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" 38103c1972SChristoph Hellwig #include "tree-checker.h" 395d4f98a2SYan Zheng 405d4f98a2SYan Zheng /* 410c891389SQu Wenruo * Relocation overview 420c891389SQu Wenruo * 430c891389SQu Wenruo * [What does relocation do] 440c891389SQu Wenruo * 450c891389SQu Wenruo * The objective of relocation is to relocate all extents of the target block 460c891389SQu Wenruo * group to other block groups. 470c891389SQu Wenruo * This is utilized by resize (shrink only), profile converting, compacting 480c891389SQu Wenruo * space, or balance routine to spread chunks over devices. 490c891389SQu Wenruo * 500c891389SQu Wenruo * Before | After 510c891389SQu Wenruo * ------------------------------------------------------------------ 520c891389SQu Wenruo * BG A: 10 data extents | BG A: deleted 530c891389SQu Wenruo * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) 540c891389SQu Wenruo * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) 550c891389SQu Wenruo * 560c891389SQu Wenruo * [How does relocation work] 570c891389SQu Wenruo * 580c891389SQu Wenruo * 1. Mark the target block group read-only 590c891389SQu Wenruo * New extents won't be allocated from the target block group. 600c891389SQu Wenruo * 610c891389SQu Wenruo * 2.1 Record each extent in the target block group 620c891389SQu Wenruo * To build a proper map of extents to be relocated. 630c891389SQu Wenruo * 640c891389SQu Wenruo * 2.2 Build data reloc tree and reloc trees 650c891389SQu Wenruo * Data reloc tree will contain an inode, recording all newly relocated 660c891389SQu Wenruo * data extents. 670c891389SQu Wenruo * There will be only one data reloc tree for one data block group. 680c891389SQu Wenruo * 690c891389SQu Wenruo * Reloc tree will be a special snapshot of its source tree, containing 700c891389SQu Wenruo * relocated tree blocks. 710c891389SQu Wenruo * Each tree referring to a tree block in target block group will get its 720c891389SQu Wenruo * reloc tree built. 730c891389SQu Wenruo * 740c891389SQu Wenruo * 2.3 Swap source tree with its corresponding reloc tree 750c891389SQu Wenruo * Each involved tree only refers to new extents after swap. 760c891389SQu Wenruo * 770c891389SQu Wenruo * 3. Cleanup reloc trees and data reloc tree. 780c891389SQu Wenruo * As old extents in the target block group are still referenced by reloc 790c891389SQu Wenruo * trees, we need to clean them up before really freeing the target block 800c891389SQu Wenruo * group. 810c891389SQu Wenruo * 820c891389SQu Wenruo * The main complexity is in steps 2.2 and 2.3. 830c891389SQu Wenruo * 840c891389SQu Wenruo * The entry point of relocation is relocate_block_group() function. 850c891389SQu Wenruo */ 860c891389SQu Wenruo 870647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 882a979612SQu Wenruo /* 895d4f98a2SYan Zheng * map address of tree root to tree 905d4f98a2SYan Zheng */ 915d4f98a2SYan Zheng struct mapping_node { 92e9a28dc5SQu Wenruo struct { 935d4f98a2SYan Zheng struct rb_node rb_node; 945d4f98a2SYan Zheng u64 bytenr; 95e9a28dc5SQu Wenruo }; /* Use rb_simle_node for search/insert */ 965d4f98a2SYan Zheng void *data; 975d4f98a2SYan Zheng }; 985d4f98a2SYan Zheng 995d4f98a2SYan Zheng struct mapping_tree { 1005d4f98a2SYan Zheng struct rb_root rb_root; 1015d4f98a2SYan Zheng spinlock_t lock; 1025d4f98a2SYan Zheng }; 1035d4f98a2SYan Zheng 1045d4f98a2SYan Zheng /* 1055d4f98a2SYan Zheng * present a tree block to process 1065d4f98a2SYan Zheng */ 1075d4f98a2SYan Zheng struct tree_block { 108e9a28dc5SQu Wenruo struct { 1095d4f98a2SYan Zheng struct rb_node rb_node; 1105d4f98a2SYan Zheng u64 bytenr; 111e9a28dc5SQu Wenruo }; /* Use rb_simple_node for search/insert */ 112f7ba2d37SJosef Bacik u64 owner; 1135d4f98a2SYan Zheng struct btrfs_key key; 1145d4f98a2SYan Zheng unsigned int level:8; 1155d4f98a2SYan Zheng unsigned int key_ready:1; 1165d4f98a2SYan Zheng }; 1175d4f98a2SYan Zheng 1180257bb82SYan, Zheng #define MAX_EXTENTS 128 1190257bb82SYan, Zheng 1200257bb82SYan, Zheng struct file_extent_cluster { 1210257bb82SYan, Zheng u64 start; 1220257bb82SYan, Zheng u64 end; 1230257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1240257bb82SYan, Zheng unsigned int nr; 1250257bb82SYan, Zheng }; 1260257bb82SYan, Zheng 1275d4f98a2SYan Zheng struct reloc_control { 1285d4f98a2SYan Zheng /* block group to relocate */ 12932da5386SDavid Sterba struct btrfs_block_group *block_group; 1305d4f98a2SYan Zheng /* extent tree */ 1315d4f98a2SYan Zheng struct btrfs_root *extent_root; 1325d4f98a2SYan Zheng /* inode for moving data */ 1335d4f98a2SYan Zheng struct inode *data_inode; 1343fd0a558SYan, Zheng 1353fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1363fd0a558SYan, Zheng 137a26195a5SQu Wenruo struct btrfs_backref_cache backref_cache; 1383fd0a558SYan, Zheng 1393fd0a558SYan, Zheng struct file_extent_cluster cluster; 1405d4f98a2SYan Zheng /* tree blocks have been processed */ 1415d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1425d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1435d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1445d4f98a2SYan Zheng /* list of reloc trees */ 1455d4f98a2SYan Zheng struct list_head reloc_roots; 146d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 147d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 1483fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1493fd0a558SYan, Zheng u64 merging_rsv_size; 1503fd0a558SYan, Zheng /* size of relocated tree nodes */ 1513fd0a558SYan, Zheng u64 nodes_relocated; 1520647bf56SWang Shilong /* reserved size for block group relocation*/ 1530647bf56SWang Shilong u64 reserved_bytes; 1543fd0a558SYan, Zheng 1555d4f98a2SYan Zheng u64 search_start; 1565d4f98a2SYan Zheng u64 extents_found; 1573fd0a558SYan, Zheng 1583fd0a558SYan, Zheng unsigned int stage:8; 1593fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1603fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1615d4f98a2SYan Zheng unsigned int found_file_extent:1; 1625d4f98a2SYan Zheng }; 1635d4f98a2SYan Zheng 1645d4f98a2SYan Zheng /* stages of data relocation */ 1655d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1665d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1675d4f98a2SYan Zheng 1689569cc20SQu Wenruo static void mark_block_processed(struct reloc_control *rc, 169a26195a5SQu Wenruo struct btrfs_backref_node *node) 1709569cc20SQu Wenruo { 1719569cc20SQu Wenruo u32 blocksize; 1729569cc20SQu Wenruo 1739569cc20SQu Wenruo if (node->level == 0 || 1749569cc20SQu Wenruo in_range(node->bytenr, rc->block_group->start, 1759569cc20SQu Wenruo rc->block_group->length)) { 1769569cc20SQu Wenruo blocksize = rc->extent_root->fs_info->nodesize; 1770acd32c2SDavid Sterba set_extent_bit(&rc->processed_blocks, node->bytenr, 1781d126800SDavid Sterba node->bytenr + blocksize - 1, EXTENT_DIRTY, NULL); 1799569cc20SQu Wenruo } 1809569cc20SQu Wenruo node->processed = 1; 1819569cc20SQu Wenruo } 1829569cc20SQu Wenruo 1835d4f98a2SYan Zheng 1845d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1855d4f98a2SYan Zheng { 1866bef4d31SEric Paris tree->rb_root = RB_ROOT; 1875d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1885d4f98a2SYan Zheng } 1895d4f98a2SYan Zheng 1905d4f98a2SYan Zheng /* 1915d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 1925d4f98a2SYan Zheng */ 193a26195a5SQu Wenruo static struct btrfs_backref_node *walk_up_backref( 194a26195a5SQu Wenruo struct btrfs_backref_node *node, 195a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 1965d4f98a2SYan Zheng { 197a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 1985d4f98a2SYan Zheng int idx = *index; 1995d4f98a2SYan Zheng 2005d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 2015d4f98a2SYan Zheng edge = list_entry(node->upper.next, 202a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 2035d4f98a2SYan Zheng edges[idx++] = edge; 2045d4f98a2SYan Zheng node = edge->node[UPPER]; 2055d4f98a2SYan Zheng } 2063fd0a558SYan, Zheng BUG_ON(node->detached); 2075d4f98a2SYan Zheng *index = idx; 2085d4f98a2SYan Zheng return node; 2095d4f98a2SYan Zheng } 2105d4f98a2SYan Zheng 2115d4f98a2SYan Zheng /* 2125d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 2135d4f98a2SYan Zheng */ 214a26195a5SQu Wenruo static struct btrfs_backref_node *walk_down_backref( 215a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 2165d4f98a2SYan Zheng { 217a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 218a26195a5SQu Wenruo struct btrfs_backref_node *lower; 2195d4f98a2SYan Zheng int idx = *index; 2205d4f98a2SYan Zheng 2215d4f98a2SYan Zheng while (idx > 0) { 2225d4f98a2SYan Zheng edge = edges[idx - 1]; 2235d4f98a2SYan Zheng lower = edge->node[LOWER]; 2245d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 2255d4f98a2SYan Zheng idx--; 2265d4f98a2SYan Zheng continue; 2275d4f98a2SYan Zheng } 2285d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 229a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 2305d4f98a2SYan Zheng edges[idx - 1] = edge; 2315d4f98a2SYan Zheng *index = idx; 2325d4f98a2SYan Zheng return edge->node[UPPER]; 2335d4f98a2SYan Zheng } 2345d4f98a2SYan Zheng *index = 0; 2355d4f98a2SYan Zheng return NULL; 2365d4f98a2SYan Zheng } 2375d4f98a2SYan Zheng 238a26195a5SQu Wenruo static void update_backref_node(struct btrfs_backref_cache *cache, 239a26195a5SQu Wenruo struct btrfs_backref_node *node, u64 bytenr) 2403fd0a558SYan, Zheng { 2413fd0a558SYan, Zheng struct rb_node *rb_node; 2423fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 2433fd0a558SYan, Zheng node->bytenr = bytenr; 244e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node); 24543c04fb1SJeff Mahoney if (rb_node) 246982c92cbSQu Wenruo btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST); 2473fd0a558SYan, Zheng } 2483fd0a558SYan, Zheng 2493fd0a558SYan, Zheng /* 2503fd0a558SYan, Zheng * update backref cache after a transaction commit 2513fd0a558SYan, Zheng */ 2523fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 253a26195a5SQu Wenruo struct btrfs_backref_cache *cache) 2543fd0a558SYan, Zheng { 255a26195a5SQu Wenruo struct btrfs_backref_node *node; 2563fd0a558SYan, Zheng int level = 0; 2573fd0a558SYan, Zheng 2583fd0a558SYan, Zheng if (cache->last_trans == 0) { 2593fd0a558SYan, Zheng cache->last_trans = trans->transid; 2603fd0a558SYan, Zheng return 0; 2613fd0a558SYan, Zheng } 2623fd0a558SYan, Zheng 2633fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 2643fd0a558SYan, Zheng return 0; 2653fd0a558SYan, Zheng 2663fd0a558SYan, Zheng /* 2673fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 2683fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 2693fd0a558SYan, Zheng * so the detached nodes are no longer useful. 2703fd0a558SYan, Zheng */ 2713fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2723fd0a558SYan, Zheng node = list_entry(cache->detached.next, 273a26195a5SQu Wenruo struct btrfs_backref_node, list); 274023acb07SQu Wenruo btrfs_backref_cleanup_node(cache, node); 2753fd0a558SYan, Zheng } 2763fd0a558SYan, Zheng 2773fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 2783fd0a558SYan, Zheng node = list_entry(cache->changed.next, 279a26195a5SQu Wenruo struct btrfs_backref_node, list); 2803fd0a558SYan, Zheng list_del_init(&node->list); 2813fd0a558SYan, Zheng BUG_ON(node->pending); 2823fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2833fd0a558SYan, Zheng } 2843fd0a558SYan, Zheng 2853fd0a558SYan, Zheng /* 2863fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 2873fd0a558SYan, Zheng * errors during processing the pending nodes. 2883fd0a558SYan, Zheng */ 2893fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 2903fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 2913fd0a558SYan, Zheng BUG_ON(!node->pending); 2923fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 2933fd0a558SYan, Zheng continue; 2943fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 2953fd0a558SYan, Zheng } 2963fd0a558SYan, Zheng } 2973fd0a558SYan, Zheng 2983fd0a558SYan, Zheng cache->last_trans = 0; 2993fd0a558SYan, Zheng return 1; 3003fd0a558SYan, Zheng } 3013fd0a558SYan, Zheng 3026282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 3036282675eSQu Wenruo { 3046282675eSQu Wenruo /* 3056282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 3066282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 3076282675eSQu Wenruo * trying to access reloc_root 3086282675eSQu Wenruo */ 3096282675eSQu Wenruo smp_rmb(); 3106282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 3116282675eSQu Wenruo return true; 3126282675eSQu Wenruo return false; 3136282675eSQu Wenruo } 3146282675eSQu Wenruo 3156282675eSQu Wenruo /* 3166282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 3176282675eSQu Wenruo * 3186282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 3196282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 32055465730SQu Wenruo * from no reloc root. But btrfs_should_ignore_reloc_root() below is a 32155465730SQu Wenruo * special case. 3226282675eSQu Wenruo */ 3236282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 3246282675eSQu Wenruo { 3256282675eSQu Wenruo if (reloc_root_is_dead(root)) 3266282675eSQu Wenruo return false; 3276282675eSQu Wenruo if (!root->reloc_root) 3286282675eSQu Wenruo return false; 3296282675eSQu Wenruo return true; 3306282675eSQu Wenruo } 331f2a97a9dSDavid Sterba 33255465730SQu Wenruo int btrfs_should_ignore_reloc_root(struct btrfs_root *root) 3333fd0a558SYan, Zheng { 3343fd0a558SYan, Zheng struct btrfs_root *reloc_root; 3353fd0a558SYan, Zheng 33692a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 3373fd0a558SYan, Zheng return 0; 3383fd0a558SYan, Zheng 3396282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 3406282675eSQu Wenruo if (reloc_root_is_dead(root)) 3416282675eSQu Wenruo return 1; 3426282675eSQu Wenruo 3433fd0a558SYan, Zheng reloc_root = root->reloc_root; 3443fd0a558SYan, Zheng if (!reloc_root) 3453fd0a558SYan, Zheng return 0; 3463fd0a558SYan, Zheng 3474d4225fcSJosef Bacik if (btrfs_header_generation(reloc_root->commit_root) == 3484d4225fcSJosef Bacik root->fs_info->running_transaction->transid) 3493fd0a558SYan, Zheng return 0; 3503fd0a558SYan, Zheng /* 3513fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 3523fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 3533fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 3543fd0a558SYan, Zheng * relocation. 3553fd0a558SYan, Zheng */ 3563fd0a558SYan, Zheng return 1; 3573fd0a558SYan, Zheng } 35855465730SQu Wenruo 3595d4f98a2SYan Zheng /* 3605d4f98a2SYan Zheng * find reloc tree by address of tree root 3615d4f98a2SYan Zheng */ 3622433bea5SQu Wenruo struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) 3635d4f98a2SYan Zheng { 3642433bea5SQu Wenruo struct reloc_control *rc = fs_info->reloc_ctl; 3655d4f98a2SYan Zheng struct rb_node *rb_node; 3665d4f98a2SYan Zheng struct mapping_node *node; 3675d4f98a2SYan Zheng struct btrfs_root *root = NULL; 3685d4f98a2SYan Zheng 3692433bea5SQu Wenruo ASSERT(rc); 3705d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 371e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr); 3725d4f98a2SYan Zheng if (rb_node) { 3735d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 3740d031dc4SYu Zhe root = node->data; 3755d4f98a2SYan Zheng } 3765d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 37700246528SJosef Bacik return btrfs_grab_root(root); 3785d4f98a2SYan Zheng } 3795d4f98a2SYan Zheng 3805d4f98a2SYan Zheng /* 38129db137bSQu Wenruo * For useless nodes, do two major clean ups: 38229db137bSQu Wenruo * 38329db137bSQu Wenruo * - Cleanup the children edges and nodes 38429db137bSQu Wenruo * If child node is also orphan (no parent) during cleanup, then the child 38529db137bSQu Wenruo * node will also be cleaned up. 38629db137bSQu Wenruo * 38729db137bSQu Wenruo * - Freeing up leaves (level 0), keeps nodes detached 38829db137bSQu Wenruo * For nodes, the node is still cached as "detached" 38929db137bSQu Wenruo * 39029db137bSQu Wenruo * Return false if @node is not in the @useless_nodes list. 39129db137bSQu Wenruo * Return true if @node is in the @useless_nodes list. 39229db137bSQu Wenruo */ 39329db137bSQu Wenruo static bool handle_useless_nodes(struct reloc_control *rc, 394a26195a5SQu Wenruo struct btrfs_backref_node *node) 39529db137bSQu Wenruo { 396a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 39729db137bSQu Wenruo struct list_head *useless_node = &cache->useless_node; 39829db137bSQu Wenruo bool ret = false; 39929db137bSQu Wenruo 40029db137bSQu Wenruo while (!list_empty(useless_node)) { 401a26195a5SQu Wenruo struct btrfs_backref_node *cur; 40229db137bSQu Wenruo 403a26195a5SQu Wenruo cur = list_first_entry(useless_node, struct btrfs_backref_node, 40429db137bSQu Wenruo list); 40529db137bSQu Wenruo list_del_init(&cur->list); 40629db137bSQu Wenruo 40729db137bSQu Wenruo /* Only tree root nodes can be added to @useless_nodes */ 40829db137bSQu Wenruo ASSERT(list_empty(&cur->upper)); 40929db137bSQu Wenruo 41029db137bSQu Wenruo if (cur == node) 41129db137bSQu Wenruo ret = true; 41229db137bSQu Wenruo 41329db137bSQu Wenruo /* The node is the lowest node */ 41429db137bSQu Wenruo if (cur->lowest) { 41529db137bSQu Wenruo list_del_init(&cur->lower); 41629db137bSQu Wenruo cur->lowest = 0; 41729db137bSQu Wenruo } 41829db137bSQu Wenruo 41929db137bSQu Wenruo /* Cleanup the lower edges */ 42029db137bSQu Wenruo while (!list_empty(&cur->lower)) { 421a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 422a26195a5SQu Wenruo struct btrfs_backref_node *lower; 42329db137bSQu Wenruo 42429db137bSQu Wenruo edge = list_entry(cur->lower.next, 425a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 42629db137bSQu Wenruo list_del(&edge->list[UPPER]); 42729db137bSQu Wenruo list_del(&edge->list[LOWER]); 42829db137bSQu Wenruo lower = edge->node[LOWER]; 429741188d3SQu Wenruo btrfs_backref_free_edge(cache, edge); 43029db137bSQu Wenruo 43129db137bSQu Wenruo /* Child node is also orphan, queue for cleanup */ 43229db137bSQu Wenruo if (list_empty(&lower->upper)) 43329db137bSQu Wenruo list_add(&lower->list, useless_node); 43429db137bSQu Wenruo } 43529db137bSQu Wenruo /* Mark this block processed for relocation */ 43629db137bSQu Wenruo mark_block_processed(rc, cur); 43729db137bSQu Wenruo 43829db137bSQu Wenruo /* 43929db137bSQu Wenruo * Backref nodes for tree leaves are deleted from the cache. 44029db137bSQu Wenruo * Backref nodes for upper level tree blocks are left in the 44129db137bSQu Wenruo * cache to avoid unnecessary backref lookup. 44229db137bSQu Wenruo */ 44329db137bSQu Wenruo if (cur->level > 0) { 44429db137bSQu Wenruo list_add(&cur->list, &cache->detached); 44529db137bSQu Wenruo cur->detached = 1; 44629db137bSQu Wenruo } else { 44729db137bSQu Wenruo rb_erase(&cur->rb_node, &cache->rb_root); 448741188d3SQu Wenruo btrfs_backref_free_node(cache, cur); 44929db137bSQu Wenruo } 45029db137bSQu Wenruo } 45129db137bSQu Wenruo return ret; 45229db137bSQu Wenruo } 45329db137bSQu Wenruo 45429db137bSQu Wenruo /* 455e7d571c7SQu Wenruo * Build backref tree for a given tree block. Root of the backref tree 456e7d571c7SQu Wenruo * corresponds the tree block, leaves of the backref tree correspond roots of 457e7d571c7SQu Wenruo * b-trees that reference the tree block. 458e7d571c7SQu Wenruo * 459e7d571c7SQu Wenruo * The basic idea of this function is check backrefs of a given block to find 460e7d571c7SQu Wenruo * upper level blocks that reference the block, and then check backrefs of 461e7d571c7SQu Wenruo * these upper level blocks recursively. The recursion stops when tree root is 462e7d571c7SQu Wenruo * reached or backrefs for the block is cached. 463e7d571c7SQu Wenruo * 464e7d571c7SQu Wenruo * NOTE: if we find that backrefs for a block are cached, we know backrefs for 465e7d571c7SQu Wenruo * all upper level blocks that directly/indirectly reference the block are also 466e7d571c7SQu Wenruo * cached. 467e7d571c7SQu Wenruo */ 468a26195a5SQu Wenruo static noinline_for_stack struct btrfs_backref_node *build_backref_tree( 469e7d571c7SQu Wenruo struct reloc_control *rc, struct btrfs_key *node_key, 470e7d571c7SQu Wenruo int level, u64 bytenr) 471e7d571c7SQu Wenruo { 472e7d571c7SQu Wenruo struct btrfs_backref_iter *iter; 473a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 474e7d571c7SQu Wenruo /* For searching parent of TREE_BLOCK_REF */ 475e7d571c7SQu Wenruo struct btrfs_path *path; 476a26195a5SQu Wenruo struct btrfs_backref_node *cur; 477a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 478a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 479e7d571c7SQu Wenruo int ret; 480e7d571c7SQu Wenruo int err = 0; 481e7d571c7SQu Wenruo 482d68194b2SDavid Sterba iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info); 483e7d571c7SQu Wenruo if (!iter) 484e7d571c7SQu Wenruo return ERR_PTR(-ENOMEM); 485e7d571c7SQu Wenruo path = btrfs_alloc_path(); 486e7d571c7SQu Wenruo if (!path) { 487e7d571c7SQu Wenruo err = -ENOMEM; 488e7d571c7SQu Wenruo goto out; 489e7d571c7SQu Wenruo } 490e7d571c7SQu Wenruo 491b1818dabSQu Wenruo node = btrfs_backref_alloc_node(cache, bytenr, level); 492e7d571c7SQu Wenruo if (!node) { 493e7d571c7SQu Wenruo err = -ENOMEM; 494e7d571c7SQu Wenruo goto out; 495e7d571c7SQu Wenruo } 496e7d571c7SQu Wenruo 497e7d571c7SQu Wenruo node->lowest = 1; 498e7d571c7SQu Wenruo cur = node; 499e7d571c7SQu Wenruo 500e7d571c7SQu Wenruo /* Breadth-first search to build backref cache */ 501e7d571c7SQu Wenruo do { 5021b60d2ecSQu Wenruo ret = btrfs_backref_add_tree_node(cache, path, iter, node_key, 5031b60d2ecSQu Wenruo cur); 504e7d571c7SQu Wenruo if (ret < 0) { 505e7d571c7SQu Wenruo err = ret; 506e7d571c7SQu Wenruo goto out; 507e7d571c7SQu Wenruo } 508e7d571c7SQu Wenruo edge = list_first_entry_or_null(&cache->pending_edge, 509a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 510e7d571c7SQu Wenruo /* 511e7d571c7SQu Wenruo * The pending list isn't empty, take the first block to 512e7d571c7SQu Wenruo * process 513e7d571c7SQu Wenruo */ 514e7d571c7SQu Wenruo if (edge) { 5155d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 5165d4f98a2SYan Zheng cur = edge->node[UPPER]; 5175d4f98a2SYan Zheng } 518e7d571c7SQu Wenruo } while (edge); 5195d4f98a2SYan Zheng 5201f872924SQu Wenruo /* Finish the upper linkage of newly added edges/nodes */ 521fc997ed0SQu Wenruo ret = btrfs_backref_finish_upper_links(cache, node); 5221f872924SQu Wenruo if (ret < 0) { 5231f872924SQu Wenruo err = ret; 52475bfb9afSJosef Bacik goto out; 52575bfb9afSJosef Bacik } 52675bfb9afSJosef Bacik 52729db137bSQu Wenruo if (handle_useless_nodes(rc, node)) 5283fd0a558SYan, Zheng node = NULL; 5295d4f98a2SYan Zheng out: 53071f572a9SQu Wenruo btrfs_backref_iter_free(iter); 53171f572a9SQu Wenruo btrfs_free_path(path); 5325d4f98a2SYan Zheng if (err) { 5331b23ea18SQu Wenruo btrfs_backref_error_cleanup(cache, node); 5345d4f98a2SYan Zheng return ERR_PTR(err); 5355d4f98a2SYan Zheng } 53675bfb9afSJosef Bacik ASSERT(!node || !node->detached); 53784780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 53884780289SQu Wenruo list_empty(&cache->pending_edge)); 5395d4f98a2SYan Zheng return node; 5405d4f98a2SYan Zheng } 5415d4f98a2SYan Zheng 5425d4f98a2SYan Zheng /* 5433fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 5443fd0a558SYan, Zheng * the backref node is created by cloning backref node that 5453fd0a558SYan, Zheng * corresponds to root of source tree 5463fd0a558SYan, Zheng */ 5473fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 5483fd0a558SYan, Zheng struct reloc_control *rc, 5493fd0a558SYan, Zheng struct btrfs_root *src, 5503fd0a558SYan, Zheng struct btrfs_root *dest) 5513fd0a558SYan, Zheng { 5523fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 553a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 554a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 555a26195a5SQu Wenruo struct btrfs_backref_node *new_node; 556a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 557a26195a5SQu Wenruo struct btrfs_backref_edge *new_edge; 5583fd0a558SYan, Zheng struct rb_node *rb_node; 5593fd0a558SYan, Zheng 5603fd0a558SYan, Zheng if (cache->last_trans > 0) 5613fd0a558SYan, Zheng update_backref_cache(trans, cache); 5623fd0a558SYan, Zheng 563e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start); 5643fd0a558SYan, Zheng if (rb_node) { 565a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, rb_node); 5663fd0a558SYan, Zheng if (node->detached) 5673fd0a558SYan, Zheng node = NULL; 5683fd0a558SYan, Zheng else 5693fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 5703fd0a558SYan, Zheng } 5713fd0a558SYan, Zheng 5723fd0a558SYan, Zheng if (!node) { 573e9a28dc5SQu Wenruo rb_node = rb_simple_search(&cache->rb_root, 5743fd0a558SYan, Zheng reloc_root->commit_root->start); 5753fd0a558SYan, Zheng if (rb_node) { 576a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, 5773fd0a558SYan, Zheng rb_node); 5783fd0a558SYan, Zheng BUG_ON(node->detached); 5793fd0a558SYan, Zheng } 5803fd0a558SYan, Zheng } 5813fd0a558SYan, Zheng 5823fd0a558SYan, Zheng if (!node) 5833fd0a558SYan, Zheng return 0; 5843fd0a558SYan, Zheng 585b1818dabSQu Wenruo new_node = btrfs_backref_alloc_node(cache, dest->node->start, 586b1818dabSQu Wenruo node->level); 5873fd0a558SYan, Zheng if (!new_node) 5883fd0a558SYan, Zheng return -ENOMEM; 5893fd0a558SYan, Zheng 5903fd0a558SYan, Zheng new_node->lowest = node->lowest; 5916848ad64SYan, Zheng new_node->checked = 1; 59200246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 5930b530bc5SJosef Bacik ASSERT(new_node->root); 5943fd0a558SYan, Zheng 5953fd0a558SYan, Zheng if (!node->lowest) { 5963fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 59747254d07SQu Wenruo new_edge = btrfs_backref_alloc_edge(cache); 5983fd0a558SYan, Zheng if (!new_edge) 5993fd0a558SYan, Zheng goto fail; 6003fd0a558SYan, Zheng 601f39911e5SQu Wenruo btrfs_backref_link_edge(new_edge, edge->node[LOWER], 602f39911e5SQu Wenruo new_node, LINK_UPPER); 6033fd0a558SYan, Zheng } 60476b9e23dSMiao Xie } else { 60576b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 6063fd0a558SYan, Zheng } 6073fd0a558SYan, Zheng 608e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr, 6093fd0a558SYan, Zheng &new_node->rb_node); 61043c04fb1SJeff Mahoney if (rb_node) 611982c92cbSQu Wenruo btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST); 6123fd0a558SYan, Zheng 6133fd0a558SYan, Zheng if (!new_node->lowest) { 6143fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 6153fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 6163fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 6173fd0a558SYan, Zheng } 6183fd0a558SYan, Zheng } 6193fd0a558SYan, Zheng return 0; 6203fd0a558SYan, Zheng fail: 6213fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 6223fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 623a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 6243fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 625741188d3SQu Wenruo btrfs_backref_free_edge(cache, new_edge); 6263fd0a558SYan, Zheng } 627741188d3SQu Wenruo btrfs_backref_free_node(cache, new_node); 6283fd0a558SYan, Zheng return -ENOMEM; 6293fd0a558SYan, Zheng } 6303fd0a558SYan, Zheng 6313fd0a558SYan, Zheng /* 6325d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 6335d4f98a2SYan Zheng */ 634ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 6355d4f98a2SYan Zheng { 6360b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6375d4f98a2SYan Zheng struct rb_node *rb_node; 6385d4f98a2SYan Zheng struct mapping_node *node; 6390b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 6405d4f98a2SYan Zheng 6415d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 642ffd7b339SJeff Mahoney if (!node) 643ffd7b339SJeff Mahoney return -ENOMEM; 6445d4f98a2SYan Zheng 645ea287ab1SJosef Bacik node->bytenr = root->commit_root->start; 6465d4f98a2SYan Zheng node->data = root; 6475d4f98a2SYan Zheng 6485d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 649e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 6505d4f98a2SYan Zheng node->bytenr, &node->rb_node); 6515d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 652ffd7b339SJeff Mahoney if (rb_node) { 65357a304cfSJosef Bacik btrfs_err(fs_info, 6545d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 6555d163e0eSJeff Mahoney node->bytenr); 65657a304cfSJosef Bacik return -EEXIST; 657ffd7b339SJeff Mahoney } 6585d4f98a2SYan Zheng 6595d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 6605d4f98a2SYan Zheng return 0; 6615d4f98a2SYan Zheng } 6625d4f98a2SYan Zheng 6635d4f98a2SYan Zheng /* 664c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 6655d4f98a2SYan Zheng * mapping 6665d4f98a2SYan Zheng */ 667c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 6685d4f98a2SYan Zheng { 6690b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6705d4f98a2SYan Zheng struct rb_node *rb_node; 6715d4f98a2SYan Zheng struct mapping_node *node = NULL; 6720b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 673f44deb74SJosef Bacik bool put_ref = false; 6745d4f98a2SYan Zheng 67565c6e82bSQu Wenruo if (rc && root->node) { 6765d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 677e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 678ea287ab1SJosef Bacik root->commit_root->start); 679c974c464SWang Shilong if (rb_node) { 680c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 681c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 682ea287ab1SJosef Bacik RB_CLEAR_NODE(&node->rb_node); 683c974c464SWang Shilong } 684c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 685c78a10aeSJosef Bacik ASSERT(!node || (struct btrfs_root *)node->data == root); 686389305b2SQu Wenruo } 687c974c464SWang Shilong 688f44deb74SJosef Bacik /* 689f44deb74SJosef Bacik * We only put the reloc root here if it's on the list. There's a lot 690f44deb74SJosef Bacik * of places where the pattern is to splice the rc->reloc_roots, process 691f44deb74SJosef Bacik * the reloc roots, and then add the reloc root back onto 692f44deb74SJosef Bacik * rc->reloc_roots. If we call __del_reloc_root while it's off of the 693f44deb74SJosef Bacik * list we don't want the reference being dropped, because the guy 694f44deb74SJosef Bacik * messing with the list is in charge of the reference. 695f44deb74SJosef Bacik */ 6960b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 697f44deb74SJosef Bacik if (!list_empty(&root->root_list)) { 698f44deb74SJosef Bacik put_ref = true; 699c974c464SWang Shilong list_del_init(&root->root_list); 700f44deb74SJosef Bacik } 7010b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 702f44deb74SJosef Bacik if (put_ref) 703f44deb74SJosef Bacik btrfs_put_root(root); 704c974c464SWang Shilong kfree(node); 705c974c464SWang Shilong } 706c974c464SWang Shilong 707c974c464SWang Shilong /* 708c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 709c974c464SWang Shilong * mapping 710c974c464SWang Shilong */ 711ea287ab1SJosef Bacik static int __update_reloc_root(struct btrfs_root *root) 712c974c464SWang Shilong { 7130b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 714c974c464SWang Shilong struct rb_node *rb_node; 715c974c464SWang Shilong struct mapping_node *node = NULL; 7160b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 717c974c464SWang Shilong 718c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 719e9a28dc5SQu Wenruo rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, 720ea287ab1SJosef Bacik root->commit_root->start); 7215d4f98a2SYan Zheng if (rb_node) { 7225d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 7235d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 7245d4f98a2SYan Zheng } 7255d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 7265d4f98a2SYan Zheng 7278f71f3e0SLiu Bo if (!node) 7288f71f3e0SLiu Bo return 0; 7295d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 7305d4f98a2SYan Zheng 7315d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 732ea287ab1SJosef Bacik node->bytenr = root->node->start; 733e9a28dc5SQu Wenruo rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, 7345d4f98a2SYan Zheng node->bytenr, &node->rb_node); 7355d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 73643c04fb1SJeff Mahoney if (rb_node) 737982c92cbSQu Wenruo btrfs_backref_panic(fs_info, node->bytenr, -EEXIST); 7385d4f98a2SYan Zheng return 0; 7395d4f98a2SYan Zheng } 7405d4f98a2SYan Zheng 7413fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 7423fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 7435d4f98a2SYan Zheng { 7440b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 7455d4f98a2SYan Zheng struct btrfs_root *reloc_root; 7465d4f98a2SYan Zheng struct extent_buffer *eb; 7475d4f98a2SYan Zheng struct btrfs_root_item *root_item; 7485d4f98a2SYan Zheng struct btrfs_key root_key; 74984c50ba5SJosef Bacik int ret = 0; 75084c50ba5SJosef Bacik bool must_abort = false; 7515d4f98a2SYan Zheng 7525d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 75384c50ba5SJosef Bacik if (!root_item) 75484c50ba5SJosef Bacik return ERR_PTR(-ENOMEM); 7555d4f98a2SYan Zheng 7565d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 7575d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 7583fd0a558SYan, Zheng root_key.offset = objectid; 7595d4f98a2SYan Zheng 7603fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 761054570a1SFilipe Manana u64 commit_root_gen; 762054570a1SFilipe Manana 7633fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 7645d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 7655d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 76684c50ba5SJosef Bacik if (ret) 76784c50ba5SJosef Bacik goto fail; 76884c50ba5SJosef Bacik 769054570a1SFilipe Manana /* 770054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 771054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 772054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 773054570a1SFilipe Manana * either inside the critical section of a transaction commit 774054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 775054570a1SFilipe Manana * it's created before the transaction commit is started. 776054570a1SFilipe Manana */ 777054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 778054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 7793fd0a558SYan, Zheng } else { 7803fd0a558SYan, Zheng /* 7813fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 7823fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 7833fd0a558SYan, Zheng * modified after it was created have RELOC flag 7843fd0a558SYan, Zheng * set in their headers. so it's OK to not update 7853fd0a558SYan, Zheng * the 'last_snapshot'. 7863fd0a558SYan, Zheng */ 7873fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 7883fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 78984c50ba5SJosef Bacik if (ret) 79084c50ba5SJosef Bacik goto fail; 7913fd0a558SYan, Zheng } 7923fd0a558SYan, Zheng 79384c50ba5SJosef Bacik /* 79484c50ba5SJosef Bacik * We have changed references at this point, we must abort the 79584c50ba5SJosef Bacik * transaction if anything fails. 79684c50ba5SJosef Bacik */ 79784c50ba5SJosef Bacik must_abort = true; 79884c50ba5SJosef Bacik 7995d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 8005d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 8015d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 8025d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 8033fd0a558SYan, Zheng 8043fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 8053fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 8063fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 8073fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 808c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, 0); 8093fd0a558SYan, Zheng } 8105d4f98a2SYan Zheng 8115d4f98a2SYan Zheng btrfs_tree_unlock(eb); 8125d4f98a2SYan Zheng free_extent_buffer(eb); 8135d4f98a2SYan Zheng 8140b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 8155d4f98a2SYan Zheng &root_key, root_item); 81684c50ba5SJosef Bacik if (ret) 81784c50ba5SJosef Bacik goto fail; 81884c50ba5SJosef Bacik 8195d4f98a2SYan Zheng kfree(root_item); 8205d4f98a2SYan Zheng 8213dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 82284c50ba5SJosef Bacik if (IS_ERR(reloc_root)) { 82384c50ba5SJosef Bacik ret = PTR_ERR(reloc_root); 82484c50ba5SJosef Bacik goto abort; 82584c50ba5SJosef Bacik } 82692a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 8275d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 8283fd0a558SYan, Zheng return reloc_root; 82984c50ba5SJosef Bacik fail: 83084c50ba5SJosef Bacik kfree(root_item); 83184c50ba5SJosef Bacik abort: 83284c50ba5SJosef Bacik if (must_abort) 83384c50ba5SJosef Bacik btrfs_abort_transaction(trans, ret); 83484c50ba5SJosef Bacik return ERR_PTR(ret); 8353fd0a558SYan, Zheng } 8363fd0a558SYan, Zheng 8373fd0a558SYan, Zheng /* 8383fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 8393fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 840f44deb74SJosef Bacik * 841f44deb74SJosef Bacik * The reloc_root comes out of here with two references, one for 842f44deb74SJosef Bacik * root->reloc_root, and another for being on the rc->reloc_roots list. 8433fd0a558SYan, Zheng */ 8443fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 8453fd0a558SYan, Zheng struct btrfs_root *root) 8463fd0a558SYan, Zheng { 8470b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8483fd0a558SYan, Zheng struct btrfs_root *reloc_root; 8490b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 85020dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 8513fd0a558SYan, Zheng int clear_rsv = 0; 852ffd7b339SJeff Mahoney int ret; 8533fd0a558SYan, Zheng 854aec7db3bSJosef Bacik if (!rc) 8552abc726aSJosef Bacik return 0; 8562abc726aSJosef Bacik 8571fac4a54SQu Wenruo /* 8581fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 8591fac4a54SQu Wenruo * create/update the dead reloc tree 8601fac4a54SQu Wenruo */ 8616282675eSQu Wenruo if (reloc_root_is_dead(root)) 8621fac4a54SQu Wenruo return 0; 8631fac4a54SQu Wenruo 864aec7db3bSJosef Bacik /* 865aec7db3bSJosef Bacik * This is subtle but important. We do not do 866aec7db3bSJosef Bacik * record_root_in_transaction for reloc roots, instead we record their 867aec7db3bSJosef Bacik * corresponding fs root, and then here we update the last trans for the 868aec7db3bSJosef Bacik * reloc root. This means that we have to do this for the entire life 869aec7db3bSJosef Bacik * of the reloc root, regardless of which stage of the relocation we are 870aec7db3bSJosef Bacik * in. 871aec7db3bSJosef Bacik */ 8723fd0a558SYan, Zheng if (root->reloc_root) { 8733fd0a558SYan, Zheng reloc_root = root->reloc_root; 8743fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 8753fd0a558SYan, Zheng return 0; 8763fd0a558SYan, Zheng } 8773fd0a558SYan, Zheng 878aec7db3bSJosef Bacik /* 879aec7db3bSJosef Bacik * We are merging reloc roots, we do not need new reloc trees. Also 880aec7db3bSJosef Bacik * reloc trees never need their own reloc tree. 881aec7db3bSJosef Bacik */ 882aec7db3bSJosef Bacik if (!rc->create_reloc_tree || 883aec7db3bSJosef Bacik root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 884aec7db3bSJosef Bacik return 0; 885aec7db3bSJosef Bacik 88620dd2cbfSMiao Xie if (!trans->reloc_reserved) { 88720dd2cbfSMiao Xie rsv = trans->block_rsv; 8883fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 8893fd0a558SYan, Zheng clear_rsv = 1; 8903fd0a558SYan, Zheng } 8913fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 8923fd0a558SYan, Zheng if (clear_rsv) 89320dd2cbfSMiao Xie trans->block_rsv = rsv; 89400bb36a0SJosef Bacik if (IS_ERR(reloc_root)) 89500bb36a0SJosef Bacik return PTR_ERR(reloc_root); 8965d4f98a2SYan Zheng 897ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 89857a304cfSJosef Bacik ASSERT(ret != -EEXIST); 89900bb36a0SJosef Bacik if (ret) { 90000bb36a0SJosef Bacik /* Pairs with create_reloc_root */ 90100bb36a0SJosef Bacik btrfs_put_root(reloc_root); 90200bb36a0SJosef Bacik return ret; 90300bb36a0SJosef Bacik } 904f44deb74SJosef Bacik root->reloc_root = btrfs_grab_root(reloc_root); 9055d4f98a2SYan Zheng return 0; 9065d4f98a2SYan Zheng } 9075d4f98a2SYan Zheng 9085d4f98a2SYan Zheng /* 9095d4f98a2SYan Zheng * update root item of reloc tree 9105d4f98a2SYan Zheng */ 9115d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 9125d4f98a2SYan Zheng struct btrfs_root *root) 9135d4f98a2SYan Zheng { 9140b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 9155d4f98a2SYan Zheng struct btrfs_root *reloc_root; 9165d4f98a2SYan Zheng struct btrfs_root_item *root_item; 9175d4f98a2SYan Zheng int ret; 9185d4f98a2SYan Zheng 9196282675eSQu Wenruo if (!have_reloc_root(root)) 920592fbcd5SJosef Bacik return 0; 9215d4f98a2SYan Zheng 9225d4f98a2SYan Zheng reloc_root = root->reloc_root; 9235d4f98a2SYan Zheng root_item = &reloc_root->root_item; 9245d4f98a2SYan Zheng 925f44deb74SJosef Bacik /* 926f44deb74SJosef Bacik * We are probably ok here, but __del_reloc_root() will drop its ref of 927f44deb74SJosef Bacik * the root. We have the ref for root->reloc_root, but just in case 928f44deb74SJosef Bacik * hold it while we update the reloc root. 929f44deb74SJosef Bacik */ 930f44deb74SJosef Bacik btrfs_grab_root(reloc_root); 931f44deb74SJosef Bacik 932d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 9330b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 9343fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 935d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 9366282675eSQu Wenruo /* 9376282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 9386282675eSQu Wenruo * have_reloc_root will not touch it from now on. 9396282675eSQu Wenruo */ 9406282675eSQu Wenruo smp_wmb(); 941c974c464SWang Shilong __del_reloc_root(reloc_root); 9425d4f98a2SYan Zheng } 9435d4f98a2SYan Zheng 9445d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 945ea287ab1SJosef Bacik __update_reloc_root(reloc_root); 9465d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 9475d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 9485d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 9495d4f98a2SYan Zheng } 9505d4f98a2SYan Zheng 9510b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 9525d4f98a2SYan Zheng &reloc_root->root_key, root_item); 953f44deb74SJosef Bacik btrfs_put_root(reloc_root); 954592fbcd5SJosef Bacik return ret; 9555d4f98a2SYan Zheng } 9565d4f98a2SYan Zheng 9575d4f98a2SYan Zheng /* 9585d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 9595d4f98a2SYan Zheng * in a subvolume 9605d4f98a2SYan Zheng */ 9615d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 9625d4f98a2SYan Zheng { 9635d4f98a2SYan Zheng struct rb_node *node; 9645d4f98a2SYan Zheng struct rb_node *prev; 9655d4f98a2SYan Zheng struct btrfs_inode *entry; 9665d4f98a2SYan Zheng struct inode *inode; 9675d4f98a2SYan Zheng 9685d4f98a2SYan Zheng spin_lock(&root->inode_lock); 9695d4f98a2SYan Zheng again: 9705d4f98a2SYan Zheng node = root->inode_tree.rb_node; 9715d4f98a2SYan Zheng prev = NULL; 9725d4f98a2SYan Zheng while (node) { 9735d4f98a2SYan Zheng prev = node; 9745d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9755d4f98a2SYan Zheng 9764a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 9775d4f98a2SYan Zheng node = node->rb_left; 9784a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 9795d4f98a2SYan Zheng node = node->rb_right; 9805d4f98a2SYan Zheng else 9815d4f98a2SYan Zheng break; 9825d4f98a2SYan Zheng } 9835d4f98a2SYan Zheng if (!node) { 9845d4f98a2SYan Zheng while (prev) { 9855d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 9864a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 9875d4f98a2SYan Zheng node = prev; 9885d4f98a2SYan Zheng break; 9895d4f98a2SYan Zheng } 9905d4f98a2SYan Zheng prev = rb_next(prev); 9915d4f98a2SYan Zheng } 9925d4f98a2SYan Zheng } 9935d4f98a2SYan Zheng while (node) { 9945d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 9955d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 9965d4f98a2SYan Zheng if (inode) { 9975d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 9985d4f98a2SYan Zheng return inode; 9995d4f98a2SYan Zheng } 10005d4f98a2SYan Zheng 10014a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 10025d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 10035d4f98a2SYan Zheng goto again; 10045d4f98a2SYan Zheng 10055d4f98a2SYan Zheng node = rb_next(node); 10065d4f98a2SYan Zheng } 10075d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 10085d4f98a2SYan Zheng return NULL; 10095d4f98a2SYan Zheng } 10105d4f98a2SYan Zheng 10115d4f98a2SYan Zheng /* 10125d4f98a2SYan Zheng * get new location of data 10135d4f98a2SYan Zheng */ 10145d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 10155d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 10165d4f98a2SYan Zheng { 10175d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 10185d4f98a2SYan Zheng struct btrfs_path *path; 10195d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10205d4f98a2SYan Zheng struct extent_buffer *leaf; 10215d4f98a2SYan Zheng int ret; 10225d4f98a2SYan Zheng 10235d4f98a2SYan Zheng path = btrfs_alloc_path(); 10245d4f98a2SYan Zheng if (!path) 10255d4f98a2SYan Zheng return -ENOMEM; 10265d4f98a2SYan Zheng 10275d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1028f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1029f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 10305d4f98a2SYan Zheng if (ret < 0) 10315d4f98a2SYan Zheng goto out; 10325d4f98a2SYan Zheng if (ret > 0) { 10335d4f98a2SYan Zheng ret = -ENOENT; 10345d4f98a2SYan Zheng goto out; 10355d4f98a2SYan Zheng } 10365d4f98a2SYan Zheng 10375d4f98a2SYan Zheng leaf = path->nodes[0]; 10385d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 10395d4f98a2SYan Zheng struct btrfs_file_extent_item); 10405d4f98a2SYan Zheng 10415d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 10425d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 10435d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 10445d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 10455d4f98a2SYan Zheng 10465d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 104783d4cfd4SJosef Bacik ret = -EINVAL; 10485d4f98a2SYan Zheng goto out; 10495d4f98a2SYan Zheng } 10505d4f98a2SYan Zheng 10515d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 10525d4f98a2SYan Zheng ret = 0; 10535d4f98a2SYan Zheng out: 10545d4f98a2SYan Zheng btrfs_free_path(path); 10555d4f98a2SYan Zheng return ret; 10565d4f98a2SYan Zheng } 10575d4f98a2SYan Zheng 10585d4f98a2SYan Zheng /* 10595d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 10605d4f98a2SYan Zheng * the new locations. 10615d4f98a2SYan Zheng */ 10623fd0a558SYan, Zheng static noinline_for_stack 10633fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 10645d4f98a2SYan Zheng struct reloc_control *rc, 10655d4f98a2SYan Zheng struct btrfs_root *root, 10663fd0a558SYan, Zheng struct extent_buffer *leaf) 10675d4f98a2SYan Zheng { 10680b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 10695d4f98a2SYan Zheng struct btrfs_key key; 10705d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 10715d4f98a2SYan Zheng struct inode *inode = NULL; 10725d4f98a2SYan Zheng u64 parent; 10735d4f98a2SYan Zheng u64 bytenr; 10743fd0a558SYan, Zheng u64 new_bytenr = 0; 10755d4f98a2SYan Zheng u64 num_bytes; 10765d4f98a2SYan Zheng u64 end; 10775d4f98a2SYan Zheng u32 nritems; 10785d4f98a2SYan Zheng u32 i; 107983d4cfd4SJosef Bacik int ret = 0; 10805d4f98a2SYan Zheng int first = 1; 10815d4f98a2SYan Zheng int dirty = 0; 10825d4f98a2SYan Zheng 10835d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 10845d4f98a2SYan Zheng return 0; 10855d4f98a2SYan Zheng 10865d4f98a2SYan Zheng /* reloc trees always use full backref */ 10875d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 10885d4f98a2SYan Zheng parent = leaf->start; 10895d4f98a2SYan Zheng else 10905d4f98a2SYan Zheng parent = 0; 10915d4f98a2SYan Zheng 10925d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 10935d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 109482fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 109582fa113fSQu Wenruo 10965d4f98a2SYan Zheng cond_resched(); 10975d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 10985d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 10995d4f98a2SYan Zheng continue; 11005d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 11015d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 11025d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 11035d4f98a2SYan Zheng continue; 11045d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 11055d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 11065d4f98a2SYan Zheng if (bytenr == 0) 11075d4f98a2SYan Zheng continue; 11089569cc20SQu Wenruo if (!in_range(bytenr, rc->block_group->start, 11099569cc20SQu Wenruo rc->block_group->length)) 11105d4f98a2SYan Zheng continue; 11115d4f98a2SYan Zheng 11125d4f98a2SYan Zheng /* 1113fb12489bSMatthew Wilcox (Oracle) * if we are modifying block in fs tree, wait for read_folio 11145d4f98a2SYan Zheng * to complete and drop the extent cache 11155d4f98a2SYan Zheng */ 11165d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 11175d4f98a2SYan Zheng if (first) { 11185d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11195d4f98a2SYan Zheng first = 0; 11204a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 1121e55cf7caSDavid Sterba btrfs_add_delayed_iput(BTRFS_I(inode)); 11225d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 11235d4f98a2SYan Zheng } 11244a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 11259c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 11269c5c9604SJosef Bacik 11275d4f98a2SYan Zheng end = key.offset + 11285d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 11295d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 11300b246afaSJeff Mahoney fs_info->sectorsize)); 11310b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 11325d4f98a2SYan Zheng end--; 11335d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 11349c5c9604SJosef Bacik key.offset, end, 11359c5c9604SJosef Bacik &cached_state); 11365d4f98a2SYan Zheng if (!ret) 11375d4f98a2SYan Zheng continue; 11385d4f98a2SYan Zheng 11394c0c8cfcSFilipe Manana btrfs_drop_extent_map_range(BTRFS_I(inode), 11404c0c8cfcSFilipe Manana key.offset, end, true); 11415d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 11429c5c9604SJosef Bacik key.offset, end, &cached_state); 11435d4f98a2SYan Zheng } 11445d4f98a2SYan Zheng } 11455d4f98a2SYan Zheng 11465d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 11475d4f98a2SYan Zheng bytenr, num_bytes); 114883d4cfd4SJosef Bacik if (ret) { 114983d4cfd4SJosef Bacik /* 115083d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 115183d4cfd4SJosef Bacik * in the file extent yet. 115283d4cfd4SJosef Bacik */ 115383d4cfd4SJosef Bacik break; 11543fd0a558SYan, Zheng } 11555d4f98a2SYan Zheng 11565d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 11575d4f98a2SYan Zheng dirty = 1; 11585d4f98a2SYan Zheng 11595d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 116082fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 116182fa113fSQu Wenruo num_bytes, parent); 116282fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1163f42c5da6SNikolay Borisov key.objectid, key.offset, 1164f42c5da6SNikolay Borisov root->root_key.objectid, false); 116582fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 116683d4cfd4SJosef Bacik if (ret) { 116766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 116883d4cfd4SJosef Bacik break; 116983d4cfd4SJosef Bacik } 11705d4f98a2SYan Zheng 1171ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1172ffd4bb2aSQu Wenruo num_bytes, parent); 1173ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1174f42c5da6SNikolay Borisov key.objectid, key.offset, 1175f42c5da6SNikolay Borisov root->root_key.objectid, false); 1176ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 117783d4cfd4SJosef Bacik if (ret) { 117866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 117983d4cfd4SJosef Bacik break; 118083d4cfd4SJosef Bacik } 11815d4f98a2SYan Zheng } 11825d4f98a2SYan Zheng if (dirty) 11835d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 11843fd0a558SYan, Zheng if (inode) 1185e55cf7caSDavid Sterba btrfs_add_delayed_iput(BTRFS_I(inode)); 118683d4cfd4SJosef Bacik return ret; 11875d4f98a2SYan Zheng } 11885d4f98a2SYan Zheng 11895d4f98a2SYan Zheng static noinline_for_stack 11905d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 11915d4f98a2SYan Zheng struct btrfs_path *path, int level) 11925d4f98a2SYan Zheng { 11935d4f98a2SYan Zheng struct btrfs_disk_key key1; 11945d4f98a2SYan Zheng struct btrfs_disk_key key2; 11955d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 11965d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 11975d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 11985d4f98a2SYan Zheng } 11995d4f98a2SYan Zheng 12005d4f98a2SYan Zheng /* 12015d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 12025d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 12035d4f98a2SYan Zheng * reloc tree was create can be replaced. 12045d4f98a2SYan Zheng * 12055d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 12065d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 12075d4f98a2SYan Zheng * errors, a negative error number is returned. 12085d4f98a2SYan Zheng */ 12093fd0a558SYan, Zheng static noinline_for_stack 12103d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 12115d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 12125d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 12135d4f98a2SYan Zheng int lowest_level, int max_level) 12145d4f98a2SYan Zheng { 12150b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 12165d4f98a2SYan Zheng struct extent_buffer *eb; 12175d4f98a2SYan Zheng struct extent_buffer *parent; 121882fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 12195d4f98a2SYan Zheng struct btrfs_key key; 12205d4f98a2SYan Zheng u64 old_bytenr; 12215d4f98a2SYan Zheng u64 new_bytenr; 12225d4f98a2SYan Zheng u64 old_ptr_gen; 12235d4f98a2SYan Zheng u64 new_ptr_gen; 12245d4f98a2SYan Zheng u64 last_snapshot; 12255d4f98a2SYan Zheng u32 blocksize; 12263fd0a558SYan, Zheng int cow = 0; 12275d4f98a2SYan Zheng int level; 12285d4f98a2SYan Zheng int ret; 12295d4f98a2SYan Zheng int slot; 12305d4f98a2SYan Zheng 12317a9213a9SJosef Bacik ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 12327a9213a9SJosef Bacik ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 12335d4f98a2SYan Zheng 12345d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 12353fd0a558SYan, Zheng again: 12365d4f98a2SYan Zheng slot = path->slots[lowest_level]; 12375d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 12385d4f98a2SYan Zheng 12395d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 12405d4f98a2SYan Zheng level = btrfs_header_level(eb); 12415d4f98a2SYan Zheng 12425d4f98a2SYan Zheng if (level < lowest_level) { 12435d4f98a2SYan Zheng btrfs_tree_unlock(eb); 12445d4f98a2SYan Zheng free_extent_buffer(eb); 12455d4f98a2SYan Zheng return 0; 12465d4f98a2SYan Zheng } 12475d4f98a2SYan Zheng 12483fd0a558SYan, Zheng if (cow) { 12499631e4ccSJosef Bacik ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb, 12509631e4ccSJosef Bacik BTRFS_NESTING_COW); 125145b87c5dSJosef Bacik if (ret) { 125245b87c5dSJosef Bacik btrfs_tree_unlock(eb); 125345b87c5dSJosef Bacik free_extent_buffer(eb); 125445b87c5dSJosef Bacik return ret; 125545b87c5dSJosef Bacik } 12563fd0a558SYan, Zheng } 12575d4f98a2SYan Zheng 12585d4f98a2SYan Zheng if (next_key) { 12595d4f98a2SYan Zheng next_key->objectid = (u64)-1; 12605d4f98a2SYan Zheng next_key->type = (u8)-1; 12615d4f98a2SYan Zheng next_key->offset = (u64)-1; 12625d4f98a2SYan Zheng } 12635d4f98a2SYan Zheng 12645d4f98a2SYan Zheng parent = eb; 12655d4f98a2SYan Zheng while (1) { 12665d4f98a2SYan Zheng level = btrfs_header_level(parent); 12677a9213a9SJosef Bacik ASSERT(level >= lowest_level); 12685d4f98a2SYan Zheng 1269fdf8d595SAnand Jain ret = btrfs_bin_search(parent, 0, &key, &slot); 1270cbca7d59SFilipe Manana if (ret < 0) 1271cbca7d59SFilipe Manana break; 12725d4f98a2SYan Zheng if (ret && slot > 0) 12735d4f98a2SYan Zheng slot--; 12745d4f98a2SYan Zheng 12755d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 12765d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 12775d4f98a2SYan Zheng 12785d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 12790b246afaSJeff Mahoney blocksize = fs_info->nodesize; 12805d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 12815d4f98a2SYan Zheng 12825d4f98a2SYan Zheng if (level <= max_level) { 12835d4f98a2SYan Zheng eb = path->nodes[level]; 12845d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 12855d4f98a2SYan Zheng path->slots[level]); 12865d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 12875d4f98a2SYan Zheng path->slots[level]); 12885d4f98a2SYan Zheng } else { 12895d4f98a2SYan Zheng new_bytenr = 0; 12905d4f98a2SYan Zheng new_ptr_gen = 0; 12915d4f98a2SYan Zheng } 12925d4f98a2SYan Zheng 1293fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 12945d4f98a2SYan Zheng ret = level; 12955d4f98a2SYan Zheng break; 12965d4f98a2SYan Zheng } 12975d4f98a2SYan Zheng 12985d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 12995d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 13003fd0a558SYan, Zheng if (level <= lowest_level) { 13015d4f98a2SYan Zheng ret = 0; 13025d4f98a2SYan Zheng break; 13035d4f98a2SYan Zheng } 13045d4f98a2SYan Zheng 13056b3426beSJosef Bacik eb = btrfs_read_node_slot(parent, slot); 130664c043deSLiu Bo if (IS_ERR(eb)) { 130764c043deSLiu Bo ret = PTR_ERR(eb); 1308264813acSLiu Bo break; 1309416bc658SJosef Bacik } 13105d4f98a2SYan Zheng btrfs_tree_lock(eb); 13113fd0a558SYan, Zheng if (cow) { 13125d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 13139631e4ccSJosef Bacik slot, &eb, 13149631e4ccSJosef Bacik BTRFS_NESTING_COW); 131545b87c5dSJosef Bacik if (ret) { 131645b87c5dSJosef Bacik btrfs_tree_unlock(eb); 131745b87c5dSJosef Bacik free_extent_buffer(eb); 131845b87c5dSJosef Bacik break; 131945b87c5dSJosef Bacik } 13205d4f98a2SYan Zheng } 13215d4f98a2SYan Zheng 13225d4f98a2SYan Zheng btrfs_tree_unlock(parent); 13235d4f98a2SYan Zheng free_extent_buffer(parent); 13245d4f98a2SYan Zheng 13255d4f98a2SYan Zheng parent = eb; 13265d4f98a2SYan Zheng continue; 13275d4f98a2SYan Zheng } 13285d4f98a2SYan Zheng 13293fd0a558SYan, Zheng if (!cow) { 13303fd0a558SYan, Zheng btrfs_tree_unlock(parent); 13313fd0a558SYan, Zheng free_extent_buffer(parent); 13323fd0a558SYan, Zheng cow = 1; 13333fd0a558SYan, Zheng goto again; 13343fd0a558SYan, Zheng } 13353fd0a558SYan, Zheng 13365d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 13375d4f98a2SYan Zheng path->slots[level]); 1338b3b4aa74SDavid Sterba btrfs_release_path(path); 13395d4f98a2SYan Zheng 13405d4f98a2SYan Zheng path->lowest_level = level; 1341b40130b2SJosef Bacik set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); 13425d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 1343b40130b2SJosef Bacik clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); 13445d4f98a2SYan Zheng path->lowest_level = 0; 13450e9873e2SJosef Bacik if (ret) { 13460e9873e2SJosef Bacik if (ret > 0) 13470e9873e2SJosef Bacik ret = -ENOENT; 13480e9873e2SJosef Bacik break; 13490e9873e2SJosef Bacik } 13505d4f98a2SYan Zheng 13515d4f98a2SYan Zheng /* 1352824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1353824d8dffSQu Wenruo * 1354824d8dffSQu Wenruo * We must trace both trees. 1355824d8dffSQu Wenruo * 1) Tree reloc subtree 1356824d8dffSQu Wenruo * If not traced, we will leak data numbers 1357824d8dffSQu Wenruo * 2) Fs subtree 1358824d8dffSQu Wenruo * If not traced, we will double count old data 1359f616f5cdSQu Wenruo * 1360f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1361f616f5cdSQu Wenruo * the swapped tree blocks. 1362f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1363f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1364824d8dffSQu Wenruo */ 1365370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1366370a11b8SQu Wenruo rc->block_group, parent, slot, 1367370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1368370a11b8SQu Wenruo last_snapshot); 1369370a11b8SQu Wenruo if (ret < 0) 1370370a11b8SQu Wenruo break; 1371824d8dffSQu Wenruo /* 13725d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 13735d4f98a2SYan Zheng */ 13745d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 13755d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 13765d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 13775d4f98a2SYan Zheng 13785d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 13795d4f98a2SYan Zheng path->slots[level], old_bytenr); 13805d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 13815d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 13825d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 13835d4f98a2SYan Zheng 138482fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 138582fa113fSQu Wenruo blocksize, path->nodes[level]->start); 1386f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1387f42c5da6SNikolay Borisov 0, true); 138882fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1389253e258cSJosef Bacik if (ret) { 1390253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1391253e258cSJosef Bacik break; 1392253e258cSJosef Bacik } 139382fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 139482fa113fSQu Wenruo blocksize, 0); 1395f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0, 1396f42c5da6SNikolay Borisov true); 139782fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 1398253e258cSJosef Bacik if (ret) { 1399253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1400253e258cSJosef Bacik break; 1401253e258cSJosef Bacik } 14025d4f98a2SYan Zheng 1403ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1404ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1405f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, 1406f42c5da6SNikolay Borisov 0, true); 1407ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1408253e258cSJosef Bacik if (ret) { 1409253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1410253e258cSJosef Bacik break; 1411253e258cSJosef Bacik } 14125d4f98a2SYan Zheng 1413ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1414ffd4bb2aSQu Wenruo blocksize, 0); 1415f42c5da6SNikolay Borisov btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 1416f42c5da6SNikolay Borisov 0, true); 1417ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 1418253e258cSJosef Bacik if (ret) { 1419253e258cSJosef Bacik btrfs_abort_transaction(trans, ret); 1420253e258cSJosef Bacik break; 1421253e258cSJosef Bacik } 14225d4f98a2SYan Zheng 14235d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 14245d4f98a2SYan Zheng 14255d4f98a2SYan Zheng ret = level; 14265d4f98a2SYan Zheng break; 14275d4f98a2SYan Zheng } 14285d4f98a2SYan Zheng btrfs_tree_unlock(parent); 14295d4f98a2SYan Zheng free_extent_buffer(parent); 14305d4f98a2SYan Zheng return ret; 14315d4f98a2SYan Zheng } 14325d4f98a2SYan Zheng 14335d4f98a2SYan Zheng /* 14345d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 14355d4f98a2SYan Zheng */ 14365d4f98a2SYan Zheng static noinline_for_stack 14375d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14385d4f98a2SYan Zheng int *level) 14395d4f98a2SYan Zheng { 14405d4f98a2SYan Zheng struct extent_buffer *eb; 14415d4f98a2SYan Zheng int i; 14425d4f98a2SYan Zheng u64 last_snapshot; 14435d4f98a2SYan Zheng u32 nritems; 14445d4f98a2SYan Zheng 14455d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14465d4f98a2SYan Zheng 14475d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 14485d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14495d4f98a2SYan Zheng path->nodes[i] = NULL; 14505d4f98a2SYan Zheng } 14515d4f98a2SYan Zheng 14525d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 14535d4f98a2SYan Zheng eb = path->nodes[i]; 14545d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14555d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 14565d4f98a2SYan Zheng path->slots[i]++; 14575d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 14585d4f98a2SYan Zheng last_snapshot) 14595d4f98a2SYan Zheng continue; 14605d4f98a2SYan Zheng 14615d4f98a2SYan Zheng *level = i; 14625d4f98a2SYan Zheng return 0; 14635d4f98a2SYan Zheng } 14645d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 14655d4f98a2SYan Zheng path->nodes[i] = NULL; 14665d4f98a2SYan Zheng } 14675d4f98a2SYan Zheng return 1; 14685d4f98a2SYan Zheng } 14695d4f98a2SYan Zheng 14705d4f98a2SYan Zheng /* 14715d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 14725d4f98a2SYan Zheng */ 14735d4f98a2SYan Zheng static noinline_for_stack 14745d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 14755d4f98a2SYan Zheng int *level) 14765d4f98a2SYan Zheng { 14775d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 14785d4f98a2SYan Zheng int i; 14795d4f98a2SYan Zheng u64 ptr_gen = 0; 14805d4f98a2SYan Zheng u64 last_snapshot; 14815d4f98a2SYan Zheng u32 nritems; 14825d4f98a2SYan Zheng 14835d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 14845d4f98a2SYan Zheng 14855d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 14865d4f98a2SYan Zheng eb = path->nodes[i]; 14875d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 14885d4f98a2SYan Zheng while (path->slots[i] < nritems) { 14895d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 14905d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 14915d4f98a2SYan Zheng break; 14925d4f98a2SYan Zheng path->slots[i]++; 14935d4f98a2SYan Zheng } 14945d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 14955d4f98a2SYan Zheng if (i == *level) 14965d4f98a2SYan Zheng break; 14975d4f98a2SYan Zheng *level = i + 1; 14985d4f98a2SYan Zheng return 0; 14995d4f98a2SYan Zheng } 15005d4f98a2SYan Zheng if (i == 1) { 15015d4f98a2SYan Zheng *level = i; 15025d4f98a2SYan Zheng return 0; 15035d4f98a2SYan Zheng } 15045d4f98a2SYan Zheng 15058ef385bbSJosef Bacik eb = btrfs_read_node_slot(eb, path->slots[i]); 15068ef385bbSJosef Bacik if (IS_ERR(eb)) 150764c043deSLiu Bo return PTR_ERR(eb); 15085d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 15095d4f98a2SYan Zheng path->nodes[i - 1] = eb; 15105d4f98a2SYan Zheng path->slots[i - 1] = 0; 15115d4f98a2SYan Zheng } 15125d4f98a2SYan Zheng return 1; 15135d4f98a2SYan Zheng } 15145d4f98a2SYan Zheng 15155d4f98a2SYan Zheng /* 15165d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 15175d4f98a2SYan Zheng * [min_key, max_key) 15185d4f98a2SYan Zheng */ 15195d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 15205d4f98a2SYan Zheng struct btrfs_key *min_key, 15215d4f98a2SYan Zheng struct btrfs_key *max_key) 15225d4f98a2SYan Zheng { 15230b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15245d4f98a2SYan Zheng struct inode *inode = NULL; 15255d4f98a2SYan Zheng u64 objectid; 15265d4f98a2SYan Zheng u64 start, end; 152733345d01SLi Zefan u64 ino; 15285d4f98a2SYan Zheng 15295d4f98a2SYan Zheng objectid = min_key->objectid; 15305d4f98a2SYan Zheng while (1) { 15319c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 15329c5c9604SJosef Bacik 15335d4f98a2SYan Zheng cond_resched(); 15345d4f98a2SYan Zheng iput(inode); 15355d4f98a2SYan Zheng 15365d4f98a2SYan Zheng if (objectid > max_key->objectid) 15375d4f98a2SYan Zheng break; 15385d4f98a2SYan Zheng 15395d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 15405d4f98a2SYan Zheng if (!inode) 15415d4f98a2SYan Zheng break; 15424a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 15435d4f98a2SYan Zheng 154433345d01SLi Zefan if (ino > max_key->objectid) { 15455d4f98a2SYan Zheng iput(inode); 15465d4f98a2SYan Zheng break; 15475d4f98a2SYan Zheng } 15485d4f98a2SYan Zheng 154933345d01SLi Zefan objectid = ino + 1; 15505d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 15515d4f98a2SYan Zheng continue; 15525d4f98a2SYan Zheng 155333345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 15545d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 15555d4f98a2SYan Zheng continue; 15565d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 15575d4f98a2SYan Zheng start = 0; 15585d4f98a2SYan Zheng else { 15595d4f98a2SYan Zheng start = min_key->offset; 15600b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 15615d4f98a2SYan Zheng } 15625d4f98a2SYan Zheng } else { 15635d4f98a2SYan Zheng start = 0; 15645d4f98a2SYan Zheng } 15655d4f98a2SYan Zheng 156633345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 15675d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 15685d4f98a2SYan Zheng continue; 15695d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 15705d4f98a2SYan Zheng end = (u64)-1; 15715d4f98a2SYan Zheng } else { 15725d4f98a2SYan Zheng if (max_key->offset == 0) 15735d4f98a2SYan Zheng continue; 15745d4f98a2SYan Zheng end = max_key->offset; 15750b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 15765d4f98a2SYan Zheng end--; 15775d4f98a2SYan Zheng } 15785d4f98a2SYan Zheng } else { 15795d4f98a2SYan Zheng end = (u64)-1; 15805d4f98a2SYan Zheng } 15815d4f98a2SYan Zheng 1582fb12489bSMatthew Wilcox (Oracle) /* the lock_extent waits for read_folio to complete */ 15839c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 15844c0c8cfcSFilipe Manana btrfs_drop_extent_map_range(BTRFS_I(inode), start, end, true); 15859c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 15865d4f98a2SYan Zheng } 15875d4f98a2SYan Zheng return 0; 15885d4f98a2SYan Zheng } 15895d4f98a2SYan Zheng 15905d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 15915d4f98a2SYan Zheng struct btrfs_key *key) 15925d4f98a2SYan Zheng 15935d4f98a2SYan Zheng { 15945d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 15955d4f98a2SYan Zheng if (!path->nodes[level]) 15965d4f98a2SYan Zheng break; 15975d4f98a2SYan Zheng if (path->slots[level] + 1 < 15985d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 15995d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 16005d4f98a2SYan Zheng path->slots[level] + 1); 16015d4f98a2SYan Zheng return 0; 16025d4f98a2SYan Zheng } 16035d4f98a2SYan Zheng level++; 16045d4f98a2SYan Zheng } 16055d4f98a2SYan Zheng return 1; 16065d4f98a2SYan Zheng } 16075d4f98a2SYan Zheng 16085d4f98a2SYan Zheng /* 1609d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 1610d2311e69SQu Wenruo */ 1611ac54da6cSJosef Bacik static int insert_dirty_subvol(struct btrfs_trans_handle *trans, 1612d2311e69SQu Wenruo struct reloc_control *rc, 1613d2311e69SQu Wenruo struct btrfs_root *root) 1614d2311e69SQu Wenruo { 1615d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1616d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 16177934133fSJosef Bacik int ret; 1618d2311e69SQu Wenruo 1619d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 1620d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 1621d2311e69SQu Wenruo ASSERT(reloc_root); 1622d2311e69SQu Wenruo 1623d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 1624d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 1625d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 1626c8422684SDavid Sterba btrfs_set_root_drop_level(reloc_root_item, 0); 1627d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 16287934133fSJosef Bacik ret = btrfs_update_reloc_root(trans, root); 16297934133fSJosef Bacik if (ret) 16307934133fSJosef Bacik return ret; 1631d2311e69SQu Wenruo 1632d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 163300246528SJosef Bacik btrfs_grab_root(root); 1634d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 1635d2311e69SQu Wenruo } 1636ac54da6cSJosef Bacik 1637ac54da6cSJosef Bacik return 0; 1638d2311e69SQu Wenruo } 1639d2311e69SQu Wenruo 1640d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 1641d2311e69SQu Wenruo { 1642d2311e69SQu Wenruo struct btrfs_root *root; 1643d2311e69SQu Wenruo struct btrfs_root *next; 1644d2311e69SQu Wenruo int ret = 0; 164530d40577SQu Wenruo int ret2; 1646d2311e69SQu Wenruo 1647d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 1648d2311e69SQu Wenruo reloc_dirty_list) { 164930d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 165030d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 1651d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 1652d2311e69SQu Wenruo 1653d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 1654d2311e69SQu Wenruo root->reloc_root = NULL; 16556282675eSQu Wenruo /* 16566282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 16576282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 16586282675eSQu Wenruo */ 16596282675eSQu Wenruo smp_wmb(); 16601fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 1661f28de8d8SJosef Bacik if (reloc_root) { 1662f44deb74SJosef Bacik /* 1663f44deb74SJosef Bacik * btrfs_drop_snapshot drops our ref we hold for 1664f44deb74SJosef Bacik * ->reloc_root. If it fails however we must 1665f44deb74SJosef Bacik * drop the ref ourselves. 1666f44deb74SJosef Bacik */ 1667f28de8d8SJosef Bacik ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); 1668f44deb74SJosef Bacik if (ret2 < 0) { 1669f44deb74SJosef Bacik btrfs_put_root(reloc_root); 1670f44deb74SJosef Bacik if (!ret) 1671f28de8d8SJosef Bacik ret = ret2; 1672f28de8d8SJosef Bacik } 1673f44deb74SJosef Bacik } 167400246528SJosef Bacik btrfs_put_root(root); 167530d40577SQu Wenruo } else { 167630d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 16770078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(root, 0, 1); 1678f44deb74SJosef Bacik if (ret2 < 0) { 1679f44deb74SJosef Bacik btrfs_put_root(root); 1680f44deb74SJosef Bacik if (!ret) 168130d40577SQu Wenruo ret = ret2; 168230d40577SQu Wenruo } 1683d2311e69SQu Wenruo } 1684f44deb74SJosef Bacik } 1685d2311e69SQu Wenruo return ret; 1686d2311e69SQu Wenruo } 1687d2311e69SQu Wenruo 1688d2311e69SQu Wenruo /* 16895d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 16905d4f98a2SYan Zheng * fs tree. 16915d4f98a2SYan Zheng */ 16925d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 16935d4f98a2SYan Zheng struct btrfs_root *root) 16945d4f98a2SYan Zheng { 16950b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 16965d4f98a2SYan Zheng struct btrfs_key key; 16975d4f98a2SYan Zheng struct btrfs_key next_key; 16989e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 16995d4f98a2SYan Zheng struct btrfs_root *reloc_root; 17005d4f98a2SYan Zheng struct btrfs_root_item *root_item; 17015d4f98a2SYan Zheng struct btrfs_path *path; 17023fd0a558SYan, Zheng struct extent_buffer *leaf; 1703fca3a45dSJosef Bacik int reserve_level; 17045d4f98a2SYan Zheng int level; 17055d4f98a2SYan Zheng int max_level; 17065d4f98a2SYan Zheng int replaced = 0; 1707c6a592f2SNikolay Borisov int ret = 0; 17083fd0a558SYan, Zheng u32 min_reserved; 17095d4f98a2SYan Zheng 17105d4f98a2SYan Zheng path = btrfs_alloc_path(); 17115d4f98a2SYan Zheng if (!path) 17125d4f98a2SYan Zheng return -ENOMEM; 1713e4058b54SDavid Sterba path->reada = READA_FORWARD; 17145d4f98a2SYan Zheng 17155d4f98a2SYan Zheng reloc_root = root->reloc_root; 17165d4f98a2SYan Zheng root_item = &reloc_root->root_item; 17175d4f98a2SYan Zheng 17185d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 17195d4f98a2SYan Zheng level = btrfs_root_level(root_item); 172067439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 17215d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 17225d4f98a2SYan Zheng path->slots[level] = 0; 17235d4f98a2SYan Zheng } else { 17245d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 17255d4f98a2SYan Zheng 1726c8422684SDavid Sterba level = btrfs_root_drop_level(root_item); 17275d4f98a2SYan Zheng BUG_ON(level == 0); 17285d4f98a2SYan Zheng path->lowest_level = level; 17295d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 173033c66f43SYan Zheng path->lowest_level = 0; 17315d4f98a2SYan Zheng if (ret < 0) { 17325d4f98a2SYan Zheng btrfs_free_path(path); 17335d4f98a2SYan Zheng return ret; 17345d4f98a2SYan Zheng } 17355d4f98a2SYan Zheng 17365d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 17375d4f98a2SYan Zheng path->slots[level]); 17385d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 17395d4f98a2SYan Zheng 17405d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 17415d4f98a2SYan Zheng } 17425d4f98a2SYan Zheng 174344d354abSQu Wenruo /* 174444d354abSQu Wenruo * In merge_reloc_root(), we modify the upper level pointer to swap the 174544d354abSQu Wenruo * tree blocks between reloc tree and subvolume tree. Thus for tree 174644d354abSQu Wenruo * block COW, we COW at most from level 1 to root level for each tree. 174744d354abSQu Wenruo * 174844d354abSQu Wenruo * Thus the needed metadata size is at most root_level * nodesize, 174944d354abSQu Wenruo * and * 2 since we have two trees to COW. 175044d354abSQu Wenruo */ 1751fca3a45dSJosef Bacik reserve_level = max_t(int, 1, btrfs_root_level(root_item)); 1752fca3a45dSJosef Bacik min_reserved = fs_info->nodesize * reserve_level * 2; 17535d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 17545d4f98a2SYan Zheng 17555d4f98a2SYan Zheng while (1) { 17569270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 17579270501cSJosef Bacik min_reserved, 175844d354abSQu Wenruo BTRFS_RESERVE_FLUSH_LIMIT); 1759c6a592f2SNikolay Borisov if (ret) 17609e6a0c52SJosef Bacik goto out; 17619e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 17629e6a0c52SJosef Bacik if (IS_ERR(trans)) { 1763c6a592f2SNikolay Borisov ret = PTR_ERR(trans); 17649e6a0c52SJosef Bacik trans = NULL; 17659e6a0c52SJosef Bacik goto out; 17669e6a0c52SJosef Bacik } 17672abc726aSJosef Bacik 17682abc726aSJosef Bacik /* 17692abc726aSJosef Bacik * At this point we no longer have a reloc_control, so we can't 17702abc726aSJosef Bacik * depend on btrfs_init_reloc_root to update our last_trans. 17712abc726aSJosef Bacik * 17722abc726aSJosef Bacik * But that's ok, we started the trans handle on our 17732abc726aSJosef Bacik * corresponding fs_root, which means it's been added to the 17742abc726aSJosef Bacik * dirty list. At commit time we'll still call 17752abc726aSJosef Bacik * btrfs_update_reloc_root() and update our root item 17762abc726aSJosef Bacik * appropriately. 17772abc726aSJosef Bacik */ 17782abc726aSJosef Bacik reloc_root->last_trans = trans->transid; 17799e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 17803fd0a558SYan, Zheng 17813fd0a558SYan, Zheng replaced = 0; 17825d4f98a2SYan Zheng max_level = level; 17835d4f98a2SYan Zheng 17845d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 1785c6a592f2SNikolay Borisov if (ret < 0) 17865d4f98a2SYan Zheng goto out; 17875d4f98a2SYan Zheng if (ret > 0) 17885d4f98a2SYan Zheng break; 17895d4f98a2SYan Zheng 17905d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 17915d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 17925d4f98a2SYan Zheng ret = 0; 17935d4f98a2SYan Zheng } else { 17943d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 17953fd0a558SYan, Zheng &next_key, level, max_level); 17965d4f98a2SYan Zheng } 1797c6a592f2SNikolay Borisov if (ret < 0) 17985d4f98a2SYan Zheng goto out; 17995d4f98a2SYan Zheng if (ret > 0) { 18005d4f98a2SYan Zheng level = ret; 18015d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18025d4f98a2SYan Zheng path->slots[level]); 18035d4f98a2SYan Zheng replaced = 1; 18045d4f98a2SYan Zheng } 18055d4f98a2SYan Zheng 18065d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 18075d4f98a2SYan Zheng if (ret > 0) 18085d4f98a2SYan Zheng break; 18095d4f98a2SYan Zheng 18105d4f98a2SYan Zheng BUG_ON(level == 0); 18115d4f98a2SYan Zheng /* 18125d4f98a2SYan Zheng * save the merging progress in the drop_progress. 18135d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 18145d4f98a2SYan Zheng */ 18155d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 18165d4f98a2SYan Zheng path->slots[level]); 1817c8422684SDavid Sterba btrfs_set_root_drop_level(root_item, level); 18185d4f98a2SYan Zheng 18193a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18209e6a0c52SJosef Bacik trans = NULL; 18215d4f98a2SYan Zheng 18222ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18235d4f98a2SYan Zheng 18245d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18255d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18265d4f98a2SYan Zheng } 18275d4f98a2SYan Zheng 18285d4f98a2SYan Zheng /* 18295d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 18305d4f98a2SYan Zheng * relocated and the block is tree root. 18315d4f98a2SYan Zheng */ 18325d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 18339631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf, 18349631e4ccSJosef Bacik BTRFS_NESTING_COW); 18355d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 18365d4f98a2SYan Zheng free_extent_buffer(leaf); 18375d4f98a2SYan Zheng out: 18385d4f98a2SYan Zheng btrfs_free_path(path); 18395d4f98a2SYan Zheng 1840ac54da6cSJosef Bacik if (ret == 0) { 1841ac54da6cSJosef Bacik ret = insert_dirty_subvol(trans, rc, root); 1842ac54da6cSJosef Bacik if (ret) 1843ac54da6cSJosef Bacik btrfs_abort_transaction(trans, ret); 1844ac54da6cSJosef Bacik } 18455d4f98a2SYan Zheng 18469e6a0c52SJosef Bacik if (trans) 18473a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 18485d4f98a2SYan Zheng 18492ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 18505d4f98a2SYan Zheng 18515d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 18525d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 18535d4f98a2SYan Zheng 1854c6a592f2SNikolay Borisov return ret; 18555d4f98a2SYan Zheng } 18565d4f98a2SYan Zheng 18573fd0a558SYan, Zheng static noinline_for_stack 18583fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 18595d4f98a2SYan Zheng { 18603fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 18610b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 18623fd0a558SYan, Zheng struct btrfs_root *reloc_root; 18635d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 18643fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 18653fd0a558SYan, Zheng u64 num_bytes = 0; 18663fd0a558SYan, Zheng int ret; 18673fd0a558SYan, Zheng 18680b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 18690b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 18703fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 18710b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 18727585717fSChris Mason 18733fd0a558SYan, Zheng again: 18743fd0a558SYan, Zheng if (!err) { 18753fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 18769270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes, 187708e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 18783fd0a558SYan, Zheng if (ret) 18793fd0a558SYan, Zheng err = ret; 18803fd0a558SYan, Zheng } 18813fd0a558SYan, Zheng 18827a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 18833612b495STsutomu Itoh if (IS_ERR(trans)) { 18843612b495STsutomu Itoh if (!err) 18852ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 188663f018beSNikolay Borisov num_bytes, NULL); 18873612b495STsutomu Itoh return PTR_ERR(trans); 18883612b495STsutomu Itoh } 18893fd0a558SYan, Zheng 18903fd0a558SYan, Zheng if (!err) { 18913fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 18923a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18932ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 189463f018beSNikolay Borisov num_bytes, NULL); 18953fd0a558SYan, Zheng goto again; 18963fd0a558SYan, Zheng } 18973fd0a558SYan, Zheng } 18983fd0a558SYan, Zheng 18993fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 19003fd0a558SYan, Zheng 19013fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 19023fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 19033fd0a558SYan, Zheng struct btrfs_root, root_list); 19043fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 19053fd0a558SYan, Zheng 1906a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 1907a820feb5SDavid Sterba false); 1908e0b085b0SJosef Bacik if (IS_ERR(root)) { 1909e0b085b0SJosef Bacik /* 1910e0b085b0SJosef Bacik * Even if we have an error we need this reloc root 1911e0b085b0SJosef Bacik * back on our list so we can clean up properly. 1912e0b085b0SJosef Bacik */ 1913e0b085b0SJosef Bacik list_add(&reloc_root->root_list, &reloc_roots); 1914e0b085b0SJosef Bacik btrfs_abort_transaction(trans, (int)PTR_ERR(root)); 1915e0b085b0SJosef Bacik if (!err) 1916e0b085b0SJosef Bacik err = PTR_ERR(root); 1917e0b085b0SJosef Bacik break; 1918e0b085b0SJosef Bacik } 191905d7ce50SQu Wenruo 192005d7ce50SQu Wenruo if (unlikely(root->reloc_root != reloc_root)) { 192105d7ce50SQu Wenruo if (root->reloc_root) { 192205d7ce50SQu Wenruo btrfs_err(fs_info, 192305d7ce50SQu Wenruo "reloc tree mismatch, root %lld has reloc root key (%lld %u %llu) gen %llu, expect reloc root key (%lld %u %llu) gen %llu", 192405d7ce50SQu Wenruo root->root_key.objectid, 192505d7ce50SQu Wenruo root->reloc_root->root_key.objectid, 192605d7ce50SQu Wenruo root->reloc_root->root_key.type, 192705d7ce50SQu Wenruo root->reloc_root->root_key.offset, 192805d7ce50SQu Wenruo btrfs_root_generation( 192905d7ce50SQu Wenruo &root->reloc_root->root_item), 193005d7ce50SQu Wenruo reloc_root->root_key.objectid, 193105d7ce50SQu Wenruo reloc_root->root_key.type, 193205d7ce50SQu Wenruo reloc_root->root_key.offset, 193305d7ce50SQu Wenruo btrfs_root_generation( 193405d7ce50SQu Wenruo &reloc_root->root_item)); 193505d7ce50SQu Wenruo } else { 193605d7ce50SQu Wenruo btrfs_err(fs_info, 193705d7ce50SQu Wenruo "reloc tree mismatch, root %lld has no reloc root, expect reloc root key (%lld %u %llu) gen %llu", 193805d7ce50SQu Wenruo root->root_key.objectid, 193905d7ce50SQu Wenruo reloc_root->root_key.objectid, 194005d7ce50SQu Wenruo reloc_root->root_key.type, 194105d7ce50SQu Wenruo reloc_root->root_key.offset, 194205d7ce50SQu Wenruo btrfs_root_generation( 194305d7ce50SQu Wenruo &reloc_root->root_item)); 194405d7ce50SQu Wenruo } 194505d7ce50SQu Wenruo list_add(&reloc_root->root_list, &reloc_roots); 194605d7ce50SQu Wenruo btrfs_put_root(root); 194705d7ce50SQu Wenruo btrfs_abort_transaction(trans, -EUCLEAN); 194805d7ce50SQu Wenruo if (!err) 194905d7ce50SQu Wenruo err = -EUCLEAN; 195005d7ce50SQu Wenruo break; 195105d7ce50SQu Wenruo } 19523fd0a558SYan, Zheng 19533fd0a558SYan, Zheng /* 19543fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 19553fd0a558SYan, Zheng * knows it should resumes merging 19563fd0a558SYan, Zheng */ 19573fd0a558SYan, Zheng if (!err) 19583fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 1959bbae13f8SJosef Bacik ret = btrfs_update_reloc_root(trans, root); 19603fd0a558SYan, Zheng 1961bbae13f8SJosef Bacik /* 1962bbae13f8SJosef Bacik * Even if we have an error we need this reloc root back on our 1963bbae13f8SJosef Bacik * list so we can clean up properly. 1964bbae13f8SJosef Bacik */ 19653fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 196600246528SJosef Bacik btrfs_put_root(root); 1967bbae13f8SJosef Bacik 1968bbae13f8SJosef Bacik if (ret) { 1969bbae13f8SJosef Bacik btrfs_abort_transaction(trans, ret); 1970bbae13f8SJosef Bacik if (!err) 1971bbae13f8SJosef Bacik err = ret; 1972bbae13f8SJosef Bacik break; 1973bbae13f8SJosef Bacik } 19743fd0a558SYan, Zheng } 19753fd0a558SYan, Zheng 19763fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 19773fd0a558SYan, Zheng 19783fd0a558SYan, Zheng if (!err) 1979fb686c68SJosef Bacik err = btrfs_commit_transaction(trans); 19803fd0a558SYan, Zheng else 19813a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19823fd0a558SYan, Zheng return err; 19833fd0a558SYan, Zheng } 19843fd0a558SYan, Zheng 19853fd0a558SYan, Zheng static noinline_for_stack 1986aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 1987aca1bba6SLiu Bo { 1988a7571232SNikolay Borisov struct btrfs_root *reloc_root, *tmp; 1989aca1bba6SLiu Bo 1990a7571232SNikolay Borisov list_for_each_entry_safe(reloc_root, tmp, list, root_list) 1991bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 1992aca1bba6SLiu Bo } 1993aca1bba6SLiu Bo 1994aca1bba6SLiu Bo static noinline_for_stack 199594404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 19963fd0a558SYan, Zheng { 19970b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 19985d4f98a2SYan Zheng struct btrfs_root *root; 19995d4f98a2SYan Zheng struct btrfs_root *reloc_root; 20003fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 20013fd0a558SYan, Zheng int found = 0; 2002aca1bba6SLiu Bo int ret = 0; 20033fd0a558SYan, Zheng again: 20043fd0a558SYan, Zheng root = rc->extent_root; 20057585717fSChris Mason 20067585717fSChris Mason /* 20077585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 20087585717fSChris Mason * we have to make sure nobody is in the middle of 20097585717fSChris Mason * adding their roots to the list while we are 20107585717fSChris Mason * doing this splice 20117585717fSChris Mason */ 20120b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 20133fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 20140b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 20155d4f98a2SYan Zheng 20163fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 20173fd0a558SYan, Zheng found = 1; 20183fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 20193fd0a558SYan, Zheng struct btrfs_root, root_list); 20205d4f98a2SYan Zheng 2021a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 2022a820feb5SDavid Sterba false); 20235d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 202405d7ce50SQu Wenruo if (WARN_ON(IS_ERR(root))) { 202524213fa4SJosef Bacik /* 202624213fa4SJosef Bacik * For recovery we read the fs roots on mount, 202724213fa4SJosef Bacik * and if we didn't find the root then we marked 202824213fa4SJosef Bacik * the reloc root as a garbage root. For normal 202924213fa4SJosef Bacik * relocation obviously the root should exist in 203024213fa4SJosef Bacik * memory. However there's no reason we can't 203124213fa4SJosef Bacik * handle the error properly here just in case. 203224213fa4SJosef Bacik */ 203324213fa4SJosef Bacik ret = PTR_ERR(root); 203424213fa4SJosef Bacik goto out; 203524213fa4SJosef Bacik } 203605d7ce50SQu Wenruo if (WARN_ON(root->reloc_root != reloc_root)) { 203724213fa4SJosef Bacik /* 203805d7ce50SQu Wenruo * This can happen if on-disk metadata has some 203905d7ce50SQu Wenruo * corruption, e.g. bad reloc tree key offset. 204024213fa4SJosef Bacik */ 204124213fa4SJosef Bacik ret = -EINVAL; 204224213fa4SJosef Bacik goto out; 204324213fa4SJosef Bacik } 20443fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 204500246528SJosef Bacik btrfs_put_root(root); 2046b37b39cdSJosef Bacik if (ret) { 204725e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 204825e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 204925e293c2SWang Shilong &reloc_roots); 2050aca1bba6SLiu Bo goto out; 2051b37b39cdSJosef Bacik } 20523fd0a558SYan, Zheng } else { 205351415b6cSQu Wenruo if (!IS_ERR(root)) { 205451415b6cSQu Wenruo if (root->reloc_root == reloc_root) { 205551415b6cSQu Wenruo root->reloc_root = NULL; 205651415b6cSQu Wenruo btrfs_put_root(reloc_root); 205751415b6cSQu Wenruo } 20581dae7e0eSQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, 20591dae7e0eSQu Wenruo &root->state); 206051415b6cSQu Wenruo btrfs_put_root(root); 206151415b6cSQu Wenruo } 206251415b6cSQu Wenruo 20633fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 206430d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 206530d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 206630d40577SQu Wenruo &rc->dirty_subvol_roots); 20673fd0a558SYan, Zheng } 20685d4f98a2SYan Zheng } 20695d4f98a2SYan Zheng 20703fd0a558SYan, Zheng if (found) { 20713fd0a558SYan, Zheng found = 0; 20723fd0a558SYan, Zheng goto again; 20735d4f98a2SYan Zheng } 2074aca1bba6SLiu Bo out: 2075aca1bba6SLiu Bo if (ret) { 20760b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2077aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2078467bb1d2SWang Shilong 2079467bb1d2SWang Shilong /* new reloc root may be added */ 20800b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2081467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 20820b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2083467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2084aca1bba6SLiu Bo } 2085aca1bba6SLiu Bo 20867b7b7431SJosef Bacik /* 20877b7b7431SJosef Bacik * We used to have 20887b7b7431SJosef Bacik * 20897b7b7431SJosef Bacik * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 20907b7b7431SJosef Bacik * 20917b7b7431SJosef Bacik * here, but it's wrong. If we fail to start the transaction in 20927b7b7431SJosef Bacik * prepare_to_merge() we will have only 0 ref reloc roots, none of which 20937b7b7431SJosef Bacik * have actually been removed from the reloc_root_tree rb tree. This is 20947b7b7431SJosef Bacik * fine because we're bailing here, and we hold a reference on the root 20957b7b7431SJosef Bacik * for the list that holds it, so these roots will be cleaned up when we 20967b7b7431SJosef Bacik * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root 20977b7b7431SJosef Bacik * will be cleaned up on unmount. 20987b7b7431SJosef Bacik * 20997b7b7431SJosef Bacik * The remaining nodes will be cleaned up by free_reloc_control. 21007b7b7431SJosef Bacik */ 21015d4f98a2SYan Zheng } 21025d4f98a2SYan Zheng 21035d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 21045d4f98a2SYan Zheng { 21055d4f98a2SYan Zheng struct tree_block *block; 21065d4f98a2SYan Zheng struct rb_node *rb_node; 21075d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 21085d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 21095d4f98a2SYan Zheng rb_erase(rb_node, blocks); 21105d4f98a2SYan Zheng kfree(block); 21115d4f98a2SYan Zheng } 21125d4f98a2SYan Zheng } 21135d4f98a2SYan Zheng 21145d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 21155d4f98a2SYan Zheng struct btrfs_root *reloc_root) 21165d4f98a2SYan Zheng { 21170b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 21185d4f98a2SYan Zheng struct btrfs_root *root; 2119442b1ac5SJosef Bacik int ret; 21205d4f98a2SYan Zheng 21215d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 21225d4f98a2SYan Zheng return 0; 21235d4f98a2SYan Zheng 2124a820feb5SDavid Sterba root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); 2125404bccbcSJosef Bacik 2126404bccbcSJosef Bacik /* 2127404bccbcSJosef Bacik * This should succeed, since we can't have a reloc root without having 2128404bccbcSJosef Bacik * already looked up the actual root and created the reloc root for this 2129404bccbcSJosef Bacik * root. 2130404bccbcSJosef Bacik * 2131404bccbcSJosef Bacik * However if there's some sort of corruption where we have a ref to a 2132404bccbcSJosef Bacik * reloc root without a corresponding root this could return ENOENT. 2133404bccbcSJosef Bacik */ 2134404bccbcSJosef Bacik if (IS_ERR(root)) { 2135404bccbcSJosef Bacik ASSERT(0); 2136404bccbcSJosef Bacik return PTR_ERR(root); 2137404bccbcSJosef Bacik } 2138404bccbcSJosef Bacik if (root->reloc_root != reloc_root) { 2139404bccbcSJosef Bacik ASSERT(0); 2140404bccbcSJosef Bacik btrfs_err(fs_info, 2141404bccbcSJosef Bacik "root %llu has two reloc roots associated with it", 2142404bccbcSJosef Bacik reloc_root->root_key.offset); 2143404bccbcSJosef Bacik btrfs_put_root(root); 2144404bccbcSJosef Bacik return -EUCLEAN; 2145404bccbcSJosef Bacik } 2146442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 214700246528SJosef Bacik btrfs_put_root(root); 21485d4f98a2SYan Zheng 2149442b1ac5SJosef Bacik return ret; 21505d4f98a2SYan Zheng } 21515d4f98a2SYan Zheng 21523fd0a558SYan, Zheng static noinline_for_stack 21533fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 21543fd0a558SYan, Zheng struct reloc_control *rc, 2155a26195a5SQu Wenruo struct btrfs_backref_node *node, 2156a26195a5SQu Wenruo struct btrfs_backref_edge *edges[]) 21575d4f98a2SYan Zheng { 2158a26195a5SQu Wenruo struct btrfs_backref_node *next; 21595d4f98a2SYan Zheng struct btrfs_root *root; 21603fd0a558SYan, Zheng int index = 0; 216192de551bSJosef Bacik int ret; 21623fd0a558SYan, Zheng 21635d4f98a2SYan Zheng next = node; 21645d4f98a2SYan Zheng while (1) { 21655d4f98a2SYan Zheng cond_resched(); 21665d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 21675d4f98a2SYan Zheng root = next->root; 21688ee66afeSJosef Bacik 21698ee66afeSJosef Bacik /* 21708ee66afeSJosef Bacik * If there is no root, then our references for this block are 21718ee66afeSJosef Bacik * incomplete, as we should be able to walk all the way up to a 21728ee66afeSJosef Bacik * block that is owned by a root. 21738ee66afeSJosef Bacik * 21748ee66afeSJosef Bacik * This path is only for SHAREABLE roots, so if we come upon a 21758ee66afeSJosef Bacik * non-SHAREABLE root then we have backrefs that resolve 21768ee66afeSJosef Bacik * improperly. 21778ee66afeSJosef Bacik * 21788ee66afeSJosef Bacik * Both of these cases indicate file system corruption, or a bug 21798ee66afeSJosef Bacik * in the backref walking code. 21808ee66afeSJosef Bacik */ 21818ee66afeSJosef Bacik if (!root) { 21828ee66afeSJosef Bacik ASSERT(0); 21838ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21848ee66afeSJosef Bacik "bytenr %llu doesn't have a backref path ending in a root", 21858ee66afeSJosef Bacik node->bytenr); 21868ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21878ee66afeSJosef Bacik } 21888ee66afeSJosef Bacik if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 21898ee66afeSJosef Bacik ASSERT(0); 21908ee66afeSJosef Bacik btrfs_err(trans->fs_info, 21918ee66afeSJosef Bacik "bytenr %llu has multiple refs with one ending in a non-shareable root", 21928ee66afeSJosef Bacik node->bytenr); 21938ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 21948ee66afeSJosef Bacik } 21955d4f98a2SYan Zheng 21965d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 219792de551bSJosef Bacik ret = record_reloc_root_in_trans(trans, root); 219892de551bSJosef Bacik if (ret) 219992de551bSJosef Bacik return ERR_PTR(ret); 22005d4f98a2SYan Zheng break; 22015d4f98a2SYan Zheng } 22025d4f98a2SYan Zheng 220392de551bSJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 220492de551bSJosef Bacik if (ret) 220592de551bSJosef Bacik return ERR_PTR(ret); 22063fd0a558SYan, Zheng root = root->reloc_root; 22073fd0a558SYan, Zheng 220839200e59SJosef Bacik /* 220939200e59SJosef Bacik * We could have raced with another thread which failed, so 221039200e59SJosef Bacik * root->reloc_root may not be set, return ENOENT in this case. 221139200e59SJosef Bacik */ 221239200e59SJosef Bacik if (!root) 221339200e59SJosef Bacik return ERR_PTR(-ENOENT); 221439200e59SJosef Bacik 22153fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 22168ee66afeSJosef Bacik /* 22178ee66afeSJosef Bacik * We just created the reloc root, so we shouldn't have 22188ee66afeSJosef Bacik * ->new_bytenr set and this shouldn't be in the changed 22198ee66afeSJosef Bacik * list. If it is then we have multiple roots pointing 22208ee66afeSJosef Bacik * at the same bytenr which indicates corruption, or 22218ee66afeSJosef Bacik * we've made a mistake in the backref walking code. 22228ee66afeSJosef Bacik */ 22238ee66afeSJosef Bacik ASSERT(next->new_bytenr == 0); 22248ee66afeSJosef Bacik ASSERT(list_empty(&next->list)); 22258ee66afeSJosef Bacik if (next->new_bytenr || !list_empty(&next->list)) { 22268ee66afeSJosef Bacik btrfs_err(trans->fs_info, 22278ee66afeSJosef Bacik "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu", 22288ee66afeSJosef Bacik node->bytenr, next->bytenr); 22298ee66afeSJosef Bacik return ERR_PTR(-EUCLEAN); 22308ee66afeSJosef Bacik } 22318ee66afeSJosef Bacik 22323fd0a558SYan, Zheng next->new_bytenr = root->node->start; 223300246528SJosef Bacik btrfs_put_root(next->root); 223400246528SJosef Bacik next->root = btrfs_grab_root(root); 22350b530bc5SJosef Bacik ASSERT(next->root); 22363fd0a558SYan, Zheng list_add_tail(&next->list, 22373fd0a558SYan, Zheng &rc->backref_cache.changed); 22389569cc20SQu Wenruo mark_block_processed(rc, next); 22395d4f98a2SYan Zheng break; 22405d4f98a2SYan Zheng } 22415d4f98a2SYan Zheng 22423fd0a558SYan, Zheng WARN_ON(1); 22435d4f98a2SYan Zheng root = NULL; 22445d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 22455d4f98a2SYan Zheng if (!next || next->level <= node->level) 22465d4f98a2SYan Zheng break; 22475d4f98a2SYan Zheng } 2248cbdc2ebcSJosef Bacik if (!root) { 2249cbdc2ebcSJosef Bacik /* 2250cbdc2ebcSJosef Bacik * This can happen if there's fs corruption or if there's a bug 2251cbdc2ebcSJosef Bacik * in the backref lookup code. 2252cbdc2ebcSJosef Bacik */ 2253cbdc2ebcSJosef Bacik ASSERT(0); 2254cbdc2ebcSJosef Bacik return ERR_PTR(-ENOENT); 2255cbdc2ebcSJosef Bacik } 22565d4f98a2SYan Zheng 22573fd0a558SYan, Zheng next = node; 22583fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 22593fd0a558SYan, Zheng while (1) { 22603fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 22613fd0a558SYan, Zheng if (--index < 0) 22623fd0a558SYan, Zheng break; 22633fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 22643fd0a558SYan, Zheng } 22655d4f98a2SYan Zheng return root; 22665d4f98a2SYan Zheng } 22675d4f98a2SYan Zheng 22683fd0a558SYan, Zheng /* 226992a7cc42SQu Wenruo * Select a tree root for relocation. 227092a7cc42SQu Wenruo * 227192a7cc42SQu Wenruo * Return NULL if the block is not shareable. We should use do_relocation() in 227292a7cc42SQu Wenruo * this case. 227392a7cc42SQu Wenruo * 227492a7cc42SQu Wenruo * Return a tree root pointer if the block is shareable. 227592a7cc42SQu Wenruo * Return -ENOENT if the block is root of reloc tree. 22763fd0a558SYan, Zheng */ 22775d4f98a2SYan Zheng static noinline_for_stack 2278a26195a5SQu Wenruo struct btrfs_root *select_one_root(struct btrfs_backref_node *node) 22795d4f98a2SYan Zheng { 2280a26195a5SQu Wenruo struct btrfs_backref_node *next; 22813fd0a558SYan, Zheng struct btrfs_root *root; 22823fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 2283a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 22843fd0a558SYan, Zheng int index = 0; 22853fd0a558SYan, Zheng 22863fd0a558SYan, Zheng next = node; 22873fd0a558SYan, Zheng while (1) { 22883fd0a558SYan, Zheng cond_resched(); 22893fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 22903fd0a558SYan, Zheng root = next->root; 22918717cf44SJosef Bacik 22928717cf44SJosef Bacik /* 22938717cf44SJosef Bacik * This can occur if we have incomplete extent refs leading all 22948717cf44SJosef Bacik * the way up a particular path, in this case return -EUCLEAN. 22958717cf44SJosef Bacik */ 22968717cf44SJosef Bacik if (!root) 22978717cf44SJosef Bacik return ERR_PTR(-EUCLEAN); 22983fd0a558SYan, Zheng 229992a7cc42SQu Wenruo /* No other choice for non-shareable tree */ 230092a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 23013fd0a558SYan, Zheng return root; 23023fd0a558SYan, Zheng 23033fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 23043fd0a558SYan, Zheng fs_root = root; 23053fd0a558SYan, Zheng 23063fd0a558SYan, Zheng if (next != node) 23073fd0a558SYan, Zheng return NULL; 23083fd0a558SYan, Zheng 23093fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 23103fd0a558SYan, Zheng if (!next || next->level <= node->level) 23113fd0a558SYan, Zheng break; 23123fd0a558SYan, Zheng } 23133fd0a558SYan, Zheng 23143fd0a558SYan, Zheng if (!fs_root) 23153fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 23163fd0a558SYan, Zheng return fs_root; 23175d4f98a2SYan Zheng } 23185d4f98a2SYan Zheng 23195d4f98a2SYan Zheng static noinline_for_stack 23203fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 2321a26195a5SQu Wenruo struct btrfs_backref_node *node, int reserve) 23225d4f98a2SYan Zheng { 23230b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2324a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2325a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2326a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 23273fd0a558SYan, Zheng u64 num_bytes = 0; 23283fd0a558SYan, Zheng int index = 0; 23295d4f98a2SYan Zheng 23303fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 23313fd0a558SYan, Zheng 23323fd0a558SYan, Zheng while (next) { 23333fd0a558SYan, Zheng cond_resched(); 23345d4f98a2SYan Zheng while (1) { 23353fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 23365d4f98a2SYan Zheng break; 23375d4f98a2SYan Zheng 23380b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 23393fd0a558SYan, Zheng 23403fd0a558SYan, Zheng if (list_empty(&next->upper)) 23413fd0a558SYan, Zheng break; 23423fd0a558SYan, Zheng 23433fd0a558SYan, Zheng edge = list_entry(next->upper.next, 2344a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 23453fd0a558SYan, Zheng edges[index++] = edge; 23463fd0a558SYan, Zheng next = edge->node[UPPER]; 23475d4f98a2SYan Zheng } 23483fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 23493fd0a558SYan, Zheng } 23503fd0a558SYan, Zheng return num_bytes; 23513fd0a558SYan, Zheng } 23523fd0a558SYan, Zheng 23533fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 23543fd0a558SYan, Zheng struct reloc_control *rc, 2355a26195a5SQu Wenruo struct btrfs_backref_node *node) 23563fd0a558SYan, Zheng { 23573fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2358da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23593fd0a558SYan, Zheng u64 num_bytes; 23603fd0a558SYan, Zheng int ret; 23610647bf56SWang Shilong u64 tmp; 23623fd0a558SYan, Zheng 23633fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 23643fd0a558SYan, Zheng 23653fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 23660647bf56SWang Shilong rc->reserved_bytes += num_bytes; 23678ca17f0fSJosef Bacik 23688ca17f0fSJosef Bacik /* 23698ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 23708ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 23718ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 23728ca17f0fSJosef Bacik */ 23739270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes, 23748ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 23753fd0a558SYan, Zheng if (ret) { 2376da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 23770647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 23780647bf56SWang Shilong tmp <<= 1; 23790647bf56SWang Shilong /* 23800647bf56SWang Shilong * only one thread can access block_rsv at this point, 23810647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 23820647bf56SWang Shilong * we expand more reservation size here to allow enough 238352042d8eSAndrea Gelmini * space for relocation and we will return earlier in 23840647bf56SWang Shilong * enospc case. 23850647bf56SWang Shilong */ 2386da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 23870647bf56SWang Shilong RELOCATION_RESERVED_NODES; 23888ca17f0fSJosef Bacik return -EAGAIN; 23893fd0a558SYan, Zheng } 23903fd0a558SYan, Zheng 23913fd0a558SYan, Zheng return 0; 23923fd0a558SYan, Zheng } 23933fd0a558SYan, Zheng 23945d4f98a2SYan Zheng /* 23955d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 23965d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 23975d4f98a2SYan Zheng * 23985d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 23995d4f98a2SYan Zheng * in that case this function just updates pointers. 24005d4f98a2SYan Zheng */ 24015d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 24023fd0a558SYan, Zheng struct reloc_control *rc, 2403a26195a5SQu Wenruo struct btrfs_backref_node *node, 24045d4f98a2SYan Zheng struct btrfs_key *key, 24055d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 24065d4f98a2SYan Zheng { 2407a26195a5SQu Wenruo struct btrfs_backref_node *upper; 2408a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2409a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 24105d4f98a2SYan Zheng struct btrfs_root *root; 24115d4f98a2SYan Zheng struct extent_buffer *eb; 24125d4f98a2SYan Zheng u32 blocksize; 24135d4f98a2SYan Zheng u64 bytenr; 24145d4f98a2SYan Zheng int slot; 24158df01fddSNikolay Borisov int ret = 0; 24165d4f98a2SYan Zheng 2417ffe30dd8SJosef Bacik /* 2418ffe30dd8SJosef Bacik * If we are lowest then this is the first time we're processing this 2419ffe30dd8SJosef Bacik * block, and thus shouldn't have an eb associated with it yet. 2420ffe30dd8SJosef Bacik */ 2421ffe30dd8SJosef Bacik ASSERT(!lowest || !node->eb); 24225d4f98a2SYan Zheng 24235d4f98a2SYan Zheng path->lowest_level = node->level + 1; 24243fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 24255d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 242682fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2427581c1760SQu Wenruo 24285d4f98a2SYan Zheng cond_resched(); 24295d4f98a2SYan Zheng 24305d4f98a2SYan Zheng upper = edge->node[UPPER]; 2431dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 2432cbdc2ebcSJosef Bacik if (IS_ERR(root)) { 2433cbdc2ebcSJosef Bacik ret = PTR_ERR(root); 2434cbdc2ebcSJosef Bacik goto next; 2435cbdc2ebcSJosef Bacik } 24365d4f98a2SYan Zheng 24373fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 24383fd0a558SYan, Zheng if (!lowest) { 2439fdf8d595SAnand Jain ret = btrfs_bin_search(upper->eb, 0, key, &slot); 24408df01fddSNikolay Borisov if (ret < 0) 2441cbca7d59SFilipe Manana goto next; 24423fd0a558SYan, Zheng BUG_ON(ret); 24433fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24443fd0a558SYan, Zheng if (node->eb->start == bytenr) 24453fd0a558SYan, Zheng goto next; 24463fd0a558SYan, Zheng } 2447b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 24483fd0a558SYan, Zheng } 24495d4f98a2SYan Zheng 24505d4f98a2SYan Zheng if (!upper->eb) { 24515d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 24523561b9dbSLiu Bo if (ret) { 24538df01fddSNikolay Borisov if (ret > 0) 24548df01fddSNikolay Borisov ret = -ENOENT; 24553561b9dbSLiu Bo 24563561b9dbSLiu Bo btrfs_release_path(path); 24575d4f98a2SYan Zheng break; 24585d4f98a2SYan Zheng } 24595d4f98a2SYan Zheng 24603fd0a558SYan, Zheng if (!upper->eb) { 24613fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 24623fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 24633fd0a558SYan, Zheng } else { 24643fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 24653fd0a558SYan, Zheng } 24663fd0a558SYan, Zheng 24673fd0a558SYan, Zheng upper->locked = 1; 24683fd0a558SYan, Zheng path->locks[upper->level] = 0; 24693fd0a558SYan, Zheng 24705d4f98a2SYan Zheng slot = path->slots[upper->level]; 2471b3b4aa74SDavid Sterba btrfs_release_path(path); 24725d4f98a2SYan Zheng } else { 2473fdf8d595SAnand Jain ret = btrfs_bin_search(upper->eb, 0, key, &slot); 24748df01fddSNikolay Borisov if (ret < 0) 2475cbca7d59SFilipe Manana goto next; 24765d4f98a2SYan Zheng BUG_ON(ret); 24775d4f98a2SYan Zheng } 24785d4f98a2SYan Zheng 24795d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 24803fd0a558SYan, Zheng if (lowest) { 24814547f4d8SLiu Bo if (bytenr != node->bytenr) { 24824547f4d8SLiu Bo btrfs_err(root->fs_info, 24834547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 24844547f4d8SLiu Bo bytenr, node->bytenr, slot, 24854547f4d8SLiu Bo upper->eb->start); 24868df01fddSNikolay Borisov ret = -EIO; 24874547f4d8SLiu Bo goto next; 24884547f4d8SLiu Bo } 24895d4f98a2SYan Zheng } else { 24903fd0a558SYan, Zheng if (node->eb->start == bytenr) 24913fd0a558SYan, Zheng goto next; 24925d4f98a2SYan Zheng } 24935d4f98a2SYan Zheng 2494da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 2495c9752536SJosef Bacik eb = btrfs_read_node_slot(upper->eb, slot); 249664c043deSLiu Bo if (IS_ERR(eb)) { 24978df01fddSNikolay Borisov ret = PTR_ERR(eb); 249864c043deSLiu Bo goto next; 249997d9a8a4STsutomu Itoh } 25005d4f98a2SYan Zheng btrfs_tree_lock(eb); 25015d4f98a2SYan Zheng 25025d4f98a2SYan Zheng if (!node->eb) { 25035d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 25049631e4ccSJosef Bacik slot, &eb, BTRFS_NESTING_COW); 25053fd0a558SYan, Zheng btrfs_tree_unlock(eb); 25063fd0a558SYan, Zheng free_extent_buffer(eb); 25078df01fddSNikolay Borisov if (ret < 0) 25083fd0a558SYan, Zheng goto next; 2509ffe30dd8SJosef Bacik /* 2510ffe30dd8SJosef Bacik * We've just COWed this block, it should have updated 2511ffe30dd8SJosef Bacik * the correct backref node entry. 2512ffe30dd8SJosef Bacik */ 2513ffe30dd8SJosef Bacik ASSERT(node->eb == eb); 25145d4f98a2SYan Zheng } else { 25155d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 25165d4f98a2SYan Zheng node->eb->start); 25175d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 25185d4f98a2SYan Zheng trans->transid); 25195d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 25205d4f98a2SYan Zheng 252182fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 25225d4f98a2SYan Zheng node->eb->start, blocksize, 252382fa113fSQu Wenruo upper->eb->start); 252482fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 2525f42c5da6SNikolay Borisov btrfs_header_owner(upper->eb), 2526f42c5da6SNikolay Borisov root->root_key.objectid, false); 252782fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 2528eb6b7fb4SJosef Bacik if (!ret) 2529eb6b7fb4SJosef Bacik ret = btrfs_drop_subtree(trans, root, eb, 2530eb6b7fb4SJosef Bacik upper->eb); 2531eb6b7fb4SJosef Bacik if (ret) 2532eb6b7fb4SJosef Bacik btrfs_abort_transaction(trans, ret); 25335d4f98a2SYan Zheng } 25343fd0a558SYan, Zheng next: 25353fd0a558SYan, Zheng if (!upper->pending) 2536b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(upper); 25373fd0a558SYan, Zheng else 2538b0fe7078SQu Wenruo btrfs_backref_unlock_node_buffer(upper); 25398df01fddSNikolay Borisov if (ret) 25403fd0a558SYan, Zheng break; 25415d4f98a2SYan Zheng } 25423fd0a558SYan, Zheng 25438df01fddSNikolay Borisov if (!ret && node->pending) { 2544b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 25453fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 25463fd0a558SYan, Zheng node->pending = 0; 25475d4f98a2SYan Zheng } 25483fd0a558SYan, Zheng 25495d4f98a2SYan Zheng path->lowest_level = 0; 2550ffe30dd8SJosef Bacik 2551ffe30dd8SJosef Bacik /* 2552ffe30dd8SJosef Bacik * We should have allocated all of our space in the block rsv and thus 2553ffe30dd8SJosef Bacik * shouldn't ENOSPC. 2554ffe30dd8SJosef Bacik */ 2555ffe30dd8SJosef Bacik ASSERT(ret != -ENOSPC); 25568df01fddSNikolay Borisov return ret; 25575d4f98a2SYan Zheng } 25585d4f98a2SYan Zheng 25595d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 25603fd0a558SYan, Zheng struct reloc_control *rc, 2561a26195a5SQu Wenruo struct btrfs_backref_node *node, 25625d4f98a2SYan Zheng struct btrfs_path *path) 25635d4f98a2SYan Zheng { 25645d4f98a2SYan Zheng struct btrfs_key key; 25655d4f98a2SYan Zheng 25665d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 25673fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 25685d4f98a2SYan Zheng } 25695d4f98a2SYan Zheng 25705d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 25713fd0a558SYan, Zheng struct reloc_control *rc, 25723fd0a558SYan, Zheng struct btrfs_path *path, int err) 25735d4f98a2SYan Zheng { 25743fd0a558SYan, Zheng LIST_HEAD(list); 2575a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 2576a26195a5SQu Wenruo struct btrfs_backref_node *node; 25775d4f98a2SYan Zheng int level; 25785d4f98a2SYan Zheng int ret; 25795d4f98a2SYan Zheng 25805d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 25815d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 25825d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 2583a26195a5SQu Wenruo struct btrfs_backref_node, list); 25843fd0a558SYan, Zheng list_move_tail(&node->list, &list); 25853fd0a558SYan, Zheng BUG_ON(!node->pending); 25865d4f98a2SYan Zheng 25873fd0a558SYan, Zheng if (!err) { 25883fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 25895d4f98a2SYan Zheng if (ret < 0) 25905d4f98a2SYan Zheng err = ret; 25915d4f98a2SYan Zheng } 25925d4f98a2SYan Zheng } 25933fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 25943fd0a558SYan, Zheng } 25955d4f98a2SYan Zheng return err; 25965d4f98a2SYan Zheng } 25975d4f98a2SYan Zheng 25985d4f98a2SYan Zheng /* 25995d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 26005d4f98a2SYan Zheng * as processed. 26015d4f98a2SYan Zheng */ 26025d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 2603a26195a5SQu Wenruo struct btrfs_backref_node *node) 26045d4f98a2SYan Zheng { 2605a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2606a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2607a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26085d4f98a2SYan Zheng int index = 0; 26095d4f98a2SYan Zheng 26105d4f98a2SYan Zheng while (next) { 26115d4f98a2SYan Zheng cond_resched(); 26125d4f98a2SYan Zheng while (1) { 26135d4f98a2SYan Zheng if (next->processed) 26145d4f98a2SYan Zheng break; 26155d4f98a2SYan Zheng 26169569cc20SQu Wenruo mark_block_processed(rc, next); 26175d4f98a2SYan Zheng 26185d4f98a2SYan Zheng if (list_empty(&next->upper)) 26195d4f98a2SYan Zheng break; 26205d4f98a2SYan Zheng 26215d4f98a2SYan Zheng edge = list_entry(next->upper.next, 2622a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 26235d4f98a2SYan Zheng edges[index++] = edge; 26245d4f98a2SYan Zheng next = edge->node[UPPER]; 26255d4f98a2SYan Zheng } 26265d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 26275d4f98a2SYan Zheng } 26285d4f98a2SYan Zheng } 26295d4f98a2SYan Zheng 26307476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 26315d4f98a2SYan Zheng { 2632da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 26337476dfdaSDavid Sterba 26345d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 26359655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 26365d4f98a2SYan Zheng return 1; 26375d4f98a2SYan Zheng return 0; 26385d4f98a2SYan Zheng } 26395d4f98a2SYan Zheng 26402ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 26415d4f98a2SYan Zheng struct tree_block *block) 26425d4f98a2SYan Zheng { 2643789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 2644789d6a3aSQu Wenruo .level = block->level, 2645789d6a3aSQu Wenruo .owner_root = block->owner, 2646789d6a3aSQu Wenruo .transid = block->key.offset 2647789d6a3aSQu Wenruo }; 26485d4f98a2SYan Zheng struct extent_buffer *eb; 26495d4f98a2SYan Zheng 2650789d6a3aSQu Wenruo eb = read_tree_block(fs_info, block->bytenr, &check); 26514eb150d6SQu Wenruo if (IS_ERR(eb)) 265264c043deSLiu Bo return PTR_ERR(eb); 26534eb150d6SQu Wenruo if (!extent_buffer_uptodate(eb)) { 2654416bc658SJosef Bacik free_extent_buffer(eb); 2655416bc658SJosef Bacik return -EIO; 2656416bc658SJosef Bacik } 26575d4f98a2SYan Zheng if (block->level == 0) 26585d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 26595d4f98a2SYan Zheng else 26605d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 26615d4f98a2SYan Zheng free_extent_buffer(eb); 26625d4f98a2SYan Zheng block->key_ready = 1; 26635d4f98a2SYan Zheng return 0; 26645d4f98a2SYan Zheng } 26655d4f98a2SYan Zheng 26665d4f98a2SYan Zheng /* 26675d4f98a2SYan Zheng * helper function to relocate a tree block 26685d4f98a2SYan Zheng */ 26695d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 26705d4f98a2SYan Zheng struct reloc_control *rc, 2671a26195a5SQu Wenruo struct btrfs_backref_node *node, 26725d4f98a2SYan Zheng struct btrfs_key *key, 26735d4f98a2SYan Zheng struct btrfs_path *path) 26745d4f98a2SYan Zheng { 26755d4f98a2SYan Zheng struct btrfs_root *root; 26763fd0a558SYan, Zheng int ret = 0; 26775d4f98a2SYan Zheng 26783fd0a558SYan, Zheng if (!node) 26795d4f98a2SYan Zheng return 0; 26803fd0a558SYan, Zheng 26815f6b2e5cSJosef Bacik /* 26825f6b2e5cSJosef Bacik * If we fail here we want to drop our backref_node because we are going 26835f6b2e5cSJosef Bacik * to start over and regenerate the tree for it. 26845f6b2e5cSJosef Bacik */ 26855f6b2e5cSJosef Bacik ret = reserve_metadata_space(trans, rc, node); 26865f6b2e5cSJosef Bacik if (ret) 26875f6b2e5cSJosef Bacik goto out; 26885f6b2e5cSJosef Bacik 26893fd0a558SYan, Zheng BUG_ON(node->processed); 2690147d256eSZhaolei root = select_one_root(node); 26918717cf44SJosef Bacik if (IS_ERR(root)) { 26928717cf44SJosef Bacik ret = PTR_ERR(root); 26938717cf44SJosef Bacik 26948717cf44SJosef Bacik /* See explanation in select_one_root for the -EUCLEAN case. */ 26958717cf44SJosef Bacik ASSERT(ret == -ENOENT); 26968717cf44SJosef Bacik if (ret == -ENOENT) { 26978717cf44SJosef Bacik ret = 0; 26983fd0a558SYan, Zheng update_processed_blocks(rc, node); 26998717cf44SJosef Bacik } 27003fd0a558SYan, Zheng goto out; 27015d4f98a2SYan Zheng } 27025d4f98a2SYan Zheng 27033fd0a558SYan, Zheng if (root) { 270492a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 27051c7bfa15SJosef Bacik /* 27061c7bfa15SJosef Bacik * This block was the root block of a root, and this is 27071c7bfa15SJosef Bacik * the first time we're processing the block and thus it 27081c7bfa15SJosef Bacik * should not have had the ->new_bytenr modified and 27091c7bfa15SJosef Bacik * should have not been included on the changed list. 27101c7bfa15SJosef Bacik * 27111c7bfa15SJosef Bacik * However in the case of corruption we could have 27121c7bfa15SJosef Bacik * multiple refs pointing to the same block improperly, 27131c7bfa15SJosef Bacik * and thus we would trip over these checks. ASSERT() 27141c7bfa15SJosef Bacik * for the developer case, because it could indicate a 27151c7bfa15SJosef Bacik * bug in the backref code, however error out for a 27161c7bfa15SJosef Bacik * normal user in the case of corruption. 27171c7bfa15SJosef Bacik */ 27181c7bfa15SJosef Bacik ASSERT(node->new_bytenr == 0); 27191c7bfa15SJosef Bacik ASSERT(list_empty(&node->list)); 27201c7bfa15SJosef Bacik if (node->new_bytenr || !list_empty(&node->list)) { 27211c7bfa15SJosef Bacik btrfs_err(root->fs_info, 27221c7bfa15SJosef Bacik "bytenr %llu has improper references to it", 27231c7bfa15SJosef Bacik node->bytenr); 27241c7bfa15SJosef Bacik ret = -EUCLEAN; 27251c7bfa15SJosef Bacik goto out; 27261c7bfa15SJosef Bacik } 2727d18c7bd9SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 2728d18c7bd9SJosef Bacik if (ret) 2729d18c7bd9SJosef Bacik goto out; 273039200e59SJosef Bacik /* 273139200e59SJosef Bacik * Another thread could have failed, need to check if we 273239200e59SJosef Bacik * have reloc_root actually set. 273339200e59SJosef Bacik */ 273439200e59SJosef Bacik if (!root->reloc_root) { 273539200e59SJosef Bacik ret = -ENOENT; 273639200e59SJosef Bacik goto out; 273739200e59SJosef Bacik } 27383fd0a558SYan, Zheng root = root->reloc_root; 27393fd0a558SYan, Zheng node->new_bytenr = root->node->start; 274000246528SJosef Bacik btrfs_put_root(node->root); 274100246528SJosef Bacik node->root = btrfs_grab_root(root); 27420b530bc5SJosef Bacik ASSERT(node->root); 27433fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 27443fd0a558SYan, Zheng } else { 27455d4f98a2SYan Zheng path->lowest_level = node->level; 27462bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 27472bb2e00eSFilipe Manana btrfs_reserve_chunk_metadata(trans, false); 27485d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 2749b3b4aa74SDavid Sterba btrfs_release_path(path); 27502bb2e00eSFilipe Manana if (root == root->fs_info->chunk_root) 27512bb2e00eSFilipe Manana btrfs_trans_release_chunk_metadata(trans); 27523fd0a558SYan, Zheng if (ret > 0) 27535d4f98a2SYan Zheng ret = 0; 27543fd0a558SYan, Zheng } 27553fd0a558SYan, Zheng if (!ret) 27563fd0a558SYan, Zheng update_processed_blocks(rc, node); 27573fd0a558SYan, Zheng } else { 27583fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 27593fd0a558SYan, Zheng } 27605d4f98a2SYan Zheng out: 27610647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 2762023acb07SQu Wenruo btrfs_backref_cleanup_node(&rc->backref_cache, node); 27635d4f98a2SYan Zheng return ret; 27645d4f98a2SYan Zheng } 27655d4f98a2SYan Zheng 27665d4f98a2SYan Zheng /* 27675d4f98a2SYan Zheng * relocate a list of blocks 27685d4f98a2SYan Zheng */ 27695d4f98a2SYan Zheng static noinline_for_stack 27705d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 27715d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 27725d4f98a2SYan Zheng { 27732ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2774a26195a5SQu Wenruo struct btrfs_backref_node *node; 27755d4f98a2SYan Zheng struct btrfs_path *path; 27765d4f98a2SYan Zheng struct tree_block *block; 277798ff7b94SQu Wenruo struct tree_block *next; 27785d4f98a2SYan Zheng int ret; 27795d4f98a2SYan Zheng int err = 0; 27805d4f98a2SYan Zheng 27815d4f98a2SYan Zheng path = btrfs_alloc_path(); 2782e1a12670SLiu Bo if (!path) { 2783e1a12670SLiu Bo err = -ENOMEM; 278434c2b290SDavid Sterba goto out_free_blocks; 2785e1a12670SLiu Bo } 27865d4f98a2SYan Zheng 278798ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 278898ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 27895d4f98a2SYan Zheng if (!block->key_ready) 2790f7ba2d37SJosef Bacik btrfs_readahead_tree_block(fs_info, block->bytenr, 2791f7ba2d37SJosef Bacik block->owner, 0, 27923fbaf258SJosef Bacik block->level); 27935d4f98a2SYan Zheng } 27945d4f98a2SYan Zheng 279598ff7b94SQu Wenruo /* Get first keys */ 279698ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 279734c2b290SDavid Sterba if (!block->key_ready) { 27982ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 279934c2b290SDavid Sterba if (err) 280034c2b290SDavid Sterba goto out_free_path; 280134c2b290SDavid Sterba } 28025d4f98a2SYan Zheng } 28035d4f98a2SYan Zheng 280498ff7b94SQu Wenruo /* Do tree relocation */ 280598ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 28063fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 28075d4f98a2SYan Zheng block->level, block->bytenr); 28085d4f98a2SYan Zheng if (IS_ERR(node)) { 28095d4f98a2SYan Zheng err = PTR_ERR(node); 28105d4f98a2SYan Zheng goto out; 28115d4f98a2SYan Zheng } 28125d4f98a2SYan Zheng 28135d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 28145d4f98a2SYan Zheng path); 28155d4f98a2SYan Zheng if (ret < 0) { 28165d4f98a2SYan Zheng err = ret; 281750dbbb71SJosef Bacik break; 28185d4f98a2SYan Zheng } 28195d4f98a2SYan Zheng } 28205d4f98a2SYan Zheng out: 28213fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 28225d4f98a2SYan Zheng 282334c2b290SDavid Sterba out_free_path: 28245d4f98a2SYan Zheng btrfs_free_path(path); 282534c2b290SDavid Sterba out_free_blocks: 2826e1a12670SLiu Bo free_block_list(blocks); 28275d4f98a2SYan Zheng return err; 28285d4f98a2SYan Zheng } 28295d4f98a2SYan Zheng 2830056d9becSNikolay Borisov static noinline_for_stack int prealloc_file_extent_cluster( 2831056d9becSNikolay Borisov struct btrfs_inode *inode, 2832efa56464SYan, Zheng struct file_extent_cluster *cluster) 2833efa56464SYan, Zheng { 2834efa56464SYan, Zheng u64 alloc_hint = 0; 2835efa56464SYan, Zheng u64 start; 2836efa56464SYan, Zheng u64 end; 2837056d9becSNikolay Borisov u64 offset = inode->index_cnt; 2838efa56464SYan, Zheng u64 num_bytes; 28394e9d0d01SNikolay Borisov int nr; 2840efa56464SYan, Zheng int ret = 0; 28419d9ea1e6SQu Wenruo u64 i_size = i_size_read(&inode->vfs_inode); 2842dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 2843dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 2844214e61d0SNikolay Borisov u64 cur_offset = prealloc_start; 2845efa56464SYan, Zheng 28469d9ea1e6SQu Wenruo /* 28479d9ea1e6SQu Wenruo * For subpage case, previous i_size may not be aligned to PAGE_SIZE. 28489d9ea1e6SQu Wenruo * This means the range [i_size, PAGE_END + 1) is filled with zeros by 28499d9ea1e6SQu Wenruo * btrfs_do_readpage() call of previously relocated file cluster. 28509d9ea1e6SQu Wenruo * 28519d9ea1e6SQu Wenruo * If the current cluster starts in the above range, btrfs_do_readpage() 28529d9ea1e6SQu Wenruo * will skip the read, and relocate_one_page() will later writeback 28539d9ea1e6SQu Wenruo * the padding zeros as new data, causing data corruption. 28549d9ea1e6SQu Wenruo * 28559d9ea1e6SQu Wenruo * Here we have to manually invalidate the range (i_size, PAGE_END + 1). 28569d9ea1e6SQu Wenruo */ 2857ce394a7fSYushan Zhou if (!PAGE_ALIGNED(i_size)) { 28589d9ea1e6SQu Wenruo struct address_space *mapping = inode->vfs_inode.i_mapping; 28599d9ea1e6SQu Wenruo struct btrfs_fs_info *fs_info = inode->root->fs_info; 28609d9ea1e6SQu Wenruo const u32 sectorsize = fs_info->sectorsize; 28619d9ea1e6SQu Wenruo struct page *page; 28629d9ea1e6SQu Wenruo 28639d9ea1e6SQu Wenruo ASSERT(sectorsize < PAGE_SIZE); 28649d9ea1e6SQu Wenruo ASSERT(IS_ALIGNED(i_size, sectorsize)); 28659d9ea1e6SQu Wenruo 28669d9ea1e6SQu Wenruo /* 28679d9ea1e6SQu Wenruo * Subpage can't handle page with DIRTY but without UPTODATE 28689d9ea1e6SQu Wenruo * bit as it can lead to the following deadlock: 28699d9ea1e6SQu Wenruo * 2870fb12489bSMatthew Wilcox (Oracle) * btrfs_read_folio() 28719d9ea1e6SQu Wenruo * | Page already *locked* 28729d9ea1e6SQu Wenruo * |- btrfs_lock_and_flush_ordered_range() 28739d9ea1e6SQu Wenruo * |- btrfs_start_ordered_extent() 28749d9ea1e6SQu Wenruo * |- extent_write_cache_pages() 28759d9ea1e6SQu Wenruo * |- lock_page() 28769d9ea1e6SQu Wenruo * We try to lock the page we already hold. 28779d9ea1e6SQu Wenruo * 28789d9ea1e6SQu Wenruo * Here we just writeback the whole data reloc inode, so that 28799d9ea1e6SQu Wenruo * we will be ensured to have no dirty range in the page, and 28809d9ea1e6SQu Wenruo * are safe to clear the uptodate bits. 28819d9ea1e6SQu Wenruo * 28829d9ea1e6SQu Wenruo * This shouldn't cause too much overhead, as we need to write 28839d9ea1e6SQu Wenruo * the data back anyway. 28849d9ea1e6SQu Wenruo */ 28859d9ea1e6SQu Wenruo ret = filemap_write_and_wait(mapping); 28869d9ea1e6SQu Wenruo if (ret < 0) 28879d9ea1e6SQu Wenruo return ret; 28889d9ea1e6SQu Wenruo 28899d9ea1e6SQu Wenruo clear_extent_bits(&inode->io_tree, i_size, 28909d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - 1, 28919d9ea1e6SQu Wenruo EXTENT_UPTODATE); 28929d9ea1e6SQu Wenruo page = find_lock_page(mapping, i_size >> PAGE_SHIFT); 28939d9ea1e6SQu Wenruo /* 28949d9ea1e6SQu Wenruo * If page is freed we don't need to do anything then, as we 28959d9ea1e6SQu Wenruo * will re-read the whole page anyway. 28969d9ea1e6SQu Wenruo */ 28979d9ea1e6SQu Wenruo if (page) { 28989d9ea1e6SQu Wenruo btrfs_subpage_clear_uptodate(fs_info, page, i_size, 28999d9ea1e6SQu Wenruo round_up(i_size, PAGE_SIZE) - i_size); 29009d9ea1e6SQu Wenruo unlock_page(page); 29019d9ea1e6SQu Wenruo put_page(page); 29029d9ea1e6SQu Wenruo } 29039d9ea1e6SQu Wenruo } 29049d9ea1e6SQu Wenruo 2905efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 2906056d9becSNikolay Borisov ret = btrfs_alloc_data_chunk_ondemand(inode, 2907dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 2908efa56464SYan, Zheng if (ret) 2909214e61d0SNikolay Borisov return ret; 2910efa56464SYan, Zheng 291129b6352bSDavid Sterba btrfs_inode_lock(inode, 0); 29124e9d0d01SNikolay Borisov for (nr = 0; nr < cluster->nr; nr++) { 29139c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 29149c5c9604SJosef Bacik 2915efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 2916efa56464SYan, Zheng if (nr + 1 < cluster->nr) 2917efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 2918efa56464SYan, Zheng else 2919efa56464SYan, Zheng end = cluster->end - offset; 2920efa56464SYan, Zheng 29219c5c9604SJosef Bacik lock_extent(&inode->io_tree, start, end, &cached_state); 2922efa56464SYan, Zheng num_bytes = end + 1 - start; 2923056d9becSNikolay Borisov ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start, 2924efa56464SYan, Zheng num_bytes, num_bytes, 2925efa56464SYan, Zheng end + 1, &alloc_hint); 292618513091SWang Xiaoguang cur_offset = end + 1; 29279c5c9604SJosef Bacik unlock_extent(&inode->io_tree, start, end, &cached_state); 2928efa56464SYan, Zheng if (ret) 2929efa56464SYan, Zheng break; 2930efa56464SYan, Zheng } 2931e5d4d75bSDavid Sterba btrfs_inode_unlock(inode, 0); 2932214e61d0SNikolay Borisov 293318513091SWang Xiaoguang if (cur_offset < prealloc_end) 2934056d9becSNikolay Borisov btrfs_free_reserved_data_space_noquota(inode->root->fs_info, 2935a89ef455SFilipe Manana prealloc_end + 1 - cur_offset); 2936efa56464SYan, Zheng return ret; 2937efa56464SYan, Zheng } 2938efa56464SYan, Zheng 29394b01c44fSJohannes Thumshirn static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode, 29404b01c44fSJohannes Thumshirn u64 start, u64 end, u64 block_start) 29415d4f98a2SYan Zheng { 29425d4f98a2SYan Zheng struct extent_map *em; 29439c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 29440257bb82SYan, Zheng int ret = 0; 29455d4f98a2SYan Zheng 2946172ddd60SDavid Sterba em = alloc_extent_map(); 29470257bb82SYan, Zheng if (!em) 29480257bb82SYan, Zheng return -ENOMEM; 29490257bb82SYan, Zheng 29505d4f98a2SYan Zheng em->start = start; 29510257bb82SYan, Zheng em->len = end + 1 - start; 29520257bb82SYan, Zheng em->block_len = em->len; 29530257bb82SYan, Zheng em->block_start = block_start; 29545d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 29555d4f98a2SYan Zheng 29569c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 2957a1ba4c08SFilipe Manana ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, false); 29589c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); 2959a1ba4c08SFilipe Manana free_extent_map(em); 2960a1ba4c08SFilipe Manana 29610257bb82SYan, Zheng return ret; 29620257bb82SYan, Zheng } 29635d4f98a2SYan Zheng 2964726a3421SQu Wenruo /* 2965907d2710SDavid Sterba * Allow error injection to test balance/relocation cancellation 2966726a3421SQu Wenruo */ 29671fec12a5SJosef Bacik noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 2968726a3421SQu Wenruo { 29695cb502f4SQu Wenruo return atomic_read(&fs_info->balance_cancel_req) || 2970907d2710SDavid Sterba atomic_read(&fs_info->reloc_cancel_req) || 29715cb502f4SQu Wenruo fatal_signal_pending(current); 2972726a3421SQu Wenruo } 2973726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 2974726a3421SQu Wenruo 2975c2832898SQu Wenruo static u64 get_cluster_boundary_end(struct file_extent_cluster *cluster, 2976c2832898SQu Wenruo int cluster_nr) 2977c2832898SQu Wenruo { 2978c2832898SQu Wenruo /* Last extent, use cluster end directly */ 2979c2832898SQu Wenruo if (cluster_nr >= cluster->nr - 1) 2980c2832898SQu Wenruo return cluster->end; 2981c2832898SQu Wenruo 2982c2832898SQu Wenruo /* Use next boundary start*/ 2983c2832898SQu Wenruo return cluster->boundary[cluster_nr + 1] - 1; 2984c2832898SQu Wenruo } 2985c2832898SQu Wenruo 2986f47960f4SQu Wenruo static int relocate_one_page(struct inode *inode, struct file_ra_state *ra, 2987f47960f4SQu Wenruo struct file_extent_cluster *cluster, 2988f47960f4SQu Wenruo int *cluster_nr, unsigned long page_index) 2989f47960f4SQu Wenruo { 2990f47960f4SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 2991f47960f4SQu Wenruo u64 offset = BTRFS_I(inode)->index_cnt; 2992f47960f4SQu Wenruo const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT; 2993f47960f4SQu Wenruo gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 2994f47960f4SQu Wenruo struct page *page; 2995f47960f4SQu Wenruo u64 page_start; 2996f47960f4SQu Wenruo u64 page_end; 2997c2832898SQu Wenruo u64 cur; 2998f47960f4SQu Wenruo int ret; 2999f47960f4SQu Wenruo 3000f47960f4SQu Wenruo ASSERT(page_index <= last_index); 3001f47960f4SQu Wenruo page = find_lock_page(inode->i_mapping, page_index); 3002f47960f4SQu Wenruo if (!page) { 3003f47960f4SQu Wenruo page_cache_sync_readahead(inode->i_mapping, ra, NULL, 3004f47960f4SQu Wenruo page_index, last_index + 1 - page_index); 3005f47960f4SQu Wenruo page = find_or_create_page(inode->i_mapping, page_index, mask); 3006c2832898SQu Wenruo if (!page) 3007c2832898SQu Wenruo return -ENOMEM; 3008f47960f4SQu Wenruo } 3009f47960f4SQu Wenruo 3010f47960f4SQu Wenruo if (PageReadahead(page)) 30112ebdd1dfSMatthew Wilcox (Oracle) page_cache_async_readahead(inode->i_mapping, ra, NULL, 30122ebdd1dfSMatthew Wilcox (Oracle) page_folio(page), page_index, 30132ebdd1dfSMatthew Wilcox (Oracle) last_index + 1 - page_index); 3014f47960f4SQu Wenruo 3015f47960f4SQu Wenruo if (!PageUptodate(page)) { 3016fb12489bSMatthew Wilcox (Oracle) btrfs_read_folio(NULL, page_folio(page)); 3017f47960f4SQu Wenruo lock_page(page); 3018f47960f4SQu Wenruo if (!PageUptodate(page)) { 3019f47960f4SQu Wenruo ret = -EIO; 3020f47960f4SQu Wenruo goto release_page; 3021f47960f4SQu Wenruo } 3022f47960f4SQu Wenruo } 3023f47960f4SQu Wenruo 3024e7f1326cSJosef Bacik /* 3025e7f1326cSJosef Bacik * We could have lost page private when we dropped the lock to read the 3026e7f1326cSJosef Bacik * page above, make sure we set_page_extent_mapped here so we have any 3027e7f1326cSJosef Bacik * of the subpage blocksize stuff we need in place. 3028e7f1326cSJosef Bacik */ 3029e7f1326cSJosef Bacik ret = set_page_extent_mapped(page); 3030e7f1326cSJosef Bacik if (ret < 0) 3031e7f1326cSJosef Bacik goto release_page; 3032e7f1326cSJosef Bacik 3033f47960f4SQu Wenruo page_start = page_offset(page); 3034f47960f4SQu Wenruo page_end = page_start + PAGE_SIZE - 1; 3035f47960f4SQu Wenruo 3036c2832898SQu Wenruo /* 3037c2832898SQu Wenruo * Start from the cluster, as for subpage case, the cluster can start 3038c2832898SQu Wenruo * inside the page. 3039c2832898SQu Wenruo */ 3040c2832898SQu Wenruo cur = max(page_start, cluster->boundary[*cluster_nr] - offset); 3041c2832898SQu Wenruo while (cur <= page_end) { 30429c5c9604SJosef Bacik struct extent_state *cached_state = NULL; 3043c2832898SQu Wenruo u64 extent_start = cluster->boundary[*cluster_nr] - offset; 3044c2832898SQu Wenruo u64 extent_end = get_cluster_boundary_end(cluster, 3045c2832898SQu Wenruo *cluster_nr) - offset; 3046c2832898SQu Wenruo u64 clamped_start = max(page_start, extent_start); 3047c2832898SQu Wenruo u64 clamped_end = min(page_end, extent_end); 3048c2832898SQu Wenruo u32 clamped_len = clamped_end + 1 - clamped_start; 3049f47960f4SQu Wenruo 3050c2832898SQu Wenruo /* Reserve metadata for this range */ 3051c2832898SQu Wenruo ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 3052d4135134SFilipe Manana clamped_len, clamped_len, 3053d4135134SFilipe Manana false); 3054c2832898SQu Wenruo if (ret) 3055f47960f4SQu Wenruo goto release_page; 3056f47960f4SQu Wenruo 3057c2832898SQu Wenruo /* Mark the range delalloc and dirty for later writeback */ 30589c5c9604SJosef Bacik lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, 30599c5c9604SJosef Bacik &cached_state); 3060c2832898SQu Wenruo ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start, 30619c5c9604SJosef Bacik clamped_end, 0, &cached_state); 3062c2832898SQu Wenruo if (ret) { 30639c5c9604SJosef Bacik clear_extent_bit(&BTRFS_I(inode)->io_tree, 3064c2832898SQu Wenruo clamped_start, clamped_end, 30659c5c9604SJosef Bacik EXTENT_LOCKED | EXTENT_BOUNDARY, 30669c5c9604SJosef Bacik &cached_state); 3067c2832898SQu Wenruo btrfs_delalloc_release_metadata(BTRFS_I(inode), 3068c2832898SQu Wenruo clamped_len, true); 3069c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), 3070c2832898SQu Wenruo clamped_len); 3071c2832898SQu Wenruo goto release_page; 3072f47960f4SQu Wenruo } 3073c2832898SQu Wenruo btrfs_page_set_dirty(fs_info, page, clamped_start, clamped_len); 3074f47960f4SQu Wenruo 3075c2832898SQu Wenruo /* 3076c2832898SQu Wenruo * Set the boundary if it's inside the page. 3077c2832898SQu Wenruo * Data relocation requires the destination extents to have the 3078c2832898SQu Wenruo * same size as the source. 3079c2832898SQu Wenruo * EXTENT_BOUNDARY bit prevents current extent from being merged 3080c2832898SQu Wenruo * with previous extent. 3081c2832898SQu Wenruo */ 3082c2832898SQu Wenruo if (in_range(cluster->boundary[*cluster_nr] - offset, 3083c2832898SQu Wenruo page_start, PAGE_SIZE)) { 3084c2832898SQu Wenruo u64 boundary_start = cluster->boundary[*cluster_nr] - 3085c2832898SQu Wenruo offset; 3086c2832898SQu Wenruo u64 boundary_end = boundary_start + 3087c2832898SQu Wenruo fs_info->sectorsize - 1; 3088c2832898SQu Wenruo 30890acd32c2SDavid Sterba set_extent_bit(&BTRFS_I(inode)->io_tree, 3090c2832898SQu Wenruo boundary_start, boundary_end, 30911d126800SDavid Sterba EXTENT_BOUNDARY, NULL); 3092c2832898SQu Wenruo } 30939c5c9604SJosef Bacik unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, 30949c5c9604SJosef Bacik &cached_state); 3095c2832898SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len); 3096c2832898SQu Wenruo cur += clamped_len; 3097c2832898SQu Wenruo 3098c2832898SQu Wenruo /* Crossed extent end, go to next extent */ 3099c2832898SQu Wenruo if (cur >= extent_end) { 3100c2832898SQu Wenruo (*cluster_nr)++; 3101c2832898SQu Wenruo /* Just finished the last extent of the cluster, exit. */ 3102c2832898SQu Wenruo if (*cluster_nr >= cluster->nr) 3103c2832898SQu Wenruo break; 3104c2832898SQu Wenruo } 3105c2832898SQu Wenruo } 3106f47960f4SQu Wenruo unlock_page(page); 3107f47960f4SQu Wenruo put_page(page); 3108f47960f4SQu Wenruo 3109f47960f4SQu Wenruo balance_dirty_pages_ratelimited(inode->i_mapping); 3110f47960f4SQu Wenruo btrfs_throttle(fs_info); 3111f47960f4SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) 3112f47960f4SQu Wenruo ret = -ECANCELED; 3113f47960f4SQu Wenruo return ret; 3114f47960f4SQu Wenruo 3115f47960f4SQu Wenruo release_page: 3116f47960f4SQu Wenruo unlock_page(page); 3117f47960f4SQu Wenruo put_page(page); 3118f47960f4SQu Wenruo return ret; 3119f47960f4SQu Wenruo } 3120f47960f4SQu Wenruo 31210257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 31220257bb82SYan, Zheng struct file_extent_cluster *cluster) 31230257bb82SYan, Zheng { 31240257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 31250257bb82SYan, Zheng unsigned long index; 31260257bb82SYan, Zheng unsigned long last_index; 31270257bb82SYan, Zheng struct file_ra_state *ra; 3128f47960f4SQu Wenruo int cluster_nr = 0; 31290257bb82SYan, Zheng int ret = 0; 31300257bb82SYan, Zheng 31310257bb82SYan, Zheng if (!cluster->nr) 31320257bb82SYan, Zheng return 0; 31330257bb82SYan, Zheng 31340257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 31350257bb82SYan, Zheng if (!ra) 31360257bb82SYan, Zheng return -ENOMEM; 31370257bb82SYan, Zheng 3138056d9becSNikolay Borisov ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster); 31390257bb82SYan, Zheng if (ret) 3140efa56464SYan, Zheng goto out; 31410257bb82SYan, Zheng 31420257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 31430257bb82SYan, Zheng 31444b01c44fSJohannes Thumshirn ret = setup_relocation_extent_mapping(inode, cluster->start - offset, 3145efa56464SYan, Zheng cluster->end - offset, cluster->start); 3146efa56464SYan, Zheng if (ret) 3147efa56464SYan, Zheng goto out; 3148efa56464SYan, Zheng 314909cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 3150f47960f4SQu Wenruo for (index = (cluster->start - offset) >> PAGE_SHIFT; 3151f47960f4SQu Wenruo index <= last_index && !ret; index++) 3152f47960f4SQu Wenruo ret = relocate_one_page(inode, ra, cluster, &cluster_nr, index); 3153f47960f4SQu Wenruo if (ret == 0) 3154f47960f4SQu Wenruo WARN_ON(cluster_nr != cluster->nr); 3155efa56464SYan, Zheng out: 31560257bb82SYan, Zheng kfree(ra); 31570257bb82SYan, Zheng return ret; 31580257bb82SYan, Zheng } 31590257bb82SYan, Zheng 31600257bb82SYan, Zheng static noinline_for_stack 31610257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 31620257bb82SYan, Zheng struct file_extent_cluster *cluster) 31630257bb82SYan, Zheng { 31640257bb82SYan, Zheng int ret; 31650257bb82SYan, Zheng 31660257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 31670257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31680257bb82SYan, Zheng if (ret) 31690257bb82SYan, Zheng return ret; 31700257bb82SYan, Zheng cluster->nr = 0; 31710257bb82SYan, Zheng } 31720257bb82SYan, Zheng 31730257bb82SYan, Zheng if (!cluster->nr) 31740257bb82SYan, Zheng cluster->start = extent_key->objectid; 31750257bb82SYan, Zheng else 31760257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 31770257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 31780257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 31790257bb82SYan, Zheng cluster->nr++; 31800257bb82SYan, Zheng 31810257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 31820257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 31830257bb82SYan, Zheng if (ret) 31840257bb82SYan, Zheng return ret; 31850257bb82SYan, Zheng cluster->nr = 0; 31860257bb82SYan, Zheng } 31870257bb82SYan, Zheng return 0; 31885d4f98a2SYan Zheng } 31895d4f98a2SYan Zheng 31905d4f98a2SYan Zheng /* 31915d4f98a2SYan Zheng * helper to add a tree block to the list. 31925d4f98a2SYan Zheng * the major work is getting the generation and level of the block 31935d4f98a2SYan Zheng */ 31945d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 31955d4f98a2SYan Zheng struct btrfs_key *extent_key, 31965d4f98a2SYan Zheng struct btrfs_path *path, 31975d4f98a2SYan Zheng struct rb_root *blocks) 31985d4f98a2SYan Zheng { 31995d4f98a2SYan Zheng struct extent_buffer *eb; 32005d4f98a2SYan Zheng struct btrfs_extent_item *ei; 32015d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 32025d4f98a2SYan Zheng struct tree_block *block; 32035d4f98a2SYan Zheng struct rb_node *rb_node; 32045d4f98a2SYan Zheng u32 item_size; 32055d4f98a2SYan Zheng int level = -1; 32067fdf4b60SWang Shilong u64 generation; 3207f7ba2d37SJosef Bacik u64 owner = 0; 32085d4f98a2SYan Zheng 32095d4f98a2SYan Zheng eb = path->nodes[0]; 32103212fa14SJosef Bacik item_size = btrfs_item_size(eb, path->slots[0]); 32115d4f98a2SYan Zheng 32123173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 32133173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 3214f7ba2d37SJosef Bacik unsigned long ptr = 0, end; 3215f7ba2d37SJosef Bacik 32165d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 32175d4f98a2SYan Zheng struct btrfs_extent_item); 3218f7ba2d37SJosef Bacik end = (unsigned long)ei + item_size; 32193173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 32205d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 32215d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 3222f7ba2d37SJosef Bacik ptr = (unsigned long)(bi + 1); 32235d4f98a2SYan Zheng } else { 32243173a18fSJosef Bacik level = (int)extent_key->offset; 3225f7ba2d37SJosef Bacik ptr = (unsigned long)(ei + 1); 32263173a18fSJosef Bacik } 32273173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 3228f7ba2d37SJosef Bacik 3229f7ba2d37SJosef Bacik /* 3230f7ba2d37SJosef Bacik * We're reading random blocks without knowing their owner ahead 3231f7ba2d37SJosef Bacik * of time. This is ok most of the time, as all reloc roots and 3232f7ba2d37SJosef Bacik * fs roots have the same lock type. However normal trees do 3233f7ba2d37SJosef Bacik * not, and the only way to know ahead of time is to read the 3234f7ba2d37SJosef Bacik * inline ref offset. We know it's an fs root if 3235f7ba2d37SJosef Bacik * 3236f7ba2d37SJosef Bacik * 1. There's more than one ref. 3237f7ba2d37SJosef Bacik * 2. There's a SHARED_DATA_REF_KEY set. 3238f7ba2d37SJosef Bacik * 3. FULL_BACKREF is set on the flags. 3239f7ba2d37SJosef Bacik * 3240f7ba2d37SJosef Bacik * Otherwise it's safe to assume that the ref offset == the 3241f7ba2d37SJosef Bacik * owner of this block, so we can use that when calling 3242f7ba2d37SJosef Bacik * read_tree_block. 3243f7ba2d37SJosef Bacik */ 3244f7ba2d37SJosef Bacik if (btrfs_extent_refs(eb, ei) == 1 && 3245f7ba2d37SJosef Bacik !(btrfs_extent_flags(eb, ei) & 3246f7ba2d37SJosef Bacik BTRFS_BLOCK_FLAG_FULL_BACKREF) && 3247f7ba2d37SJosef Bacik ptr < end) { 3248f7ba2d37SJosef Bacik struct btrfs_extent_inline_ref *iref; 3249f7ba2d37SJosef Bacik int type; 3250f7ba2d37SJosef Bacik 3251f7ba2d37SJosef Bacik iref = (struct btrfs_extent_inline_ref *)ptr; 3252f7ba2d37SJosef Bacik type = btrfs_get_extent_inline_ref_type(eb, iref, 3253f7ba2d37SJosef Bacik BTRFS_REF_TYPE_BLOCK); 3254f7ba2d37SJosef Bacik if (type == BTRFS_REF_TYPE_INVALID) 3255f7ba2d37SJosef Bacik return -EINVAL; 3256f7ba2d37SJosef Bacik if (type == BTRFS_TREE_BLOCK_REF_KEY) 3257f7ba2d37SJosef Bacik owner = btrfs_extent_inline_ref_offset(eb, iref); 3258f7ba2d37SJosef Bacik } 32593173a18fSJosef Bacik } else { 3260*182741d2SQu Wenruo btrfs_print_leaf(eb); 3261*182741d2SQu Wenruo btrfs_err(rc->block_group->fs_info, 3262*182741d2SQu Wenruo "unrecognized tree backref at tree block %llu slot %u", 3263*182741d2SQu Wenruo eb->start, path->slots[0]); 3264*182741d2SQu Wenruo btrfs_release_path(path); 3265*182741d2SQu Wenruo return -EUCLEAN; 32665d4f98a2SYan Zheng } 32675d4f98a2SYan Zheng 3268b3b4aa74SDavid Sterba btrfs_release_path(path); 32695d4f98a2SYan Zheng 32705d4f98a2SYan Zheng BUG_ON(level == -1); 32715d4f98a2SYan Zheng 32725d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 32735d4f98a2SYan Zheng if (!block) 32745d4f98a2SYan Zheng return -ENOMEM; 32755d4f98a2SYan Zheng 32765d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3277da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 32785d4f98a2SYan Zheng block->key.offset = generation; 32795d4f98a2SYan Zheng block->level = level; 32805d4f98a2SYan Zheng block->key_ready = 0; 3281f7ba2d37SJosef Bacik block->owner = owner; 32825d4f98a2SYan Zheng 3283e9a28dc5SQu Wenruo rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node); 328443c04fb1SJeff Mahoney if (rb_node) 3285982c92cbSQu Wenruo btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr, 3286982c92cbSQu Wenruo -EEXIST); 32875d4f98a2SYan Zheng 32885d4f98a2SYan Zheng return 0; 32895d4f98a2SYan Zheng } 32905d4f98a2SYan Zheng 32915d4f98a2SYan Zheng /* 32925d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 32935d4f98a2SYan Zheng */ 32945d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 32955d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 32965d4f98a2SYan Zheng struct rb_root *blocks) 32975d4f98a2SYan Zheng { 32980b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 32995d4f98a2SYan Zheng struct btrfs_path *path; 33005d4f98a2SYan Zheng struct btrfs_key key; 33015d4f98a2SYan Zheng int ret; 33020b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 33035d4f98a2SYan Zheng 33047476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 33055d4f98a2SYan Zheng return 0; 33065d4f98a2SYan Zheng 3307e9a28dc5SQu Wenruo if (rb_simple_search(blocks, bytenr)) 33085d4f98a2SYan Zheng return 0; 33095d4f98a2SYan Zheng 33105d4f98a2SYan Zheng path = btrfs_alloc_path(); 33115d4f98a2SYan Zheng if (!path) 33125d4f98a2SYan Zheng return -ENOMEM; 3313aee68ee5SJosef Bacik again: 33145d4f98a2SYan Zheng key.objectid = bytenr; 3315aee68ee5SJosef Bacik if (skinny) { 3316aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3317aee68ee5SJosef Bacik key.offset = (u64)-1; 3318aee68ee5SJosef Bacik } else { 33195d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 33205d4f98a2SYan Zheng key.offset = blocksize; 3321aee68ee5SJosef Bacik } 33225d4f98a2SYan Zheng 33235d4f98a2SYan Zheng path->search_commit_root = 1; 33245d4f98a2SYan Zheng path->skip_locking = 1; 33255d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 33265d4f98a2SYan Zheng if (ret < 0) 33275d4f98a2SYan Zheng goto out; 33285d4f98a2SYan Zheng 3329aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3330aee68ee5SJosef Bacik if (path->slots[0]) { 3331aee68ee5SJosef Bacik path->slots[0]--; 3332aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3333aee68ee5SJosef Bacik path->slots[0]); 33343173a18fSJosef Bacik if (key.objectid == bytenr && 3335aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3336aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3337aee68ee5SJosef Bacik key.offset == blocksize))) 33383173a18fSJosef Bacik ret = 0; 33393173a18fSJosef Bacik } 3340aee68ee5SJosef Bacik 3341aee68ee5SJosef Bacik if (ret) { 3342aee68ee5SJosef Bacik skinny = false; 3343aee68ee5SJosef Bacik btrfs_release_path(path); 3344aee68ee5SJosef Bacik goto again; 3345aee68ee5SJosef Bacik } 3346aee68ee5SJosef Bacik } 3347cdccee99SLiu Bo if (ret) { 3348cdccee99SLiu Bo ASSERT(ret == 1); 3349cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3350cdccee99SLiu Bo btrfs_err(fs_info, 3351cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3352cdccee99SLiu Bo bytenr); 3353cdccee99SLiu Bo WARN_ON(1); 3354cdccee99SLiu Bo ret = -EINVAL; 3355cdccee99SLiu Bo goto out; 3356cdccee99SLiu Bo } 33573173a18fSJosef Bacik 33585d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 33595d4f98a2SYan Zheng out: 33605d4f98a2SYan Zheng btrfs_free_path(path); 33615d4f98a2SYan Zheng return ret; 33625d4f98a2SYan Zheng } 33635d4f98a2SYan Zheng 33640af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 336532da5386SDavid Sterba struct btrfs_block_group *block_group, 33661bbc621eSChris Mason struct inode *inode, 33671bbc621eSChris Mason u64 ino) 33680af3d00bSJosef Bacik { 33690af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 33700af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 33710af3d00bSJosef Bacik int ret = 0; 33720af3d00bSJosef Bacik 33730af3d00bSJosef Bacik if (inode) 33740af3d00bSJosef Bacik goto truncate; 33750af3d00bSJosef Bacik 33760202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, ino, root); 33772e19f1f9SAl Viro if (IS_ERR(inode)) 33780af3d00bSJosef Bacik return -ENOENT; 33790af3d00bSJosef Bacik 33800af3d00bSJosef Bacik truncate: 33812ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 33827b61cd92SMiao Xie &fs_info->global_block_rsv); 33837b61cd92SMiao Xie if (ret) 33847b61cd92SMiao Xie goto out; 33857b61cd92SMiao Xie 33867a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 33870af3d00bSJosef Bacik if (IS_ERR(trans)) { 33883612b495STsutomu Itoh ret = PTR_ERR(trans); 33890af3d00bSJosef Bacik goto out; 33900af3d00bSJosef Bacik } 33910af3d00bSJosef Bacik 339277ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 33930af3d00bSJosef Bacik 33943a45bb20SJeff Mahoney btrfs_end_transaction(trans); 33952ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 33960af3d00bSJosef Bacik out: 33970af3d00bSJosef Bacik iput(inode); 33980af3d00bSJosef Bacik return ret; 33990af3d00bSJosef Bacik } 34000af3d00bSJosef Bacik 34015d4f98a2SYan Zheng /* 340219b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 340319b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 34045d4f98a2SYan Zheng */ 340519b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 340619b546d7SQu Wenruo struct btrfs_block_group *block_group, 340719b546d7SQu Wenruo u64 data_bytenr) 34085d4f98a2SYan Zheng { 340919b546d7SQu Wenruo u64 space_cache_ino; 341019b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 34115d4f98a2SYan Zheng struct btrfs_key key; 341219b546d7SQu Wenruo bool found = false; 341319b546d7SQu Wenruo int i; 34145d4f98a2SYan Zheng int ret; 34155d4f98a2SYan Zheng 341619b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 341719b546d7SQu Wenruo return 0; 34185d4f98a2SYan Zheng 341919b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 342050e31ef4SQu Wenruo u8 type; 342150e31ef4SQu Wenruo 342219b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 342319b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 342419b546d7SQu Wenruo continue; 342519b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 342650e31ef4SQu Wenruo type = btrfs_file_extent_type(leaf, ei); 342750e31ef4SQu Wenruo 342850e31ef4SQu Wenruo if ((type == BTRFS_FILE_EXTENT_REG || 342950e31ef4SQu Wenruo type == BTRFS_FILE_EXTENT_PREALLOC) && 343019b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 343119b546d7SQu Wenruo found = true; 343219b546d7SQu Wenruo space_cache_ino = key.objectid; 343319b546d7SQu Wenruo break; 343419b546d7SQu Wenruo } 343519b546d7SQu Wenruo } 343619b546d7SQu Wenruo if (!found) 343719b546d7SQu Wenruo return -ENOENT; 343819b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 343919b546d7SQu Wenruo space_cache_ino); 34400af3d00bSJosef Bacik return ret; 34415d4f98a2SYan Zheng } 34425d4f98a2SYan Zheng 34435d4f98a2SYan Zheng /* 34442c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 34455d4f98a2SYan Zheng */ 34465d4f98a2SYan Zheng static noinline_for_stack 34475d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 34485d4f98a2SYan Zheng struct btrfs_key *extent_key, 34495d4f98a2SYan Zheng struct btrfs_path *path, 34505d4f98a2SYan Zheng struct rb_root *blocks) 34515d4f98a2SYan Zheng { 3452a2c8d27eSFilipe Manana struct btrfs_backref_walk_ctx ctx = { 0 }; 345319b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 345419b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 3455a2c8d27eSFilipe Manana const u32 blocksize = rc->extent_root->fs_info->nodesize; 3456647f63bdSFilipe David Borba Manana int ret = 0; 34575d4f98a2SYan Zheng 3458b3b4aa74SDavid Sterba btrfs_release_path(path); 3459a2c8d27eSFilipe Manana 3460a2c8d27eSFilipe Manana ctx.bytenr = extent_key->objectid; 34610cad8f14SFilipe Manana ctx.skip_inode_ref_list = true; 3462a2c8d27eSFilipe Manana ctx.fs_info = rc->extent_root->fs_info; 3463a2c8d27eSFilipe Manana 3464a2c8d27eSFilipe Manana ret = btrfs_find_all_leafs(&ctx); 346519b546d7SQu Wenruo if (ret < 0) 346619b546d7SQu Wenruo return ret; 346719b546d7SQu Wenruo 346819b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 3469a2c8d27eSFilipe Manana while ((ref_node = ulist_next(ctx.refs, &leaf_uiter))) { 3470789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 0 }; 347119b546d7SQu Wenruo struct extent_buffer *eb; 347219b546d7SQu Wenruo 3473789d6a3aSQu Wenruo eb = read_tree_block(ctx.fs_info, ref_node->val, &check); 347419b546d7SQu Wenruo if (IS_ERR(eb)) { 347519b546d7SQu Wenruo ret = PTR_ERR(eb); 347619b546d7SQu Wenruo break; 347719b546d7SQu Wenruo } 347819b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 347919b546d7SQu Wenruo extent_key->objectid); 348019b546d7SQu Wenruo free_extent_buffer(eb); 348119b546d7SQu Wenruo if (ret < 0) 348219b546d7SQu Wenruo break; 348319b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 348419b546d7SQu Wenruo if (ret < 0) 348519b546d7SQu Wenruo break; 348619b546d7SQu Wenruo } 348719b546d7SQu Wenruo if (ret < 0) 34885d4f98a2SYan Zheng free_block_list(blocks); 3489a2c8d27eSFilipe Manana ulist_free(ctx.refs); 349019b546d7SQu Wenruo return ret; 34915d4f98a2SYan Zheng } 34925d4f98a2SYan Zheng 34935d4f98a2SYan Zheng /* 34942c016dc2SLiu Bo * helper to find next unprocessed extent 34955d4f98a2SYan Zheng */ 34965d4f98a2SYan Zheng static noinline_for_stack 3497147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 34983fd0a558SYan, Zheng struct btrfs_key *extent_key) 34995d4f98a2SYan Zheng { 35000b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35015d4f98a2SYan Zheng struct btrfs_key key; 35025d4f98a2SYan Zheng struct extent_buffer *leaf; 35035d4f98a2SYan Zheng u64 start, end, last; 35045d4f98a2SYan Zheng int ret; 35055d4f98a2SYan Zheng 3506b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 35075d4f98a2SYan Zheng while (1) { 3508e5860f82SFilipe Manana bool block_found; 3509e5860f82SFilipe Manana 35105d4f98a2SYan Zheng cond_resched(); 35115d4f98a2SYan Zheng if (rc->search_start >= last) { 35125d4f98a2SYan Zheng ret = 1; 35135d4f98a2SYan Zheng break; 35145d4f98a2SYan Zheng } 35155d4f98a2SYan Zheng 35165d4f98a2SYan Zheng key.objectid = rc->search_start; 35175d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 35185d4f98a2SYan Zheng key.offset = 0; 35195d4f98a2SYan Zheng 35205d4f98a2SYan Zheng path->search_commit_root = 1; 35215d4f98a2SYan Zheng path->skip_locking = 1; 35225d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 35235d4f98a2SYan Zheng 0, 0); 35245d4f98a2SYan Zheng if (ret < 0) 35255d4f98a2SYan Zheng break; 35265d4f98a2SYan Zheng next: 35275d4f98a2SYan Zheng leaf = path->nodes[0]; 35285d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 35295d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 35305d4f98a2SYan Zheng if (ret != 0) 35315d4f98a2SYan Zheng break; 35325d4f98a2SYan Zheng leaf = path->nodes[0]; 35335d4f98a2SYan Zheng } 35345d4f98a2SYan Zheng 35355d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 35365d4f98a2SYan Zheng if (key.objectid >= last) { 35375d4f98a2SYan Zheng ret = 1; 35385d4f98a2SYan Zheng break; 35395d4f98a2SYan Zheng } 35405d4f98a2SYan Zheng 35413173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 35423173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 35433173a18fSJosef Bacik path->slots[0]++; 35443173a18fSJosef Bacik goto next; 35453173a18fSJosef Bacik } 35463173a18fSJosef Bacik 35473173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 35485d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 35495d4f98a2SYan Zheng path->slots[0]++; 35505d4f98a2SYan Zheng goto next; 35515d4f98a2SYan Zheng } 35525d4f98a2SYan Zheng 35533173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 35540b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 35553173a18fSJosef Bacik rc->search_start) { 35563173a18fSJosef Bacik path->slots[0]++; 35573173a18fSJosef Bacik goto next; 35583173a18fSJosef Bacik } 35593173a18fSJosef Bacik 3560e5860f82SFilipe Manana block_found = find_first_extent_bit(&rc->processed_blocks, 35615d4f98a2SYan Zheng key.objectid, &start, &end, 3562e6138876SJosef Bacik EXTENT_DIRTY, NULL); 35635d4f98a2SYan Zheng 3564e5860f82SFilipe Manana if (block_found && start <= key.objectid) { 3565b3b4aa74SDavid Sterba btrfs_release_path(path); 35665d4f98a2SYan Zheng rc->search_start = end + 1; 35675d4f98a2SYan Zheng } else { 35683173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 35695d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 35703173a18fSJosef Bacik else 35713173a18fSJosef Bacik rc->search_start = key.objectid + 35720b246afaSJeff Mahoney fs_info->nodesize; 35733fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 35745d4f98a2SYan Zheng return 0; 35755d4f98a2SYan Zheng } 35765d4f98a2SYan Zheng } 3577b3b4aa74SDavid Sterba btrfs_release_path(path); 35785d4f98a2SYan Zheng return ret; 35795d4f98a2SYan Zheng } 35805d4f98a2SYan Zheng 35815d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 35825d4f98a2SYan Zheng { 35835d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35847585717fSChris Mason 35857585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35865d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 35877585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35885d4f98a2SYan Zheng } 35895d4f98a2SYan Zheng 35905d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 35915d4f98a2SYan Zheng { 35925d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35937585717fSChris Mason 35947585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 35955d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 35967585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 35975d4f98a2SYan Zheng } 35985d4f98a2SYan Zheng 35993fd0a558SYan, Zheng static noinline_for_stack 36003fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 36013fd0a558SYan, Zheng { 36023fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3603ac2fabacSJosef Bacik int ret; 36043fd0a558SYan, Zheng 36052ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 360666d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 36073fd0a558SYan, Zheng if (!rc->block_rsv) 36083fd0a558SYan, Zheng return -ENOMEM; 36093fd0a558SYan, Zheng 36103fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3611b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 36123fd0a558SYan, Zheng rc->extents_found = 0; 36133fd0a558SYan, Zheng rc->nodes_relocated = 0; 36143fd0a558SYan, Zheng rc->merging_rsv_size = 0; 36150647bf56SWang Shilong rc->reserved_bytes = 0; 3616da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 36170647bf56SWang Shilong RELOCATION_RESERVED_NODES; 36189270501cSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root->fs_info, 3619ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3620ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3621ac2fabacSJosef Bacik if (ret) 3622ac2fabacSJosef Bacik return ret; 36233fd0a558SYan, Zheng 36243fd0a558SYan, Zheng rc->create_reloc_tree = 1; 36253fd0a558SYan, Zheng set_reloc_control(rc); 36263fd0a558SYan, Zheng 36277a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 362828818947SLiu Bo if (IS_ERR(trans)) { 362928818947SLiu Bo unset_reloc_control(rc); 363028818947SLiu Bo /* 363128818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 363228818947SLiu Bo * cleanup. And callers are responsible to free the above 363328818947SLiu Bo * block rsv. 363428818947SLiu Bo */ 363528818947SLiu Bo return PTR_ERR(trans); 363628818947SLiu Bo } 363785f02d6cSZixuan Fu 363885f02d6cSZixuan Fu ret = btrfs_commit_transaction(trans); 363985f02d6cSZixuan Fu if (ret) 364085f02d6cSZixuan Fu unset_reloc_control(rc); 364185f02d6cSZixuan Fu 364285f02d6cSZixuan Fu return ret; 36433fd0a558SYan, Zheng } 364476dda93cSYan, Zheng 36455d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 36465d4f98a2SYan Zheng { 36472ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 36485d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 36495d4f98a2SYan Zheng struct btrfs_key key; 36505d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 36515d4f98a2SYan Zheng struct btrfs_path *path; 36525d4f98a2SYan Zheng struct btrfs_extent_item *ei; 36535d4f98a2SYan Zheng u64 flags; 36545d4f98a2SYan Zheng int ret; 36555d4f98a2SYan Zheng int err = 0; 3656c87f08caSChris Mason int progress = 0; 36575d4f98a2SYan Zheng 36585d4f98a2SYan Zheng path = btrfs_alloc_path(); 36593fd0a558SYan, Zheng if (!path) 36605d4f98a2SYan Zheng return -ENOMEM; 3661e4058b54SDavid Sterba path->reada = READA_FORWARD; 36623fd0a558SYan, Zheng 36633fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 36643fd0a558SYan, Zheng if (ret) { 36653fd0a558SYan, Zheng err = ret; 36663fd0a558SYan, Zheng goto out_free; 36672423fdfbSJiri Slaby } 36685d4f98a2SYan Zheng 36695d4f98a2SYan Zheng while (1) { 36700647bf56SWang Shilong rc->reserved_bytes = 0; 36719270501cSJosef Bacik ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, 36729270501cSJosef Bacik rc->block_rsv->size, 36730647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 36740647bf56SWang Shilong if (ret) { 36750647bf56SWang Shilong err = ret; 36760647bf56SWang Shilong break; 36770647bf56SWang Shilong } 3678c87f08caSChris Mason progress++; 3679a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 36800f788c58SLiu Bo if (IS_ERR(trans)) { 36810f788c58SLiu Bo err = PTR_ERR(trans); 36820f788c58SLiu Bo trans = NULL; 36830f788c58SLiu Bo break; 36840f788c58SLiu Bo } 3685c87f08caSChris Mason restart: 36863fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 36873a45bb20SJeff Mahoney btrfs_end_transaction(trans); 368842a657f5SPan Bian trans = NULL; 36893fd0a558SYan, Zheng continue; 36903fd0a558SYan, Zheng } 36913fd0a558SYan, Zheng 3692147d256eSZhaolei ret = find_next_extent(rc, path, &key); 36935d4f98a2SYan Zheng if (ret < 0) 36945d4f98a2SYan Zheng err = ret; 36955d4f98a2SYan Zheng if (ret != 0) 36965d4f98a2SYan Zheng break; 36975d4f98a2SYan Zheng 36985d4f98a2SYan Zheng rc->extents_found++; 36995d4f98a2SYan Zheng 37005d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 37015d4f98a2SYan Zheng struct btrfs_extent_item); 37025d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 37035d4f98a2SYan Zheng 37045d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 37055d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 37065d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 37075d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 37085d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 37095d4f98a2SYan Zheng } else { 3710b3b4aa74SDavid Sterba btrfs_release_path(path); 37115d4f98a2SYan Zheng ret = 0; 37125d4f98a2SYan Zheng } 37135d4f98a2SYan Zheng if (ret < 0) { 37143fd0a558SYan, Zheng err = ret; 37155d4f98a2SYan Zheng break; 37165d4f98a2SYan Zheng } 37175d4f98a2SYan Zheng 37185d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 37195d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 37205d4f98a2SYan Zheng if (ret < 0) { 37213fd0a558SYan, Zheng if (ret != -EAGAIN) { 37225d4f98a2SYan Zheng err = ret; 37235d4f98a2SYan Zheng break; 37245d4f98a2SYan Zheng } 37253fd0a558SYan, Zheng rc->extents_found--; 37263fd0a558SYan, Zheng rc->search_start = key.objectid; 37273fd0a558SYan, Zheng } 37285d4f98a2SYan Zheng } 37295d4f98a2SYan Zheng 37303a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 37312ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 37323fd0a558SYan, Zheng trans = NULL; 37335d4f98a2SYan Zheng 37345d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 37355d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 37365d4f98a2SYan Zheng rc->found_file_extent = 1; 37370257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 37383fd0a558SYan, Zheng &key, &rc->cluster); 37395d4f98a2SYan Zheng if (ret < 0) { 37405d4f98a2SYan Zheng err = ret; 37415d4f98a2SYan Zheng break; 37425d4f98a2SYan Zheng } 37435d4f98a2SYan Zheng } 3744f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 3745f31ea088SQu Wenruo err = -ECANCELED; 3746f31ea088SQu Wenruo break; 3747f31ea088SQu Wenruo } 37485d4f98a2SYan Zheng } 3749c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 375043a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 37519689457bSShilong Wang if (ret == 1) { 3752c87f08caSChris Mason err = 0; 3753c87f08caSChris Mason progress = 0; 3754c87f08caSChris Mason goto restart; 3755c87f08caSChris Mason } 3756c87f08caSChris Mason } 37573fd0a558SYan, Zheng 3758b3b4aa74SDavid Sterba btrfs_release_path(path); 375991166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 37605d4f98a2SYan Zheng 37615d4f98a2SYan Zheng if (trans) { 37623a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 37632ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 37645d4f98a2SYan Zheng } 37655d4f98a2SYan Zheng 37660257bb82SYan, Zheng if (!err) { 37673fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 37683fd0a558SYan, Zheng &rc->cluster); 37690257bb82SYan, Zheng if (ret < 0) 37700257bb82SYan, Zheng err = ret; 37710257bb82SYan, Zheng } 37720257bb82SYan, Zheng 37733fd0a558SYan, Zheng rc->create_reloc_tree = 0; 37743fd0a558SYan, Zheng set_reloc_control(rc); 37750257bb82SYan, Zheng 377613fe1bdbSQu Wenruo btrfs_backref_release_cache(&rc->backref_cache); 377763f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37785d4f98a2SYan Zheng 37797f913c7cSQu Wenruo /* 37807f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 37817f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 37827f913c7cSQu Wenruo * 37837f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 37847f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 37857f913c7cSQu Wenruo * merge_reloc_roots() 37867f913c7cSQu Wenruo */ 37873fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 37885d4f98a2SYan Zheng 37895d4f98a2SYan Zheng merge_reloc_roots(rc); 37905d4f98a2SYan Zheng 37913fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 37925d4f98a2SYan Zheng unset_reloc_control(rc); 379363f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 37945d4f98a2SYan Zheng 37955d4f98a2SYan Zheng /* get rid of pinned extents */ 37967a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 379762b99540SQu Wenruo if (IS_ERR(trans)) { 37983612b495STsutomu Itoh err = PTR_ERR(trans); 379962b99540SQu Wenruo goto out_free; 380062b99540SQu Wenruo } 3801fb686c68SJosef Bacik ret = btrfs_commit_transaction(trans); 3802fb686c68SJosef Bacik if (ret && !err) 3803fb686c68SJosef Bacik err = ret; 38046217b0faSJosef Bacik out_free: 3805d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 3806d2311e69SQu Wenruo if (ret < 0 && !err) 3807d2311e69SQu Wenruo err = ret; 38082ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 38093fd0a558SYan, Zheng btrfs_free_path(path); 38105d4f98a2SYan Zheng return err; 38115d4f98a2SYan Zheng } 38125d4f98a2SYan Zheng 38135d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 38140257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 38155d4f98a2SYan Zheng { 38165d4f98a2SYan Zheng struct btrfs_path *path; 38175d4f98a2SYan Zheng struct btrfs_inode_item *item; 38185d4f98a2SYan Zheng struct extent_buffer *leaf; 38195d4f98a2SYan Zheng int ret; 38205d4f98a2SYan Zheng 38215d4f98a2SYan Zheng path = btrfs_alloc_path(); 38225d4f98a2SYan Zheng if (!path) 38235d4f98a2SYan Zheng return -ENOMEM; 38245d4f98a2SYan Zheng 38255d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 38265d4f98a2SYan Zheng if (ret) 38275d4f98a2SYan Zheng goto out; 38285d4f98a2SYan Zheng 38295d4f98a2SYan Zheng leaf = path->nodes[0]; 38305d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 3831b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 38325d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 38330257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 38345d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 3835960a3166SJohannes Thumshirn btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 3836960a3166SJohannes Thumshirn BTRFS_INODE_PREALLOC); 38375d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 38385d4f98a2SYan Zheng out: 38395d4f98a2SYan Zheng btrfs_free_path(path); 38405d4f98a2SYan Zheng return ret; 38415d4f98a2SYan Zheng } 38425d4f98a2SYan Zheng 3843790c1b8cSJosef Bacik static void delete_orphan_inode(struct btrfs_trans_handle *trans, 3844790c1b8cSJosef Bacik struct btrfs_root *root, u64 objectid) 3845790c1b8cSJosef Bacik { 3846790c1b8cSJosef Bacik struct btrfs_path *path; 3847790c1b8cSJosef Bacik struct btrfs_key key; 3848790c1b8cSJosef Bacik int ret = 0; 3849790c1b8cSJosef Bacik 3850790c1b8cSJosef Bacik path = btrfs_alloc_path(); 3851790c1b8cSJosef Bacik if (!path) { 3852790c1b8cSJosef Bacik ret = -ENOMEM; 3853790c1b8cSJosef Bacik goto out; 3854790c1b8cSJosef Bacik } 3855790c1b8cSJosef Bacik 3856790c1b8cSJosef Bacik key.objectid = objectid; 3857790c1b8cSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 3858790c1b8cSJosef Bacik key.offset = 0; 3859790c1b8cSJosef Bacik ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3860790c1b8cSJosef Bacik if (ret) { 3861790c1b8cSJosef Bacik if (ret > 0) 3862790c1b8cSJosef Bacik ret = -ENOENT; 3863790c1b8cSJosef Bacik goto out; 3864790c1b8cSJosef Bacik } 3865790c1b8cSJosef Bacik ret = btrfs_del_item(trans, root, path); 3866790c1b8cSJosef Bacik out: 3867790c1b8cSJosef Bacik if (ret) 3868790c1b8cSJosef Bacik btrfs_abort_transaction(trans, ret); 3869790c1b8cSJosef Bacik btrfs_free_path(path); 3870790c1b8cSJosef Bacik } 3871790c1b8cSJosef Bacik 38725d4f98a2SYan Zheng /* 38735d4f98a2SYan Zheng * helper to create inode for data relocation. 38745d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 38755d4f98a2SYan Zheng */ 38763fd0a558SYan, Zheng static noinline_for_stack 38773fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 387832da5386SDavid Sterba struct btrfs_block_group *group) 38795d4f98a2SYan Zheng { 38805d4f98a2SYan Zheng struct inode *inode = NULL; 38815d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 38825d4f98a2SYan Zheng struct btrfs_root *root; 38834624900dSZhaolei u64 objectid; 38845d4f98a2SYan Zheng int err = 0; 38855d4f98a2SYan Zheng 3886aeb935a4SQu Wenruo root = btrfs_grab_root(fs_info->data_reloc_root); 3887a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 388876deacf0SJosef Bacik if (IS_ERR(trans)) { 388900246528SJosef Bacik btrfs_put_root(root); 38903fd0a558SYan, Zheng return ERR_CAST(trans); 389176deacf0SJosef Bacik } 38925d4f98a2SYan Zheng 3893543068a2SNikolay Borisov err = btrfs_get_free_objectid(root, &objectid); 38945d4f98a2SYan Zheng if (err) 38955d4f98a2SYan Zheng goto out; 38965d4f98a2SYan Zheng 38970257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 3898790c1b8cSJosef Bacik if (err) 3899790c1b8cSJosef Bacik goto out; 39005d4f98a2SYan Zheng 39010202e83fSDavid Sterba inode = btrfs_iget(fs_info->sb, objectid, root); 3902790c1b8cSJosef Bacik if (IS_ERR(inode)) { 3903790c1b8cSJosef Bacik delete_orphan_inode(trans, root, objectid); 3904790c1b8cSJosef Bacik err = PTR_ERR(inode); 3905790c1b8cSJosef Bacik inode = NULL; 3906790c1b8cSJosef Bacik goto out; 3907790c1b8cSJosef Bacik } 3908b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 39095d4f98a2SYan Zheng 391073f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 39115d4f98a2SYan Zheng out: 391200246528SJosef Bacik btrfs_put_root(root); 39133a45bb20SJeff Mahoney btrfs_end_transaction(trans); 39142ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 39155d4f98a2SYan Zheng if (err) { 39165d4f98a2SYan Zheng iput(inode); 39175d4f98a2SYan Zheng inode = ERR_PTR(err); 39185d4f98a2SYan Zheng } 39195d4f98a2SYan Zheng return inode; 39205d4f98a2SYan Zheng } 39215d4f98a2SYan Zheng 3922907d2710SDavid Sterba /* 3923907d2710SDavid Sterba * Mark start of chunk relocation that is cancellable. Check if the cancellation 3924907d2710SDavid Sterba * has been requested meanwhile and don't start in that case. 3925907d2710SDavid Sterba * 3926907d2710SDavid Sterba * Return: 3927907d2710SDavid Sterba * 0 success 3928907d2710SDavid Sterba * -EINPROGRESS operation is already in progress, that's probably a bug 3929907d2710SDavid Sterba * -ECANCELED cancellation request was set before the operation started 3930907d2710SDavid Sterba */ 3931907d2710SDavid Sterba static int reloc_chunk_start(struct btrfs_fs_info *fs_info) 3932907d2710SDavid Sterba { 3933907d2710SDavid Sterba if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) { 3934907d2710SDavid Sterba /* This should not happen */ 3935907d2710SDavid Sterba btrfs_err(fs_info, "reloc already running, cannot start"); 3936907d2710SDavid Sterba return -EINPROGRESS; 3937907d2710SDavid Sterba } 3938907d2710SDavid Sterba 3939907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) { 3940907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled on start"); 3941907d2710SDavid Sterba /* 3942907d2710SDavid Sterba * On cancel, clear all requests but let the caller mark 3943907d2710SDavid Sterba * the end after cleanup operations. 3944907d2710SDavid Sterba */ 3945907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3946907d2710SDavid Sterba return -ECANCELED; 3947907d2710SDavid Sterba } 3948907d2710SDavid Sterba return 0; 3949907d2710SDavid Sterba } 3950907d2710SDavid Sterba 3951907d2710SDavid Sterba /* 3952907d2710SDavid Sterba * Mark end of chunk relocation that is cancellable and wake any waiters. 3953907d2710SDavid Sterba */ 3954907d2710SDavid Sterba static void reloc_chunk_end(struct btrfs_fs_info *fs_info) 3955907d2710SDavid Sterba { 3956907d2710SDavid Sterba /* Requested after start, clear bit first so any waiters can continue */ 3957907d2710SDavid Sterba if (atomic_read(&fs_info->reloc_cancel_req) > 0) 3958907d2710SDavid Sterba btrfs_info(fs_info, "chunk relocation canceled during operation"); 3959907d2710SDavid Sterba clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags); 3960907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 3961907d2710SDavid Sterba } 3962907d2710SDavid Sterba 3963c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 39643fd0a558SYan, Zheng { 39653fd0a558SYan, Zheng struct reloc_control *rc; 39663fd0a558SYan, Zheng 39673fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 39683fd0a558SYan, Zheng if (!rc) 39693fd0a558SYan, Zheng return NULL; 39703fd0a558SYan, Zheng 39713fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 3972d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 3973584fb121SQu Wenruo btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1); 39743fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 397535da5a7eSDavid Sterba extent_io_tree_init(fs_info, &rc->processed_blocks, IO_TREE_RELOC_BLOCKS); 39763fd0a558SYan, Zheng return rc; 39773fd0a558SYan, Zheng } 39783fd0a558SYan, Zheng 39791a0afa0eSJosef Bacik static void free_reloc_control(struct reloc_control *rc) 39801a0afa0eSJosef Bacik { 39811a0afa0eSJosef Bacik struct mapping_node *node, *tmp; 39821a0afa0eSJosef Bacik 39831a0afa0eSJosef Bacik free_reloc_roots(&rc->reloc_roots); 39841a0afa0eSJosef Bacik rbtree_postorder_for_each_entry_safe(node, tmp, 39851a0afa0eSJosef Bacik &rc->reloc_root_tree.rb_root, rb_node) 39861a0afa0eSJosef Bacik kfree(node); 39871a0afa0eSJosef Bacik 39881a0afa0eSJosef Bacik kfree(rc); 39891a0afa0eSJosef Bacik } 39901a0afa0eSJosef Bacik 39915d4f98a2SYan Zheng /* 3992ebce0e01SAdam Borowski * Print the block group being relocated 3993ebce0e01SAdam Borowski */ 3994ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 399532da5386SDavid Sterba struct btrfs_block_group *block_group) 3996ebce0e01SAdam Borowski { 3997f89e09cfSAnand Jain char buf[128] = {'\0'}; 3998ebce0e01SAdam Borowski 3999f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4000ebce0e01SAdam Borowski 4001ebce0e01SAdam Borowski btrfs_info(fs_info, 4002ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4003b3470b5dSDavid Sterba block_group->start, buf); 4004ebce0e01SAdam Borowski } 4005ebce0e01SAdam Borowski 4006430640e3SQu Wenruo static const char *stage_to_string(int stage) 4007430640e3SQu Wenruo { 4008430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 4009430640e3SQu Wenruo return "move data extents"; 4010430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 4011430640e3SQu Wenruo return "update data pointers"; 4012430640e3SQu Wenruo return "unknown"; 4013430640e3SQu Wenruo } 4014430640e3SQu Wenruo 4015ebce0e01SAdam Borowski /* 40165d4f98a2SYan Zheng * function to relocate all extents in a block group. 40175d4f98a2SYan Zheng */ 40186bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 40195d4f98a2SYan Zheng { 402032da5386SDavid Sterba struct btrfs_block_group *bg; 402129cbcf40SJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start); 40225d4f98a2SYan Zheng struct reloc_control *rc; 40230af3d00bSJosef Bacik struct inode *inode; 40240af3d00bSJosef Bacik struct btrfs_path *path; 40255d4f98a2SYan Zheng int ret; 4026f0486c68SYan, Zheng int rw = 0; 40275d4f98a2SYan Zheng int err = 0; 40285d4f98a2SYan Zheng 4029b4be6aefSJosef Bacik /* 4030b4be6aefSJosef Bacik * This only gets set if we had a half-deleted snapshot on mount. We 4031b4be6aefSJosef Bacik * cannot allow relocation to start while we're still trying to clean up 4032b4be6aefSJosef Bacik * these pending deletions. 4033b4be6aefSJosef Bacik */ 4034b4be6aefSJosef Bacik ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE); 4035b4be6aefSJosef Bacik if (ret) 4036b4be6aefSJosef Bacik return ret; 4037b4be6aefSJosef Bacik 4038b4be6aefSJosef Bacik /* We may have been woken up by close_ctree, so bail if we're closing. */ 4039b4be6aefSJosef Bacik if (btrfs_fs_closing(fs_info)) 4040b4be6aefSJosef Bacik return -EINTR; 4041b4be6aefSJosef Bacik 4042eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4043eede2bf3SOmar Sandoval if (!bg) 4044eede2bf3SOmar Sandoval return -ENOENT; 4045eede2bf3SOmar Sandoval 40460320b353SNaohiro Aota /* 40470320b353SNaohiro Aota * Relocation of a data block group creates ordered extents. Without 40480320b353SNaohiro Aota * sb_start_write(), we can freeze the filesystem while unfinished 40490320b353SNaohiro Aota * ordered extents are left. Such ordered extents can cause a deadlock 40500320b353SNaohiro Aota * e.g. when syncfs() is waiting for their completion but they can't 40510320b353SNaohiro Aota * finish because they block when joining a transaction, due to the 40520320b353SNaohiro Aota * fact that the freeze locks are being held in write mode. 40530320b353SNaohiro Aota */ 40540320b353SNaohiro Aota if (bg->flags & BTRFS_BLOCK_GROUP_DATA) 40550320b353SNaohiro Aota ASSERT(sb_write_started(fs_info->sb)); 40560320b353SNaohiro Aota 4057eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4058eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4059eede2bf3SOmar Sandoval return -ETXTBSY; 4060eede2bf3SOmar Sandoval } 4061eede2bf3SOmar Sandoval 4062c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4063eede2bf3SOmar Sandoval if (!rc) { 4064eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 40655d4f98a2SYan Zheng return -ENOMEM; 4066eede2bf3SOmar Sandoval } 40675d4f98a2SYan Zheng 4068907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4069907d2710SDavid Sterba if (ret < 0) { 4070907d2710SDavid Sterba err = ret; 4071907d2710SDavid Sterba goto out_put_bg; 4072907d2710SDavid Sterba } 4073907d2710SDavid Sterba 4074f0486c68SYan, Zheng rc->extent_root = extent_root; 4075eede2bf3SOmar Sandoval rc->block_group = bg; 40765d4f98a2SYan Zheng 4077b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4078f0486c68SYan, Zheng if (ret) { 4079f0486c68SYan, Zheng err = ret; 4080f0486c68SYan, Zheng goto out; 4081f0486c68SYan, Zheng } 4082f0486c68SYan, Zheng rw = 1; 4083f0486c68SYan, Zheng 40840af3d00bSJosef Bacik path = btrfs_alloc_path(); 40850af3d00bSJosef Bacik if (!path) { 40860af3d00bSJosef Bacik err = -ENOMEM; 40870af3d00bSJosef Bacik goto out; 40880af3d00bSJosef Bacik } 40890af3d00bSJosef Bacik 40907949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 40910af3d00bSJosef Bacik btrfs_free_path(path); 40920af3d00bSJosef Bacik 40930af3d00bSJosef Bacik if (!IS_ERR(inode)) 40941bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 40950af3d00bSJosef Bacik else 40960af3d00bSJosef Bacik ret = PTR_ERR(inode); 40970af3d00bSJosef Bacik 40980af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 40990af3d00bSJosef Bacik err = ret; 41000af3d00bSJosef Bacik goto out; 41010af3d00bSJosef Bacik } 41020af3d00bSJosef Bacik 41035d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 41045d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 41055d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 41065d4f98a2SYan Zheng rc->data_inode = NULL; 41075d4f98a2SYan Zheng goto out; 41085d4f98a2SYan Zheng } 41095d4f98a2SYan Zheng 41100b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 41115d4f98a2SYan Zheng 41129cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4113f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 41146374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4115b3470b5dSDavid Sterba rc->block_group->start, 4116b3470b5dSDavid Sterba rc->block_group->length); 41175d4f98a2SYan Zheng 41187ae9bd18SNaohiro Aota ret = btrfs_zone_finish(rc->block_group); 41197ae9bd18SNaohiro Aota WARN_ON(ret && ret != -EAGAIN); 41207ae9bd18SNaohiro Aota 41215d4f98a2SYan Zheng while (1) { 4122430640e3SQu Wenruo int finishes_stage; 4123430640e3SQu Wenruo 412476dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 41255d4f98a2SYan Zheng ret = relocate_block_group(rc); 412676dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4127ff612ba7SJosef Bacik if (ret < 0) 41285d4f98a2SYan Zheng err = ret; 4129ff612ba7SJosef Bacik 4130430640e3SQu Wenruo finishes_stage = rc->stage; 4131ff612ba7SJosef Bacik /* 4132ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4133ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4134ff612ba7SJosef Bacik * different block group we could end up hitting the 4135ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4136ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4137ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4138ff612ba7SJosef Bacik * out of the loop if we hit an error. 4139ff612ba7SJosef Bacik */ 4140ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4141ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4142ff612ba7SJosef Bacik (u64)-1); 4143ff612ba7SJosef Bacik if (ret) 4144ff612ba7SJosef Bacik err = ret; 4145ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4146ff612ba7SJosef Bacik 0, -1); 4147ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 41485d4f98a2SYan Zheng } 41495d4f98a2SYan Zheng 4150ff612ba7SJosef Bacik if (err < 0) 4151ff612ba7SJosef Bacik goto out; 4152ff612ba7SJosef Bacik 41535d4f98a2SYan Zheng if (rc->extents_found == 0) 41545d4f98a2SYan Zheng break; 41555d4f98a2SYan Zheng 4156430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4157430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 41585d4f98a2SYan Zheng } 41595d4f98a2SYan Zheng 41605d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 41615d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4162bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 41635d4f98a2SYan Zheng out: 4164f0486c68SYan, Zheng if (err && rw) 41652ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 41665d4f98a2SYan Zheng iput(rc->data_inode); 4167907d2710SDavid Sterba out_put_bg: 4168907d2710SDavid Sterba btrfs_put_block_group(bg); 4169907d2710SDavid Sterba reloc_chunk_end(fs_info); 41701a0afa0eSJosef Bacik free_reloc_control(rc); 41715d4f98a2SYan Zheng return err; 41725d4f98a2SYan Zheng } 41735d4f98a2SYan Zheng 417476dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 417576dda93cSYan, Zheng { 41760b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 417776dda93cSYan, Zheng struct btrfs_trans_handle *trans; 417879787eaaSJeff Mahoney int ret, err; 417976dda93cSYan, Zheng 41800b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 418179787eaaSJeff Mahoney if (IS_ERR(trans)) 418279787eaaSJeff Mahoney return PTR_ERR(trans); 418376dda93cSYan, Zheng 418476dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 418576dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 4186c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 418776dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 41880b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 418976dda93cSYan, Zheng &root->root_key, &root->root_item); 419076dda93cSYan, Zheng 41913a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 419279787eaaSJeff Mahoney if (err) 419379787eaaSJeff Mahoney return err; 419479787eaaSJeff Mahoney return ret; 419576dda93cSYan, Zheng } 419676dda93cSYan, Zheng 41975d4f98a2SYan Zheng /* 41985d4f98a2SYan Zheng * recover relocation interrupted by system crash. 41995d4f98a2SYan Zheng * 42005d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 42015d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 42025d4f98a2SYan Zheng */ 42037eefae6bSJosef Bacik int btrfs_recover_relocation(struct btrfs_fs_info *fs_info) 42045d4f98a2SYan Zheng { 42055d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 42065d4f98a2SYan Zheng struct btrfs_key key; 42075d4f98a2SYan Zheng struct btrfs_root *fs_root; 42085d4f98a2SYan Zheng struct btrfs_root *reloc_root; 42095d4f98a2SYan Zheng struct btrfs_path *path; 42105d4f98a2SYan Zheng struct extent_buffer *leaf; 42115d4f98a2SYan Zheng struct reloc_control *rc = NULL; 42125d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42135d4f98a2SYan Zheng int ret; 42145d4f98a2SYan Zheng int err = 0; 42155d4f98a2SYan Zheng 42165d4f98a2SYan Zheng path = btrfs_alloc_path(); 42175d4f98a2SYan Zheng if (!path) 42185d4f98a2SYan Zheng return -ENOMEM; 4219e4058b54SDavid Sterba path->reada = READA_BACK; 42205d4f98a2SYan Zheng 42215d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 42225d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 42235d4f98a2SYan Zheng key.offset = (u64)-1; 42245d4f98a2SYan Zheng 42255d4f98a2SYan Zheng while (1) { 42260b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 42275d4f98a2SYan Zheng path, 0, 0); 42285d4f98a2SYan Zheng if (ret < 0) { 42295d4f98a2SYan Zheng err = ret; 42305d4f98a2SYan Zheng goto out; 42315d4f98a2SYan Zheng } 42325d4f98a2SYan Zheng if (ret > 0) { 42335d4f98a2SYan Zheng if (path->slots[0] == 0) 42345d4f98a2SYan Zheng break; 42355d4f98a2SYan Zheng path->slots[0]--; 42365d4f98a2SYan Zheng } 42375d4f98a2SYan Zheng leaf = path->nodes[0]; 42385d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4239b3b4aa74SDavid Sterba btrfs_release_path(path); 42405d4f98a2SYan Zheng 42415d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 42425d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 42435d4f98a2SYan Zheng break; 42445d4f98a2SYan Zheng 42457eefae6bSJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key); 42465d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 42475d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 42485d4f98a2SYan Zheng goto out; 42495d4f98a2SYan Zheng } 42505d4f98a2SYan Zheng 425192a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); 42525d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 42535d4f98a2SYan Zheng 42545d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 4255a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, 4256a820feb5SDavid Sterba reloc_root->root_key.offset, false); 42575d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 425876dda93cSYan, Zheng ret = PTR_ERR(fs_root); 425976dda93cSYan, Zheng if (ret != -ENOENT) { 426076dda93cSYan, Zheng err = ret; 42615d4f98a2SYan Zheng goto out; 42625d4f98a2SYan Zheng } 426379787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 426479787eaaSJeff Mahoney if (ret < 0) { 426579787eaaSJeff Mahoney err = ret; 426679787eaaSJeff Mahoney goto out; 426779787eaaSJeff Mahoney } 4268932fd26dSJosef Bacik } else { 426900246528SJosef Bacik btrfs_put_root(fs_root); 427076dda93cSYan, Zheng } 42715d4f98a2SYan Zheng } 42725d4f98a2SYan Zheng 42735d4f98a2SYan Zheng if (key.offset == 0) 42745d4f98a2SYan Zheng break; 42755d4f98a2SYan Zheng 42765d4f98a2SYan Zheng key.offset--; 42775d4f98a2SYan Zheng } 4278b3b4aa74SDavid Sterba btrfs_release_path(path); 42795d4f98a2SYan Zheng 42805d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 42815d4f98a2SYan Zheng goto out; 42825d4f98a2SYan Zheng 4283c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 42845d4f98a2SYan Zheng if (!rc) { 42855d4f98a2SYan Zheng err = -ENOMEM; 42865d4f98a2SYan Zheng goto out; 42875d4f98a2SYan Zheng } 42885d4f98a2SYan Zheng 4289907d2710SDavid Sterba ret = reloc_chunk_start(fs_info); 4290907d2710SDavid Sterba if (ret < 0) { 4291907d2710SDavid Sterba err = ret; 4292907d2710SDavid Sterba goto out_end; 4293907d2710SDavid Sterba } 4294907d2710SDavid Sterba 429529cbcf40SJosef Bacik rc->extent_root = btrfs_extent_root(fs_info, 0); 42965d4f98a2SYan Zheng 42975d4f98a2SYan Zheng set_reloc_control(rc); 42985d4f98a2SYan Zheng 42997a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 43003612b495STsutomu Itoh if (IS_ERR(trans)) { 43013612b495STsutomu Itoh err = PTR_ERR(trans); 4302fb2d83eeSJosef Bacik goto out_unset; 43033612b495STsutomu Itoh } 43043fd0a558SYan, Zheng 43053fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 43063fd0a558SYan, Zheng 43075d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 43085d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 43095d4f98a2SYan Zheng struct btrfs_root, root_list); 43105d4f98a2SYan Zheng list_del(&reloc_root->root_list); 43115d4f98a2SYan Zheng 43125d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 43135d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 43145d4f98a2SYan Zheng &rc->reloc_roots); 43155d4f98a2SYan Zheng continue; 43165d4f98a2SYan Zheng } 43175d4f98a2SYan Zheng 4318a820feb5SDavid Sterba fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, 4319a820feb5SDavid Sterba false); 432079787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 432179787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4322ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 43231402d17dSXiyu Yang btrfs_end_transaction(trans); 4324fb2d83eeSJosef Bacik goto out_unset; 432579787eaaSJeff Mahoney } 43265d4f98a2SYan Zheng 4327ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 432857a304cfSJosef Bacik ASSERT(err != -EEXIST); 43293c925863SJosef Bacik if (err) { 43303c925863SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 43313c925863SJosef Bacik btrfs_put_root(fs_root); 43323c925863SJosef Bacik btrfs_end_transaction(trans); 43333c925863SJosef Bacik goto out_unset; 43343c925863SJosef Bacik } 4335f44deb74SJosef Bacik fs_root->reloc_root = btrfs_grab_root(reloc_root); 433600246528SJosef Bacik btrfs_put_root(fs_root); 43375d4f98a2SYan Zheng } 43385d4f98a2SYan Zheng 43393a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 434079787eaaSJeff Mahoney if (err) 4341fb2d83eeSJosef Bacik goto out_unset; 43425d4f98a2SYan Zheng 43435d4f98a2SYan Zheng merge_reloc_roots(rc); 43445d4f98a2SYan Zheng 43455d4f98a2SYan Zheng unset_reloc_control(rc); 43465d4f98a2SYan Zheng 43477a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 434862b99540SQu Wenruo if (IS_ERR(trans)) { 43493612b495STsutomu Itoh err = PTR_ERR(trans); 43506217b0faSJosef Bacik goto out_clean; 435162b99540SQu Wenruo } 43523a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 43536217b0faSJosef Bacik out_clean: 4354d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4355d2311e69SQu Wenruo if (ret < 0 && !err) 4356d2311e69SQu Wenruo err = ret; 4357fb2d83eeSJosef Bacik out_unset: 4358fb2d83eeSJosef Bacik unset_reloc_control(rc); 4359907d2710SDavid Sterba out_end: 4360907d2710SDavid Sterba reloc_chunk_end(fs_info); 43611a0afa0eSJosef Bacik free_reloc_control(rc); 43623612b495STsutomu Itoh out: 4363aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4364aca1bba6SLiu Bo 43655d4f98a2SYan Zheng btrfs_free_path(path); 43665d4f98a2SYan Zheng 43675d4f98a2SYan Zheng if (err == 0) { 43685d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 4369aeb935a4SQu Wenruo fs_root = btrfs_grab_root(fs_info->data_reloc_root); 4370aeb935a4SQu Wenruo ASSERT(fs_root); 437166b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 437200246528SJosef Bacik btrfs_put_root(fs_root); 4373932fd26dSJosef Bacik } 43745d4f98a2SYan Zheng return err; 43755d4f98a2SYan Zheng } 43765d4f98a2SYan Zheng 43775d4f98a2SYan Zheng /* 43785d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 43795d4f98a2SYan Zheng * 43805d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 43815d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 43825d4f98a2SYan Zheng */ 438334bfaf15SChristoph Hellwig int btrfs_reloc_clone_csums(struct btrfs_ordered_extent *ordered) 43845d4f98a2SYan Zheng { 438534bfaf15SChristoph Hellwig struct btrfs_inode *inode = BTRFS_I(ordered->inode); 43867bfa9535SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 438734bfaf15SChristoph Hellwig u64 disk_bytenr = ordered->file_offset + inode->index_cnt; 438834bfaf15SChristoph Hellwig struct btrfs_root *csum_root = btrfs_csum_root(fs_info, disk_bytenr); 43895d4f98a2SYan Zheng LIST_HEAD(list); 439034bfaf15SChristoph Hellwig int ret; 43915d4f98a2SYan Zheng 439297e38239SQu Wenruo ret = btrfs_lookup_csums_list(csum_root, disk_bytenr, 439334bfaf15SChristoph Hellwig disk_bytenr + ordered->num_bytes - 1, 439434bfaf15SChristoph Hellwig &list, 0, false); 439579787eaaSJeff Mahoney if (ret) 439634bfaf15SChristoph Hellwig return ret; 43975d4f98a2SYan Zheng 43985d4f98a2SYan Zheng while (!list_empty(&list)) { 439934bfaf15SChristoph Hellwig struct btrfs_ordered_sum *sums = 440034bfaf15SChristoph Hellwig list_entry(list.next, struct btrfs_ordered_sum, list); 440134bfaf15SChristoph Hellwig 44025d4f98a2SYan Zheng list_del_init(&sums->list); 44035d4f98a2SYan Zheng 44044577b014SJosef Bacik /* 44054577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 44064577b014SJosef Bacik * We need to do this because we will read in entire prealloc 44074577b014SJosef Bacik * extents but we may have written to say the middle of the 44084577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 44094577b014SJosef Bacik * the right disk offset. 44104577b014SJosef Bacik * 44114577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 44124577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 44134577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 44144577b014SJosef Bacik * disk length. 44154577b014SJosef Bacik */ 441634bfaf15SChristoph Hellwig sums->logical = ordered->disk_bytenr + sums->logical - disk_bytenr; 4417f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 44185d4f98a2SYan Zheng } 441934bfaf15SChristoph Hellwig 442034bfaf15SChristoph Hellwig return 0; 44215d4f98a2SYan Zheng } 44223fd0a558SYan, Zheng 442383d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 44243fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 44253fd0a558SYan, Zheng struct extent_buffer *cow) 44263fd0a558SYan, Zheng { 44270b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 44283fd0a558SYan, Zheng struct reloc_control *rc; 4429a26195a5SQu Wenruo struct btrfs_backref_node *node; 44303fd0a558SYan, Zheng int first_cow = 0; 44313fd0a558SYan, Zheng int level; 443283d4cfd4SJosef Bacik int ret = 0; 44333fd0a558SYan, Zheng 44340b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 44353fd0a558SYan, Zheng if (!rc) 443683d4cfd4SJosef Bacik return 0; 44373fd0a558SYan, Zheng 443837f00a6dSJohannes Thumshirn BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root)); 44393fd0a558SYan, Zheng 44403fd0a558SYan, Zheng level = btrfs_header_level(buf); 44413fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 44423fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 44433fd0a558SYan, Zheng first_cow = 1; 44443fd0a558SYan, Zheng 44453fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 44463fd0a558SYan, Zheng rc->create_reloc_tree) { 44473fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 44483fd0a558SYan, Zheng 44493fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 44503fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 44513fd0a558SYan, Zheng node->new_bytenr != buf->start); 44523fd0a558SYan, Zheng 4453b0fe7078SQu Wenruo btrfs_backref_drop_node_buffer(node); 445467439dadSDavid Sterba atomic_inc(&cow->refs); 44553fd0a558SYan, Zheng node->eb = cow; 44563fd0a558SYan, Zheng node->new_bytenr = cow->start; 44573fd0a558SYan, Zheng 44583fd0a558SYan, Zheng if (!node->pending) { 44593fd0a558SYan, Zheng list_move_tail(&node->list, 44603fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 44613fd0a558SYan, Zheng node->pending = 1; 44623fd0a558SYan, Zheng } 44633fd0a558SYan, Zheng 44643fd0a558SYan, Zheng if (first_cow) 44659569cc20SQu Wenruo mark_block_processed(rc, node); 44663fd0a558SYan, Zheng 44673fd0a558SYan, Zheng if (first_cow && level > 0) 44683fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 44693fd0a558SYan, Zheng } 44703fd0a558SYan, Zheng 447183d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 44723fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 447383d4cfd4SJosef Bacik return ret; 44743fd0a558SYan, Zheng } 44753fd0a558SYan, Zheng 44763fd0a558SYan, Zheng /* 44773fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 447801327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 44793fd0a558SYan, Zheng */ 4480147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 44813fd0a558SYan, Zheng u64 *bytes_to_reserve) 44823fd0a558SYan, Zheng { 448310995c04SQu Wenruo struct btrfs_root *root = pending->root; 448410995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 44853fd0a558SYan, Zheng 44866282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 44873fd0a558SYan, Zheng return; 44883fd0a558SYan, Zheng 44893fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 44903fd0a558SYan, Zheng return; 44913fd0a558SYan, Zheng 44923fd0a558SYan, Zheng root = root->reloc_root; 44933fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 44943fd0a558SYan, Zheng /* 44953fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 44963fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 44973fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 44983fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 44993fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 45003fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 45013fd0a558SYan, Zheng * tree may use more space than it frees. so we need 45023fd0a558SYan, Zheng * reserve extra space. 45033fd0a558SYan, Zheng */ 45043fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 45053fd0a558SYan, Zheng } 45063fd0a558SYan, Zheng 45073fd0a558SYan, Zheng /* 45083fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 45093fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 4510f44deb74SJosef Bacik * 4511f44deb74SJosef Bacik * This is similar to btrfs_init_reloc_root(), we come out of here with two 4512f44deb74SJosef Bacik * references held on the reloc_root, one for root->reloc_root and one for 4513f44deb74SJosef Bacik * rc->reloc_roots. 45143fd0a558SYan, Zheng */ 451549b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 45163fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 45173fd0a558SYan, Zheng { 45183fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 45193fd0a558SYan, Zheng struct btrfs_root *reloc_root; 45203fd0a558SYan, Zheng struct btrfs_root *new_root; 452110995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 45223fd0a558SYan, Zheng int ret; 45233fd0a558SYan, Zheng 45246282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 452549b25e05SJeff Mahoney return 0; 45263fd0a558SYan, Zheng 45273fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 45283fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 45293fd0a558SYan, Zheng 45303fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 45313fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 45323fd0a558SYan, Zheng rc->block_rsv, 45333a584174SLu Fengqi rc->nodes_relocated, true); 453449b25e05SJeff Mahoney if (ret) 453549b25e05SJeff Mahoney return ret; 45363fd0a558SYan, Zheng } 45373fd0a558SYan, Zheng 45383fd0a558SYan, Zheng new_root = pending->snap; 45393fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 45403fd0a558SYan, Zheng new_root->root_key.objectid); 454149b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 454249b25e05SJeff Mahoney return PTR_ERR(reloc_root); 45433fd0a558SYan, Zheng 4544ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 454557a304cfSJosef Bacik ASSERT(ret != -EEXIST); 45463c925863SJosef Bacik if (ret) { 45473c925863SJosef Bacik /* Pairs with create_reloc_root */ 45483c925863SJosef Bacik btrfs_put_root(reloc_root); 45493c925863SJosef Bacik return ret; 45503c925863SJosef Bacik } 4551f44deb74SJosef Bacik new_root->reloc_root = btrfs_grab_root(reloc_root); 45523fd0a558SYan, Zheng 455349b25e05SJeff Mahoney if (rc->create_reloc_tree) 45543fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 455549b25e05SJeff Mahoney return ret; 45563fd0a558SYan, Zheng } 4557b9a9a850SQu Wenruo 4558b9a9a850SQu Wenruo /* 4559b9a9a850SQu Wenruo * Get the current bytenr for the block group which is being relocated. 4560b9a9a850SQu Wenruo * 4561b9a9a850SQu Wenruo * Return U64_MAX if no running relocation. 4562b9a9a850SQu Wenruo */ 4563b9a9a850SQu Wenruo u64 btrfs_get_reloc_bg_bytenr(struct btrfs_fs_info *fs_info) 4564b9a9a850SQu Wenruo { 4565b9a9a850SQu Wenruo u64 logical = U64_MAX; 4566b9a9a850SQu Wenruo 4567b9a9a850SQu Wenruo lockdep_assert_held(&fs_info->reloc_mutex); 4568b9a9a850SQu Wenruo 4569b9a9a850SQu Wenruo if (fs_info->reloc_ctl && fs_info->reloc_ctl->block_group) 4570b9a9a850SQu Wenruo logical = fs_info->reloc_ctl->block_group->start; 4571b9a9a850SQu Wenruo return logical; 4572b9a9a850SQu Wenruo } 4573