db137bSQu Wenruo } 45029db137bSQu Wenruo return ret; 45129db137bSQu Wenruo } 45229db137bSQu Wenruo 45329db137bSQu Wenruo /* 454e7d571c7SQu Wenruo * Build backref tree for a given tree block. Root of the backref tree 455e7d571c7SQu Wenruo * corresponds the tree block, leaves of the backref tree correspond roots of 456e7d571c7SQu Wenruo * b-trees that reference the tree block. 457e7d571c7SQu Wenruo * 458e7d571c7SQu Wenruo * The basic idea of this function is check backrefs of a given block to find 459e7d571c7SQu Wenruo * upper level blocks that reference the block, and then check backrefs of 460e7d571c7SQu Wenruo * these upper level blocks recursively. The recursion stops when tree root is 461e7d571c7SQu Wenruo * reached or backrefs for the block is cached. 462e7d571c7SQu Wenruo * 463e7d571c7SQu Wenruo * NOTE: if we find that backrefs for a block are cached, we know backrefs for 464e7d571c7SQu Wenruo * all upper level blocks that directly/indirectly reference the block are also 465e7d571c7SQu Wenruo * cached. 466e7d571c7SQu Wenruo */ 467a26195a5SQu Wenruo static noinline_for_stack struct btrfs_backref_node *build_backref_tree( 468e7d571c7SQu Wenruo struct reloc_control *rc, struct btrfs_key *node_key, 469e7d571c7SQu Wenruo int level, u64 bytenr) 470e7d571c7SQu Wenruo { 471e7d571c7SQu Wenruo struct btrfs_backref_iter *iter; 472a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 473e7d571c7SQu Wenruo /* For searching parent of TREE_BLOCK_REF */ 474e7d571c7SQu Wenruo struct btrfs_path *path; 475a26195a5SQu Wenruo struct btrfs_backref_node *cur; 476a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 477a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 478e7d571c7SQu Wenruo int ret; 479e7d571c7SQu Wenruo int err = 0; 480e7d571c7SQu Wenruo 481d68194b2SDavid Sterba iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info); 482e7d571c7SQu Wenruo if (!iter) 483e7d571c7SQu Wenruo return ERR_PTR(-ENOMEM); 484e7d571c7SQu Wenruo path = btrfs_alloc_path(); 485e7d571c7SQu Wenruo if (!path) { 486e7d571c7SQu Wenruo err = -ENOMEM; 487e7d571c7SQu Wenruo goto out; 488e7d571c7SQu Wenruo } 489e7d571c7SQu Wenruo 490b1818dabSQu Wenruo node = btrfs_backref_alloc_node(cache, bytenr, level); 491e7d571c7SQu Wenruo if (!node) { 492e7d571c7SQu Wenruo err = -ENOMEM; 493e7d571c7SQu Wenruo goto out; 494e7d571c7SQu Wenruo } 495e7d571c7SQu Wenruo 496e7d571c7SQu Wenruo node->lowest = 1; 497e7d571c7SQu Wenruo cur = node; 498e7d571c7SQu Wenruo 499e7d571c7SQu Wenruo /* Breadth-first search to build backref cache */ 500e7d571c7SQu Wenruo do { 5011b60d2ecSQu Wenruo ret = btrfs_backref_add_tree_node(cache, path, iter, node_key, 5021b60d2ecSQu Wenruo cur); 503e7d571c7SQu Wenruo if (ret < 0) { 504e7d571c7SQu Wenruo err = ret; 505e7d571c7SQu Wenruo goto out; 506e7d571c7SQu Wenruo } 507e7d571c7SQu Wenruo edge = list_first_entry_or_null(&cache->pending_edge, 508a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 509e7d571c7SQu Wenruo /* 510e7d571c7SQu Wenruo * The pending list isn't empty, take the first block to 511e7d571c7SQu Wenruo * process 512e7d571c7SQu Wenruo */ 513e7d571c7SQu Wenruo if (edge) { 5145d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 5155d4f98a2SYan Zheng cur = edge->node[UPPER]; 5165d4f98a2SYan Zheng } 517e7d571c7SQu Wenruo } while (edge); 5185d4f98a2SYan Zheng 5191f872924SQu Wenruo /* Finish the upper linkage of newly added edges/nodes */ 520fc997ed0SQu Wenruo ret = btrfs_backref_finish_upper_links(cache, node); 5211f872924SQu Wenruo if (ret < 0) { 5221f872924SQu Wenruo err = ret; 52375bfb9afSJosef Bacik goto out; 52475bfb9afSJosef Bacik } 52575bfb9afSJosef Bacik 52629db137bSQu Wenruo if (handle_useless_nodes(rc, node)) 5273fd0a558SYan, Zheng node = NULL; 5285d4f98a2SYan Zheng out: 52971f572a9SQu Wenruo btrfs_backref_iter_free(iter); 53071f572a9SQu Wenruo btrfs_free_path(path); 5315d4f98a2SYan Zheng if (err) { 5321b23ea18SQu Wenruo btrfs_backref_error_cleanup(cache, node); 5335d4f98a2SYan Zheng return ERR_PTR(err); 5345d4f98a2SYan Zheng } 53575bfb9afSJosef Bacik ASSERT(!node || !node->detached); 53684780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 53784780289SQu Wenruo list_empty(&cache->pending_edge)); 5385d4f98a2SYan Zheng return node; 5395d4f98a2SYan Zheng } 5405d4f98a2SYan Zheng 5415d4f98a2SYan Zheng /* 5423fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 5433fd0a558SYan, Zheng * the backref node is created by cloning backref node that 5443fd0a558SYan, Zheng * corresponds to root of source tree 5453fd0a558SYan, Zheng */ 5463fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 5473fd0a558SYan, Zheng struct reloc_control *rc, 5483fd0a558SYan, Zheng struct btrfs_root *src, 5493fd0a558SYan, Zheng struct btrfs_root *dest) 5503fd0a558SYan, Zheng { 5513fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 552a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 553a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 554a26195a5SQu Wenruo struct btrfs_backref_node *new_node; 555a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 556a26195a5SQu Wenruo struct btrfs_backref_edge *new_edge; 5573fd0a558SYan, Zheng struct rb_node *rb_node; 5583fd0a558SYan, Zheng 5593fd0a558SYan, Zheng if (cache->last_trans > 0) 5603fd0a558SYan, Zheng update_backref_cache(trans, cache); 5613fd0a558SYan, Zheng 562e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start); 5633fd0a558SYan, Zheng if (rb_node) { 564a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, rb_node); 5653fd0a558SYan, Zheng if (node->detached) 5663fd0a558SYan, Zheng node = NULL; 5673fd0a558SYan, Zheng else 5683fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 5693fd0a558SYan, Zheng } 5703fd0a558SYan, Zheng 5713fd0a558SYan, Zheng if (!node) { 572e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, 5733fd0a558SYan, Zheng reloc_root->commit_root->start); 5743fd0a558SYan, Zheng if (rb_node) { 575a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, 5763fd0a558SYan, Zheng rb_node); 5773fd0a558SYan, Zheng BUG_ON(node->detached); 5783fd0a558SYan, Zheng } 5793fd0a558SYan, Zheng } 5803fd0a558SYan, Zheng 5813fd0a558SYan, Zheng if (!node) 5823fd0a558SYan, Zheng return 0; 5833fd0a558SYan, Zheng 584b1818dabSQu Wenruo new_node = btrfs_backref_alloc_node(cache, dest->node->start, 585b1818dabSQu Wenruo node->level); 5863fd0a558SYan, Zheng if (!new_node) 5873fd0a558SYan, Zheng return -ENOMEM; 5883fd0a558SYan, Zheng 5893fd0a558SYan, Zheng new_node->lowest = node->lowest; 5906848ad64SYan, Zheng new_node->checked = 1; 59100246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 5920b530bc5SJosef Bacik ASSERT(new_node->root); 5933fd0a558SYan, Zheng 5943fd0a558SYan, Zheng if (!node->lowest) { 5953fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 59647254d07SQu Wenruo new_edge = btrfs_backref_alloc_edge(cache); 5973fd0a558SYan, Zheng if (!new_edge) 5983fd0a558SYan, Zheng goto fail; 5993fd0a558SYan, Zheng 600f39911e5SQu Wenruo btrfs_backref_link_edge(new_edge, edge->node[LOWER], 601f39911e5SQu Wenruo new_node, LINK_UPPER); 6023fd0a558SYan, Zheng } 60376b9e23dSMiao Xie } else { 60476b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 6053fd0a558SYan, Zheng } 6063fd0a558SYan, Zheng 607e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr, 6083fd0a558SYan, Zheng &new_node->rb_node); 60943c04fb1SJeff Mahoney if (rb_node) 610982c92cbSQu Wenruo btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST); 6113fd0a558SYan, Zheng 6123fd0a558SYan, Zheng if (!new_node->lowest) { 6133fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 6143fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 6153fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 6163fd0a558SYan, Zheng } 6173fd0a558SYan, Zheng } 6183fd0a558SYan, Zheng return 0; 6193fd0a558SYan, Zheng fail: 6203fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 6213fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 622a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 6233fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 624741188d3SQu Wenruo btrfs_backref_free_edge(cache, new_edge); 6253fd0a558SYan, Zheng } 626741188d3SQu Wenruo btrfs_backref_free_node(cache, new_node); 6273fd0a558SYan, Zheng return -ENOMEM; 6283fd0a558SYan, Zheng } 6293fd0a558SYan, Zheng 6303fd0a558SYan, Zheng /* 6315d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 6325d4f98a2SYan Zheng */ 633ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 6345d4f98a2SYan Zheng { 6350b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6365d4f98a2SYan Zheng struct rb_node *rb_node; 6375d4f98a2SYan Zheng struct mapping_node *node; 6380b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 6395d4f98a2SYan Zheng 6405d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 641ffd7b339SJeff Mahoney if (!node) 642ffd7b339SJeff Mahoney return -ENOMEM; 6435d4f98a2SYan Zheng 644ea287ab1SJosef Bacik node->bytenr = root->commit_root->start; 6455d4f98a2SYan Zheng node->data = root; 6465d4f98a2SYan Zheng 6475d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 648e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 6495d4f98a2SYan Zheng node->bytenr, &node->rb_node); 6505d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 651ffd7b339SJeff Mahoney if (rb_node) { 65257a304cfSJosef Bacik btrfs_err(fs_info, 6535d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 6545d163e0eSJeff Mahoney node->bytenr); 65557a304cfSJosef Bacik return -EEXIST; 656ffd7b339SJeff Mahoney } 6575d4f98a2SYan Zheng 6585d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 6595d4f98a2SYan Zheng return 0; 6605d4f98a2SYan Zheng } 6615d4f98a2SYan Zheng 6625d4f98a2SYan Zheng /* 663c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 6645d4f98a2SYan Zheng * mapping 6655d4f98a2SYan Zheng */ 666c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 6675d4f98a2SYan Zheng { 6680b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6695d4f98a2SYan Zheng struct rb_node *rb_node; 6705d4f98a2SYan Zheng struct mapping_node *node = NULL; 6710b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 672f44deb74SJosef Bacik bool put_ref = false; 6735d4f98a2SYan Zheng 67465c6e82bSQu Wenruo if (rc && root->node) { 6755d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 676e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 677ea287ab1SJosef Bacik root->commit_root->start); 678c974c464SWang Shilong if (rb_node) { 679c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 680c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 681ea287ab1SJosef Bacik RB_CLEAR_NODE(&node->rb_node); 682c974c464SWang Shilong } 683c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 684c78a10aeSJosef Bacik ASSERT(!node || (struct btrfs_root *)node->data == root); 685389305b2SQu Wenruo } 686c974c464SWang Shilong 687f44deb74SJosef Bacik /* 688f44deb74SJosef Bacik * We only put the reloc root here if it's on the list. There's a lot 689f44deb74SJosef Bacik * of places where the pattern is to splice the rc->reloc_roots, process 690f44deb74SJosef Bacik * the reloc roots, and then add the reloc root back onto 691f44deb74SJosef Bacik * rc->reloc_roots. If we call __del_reloc_root while it's off of the 692f44deb74SJosef Bacik * list we don't want the reference being dropped, because the guy 693f44deb74SJosef Bacik * messing with the list is in charge of the reference. 694f44deb74SJosef Bacik */ 6950b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 696f44deb74SJosef Bacik if (!list_empty(&root->root_list)) { 697f44deb74SJosef Bacik put_ref = true; 698c974c464SWang Shilong list_del_init(&root->root_list); 699f44deb74SJosef Bacik } 7000b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 701f44deb74SJosef Bacik if (put_ref) 702f44deb74SJosef Bacik btrfs_put_root(root); 703c974c464SWang Shilong kfree(node); 704c974c464SWang Shilong } 705c974c464SWang Shilong 706c974c464SWang Shilong /* 707c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 708c974c464SWang Shilong * mapping 709c974c464SWang Shilong */ 710ea287ab1SJosef Bacik static int __update_reloc_root(struct btrfs_root *root) 711c974c464SWang Shilong { 7120b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 713c974c464SWang Shilong struct rb_node *rb_node; 714c974c464SWang Shilong struct mapping_node *node = NULL; 7150b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 716c974c464SWang Shilong 717c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 718e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 719ea287ab1SJosef Bacik root->commit_root->start); 7205d4f98a2SYan Zheng if (rb_node) { 7215d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 7225d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 7235d4f98a2SYan Zheng } 7245d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 7255d4f98a2SYan Zheng 7268f71f3e0SLiu Bo if (!node) 7278f71f3e0SLiu Bo return 0; 7285d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 7295d4f98a2SYan Zheng 7305d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 731ea287ab1SJosef Bacik node->bytenr = root->node->start; 732e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 7335d4f98a2SYan Zheng node->bytenr, &node->rb_node); 7345d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 73543c04fb1SJeff Mahoney if (rb_node) 736982c92cbSQu Wenruo btrfs_backref_panic(fs_info, node->bytenr, -EEXIST); 7375d4f98a2SYan Zheng return 0; 7385d4f98a2SYan Zheng } 7395d4f98a2SYan Zheng 7403fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 7413fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 7425d4f98a2SYan Zheng { 7430b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 7445d4f98a2SYan Zheng struct btrfs_root *reloc_root; 7455d4f98a2SYan Zheng struct extent_buffer *eb; 7465d4f98a2SYan Zheng struct btrfs_root_item *root_item; 7475d4f98a2SYan Zheng struct btrfs_key root_key; 74884c50ba5SJosef Bacik int ret = 0; 74984c50ba5SJosef Bacik bool must_abort = false; 7505d4f98a2SYan Zheng 7515d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 75284c50ba5SJosef Bacik if (!root_item) 75384c50ba5SJosef Bacik return ERR_PTR(-ENOMEM); 7545d4f98a2SYan Zheng 7555d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 7565d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 7573fd0a558SYan, Zheng root_key.offset = objectid; 7585d4f98a2SYan Zheng 7593fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 760054570a1SFilipe Manana u64 commit_root_gen; 761054570a1SFilipe Manana 7623fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 7635d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 7645d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 76584c50ba5SJosef Bacik if (ret) 76684c50ba5SJosef Bacik goto fail; 76784c50ba5SJosef Bacik 768054570a1SFilipe Manana /* 769054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 770054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 771054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 772054570a1SFilipe Manana * either inside the critical section of a transaction commit 773054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 774054570a1SFilipe Manana * it's created before the transaction commit is started. 775054570a1SFilipe Manana */ 776054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 777054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 7783fd0a558SYan, Zheng } else { 7793fd0a558SYan, Zheng /* 7803fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 7813fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 7823fd0a558SYan, Zheng * modified after it was created have RELOC flag 7833fd0a558SYan, Zheng * set in their headers. so it's OK to not update 7843fd0a558SYan, Zheng * the 'last_snapshot'. 7853fd0a558SYan, Zheng */ 7863fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 7873fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 78884c50ba5SJosef Bacik if (ret) 78984c50ba5SJosef Bacik goto fail; 7903fd0a558SYan, Zheng } 7913fd0a558SYan, Zheng 79284c50ba5SJosef Bacik /* 79384c50ba5SJosef Bacik * We have changed references at this point, we must abort the 79484c50ba5SJosef Bacik * transaction if anything fails. 79584c50ba5SJosef Bacik */ 79684c50ba5SJosef Bacik must_abort = true; 79784c50ba5SJosef Bacik 7985d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 7995d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 8005d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 8015d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 8023fd0a558SYan, Zheng 8033fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 8043fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 8053fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 8063fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 807c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, 0); 8083fd0a558SYan, Zheng } 8095d4f98a2SYan Zheng 8105d4f98a2SYan Zheng btrfs_tree_unlock(eb); 8115d4f98a2SYan Zheng free_extent_buffer(eb); 8125d4f98a2SYan Zheng 8130b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 8145d4f98a2SYan Zheng &root_key, root_item); 81584c50ba5SJosef Bacik if (ret) 81684c50ba5SJosef Bacik goto fail; 81784c50ba5SJosef Bacik 8185d4f98a2SYan Zheng kfree(root_item); 8195d4f98a2SYan Zheng 8203dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 82184c50ba5SJosef Bacik if (IS_ERR(reloc_root)) { 82284c50ba5SJosef Bacik ret = PTR_ERR(reloc_root); 82384c50ba5SJosef Bacik goto abort; 82484c50ba5SJosef Bacik } 82592a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 8265d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 8273fd0a558SYan, Zheng return reloc_root; 82884c50ba5SJosef Bacik fail: 82984c50ba5SJosef Bacik kfree(root_item); 83084c50ba5SJosef Bacik abort: 83184c50ba5SJosef Bacik if (must_abort) 83284c50ba5SJosef Bacik btrfs_abort_transaction(trans, ret); 83384c50ba5SJosef Bacik return ERR_PTR(ret); 8343fd0a558SYan, Zheng } 8353fd0a558SYan, Zheng 8363fd0a558SYan, Zheng /* 8373fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 8383fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 839f44deb74SJosef Bacik * 840f44deb74SJosef Bacik * The reloc_root comes out of here with two references, one for 841f44deb74SJosef Bacik * root->reloc_root, and another for being on the rc->reloc_roots list. 8423fd0a558SYan, Zheng */ 8433fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 8443fd0a558SYan, Zheng struct btrfs_root *root) 8453fd0a558SYan, Zheng { 8460b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8473fd0a558SYan, Zheng struct btrfs_root *reloc_root; 8480b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 84920dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 8503fd0a558SYan, Zheng int clear_rsv = 0; 851ffd7b339SJeff Mahoney int ret; 8523fd0a558SYan, Zheng 853aec7db3bSJosef Bacik if (!rc) 8542abc726aSJosef Bacik return 0; 8552abc726aSJosef Bacik 8561fac4a54SQu Wenruo /* 8571fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 8581fac4a54SQu Wenruo * create/update the dead reloc tree 8591fac4a54SQu Wenruo */ 8606282675eSQu Wenruo if (reloc_root_is_dead(root)) 8611fac4a54SQu Wenruo return 0; 8621fac4a54SQu Wenruo 863aec7db3bSJosef Bacik /* 864aec7db3bSJosef Bacik * This is subtle but important. We do not do 865aec7db3bSJosef Bacik * record_root_in_transaction for reloc roots, instead we record their 866aec7db3bSJosef Bacik * corresponding fs root, and then here we update the last trans for the 867aec7db3bSJosef Bacik * reloc root. This means that we have to do this for the entire life 868aec7db3bSJosef Bacik * of the reloc root, regardless of which stage of the relocation we are 869aec7db3bSJosef Bacik * in. 870aec7db3bSJosef Bacik */ 8713fd0a558SYan, Zheng if (root->reloc_root) { 8723fd0a558SYan, Zheng reloc_root = root->reloc_root; 8733fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 8743fd0a558SYan, Zheng return 0; 8753fd0a558SYan, Zheng } 8763fd0a558SYan, Zheng 877aec7db3bSJosef Bacik /* 878aec7db3bSJosef Bacik * We are merging reloc roots, we do not need new reloc trees. Also 879aec7db3bSJosef Bacik * reloc trees never need their own reloc tree. 880aec7db3bSJosef Bacik */ 881aec7db3bSJosef Bacik if (!rc->create_reloc_tree || 882aec7db3bSJosef Bacik root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 883aec7db3bSJosef Bacik return 0; 884aec7db3bSJosef Bacik 88520dd2cbfSMiao Xie if (!trans->reloc_reserved) { 88620dd2cbfSMiao Xie rsv = trans->block_rsv; 8873fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 8883fd0a558SYan, Zheng clear_rsv = 1; 8893fd0a558SYan, Zheng } 8903fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 8913fd0a558SYan, Zheng if (clear_rsv) 89220dd2cbfSMiao Xie trans->block_rsv = rsv; 89300bb36a0SJosef Bacik if (IS_ERR(reloc_root)) 89400bb36a0SJosef Bacik return PTR_ERR(reloc_root); 8955d4f98a2SYan Zheng 896ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 89757a304cfSJosef Bacik ASSERT(ret != -EEXIST); 89800bb36a0SJosef Bacik if (ret) { 89900bb36a0SJosef Bacik /* Pairs with create_reloc_root */ 90000bb36a0SJosef Bacik btrfs_put_root(reloc_root); 90100bb36a0SJosef Bacik return ret; 90200bb36a0SJosef Bacik } 903f44deb74SJosef Bacik root->reloc_root = btrfs_grab_root(reloc_root); 9045d4f98a2SYan Zheng return 0; 9055d4f98a2SYan Zheng } 9065d4f98a2SYan Zheng 9075d4f98a2SYan Zheng /* 9085d4f98a2SYan Zheng * update root item of reloc tree 9095d4f98a2SYan Zheng */ 9105d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 9115d4f98a2SYan Zheng struct btrfs_root *root) 9125d4f98a2SYan Zheng { 9130b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 9145d4f98a2SYan Zheng struct btrfs_root *reloc_root; 9155d4f98a2SYan Zheng struct btrfs_root_item *root_item; 9165d4f98a2SYan Zheng int ret; 9175d4f98a2SYan Zheng 9186282675eSQu Wenruo if (!have_reloc_root(root)) 919592fbcd5SJosef Bacik return 0; 9205d4f98a2SYan Zheng 9215d4f98a2SYan Zheng reloc_root = root->reloc_root; 9225d4f98a2SYan Zheng root_item = &reloc_root->root_item; 9235d4f98a2SYan Zheng 924f44deb74SJosef Bacik /* 925f44deb74SJosef Bacik * We are probably ok here, but __del_reloc_root() will drop its ref of 926f44deb74SJosef Bacik * the root. We have the ref for root->reloc_root, but just in case 927f44deb74SJosef Bacik * hold it while we update the reloc root. 928f44deb74SJosef Bacik */ 929f44deb74SJosef Bacik btrfs_grab_root(reloc_root); 930f44deb74SJosef Bacik 931d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 9320b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 9333fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 934d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 9356282675eSQu Wenruo /* 9366282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 9376282675eSQu Wenruo * have_reloc_root will not touch it from now on. 9386282675eSQu Wenruo */ 9396282675eSQu Wenruo smp_wmb(); 940c974c464SWang Shilong __del_reloc_root(reloc_root); 9415d4f98a2SYan Zheng } 9425d4f98a2SYan Zheng 9435d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 944ea287ab1SJosef Bacik __update_reloc_root(reloc_root); 9455d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 9465d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 9475d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 9485d4f98a2SYan Zheng } 9495d4f98a2SYan Zheng 9500b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 9515d4f98a2SYan Zheng &reloc_root->root_key, root_item); 952f44deb74SJosef Bacik btrfs_put_root(reloc_root); 953592fbcd5SJosef Bacik return ret; 9545d4f98a2SYan Zheng } 9555d4f98a2SYan Zheng 9565d4f98a2SYan Zheng /* 9575d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 9585d4f98a2SYan Zheng * in a subvolume 9595d4f98a2SYan Zheng */ 9605d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 9615d4f98a2SYan Zheng { 9625d4f98a2SYan Zheng struct rb_node *node; 9635d4f98a2SYan Zheng struct rb_node *prev; 9645d4f98a2SYan Zheng struct btrfs_inode *entry; 9655d4f98a2SYan Zheng struct inode *inode; 9665d4f98a2SYan Zheng 9675d4f98a2SYan Zheng spin_lock(&root->inode_lock); 9685d4f98a2SYan Zheng again: 9695d4f98a2SYan Zheng node = root->inode_tree.rb_node; 9705d4f98a2SYan Zheng prev = NULL; 9715d4f98a2SYan Zheng while (node) { 9725d4f98a2SYan Zheng prev = node; 9735d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9745d4f98a2SYan Zheng 9754a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 9765d4f98a2SYan Zheng node = node->rb_left; 9774a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 9785d4f98a2SYan Zheng node = node->rb_right; 9795d4f98a2SYan Zheng else 9805d4f98a2SYan Zheng break; 9815d4f98a2SYan Zheng } 9825d4f98a2SYan Zheng if (!node) { 9835d4f98a2SYan Zheng while (prev) { 9845d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 9854a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 9865d4f98a2SYan Zheng node = prev; 9875d4f98a2SYan Zheng break; 9885d4f98a2SYan Zheng } 9895d4f98a2SYan Zheng prev = rb_next(prev); 9905d4f98a2SYan Zheng } 9915d4f98a2SYan Zheng } 9925d4f98a2SYan Zheng while (node) { 9935d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9945d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 9955d4f98a2SYan Zheng if (inode) { 9965d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 9975d4f98a2SYan Zheng return inode; 9985d4f98a2SYan Zheng } 9995d4f98a2SYan Zheng 10004a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 10015d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 10025d4f98a2SYan Zheng goto again; 10035d4f98a2SYan Zheng 10045d4f98a2SYan Zheng node = rb_next(node); 10055d4f98a2SYan Zheng } 10065d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 10075d4f98a2SYan Zheng return NULL; 10085d4f98a2SYan Zheng } 10095d4f98a2SYan Zheng 10105d4f98a2SYan Zheng /* 10115d4f98a2SYan Zheng * get new location of data 10125d4f98a2SYan Zheng */ 10135d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 10145d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 10155d4f98a2SYan Zheng { 10165d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 10175d4f98a2SYan Zheng struct btrfs_path *path; 10185d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10195d4f98a2SYan Zheng struct extent_buffer *leaf; 10205d4f98a2SYan Zheng int ret; 10215d4f98a2SYan Zheng 10225d4f98a2SYan Zheng path = btrfs_alloc_path(); 10235d4f98a2SYan Zheng if (!path) 10245d4f98a2SYan Zheng return -ENOMEM; 10255d4f98a2SYan Zheng 10265d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1027f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1028f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 10295d4f98a2SYan Zheng if (ret < 0) 10305d4f98a2SYan Zheng goto out; 10315d4f98a2SYan Zheng if (ret > 0) { 10325d4f98a2SYan Zheng ret = -ENOENT; 10335d4f98a2SYan Zheng goto out; 10345d4f98a2SYan Zheng } 10355d4f98a2SYan Zheng 10365d4f98a2SYan Zheng leaf = path->nodes[0]; 10375d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 10385d4f98a2SYan Zheng struct btrfs_file_extent_item); 10395d4f98a2SYan Zheng 10405d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 10415d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 10425d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 10435d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 10445d4f98a2SYan Zheng 10455d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 104683d4cfd4SJosef Bacik ret = -EINVAL; 10475d4f98a2SYan Zheng goto out; 10485d4f98a2SYan Zheng } 10495d4f98a2SYan Zheng 10505d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 10515d4f98a2SYan Zheng ret = 0; 10525d4f98a2SYan Zheng out: 10535d4f98a2SYan Zheng btrfs_free_path(path); 10545d4f98a2SYan Zheng return ret; 10555d4f98a2SYan Zheng } 10565d4f98a2SYan Zheng 10575d4f98a2SYan Zheng /* 10585d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 10595d4f98a2SYan Zheng * the new locations. 10605d4f98a2SYan Zheng */ 10613fd0a558SYan, Zheng static noinline_for_stack 10623fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 10635d4f98a2SYan Zheng struct reloc_control *rc, 10645d4f98a2SYan Zheng struct btrfs_root *root, 10653fd0a558SYan, Zheng struct extent_buffer *leaf) 10665d4f98a2SYan Zheng { 10670b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 10685d4f98a2SYan Zheng struct btrfs_key key; 10695d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10705d4f98a2SYan Zheng struct inode *inode = NULL; 10715d4f98a2SYan Zheng u64 parent; 10725d4f98a2SYan Zheng u64 bytenr; 10733fd0a558SYan, Zheng u64 new_bytenr = 0; 10745d4f98a2SYan Zheng u64 num_bytes; 10755d4f98a2SYan Zheng u64 end; 10765d4f98a2SYan Zheng u32 nritems; 10775d4f98a2SYan Zheng u32 i; 107883d4cfd4SJosef Bacik int ret = 0; 10795d4f98a2SYan Zheng int first = 1; 10805d4f98a2SYan Zheng int dirty = 0; 10815d4f98a2SYan Zheng 10825d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 10835d4f98a2SYan Zheng return 0; 10845d4f98a2SYan Zheng 10855d4f98a2SYan Zheng /* reloc trees always use full backref */ 10865d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 10875d4f98a2SYan Zheng parent = leaf->start; 10885d4f98a2SYan Zheng else 10895d4f98a2SYan Zheng parent = 0; 10905d4f98a2SYan Zheng 10915d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 10925d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 109382fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 109482fa113fSQu Wenruo 10955d4f98a2SYan Zheng cond_resched(); 10965d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 10975d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 10985d4f98a2SYan Zheng continue; 10995d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 11005d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 11015d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 11025d4f98a2SYan Zheng continue; 11035d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 11045d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 11055d4f98a2SYan Zheng if (bytenr == 0) 11065d4f98a2SYan Zheng continue; 11079569cc20SQu Wenruo if (!in_range(bytenr, rc->block_group->start, 11089569cc20SQu Wenruo rc->block_group->length)) 11095d4f98a2SYan Zheng continue; 11105d4f98a2SYan Zheng 11115d4f98a2SYan Zheng /* 1112fb12489bSMatthew Wilcox (Oracle) * if we are modifying block in fs tree, wait for read_folio 11135d4f98a2SYan Zheng * to complete and drop the extent cache 11145d4f98a2SYan Zheng */ 11155d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 11165d4f98a2SYan Zheng if (first) { 11175d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11185d4f98a2SYan Zheng first = 0; 11194a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 11203fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 11215d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11225d4f98a2SYan Zheng } 11234a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 11249c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 11259c5c9604SJosef Bacik 11265d4f98a2SYan Zheng end = key.offset + 11275d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 11285d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 11290b246afaSJeff Mahoney fs_info->sectorsize)); 11300b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 11315d4f98a2SYan Zheng end--; 11325d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 11339c5c9604SJosef Bacik key.offset, end, 11349c5c9604SJosef Bacik &cached_state); 11355d4f98a2SYan Zheng if (!ret) 11365d4f98a2SYan Zheng continue; 11375d4f98a2SYan Zheng 11384c0c8cfcSFilipe Manana btrfs_drop_extent_map_range(BTRFS_I(inode), 11394c0c8cfcSFilipe Manana key.offset, end, true); 11405d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 11419c5c9604SJosef Bacik key.offset, end, &cached_state); 11425d4f98a2SYan Zheng } 11435d4f98a2SYan Zheng } 11445d4f98a2SYan Zheng 11455d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 11465d4f98a2SYan Zheng bytenr, num_bytes); 114783d4cfd4SJosef Bacik if (ret) { 114883d4cfd4SJosef Bacik /* 114983d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 115083d4cfd4SJosef Bacik * in the file extent yet. 115183d4cfd4SJosef Bacik */ 115283d4cfd4SJosef Bacik break; 11533fd0a558SYan, Zheng } 11545d4f98a2SYan Zheng 11555d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 11565d4f98a2SYan Zheng dirty = 1; 11575d4f98a2SYan Zheng 11585d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 115982fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 116082fa113fSQu Wenruo num_bytes, parent); 116182fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1162f42c5da6SNikolay Borisov key.objectid, key.offset, 1163f42c5da6SNikolay Borisov root->root_key.objectid, false); 116482fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 116583d4cfd4SJosef Bacik if (ret) { 116666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 116783d4cfd4SJosef Bacik break; 116883d4cfd4SJosef Bacik } 11695d4f98a2SYan Zheng 1170ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1171ffd4bb2aSQu Wenruo num_bytes, parent); 1172ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1173f42c5da6SNikolay Borisov key.objectid, key.offset, 1174f42c5da6SNikolay Borisov root->root_key.objectid, false); 1175ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 117683d4cfd4SJosef Bacik if (ret) { 117766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 117883d4cfd4SJosef Bacik break; 117983d4cfd4SJosef Bacik } 11805d4f98a2SYan Zheng } 11815d4f98a2SYan Zheng if (dirty) 11825d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 11833fd0a558SYan, Zheng if (inode) 11843fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 118583d4cfd4SJosef Bacik return ret; 11865d4f98a2SYan Zheng } 11875d4f98a2SYan Zheng 11885d4f98a2SYan Zheng static noinline_for_stack 11895d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 11905d4f98a2SYan Zheng struct btrfs_path *path, int level) 11915d4f98a2SYan Zheng { 11925d4f98a2SYan Zheng struct btrfs_disk_key key1; 11935d4f98a2SYan Zheng struct btrfs_disk_key key2; 11945d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 11955d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 11965d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 11975d4f98a2SYan Zheng } 11985d4f98a2SYan Zheng 11995d4f98a2SYan Zheng /* 12005d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 12015d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 12025d4f98a2SYan Zheng * reloc tree was create can be replaced. 12035d4f98a2SYan Zheng * 12045d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 12055d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 12065d4f98a2SYan Zheng * errors, a negative error number is returned. 12075d4f98a2SYan Zheng */ 12083fd0a558SYan, Zheng static noinline_for_stack 12093d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 12105d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 12115d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 12125d4f98a2SYan Zheng int lowest_level, int max_level) 12135d4f98a2SYan Zheng { 12140b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 12155d4f98a2SYan Zheng struct extent_buffer *eb; 12165d4f98a2SYan Zheng struct extent_buffer *parent; 121782fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 12185d4f98a2SYan Zheng struct btrfs_key key; 12195d4f98a2SYan Zheng u64 old_bytenr; 12205d4f98a2SYan Zheng u64 new_bytenr; 12215d4f98a2SYan Zheng u64 old_ptr_gen; 12225d4f98a2SYan Zheng u64 new_ptr_gen; 12235d4f98a2SYan Zheng u64 last_snapshot; 12245d4f98a2SYan Zheng u32 blocksize; 12253fd0a558SYan, Zheng int cow = 0; 12265d4f98a2SYan Zheng int level; 12275d4f98a2SYan Zheng int ret; 12285d4f98a2SYan Zheng int slot; 12295d4f98a2SYan Zheng 12307a9213a9SJosef Bacik ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 12317a9213a9SJosef Bacik ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 12325d4f98a2SYan Zheng 12335d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 12343fd0a558SYan, Zheng again: 12355d4f98a2SYan Zheng slot = path->slots[lowest_level]; 12365d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 12375d4f98a2SYan Zheng 12385d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 12395d4f98a2SYan Zheng level = btrfs_header_level(eb); 12405d4f98a2SYan Zheng 12415d4f98a2SYan Zheng if (level < lowest_level) { 12425d4f98a2SYan Zheng btrfs_tree_unlock(eb); 12435d4f98a2SYan Zheng free_extent_buffer(eb); 12445d4f98a2SYan Zheng return 0; 12455d4f98a2SYan Zheng } 12465d4f98a2SYan Zheng 12473fd0a558SYan, Zheng if (cow) { 12489631e4ccSJosef Bacik ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb, 12499631e4ccSJosef Bacik BTRFS_NESTING_COW); 125045b87c5dSJosef Bacik if (ret) { 125145b87c5dSJosef Bacik btrfs_tree_unlock(eb); 125245b87c5dSJosef Bacik free_extent_buffer(eb); 125345b87c5dSJosef Bacik return ret; 125445b87c5dSJosef Bacik } 12553fd0a558SYan, Zheng } 12565d4f98a2SYan Zheng 12575d4f98a2SYan Zheng if (next_key) { 12585d4f98a2SYan Zheng next_key->objectid = (u64)-1; 12595d4f98a2SYan Zheng next_key->type = (u8)-1; 12605d4f98a2SYan Zheng next_key->offset = (u64)-1; 12615d4f98a2SYan Zheng } 12625d4f98a2SYan Zheng 12635d4f98a2SYan Zheng parent = eb; 12645d4f98a2SYan Zheng while (1) { 12655d4f98a2SYan Zheng level = btrfs_header_level(parent); 12667a9213a9SJosef Bacik ASSERT(level >= lowest_level); 12675d4f98a2SYan Zheng 1268e3b83361SQu Wenruo ret = btrfs_bin_search(parent, &key, &slot); 1269cbca7d59SFilipe Manana if (ret < 0) 1270cbca7d59SFilipe Manana break; 12715d4f98a2SYan Zheng if (ret && slot > 0) 12725d4f98a2SYan Zheng slot--; 12735d4f98a2SYan Zheng 12745d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 12755d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 12765d4f98a2SYan Zheng 12775d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 12780b246afaSJeff Mahoney blocksize = fs_info->nodesize; 12795d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 12805d4f98a2SYan Zheng 12815d4f98a2SYan Zheng if (level <= max_level) { 12825d4f98a2SYan Zheng eb = path->nodes[level]; 12835d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 12845d4f98a2SYan Zheng path->slots[level]); 12855d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 12865d4f98a2SYan Zheng path->slots[level]); 12875d4f98a2SYan Zheng } else { 12885d4f98a2SYan Zheng new_bytenr = 0; 12895d4f98a2SYan Zheng new_ptr_gen = 0; 12905d4f98a2SYan Zheng } 12915d4f98a2SYan Zheng 1292fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 12935d4f98a2SYan Zheng ret = level; 12945d4f98a2SYan Zheng break; 12955d4f98a2SYan Zheng } 12965d4f98a2SYan Zheng 12975d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 12985d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 12993fd0a558SYan, Zheng if (level <= lowest_level) { 13005d4f98a2SYan Zheng ret = 0; 13015d4f98a2SYan Zheng break; 13025d4f98a2SYan Zheng } 13035d4f98a2SYan Zheng 13046b3426beSJosef Bacik eb = btrfs_read_node_slot(parent, slot); 130564c043deSLiu Bo if (IS_ERR(eb)) { 130664c043deSLiu Bo ret = PTR_ERR(eb); 1307264813acSLiu Bo break; 1308416bc658SJosef Bacik } 13095d4f98a2SYan Zheng btrfs_tree_lock(eb); 13103fd0a558SYan, Zheng if (cow) { 13115d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 13129631e4ccSJosef Bacik slot, &eb, 13139631e4ccSJosef Bacik BTRFS_NESTING_COW); 131445b87c5dSJosef Bacik if (ret) { 131545b87c5dSJosef Bacik btrfs_tree_unlock(eb); 131645b87c5dSJosef Bacik free_extent_buffer(eb); 131745b87c5dSJosef Bacik break; 131845b87c5dSJosef Bacik } 13195d4f98a2SYan Zheng } 13205d4f98a2SYan Zheng 13215d4f98a2SYan Zheng btrfs_tree_unlock(parent); 13225d4f98a2SYan Zheng free_extent_buffer(parent); 13235d4f98a2SYan Zheng 13245d4f98a2SYan Zheng parent = eb; 13255d4f98a2SYan Zheng continue; 13265d4f98a2SYan Zheng } 13275d4f98a2SYan Zheng 13283fd0a558SYan, Zheng if (!cow) { 13293fd0a558SYan, Zheng btrfs_tree_unlock(parent); 13303fd0a558SYan, Zheng free_extent_buffer(parent); 13313fd0a558SYan, Zheng cow = 1; 13323fd0a558SYan, Zheng goto again; 13333fd0a558SYan, Zheng } 13343fd0a558SYan, Zheng 13355d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 13365d4f98a2SYan Zheng path->slots[level]); 1337b3b4aa74SDavid Sterba btrfs_release_path(path); 13385d4f98a2SYan Zheng 13395d4f98a2SYan Zheng path->lowest_level = level; 1340b40130b2SJosef Bacik set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); 13415d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 1342b40130b2SJosef Bacik clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); 13435d4f98a2SYan Zheng path->lowest_level = 0; 13440e9873e2SJosef Bacik if (ret) { 13450e9873e2SJosef Bacik if (ret > 0) 13460e9873e2SJosef Bacik ret = -ENOENT; 13470e9873e2SJosef Bacik break; 13480e9873e2SJosef Bacik } 13495d4f98a2SYan Zheng 13505d4f98a2SYan Zheng /* 1351824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1352824d8dffSQu Wenruo * 1353824d8dffSQu Wenruo * We must trace both trees. 1354824d8dffSQu Wenruo * 1) Tree reloc subtree 1355824d8dffSQu Wenruo * If not traced, we will leak data numbers 1356824d8dffSQu Wenruo * 2) Fs subtree 1357824d8dffSQu Wenruo * If not traced, we will double count old data 1358f616f5cdSQu Wenruo * 1359f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1360f616f5cdSQu Wenruo * the swapped tree blocks. 1361f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1362f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1363824d8dffSQu Wenruo */ 1364370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1365370a11b8SQu Wenruo rc->block_group, parent, slot, 1366370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1367370a11b8SQu Wenruo last_snapshot); 1368370a11b8SQu Wenruo if (ret < 0) 1369370a11b8SQu Wenruo break; 1370824d8dffSQu Wenruo /* 13715d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 13725d4f98a2SYan Zheng */ 13735d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 13745d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 13755d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 13765d4f98a2SYan Zheng 13775d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 13785d4f98a2SYan Zheng path->slots[level], old_bytenr); 13795d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 13805d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 13815d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 13825d4f98a2SYan Zheng 138382fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 138482fa113fSQu Wenruo blocksize, path->nodes[level]->start); 1385f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1386f42c5da6SNikolay Borisov 0, true); 138782fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1388253e258cSJosef Bacik if (ret) { 1389253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1390253e258cSJosef Bacik break; 1391253e258cSJosef Bacik } 139282fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 139382fa113fSQu Wenruo blocksize, 0); 1394f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0, 1395f42c5da6SNikolay Borisov true); 139682fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1397253e258cSJosef Bacik if (ret) { 1398253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1399253e258cSJosef Bacik break; 1400253e258cSJosef Bacik } 14015d4f98a2SYan Zheng 1402ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1403ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1404f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1405f42c5da6SNikolay Borisov 0, true); 1406ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1407253e258cSJosef Bacik if (ret) { 1408253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1409253e258cSJosef Bacik break; 1410253e258cSJosef Bacik } 14115d4f98a2SYan Zheng 1412ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1413ffd4bb2aSQu Wenruo blocksize, 0); 1414f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 1415f42c5da6SNikolay Borisov 0, true); 1416ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1417253e258cSJosef Bacik if (ret) { 1418253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1419253e258cSJosef Bacik break; 1420253e258cSJosef Bacik } 14215d4f98a2SYan Zheng 14225d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 14235d4f98a2SYan Zheng 14245d4f98a2SYan Zheng ret = level; 14255d4f98a2SYan Zheng break; 14265d4f98a2SYan Zheng } 14275d4f98a2SYan Zheng btrfs_tree_unlock(parent); 14285d4f98a2SYan Zheng free_extent_buffer(parent); 14295d4f98a2SYan Zheng return ret; 14305d4f98a2SYan Zheng } 14315d4f98a2SYan Zheng 14325d4f98a2SYan Zheng /* 14335d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 14345d4f98a2SYan Zheng */ 14355d4f98a2SYan Zheng static noinline_for_stack 14365d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14375d4f98a2SYan Zheng int *level) 14385d4f98a2SYan Zheng { 14395d4f98a2SYan Zheng struct extent_buffer *eb; 14405d4f98a2SYan Zheng int i; 14415d4f98a2SYan Zheng u64 last_snapshot; 14425d4f98a2SYan Zheng u32 nritems; 14435d4f98a2SYan Zheng 14445d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14455d4f98a2SYan Zheng 14465d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 14475d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14485d4f98a2SYan Zheng path->nodes[i] = NULL; 14495d4f98a2SYan Zheng } 14505d4f98a2SYan Zheng 14515d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 14525d4f98a2SYan Zheng eb = path->nodes[i]; 14535d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14545d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 14555d4f98a2SYan Zheng path->slots[i]++; 14565d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 14575d4f98a2SYan Zheng last_snapshot) 14585d4f98a2SYan Zheng continue; 14595d4f98a2SYan Zheng 14605d4f98a2SYan Zheng *level = i; 14615d4f98a2SYan Zheng return 0; 14625d4f98a2SYan Zheng } 14635d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14645d4f98a2SYan Zheng path->nodes[i] = NULL; 14655d4f98a2SYan Zheng } 14665d4f98a2SYan Zheng return 1; 14675d4f98a2SYan Zheng } 14685d4f98a2SYan Zheng 14695d4f98a2SYan Zheng /* 14705d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 14715d4f98a2SYan Zheng */ 14725d4f98a2SYan Zheng static noinline_for_stack 14735d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14745d4f98a2SYan Zheng int *level) 14755d4f98a2SYan Zheng { 14765d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 14775d4f98a2SYan Zheng int i; 14785d4f98a2SYan Zheng u64 ptr_gen = 0; 14795d4f98a2SYan Zheng u64 last_snapshot; 14805d4f98a2SYan Zheng u32 nritems; 14815d4f98a2SYan Zheng 14825d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14835d4f98a2SYan Zheng 14845d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 14855d4f98a2SYan Zheng eb = path->nodes[i]; 14865d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14875d4f98a2SYan Zheng while (path->slots[i] < nritems) { 14885d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 14895d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 14905d4f98a2SYan Zheng break; 14915d4f98a2SYan Zheng path->slots[i]++; 14925d4f98a2SYan Zheng } 14935d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 14945d4f98a2SYan Zheng if (i == *level) 14955d4f98a2SYan Zheng break; 14965d4f98a2SYan Zheng *level = i + 1; 14975d4f98a2SYan Zheng return 0; 14985d4f98a2SYan Zheng } 14995d4f98a2SYan Zheng if (i == 1) { 15005d4f98a2SYan Zheng *level = i; 15015d4f98a2SYan Zheng return 0; 15025d4f98a2SYan Zheng } 15035d4f98a2SYan Zheng 15048ef385bbSJosef Bacik eb = btrfs_read_node_slot(eb, path->slots[i]); 15058ef385bbSJosef Bacik if (IS_ERR(eb)) 150664c043deSLiu Bo return PTR_ERR(eb); 15075d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 15085d4f98a2SYan Zheng path->nodes[i - 1] = eb; 15095d4f98a2SYan Zheng path->slots[i - 1] = 0; 15105d4f98a2SYan Zheng } 15115d4f98a2SYan Zheng return 1; 15125d4f98a2SYan Zheng } 15135d4f98a2SYan Zheng 15145d4f98a2SYan Zheng /* 15155d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 15165d4f98a2SYan Zheng * [min_key, max_key) 15175d4f98a2SYan Zheng */ 15185d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 15195d4f98a2SYan Zheng struct btrfs_key *min_key, 15205d4f98a2SYan Zheng struct btrfs_key *max_key) 15215d4f98a2SYan Zheng { 15220b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15235d4f98a2SYan Zheng struct inode *inode = NULL; 15245d4f98a2SYan Zheng u64 objectid; 15255d4f98a2SYan Zheng u64 start, end; 152633345d01SLi Zefan u64 ino; 15275d4f98a2SYan Zheng 15285d4f98a2SYan Zheng objectid = min_key->objectid; 15295d4f98a2SYan Zheng while (1) { 15309c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 15319c5c9604SJosef Bacik 15325d4f98a2SYan Zheng cond_resched(); 15335d4f98a2SYan Zheng iput(inode); 15345d4f98a2SYan Zheng 15355d4f98a2SYan Zheng if (objectid > max_key->objectid) 15365d4f98a2SYan Zheng break; 15375d4f98a2SYan Zheng 15385d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 15395d4f98a2SYan Zheng if (!inode) 15405d4f98a2SYan Zheng break; 15414a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 15425d4f98a2SYan Zheng 154333345d01SLi Zefan if (ino > max_key->objectid) { 15445d4f98a2SYan Zheng iput(inode); 15455d4f98a2SYan Zheng break; 15465d4f98a2SYan Zheng } 15475d4f98a2SYan Zheng 154833345d01SLi Zefan objectid = ino + 1; 15495d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 15505d4f98a2SYan Zheng continue; 15515d4f98a2SYan Zheng 155233345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 15535d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 15545d4f98a2SYan Zheng continue; 15555d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 15565d4f98a2SYan Zheng start = 0; 15575d4f98a2SYan Zheng else { 15585d4f98a2SYan Zheng start = min_key->offset; 15590b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 15605d4f98a2SYan Zheng } 15615d4f98a2SYan Zheng } else { 15625d4f98a2SYan Zheng start = 0; 15635d4f98a2SYan Zheng } 15645d4f98a2SYan Zheng 156533345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 15665d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 15675d4f98a2SYan Zheng continue; 15685d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 15695d4f98a2SYan Zheng end = (u64)-1; 15705d4f98a2SYan Zheng } else { 15715d4f98a2SYan Zheng if (max_key->offset == 0) 15725d4f98a2SYan Zheng continue; 15735d4f98a2SYan Zheng end = max_key->offset; 15740b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 15755d4f98a2SYan Zheng end--; 15765d4f98a2SYan Zheng } 15775d4f98a2SYan Zheng } else { 15785d4f98a2SYan Zheng end = (u64)-1; 15795d4f98a2SYan Zheng } 15805d4f98a2SYan Zheng 1581fb12489bSMatthew Wilcox (Oracle) /* the lock_extent waits for read_folio to complete */ 15829c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 15834c0c8cfcSFilipe Manana btrfs_drop_extent_map_range(BTRFS_I(inode), start, end, true); 15849c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 15855d4f98a2SYan Zheng } 15865d4f98a2SYan Zheng return 0; 15875d4f98a2SYan Zheng } 15885d4f98a2SYan Zheng 15895d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 15905d4f98a2SYan Zheng struct btrfs_key *key) 15915d4f98a2SYan Zheng 15925d4f98a2SYan Zheng { 15935d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 15945d4f98a2SYan Zheng if (!path->nodes[level]) 15955d4f98a2SYan Zheng break; 15965d4f98a2SYan Zheng if (path->slots[level] + 1 < 15975d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 15985d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 15995d4f98a2SYan Zheng path->slots[level] + 1); 16005d4f98a2SYan Zheng return 0; 16015d4f98a2SYan Zheng } 16025d4f98a2SYan Zheng level++; 16035d4f98a2SYan Zheng } 16045d4f98a2SYan Zheng return 1; 16055d4f98a2SYan Zheng } 16065d4f98a2SYan Zheng 16075d4f98a2SYan Zheng /* 1608d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 1609d2311e69SQu Wenruo */ 1610ac54da6cSJosef Bacik static int insert_dirty_subvol(struct btrfs_trans_handle *trans, 1611d2311e69SQu Wenruo struct reloc_control *rc, 1612d2311e69SQu Wenruo struct btrfs_root *root) 1613d2311e69SQu Wenruo { 1614d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1615d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 16167934133fSJosef Bacik int ret; 1617d2311e69SQu Wenruo 1618d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 1619d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 1620d2311e69SQu Wenruo ASSERT(reloc_root); 1621d2311e69SQu Wenruo 1622d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 1623d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 1624d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 1625c8422684SDavid Sterba btrfs_set_root_drop_level(reloc_root_item, 0); 1626d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 16277934133fSJosef Bacik ret = btrfs_update_reloc_root(trans, root); 16287934133fSJosef Bacik if (ret) 16297934133fSJosef Bacik return ret; 1630d2311e69SQu Wenruo 1631d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 163200246528SJosef Bacik btrfs_grab_root(root); 1633d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 1634d2311e69SQu Wenruo } 1635ac54da6cSJosef Bacik 1636ac54da6cSJosef Bacik return 0; 1637d2311e69SQu Wenruo } 1638d2311e69SQu Wenruo 1639d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 1640d2311e69SQu Wenruo { 1641d2311e69SQu Wenruo struct btrfs_root *root; 1642d2311e69SQu Wenruo struct btrfs_root *next; 1643d2311e69SQu Wenruo int ret = 0; 164430d40577SQu Wenruo int ret2; 1645d2311e69SQu Wenruo 1646d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 1647d2311e69SQu Wenruo reloc_dirty_list) { 164830d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 164930d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 1650d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1651d2311e69SQu Wenruo 1652d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 1653d2311e69SQu Wenruo root->reloc_root = NULL; 16546282675eSQu Wenruo /* 16556282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 16566282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 16576282675eSQu Wenruo */ 16586282675eSQu Wenruo smp_wmb(); 16591fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 1660f28de8d8SJosef Bacik if (reloc_root) { 1661f44deb74SJosef Bacik /* 1662f44deb74SJosef Bacik * btrfs_drop_snapshot drops our ref we hold for 1663f44deb74SJosef Bacik * ->reloc_root. If it fails however we must 1664f44deb74SJosef Bacik * drop the ref ourselves. 1665f44deb74SJosef Bacik */ 1666f28de8d8SJosef Bacik ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); 1667f44deb74SJosef Bacik if (ret2 < 0) { 1668f44deb74SJosef Bacik btrfs_put_root(reloc_root); 1669f44deb74SJosef Bacik if (!ret) 1670f28de8d8SJosef Bacik ret = ret2; 1671f28de8d8SJosef Bacik } 1672f44deb74SJosef Bacik } 167300246528SJosef Bacik btrfs_put_root(root); 167430d40577SQu Wenruo } else { 167530d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 16760078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(root, 0, 1); 1677f44deb74SJosef Bacik if (ret2 < 0) { 1678f44deb74SJosef Bacik btrfs_put_root(root); 1679f44deb74SJosef Bacik if (!ret) 168030d40577SQu Wenruo ret = ret2; 168130d40577SQu Wenruo } 1682d2311e69SQu Wenruo } 1683f44deb74SJosef Bacik } 1684d2311e69SQu Wenruo return ret; 1685d2311e69SQu Wenruo } 1686d2311e69SQu Wenruo 1687d2311e69SQu Wenruo /* 16885d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 16895d4f98a2SYan Zheng * fs tree. 16905d4f98a2SYan Zheng */ 16915d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 16925d4f98a2SYan Zheng struct btrfs_root *root) 16935d4f98a2SYan Zheng { 16940b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 16955d4f98a2SYan Zheng struct btrfs_key key; 16965d4f98a2SYan Zheng struct btrfs_key next_key; 16979e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 16985d4f98a2SYan Zheng struct btrfs_root *reloc_root; 16995d4f98a2SYan Zheng struct btrfs_root_item *root_item; 17005d4f98a2SYan Zheng struct btrfs_path *path; 17013fd0a558SYan, Zheng struct extent_buffer *leaf; 1702fca3a45dSJosef Bacik int reserve_level; 17035d4f98a2SYan Zheng int level; 17045d4f98a2SYan Zheng int max_level; 17055d4f98a2SYan Zheng int replaced = 0; 1706c6a592f2SNikolay Borisov int ret = 0; 17073fd0a558SYan, Zheng u32 min_reserved; 17085d4f98a2SYan Zheng 17095d4f98a2SYan Zheng path = btrfs_alloc_path(); 17105d4f98a2SYan Zheng if (!path) 17115d4f98a2SYan Zheng return -ENOMEM; 1712e4058b54SDavid Sterba path->reada = READA_FORWARD; 17135d4f98a2SYan Zheng 17145d4f98a2SYan Zheng reloc_root = root->reloc_root; 17155d4f98a2SYan Zheng root_item = &reloc_root->root_item; 17165d4f98a2SYan Zheng 17175d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 17185d4f98a2SYan Zheng level = btrfs_root_level(root_item); 171967439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 17205d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 17215d4f98a2SYan Zheng path->slots[level] = 0; 17225d4f98a2SYan Zheng } else { 17235d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 17245d4f98a2SYan Zheng 1725c8422684SDavid Sterba level = btrfs_root_drop_level(root_item); 17265d4f98a2SYan Zheng BUG_ON(level == 0); 17275d4f98a2SYan Zheng path->lowest_level = level; 17285d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 172933c66f43SYan Zheng path->lowest_level = 0; 17305d4f98a2SYan Zheng if (ret < 0) { 17315d4f98a2SYan Zheng btrfs_free_path(path); 17325d4f98a2SYan Zheng return ret; 17335d4f98a2SYan Zheng } 17345d4f98a2SYan Zheng 17355d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 17365d4f98a2SYan Zheng path->slots[level]); 17375d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 17385d4f98a2SYan Zheng 17395d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 17405d4f98a2SYan Zheng } 17415d4f98a2SYan Zheng 174244d354abSQu Wenruo /* 174344d354abSQu Wenruo * In merge_reloc_root(), we modify the upper level pointer to swap the 174444d354abSQu Wenruo * tree blocks between reloc tree and subvolume tree. Thus for tree 174544d354abSQu Wenruo * block COW, we COW at most from level 1 to root level for each tree. 174644d354abSQu Wenruo * 174744d354abSQu Wenruo * Thus the needed metadata size is at most root_level * nodesize, 174844d354abSQu Wenruo * and * 2 since we have two trees to COW. 174944d354abSQu Wenruo */ 1750fca3a45dSJosef Bacik reserve_level = max_t(int, 1, btrfs_root_level(root_item)); 1751fca3a45dSJosef Bacik min_reserved = fs_info->nodesize * reserve_level * 2; 17525d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 17535d4f98a2SYan Zheng 17545d4f98a2SYan Zheng while (1) { 17559270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 17569270501cSJosef Bacik min_reserved, 175744d354abSQu Wenruo BTRFS_RESERVE_FLUSH_LIMIT); 1758c6a592f2SNikolay Borisov if (ret) 17599e6a0c52SJosef Bacik goto out; 17609e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 17619e6a0c52SJosef Bacik if (IS_ERR(trans)) { 1762c6a592f2SNikolay Borisov ret = PTR_ERR(trans); 17639e6a0c52SJosef Bacik trans = NULL; 17649e6a0c52SJosef Bacik goto out; 17659e6a0c52SJosef Bacik } 17662abc726aSJosef Bacik 17672abc726aSJosef Bacik /* 17682abc726aSJosef Bacik * At this point we no longer have a reloc_control, so we can't 17692abc726aSJosef Bacik * depend on btrfs_init_reloc_root to update our last_trans. 17702abc726aSJosef Bacik * 17712abc726aSJosef Bacik * But that's ok, we started the trans handle on our 17722abc726aSJosef Bacik * corresponding fs_root, which means it's been added to the 17732abc726aSJosef Bacik * dirty list. At commit time we'll still call 17742abc726aSJosef Bacik * btrfs_update_reloc_root() and update our root item 17752abc726aSJosef Bacik * appropriately. 17762abc726aSJosef Bacik */ 17772abc726aSJosef Bacik reloc_root->last_trans = trans->transid; 17789e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 17793fd0a558SYan, Zheng 17803fd0a558SYan, Zheng replaced = 0; 17815d4f98a2SYan Zheng max_level = level; 17825d4f98a2SYan Zheng 17835d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 1784c6a592f2SNikolay Borisov if (ret < 0) 17855d4f98a2SYan Zheng goto out; 17865d4f98a2SYan Zheng if (ret > 0) 17875d4f98a2SYan Zheng break; 17885d4f98a2SYan Zheng 17895d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 17905d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 17915d4f98a2SYan Zheng ret = 0; 17925d4f98a2SYan Zheng } else { 17933d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 17943fd0a558SYan, Zheng &next_key, level, max_level); 17955d4f98a2SYan Zheng } 1796c6a592f2SNikolay Borisov if (ret < 0) 17975d4f98a2SYan Zheng goto out; 17985d4f98a2SYan Zheng if (ret > 0) { 17995d4f98a2SYan Zheng level = ret; 18005d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18015d4f98a2SYan Zheng path->slots[level]); 18025d4f98a2SYan Zheng replaced = 1; 18035d4f98a2SYan Zheng } 18045d4f98a2SYan Zheng 18055d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 18065d4f98a2SYan Zheng if (ret > 0) 18075d4f98a2SYan Zheng break; 18085d4f98a2SYan Zheng 18095d4f98a2SYan Zheng BUG_ON(level == 0); 18105d4f98a2SYan Zheng /* 18115d4f98a2SYan Zheng * save the merging progress in the drop_progress. 18125d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 18135d4f98a2SYan Zheng */ 18145d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 18155d4f98a2SYan Zheng path->slots[level]); 1816c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, level); 18175d4f98a2SYan Zheng 18183a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18199e6a0c52SJosef Bacik trans = NULL; 18205d4f98a2SYan Zheng 18212ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18225d4f98a2SYan Zheng 18235d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18245d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18255d4f98a2SYan Zheng } 18265d4f98a2SYan Zheng 18275d4f98a2SYan Zheng /* 18285d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 18295d4f98a2SYan Zheng * relocated and the block is tree root. 18305d4f98a2SYan Zheng */ 18315d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 18329631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf, 18339631e4ccSJosef Bacik BTRFS_NESTING_COW); 18345d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 18355d4f98a2SYan Zheng free_extent_buffer(leaf); 18365d4f98a2SYan Zheng out: 18375d4f98a2SYan Zheng btrfs_free_path(path); 18385d4f98a2SYan Zheng 1839ac54da6cSJosef Bacik if (ret == 0) { 1840ac54da6cSJosef Bacik ret = insert_dirty_subvol(trans, rc, root); 1841ac54da6cSJosef Bacik if (ret) 1842ac54da6cSJosef Bacik btrfs_abort_transaction(trans, ret); 1843ac54da6cSJosef Bacik } 18445d4f98a2SYan Zheng 18459e6a0c52SJosef Bacik if (trans) 18463a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18475d4f98a2SYan Zheng 18482ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18495d4f98a2SYan Zheng 18505d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18515d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18525d4f98a2SYan Zheng 1853c6a592f2SNikolay Borisov return ret; 18545d4f98a2SYan Zheng } 18555d4f98a2SYan Zheng 18563fd0a558SYan, Zheng static noinline_for_stack 18573fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 18585d4f98a2SYan Zheng { 18593fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 18600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 18613fd0a558SYan, Zheng struct btrfs_root *reloc_root; 18625d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 18633fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 18643fd0a558SYan, Zheng u64 num_bytes = 0; 18653fd0a558SYan, Zheng int ret; 18663fd0a558SYan, Zheng 18670b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 18680b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 18693fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 18700b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 18717585717fSChris Mason 18723fd0a558SYan, Zheng again: 18733fd0a558SYan, Zheng if (!err) { 18743fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 18759270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes, 187608e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 18773fd0a558SYan, Zheng if (ret) 18783fd0a558SYan, Zheng err = ret; 18793fd0a558SYan, Zheng } 18803fd0a558SYan, Zheng 18817a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 18823612b495STsutomu Itoh if (IS_ERR(trans)) { 18833612b495STsutomu Itoh if (!err) 18842ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 188563f018beSNikolay Borisov num_bytes, NULL); 18863612b495STsutomu Itoh return PTR_ERR(trans); 18873612b495STsutomu Itoh } 18883fd0a558SYan, Zheng 18893fd0a558SYan, Zheng if (!err) { 18903fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 18913a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18922ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 189363f018beSNikolay Borisov num_bytes, NULL); 18943fd0a558SYan, Zheng goto again; 18953fd0a558SYan, Zheng } 18963fd0a558SYan, Zheng } 18973fd0a558SYan, Zheng 18983fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 18993fd0a558SYan, Zheng 19003fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 19013fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 19023fd0a558SYan, Zheng struct btrfs_root, root_list); 19033fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 19043fd0a558SYan, Zheng 1905a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1906a820feb5SDavid Sterba false); 1907e0b085b0SJosef Bacik if (IS_ERR(root)) { 1908e0b085b0SJosef Bacik /* 1909e0b085b0SJosef Bacik * Even if we have an error we need this reloc root 1910e0b085b0SJosef Bacik * back on our list so we can clean up properly. 1911e0b085b0SJosef Bacik */ 1912e0b085b0SJosef Bacik list_add(&reloc_root->root_list, &reloc_roots); 1913e0b085b0SJosef Bacik btrfs_abort_transaction(trans, (int)PTR_ERR(root)); 1914e0b085b0SJosef Bacik if (!err) 1915e0b085b0SJosef Bacik err = PTR_ERR(root); 1916e0b085b0SJosef Bacik break; 1917e0b085b0SJosef Bacik } 1918e0b085b0SJosef Bacik ASSERT(root->reloc_root == reloc_root); 19193fd0a558SYan, Zheng 19203fd0a558SYan, Zheng /* 19213fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 19223fd0a558SYan, Zheng * knows it should resumes merging 19233fd0a558SYan, Zheng */ 19243fd0a558SYan, Zheng if (!err) 19253fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 1926bbae13f8SJosef Bacik ret = btrfs_update_reloc_root(trans, root); 19273fd0a558SYan, Zheng 1928bbae13f8SJosef Bacik /* 1929bbae13f8SJosef Bacik * Even if we have an error we need this reloc root back on our 1930bbae13f8SJosef Bacik * list so we can clean up properly. 1931bbae13f8SJosef Bacik */ 19323fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 193300246528SJosef Bacik btrfs_put_root(root); 1934bbae13f8SJosef Bacik 1935bbae13f8SJosef Bacik if (ret) { 1936bbae13f8SJosef Bacik btrfs_abort_transaction(trans, ret); 1937bbae13f8SJosef Bacik if (!err) 1938bbae13f8SJosef Bacik err = ret; 1939bbae13f8SJosef Bacik break; 1940bbae13f8SJosef Bacik } 19413fd0a558SYan, Zheng } 19423fd0a558SYan, Zheng 19433fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 19443fd0a558SYan, Zheng 19453fd0a558SYan, Zheng if (!err) 1946fb686c68SJosef Bacik err = btrfs_commit_transaction(trans); 19473fd0a558SYan, Zheng else 19483a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19493fd0a558SYan, Zheng return err; 19503fd0a558SYan, Zheng } 19513fd0a558SYan, Zheng 19523fd0a558SYan, Zheng static noinline_for_stack 1953aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 1954aca1bba6SLiu Bo { 1955a7571232SNikolay Borisov struct btrfs_root *reloc_root, *tmp; 1956aca1bba6SLiu Bo 1957a7571232SNikolay Borisov list_for_each_entry_safe(reloc_root, tmp, list, root_list) 1958bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 1959aca1bba6SLiu Bo } 1960aca1bba6SLiu Bo 1961aca1bba6SLiu Bo static noinline_for_stack 196294404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 19633fd0a558SYan, Zheng { 19640b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 19655d4f98a2SYan Zheng struct btrfs_root *root; 19665d4f98a2SYan Zheng struct btrfs_root *reloc_root; 19673fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 19683fd0a558SYan, Zheng int found = 0; 1969aca1bba6SLiu Bo int ret = 0; 19703fd0a558SYan, Zheng again: 19713fd0a558SYan, Zheng root = rc->extent_root; 19727585717fSChris Mason 19737585717fSChris Mason /* 19747585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 19757585717fSChris Mason * we have to make sure nobody is in the middle of 19767585717fSChris Mason * adding their roots to the list while we are 19777585717fSChris Mason * doing this splice 19787585717fSChris Mason */ 19790b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 19803fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 19810b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 19825d4f98a2SYan Zheng 19833fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 19843fd0a558SYan, Zheng found = 1; 19853fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 19863fd0a558SYan, Zheng struct btrfs_root, root_list); 19875d4f98a2SYan Zheng 1988a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1989a820feb5SDavid Sterba false); 19905d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 199124213fa4SJosef Bacik if (IS_ERR(root)) { 199224213fa4SJosef Bacik /* 199324213fa4SJosef Bacik * For recovery we read the fs roots on mount, 199424213fa4SJosef Bacik * and if we didn't find the root then we marked 199524213fa4SJosef Bacik * the reloc root as a garbage root. For normal 199624213fa4SJosef Bacik * relocation obviously the root should exist in 199724213fa4SJosef Bacik * memory. However there's no reason we can't 199824213fa4SJosef Bacik * handle the error properly here just in case. 199924213fa4SJosef Bacik */ 200024213fa4SJosef Bacik ASSERT(0); 200124213fa4SJosef Bacik ret = PTR_ERR(root); 200224213fa4SJosef Bacik goto out; 200324213fa4SJosef Bacik } 200424213fa4SJosef Bacik if (root->reloc_root != reloc_root) { 200524213fa4SJosef Bacik /* 200624213fa4SJosef Bacik * This is actually impossible without something 200724213fa4SJosef Bacik * going really wrong (like weird race condition 200824213fa4SJosef Bacik * or cosmic rays). 200924213fa4SJosef Bacik */ 201024213fa4SJosef Bacik ASSERT(0); 201124213fa4SJosef Bacik ret = -EINVAL; 201224213fa4SJosef Bacik goto out; 201324213fa4SJosef Bacik } 20143fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 201500246528SJosef Bacik btrfs_put_root(root); 2016b37b39cdSJosef Bacik if (ret) { 201725e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 201825e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 201925e293c2SWang Shilong &reloc_roots); 2020aca1bba6SLiu Bo goto out; 2021b37b39cdSJosef Bacik } 20223fd0a558SYan, Zheng } else { 202351415b6cSQu Wenruo if (!IS_ERR(root)) { 202451415b6cSQu Wenruo if (root->reloc_root == reloc_root) { 202551415b6cSQu Wenruo root->reloc_root = NULL; 202651415b6cSQu Wenruo btrfs_put_root(reloc_root); 202751415b6cSQu Wenruo } 20281dae7e0eSQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, 20291dae7e0eSQu Wenruo &root->state); 203051415b6cSQu Wenruo btrfs_put_root(root); 203151415b6cSQu Wenruo } 203251415b6cSQu Wenruo 20333fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 203430d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 203530d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 203630d40577SQu Wenruo &rc->dirty_subvol_roots); 20373fd0a558SYan, Zheng } 20385d4f98a2SYan Zheng } 20395d4f98a2SYan Zheng 20403fd0a558SYan, Zheng if (found) { 20413fd0a558SYan, Zheng found = 0; 20423fd0a558SYan, Zheng goto again; 20435d4f98a2SYan Zheng } 2044aca1bba6SLiu Bo out: 2045aca1bba6SLiu Bo if (ret) { 20460b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2047aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2048467bb1d2SWang Shilong 2049467bb1d2SWang Shilong /* new reloc root may be added */ 20500b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2051467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 20520b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2053467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2054aca1bba6SLiu Bo } 2055aca1bba6SLiu Bo 20567b7b7431SJosef Bacik /* 20577b7b7431SJosef Bacik * We used to have 20587b7b7431SJosef Bacik * 20597b7b7431SJosef Bacik * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 20607b7b7431SJosef Bacik * 20617b7b7431SJosef Bacik * here, but it's wrong. If we fail to start the transaction in 20627b7b7431SJosef Bacik * prepare_to_merge() we will have only 0 ref reloc roots, none of which 20637b7b7431SJosef Bacik * have actually been removed from the reloc_root_tree rb tree. This is 20647b7b7431SJosef Bacik * fine because we're bailing here, and we hold a reference on the root 20657b7b7431SJosef Bacik * for the list that holds it, so these roots will be cleaned up when we 20667b7b7431SJosef Bacik * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root 20677b7b7431SJosef Bacik * will be cleaned up on unmount. 20687b7b7431SJosef Bacik * 20697b7b7431SJosef Bacik * The remaining nodes will be cleaned up by free_reloc_control. 20707b7b7431SJosef Bacik */ 20715d4f98a2SYan Zheng } 20725d4f98a2SYan Zheng 20735d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 20745d4f98a2SYan Zheng { 20755d4f98a2SYan Zheng struct tree_block *block; 20765d4f98a2SYan Zheng struct rb_node *rb_node; 20775d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 20785d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 20795d4f98a2SYan Zheng rb_erase(rb_node, blocks); 20805d4f98a2SYan Zheng kfree(block); 20815d4f98a2SYan Zheng } 20825d4f98a2SYan Zheng } 20835d4f98a2SYan Zheng 20845d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 20855d4f98a2SYan Zheng struct btrfs_root *reloc_root) 20865d4f98a2SYan Zheng { 20870b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 20885d4f98a2SYan Zheng struct btrfs_root *root; 2089442b1ac5SJosef Bacik int ret; 20905d4f98a2SYan Zheng 20915d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 20925d4f98a2SYan Zheng return 0; 20935d4f98a2SYan Zheng 2094a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); 2095404bccbcSJosef Bacik 2096404bccbcSJosef Bacik /* 2097404bccbcSJosef Bacik * This should succeed, since we can't have a reloc root without having 2098404bccbcSJosef Bacik * already looked up the actual root and created the reloc root for this 2099404bccbcSJosef Bacik * root. 2100404bccbcSJosef Bacik * 2101404bccbcSJosef Bacik * However if there's some sort of corruption where we have a ref to a 2102404bccbcSJosef Bacik * reloc root without a corresponding root this could return ENOENT. 2103404bccbcSJosef Bacik */ 2104404bccbcSJosef Bacik if (IS_ERR(root)) { 2105404bccbcSJosef Bacik ASSERT(0); 2106404bccbcSJosef Bacik return PTR_ERR(root); 2107404bccbcSJosef Bacik } 2108404bccbcSJosef Bacik if (root->reloc_root != reloc_root) { 2109404bccbcSJosef Bacik ASSERT(0); 2110404bccbcSJosef Bacik btrfs_err(fs_info, 2111404bccbcSJosef Bacik "root %llu has two reloc roots associated with it", 2112404bccbcSJosef Bacik reloc_root->root_key.offset); 2113404bccbcSJosef Bacik btrfs_put_root(root); 2114404bccbcSJosef Bacik return -EUCLEAN; 2115404bccbcSJosef Bacik } 2116442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 211700246528SJosef Bacik btrfs_put_root(root); 21185d4f98a2SYan Zheng 2119442b1ac5SJosef Bacik return ret; 21205d4f98a2SYan Zheng } 21215d4f98a2SYan Zheng 21223fd0a558SYan, Zheng static noinline_for_stack 21233fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 21243fd0a558SYan, Zheng struct reloc_control *rc, 2125a26195a5SQu Wenruo struct btrfs_backref_node *node, 2126a26195a5SQu Wenruo struct btrfs_backref_edge *edges[]) 21275d4f98a2SYan Zheng { 2128a26195a5SQu Wenruo struct btrfs_backref_node *next; 21295d4f98a2SYan Zheng struct btrfs_root *root; 21303fd0a558SYan, Zheng int index = 0; 213192de551bSJosef Bacik int ret; 21323fd0a558SYan, Zheng 21335d4f98a2SYan Zheng next = node; 21345d4f98a2SYan Zheng while (1) { 21355d4f98a2SYan Zheng cond_resched(); 21365d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 21375d4f98a2SYan Zheng root = next->root; 21388ee66afeSJosef Bacik 21398ee66afeSJosef Bacik /* 21408ee66afeSJosef Bacik * If there is no root, then our references for this block are 21418ee66afeSJosef Bacik * incomplete, as we should be able to walk all the way up to a 21428ee66afeSJosef Bacik * block that is owned by a root. 21438ee66afeSJosef Bacik * 21448ee66afeSJosef Bacik * This path is only for SHAREABLE roots, so if we come upon a 21458ee66afeSJosef Bacik * non-SHAREABLE root then we have backrefs that resolve 21468ee66afeSJosef Bacik * improperly. 21478ee66afeSJosef Bacik * 21488ee66afeSJosef Bacik * Both of these cases indicate file system corruption, or a bug 21498ee66afeSJosef Bacik * in the backref walking code. 21508ee66afeSJosef Bacik */ 21518ee66afeSJosef Bacik if (!root) { 21528ee66afeSJosef Bacik ASSERT(0); 21538ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21548ee66afeSJosef Bacik "bytenr %llu doesn't have a backref path ending in a root", 21558ee66afeSJosef Bacik node->bytenr); 21568ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21578ee66afeSJosef Bacik } 21588ee66afeSJosef Bacik if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 21598ee66afeSJosef Bacik ASSERT(0); 21608ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21618ee66afeSJosef Bacik "bytenr %llu has multiple refs with one ending in a non-shareable root", 21628ee66afeSJosef Bacik node->bytenr); 21638ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21648ee66afeSJosef Bacik } 21655d4f98a2SYan Zheng 21665d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 216792de551bSJosef Bacik ret = record_reloc_root_in_trans(trans, root); 216892de551bSJosef Bacik if (ret) 216992de551bSJosef Bacik return ERR_PTR(ret); 21705d4f98a2SYan Zheng break; 21715d4f98a2SYan Zheng } 21725d4f98a2SYan Zheng 217392de551bSJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 217492de551bSJosef Bacik if (ret) 217592de551bSJosef Bacik return ERR_PTR(ret); 21763fd0a558SYan, Zheng root = root->reloc_root; 21773fd0a558SYan, Zheng 217839200e59SJosef Bacik /* 217939200e59SJosef Bacik * We could have raced with another thread which failed, so 218039200e59SJosef Bacik * root->reloc_root may not be set, return ENOENT in this case. 218139200e59SJosef Bacik */ 218239200e59SJosef Bacik if (!root) 218339200e59SJosef Bacik return ERR_PTR(-ENOENT); 218439200e59SJosef Bacik 21853fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 21868ee66afeSJosef Bacik /* 21878ee66afeSJosef Bacik * We just created the reloc root, so we shouldn't have 21888ee66afeSJosef Bacik * ->new_bytenr set and this shouldn't be in the changed 21898ee66afeSJosef Bacik * list. If it is then we have multiple roots pointing 21908ee66afeSJosef Bacik * at the same bytenr which indicates corruption, or 21918ee66afeSJosef Bacik * we've made a mistake in the backref walking code. 21928ee66afeSJosef Bacik */ 21938ee66afeSJosef Bacik ASSERT(next->new_bytenr == 0); 21948ee66afeSJosef Bacik ASSERT(list_empty(&next->list)); 21958ee66afeSJosef Bacik if (next->new_bytenr || !list_empty(&next->list)) { 21968ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21978ee66afeSJosef Bacik "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu", 21988ee66afeSJosef Bacik node->bytenr, next->bytenr); 21998ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 22008ee66afeSJosef Bacik } 22018ee66afeSJosef Bacik 22023fd0a558SYan, Zheng next->new_bytenr = root->node->start; 220300246528SJosef Bacik btrfs_put_root(next->root); 220400246528SJosef Bacik next->root = btrfs_grab_root(root); 22050b530bc5SJosef Bacik ASSERT(next->root); 22063fd0a558SYan, Zheng list_add_tail(&next->list, 22073fd0a558SYan, Zheng &rc->backref_cache.changed); 22089569cc20SQu Wenruo mark_block_processed(rc, next); 22095d4f98a2SYan Zheng break; 22105d4f98a2SYan Zheng } 22115d4f98a2SYan Zheng 22123fd0a558SYan, Zheng WARN_ON(1); 22135d4f98a2SYan Zheng root = NULL; 22145d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 22155d4f98a2SYan Zheng if (!next || next->level <= node->level) 22165d4f98a2SYan Zheng break; 22175d4f98a2SYan Zheng } 2218cbdc2ebcSJosef Bacik if (!root) { 2219cbdc2ebcSJosef Bacik /* 2220cbdc2ebcSJosef Bacik * This can happen if there's fs corruption or if there's a bug 2221cbdc2ebcSJosef Bacik * in the backref lookup code. 2222cbdc2ebcSJosef Bacik */ 2223cbdc2ebcSJosef Bacik ASSERT(0); 2224cbdc2ebcSJosef Bacik return ERR_PTR(-ENOENT); 2225cbdc2ebcSJosef Bacik } 22265d4f98a2SYan Zheng 22273fd0a558SYan, Zheng next = node; 22283fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 22293fd0a558SYan, Zheng while (1) { 22303fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 22313fd0a558SYan, Zheng if (--index < 0) 22323fd0a558SYan, Zheng break; 22333fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 22343fd0a558SYan, Zheng } 22355d4f98a2SYan Zheng return root; 22365d4f98a2SYan Zheng } 22375d4f98a2SYan Zheng 22383fd0a558SYan, Zheng /* 223992a7cc42SQu Wenruo * Select a tree root for relocation. 224092a7cc42SQu Wenruo * 224192a7cc42SQu Wenruo * Return NULL if the block is not shareable. We should use do_relocation() in 224292a7cc42SQu Wenruo * this case. 224392a7cc42SQu Wenruo * 224492a7cc42SQu Wenruo * Return a tree root pointer if the block is shareable. 224592a7cc42SQu Wenruo * Return -ENOENT if the block is root of reloc tree. 22463fd0a558SYan, Zheng */ 22475d4f98a2SYan Zheng static noinline_for_stack 2248a26195a5SQu Wenruo struct btrfs_root *select_one_root(struct btrfs_backref_node *node) 22495d4f98a2SYan Zheng { 2250a26195a5SQu Wenruo struct btrfs_backref_node *next; 22513fd0a558SYan, Zheng struct btrfs_root *root; 22523fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 2253a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22543fd0a558SYan, Zheng int index = 0; 22553fd0a558SYan, Zheng 22563fd0a558SYan, Zheng next = node; 22573fd0a558SYan, Zheng while (1) { 22583fd0a558SYan, Zheng cond_resched(); 22593fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 22603fd0a558SYan, Zheng root = next->root; 22618717cf44SJosef Bacik 22628717cf44SJosef Bacik /* 22638717cf44SJosef Bacik * This can occur if we have incomplete extent refs leading all 22648717cf44SJosef Bacik * the way up a particular path, in this case return -EUCLEAN. 22658717cf44SJosef Bacik */ 22668717cf44SJosef Bacik if (!root) 22678717cf44SJosef Bacik return ERR_PTR(-EUCLEAN); 22683fd0a558SYan, Zheng 226992a7cc42SQu Wenruo /* No other choice for non-shareable tree */ 227092a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 22713fd0a558SYan, Zheng return root; 22723fd0a558SYan, Zheng 22733fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 22743fd0a558SYan, Zheng fs_root = root; 22753fd0a558SYan, Zheng 22763fd0a558SYan, Zheng if (next != node) 22773fd0a558SYan, Zheng return NULL; 22783fd0a558SYan, Zheng 22793fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 22803fd0a558SYan, Zheng if (!next || next->level <= node->level) 22813fd0a558SYan, Zheng break; 22823fd0a558SYan, Zheng } 22833fd0a558SYan, Zheng 22843fd0a558SYan, Zheng if (!fs_root) 22853fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 22863fd0a558SYan, Zheng return fs_root; 22875d4f98a2SYan Zheng } 22885d4f98a2SYan Zheng 22895d4f98a2SYan Zheng static noinline_for_stack 22903fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 2291a26195a5SQu Wenruo struct btrfs_backref_node *node, int reserve) 22925d4f98a2SYan Zheng { 22930b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2294a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2295a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2296a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22973fd0a558SYan, Zheng u64 num_bytes = 0; 22983fd0a558SYan, Zheng int index = 0; 22995d4f98a2SYan Zheng 23003fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 23013fd0a558SYan, Zheng 23023fd0a558SYan, Zheng while (next) { 23033fd0a558SYan, Zheng cond_resched(); 23045d4f98a2SYan Zheng while (1) { 23053fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 23065d4f98a2SYan Zheng break; 23075d4f98a2SYan Zheng 23080b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 23093fd0a558SYan, Zheng 23103fd0a558SYan, Zheng if (list_empty(&next->upper)) 23113fd0a558SYan, Zheng break; 23123fd0a558SYan, Zheng 23133fd0a558SYan, Zheng edge = list_entry(next->upper.next, 2314a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 23153fd0a558SYan, Zheng edges[index++] = edge; 23163fd0a558SYan, Zheng next = edge->node[UPPER]; 23175d4f98a2SYan Zheng } 23183fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 23193fd0a558SYan, Zheng } 23203fd0a558SYan, Zheng return num_bytes; 23213fd0a558SYan, Zheng } 23223fd0a558SYan, Zheng 23233fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 23243fd0a558SYan, Zheng struct reloc_control *rc, 2325a26195a5SQu Wenruo struct btrfs_backref_node *node) 23263fd0a558SYan, Zheng { 23273fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2328da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23293fd0a558SYan, Zheng u64 num_bytes; 23303fd0a558SYan, Zheng int ret; 23310647bf56SWang Shilong u64 tmp; 23323fd0a558SYan, Zheng 23333fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 23343fd0a558SYan, Zheng 23353fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 23360647bf56SWang Shilong rc->reserved_bytes += num_bytes; 23378ca17f0fSJosef Bacik 23388ca17f0fSJosef Bacik /* 23398ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 23408ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 23418ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 23428ca17f0fSJosef Bacik */ 23439270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes, 23448ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 23453fd0a558SYan, Zheng if (ret) { 2346da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 23470647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 23480647bf56SWang Shilong tmp <<= 1; 23490647bf56SWang Shilong /* 23500647bf56SWang Shilong * only one thread can access block_rsv at this point, 23510647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 23520647bf56SWang Shilong * we expand more reservation size here to allow enough 235352042d8eSAndrea Gelmini * space for relocation and we will return earlier in 23540647bf56SWang Shilong * enospc case. 23550647bf56SWang Shilong */ 2356da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 23570647bf56SWang Shilong RELOCATION_RESERVED_NODES; 23588ca17f0fSJosef Bacik return -EAGAIN; 23593fd0a558SYan, Zheng } 23603fd0a558SYan, Zheng 23613fd0a558SYan, Zheng return 0; 23623fd0a558SYan, Zheng } 23633fd0a558SYan, Zheng 23645d4f98a2SYan Zheng /* 23655d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 23665d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 23675d4f98a2SYan Zheng * 23685d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 23695d4f98a2SYan Zheng * in that case this function just updates pointers. 23705d4f98a2SYan Zheng */ 23715d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 23723fd0a558SYan, Zheng struct reloc_control *rc, 2373a26195a5SQu Wenruo struct btrfs_backref_node *node, 23745d4f98a2SYan Zheng struct btrfs_key *key, 23755d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 23765d4f98a2SYan Zheng { 2377a26195a5SQu Wenruo struct btrfs_backref_node *upper; 2378a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2379a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 23805d4f98a2SYan Zheng struct btrfs_root *root; 23815d4f98a2SYan Zheng struct extent_buffer *eb; 23825d4f98a2SYan Zheng u32 blocksize; 23835d4f98a2SYan Zheng u64 bytenr; 23845d4f98a2SYan Zheng int slot; 23858df01fddSNikolay Borisov int ret = 0; 23865d4f98a2SYan Zheng 2387ffe30dd8SJosef Bacik /* 2388ffe30dd8SJosef Bacik * If we are lowest then this is the first time we're processing this 2389ffe30dd8SJosef Bacik * block, and thus shouldn't have an eb associated with it yet. 2390ffe30dd8SJosef Bacik */ 2391ffe30dd8SJosef Bacik ASSERT(!lowest || !node->eb); 23925d4f98a2SYan Zheng 23935d4f98a2SYan Zheng path->lowest_level = node->level + 1; 23943fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 23955d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 239682fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2397581c1760SQu Wenruo 23985d4f98a2SYan Zheng cond_resched(); 23995d4f98a2SYan Zheng 24005d4f98a2SYan Zheng upper = edge->node[UPPER]; 2401dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 2402cbdc2ebcSJosef Bacik if (IS_ERR(root)) { 2403cbdc2ebcSJosef Bacik ret = PTR_ERR(root); 2404cbdc2ebcSJosef Bacik goto next; 2405cbdc2ebcSJosef Bacik } 24065d4f98a2SYan Zheng 24073fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 24083fd0a558SYan, Zheng if (!lowest) { 2409e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 24108df01fddSNikolay Borisov if (ret < 0) 2411cbca7d59SFilipe Manana goto next; 24123fd0a558SYan, Zheng BUG_ON(ret); 24133fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24143fd0a558SYan, Zheng if (node->eb->start == bytenr) 24153fd0a558SYan, Zheng goto next; 24163fd0a558SYan, Zheng } 2417b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 24183fd0a558SYan, Zheng } 24195d4f98a2SYan Zheng 24205d4f98a2SYan Zheng if (!upper->eb) { 24215d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 24223561b9dbSLiu Bo if (ret) { 24238df01fddSNikolay Borisov if (ret > 0) 24248df01fddSNikolay Borisov ret = -ENOENT; 24253561b9dbSLiu Bo 24263561b9dbSLiu Bo btrfs_release_path(path); 24275d4f98a2SYan Zheng break; 24285d4f98a2SYan Zheng } 24295d4f98a2SYan Zheng 24303fd0a558SYan, Zheng if (!upper->eb) { 24313fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 24323fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 24333fd0a558SYan, Zheng } else { 24343fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 24353fd0a558SYan, Zheng } 24363fd0a558SYan, Zheng 24373fd0a558SYan, Zheng upper->locked = 1; 24383fd0a558SYan, Zheng path->locks[upper->level] = 0; 24393fd0a558SYan, Zheng 24405d4f98a2SYan Zheng slot = path->slots[upper->level]; 2441b3b4aa74SDavid Sterba btrfs_release_path(path); 24425d4f98a2SYan Zheng } else { 2443e3b83361SQu Wenruo ret = btrfs_bin_search(upper->eb, key, &slot); 24448df01fddSNikolay Borisov if (ret < 0) 2445cbca7d59SFilipe Manana goto next; 24465d4f98a2SYan Zheng BUG_ON(ret); 24475d4f98a2SYan Zheng } 24485d4f98a2SYan Zheng 24495d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24503fd0a558SYan, Zheng if (lowest) { 24514547f4d8SLiu Bo if (bytenr != node->bytenr) { 24524547f4d8SLiu Bo btrfs_err(root->fs_info, 24534547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 24544547f4d8SLiu Bo bytenr, node->bytenr, slot, 24554547f4d8SLiu Bo upper->eb->start); 24568df01fddSNikolay Borisov ret = -EIO; 24574547f4d8SLiu Bo goto next; 24584547f4d8SLiu Bo } 24595d4f98a2SYan Zheng } else { 24603fd0a558SYan, Zheng if (node->eb->start == bytenr) 24613fd0a558SYan, Zheng goto next; 24625d4f98a2SYan Zheng } 24635d4f98a2SYan Zheng 2464da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 2465c9752536SJosef Bacik eb = btrfs_read_node_slot(upper->eb, slot); 246664c043deSLiu Bo if (IS_ERR(eb)) { 24678df01fddSNikolay Borisov ret = PTR_ERR(eb); 246864c043deSLiu Bo goto next; 246997d9a8a4STsutomu Itoh } 24705d4f98a2SYan Zheng btrfs_tree_lock(eb); 24715d4f98a2SYan Zheng 24725d4f98a2SYan Zheng if (!node->eb) { 24735d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 24749631e4ccSJosef Bacik slot, &eb, BTRFS_NESTING_COW); 24753fd0a558SYan, Zheng btrfs_tree_unlock(eb); 24763fd0a558SYan, Zheng free_extent_buffer(eb); 24778df01fddSNikolay Borisov if (ret < 0) 24783fd0a558SYan, Zheng goto next; 2479ffe30dd8SJosef Bacik /* 2480ffe30dd8SJosef Bacik * We've just COWed this block, it should have updated 2481ffe30dd8SJosef Bacik * the correct backref node entry. 2482ffe30dd8SJosef Bacik */ 2483ffe30dd8SJosef Bacik ASSERT(node->eb == eb); 24845d4f98a2SYan Zheng } else { 24855d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 24865d4f98a2SYan Zheng node->eb->start); 24875d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 24885d4f98a2SYan Zheng trans->transid); 24895d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 24905d4f98a2SYan Zheng 249182fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 24925d4f98a2SYan Zheng node->eb->start, blocksize, 249382fa113fSQu Wenruo upper->eb->start); 249482fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 2495f42c5da6SNikolay Borisov btrfs_header_owner(upper->eb), 2496f42c5da6SNikolay Borisov root->root_key.objectid, false); 249782fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 2498eb6b7fb4SJosef Bacik if (!ret) 2499eb6b7fb4SJosef Bacik ret = btrfs_drop_subtree(trans, root, eb, 2500eb6b7fb4SJosef Bacik upper->eb); 2501eb6b7fb4SJosef Bacik if (ret) 2502eb6b7fb4SJosef Bacik btrfs_abort_transaction(trans, ret); 25035d4f98a2SYan Zheng } 25043fd0a558SYan, Zheng next: 25053fd0a558SYan, Zheng if (!upper->pending) 2506b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 25073fd0a558SYan, Zheng else 2508b0fe7078SQu Wenruo btrfs_backref_unlock_node_buffer(upper); 25098df01fddSNikolay Borisov if (ret) 25103fd0a558SYan, Zheng break; 25115d4f98a2SYan Zheng } 25123fd0a558SYan, Zheng 25138df01fddSNikolay Borisov if (!ret && node->pending) { 2514b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 25153fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 25163fd0a558SYan, Zheng node->pending = 0; 25175d4f98a2SYan Zheng } 25183fd0a558SYan, Zheng 25195d4f98a2SYan Zheng path->lowest_level = 0; 2520ffe30dd8SJosef Bacik 2521ffe30dd8SJosef Bacik /* 2522ffe30dd8SJosef Bacik * We should have allocated all of our space in the block rsv and thus 2523ffe30dd8SJosef Bacik * shouldn't ENOSPC. 2524ffe30dd8SJosef Bacik */ 2525ffe30dd8SJosef Bacik ASSERT(ret != -ENOSPC); 25268df01fddSNikolay Borisov return ret; 25275d4f98a2SYan Zheng } 25285d4f98a2SYan Zheng 25295d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 25303fd0a558SYan, Zheng struct reloc_control *rc, 2531a26195a5SQu Wenruo struct btrfs_backref_node *node, 25325d4f98a2SYan Zheng struct btrfs_path *path) 25335d4f98a2SYan Zheng { 25345d4f98a2SYan Zheng struct btrfs_key key; 25355d4f98a2SYan Zheng 25365d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 25373fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 25385d4f98a2SYan Zheng } 25395d4f98a2SYan Zheng 25405d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 25413fd0a558SYan, Zheng struct reloc_control *rc, 25423fd0a558SYan, Zheng struct btrfs_path *path, int err) 25435d4f98a2SYan Zheng { 25443fd0a558SYan, Zheng LIST_HEAD(list); 2545a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 2546a26195a5SQu Wenruo struct btrfs_backref_node *node; 25475d4f98a2SYan Zheng int level; 25485d4f98a2SYan Zheng int ret; 25495d4f98a2SYan Zheng 25505d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 25515d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 25525d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 2553a26195a5SQu Wenruo struct btrfs_backref_node, list); 25543fd0a558SYan, Zheng list_move_tail(&node->list, &list); 25553fd0a558SYan, Zheng BUG_ON(!node->pending); 25565d4f98a2SYan Zheng 25573fd0a558SYan, Zheng if (!err) { 25583fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 25595d4f98a2SYan Zheng if (ret < 0) 25605d4f98a2SYan Zheng err = ret; 25615d4f98a2SYan Zheng } 25625d4f98a2SYan Zheng } 25633fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 25643fd0a558SYan, Zheng } 25655d4f98a2SYan Zheng return err; 25665d4f98a2SYan Zheng } 25675d4f98a2SYan Zheng 25685d4f98a2SYan Zheng /* 25695d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 25705d4f98a2SYan Zheng * as processed. 25715d4f98a2SYan Zheng */ 25725d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 2573a26195a5SQu Wenruo struct btrfs_backref_node *node) 25745d4f98a2SYan Zheng { 2575a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2576a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2577a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25785d4f98a2SYan Zheng int index = 0; 25795d4f98a2SYan Zheng 25805d4f98a2SYan Zheng while (next) { 25815d4f98a2SYan Zheng cond_resched(); 25825d4f98a2SYan Zheng while (1) { 25835d4f98a2SYan Zheng if (next->processed) 25845d4f98a2SYan Zheng break; 25855d4f98a2SYan Zheng 25869569cc20SQu Wenruo mark_block_processed(rc, next); 25875d4f98a2SYan Zheng 25885d4f98a2SYan Zheng if (list_empty(&next->upper)) 25895d4f98a2SYan Zheng break; 25905d4f98a2SYan Zheng 25915d4f98a2SYan Zheng edge = list_entry(next->upper.next, 2592a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 25935d4f98a2SYan Zheng edges[index++] = edge; 25945d4f98a2SYan Zheng next = edge->node[UPPER]; 25955d4f98a2SYan Zheng } 25965d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25975d4f98a2SYan Zheng } 25985d4f98a2SYan Zheng } 25995d4f98a2SYan Zheng 26007476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 26015d4f98a2SYan Zheng { 2602da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 26037476dfdaSDavid Sterba 26045d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 26059655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 26065d4f98a2SYan Zheng return 1; 26075d4f98a2SYan Zheng return 0; 26085d4f98a2SYan Zheng } 26095d4f98a2SYan Zheng 26102ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 26115d4f98a2SYan Zheng struct tree_block *block) 26125d4f98a2SYan Zheng { 26135d4f98a2SYan Zheng struct extent_buffer *eb; 26145d4f98a2SYan Zheng 2615f7ba2d37SJosef Bacik eb = read_tree_block(fs_info, block->bytenr, block->owner, 2616f7ba2d37SJosef Bacik block->key.offset, block->level, NULL); 26174eb150d6SQu Wenruo if (IS_ERR(eb)) 261864c043deSLiu Bo return PTR_ERR(eb); 26194eb150d6SQu Wenruo if (!extent_buffer_uptodate(eb)) { 2620416bc658SJosef Bacik free_extent_buffer(eb); 2621416bc658SJosef Bacik return -EIO; 2622416bc658SJosef Bacik } 26235d4f98a2SYan Zheng if (block->level == 0) 26245d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 26255d4f98a2SYan Zheng else 26265d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 26275d4f98a2SYan Zheng free_extent_buffer(eb); 26285d4f98a2SYan Zheng block->key_ready = 1; 26295d4f98a2SYan Zheng return 0; 26305d4f98a2SYan Zheng } 26315d4f98a2SYan Zheng 26325d4f98a2SYan Zheng /* 26335d4f98a2SYan Zheng * helper function to relocate a tree block 26345d4f98a2SYan Zheng */ 26355d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 26365d4f98a2SYan Zheng struct reloc_control *rc, 2637a26195a5SQu Wenruo struct btrfs_backref_node *node, 26385d4f98a2SYan Zheng struct btrfs_key *key, 26395d4f98a2SYan Zheng struct btrfs_path *path) 26405d4f98a2SYan Zheng { 26415d4f98a2SYan Zheng struct btrfs_root *root; 26423fd0a558SYan, Zheng int ret = 0; 26435d4f98a2SYan Zheng 26443fd0a558SYan, Zheng if (!node) 26455d4f98a2SYan Zheng return 0; 26463fd0a558SYan, Zheng 26475f6b2e5cSJosef Bacik /* 26485f6b2e5cSJosef Bacik * If we fail here we want to drop our backref_node because we are going 26495f6b2e5cSJosef Bacik * to start over and regenerate the tree for it. 26505f6b2e5cSJosef Bacik */ 26515f6b2e5cSJosef Bacik ret = reserve_metadata_space(trans, rc, node); 26525f6b2e5cSJosef Bacik if (ret) 26535f6b2e5cSJosef Bacik goto out; 26545f6b2e5cSJosef Bacik 26553fd0a558SYan, Zheng BUG_ON(node->processed); 2656147d256eSZhaolei root = select_one_root(node); 26578717cf44SJosef Bacik if (IS_ERR(root)) { 26588717cf44SJosef Bacik ret = PTR_ERR(root); 26598717cf44SJosef Bacik 26608717cf44SJosef Bacik /* See explanation in select_one_root for the -EUCLEAN case. */ 26618717cf44SJosef Bacik ASSERT(ret == -ENOENT); 26628717cf44SJosef Bacik if (ret == -ENOENT) { 26638717cf44SJosef Bacik ret = 0; 26643fd0a558SYan, Zheng update_processed_blocks(rc, node); 26658717cf44SJosef Bacik } 26663fd0a558SYan, Zheng goto out; 26675d4f98a2SYan Zheng } 26685d4f98a2SYan Zheng 26693fd0a558SYan, Zheng if (root) { 267092a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 26711c7bfa15SJosef Bacik /* 26721c7bfa15SJosef Bacik * This block was the root block of a root, and this is 26731c7bfa15SJosef Bacik * the first time we're processing the block and thus it 26741c7bfa15SJosef Bacik * should not have had the ->new_bytenr modified and 26751c7bfa15SJosef Bacik * should have not been included on the changed list. 26761c7bfa15SJosef Bacik * 26771c7bfa15SJosef Bacik * However in the case of corruption we could have 26781c7bfa15SJosef Bacik * multiple refs pointing to the same block improperly, 26791c7bfa15SJosef Bacik * and thus we would trip over these checks. ASSERT() 26801c7bfa15SJosef Bacik * for the developer case, because it could indicate a 26811c7bfa15SJosef Bacik * bug in the backref code, however error out for a 26821c7bfa15SJosef Bacik * normal user in the case of corruption. 26831c7bfa15SJosef Bacik */ 26841c7bfa15SJosef Bacik ASSERT(node->new_bytenr == 0); 26851c7bfa15SJosef Bacik ASSERT(list_empty(&node->list)); 26861c7bfa15SJosef Bacik if (node->new_bytenr || !list_empty(&node->list)) { 26871c7bfa15SJosef Bacik btrfs_err(root->fs_info, 26881c7bfa15SJosef Bacik "bytenr %llu has improper references to it", 26891c7bfa15SJosef Bacik node->bytenr); 26901c7bfa15SJosef Bacik ret = -EUCLEAN; 26911c7bfa15SJosef Bacik goto out; 26921c7bfa15SJosef Bacik } 2693d18c7bd9SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 2694d18c7bd9SJosef Bacik if (ret) 2695d18c7bd9SJosef Bacik goto out; 269639200e59SJosef Bacik /* 269739200e59SJosef Bacik * Another thread could have failed, need to check if we 269839200e59SJosef Bacik * have reloc_root actually set. 269939200e59SJosef Bacik */ 270039200e59SJosef Bacik if (!root->reloc_root) { 270139200e59SJosef Bacik ret = -ENOENT; 270239200e59SJosef Bacik goto out; 270339200e59SJosef Bacik } 27043fd0a558SYan, Zheng root = root->reloc_root; 27053fd0a558SYan, Zheng node->new_bytenr = root->node->start; 270600246528SJosef Bacik btrfs_put_root(node->root); 270700246528SJosef Bacik node->root = btrfs_grab_root(root); 27080b530bc5SJosef Bacik ASSERT(node->root); 27093fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 27103fd0a558SYan, Zheng } else { 27115d4f98a2SYan Zheng path->lowest_level = node->level; 27122bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 27132bb2e00eSFilipe Manana btrfs_reserve_chunk_metadata(trans, false); 27145d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2715b3b4aa74SDavid Sterba btrfs_release_path(path); 27162bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 27172bb2e00eSFilipe Manana btrfs_trans_release_chunk_metadata(trans); 27183fd0a558SYan, Zheng if (ret > 0) 27195d4f98a2SYan Zheng ret = 0; 27203fd0a558SYan, Zheng } 27213fd0a558SYan, Zheng if (!ret) 27223fd0a558SYan, Zheng update_processed_blocks(rc, node); 27233fd0a558SYan, Zheng } else { 27243fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 27253fd0a558SYan, Zheng } 27265d4f98a2SYan Zheng out: 27270647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 2728023acb07SQu Wenruo btrfs_backref_cleanup_node(&rc->backref_cache, node); 27295d4f98a2SYan Zheng return ret; 27305d4f98a2SYan Zheng } 27315d4f98a2SYan Zheng 27325d4f98a2SYan Zheng /* 27335d4f98a2SYan Zheng * relocate a list of blocks 27345d4f98a2SYan Zheng */ 27355d4f98a2SYan Zheng static noinline_for_stack 27365d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 27375d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 27385d4f98a2SYan Zheng { 27392ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2740a26195a5SQu Wenruo struct btrfs_backref_node *node; 27415d4f98a2SYan Zheng struct btrfs_path *path; 27425d4f98a2SYan Zheng struct tree_block *block; 274398ff7b94SQu Wenruo struct tree_block *next; 27445d4f98a2SYan Zheng int ret; 27455d4f98a2SYan Zheng int err = 0; 27465d4f98a2SYan Zheng 27475d4f98a2SYan Zheng path = btrfs_alloc_path(); 2748e1a12670SLiu Bo if (!path) { 2749e1a12670SLiu Bo err = -ENOMEM; 275034c2b290SDavid Sterba goto out_free_blocks; 2751e1a12670SLiu Bo } 27525d4f98a2SYan Zheng 275398ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 275498ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27555d4f98a2SYan Zheng if (!block->key_ready) 2756f7ba2d37SJosef Bacik btrfs_readahead_tree_block(fs_info, block->bytenr, 2757f7ba2d37SJosef Bacik block->owner, 0, 27583fbaf258SJosef Bacik block->level); 27595d4f98a2SYan Zheng } 27605d4f98a2SYan Zheng 276198ff7b94SQu Wenruo /* Get first keys */ 276298ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 276334c2b290SDavid Sterba if (!block->key_ready) { 27642ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 276534c2b290SDavid Sterba if (err) 276634c2b290SDavid Sterba goto out_free_path; 276734c2b290SDavid Sterba } 27685d4f98a2SYan Zheng } 27695d4f98a2SYan Zheng 277098ff7b94SQu Wenruo /* Do tree relocation */ 277198ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27723fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 27735d4f98a2SYan Zheng block->level, block->bytenr); 27745d4f98a2SYan Zheng if (IS_ERR(node)) { 27755d4f98a2SYan Zheng err = PTR_ERR(node); 27765d4f98a2SYan Zheng goto out; 27775d4f98a2SYan Zheng } 27785d4f98a2SYan Zheng 27795d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 27805d4f98a2SYan Zheng path); 27815d4f98a2SYan Zheng if (ret < 0) { 27825d4f98a2SYan Zheng err = ret; 278350dbbb71SJosef Bacik break; 27845d4f98a2SYan Zheng } 27855d4f98a2SYan Zheng } 27865d4f98a2SYan Zheng out: 27873fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 27885d4f98a2SYan Zheng 278934c2b290SDavid Sterba out_free_path: 27905d4f98a2SYan Zheng btrfs_free_path(path); 279134c2b290SDavid Sterba out_free_blocks: 2792e1a12670SLiu Bo free_block_list(blocks); 27935d4f98a2SYan Zheng return err; 27945d4f98a2SYan Zheng } 27955d4f98a2SYan Zheng 2796056d9becSNikolay Borisov static noinline_for_stack int prealloc_file_extent_cluster( 2797056d9becSNikolay Borisov struct btrfs_inode *inode, 2798efa56464SYan, Zheng struct file_extent_cluster *cluster) 2799efa56464SYan, Zheng { 2800efa56464SYan, Zheng u64 alloc_hint = 0; 2801efa56464SYan, Zheng u64 start; 2802efa56464SYan, Zheng u64 end; 2803056d9becSNikolay Borisov u64 offset = inode->index_cnt; 2804efa56464SYan, Zheng u64 num_bytes; 28054e9d0d01SNikolay Borisov int nr; 2806efa56464SYan, Zheng int ret = 0; 28079d9ea1e6SQu Wenruo u64 i_size = i_size_read(&inode->vfs_inode); 2808dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 2809dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 2810214e61d0SNikolay Borisov u64 cur_offset = prealloc_start; 2811efa56464SYan, Zheng 28129d9ea1e6SQu Wenruo /* 28139d9ea1e6SQu Wenruo * For subpage case, previous i_size may not be aligned to PAGE_SIZE. 28149d9ea1e6SQu Wenruo * This means the range [i_size, PAGE_END + 1) is filled with zeros by 28159d9ea1e6SQu Wenruo * btrfs_do_readpage() call of previously relocated file cluster. 28169d9ea1e6SQu Wenruo * 28179d9ea1e6SQu Wenruo * If the current cluster starts in the above range, btrfs_do_readpage() 28189d9ea1e6SQu Wenruo * will skip the read, and relocate_one_page() will later writeback 28199d9ea1e6SQu Wenruo * the padding zeros as new data, causing data corruption. 28209d9ea1e6SQu Wenruo * 28219d9ea1e6SQu Wenruo * Here we have to manually invalidate the range (i_size, PAGE_END + 1). 28229d9ea1e6SQu Wenruo */ 28239d9ea1e6SQu Wenruo if (!IS_ALIGNED(i_size, PAGE_SIZE)) { 28249d9ea1e6SQu Wenruo struct address_space *mapping = inode->vfs_inode.i_mapping; 28259d9ea1e6SQu Wenruo struct btrfs_fs_info *fs_info = inode->root->fs_info; 28269d9ea1e6SQu Wenruo const u32 sectorsize = fs_info->sectorsize; 28279d9ea1e6SQu Wenruo struct page *page; 28289d9ea1e6SQu Wenruo 28299d9ea1e6SQu Wenruo ASSERT(sectorsize < PAGE_SIZE); 28309d9ea1e6SQu Wenruo ASSERT(IS_ALIGNED(i_size, sectorsize)); 28319d9ea1e6SQu Wenruo 28329d9ea1e6SQu Wenruo /* 28339d9ea1e6SQu Wenruo * Subpage can't handle page with DIRTY but without UPTODATE 28349d9ea1e6SQu Wenruo * bit as it can lead to the following deadlock: 28359d9ea1e6SQu Wenruo * 2836fb12489bSMatthew Wilcox (Oracle) * btrfs_read_folio() 28379d9ea1e6SQu Wenruo * | Page already *locked* 28389d9ea1e6SQu Wenruo * |- btrfs_lock_and_flush_ordered_range() 28399d9ea1e6SQu Wenruo * |- btrfs_start_ordered_extent() 28409d9ea1e6SQu Wenruo * |- extent_write_cache_pages() 28419d9ea1e6SQu Wenruo * |- lock_page() 28429d9ea1e6SQu Wenruo * We try to lock the page we already hold. 28439d9ea1e6SQu Wenruo * 28449d9ea1e6SQu Wenruo * Here we just writeback the whole data reloc inode, so that 28459d9ea1e6SQu Wenruo * we will be ensured to have no dirty range in the page, and 28469d9ea1e6SQu Wenruo * are safe to clear the uptodate bits. 28479d9ea1e6SQu Wenruo * 28489d9ea1e6SQu Wenruo * This shouldn't cause too much overhead, as we need to write 28499d9ea1e6SQu Wenruo * the data back anyway. 28509d9ea1e6SQu Wenruo */ 28519d9ea1e6SQu Wenruo ret = filemap_write_and_wait(mapping); 28529d9ea1e6SQu Wenruo if (ret < 0) 28539d9ea1e6SQu Wenruo return ret; 28549d9ea1e6SQu Wenruo 28559d9ea1e6SQu Wenruo clear_extent_bits(&inode->io_tree, i_size, 28569d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - 1, 28579d9ea1e6SQu Wenruo EXTENT_UPTODATE); 28589d9ea1e6SQu Wenruo page = find_lock_page(mapping, i_size >> PAGE_SHIFT); 28599d9ea1e6SQu Wenruo /* 28609d9ea1e6SQu Wenruo * If page is freed we don't need to do anything then, as we 28619d9ea1e6SQu Wenruo * will re-read the whole page anyway. 28629d9ea1e6SQu Wenruo */ 28639d9ea1e6SQu Wenruo if (page) { 28649d9ea1e6SQu Wenruo btrfs_subpage_clear_uptodate(fs_info, page, i_size, 28659d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - i_size); 28669d9ea1e6SQu Wenruo unlock_page(page); 28679d9ea1e6SQu Wenruo put_page(page); 28689d9ea1e6SQu Wenruo } 28699d9ea1e6SQu Wenruo } 28709d9ea1e6SQu Wenruo 2871efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 2872056d9becSNikolay Borisov ret = btrfs_alloc_data_chunk_ondemand(inode, 2873dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 2874efa56464SYan, Zheng if (ret) 2875214e61d0SNikolay Borisov return ret; 2876efa56464SYan, Zheng 287729b6352bSDavid Sterba btrfs_inode_lock(inode, 0); 28784e9d0d01SNikolay Borisov for (nr = 0; nr < cluster->nr; nr++) { 28799c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 28809c5c9604SJosef Bacik 2881efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 2882efa56464SYan, Zheng if (nr + 1 < cluster->nr) 2883efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 2884efa56464SYan, Zheng else 2885efa56464SYan, Zheng end = cluster->end - offset; 2886efa56464SYan, Zheng 28879c5c9604SJosef Bacik lock_extent(&inode->io_tree, start, end, &cached_state); 2888efa56464SYan, Zheng num_bytes = end + 1 - start; 2889056d9becSNikolay Borisov ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start, 2890efa56464SYan, Zheng num_bytes, num_bytes, 2891efa56464SYan, Zheng end + 1, &alloc_hint); 289218513091SWang Xiaoguang cur_offset = end + 1; 28939c5c9604SJosef Bacik unlock_extent(&inode->io_tree, start, end, &cached_state); 2894efa56464SYan, Zheng if (ret) 2895efa56464SYan, Zheng break; 2896efa56464SYan, Zheng } 2897*e5d4d75bSDavid Sterba btrfs_inode_unlock(inode, 0); 2898214e61d0SNikolay Borisov 289918513091SWang Xiaoguang if (cur_offset < prealloc_end) 2900056d9becSNikolay Borisov btrfs_free_reserved_data_space_noquota(inode->root->fs_info, 2901a89ef455SFilipe Manana prealloc_end + 1 - cur_offset); 2902efa56464SYan, Zheng return ret; 2903efa56464SYan, Zheng } 2904efa56464SYan, Zheng 29054b01c44fSJohannes Thumshirn static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode, 29064b01c44fSJohannes Thumshirn u64 start, u64 end, u64 block_start) 29075d4f98a2SYan Zheng { 29085d4f98a2SYan Zheng struct extent_map *em; 29099c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 29100257bb82SYan, Zheng int ret = 0; 29115d4f98a2SYan Zheng 2912172ddd60SDavid Sterba em = alloc_extent_map(); 29130257bb82SYan, Zheng if (!em) 29140257bb82SYan, Zheng return -ENOMEM; 29150257bb82SYan, Zheng 29165d4f98a2SYan Zheng em->start = start; 29170257bb82SYan, Zheng em->len = end + 1 - start; 29180257bb82SYan, Zheng em->block_len = em->len; 29190257bb82SYan, Zheng em->block_start = block_start; 29205d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 29215d4f98a2SYan Zheng 29229c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 2923a1ba4c08SFilipe Manana ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, false); 29249c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 2925a1ba4c08SFilipe Manana free_extent_map(em); 2926a1ba4c08SFilipe Manana 29270257bb82SYan, Zheng return ret; 29280257bb82SYan, Zheng } 29295d4f98a2SYan Zheng 2930726a3421SQu Wenruo /* 2931907d2710SDavid Sterba * Allow error injection to test balance/relocation cancellation 2932726a3421SQu Wenruo */ 29331fec12a5SJosef Bacik noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 2934726a3421SQu Wenruo { 29355cb502f4SQu Wenruo return atomic_read(&fs_info->balance_cancel_req) || 2936907d2710SDavid Sterba atomic_read(&fs_info->reloc_cancel_req) || 29375cb502f4SQu Wenruo fatal_signal_pending(current); 2938726a3421SQu Wenruo } 2939726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 2940726a3421SQu Wenruo 2941c2832898SQu Wenruo static u64 get_cluster_boundary_end(struct file_extent_cluster *cluster, 2942c2832898SQu Wenruo int cluster_nr) 2943c2832898SQu Wenruo { 2944c2832898SQu Wenruo /* Last extent, use cluster end directly */ 2945c2832898SQu Wenruo if (cluster_nr >= cluster->nr - 1) 2946c2832898SQu Wenruo return cluster->end; 2947c2832898SQu Wenruo 2948c2832898SQu Wenruo /* Use next boundary start*/ 2949c2832898SQu Wenruo return cluster->boundary[cluster_nr + 1] - 1; 2950c2832898SQu Wenruo } 2951c2832898SQu Wenruo 2952f47960f4SQu Wenruo static int relocate_one_page(struct inode *inode, struct file_ra_state *ra, 2953f47960f4SQu Wenruo struct file_extent_cluster *cluster, 2954f47960f4SQu Wenruo int *cluster_nr, unsigned long page_index) 2955f47960f4SQu Wenruo { 2956f47960f4SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 2957f47960f4SQu Wenruo u64 offset = BTRFS_I(inode)->index_cnt; 2958f47960f4SQu Wenruo const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT; 2959f47960f4SQu Wenruo gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 2960f47960f4SQu Wenruo struct page *page; 2961f47960f4SQu Wenruo u64 page_start; 2962f47960f4SQu Wenruo u64 page_end; 2963c2832898SQu Wenruo u64 cur; 2964f47960f4SQu Wenruo int ret; 2965f47960f4SQu Wenruo 2966f47960f4SQu Wenruo ASSERT(page_index <= last_index); 2967f47960f4SQu Wenruo page = find_lock_page(inode->i_mapping, page_index); 2968f47960f4SQu Wenruo if (!page) { 2969f47960f4SQu Wenruo page_cache_sync_readahead(inode->i_mapping, ra, NULL, 2970f47960f4SQu Wenruo page_index, last_index + 1 - page_index); 2971f47960f4SQu Wenruo page = find_or_create_page(inode->i_mapping, page_index, mask); 2972c2832898SQu Wenruo if (!page) 2973c2832898SQu Wenruo return -ENOMEM; 2974f47960f4SQu Wenruo } 2975f47960f4SQu Wenruo ret = set_page_extent_mapped(page); 2976f47960f4SQu Wenruo if (ret < 0) 2977f47960f4SQu Wenruo goto release_page; 2978f47960f4SQu Wenruo 2979f47960f4SQu Wenruo if (PageReadahead(page)) 29802ebdd1dfSMatthew Wilcox (Oracle) page_cache_async_readahead(inode->i_mapping, ra, NULL, 29812ebdd1dfSMatthew Wilcox (Oracle) page_folio(page), page_index, 29822ebdd1dfSMatthew Wilcox (Oracle) last_index + 1 - page_index); 2983f47960f4SQu Wenruo 2984f47960f4SQu Wenruo if (!PageUptodate(page)) { 2985fb12489bSMatthew Wilcox (Oracle) btrfs_read_folio(NULL, page_folio(page)); 2986f47960f4SQu Wenruo lock_page(page); 2987f47960f4SQu Wenruo if (!PageUptodate(page)) { 2988f47960f4SQu Wenruo ret = -EIO; 2989f47960f4SQu Wenruo goto release_page; 2990f47960f4SQu Wenruo } 2991f47960f4SQu Wenruo } 2992f47960f4SQu Wenruo 2993f47960f4SQu Wenruo page_start = page_offset(page); 2994f47960f4SQu Wenruo page_end = page_start + PAGE_SIZE - 1; 2995f47960f4SQu Wenruo 2996c2832898SQu Wenruo /* 2997c2832898SQu Wenruo * Start from the cluster, as for subpage case, the cluster can start 2998c2832898SQu Wenruo * inside the page. 2999c2832898SQu Wenruo */ 3000c2832898SQu Wenruo cur = max(page_start, cluster->boundary[*cluster_nr] - offset); 3001c2832898SQu Wenruo while (cur <= page_end) { 30029c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 3003c2832898SQu Wenruo u64 extent_start = cluster->boundary[*cluster_nr] - offset; 3004c2832898SQu Wenruo u64 extent_end = get_cluster_boundary_end(cluster, 3005c2832898SQu Wenruo *cluster_nr) - offset; 3006c2832898SQu Wenruo u64 clamped_start = max(page_start, extent_start); 3007c2832898SQu Wenruo u64 clamped_end = min(page_end, extent_end); 3008c2832898SQu Wenruo u32 clamped_len = clamped_end + 1 - clamped_start; 3009f47960f4SQu Wenruo 3010c2832898SQu Wenruo /* Reserve metadata for this range */ 3011c2832898SQu Wenruo ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 3012d4135134SFilipe Manana clamped_len, clamped_len, 3013d4135134SFilipe Manana false); 3014c2832898SQu Wenruo if (ret) 3015f47960f4SQu Wenruo goto release_page; 3016f47960f4SQu Wenruo 3017c2832898SQu Wenruo /* Mark the range delalloc and dirty for later writeback */ 30189c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, 30199c5c9604SJosef Bacik &cached_state); 3020c2832898SQu Wenruo ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start, 30219c5c9604SJosef Bacik clamped_end, 0, &cached_state); 3022c2832898SQu Wenruo if (ret) { 30239c5c9604SJosef Bacik clear_extent_bit(&BTRFS_I(inode)->io_tree, 3024c2832898SQu Wenruo clamped_start, clamped_end, 30259c5c9604SJosef Bacik EXTENT_LOCKED | EXTENT_BOUNDARY, 30269c5c9604SJosef Bacik &cached_state); 3027c2832898SQu Wenruo btrfs_delalloc_release_metadata(BTRFS_I(inode), 3028c2832898SQu Wenruo clamped_len, true); 3029c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), 3030c2832898SQu Wenruo clamped_len); 3031c2832898SQu Wenruo goto release_page; 3032f47960f4SQu Wenruo } 3033c2832898SQu Wenruo btrfs_page_set_dirty(fs_info, page, clamped_start, clamped_len); 3034f47960f4SQu Wenruo 3035c2832898SQu Wenruo /* 3036c2832898SQu Wenruo * Set the boundary if it's inside the page. 3037c2832898SQu Wenruo * Data relocation requires the destination extents to have the 3038c2832898SQu Wenruo * same size as the source. 3039c2832898SQu Wenruo * EXTENT_BOUNDARY bit prevents current extent from being merged 3040c2832898SQu Wenruo * with previous extent. 3041c2832898SQu Wenruo */ 3042c2832898SQu Wenruo if (in_range(cluster->boundary[*cluster_nr] - offset, 3043c2832898SQu Wenruo page_start, PAGE_SIZE)) { 3044c2832898SQu Wenruo u64 boundary_start = cluster->boundary[*cluster_nr] - 3045c2832898SQu Wenruo offset; 3046c2832898SQu Wenruo u64 boundary_end = boundary_start + 3047c2832898SQu Wenruo fs_info->sectorsize - 1; 3048c2832898SQu Wenruo 3049c2832898SQu Wenruo set_extent_bits(&BTRFS_I(inode)->io_tree, 3050c2832898SQu Wenruo boundary_start, boundary_end, 3051c2832898SQu Wenruo EXTENT_BOUNDARY); 3052c2832898SQu Wenruo } 30539c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, 30549c5c9604SJosef Bacik &cached_state); 3055c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len); 3056c2832898SQu Wenruo cur += clamped_len; 3057c2832898SQu Wenruo 3058c2832898SQu Wenruo /* Crossed extent end, go to next extent */ 3059c2832898SQu Wenruo if (cur >= extent_end) { 3060c2832898SQu Wenruo (*cluster_nr)++; 3061c2832898SQu Wenruo /* Just finished the last extent of the cluster, exit. */ 3062c2832898SQu Wenruo if (*cluster_nr >= cluster->nr) 3063c2832898SQu Wenruo break; 3064c2832898SQu Wenruo } 3065c2832898SQu Wenruo } 3066f47960f4SQu Wenruo unlock_page(page); 3067f47960f4SQu Wenruo put_page(page); 3068f47960f4SQu Wenruo 3069f47960f4SQu Wenruo balance_dirty_pages_ratelimited(inode->i_mapping); 3070f47960f4SQu Wenruo btrfs_throttle(fs_info); 3071f47960f4SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) 3072f47960f4SQu Wenruo ret = -ECANCELED; 3073f47960f4SQu Wenruo return ret; 3074f47960f4SQu Wenruo 3075f47960f4SQu Wenruo release_page: 3076f47960f4SQu Wenruo unlock_page(page); 3077f47960f4SQu Wenruo put_page(page); 3078f47960f4SQu Wenruo return ret; 3079f47960f4SQu Wenruo } 3080f47960f4SQu Wenruo 30810257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 30820257bb82SYan, Zheng struct file_extent_cluster *cluster) 30830257bb82SYan, Zheng { 30840257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 30850257bb82SYan, Zheng unsigned long index; 30860257bb82SYan, Zheng unsigned long last_index; 30870257bb82SYan, Zheng struct file_ra_state *ra; 3088f47960f4SQu Wenruo int cluster_nr = 0; 30890257bb82SYan, Zheng int ret = 0; 30900257bb82SYan, Zheng 30910257bb82SYan, Zheng if (!cluster->nr) 30920257bb82SYan, Zheng return 0; 30930257bb82SYan, Zheng 30940257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 30950257bb82SYan, Zheng if (!ra) 30960257bb82SYan, Zheng return -ENOMEM; 30970257bb82SYan, Zheng 3098056d9becSNikolay Borisov ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster); 30990257bb82SYan, Zheng if (ret) 3100efa56464SYan, Zheng goto out; 31010257bb82SYan, Zheng 31020257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 31030257bb82SYan, Zheng 31044b01c44fSJohannes Thumshirn ret = setup_relocation_extent_mapping(inode, cluster->start - offset, 3105efa56464SYan, Zheng cluster->end - offset, cluster->start); 3106efa56464SYan, Zheng if (ret) 3107efa56464SYan, Zheng goto out; 3108efa56464SYan, Zheng 310909cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 3110f47960f4SQu Wenruo for (index = (cluster->start - offset) >> PAGE_SHIFT; 3111f47960f4SQu Wenruo index <= last_index && !ret; index++) 3112f47960f4SQu Wenruo ret = relocate_one_page(inode, ra, cluster, &cluster_nr, index); 3113f47960f4SQu Wenruo if (ret == 0) 3114f47960f4SQu Wenruo WARN_ON(cluster_nr != cluster->nr); 3115efa56464SYan, Zheng out: 31160257bb82SYan, Zheng kfree(ra); 31170257bb82SYan, Zheng return ret; 31180257bb82SYan, Zheng } 31190257bb82SYan, Zheng 31200257bb82SYan, Zheng static noinline_for_stack 31210257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 31220257bb82SYan, Zheng struct file_extent_cluster *cluster) 31230257bb82SYan, Zheng { 31240257bb82SYan, Zheng int ret; 31250257bb82SYan, Zheng 31260257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 31270257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31280257bb82SYan, Zheng if (ret) 31290257bb82SYan, Zheng return ret; 31300257bb82SYan, Zheng cluster->nr = 0; 31310257bb82SYan, Zheng } 31320257bb82SYan, Zheng 31330257bb82SYan, Zheng if (!cluster->nr) 31340257bb82SYan, Zheng cluster->start = extent_key->objectid; 31350257bb82SYan, Zheng else 31360257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 31370257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 31380257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 31390257bb82SYan, Zheng cluster->nr++; 31400257bb82SYan, Zheng 31410257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 31420257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31430257bb82SYan, Zheng if (ret) 31440257bb82SYan, Zheng return ret; 31450257bb82SYan, Zheng cluster->nr = 0; 31460257bb82SYan, Zheng } 31470257bb82SYan, Zheng return 0; 31485d4f98a2SYan Zheng } 31495d4f98a2SYan Zheng 31505d4f98a2SYan Zheng /* 31515d4f98a2SYan Zheng * helper to add a tree block to the list. 31525d4f98a2SYan Zheng * the major work is getting the generation and level of the block 31535d4f98a2SYan Zheng */ 31545d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 31555d4f98a2SYan Zheng struct btrfs_key *extent_key, 31565d4f98a2SYan Zheng struct btrfs_path *path, 31575d4f98a2SYan Zheng struct rb_root *blocks) 31585d4f98a2SYan Zheng { 31595d4f98a2SYan Zheng struct extent_buffer *eb; 31605d4f98a2SYan Zheng struct btrfs_extent_item *ei; 31615d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 31625d4f98a2SYan Zheng struct tree_block *block; 31635d4f98a2SYan Zheng struct rb_node *rb_node; 31645d4f98a2SYan Zheng u32 item_size; 31655d4f98a2SYan Zheng int level = -1; 31667fdf4b60SWang Shilong u64 generation; 3167f7ba2d37SJosef Bacik u64 owner = 0; 31685d4f98a2SYan Zheng 31695d4f98a2SYan Zheng eb = path->nodes[0]; 31703212fa14SJosef Bacik item_size = btrfs_item_size(eb, path->slots[0]); 31715d4f98a2SYan Zheng 31723173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 31733173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 3174f7ba2d37SJosef Bacik unsigned long ptr = 0, end; 3175f7ba2d37SJosef Bacik 31765d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 31775d4f98a2SYan Zheng struct btrfs_extent_item); 3178f7ba2d37SJosef Bacik end = (unsigned long)ei + item_size; 31793173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 31805d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 31815d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 3182f7ba2d37SJosef Bacik ptr = (unsigned long)(bi + 1); 31835d4f98a2SYan Zheng } else { 31843173a18fSJosef Bacik level = (int)extent_key->offset; 3185f7ba2d37SJosef Bacik ptr = (unsigned long)(ei + 1); 31863173a18fSJosef Bacik } 31873173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 3188f7ba2d37SJosef Bacik 3189f7ba2d37SJosef Bacik /* 3190f7ba2d37SJosef Bacik * We're reading random blocks without knowing their owner ahead 3191f7ba2d37SJosef Bacik * of time. This is ok most of the time, as all reloc roots and 3192f7ba2d37SJosef Bacik * fs roots have the same lock type. However normal trees do 3193f7ba2d37SJosef Bacik * not, and the only way to know ahead of time is to read the 3194f7ba2d37SJosef Bacik * inline ref offset. We know it's an fs root if 3195f7ba2d37SJosef Bacik * 3196f7ba2d37SJosef Bacik * 1. There's more than one ref. 3197f7ba2d37SJosef Bacik * 2. There's a SHARED_DATA_REF_KEY set. 3198f7ba2d37SJosef Bacik * 3. FULL_BACKREF is set on the flags. 3199f7ba2d37SJosef Bacik * 3200f7ba2d37SJosef Bacik * Otherwise it's safe to assume that the ref offset == the 3201f7ba2d37SJosef Bacik * owner of this block, so we can use that when calling 3202f7ba2d37SJosef Bacik * read_tree_block. 3203f7ba2d37SJosef Bacik */ 3204f7ba2d37SJosef Bacik if (btrfs_extent_refs(eb, ei) == 1 && 3205f7ba2d37SJosef Bacik !(btrfs_extent_flags(eb, ei) & 3206f7ba2d37SJosef Bacik BTRFS_BLOCK_FLAG_FULL_BACKREF) && 3207f7ba2d37SJosef Bacik ptr < end) { 3208f7ba2d37SJosef Bacik struct btrfs_extent_inline_ref *iref; 3209f7ba2d37SJosef Bacik int type; 3210f7ba2d37SJosef Bacik 3211f7ba2d37SJosef Bacik iref = (struct btrfs_extent_inline_ref *)ptr; 3212f7ba2d37SJosef Bacik type = btrfs_get_extent_inline_ref_type(eb, iref, 3213f7ba2d37SJosef Bacik BTRFS_REF_TYPE_BLOCK); 3214f7ba2d37SJosef Bacik if (type == BTRFS_REF_TYPE_INVALID) 3215f7ba2d37SJosef Bacik return -EINVAL; 3216f7ba2d37SJosef Bacik if (type == BTRFS_TREE_BLOCK_REF_KEY) 3217f7ba2d37SJosef Bacik owner = btrfs_extent_inline_ref_offset(eb, iref); 3218f7ba2d37SJosef Bacik } 32196d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3220ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3221ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3222ba3c2b19SNikolay Borisov return -EINVAL; 32233173a18fSJosef Bacik } else { 32245d4f98a2SYan Zheng BUG(); 32255d4f98a2SYan Zheng } 32265d4f98a2SYan Zheng 3227b3b4aa74SDavid Sterba btrfs_release_path(path); 32285d4f98a2SYan Zheng 32295d4f98a2SYan Zheng BUG_ON(level == -1); 32305d4f98a2SYan Zheng 32315d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 32325d4f98a2SYan Zheng if (!block) 32335d4f98a2SYan Zheng return -ENOMEM; 32345d4f98a2SYan Zheng 32355d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3236da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 32375d4f98a2SYan Zheng block->key.offset = generation; 32385d4f98a2SYan Zheng block->level = level; 32395d4f98a2SYan Zheng block->key_ready = 0; 3240f7ba2d37SJosef Bacik block->owner = owner; 32415d4f98a2SYan Zheng 3242e9a28dc5SQu Wenruo rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node); 324343c04fb1SJeff Mahoney if (rb_node) 3244982c92cbSQu Wenruo btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr, 3245982c92cbSQu Wenruo -EEXIST); 32465d4f98a2SYan Zheng 32475d4f98a2SYan Zheng return 0; 32485d4f98a2SYan Zheng } 32495d4f98a2SYan Zheng 32505d4f98a2SYan Zheng /* 32515d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 32525d4f98a2SYan Zheng */ 32535d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 32545d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 32555d4f98a2SYan Zheng struct rb_root *blocks) 32565d4f98a2SYan Zheng { 32570b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 32585d4f98a2SYan Zheng struct btrfs_path *path; 32595d4f98a2SYan Zheng struct btrfs_key key; 32605d4f98a2SYan Zheng int ret; 32610b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 32625d4f98a2SYan Zheng 32637476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 32645d4f98a2SYan Zheng return 0; 32655d4f98a2SYan Zheng 3266e9a28dc5SQu Wenruo if (rb_simple_search(blocks, bytenr)) 32675d4f98a2SYan Zheng return 0; 32685d4f98a2SYan Zheng 32695d4f98a2SYan Zheng path = btrfs_alloc_path(); 32705d4f98a2SYan Zheng if (!path) 32715d4f98a2SYan Zheng return -ENOMEM; 3272aee68ee5SJosef Bacik again: 32735d4f98a2SYan Zheng key.objectid = bytenr; 3274aee68ee5SJosef Bacik if (skinny) { 3275aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3276aee68ee5SJosef Bacik key.offset = (u64)-1; 3277aee68ee5SJosef Bacik } else { 32785d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 32795d4f98a2SYan Zheng key.offset = blocksize; 3280aee68ee5SJosef Bacik } 32815d4f98a2SYan Zheng 32825d4f98a2SYan Zheng path->search_commit_root = 1; 32835d4f98a2SYan Zheng path->skip_locking = 1; 32845d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 32855d4f98a2SYan Zheng if (ret < 0) 32865d4f98a2SYan Zheng goto out; 32875d4f98a2SYan Zheng 3288aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3289aee68ee5SJosef Bacik if (path->slots[0]) { 3290aee68ee5SJosef Bacik path->slots[0]--; 3291aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3292aee68ee5SJosef Bacik path->slots[0]); 32933173a18fSJosef Bacik if (key.objectid == bytenr && 3294aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3295aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3296aee68ee5SJosef Bacik key.offset == blocksize))) 32973173a18fSJosef Bacik ret = 0; 32983173a18fSJosef Bacik } 3299aee68ee5SJosef Bacik 3300aee68ee5SJosef Bacik if (ret) { 3301aee68ee5SJosef Bacik skinny = false; 3302aee68ee5SJosef Bacik btrfs_release_path(path); 3303aee68ee5SJosef Bacik goto again; 3304aee68ee5SJosef Bacik } 3305aee68ee5SJosef Bacik } 3306cdccee99SLiu Bo if (ret) { 3307cdccee99SLiu Bo ASSERT(ret == 1); 3308cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3309cdccee99SLiu Bo btrfs_err(fs_info, 3310cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3311cdccee99SLiu Bo bytenr); 3312cdccee99SLiu Bo WARN_ON(1); 3313cdccee99SLiu Bo ret = -EINVAL; 3314cdccee99SLiu Bo goto out; 3315cdccee99SLiu Bo } 33163173a18fSJosef Bacik 33175d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 33185d4f98a2SYan Zheng out: 33195d4f98a2SYan Zheng btrfs_free_path(path); 33205d4f98a2SYan Zheng return ret; 33215d4f98a2SYan Zheng } 33225d4f98a2SYan Zheng 33230af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 332432da5386SDavid Sterba struct btrfs_block_group *block_group, 33251bbc621eSChris Mason struct inode *inode, 33261bbc621eSChris Mason u64 ino) 33270af3d00bSJosef Bacik { 33280af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 33290af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 33300af3d00bSJosef Bacik int ret = 0; 33310af3d00bSJosef Bacik 33320af3d00bSJosef Bacik if (inode) 33330af3d00bSJosef Bacik goto truncate; 33340af3d00bSJosef Bacik 33350202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, ino, root); 33362e19f1f9SAl Viro if (IS_ERR(inode)) 33370af3d00bSJosef Bacik return -ENOENT; 33380af3d00bSJosef Bacik 33390af3d00bSJosef Bacik truncate: 33402ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 33417b61cd92SMiao Xie &fs_info->global_block_rsv); 33427b61cd92SMiao Xie if (ret) 33437b61cd92SMiao Xie goto out; 33447b61cd92SMiao Xie 33457a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 33460af3d00bSJosef Bacik if (IS_ERR(trans)) { 33473612b495STsutomu Itoh ret = PTR_ERR(trans); 33480af3d00bSJosef Bacik goto out; 33490af3d00bSJosef Bacik } 33500af3d00bSJosef Bacik 335177ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 33520af3d00bSJosef Bacik 33533a45bb20SJeff Mahoney btrfs_end_transaction(trans); 33542ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 33550af3d00bSJosef Bacik out: 33560af3d00bSJosef Bacik iput(inode); 33570af3d00bSJosef Bacik return ret; 33580af3d00bSJosef Bacik } 33590af3d00bSJosef Bacik 33605d4f98a2SYan Zheng /* 336119b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 336219b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 33635d4f98a2SYan Zheng */ 336419b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 336519b546d7SQu Wenruo struct btrfs_block_group *block_group, 336619b546d7SQu Wenruo u64 data_bytenr) 33675d4f98a2SYan Zheng { 336819b546d7SQu Wenruo u64 space_cache_ino; 336919b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 33705d4f98a2SYan Zheng struct btrfs_key key; 337119b546d7SQu Wenruo bool found = false; 337219b546d7SQu Wenruo int i; 33735d4f98a2SYan Zheng int ret; 33745d4f98a2SYan Zheng 337519b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 337619b546d7SQu Wenruo return 0; 33775d4f98a2SYan Zheng 337819b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 337950e31ef4SQu Wenruo u8 type; 338050e31ef4SQu Wenruo 338119b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 338219b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 338319b546d7SQu Wenruo continue; 338419b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 338550e31ef4SQu Wenruo type = btrfs_file_extent_type(leaf, ei); 338650e31ef4SQu Wenruo 338750e31ef4SQu Wenruo if ((type == BTRFS_FILE_EXTENT_REG || 338850e31ef4SQu Wenruo type == BTRFS_FILE_EXTENT_PREALLOC) && 338919b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 339019b546d7SQu Wenruo found = true; 339119b546d7SQu Wenruo space_cache_ino = key.objectid; 339219b546d7SQu Wenruo break; 339319b546d7SQu Wenruo } 339419b546d7SQu Wenruo } 339519b546d7SQu Wenruo if (!found) 339619b546d7SQu Wenruo return -ENOENT; 339719b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 339819b546d7SQu Wenruo space_cache_ino); 33990af3d00bSJosef Bacik return ret; 34005d4f98a2SYan Zheng } 34015d4f98a2SYan Zheng 34025d4f98a2SYan Zheng /* 34032c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 34045d4f98a2SYan Zheng */ 34055d4f98a2SYan Zheng static noinline_for_stack 34065d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 34075d4f98a2SYan Zheng struct btrfs_key *extent_key, 34085d4f98a2SYan Zheng struct btrfs_path *path, 34095d4f98a2SYan Zheng struct rb_root *blocks) 34105d4f98a2SYan Zheng { 3411a2c8d27eSFilipe Manana struct btrfs_backref_walk_ctx ctx = { 0 }; 341219b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 341319b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 3414a2c8d27eSFilipe Manana const u32 blocksize = rc->extent_root->fs_info->nodesize; 3415647f63bdSFilipe David Borba Manana int ret = 0; 34165d4f98a2SYan Zheng 3417b3b4aa74SDavid Sterba btrfs_release_path(path); 3418a2c8d27eSFilipe Manana 3419a2c8d27eSFilipe Manana ctx.bytenr = extent_key->objectid; 3420a2c8d27eSFilipe Manana ctx.ignore_extent_item_pos = true; 3421a2c8d27eSFilipe Manana ctx.fs_info = rc->extent_root->fs_info; 3422a2c8d27eSFilipe Manana 3423a2c8d27eSFilipe Manana ret = btrfs_find_all_leafs(&ctx); 342419b546d7SQu Wenruo if (ret < 0) 342519b546d7SQu Wenruo return ret; 342619b546d7SQu Wenruo 342719b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 3428a2c8d27eSFilipe Manana while ((ref_node = ulist_next(ctx.refs, &leaf_uiter))) { 342919b546d7SQu Wenruo struct extent_buffer *eb; 343019b546d7SQu Wenruo 3431a2c8d27eSFilipe Manana eb = read_tree_block(ctx.fs_info, ref_node->val, 0, 0, 0, NULL); 343219b546d7SQu Wenruo if (IS_ERR(eb)) { 343319b546d7SQu Wenruo ret = PTR_ERR(eb); 343419b546d7SQu Wenruo break; 343519b546d7SQu Wenruo } 343619b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 343719b546d7SQu Wenruo extent_key->objectid); 343819b546d7SQu Wenruo free_extent_buffer(eb); 343919b546d7SQu Wenruo if (ret < 0) 344019b546d7SQu Wenruo break; 344119b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 344219b546d7SQu Wenruo if (ret < 0) 344319b546d7SQu Wenruo break; 344419b546d7SQu Wenruo } 344519b546d7SQu Wenruo if (ret < 0) 34465d4f98a2SYan Zheng free_block_list(blocks); 3447a2c8d27eSFilipe Manana ulist_free(ctx.refs); 344819b546d7SQu Wenruo return ret; 34495d4f98a2SYan Zheng } 34505d4f98a2SYan Zheng 34515d4f98a2SYan Zheng /* 34522c016dc2SLiu Bo * helper to find next unprocessed extent 34535d4f98a2SYan Zheng */ 34545d4f98a2SYan Zheng static noinline_for_stack 3455147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 34563fd0a558SYan, Zheng struct btrfs_key *extent_key) 34575d4f98a2SYan Zheng { 34580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34595d4f98a2SYan Zheng struct btrfs_key key; 34605d4f98a2SYan Zheng struct extent_buffer *leaf; 34615d4f98a2SYan Zheng u64 start, end, last; 34625d4f98a2SYan Zheng int ret; 34635d4f98a2SYan Zheng 3464b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 34655d4f98a2SYan Zheng while (1) { 34665d4f98a2SYan Zheng cond_resched(); 34675d4f98a2SYan Zheng if (rc->search_start >= last) { 34685d4f98a2SYan Zheng ret = 1; 34695d4f98a2SYan Zheng break; 34705d4f98a2SYan Zheng } 34715d4f98a2SYan Zheng 34725d4f98a2SYan Zheng key.objectid = rc->search_start; 34735d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34745d4f98a2SYan Zheng key.offset = 0; 34755d4f98a2SYan Zheng 34765d4f98a2SYan Zheng path->search_commit_root = 1; 34775d4f98a2SYan Zheng path->skip_locking = 1; 34785d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 34795d4f98a2SYan Zheng 0, 0); 34805d4f98a2SYan Zheng if (ret < 0) 34815d4f98a2SYan Zheng break; 34825d4f98a2SYan Zheng next: 34835d4f98a2SYan Zheng leaf = path->nodes[0]; 34845d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 34855d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 34865d4f98a2SYan Zheng if (ret != 0) 34875d4f98a2SYan Zheng break; 34885d4f98a2SYan Zheng leaf = path->nodes[0]; 34895d4f98a2SYan Zheng } 34905d4f98a2SYan Zheng 34915d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 34925d4f98a2SYan Zheng if (key.objectid >= last) { 34935d4f98a2SYan Zheng ret = 1; 34945d4f98a2SYan Zheng break; 34955d4f98a2SYan Zheng } 34965d4f98a2SYan Zheng 34973173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 34983173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 34993173a18fSJosef Bacik path->slots[0]++; 35003173a18fSJosef Bacik goto next; 35013173a18fSJosef Bacik } 35023173a18fSJosef Bacik 35033173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 35045d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 35055d4f98a2SYan Zheng path->slots[0]++; 35065d4f98a2SYan Zheng goto next; 35075d4f98a2SYan Zheng } 35085d4f98a2SYan Zheng 35093173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 35100b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 35113173a18fSJosef Bacik rc->search_start) { 35123173a18fSJosef Bacik path->slots[0]++; 35133173a18fSJosef Bacik goto next; 35143173a18fSJosef Bacik } 35153173a18fSJosef Bacik 35165d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 35175d4f98a2SYan Zheng key.objectid, &start, &end, 3518e6138876SJosef Bacik EXTENT_DIRTY, NULL); 35195d4f98a2SYan Zheng 35205d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3521b3b4aa74SDavid Sterba btrfs_release_path(path); 35225d4f98a2SYan Zheng rc->search_start = end + 1; 35235d4f98a2SYan Zheng } else { 35243173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 35255d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 35263173a18fSJosef Bacik else 35273173a18fSJosef Bacik rc->search_start = key.objectid + 35280b246afaSJeff Mahoney fs_info->nodesize; 35293fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 35305d4f98a2SYan Zheng return 0; 35315d4f98a2SYan Zheng } 35325d4f98a2SYan Zheng } 3533b3b4aa74SDavid Sterba btrfs_release_path(path); 35345d4f98a2SYan Zheng return ret; 35355d4f98a2SYan Zheng } 35365d4f98a2SYan Zheng 35375d4f98a2SYan Zheng static void set_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 = rc; 35437585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35445d4f98a2SYan Zheng } 35455d4f98a2SYan Zheng 35465d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 35475d4f98a2SYan Zheng { 35485d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35497585717fSChris Mason 35507585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35515d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 35527585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35535d4f98a2SYan Zheng } 35545d4f98a2SYan Zheng 35553fd0a558SYan, Zheng static noinline_for_stack 35563fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 35573fd0a558SYan, Zheng { 35583fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3559ac2fabacSJosef Bacik int ret; 35603fd0a558SYan, Zheng 35612ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 356266d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 35633fd0a558SYan, Zheng if (!rc->block_rsv) 35643fd0a558SYan, Zheng return -ENOMEM; 35653fd0a558SYan, Zheng 35663fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3567b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 35683fd0a558SYan, Zheng rc->extents_found = 0; 35693fd0a558SYan, Zheng rc->nodes_relocated = 0; 35703fd0a558SYan, Zheng rc->merging_rsv_size = 0; 35710647bf56SWang Shilong rc->reserved_bytes = 0; 3572da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 35730647bf56SWang Shilong RELOCATION_RESERVED_NODES; 35749270501cSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root->fs_info, 3575ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3576ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3577ac2fabacSJosef Bacik if (ret) 3578ac2fabacSJosef Bacik return ret; 35793fd0a558SYan, Zheng 35803fd0a558SYan, Zheng rc->create_reloc_tree = 1; 35813fd0a558SYan, Zheng set_reloc_control(rc); 35823fd0a558SYan, Zheng 35837a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 358428818947SLiu Bo if (IS_ERR(trans)) { 358528818947SLiu Bo unset_reloc_control(rc); 358628818947SLiu Bo /* 358728818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 358828818947SLiu Bo * cleanup. And callers are responsible to free the above 358928818947SLiu Bo * block rsv. 359028818947SLiu Bo */ 359128818947SLiu Bo return PTR_ERR(trans); 359228818947SLiu Bo } 359385f02d6cSZixuan Fu 359485f02d6cSZixuan Fu ret = btrfs_commit_transaction(trans); 359585f02d6cSZixuan Fu if (ret) 359685f02d6cSZixuan Fu unset_reloc_control(rc); 359785f02d6cSZixuan Fu 359885f02d6cSZixuan Fu return ret; 35993fd0a558SYan, Zheng } 360076dda93cSYan, Zheng 36015d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 36025d4f98a2SYan Zheng { 36032ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 36045d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 36055d4f98a2SYan Zheng struct btrfs_key key; 36065d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 36075d4f98a2SYan Zheng struct btrfs_path *path; 36085d4f98a2SYan Zheng struct btrfs_extent_item *ei; 36095d4f98a2SYan Zheng u64 flags; 36105d4f98a2SYan Zheng int ret; 36115d4f98a2SYan Zheng int err = 0; 3612c87f08caSChris Mason int progress = 0; 36135d4f98a2SYan Zheng 36145d4f98a2SYan Zheng path = btrfs_alloc_path(); 36153fd0a558SYan, Zheng if (!path) 36165d4f98a2SYan Zheng return -ENOMEM; 3617e4058b54SDavid Sterba path->reada = READA_FORWARD; 36183fd0a558SYan, Zheng 36193fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 36203fd0a558SYan, Zheng if (ret) { 36213fd0a558SYan, Zheng err = ret; 36223fd0a558SYan, Zheng goto out_free; 36232423fdfbSJiri Slaby } 36245d4f98a2SYan Zheng 36255d4f98a2SYan Zheng while (1) { 36260647bf56SWang Shilong rc->reserved_bytes = 0; 36279270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 36289270501cSJosef Bacik rc->block_rsv->size, 36290647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 36300647bf56SWang Shilong if (ret) { 36310647bf56SWang Shilong err = ret; 36320647bf56SWang Shilong break; 36330647bf56SWang Shilong } 3634c87f08caSChris Mason progress++; 3635a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 36360f788c58SLiu Bo if (IS_ERR(trans)) { 36370f788c58SLiu Bo err = PTR_ERR(trans); 36380f788c58SLiu Bo trans = NULL; 36390f788c58SLiu Bo break; 36400f788c58SLiu Bo } 3641c87f08caSChris Mason restart: 36423fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 36433a45bb20SJeff Mahoney btrfs_end_transaction(trans); 364442a657f5SPan Bian trans = NULL; 36453fd0a558SYan, Zheng continue; 36463fd0a558SYan, Zheng } 36473fd0a558SYan, Zheng 3648147d256eSZhaolei ret = find_next_extent(rc, path, &key); 36495d4f98a2SYan Zheng if (ret < 0) 36505d4f98a2SYan Zheng err = ret; 36515d4f98a2SYan Zheng if (ret != 0) 36525d4f98a2SYan Zheng break; 36535d4f98a2SYan Zheng 36545d4f98a2SYan Zheng rc->extents_found++; 36555d4f98a2SYan Zheng 36565d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 36575d4f98a2SYan Zheng struct btrfs_extent_item); 36585d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 36595d4f98a2SYan Zheng 36605d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 36615d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 36625d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 36635d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36645d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 36655d4f98a2SYan Zheng } else { 3666b3b4aa74SDavid Sterba btrfs_release_path(path); 36675d4f98a2SYan Zheng ret = 0; 36685d4f98a2SYan Zheng } 36695d4f98a2SYan Zheng if (ret < 0) { 36703fd0a558SYan, Zheng err = ret; 36715d4f98a2SYan Zheng break; 36725d4f98a2SYan Zheng } 36735d4f98a2SYan Zheng 36745d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 36755d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 36765d4f98a2SYan Zheng if (ret < 0) { 36773fd0a558SYan, Zheng if (ret != -EAGAIN) { 36785d4f98a2SYan Zheng err = ret; 36795d4f98a2SYan Zheng break; 36805d4f98a2SYan Zheng } 36813fd0a558SYan, Zheng rc->extents_found--; 36823fd0a558SYan, Zheng rc->search_start = key.objectid; 36833fd0a558SYan, Zheng } 36845d4f98a2SYan Zheng } 36855d4f98a2SYan Zheng 36863a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 36872ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36883fd0a558SYan, Zheng trans = NULL; 36895d4f98a2SYan Zheng 36905d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 36915d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 36925d4f98a2SYan Zheng rc->found_file_extent = 1; 36930257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 36943fd0a558SYan, Zheng &key, &rc->cluster); 36955d4f98a2SYan Zheng if (ret < 0) { 36965d4f98a2SYan Zheng err = ret; 36975d4f98a2SYan Zheng break; 36985d4f98a2SYan Zheng } 36995d4f98a2SYan Zheng } 3700f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 3701f31ea088SQu Wenruo err = -ECANCELED; 3702f31ea088SQu Wenruo break; 3703f31ea088SQu Wenruo } 37045d4f98a2SYan Zheng } 3705c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 370643a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 37079689457bSShilong Wang if (ret == 1) { 3708c87f08caSChris Mason err = 0; 3709c87f08caSChris Mason progress = 0; 3710c87f08caSChris Mason goto restart; 3711c87f08caSChris Mason } 3712c87f08caSChris Mason } 37133fd0a558SYan, Zheng 3714b3b4aa74SDavid Sterba btrfs_release_path(path); 371591166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 37165d4f98a2SYan Zheng 37175d4f98a2SYan Zheng if (trans) { 37183a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 37192ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 37205d4f98a2SYan Zheng } 37215d4f98a2SYan Zheng 37220257bb82SYan, Zheng if (!err) { 37233fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 37243fd0a558SYan, Zheng &rc->cluster); 37250257bb82SYan, Zheng if (ret < 0) 37260257bb82SYan, Zheng err = ret; 37270257bb82SYan, Zheng } 37280257bb82SYan, Zheng 37293fd0a558SYan, Zheng rc->create_reloc_tree = 0; 37303fd0a558SYan, Zheng set_reloc_control(rc); 37310257bb82SYan, Zheng 373213fe1bdbSQu Wenruo btrfs_backref_release_cache(&rc->backref_cache); 373363f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37345d4f98a2SYan Zheng 37357f913c7cSQu Wenruo /* 37367f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 37377f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 37387f913c7cSQu Wenruo * 37397f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 37407f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 37417f913c7cSQu Wenruo * merge_reloc_roots() 37427f913c7cSQu Wenruo */ 37433fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 37445d4f98a2SYan Zheng 37455d4f98a2SYan Zheng merge_reloc_roots(rc); 37465d4f98a2SYan Zheng 37473fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 37485d4f98a2SYan Zheng unset_reloc_control(rc); 374963f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37505d4f98a2SYan Zheng 37515d4f98a2SYan Zheng /* get rid of pinned extents */ 37527a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 375362b99540SQu Wenruo if (IS_ERR(trans)) { 37543612b495STsutomu Itoh err = PTR_ERR(trans); 375562b99540SQu Wenruo goto out_free; 375662b99540SQu Wenruo } 3757fb686c68SJosef Bacik ret = btrfs_commit_transaction(trans); 3758fb686c68SJosef Bacik if (ret && !err) 3759fb686c68SJosef Bacik err = ret; 37606217b0faSJosef Bacik out_free: 3761d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 3762d2311e69SQu Wenruo if (ret < 0 && !err) 3763d2311e69SQu Wenruo err = ret; 37642ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 37653fd0a558SYan, Zheng btrfs_free_path(path); 37665d4f98a2SYan Zheng return err; 37675d4f98a2SYan Zheng } 37685d4f98a2SYan Zheng 37695d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 37700257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 37715d4f98a2SYan Zheng { 37725d4f98a2SYan Zheng struct btrfs_path *path; 37735d4f98a2SYan Zheng struct btrfs_inode_item *item; 37745d4f98a2SYan Zheng struct extent_buffer *leaf; 37755d4f98a2SYan Zheng int ret; 37765d4f98a2SYan Zheng 37775d4f98a2SYan Zheng path = btrfs_alloc_path(); 37785d4f98a2SYan Zheng if (!path) 37795d4f98a2SYan Zheng return -ENOMEM; 37805d4f98a2SYan Zheng 37815d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 37825d4f98a2SYan Zheng if (ret) 37835d4f98a2SYan Zheng goto out; 37845d4f98a2SYan Zheng 37855d4f98a2SYan Zheng leaf = path->nodes[0]; 37865d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 3787b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 37885d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 37890257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 37905d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 3791960a3166SJohannes Thumshirn btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 3792960a3166SJohannes Thumshirn BTRFS_INODE_PREALLOC); 37935d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 37945d4f98a2SYan Zheng out: 37955d4f98a2SYan Zheng btrfs_free_path(path); 37965d4f98a2SYan Zheng return ret; 37975d4f98a2SYan Zheng } 37985d4f98a2SYan Zheng 3799790c1b8cSJosef Bacik static void delete_orphan_inode(struct btrfs_trans_handle *trans, 3800790c1b8cSJosef Bacik struct btrfs_root *root, u64 objectid) 3801790c1b8cSJosef Bacik { 3802790c1b8cSJosef Bacik struct btrfs_path *path; 3803790c1b8cSJosef Bacik struct btrfs_key key; 3804790c1b8cSJosef Bacik int ret = 0; 3805790c1b8cSJosef Bacik 3806790c1b8cSJosef Bacik path = btrfs_alloc_path(); 3807790c1b8cSJosef Bacik if (!path) { 3808790c1b8cSJosef Bacik ret = -ENOMEM; 3809790c1b8cSJosef Bacik goto out; 3810790c1b8cSJosef Bacik } 3811790c1b8cSJosef Bacik 3812790c1b8cSJosef Bacik key.objectid = objectid; 3813790c1b8cSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 3814790c1b8cSJosef Bacik key.offset = 0; 3815790c1b8cSJosef Bacik ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3816790c1b8cSJosef Bacik if (ret) { 3817790c1b8cSJosef Bacik if (ret > 0) 3818790c1b8cSJosef Bacik ret = -ENOENT; 3819790c1b8cSJosef Bacik goto out; 3820790c1b8cSJosef Bacik } 3821790c1b8cSJosef Bacik ret = btrfs_del_item(trans, root, path); 3822790c1b8cSJosef Bacik out: 3823790c1b8cSJosef Bacik if (ret) 3824790c1b8cSJosef Bacik btrfs_abort_transaction(trans, ret); 3825790c1b8cSJosef Bacik btrfs_free_path(path); 3826790c1b8cSJosef Bacik } 3827790c1b8cSJosef Bacik 38285d4f98a2SYan Zheng /* 38295d4f98a2SYan Zheng * helper to create inode for data relocation. 38305d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 38315d4f98a2SYan Zheng */ 38323fd0a558SYan, Zheng static noinline_for_stack 38333fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 383432da5386SDavid Sterba struct btrfs_block_group *group) 38355d4f98a2SYan Zheng { 38365d4f98a2SYan Zheng struct inode *inode = NULL; 38375d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 38385d4f98a2SYan Zheng struct btrfs_root *root; 38394624900dSZhaolei u64 objectid; 38405d4f98a2SYan Zheng int err = 0; 38415d4f98a2SYan Zheng 3842aeb935a4SQu Wenruo root = btrfs_grab_root(fs_info->data_reloc_root); 3843a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 384476deacf0SJosef Bacik if (IS_ERR(trans)) { 384500246528SJosef Bacik btrfs_put_root(root); 38463fd0a558SYan, Zheng return ERR_CAST(trans); 384776deacf0SJosef Bacik } 38485d4f98a2SYan Zheng 3849543068a2SNikolay Borisov err = btrfs_get_free_objectid(root, &objectid); 38505d4f98a2SYan Zheng if (err) 38515d4f98a2SYan Zheng goto out; 38525d4f98a2SYan Zheng 38530257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 3854790c1b8cSJosef Bacik if (err) 3855790c1b8cSJosef Bacik goto out; 38565d4f98a2SYan Zheng 38570202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, objectid, root); 3858790c1b8cSJosef Bacik if (IS_ERR(inode)) { 3859790c1b8cSJosef Bacik delete_orphan_inode(trans, root, objectid); 3860790c1b8cSJosef Bacik err = PTR_ERR(inode); 3861790c1b8cSJosef Bacik inode = NULL; 3862790c1b8cSJosef Bacik goto out; 3863790c1b8cSJosef Bacik } 3864b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 38655d4f98a2SYan Zheng 386673f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 38675d4f98a2SYan Zheng out: 386800246528SJosef Bacik btrfs_put_root(root); 38693a45bb20SJeff Mahoney btrfs_end_transaction(trans); 38702ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 38715d4f98a2SYan Zheng if (err) { 38725d4f98a2SYan Zheng iput(inode); 38735d4f98a2SYan Zheng inode = ERR_PTR(err); 38745d4f98a2SYan Zheng } 38755d4f98a2SYan Zheng return inode; 38765d4f98a2SYan Zheng } 38775d4f98a2SYan Zheng 3878907d2710SDavid Sterba /* 3879907d2710SDavid Sterba * Mark start of chunk relocation that is cancellable. Check if the cancellation 3880907d2710SDavid Sterba * has been requested meanwhile and don't start in that case. 3881907d2710SDavid Sterba * 3882907d2710SDavid Sterba * Return: 3883907d2710SDavid Sterba * 0 success 3884907d2710SDavid Sterba * -EINPROGRESS operation is already in progress, that's probably a bug 3885907d2710SDavid Sterba * -ECANCELED cancellation request was set before the operation started 3886907d2710SDavid Sterba */ 3887907d2710SDavid Sterba static int reloc_chunk_start(struct btrfs_fs_info *fs_info) 3888907d2710SDavid Sterba { 3889907d2710SDavid Sterba if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) { 3890907d2710SDavid Sterba /* This should not happen */ 3891907d2710SDavid Sterba btrfs_err(fs_info, "reloc already running, cannot start"); 3892907d2710SDavid Sterba return -EINPROGRESS; 3893907d2710SDavid Sterba } 3894907d2710SDavid Sterba 3895907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) { 3896907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled on start"); 3897907d2710SDavid Sterba /* 3898907d2710SDavid Sterba * On cancel, clear all requests but let the caller mark 3899907d2710SDavid Sterba * the end after cleanup operations. 3900907d2710SDavid Sterba */ 3901907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3902907d2710SDavid Sterba return -ECANCELED; 3903907d2710SDavid Sterba } 3904907d2710SDavid Sterba return 0; 3905907d2710SDavid Sterba } 3906907d2710SDavid Sterba 3907907d2710SDavid Sterba /* 3908907d2710SDavid Sterba * Mark end of chunk relocation that is cancellable and wake any waiters. 3909907d2710SDavid Sterba */ 3910907d2710SDavid Sterba static void reloc_chunk_end(struct btrfs_fs_info *fs_info) 3911907d2710SDavid Sterba { 3912907d2710SDavid Sterba /* Requested after start, clear bit first so any waiters can continue */ 3913907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) 3914907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled during operation"); 3915907d2710SDavid Sterba clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags); 3916907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3917907d2710SDavid Sterba } 3918907d2710SDavid Sterba 3919c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 39203fd0a558SYan, Zheng { 39213fd0a558SYan, Zheng struct reloc_control *rc; 39223fd0a558SYan, Zheng 39233fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 39243fd0a558SYan, Zheng if (!rc) 39253fd0a558SYan, Zheng return NULL; 39263fd0a558SYan, Zheng 39273fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 3928d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 3929584fb121SQu Wenruo btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1); 39303fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 393143eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 393243eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 39333fd0a558SYan, Zheng return rc; 39343fd0a558SYan, Zheng } 39353fd0a558SYan, Zheng 39361a0afa0eSJosef Bacik static void free_reloc_control(struct reloc_control *rc) 39371a0afa0eSJosef Bacik { 39381a0afa0eSJosef Bacik struct mapping_node *node, *tmp; 39391a0afa0eSJosef Bacik 39401a0afa0eSJosef Bacik free_reloc_roots(&rc->reloc_roots); 39411a0afa0eSJosef Bacik rbtree_postorder_for_each_entry_safe(node, tmp, 39421a0afa0eSJosef Bacik &rc->reloc_root_tree.rb_root, rb_node) 39431a0afa0eSJosef Bacik kfree(node); 39441a0afa0eSJosef Bacik 39451a0afa0eSJosef Bacik kfree(rc); 39461a0afa0eSJosef Bacik } 39471a0afa0eSJosef Bacik 39485d4f98a2SYan Zheng /* 3949ebce0e01SAdam Borowski * Print the block group being relocated 3950ebce0e01SAdam Borowski */ 3951ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 395232da5386SDavid Sterba struct btrfs_block_group *block_group) 3953ebce0e01SAdam Borowski { 3954f89e09cfSAnand Jain char buf[128] = {'\0'}; 3955ebce0e01SAdam Borowski 3956f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 3957ebce0e01SAdam Borowski 3958ebce0e01SAdam Borowski btrfs_info(fs_info, 3959ebce0e01SAdam Borowski "relocating block group %llu flags %s", 3960b3470b5dSDavid Sterba block_group->start, buf); 3961ebce0e01SAdam Borowski } 3962ebce0e01SAdam Borowski 3963430640e3SQu Wenruo static const char *stage_to_string(int stage) 3964430640e3SQu Wenruo { 3965430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 3966430640e3SQu Wenruo return "move data extents"; 3967430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 3968430640e3SQu Wenruo return "update data pointers"; 3969430640e3SQu Wenruo return "unknown"; 3970430640e3SQu Wenruo } 3971430640e3SQu Wenruo 3972ebce0e01SAdam Borowski /* 39735d4f98a2SYan Zheng * function to relocate all extents in a block group. 39745d4f98a2SYan Zheng */ 39756bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 39765d4f98a2SYan Zheng { 397732da5386SDavid Sterba struct btrfs_block_group *bg; 397829cbcf40SJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start); 39795d4f98a2SYan Zheng struct reloc_control *rc; 39800af3d00bSJosef Bacik struct inode *inode; 39810af3d00bSJosef Bacik struct btrfs_path *path; 39825d4f98a2SYan Zheng int ret; 3983f0486c68SYan, Zheng int rw = 0; 39845d4f98a2SYan Zheng int err = 0; 39855d4f98a2SYan Zheng 3986b4be6aefSJosef Bacik /* 3987b4be6aefSJosef Bacik * This only gets set if we had a half-deleted snapshot on mount. We 3988b4be6aefSJosef Bacik * cannot allow relocation to start while we're still trying to clean up 3989b4be6aefSJosef Bacik * these pending deletions. 3990b4be6aefSJosef Bacik */ 3991b4be6aefSJosef Bacik ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE); 3992b4be6aefSJosef Bacik if (ret) 3993b4be6aefSJosef Bacik return ret; 3994b4be6aefSJosef Bacik 3995b4be6aefSJosef Bacik /* We may have been woken up by close_ctree, so bail if we're closing. */ 3996b4be6aefSJosef Bacik if (btrfs_fs_closing(fs_info)) 3997b4be6aefSJosef Bacik return -EINTR; 3998b4be6aefSJosef Bacik 3999eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4000eede2bf3SOmar Sandoval if (!bg) 4001eede2bf3SOmar Sandoval return -ENOENT; 4002eede2bf3SOmar Sandoval 40030320b353SNaohiro Aota /* 40040320b353SNaohiro Aota * Relocation of a data block group creates ordered extents. Without 40050320b353SNaohiro Aota * sb_start_write(), we can freeze the filesystem while unfinished 40060320b353SNaohiro Aota * ordered extents are left. Such ordered extents can cause a deadlock 40070320b353SNaohiro Aota * e.g. when syncfs() is waiting for their completion but they can't 40080320b353SNaohiro Aota * finish because they block when joining a transaction, due to the 40090320b353SNaohiro Aota * fact that the freeze locks are being held in write mode. 40100320b353SNaohiro Aota */ 40110320b353SNaohiro Aota if (bg->flags & BTRFS_BLOCK_GROUP_DATA) 40120320b353SNaohiro Aota ASSERT(sb_write_started(fs_info->sb)); 40130320b353SNaohiro Aota 4014eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4015eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4016eede2bf3SOmar Sandoval return -ETXTBSY; 4017eede2bf3SOmar Sandoval } 4018eede2bf3SOmar Sandoval 4019c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4020eede2bf3SOmar Sandoval if (!rc) { 4021eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 40225d4f98a2SYan Zheng return -ENOMEM; 4023eede2bf3SOmar Sandoval } 40245d4f98a2SYan Zheng 4025907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4026907d2710SDavid Sterba if (ret < 0) { 4027907d2710SDavid Sterba err = ret; 4028907d2710SDavid Sterba goto out_put_bg; 4029907d2710SDavid Sterba } 4030907d2710SDavid Sterba 4031f0486c68SYan, Zheng rc->extent_root = extent_root; 4032eede2bf3SOmar Sandoval rc->block_group = bg; 40335d4f98a2SYan Zheng 4034b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4035f0486c68SYan, Zheng if (ret) { 4036f0486c68SYan, Zheng err = ret; 4037f0486c68SYan, Zheng goto out; 4038f0486c68SYan, Zheng } 4039f0486c68SYan, Zheng rw = 1; 4040f0486c68SYan, Zheng 40410af3d00bSJosef Bacik path = btrfs_alloc_path(); 40420af3d00bSJosef Bacik if (!path) { 40430af3d00bSJosef Bacik err = -ENOMEM; 40440af3d00bSJosef Bacik goto out; 40450af3d00bSJosef Bacik } 40460af3d00bSJosef Bacik 40477949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 40480af3d00bSJosef Bacik btrfs_free_path(path); 40490af3d00bSJosef Bacik 40500af3d00bSJosef Bacik if (!IS_ERR(inode)) 40511bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 40520af3d00bSJosef Bacik else 40530af3d00bSJosef Bacik ret = PTR_ERR(inode); 40540af3d00bSJosef Bacik 40550af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 40560af3d00bSJosef Bacik err = ret; 40570af3d00bSJosef Bacik goto out; 40580af3d00bSJosef Bacik } 40590af3d00bSJosef Bacik 40605d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 40615d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 40625d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 40635d4f98a2SYan Zheng rc->data_inode = NULL; 40645d4f98a2SYan Zheng goto out; 40655d4f98a2SYan Zheng } 40665d4f98a2SYan Zheng 40670b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 40685d4f98a2SYan Zheng 40699cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4070f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 40716374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4072b3470b5dSDavid Sterba rc->block_group->start, 4073b3470b5dSDavid Sterba rc->block_group->length); 40745d4f98a2SYan Zheng 40757ae9bd18SNaohiro Aota ret = btrfs_zone_finish(rc->block_group); 40767ae9bd18SNaohiro Aota WARN_ON(ret && ret != -EAGAIN); 40777ae9bd18SNaohiro Aota 40785d4f98a2SYan Zheng while (1) { 4079430640e3SQu Wenruo int finishes_stage; 4080430640e3SQu Wenruo 408176dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 40825d4f98a2SYan Zheng ret = relocate_block_group(rc); 408376dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4084ff612ba7SJosef Bacik if (ret < 0) 40855d4f98a2SYan Zheng err = ret; 4086ff612ba7SJosef Bacik 4087430640e3SQu Wenruo finishes_stage = rc->stage; 4088ff612ba7SJosef Bacik /* 4089ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4090ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4091ff612ba7SJosef Bacik * different block group we could end up hitting the 4092ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4093ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4094ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4095ff612ba7SJosef Bacik * out of the loop if we hit an error. 4096ff612ba7SJosef Bacik */ 4097ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4098ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4099ff612ba7SJosef Bacik (u64)-1); 4100ff612ba7SJosef Bacik if (ret) 4101ff612ba7SJosef Bacik err = ret; 4102ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4103ff612ba7SJosef Bacik 0, -1); 4104ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 41055d4f98a2SYan Zheng } 41065d4f98a2SYan Zheng 4107ff612ba7SJosef Bacik if (err < 0) 4108ff612ba7SJosef Bacik goto out; 4109ff612ba7SJosef Bacik 41105d4f98a2SYan Zheng if (rc->extents_found == 0) 41115d4f98a2SYan Zheng break; 41125d4f98a2SYan Zheng 4113430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4114430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 41155d4f98a2SYan Zheng } 41165d4f98a2SYan Zheng 41175d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 41185d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4119bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 41205d4f98a2SYan Zheng out: 4121f0486c68SYan, Zheng if (err && rw) 41222ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 41235d4f98a2SYan Zheng iput(rc->data_inode); 4124907d2710SDavid Sterba out_put_bg: 4125907d2710SDavid Sterba btrfs_put_block_group(bg); 4126907d2710SDavid Sterba reloc_chunk_end(fs_info); 41271a0afa0eSJosef Bacik free_reloc_control(rc); 41285d4f98a2SYan Zheng return err; 41295d4f98a2SYan Zheng } 41305d4f98a2SYan Zheng 413176dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 413276dda93cSYan, Zheng { 41330b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 413476dda93cSYan, Zheng struct btrfs_trans_handle *trans; 413579787eaaSJeff Mahoney int ret, err; 413676dda93cSYan, Zheng 41370b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 413879787eaaSJeff Mahoney if (IS_ERR(trans)) 413979787eaaSJeff Mahoney return PTR_ERR(trans); 414076dda93cSYan, Zheng 414176dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 414276dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 4143c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 414476dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 41450b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 414676dda93cSYan, Zheng &root->root_key, &root->root_item); 414776dda93cSYan, Zheng 41483a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 414979787eaaSJeff Mahoney if (err) 415079787eaaSJeff Mahoney return err; 415179787eaaSJeff Mahoney return ret; 415276dda93cSYan, Zheng } 415376dda93cSYan, Zheng 41545d4f98a2SYan Zheng /* 41555d4f98a2SYan Zheng * recover relocation interrupted by system crash. 41565d4f98a2SYan Zheng * 41575d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 41585d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 41595d4f98a2SYan Zheng */ 41607eefae6bSJosef Bacik int btrfs_recover_relocation(struct btrfs_fs_info *fs_info) 41615d4f98a2SYan Zheng { 41625d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 41635d4f98a2SYan Zheng struct btrfs_key key; 41645d4f98a2SYan Zheng struct btrfs_root *fs_root; 41655d4f98a2SYan Zheng struct btrfs_root *reloc_root; 41665d4f98a2SYan Zheng struct btrfs_path *path; 41675d4f98a2SYan Zheng struct extent_buffer *leaf; 41685d4f98a2SYan Zheng struct reloc_control *rc = NULL; 41695d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41705d4f98a2SYan Zheng int ret; 41715d4f98a2SYan Zheng int err = 0; 41725d4f98a2SYan Zheng 41735d4f98a2SYan Zheng path = btrfs_alloc_path(); 41745d4f98a2SYan Zheng if (!path) 41755d4f98a2SYan Zheng return -ENOMEM; 4176e4058b54SDavid Sterba path->reada = READA_BACK; 41775d4f98a2SYan Zheng 41785d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 41795d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 41805d4f98a2SYan Zheng key.offset = (u64)-1; 41815d4f98a2SYan Zheng 41825d4f98a2SYan Zheng while (1) { 41830b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 41845d4f98a2SYan Zheng path, 0, 0); 41855d4f98a2SYan Zheng if (ret < 0) { 41865d4f98a2SYan Zheng err = ret; 41875d4f98a2SYan Zheng goto out; 41885d4f98a2SYan Zheng } 41895d4f98a2SYan Zheng if (ret > 0) { 41905d4f98a2SYan Zheng if (path->slots[0] == 0) 41915d4f98a2SYan Zheng break; 41925d4f98a2SYan Zheng path->slots[0]--; 41935d4f98a2SYan Zheng } 41945d4f98a2SYan Zheng leaf = path->nodes[0]; 41955d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4196b3b4aa74SDavid Sterba btrfs_release_path(path); 41975d4f98a2SYan Zheng 41985d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 41995d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 42005d4f98a2SYan Zheng break; 42015d4f98a2SYan Zheng 42027eefae6bSJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key); 42035d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 42045d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 42055d4f98a2SYan Zheng goto out; 42065d4f98a2SYan Zheng } 42075d4f98a2SYan Zheng 420892a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 42095d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 42105d4f98a2SYan Zheng 42115d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 4212a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, 4213a820feb5SDavid Sterba reloc_root->root_key.offset, false); 42145d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 421576dda93cSYan, Zheng ret = PTR_ERR(fs_root); 421676dda93cSYan, Zheng if (ret != -ENOENT) { 421776dda93cSYan, Zheng err = ret; 42185d4f98a2SYan Zheng goto out; 42195d4f98a2SYan Zheng } 422079787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 422179787eaaSJeff Mahoney if (ret < 0) { 422279787eaaSJeff Mahoney err = ret; 422379787eaaSJeff Mahoney goto out; 422479787eaaSJeff Mahoney } 4225932fd26dSJosef Bacik } else { 422600246528SJosef Bacik btrfs_put_root(fs_root); 422776dda93cSYan, Zheng } 42285d4f98a2SYan Zheng } 42295d4f98a2SYan Zheng 42305d4f98a2SYan Zheng if (key.offset == 0) 42315d4f98a2SYan Zheng break; 42325d4f98a2SYan Zheng 42335d4f98a2SYan Zheng key.offset--; 42345d4f98a2SYan Zheng } 4235b3b4aa74SDavid Sterba btrfs_release_path(path); 42365d4f98a2SYan Zheng 42375d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 42385d4f98a2SYan Zheng goto out; 42395d4f98a2SYan Zheng 4240c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 42415d4f98a2SYan Zheng if (!rc) { 42425d4f98a2SYan Zheng err = -ENOMEM; 42435d4f98a2SYan Zheng goto out; 42445d4f98a2SYan Zheng } 42455d4f98a2SYan Zheng 4246907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4247907d2710SDavid Sterba if (ret < 0) { 4248907d2710SDavid Sterba err = ret; 4249907d2710SDavid Sterba goto out_end; 4250907d2710SDavid Sterba } 4251907d2710SDavid Sterba 425229cbcf40SJosef Bacik rc->extent_root = btrfs_extent_root(fs_info, 0); 42535d4f98a2SYan Zheng 42545d4f98a2SYan Zheng set_reloc_control(rc); 42555d4f98a2SYan Zheng 42567a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 42573612b495STsutomu Itoh if (IS_ERR(trans)) { 42583612b495STsutomu Itoh err = PTR_ERR(trans); 4259fb2d83eeSJosef Bacik goto out_unset; 42603612b495STsutomu Itoh } 42613fd0a558SYan, Zheng 42623fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 42633fd0a558SYan, Zheng 42645d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 42655d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 42665d4f98a2SYan Zheng struct btrfs_root, root_list); 42675d4f98a2SYan Zheng list_del(&reloc_root->root_list); 42685d4f98a2SYan Zheng 42695d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 42705d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 42715d4f98a2SYan Zheng &rc->reloc_roots); 42725d4f98a2SYan Zheng continue; 42735d4f98a2SYan Zheng } 42745d4f98a2SYan Zheng 4275a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 4276a820feb5SDavid Sterba false); 427779787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 427879787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4279ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42801402d17dSXiyu Yang btrfs_end_transaction(trans); 4281fb2d83eeSJosef Bacik goto out_unset; 428279787eaaSJeff Mahoney } 42835d4f98a2SYan Zheng 4284ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 428557a304cfSJosef Bacik ASSERT(err != -EEXIST); 42863c925863SJosef Bacik if (err) { 42873c925863SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 42883c925863SJosef Bacik btrfs_put_root(fs_root); 42893c925863SJosef Bacik btrfs_end_transaction(trans); 42903c925863SJosef Bacik goto out_unset; 42913c925863SJosef Bacik } 4292f44deb74SJosef Bacik fs_root->reloc_root = btrfs_grab_root(reloc_root); 429300246528SJosef Bacik btrfs_put_root(fs_root); 42945d4f98a2SYan Zheng } 42955d4f98a2SYan Zheng 42963a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 429779787eaaSJeff Mahoney if (err) 4298fb2d83eeSJosef Bacik goto out_unset; 42995d4f98a2SYan Zheng 43005d4f98a2SYan Zheng merge_reloc_roots(rc); 43015d4f98a2SYan Zheng 43025d4f98a2SYan Zheng unset_reloc_control(rc); 43035d4f98a2SYan Zheng 43047a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 430562b99540SQu Wenruo if (IS_ERR(trans)) { 43063612b495STsutomu Itoh err = PTR_ERR(trans); 43076217b0faSJosef Bacik goto out_clean; 430862b99540SQu Wenruo } 43093a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 43106217b0faSJosef Bacik out_clean: 4311d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4312d2311e69SQu Wenruo if (ret < 0 && !err) 4313d2311e69SQu Wenruo err = ret; 4314fb2d83eeSJosef Bacik out_unset: 4315fb2d83eeSJosef Bacik unset_reloc_control(rc); 4316907d2710SDavid Sterba out_end: 4317907d2710SDavid Sterba reloc_chunk_end(fs_info); 43181a0afa0eSJosef Bacik free_reloc_control(rc); 43193612b495STsutomu Itoh out: 4320aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4321aca1bba6SLiu Bo 43225d4f98a2SYan Zheng btrfs_free_path(path); 43235d4f98a2SYan Zheng 43245d4f98a2SYan Zheng if (err == 0) { 43255d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 4326aeb935a4SQu Wenruo fs_root = btrfs_grab_root(fs_info->data_reloc_root); 4327aeb935a4SQu Wenruo ASSERT(fs_root); 432866b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 432900246528SJosef Bacik btrfs_put_root(fs_root); 4330932fd26dSJosef Bacik } 43315d4f98a2SYan Zheng return err; 43325d4f98a2SYan Zheng } 43335d4f98a2SYan Zheng 43345d4f98a2SYan Zheng /* 43355d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 43365d4f98a2SYan Zheng * 43375d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 43385d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 43395d4f98a2SYan Zheng */ 43407bfa9535SNikolay Borisov int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len) 43415d4f98a2SYan Zheng { 43427bfa9535SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 4343fc28b25eSJosef Bacik struct btrfs_root *csum_root; 43445d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 43455d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 43465d4f98a2SYan Zheng int ret; 43475d4f98a2SYan Zheng u64 disk_bytenr; 43484577b014SJosef Bacik u64 new_bytenr; 43495d4f98a2SYan Zheng LIST_HEAD(list); 43505d4f98a2SYan Zheng 43517bfa9535SNikolay Borisov ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4352bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 43535d4f98a2SYan Zheng 43547bfa9535SNikolay Borisov disk_bytenr = file_pos + inode->index_cnt; 4355fc28b25eSJosef Bacik csum_root = btrfs_csum_root(fs_info, disk_bytenr); 4356fc28b25eSJosef Bacik ret = btrfs_lookup_csums_range(csum_root, disk_bytenr, 435726ce9114SJosef Bacik disk_bytenr + len - 1, &list, 0, false); 435879787eaaSJeff Mahoney if (ret) 435979787eaaSJeff Mahoney goto out; 43605d4f98a2SYan Zheng 43615d4f98a2SYan Zheng while (!list_empty(&list)) { 43625d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 43635d4f98a2SYan Zheng list_del_init(&sums->list); 43645d4f98a2SYan Zheng 43654577b014SJosef Bacik /* 43664577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 43674577b014SJosef Bacik * We need to do this because we will read in entire prealloc 43684577b014SJosef Bacik * extents but we may have written to say the middle of the 43694577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 43704577b014SJosef Bacik * the right disk offset. 43714577b014SJosef Bacik * 43724577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 43734577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 43744577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 43754577b014SJosef Bacik * disk length. 43764577b014SJosef Bacik */ 4377bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 43784577b014SJosef Bacik sums->bytenr = new_bytenr; 43795d4f98a2SYan Zheng 4380f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 43815d4f98a2SYan Zheng } 438279787eaaSJeff Mahoney out: 43835d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4384411fc6bcSAndi Kleen return ret; 43855d4f98a2SYan Zheng } 43863fd0a558SYan, Zheng 438783d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 43883fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 43893fd0a558SYan, Zheng struct extent_buffer *cow) 43903fd0a558SYan, Zheng { 43910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 43923fd0a558SYan, Zheng struct reloc_control *rc; 4393a26195a5SQu Wenruo struct btrfs_backref_node *node; 43943fd0a558SYan, Zheng int first_cow = 0; 43953fd0a558SYan, Zheng int level; 439683d4cfd4SJosef Bacik int ret = 0; 43973fd0a558SYan, Zheng 43980b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 43993fd0a558SYan, Zheng if (!rc) 440083d4cfd4SJosef Bacik return 0; 44013fd0a558SYan, Zheng 440237f00a6dSJohannes Thumshirn BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root)); 44033fd0a558SYan, Zheng 44043fd0a558SYan, Zheng level = btrfs_header_level(buf); 44053fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 44063fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 44073fd0a558SYan, Zheng first_cow = 1; 44083fd0a558SYan, Zheng 44093fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 44103fd0a558SYan, Zheng rc->create_reloc_tree) { 44113fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 44123fd0a558SYan, Zheng 44133fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 44143fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 44153fd0a558SYan, Zheng node->new_bytenr != buf->start); 44163fd0a558SYan, Zheng 4417b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 441867439dadSDavid Sterba atomic_inc(&cow->refs); 44193fd0a558SYan, Zheng node->eb = cow; 44203fd0a558SYan, Zheng node->new_bytenr = cow->start; 44213fd0a558SYan, Zheng 44223fd0a558SYan, Zheng if (!node->pending) { 44233fd0a558SYan, Zheng list_move_tail(&node->list, 44243fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 44253fd0a558SYan, Zheng node->pending = 1; 44263fd0a558SYan, Zheng } 44273fd0a558SYan, Zheng 44283fd0a558SYan, Zheng if (first_cow) 44299569cc20SQu Wenruo mark_block_processed(rc, node); 44303fd0a558SYan, Zheng 44313fd0a558SYan, Zheng if (first_cow && level > 0) 44323fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 44333fd0a558SYan, Zheng } 44343fd0a558SYan, Zheng 443583d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 44363fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 443783d4cfd4SJosef Bacik return ret; 44383fd0a558SYan, Zheng } 44393fd0a558SYan, Zheng 44403fd0a558SYan, Zheng /* 44413fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 444201327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 44433fd0a558SYan, Zheng */ 4444147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 44453fd0a558SYan, Zheng u64 *bytes_to_reserve) 44463fd0a558SYan, Zheng { 444710995c04SQu Wenruo struct btrfs_root *root = pending->root; 444810995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44493fd0a558SYan, Zheng 44506282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 44513fd0a558SYan, Zheng return; 44523fd0a558SYan, Zheng 44533fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 44543fd0a558SYan, Zheng return; 44553fd0a558SYan, Zheng 44563fd0a558SYan, Zheng root = root->reloc_root; 44573fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 44583fd0a558SYan, Zheng /* 44593fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 44603fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 44613fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 44623fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 44633fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 44643fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 44653fd0a558SYan, Zheng * tree may use more space than it frees. so we need 44663fd0a558SYan, Zheng * reserve extra space. 44673fd0a558SYan, Zheng */ 44683fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 44693fd0a558SYan, Zheng } 44703fd0a558SYan, Zheng 44713fd0a558SYan, Zheng /* 44723fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 44733fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 4474f44deb74SJosef Bacik * 4475f44deb74SJosef Bacik * This is similar to btrfs_init_reloc_root(), we come out of here with two 4476f44deb74SJosef Bacik * references held on the reloc_root, one for root->reloc_root and one for 4477f44deb74SJosef Bacik * rc->reloc_roots. 44783fd0a558SYan, Zheng */ 447949b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 44803fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 44813fd0a558SYan, Zheng { 44823fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 44833fd0a558SYan, Zheng struct btrfs_root *reloc_root; 44843fd0a558SYan, Zheng struct btrfs_root *new_root; 448510995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44863fd0a558SYan, Zheng int ret; 44873fd0a558SYan, Zheng 44886282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 448949b25e05SJeff Mahoney return 0; 44903fd0a558SYan, Zheng 44913fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 44923fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 44933fd0a558SYan, Zheng 44943fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 44953fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 44963fd0a558SYan, Zheng rc->block_rsv, 44973a584174SLu Fengqi rc->nodes_relocated, true); 449849b25e05SJeff Mahoney if (ret) 449949b25e05SJeff Mahoney return ret; 45003fd0a558SYan, Zheng } 45013fd0a558SYan, Zheng 45023fd0a558SYan, Zheng new_root = pending->snap; 45033fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 45043fd0a558SYan, Zheng new_root->root_key.objectid); 450549b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 450649b25e05SJeff Mahoney return PTR_ERR(reloc_root); 45073fd0a558SYan, Zheng 4508ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 450957a304cfSJosef Bacik ASSERT(ret != -EEXIST); 45103c925863SJosef Bacik if (ret) { 45113c925863SJosef Bacik /* Pairs with create_reloc_root */ 45123c925863SJosef Bacik btrfs_put_root(reloc_root); 45133c925863SJosef Bacik return ret; 45143c925863SJosef Bacik } 4515f44deb74SJosef Bacik new_root->reloc_root = btrfs_grab_root(reloc_root); 45163fd0a558SYan, Zheng 451749b25e05SJeff Mahoney if (rc->create_reloc_tree) 45183fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 451949b25e05SJeff Mahoney return ret; 45203fd0a558SYan, Zheng } 4521