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" 21581bb050SLi Zefan #include "inode-map.h" 2262b99540SQu Wenruo #include "qgroup.h" 23cdccee99SLiu Bo #include "print-tree.h" 2486736342SJosef Bacik #include "delalloc-space.h" 25aac0023cSJosef Bacik #include "block-group.h" 2619b546d7SQu Wenruo #include "backref.h" 275d4f98a2SYan Zheng 285d4f98a2SYan Zheng /* 290c891389SQu Wenruo * Relocation overview 300c891389SQu Wenruo * 310c891389SQu Wenruo * [What does relocation do] 320c891389SQu Wenruo * 330c891389SQu Wenruo * The objective of relocation is to relocate all extents of the target block 340c891389SQu Wenruo * group to other block groups. 350c891389SQu Wenruo * This is utilized by resize (shrink only), profile converting, compacting 360c891389SQu Wenruo * space, or balance routine to spread chunks over devices. 370c891389SQu Wenruo * 380c891389SQu Wenruo * Before | After 390c891389SQu Wenruo * ------------------------------------------------------------------ 400c891389SQu Wenruo * BG A: 10 data extents | BG A: deleted 410c891389SQu Wenruo * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) 420c891389SQu Wenruo * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) 430c891389SQu Wenruo * 440c891389SQu Wenruo * [How does relocation work] 450c891389SQu Wenruo * 460c891389SQu Wenruo * 1. Mark the target block group read-only 470c891389SQu Wenruo * New extents won't be allocated from the target block group. 480c891389SQu Wenruo * 490c891389SQu Wenruo * 2.1 Record each extent in the target block group 500c891389SQu Wenruo * To build a proper map of extents to be relocated. 510c891389SQu Wenruo * 520c891389SQu Wenruo * 2.2 Build data reloc tree and reloc trees 530c891389SQu Wenruo * Data reloc tree will contain an inode, recording all newly relocated 540c891389SQu Wenruo * data extents. 550c891389SQu Wenruo * There will be only one data reloc tree for one data block group. 560c891389SQu Wenruo * 570c891389SQu Wenruo * Reloc tree will be a special snapshot of its source tree, containing 580c891389SQu Wenruo * relocated tree blocks. 590c891389SQu Wenruo * Each tree referring to a tree block in target block group will get its 600c891389SQu Wenruo * reloc tree built. 610c891389SQu Wenruo * 620c891389SQu Wenruo * 2.3 Swap source tree with its corresponding reloc tree 630c891389SQu Wenruo * Each involved tree only refers to new extents after swap. 640c891389SQu Wenruo * 650c891389SQu Wenruo * 3. Cleanup reloc trees and data reloc tree. 660c891389SQu Wenruo * As old extents in the target block group are still referenced by reloc 670c891389SQu Wenruo * trees, we need to clean them up before really freeing the target block 680c891389SQu Wenruo * group. 690c891389SQu Wenruo * 700c891389SQu Wenruo * The main complexity is in steps 2.2 and 2.3. 710c891389SQu Wenruo * 720c891389SQu Wenruo * The entry point of relocation is relocate_block_group() function. 730c891389SQu Wenruo */ 740c891389SQu Wenruo 750c891389SQu Wenruo /* 76a26195a5SQu Wenruo * btrfs_backref_node, mapping_node and tree_block start with this 775d4f98a2SYan Zheng */ 785d4f98a2SYan Zheng struct tree_entry { 795d4f98a2SYan Zheng struct rb_node rb_node; 805d4f98a2SYan Zheng u64 bytenr; 815d4f98a2SYan Zheng }; 825d4f98a2SYan Zheng 835d4f98a2SYan Zheng /* 845d4f98a2SYan Zheng * present a tree block in the backref cache 855d4f98a2SYan Zheng */ 86a26195a5SQu Wenruo struct btrfs_backref_node { 875d4f98a2SYan Zheng struct rb_node rb_node; 885d4f98a2SYan Zheng u64 bytenr; 893fd0a558SYan, Zheng 903fd0a558SYan, Zheng u64 new_bytenr; 913fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 925d4f98a2SYan Zheng u64 owner; 933fd0a558SYan, Zheng /* link to pending, changed or detached list */ 943fd0a558SYan, Zheng struct list_head list; 952a979612SQu Wenruo 962a979612SQu Wenruo /* List of upper level edges, which link this node to its parents */ 975d4f98a2SYan Zheng struct list_head upper; 982a979612SQu Wenruo /* List of lower level edges, which link this node to its children */ 995d4f98a2SYan Zheng struct list_head lower; 1002a979612SQu Wenruo 1015d4f98a2SYan Zheng /* NULL if this node is not tree root */ 1025d4f98a2SYan Zheng struct btrfs_root *root; 1035d4f98a2SYan Zheng /* extent buffer got by COW the block */ 1045d4f98a2SYan Zheng struct extent_buffer *eb; 1055d4f98a2SYan Zheng /* level of tree block */ 1065d4f98a2SYan Zheng unsigned int level:8; 1073fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 1083fd0a558SYan, Zheng unsigned int cowonly:1; 1093fd0a558SYan, Zheng /* 1 if no child node in the cache */ 1105d4f98a2SYan Zheng unsigned int lowest:1; 1115d4f98a2SYan Zheng /* is the extent buffer locked */ 1125d4f98a2SYan Zheng unsigned int locked:1; 1135d4f98a2SYan Zheng /* has the block been processed */ 1145d4f98a2SYan Zheng unsigned int processed:1; 1155d4f98a2SYan Zheng /* have backrefs of this block been checked */ 1165d4f98a2SYan Zheng unsigned int checked:1; 1173fd0a558SYan, Zheng /* 1183fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 1193fd0a558SYan, Zheng * level block pointers may not point to the new location 1203fd0a558SYan, Zheng */ 1213fd0a558SYan, Zheng unsigned int pending:1; 1223fd0a558SYan, Zheng /* 1233fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 1243fd0a558SYan, Zheng * backref node. 1253fd0a558SYan, Zheng */ 1263fd0a558SYan, Zheng unsigned int detached:1; 1272433bea5SQu Wenruo 1282433bea5SQu Wenruo /* 1292433bea5SQu Wenruo * For generic purpose backref cache, where we only care if it's a reloc 1302433bea5SQu Wenruo * root, doesn't care the source subvolid. 1312433bea5SQu Wenruo */ 1322433bea5SQu Wenruo unsigned int is_reloc_root:1; 1335d4f98a2SYan Zheng }; 1345d4f98a2SYan Zheng 1355d4f98a2SYan Zheng #define LOWER 0 1365d4f98a2SYan Zheng #define UPPER 1 1370647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 1382a979612SQu Wenruo /* 1392a979612SQu Wenruo * present an edge connecting upper and lower backref nodes. 1402a979612SQu Wenruo */ 141a26195a5SQu Wenruo struct btrfs_backref_edge { 1422a979612SQu Wenruo /* 143a26195a5SQu Wenruo * list[LOWER] is linked to btrfs_backref_node::upper of lower level 144a26195a5SQu Wenruo * node, and list[UPPER] is linked to btrfs_backref_node::lower of 145a26195a5SQu Wenruo * upper level node. 1462a979612SQu Wenruo * 1472a979612SQu Wenruo * Also, build_backref_tree() uses list[UPPER] for pending edges, before 1482a979612SQu Wenruo * linking list[UPPER] to its upper level nodes. 1492a979612SQu Wenruo */ 1502a979612SQu Wenruo struct list_head list[2]; 1512a979612SQu Wenruo 1522a979612SQu Wenruo /* Two related nodes */ 153a26195a5SQu Wenruo struct btrfs_backref_node *node[2]; 1542a979612SQu Wenruo }; 1552a979612SQu Wenruo 1565d4f98a2SYan Zheng 157a26195a5SQu Wenruo struct btrfs_backref_cache { 1585d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 1595d4f98a2SYan Zheng struct rb_root rb_root; 1603fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 161a26195a5SQu Wenruo struct btrfs_backref_node *path[BTRFS_MAX_LEVEL]; 1623fd0a558SYan, Zheng /* 1633fd0a558SYan, Zheng * list of blocks that have been cowed but some block 1643fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 1653fd0a558SYan, Zheng * new location 1663fd0a558SYan, Zheng */ 1675d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1683fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1693fd0a558SYan, Zheng struct list_head leaves; 1703fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1713fd0a558SYan, Zheng struct list_head changed; 1723fd0a558SYan, Zheng /* list of detached backref node. */ 1733fd0a558SYan, Zheng struct list_head detached; 1743fd0a558SYan, Zheng 1753fd0a558SYan, Zheng u64 last_trans; 1763fd0a558SYan, Zheng 1773fd0a558SYan, Zheng int nr_nodes; 1783fd0a558SYan, Zheng int nr_edges; 17984780289SQu Wenruo 18084780289SQu Wenruo /* The list of unchecked backref edges during backref cache build */ 18184780289SQu Wenruo struct list_head pending_edge; 18284780289SQu Wenruo 18384780289SQu Wenruo /* The list of useless backref nodes during backref cache build */ 18484780289SQu Wenruo struct list_head useless_node; 18533a0f1f7SQu Wenruo 18633a0f1f7SQu Wenruo struct btrfs_fs_info *fs_info; 1872433bea5SQu Wenruo 1882433bea5SQu Wenruo /* 1892433bea5SQu Wenruo * Whether this cache is for relocation 1902433bea5SQu Wenruo * 1912433bea5SQu Wenruo * Reloction backref cache require more info for reloc root compared 1922433bea5SQu Wenruo * to generic backref cache. 1932433bea5SQu Wenruo */ 1942433bea5SQu Wenruo unsigned int is_reloc; 1955d4f98a2SYan Zheng }; 1965d4f98a2SYan Zheng 1975d4f98a2SYan Zheng /* 1985d4f98a2SYan Zheng * map address of tree root to tree 1995d4f98a2SYan Zheng */ 2005d4f98a2SYan Zheng struct mapping_node { 2015d4f98a2SYan Zheng struct rb_node rb_node; 2025d4f98a2SYan Zheng u64 bytenr; 2035d4f98a2SYan Zheng void *data; 2045d4f98a2SYan Zheng }; 2055d4f98a2SYan Zheng 2065d4f98a2SYan Zheng struct mapping_tree { 2075d4f98a2SYan Zheng struct rb_root rb_root; 2085d4f98a2SYan Zheng spinlock_t lock; 2095d4f98a2SYan Zheng }; 2105d4f98a2SYan Zheng 2115d4f98a2SYan Zheng /* 2125d4f98a2SYan Zheng * present a tree block to process 2135d4f98a2SYan Zheng */ 2145d4f98a2SYan Zheng struct tree_block { 2155d4f98a2SYan Zheng struct rb_node rb_node; 2165d4f98a2SYan Zheng u64 bytenr; 2175d4f98a2SYan Zheng struct btrfs_key key; 2185d4f98a2SYan Zheng unsigned int level:8; 2195d4f98a2SYan Zheng unsigned int key_ready:1; 2205d4f98a2SYan Zheng }; 2215d4f98a2SYan Zheng 2220257bb82SYan, Zheng #define MAX_EXTENTS 128 2230257bb82SYan, Zheng 2240257bb82SYan, Zheng struct file_extent_cluster { 2250257bb82SYan, Zheng u64 start; 2260257bb82SYan, Zheng u64 end; 2270257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 2280257bb82SYan, Zheng unsigned int nr; 2290257bb82SYan, Zheng }; 2300257bb82SYan, Zheng 2315d4f98a2SYan Zheng struct reloc_control { 2325d4f98a2SYan Zheng /* block group to relocate */ 23332da5386SDavid Sterba struct btrfs_block_group *block_group; 2345d4f98a2SYan Zheng /* extent tree */ 2355d4f98a2SYan Zheng struct btrfs_root *extent_root; 2365d4f98a2SYan Zheng /* inode for moving data */ 2375d4f98a2SYan Zheng struct inode *data_inode; 2383fd0a558SYan, Zheng 2393fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 2403fd0a558SYan, Zheng 241a26195a5SQu Wenruo struct btrfs_backref_cache backref_cache; 2423fd0a558SYan, Zheng 2433fd0a558SYan, Zheng struct file_extent_cluster cluster; 2445d4f98a2SYan Zheng /* tree blocks have been processed */ 2455d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 2465d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 2475d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 2485d4f98a2SYan Zheng /* list of reloc trees */ 2495d4f98a2SYan Zheng struct list_head reloc_roots; 250d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 251d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 2523fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 2533fd0a558SYan, Zheng u64 merging_rsv_size; 2543fd0a558SYan, Zheng /* size of relocated tree nodes */ 2553fd0a558SYan, Zheng u64 nodes_relocated; 2560647bf56SWang Shilong /* reserved size for block group relocation*/ 2570647bf56SWang Shilong u64 reserved_bytes; 2583fd0a558SYan, Zheng 2595d4f98a2SYan Zheng u64 search_start; 2605d4f98a2SYan Zheng u64 extents_found; 2613fd0a558SYan, Zheng 2623fd0a558SYan, Zheng unsigned int stage:8; 2633fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 2643fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 2655d4f98a2SYan Zheng unsigned int found_file_extent:1; 2665d4f98a2SYan Zheng }; 2675d4f98a2SYan Zheng 2685d4f98a2SYan Zheng /* stages of data relocation */ 2695d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 2705d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 2715d4f98a2SYan Zheng 272a26195a5SQu Wenruo static void remove_backref_node(struct btrfs_backref_cache *cache, 273a26195a5SQu Wenruo struct btrfs_backref_node *node); 2749569cc20SQu Wenruo 2759569cc20SQu Wenruo static void mark_block_processed(struct reloc_control *rc, 276a26195a5SQu Wenruo struct btrfs_backref_node *node) 2779569cc20SQu Wenruo { 2789569cc20SQu Wenruo u32 blocksize; 2799569cc20SQu Wenruo 2809569cc20SQu Wenruo if (node->level == 0 || 2819569cc20SQu Wenruo in_range(node->bytenr, rc->block_group->start, 2829569cc20SQu Wenruo rc->block_group->length)) { 2839569cc20SQu Wenruo blocksize = rc->extent_root->fs_info->nodesize; 2849569cc20SQu Wenruo set_extent_bits(&rc->processed_blocks, node->bytenr, 2859569cc20SQu Wenruo node->bytenr + blocksize - 1, EXTENT_DIRTY); 2869569cc20SQu Wenruo } 2879569cc20SQu Wenruo node->processed = 1; 2889569cc20SQu Wenruo } 2899569cc20SQu Wenruo 2905d4f98a2SYan Zheng 2915d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 2925d4f98a2SYan Zheng { 2936bef4d31SEric Paris tree->rb_root = RB_ROOT; 2945d4f98a2SYan Zheng spin_lock_init(&tree->lock); 2955d4f98a2SYan Zheng } 2965d4f98a2SYan Zheng 29733a0f1f7SQu Wenruo static void backref_cache_init(struct btrfs_fs_info *fs_info, 298a26195a5SQu Wenruo struct btrfs_backref_cache *cache, int is_reloc) 2995d4f98a2SYan Zheng { 3005d4f98a2SYan Zheng int i; 3016bef4d31SEric Paris cache->rb_root = RB_ROOT; 3025d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 3035d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 3043fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 3053fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 3063fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 30784780289SQu Wenruo INIT_LIST_HEAD(&cache->pending_edge); 30884780289SQu Wenruo INIT_LIST_HEAD(&cache->useless_node); 30933a0f1f7SQu Wenruo cache->fs_info = fs_info; 3102433bea5SQu Wenruo cache->is_reloc = is_reloc; 3115d4f98a2SYan Zheng } 3125d4f98a2SYan Zheng 313a26195a5SQu Wenruo static void backref_cache_cleanup(struct btrfs_backref_cache *cache) 3145d4f98a2SYan Zheng { 315a26195a5SQu Wenruo struct btrfs_backref_node *node; 3163fd0a558SYan, Zheng int i; 3173fd0a558SYan, Zheng 3183fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 3193fd0a558SYan, Zheng node = list_entry(cache->detached.next, 320a26195a5SQu Wenruo struct btrfs_backref_node, list); 3213fd0a558SYan, Zheng remove_backref_node(cache, node); 3223fd0a558SYan, Zheng } 3233fd0a558SYan, Zheng 3243fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 3253fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 326a26195a5SQu Wenruo struct btrfs_backref_node, lower); 3273fd0a558SYan, Zheng remove_backref_node(cache, node); 3283fd0a558SYan, Zheng } 3293fd0a558SYan, Zheng 3303fd0a558SYan, Zheng cache->last_trans = 0; 3313fd0a558SYan, Zheng 3323fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 333f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 33484780289SQu Wenruo ASSERT(list_empty(&cache->pending_edge)); 33584780289SQu Wenruo ASSERT(list_empty(&cache->useless_node)); 336f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 337f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 338f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 339f4907095SLiu Bo ASSERT(!cache->nr_nodes); 340f4907095SLiu Bo ASSERT(!cache->nr_edges); 3413fd0a558SYan, Zheng } 3423fd0a558SYan, Zheng 343a26195a5SQu Wenruo static struct btrfs_backref_node *alloc_backref_node( 344a26195a5SQu Wenruo struct btrfs_backref_cache *cache, u64 bytenr, int level) 3453fd0a558SYan, Zheng { 346a26195a5SQu Wenruo struct btrfs_backref_node *node; 3473fd0a558SYan, Zheng 3480304f2d8SQu Wenruo ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL); 3493fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 3500304f2d8SQu Wenruo if (!node) 3510304f2d8SQu Wenruo return node; 3520304f2d8SQu Wenruo 3533fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 3545d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 3555d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 3565d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 3573fd0a558SYan, Zheng cache->nr_nodes++; 3580304f2d8SQu Wenruo node->level = level; 3590304f2d8SQu Wenruo node->bytenr = bytenr; 3600304f2d8SQu Wenruo 3613fd0a558SYan, Zheng return node; 3623fd0a558SYan, Zheng } 3633fd0a558SYan, Zheng 364a26195a5SQu Wenruo static void free_backref_node(struct btrfs_backref_cache *cache, 365a26195a5SQu Wenruo struct btrfs_backref_node *node) 3663fd0a558SYan, Zheng { 3673fd0a558SYan, Zheng if (node) { 3683fd0a558SYan, Zheng cache->nr_nodes--; 36900246528SJosef Bacik btrfs_put_root(node->root); 3703fd0a558SYan, Zheng kfree(node); 3713fd0a558SYan, Zheng } 3723fd0a558SYan, Zheng } 3733fd0a558SYan, Zheng 374a26195a5SQu Wenruo static struct btrfs_backref_edge *alloc_backref_edge( 375a26195a5SQu Wenruo struct btrfs_backref_cache *cache) 3763fd0a558SYan, Zheng { 377a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 3783fd0a558SYan, Zheng 3793fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 3803fd0a558SYan, Zheng if (edge) 3813fd0a558SYan, Zheng cache->nr_edges++; 3823fd0a558SYan, Zheng return edge; 3833fd0a558SYan, Zheng } 3843fd0a558SYan, Zheng 3852a979612SQu Wenruo #define LINK_LOWER (1 << 0) 3862a979612SQu Wenruo #define LINK_UPPER (1 << 1) 387a26195a5SQu Wenruo static void link_backref_edge(struct btrfs_backref_edge *edge, 388a26195a5SQu Wenruo struct btrfs_backref_node *lower, 389a26195a5SQu Wenruo struct btrfs_backref_node *upper, 3902a979612SQu Wenruo int link_which) 3912a979612SQu Wenruo { 3922a979612SQu Wenruo ASSERT(upper && lower && upper->level == lower->level + 1); 3932a979612SQu Wenruo edge->node[LOWER] = lower; 3942a979612SQu Wenruo edge->node[UPPER] = upper; 3952a979612SQu Wenruo if (link_which & LINK_LOWER) 3962a979612SQu Wenruo list_add_tail(&edge->list[LOWER], &lower->upper); 3972a979612SQu Wenruo if (link_which & LINK_UPPER) 3982a979612SQu Wenruo list_add_tail(&edge->list[UPPER], &upper->lower); 3992a979612SQu Wenruo } 4002a979612SQu Wenruo 401a26195a5SQu Wenruo static void free_backref_edge(struct btrfs_backref_cache *cache, 402a26195a5SQu Wenruo struct btrfs_backref_edge *edge) 4033fd0a558SYan, Zheng { 4043fd0a558SYan, Zheng if (edge) { 4053fd0a558SYan, Zheng cache->nr_edges--; 4063fd0a558SYan, Zheng kfree(edge); 4073fd0a558SYan, Zheng } 4085d4f98a2SYan Zheng } 4095d4f98a2SYan Zheng 4105d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 4115d4f98a2SYan Zheng struct rb_node *node) 4125d4f98a2SYan Zheng { 4135d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 4145d4f98a2SYan Zheng struct rb_node *parent = NULL; 4155d4f98a2SYan Zheng struct tree_entry *entry; 4165d4f98a2SYan Zheng 4175d4f98a2SYan Zheng while (*p) { 4185d4f98a2SYan Zheng parent = *p; 4195d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 4205d4f98a2SYan Zheng 4215d4f98a2SYan Zheng if (bytenr < entry->bytenr) 4225d4f98a2SYan Zheng p = &(*p)->rb_left; 4235d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 4245d4f98a2SYan Zheng p = &(*p)->rb_right; 4255d4f98a2SYan Zheng else 4265d4f98a2SYan Zheng return parent; 4275d4f98a2SYan Zheng } 4285d4f98a2SYan Zheng 4295d4f98a2SYan Zheng rb_link_node(node, parent, p); 4305d4f98a2SYan Zheng rb_insert_color(node, root); 4315d4f98a2SYan Zheng return NULL; 4325d4f98a2SYan Zheng } 4335d4f98a2SYan Zheng 4345d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 4355d4f98a2SYan Zheng { 4365d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 4375d4f98a2SYan Zheng struct tree_entry *entry; 4385d4f98a2SYan Zheng 4395d4f98a2SYan Zheng while (n) { 4405d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 4415d4f98a2SYan Zheng 4425d4f98a2SYan Zheng if (bytenr < entry->bytenr) 4435d4f98a2SYan Zheng n = n->rb_left; 4445d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 4455d4f98a2SYan Zheng n = n->rb_right; 4465d4f98a2SYan Zheng else 4475d4f98a2SYan Zheng return n; 4485d4f98a2SYan Zheng } 4495d4f98a2SYan Zheng return NULL; 4505d4f98a2SYan Zheng } 4515d4f98a2SYan Zheng 45248a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 45343c04fb1SJeff Mahoney { 45443c04fb1SJeff Mahoney 45543c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 456a26195a5SQu Wenruo struct btrfs_backref_node *bnode = rb_entry(rb_node, 457a26195a5SQu Wenruo struct btrfs_backref_node, rb_node); 45843c04fb1SJeff Mahoney if (bnode->root) 45943c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 4605d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 4615d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 4625d163e0eSJeff Mahoney bytenr); 46343c04fb1SJeff Mahoney } 46443c04fb1SJeff Mahoney 4655d4f98a2SYan Zheng /* 4665d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 4675d4f98a2SYan Zheng */ 468a26195a5SQu Wenruo static struct btrfs_backref_node *walk_up_backref( 469a26195a5SQu Wenruo struct btrfs_backref_node *node, 470a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 4715d4f98a2SYan Zheng { 472a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 4735d4f98a2SYan Zheng int idx = *index; 4745d4f98a2SYan Zheng 4755d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4765d4f98a2SYan Zheng edge = list_entry(node->upper.next, 477a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 4785d4f98a2SYan Zheng edges[idx++] = edge; 4795d4f98a2SYan Zheng node = edge->node[UPPER]; 4805d4f98a2SYan Zheng } 4813fd0a558SYan, Zheng BUG_ON(node->detached); 4825d4f98a2SYan Zheng *index = idx; 4835d4f98a2SYan Zheng return node; 4845d4f98a2SYan Zheng } 4855d4f98a2SYan Zheng 4865d4f98a2SYan Zheng /* 4875d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 4885d4f98a2SYan Zheng */ 489a26195a5SQu Wenruo static struct btrfs_backref_node *walk_down_backref( 490a26195a5SQu Wenruo struct btrfs_backref_edge *edges[], int *index) 4915d4f98a2SYan Zheng { 492a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 493a26195a5SQu Wenruo struct btrfs_backref_node *lower; 4945d4f98a2SYan Zheng int idx = *index; 4955d4f98a2SYan Zheng 4965d4f98a2SYan Zheng while (idx > 0) { 4975d4f98a2SYan Zheng edge = edges[idx - 1]; 4985d4f98a2SYan Zheng lower = edge->node[LOWER]; 4995d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 5005d4f98a2SYan Zheng idx--; 5015d4f98a2SYan Zheng continue; 5025d4f98a2SYan Zheng } 5035d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 504a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 5055d4f98a2SYan Zheng edges[idx - 1] = edge; 5065d4f98a2SYan Zheng *index = idx; 5075d4f98a2SYan Zheng return edge->node[UPPER]; 5085d4f98a2SYan Zheng } 5095d4f98a2SYan Zheng *index = 0; 5105d4f98a2SYan Zheng return NULL; 5115d4f98a2SYan Zheng } 5125d4f98a2SYan Zheng 513a26195a5SQu Wenruo static void unlock_node_buffer(struct btrfs_backref_node *node) 5145d4f98a2SYan Zheng { 5155d4f98a2SYan Zheng if (node->locked) { 5165d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 5175d4f98a2SYan Zheng node->locked = 0; 5185d4f98a2SYan Zheng } 5193fd0a558SYan, Zheng } 5203fd0a558SYan, Zheng 521a26195a5SQu Wenruo static void drop_node_buffer(struct btrfs_backref_node *node) 5223fd0a558SYan, Zheng { 5233fd0a558SYan, Zheng if (node->eb) { 5243fd0a558SYan, Zheng unlock_node_buffer(node); 5255d4f98a2SYan Zheng free_extent_buffer(node->eb); 5265d4f98a2SYan Zheng node->eb = NULL; 5275d4f98a2SYan Zheng } 5285d4f98a2SYan Zheng } 5295d4f98a2SYan Zheng 530a26195a5SQu Wenruo static void drop_backref_node(struct btrfs_backref_cache *tree, 531a26195a5SQu Wenruo struct btrfs_backref_node *node) 5325d4f98a2SYan Zheng { 5335d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 5345d4f98a2SYan Zheng 5355d4f98a2SYan Zheng drop_node_buffer(node); 5363fd0a558SYan, Zheng list_del(&node->list); 5375d4f98a2SYan Zheng list_del(&node->lower); 5383fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 5395d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 5403fd0a558SYan, Zheng free_backref_node(tree, node); 5415d4f98a2SYan Zheng } 5425d4f98a2SYan Zheng 5435d4f98a2SYan Zheng /* 5445d4f98a2SYan Zheng * remove a backref node from the backref cache 5455d4f98a2SYan Zheng */ 546a26195a5SQu Wenruo static void remove_backref_node(struct btrfs_backref_cache *cache, 547a26195a5SQu Wenruo struct btrfs_backref_node *node) 5485d4f98a2SYan Zheng { 549a26195a5SQu Wenruo struct btrfs_backref_node *upper; 550a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 5515d4f98a2SYan Zheng 5525d4f98a2SYan Zheng if (!node) 5535d4f98a2SYan Zheng return; 5545d4f98a2SYan Zheng 5553fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 5565d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 557a26195a5SQu Wenruo edge = list_entry(node->upper.next, struct btrfs_backref_edge, 5585d4f98a2SYan Zheng list[LOWER]); 5595d4f98a2SYan Zheng upper = edge->node[UPPER]; 5605d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 5615d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 5623fd0a558SYan, Zheng free_backref_edge(cache, edge); 5633fd0a558SYan, Zheng 5643fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 5653fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 5663fd0a558SYan, Zheng drop_backref_node(cache, node); 5673fd0a558SYan, Zheng node = upper; 5683fd0a558SYan, Zheng node->lowest = 1; 5693fd0a558SYan, Zheng continue; 5703fd0a558SYan, Zheng } 5715d4f98a2SYan Zheng /* 5723fd0a558SYan, Zheng * add the node to leaf node list if no other 5735d4f98a2SYan Zheng * child block cached. 5745d4f98a2SYan Zheng */ 5755d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 5763fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 5775d4f98a2SYan Zheng upper->lowest = 1; 5785d4f98a2SYan Zheng } 5795d4f98a2SYan Zheng } 5803fd0a558SYan, Zheng 5815d4f98a2SYan Zheng drop_backref_node(cache, node); 5825d4f98a2SYan Zheng } 5835d4f98a2SYan Zheng 584a26195a5SQu Wenruo static void update_backref_node(struct btrfs_backref_cache *cache, 585a26195a5SQu Wenruo struct btrfs_backref_node *node, u64 bytenr) 5863fd0a558SYan, Zheng { 5873fd0a558SYan, Zheng struct rb_node *rb_node; 5883fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 5893fd0a558SYan, Zheng node->bytenr = bytenr; 5903fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 59143c04fb1SJeff Mahoney if (rb_node) 59243c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 5933fd0a558SYan, Zheng } 5943fd0a558SYan, Zheng 5953fd0a558SYan, Zheng /* 5963fd0a558SYan, Zheng * update backref cache after a transaction commit 5973fd0a558SYan, Zheng */ 5983fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 599a26195a5SQu Wenruo struct btrfs_backref_cache *cache) 6003fd0a558SYan, Zheng { 601a26195a5SQu Wenruo struct btrfs_backref_node *node; 6023fd0a558SYan, Zheng int level = 0; 6033fd0a558SYan, Zheng 6043fd0a558SYan, Zheng if (cache->last_trans == 0) { 6053fd0a558SYan, Zheng cache->last_trans = trans->transid; 6063fd0a558SYan, Zheng return 0; 6073fd0a558SYan, Zheng } 6083fd0a558SYan, Zheng 6093fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 6103fd0a558SYan, Zheng return 0; 6113fd0a558SYan, Zheng 6123fd0a558SYan, Zheng /* 6133fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 6143fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 6153fd0a558SYan, Zheng * so the detached nodes are no longer useful. 6163fd0a558SYan, Zheng */ 6173fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 6183fd0a558SYan, Zheng node = list_entry(cache->detached.next, 619a26195a5SQu Wenruo struct btrfs_backref_node, list); 6203fd0a558SYan, Zheng remove_backref_node(cache, node); 6213fd0a558SYan, Zheng } 6223fd0a558SYan, Zheng 6233fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 6243fd0a558SYan, Zheng node = list_entry(cache->changed.next, 625a26195a5SQu Wenruo struct btrfs_backref_node, list); 6263fd0a558SYan, Zheng list_del_init(&node->list); 6273fd0a558SYan, Zheng BUG_ON(node->pending); 6283fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 6293fd0a558SYan, Zheng } 6303fd0a558SYan, Zheng 6313fd0a558SYan, Zheng /* 6323fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 6333fd0a558SYan, Zheng * errors during processing the pending nodes. 6343fd0a558SYan, Zheng */ 6353fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 6363fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 6373fd0a558SYan, Zheng BUG_ON(!node->pending); 6383fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 6393fd0a558SYan, Zheng continue; 6403fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 6413fd0a558SYan, Zheng } 6423fd0a558SYan, Zheng } 6433fd0a558SYan, Zheng 6443fd0a558SYan, Zheng cache->last_trans = 0; 6453fd0a558SYan, Zheng return 1; 6463fd0a558SYan, Zheng } 6473fd0a558SYan, Zheng 6486282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 6496282675eSQu Wenruo { 6506282675eSQu Wenruo /* 6516282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 6526282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 6536282675eSQu Wenruo * trying to access reloc_root 6546282675eSQu Wenruo */ 6556282675eSQu Wenruo smp_rmb(); 6566282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 6576282675eSQu Wenruo return true; 6586282675eSQu Wenruo return false; 6596282675eSQu Wenruo } 6606282675eSQu Wenruo 6616282675eSQu Wenruo /* 6626282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 6636282675eSQu Wenruo * 6646282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 6656282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 6666282675eSQu Wenruo * from no reloc root. But should_ignore_root() below is a special case. 6676282675eSQu Wenruo */ 6686282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 6696282675eSQu Wenruo { 6706282675eSQu Wenruo if (reloc_root_is_dead(root)) 6716282675eSQu Wenruo return false; 6726282675eSQu Wenruo if (!root->reloc_root) 6736282675eSQu Wenruo return false; 6746282675eSQu Wenruo return true; 6756282675eSQu Wenruo } 676f2a97a9dSDavid Sterba 6773fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 6783fd0a558SYan, Zheng { 6793fd0a558SYan, Zheng struct btrfs_root *reloc_root; 6803fd0a558SYan, Zheng 68127cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 6823fd0a558SYan, Zheng return 0; 6833fd0a558SYan, Zheng 6846282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 6856282675eSQu Wenruo if (reloc_root_is_dead(root)) 6866282675eSQu Wenruo return 1; 6876282675eSQu Wenruo 6883fd0a558SYan, Zheng reloc_root = root->reloc_root; 6893fd0a558SYan, Zheng if (!reloc_root) 6903fd0a558SYan, Zheng return 0; 6913fd0a558SYan, Zheng 6924d4225fcSJosef Bacik if (btrfs_header_generation(reloc_root->commit_root) == 6934d4225fcSJosef Bacik root->fs_info->running_transaction->transid) 6943fd0a558SYan, Zheng return 0; 6953fd0a558SYan, Zheng /* 6963fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 6973fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 6983fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 6993fd0a558SYan, Zheng * relocation. 7003fd0a558SYan, Zheng */ 7013fd0a558SYan, Zheng return 1; 7023fd0a558SYan, Zheng } 7035d4f98a2SYan Zheng /* 7045d4f98a2SYan Zheng * find reloc tree by address of tree root 7055d4f98a2SYan Zheng */ 7062433bea5SQu Wenruo struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) 7075d4f98a2SYan Zheng { 7082433bea5SQu Wenruo struct reloc_control *rc = fs_info->reloc_ctl; 7095d4f98a2SYan Zheng struct rb_node *rb_node; 7105d4f98a2SYan Zheng struct mapping_node *node; 7115d4f98a2SYan Zheng struct btrfs_root *root = NULL; 7125d4f98a2SYan Zheng 7132433bea5SQu Wenruo ASSERT(rc); 7145d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 7155d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 7165d4f98a2SYan Zheng if (rb_node) { 7175d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 7185d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 7195d4f98a2SYan Zheng } 7205d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 72100246528SJosef Bacik return btrfs_grab_root(root); 7225d4f98a2SYan Zheng } 7235d4f98a2SYan Zheng 7245d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 7255d4f98a2SYan Zheng u64 root_objectid) 7265d4f98a2SYan Zheng { 7275d4f98a2SYan Zheng struct btrfs_key key; 7285d4f98a2SYan Zheng 7295d4f98a2SYan Zheng key.objectid = root_objectid; 7305d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 7315d4f98a2SYan Zheng key.offset = (u64)-1; 7325d4f98a2SYan Zheng 733bc44d7c4SJosef Bacik return btrfs_get_fs_root(fs_info, &key, false); 7345d4f98a2SYan Zheng } 7355d4f98a2SYan Zheng 7365d4f98a2SYan Zheng /* 7374007ea87SQu Wenruo * Handle direct tree backref 7384007ea87SQu Wenruo * 7394007ea87SQu Wenruo * Direct tree backref means, the backref item shows its parent bytenr 7404007ea87SQu Wenruo * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined). 7414007ea87SQu Wenruo * 7424007ea87SQu Wenruo * @ref_key: The converted backref key. 7434007ea87SQu Wenruo * For keyed backref, it's the item key. 7444007ea87SQu Wenruo * For inlined backref, objectid is the bytenr, 7454007ea87SQu Wenruo * type is btrfs_inline_ref_type, offset is 7464007ea87SQu Wenruo * btrfs_inline_ref_offset. 7474007ea87SQu Wenruo */ 748a26195a5SQu Wenruo static int handle_direct_tree_backref(struct btrfs_backref_cache *cache, 7494007ea87SQu Wenruo struct btrfs_key *ref_key, 750a26195a5SQu Wenruo struct btrfs_backref_node *cur) 7514007ea87SQu Wenruo { 752a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 753a26195a5SQu Wenruo struct btrfs_backref_node *upper; 7544007ea87SQu Wenruo struct rb_node *rb_node; 7554007ea87SQu Wenruo 7564007ea87SQu Wenruo ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY); 7574007ea87SQu Wenruo 7584007ea87SQu Wenruo /* Only reloc root uses backref pointing to itself */ 7594007ea87SQu Wenruo if (ref_key->objectid == ref_key->offset) { 7604007ea87SQu Wenruo struct btrfs_root *root; 7614007ea87SQu Wenruo 7624007ea87SQu Wenruo cur->is_reloc_root = 1; 7634007ea87SQu Wenruo /* Only reloc backref cache cares about a specific root */ 7644007ea87SQu Wenruo if (cache->is_reloc) { 7654007ea87SQu Wenruo root = find_reloc_root(cache->fs_info, cur->bytenr); 7664007ea87SQu Wenruo if (WARN_ON(!root)) 7674007ea87SQu Wenruo return -ENOENT; 7684007ea87SQu Wenruo cur->root = root; 7694007ea87SQu Wenruo } else { 7704007ea87SQu Wenruo /* 7714007ea87SQu Wenruo * For generic purpose backref cache, reloc root node 7724007ea87SQu Wenruo * is useless. 7734007ea87SQu Wenruo */ 7744007ea87SQu Wenruo list_add(&cur->list, &cache->useless_node); 7754007ea87SQu Wenruo } 7764007ea87SQu Wenruo return 0; 7774007ea87SQu Wenruo } 7784007ea87SQu Wenruo 7794007ea87SQu Wenruo edge = alloc_backref_edge(cache); 7804007ea87SQu Wenruo if (!edge) 7814007ea87SQu Wenruo return -ENOMEM; 7824007ea87SQu Wenruo 7834007ea87SQu Wenruo rb_node = tree_search(&cache->rb_root, ref_key->offset); 7844007ea87SQu Wenruo if (!rb_node) { 7854007ea87SQu Wenruo /* Parent node not yet cached */ 7860304f2d8SQu Wenruo upper = alloc_backref_node(cache, ref_key->offset, 7870304f2d8SQu Wenruo cur->level + 1); 7884007ea87SQu Wenruo if (!upper) { 7894007ea87SQu Wenruo free_backref_edge(cache, edge); 7904007ea87SQu Wenruo return -ENOMEM; 7914007ea87SQu Wenruo } 7924007ea87SQu Wenruo 7934007ea87SQu Wenruo /* 7944007ea87SQu Wenruo * Backrefs for the upper level block isn't cached, add the 7954007ea87SQu Wenruo * block to pending list 7964007ea87SQu Wenruo */ 7974007ea87SQu Wenruo list_add_tail(&edge->list[UPPER], &cache->pending_edge); 7984007ea87SQu Wenruo } else { 7994007ea87SQu Wenruo /* Parent node already cached */ 800a26195a5SQu Wenruo upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node); 8014007ea87SQu Wenruo ASSERT(upper->checked); 8024007ea87SQu Wenruo INIT_LIST_HEAD(&edge->list[UPPER]); 8034007ea87SQu Wenruo } 8042a979612SQu Wenruo link_backref_edge(edge, cur, upper, LINK_LOWER); 8054007ea87SQu Wenruo return 0; 8064007ea87SQu Wenruo } 8074007ea87SQu Wenruo 8084007ea87SQu Wenruo /* 8094d81ea8bSQu Wenruo * Handle indirect tree backref 8104d81ea8bSQu Wenruo * 8114d81ea8bSQu Wenruo * Indirect tree backref means, we only know which tree the node belongs to. 8124d81ea8bSQu Wenruo * We still need to do a tree search to find out the parents. This is for 8134d81ea8bSQu Wenruo * TREE_BLOCK_REF backref (keyed or inlined). 8144d81ea8bSQu Wenruo * 8154d81ea8bSQu Wenruo * @ref_key: The same as @ref_key in handle_direct_tree_backref() 8164d81ea8bSQu Wenruo * @tree_key: The first key of this tree block. 8174d81ea8bSQu Wenruo * @path: A clean (released) path, to avoid allocating path everytime 8184d81ea8bSQu Wenruo * the function get called. 8194d81ea8bSQu Wenruo */ 820a26195a5SQu Wenruo static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache, 8214d81ea8bSQu Wenruo struct btrfs_path *path, 8224d81ea8bSQu Wenruo struct btrfs_key *ref_key, 8234d81ea8bSQu Wenruo struct btrfs_key *tree_key, 824a26195a5SQu Wenruo struct btrfs_backref_node *cur) 8254d81ea8bSQu Wenruo { 8264d81ea8bSQu Wenruo struct btrfs_fs_info *fs_info = cache->fs_info; 827a26195a5SQu Wenruo struct btrfs_backref_node *upper; 828a26195a5SQu Wenruo struct btrfs_backref_node *lower; 829a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 8304d81ea8bSQu Wenruo struct extent_buffer *eb; 8314d81ea8bSQu Wenruo struct btrfs_root *root; 8324d81ea8bSQu Wenruo struct rb_node *rb_node; 8334d81ea8bSQu Wenruo int level; 8344d81ea8bSQu Wenruo bool need_check = true; 8354d81ea8bSQu Wenruo int ret; 8364d81ea8bSQu Wenruo 8374d81ea8bSQu Wenruo root = read_fs_root(fs_info, ref_key->offset); 8384d81ea8bSQu Wenruo if (IS_ERR(root)) 8394d81ea8bSQu Wenruo return PTR_ERR(root); 8404d81ea8bSQu Wenruo if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 8414d81ea8bSQu Wenruo cur->cowonly = 1; 8424d81ea8bSQu Wenruo 8434d81ea8bSQu Wenruo if (btrfs_root_level(&root->root_item) == cur->level) { 8444d81ea8bSQu Wenruo /* Tree root */ 8454d81ea8bSQu Wenruo ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr); 8464d81ea8bSQu Wenruo if (should_ignore_root(root)) { 8474d81ea8bSQu Wenruo btrfs_put_root(root); 8484d81ea8bSQu Wenruo list_add(&cur->list, &cache->useless_node); 8494d81ea8bSQu Wenruo } else { 8504d81ea8bSQu Wenruo cur->root = root; 8514d81ea8bSQu Wenruo } 8524d81ea8bSQu Wenruo return 0; 8534d81ea8bSQu Wenruo } 8544d81ea8bSQu Wenruo 8554d81ea8bSQu Wenruo level = cur->level + 1; 8564d81ea8bSQu Wenruo 8574d81ea8bSQu Wenruo /* Search the tree to find parent blocks referring to the block */ 8584d81ea8bSQu Wenruo path->search_commit_root = 1; 8594d81ea8bSQu Wenruo path->skip_locking = 1; 8604d81ea8bSQu Wenruo path->lowest_level = level; 8614d81ea8bSQu Wenruo ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0); 8624d81ea8bSQu Wenruo path->lowest_level = 0; 8634d81ea8bSQu Wenruo if (ret < 0) { 8644d81ea8bSQu Wenruo btrfs_put_root(root); 8654d81ea8bSQu Wenruo return ret; 8664d81ea8bSQu Wenruo } 8674d81ea8bSQu Wenruo if (ret > 0 && path->slots[level] > 0) 8684d81ea8bSQu Wenruo path->slots[level]--; 8694d81ea8bSQu Wenruo 8704d81ea8bSQu Wenruo eb = path->nodes[level]; 8714d81ea8bSQu Wenruo if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) { 8724d81ea8bSQu Wenruo btrfs_err(fs_info, 8734d81ea8bSQu Wenruo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 8744d81ea8bSQu Wenruo cur->bytenr, level - 1, root->root_key.objectid, 8754d81ea8bSQu Wenruo tree_key->objectid, tree_key->type, tree_key->offset); 8764d81ea8bSQu Wenruo btrfs_put_root(root); 8774d81ea8bSQu Wenruo ret = -ENOENT; 8784d81ea8bSQu Wenruo goto out; 8794d81ea8bSQu Wenruo } 8804d81ea8bSQu Wenruo lower = cur; 8814d81ea8bSQu Wenruo 8824d81ea8bSQu Wenruo /* Add all nodes and edges in the path */ 8834d81ea8bSQu Wenruo for (; level < BTRFS_MAX_LEVEL; level++) { 8844d81ea8bSQu Wenruo if (!path->nodes[level]) { 8854d81ea8bSQu Wenruo ASSERT(btrfs_root_bytenr(&root->root_item) == 8864d81ea8bSQu Wenruo lower->bytenr); 8874d81ea8bSQu Wenruo if (should_ignore_root(root)) { 8884d81ea8bSQu Wenruo btrfs_put_root(root); 8894d81ea8bSQu Wenruo list_add(&lower->list, &cache->useless_node); 8904d81ea8bSQu Wenruo } else { 8914d81ea8bSQu Wenruo lower->root = root; 8924d81ea8bSQu Wenruo } 8934d81ea8bSQu Wenruo break; 8944d81ea8bSQu Wenruo } 8954d81ea8bSQu Wenruo 8964d81ea8bSQu Wenruo edge = alloc_backref_edge(cache); 8974d81ea8bSQu Wenruo if (!edge) { 8984d81ea8bSQu Wenruo btrfs_put_root(root); 8994d81ea8bSQu Wenruo ret = -ENOMEM; 9004d81ea8bSQu Wenruo goto out; 9014d81ea8bSQu Wenruo } 9024d81ea8bSQu Wenruo 9034d81ea8bSQu Wenruo eb = path->nodes[level]; 9044d81ea8bSQu Wenruo rb_node = tree_search(&cache->rb_root, eb->start); 9054d81ea8bSQu Wenruo if (!rb_node) { 9060304f2d8SQu Wenruo upper = alloc_backref_node(cache, eb->start, 9070304f2d8SQu Wenruo lower->level + 1); 9084d81ea8bSQu Wenruo if (!upper) { 9094d81ea8bSQu Wenruo btrfs_put_root(root); 9104d81ea8bSQu Wenruo free_backref_edge(cache, edge); 9114d81ea8bSQu Wenruo ret = -ENOMEM; 9124d81ea8bSQu Wenruo goto out; 9134d81ea8bSQu Wenruo } 9144d81ea8bSQu Wenruo upper->owner = btrfs_header_owner(eb); 9154d81ea8bSQu Wenruo if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 9164d81ea8bSQu Wenruo upper->cowonly = 1; 9174d81ea8bSQu Wenruo 9184d81ea8bSQu Wenruo /* 9194d81ea8bSQu Wenruo * If we know the block isn't shared we can avoid 9204d81ea8bSQu Wenruo * checking its backrefs. 9214d81ea8bSQu Wenruo */ 9224d81ea8bSQu Wenruo if (btrfs_block_can_be_shared(root, eb)) 9234d81ea8bSQu Wenruo upper->checked = 0; 9244d81ea8bSQu Wenruo else 9254d81ea8bSQu Wenruo upper->checked = 1; 9264d81ea8bSQu Wenruo 9274d81ea8bSQu Wenruo /* 9284d81ea8bSQu Wenruo * Add the block to pending list if we need to check its 9294d81ea8bSQu Wenruo * backrefs, we only do this once while walking up a 9304d81ea8bSQu Wenruo * tree as we will catch anything else later on. 9314d81ea8bSQu Wenruo */ 9324d81ea8bSQu Wenruo if (!upper->checked && need_check) { 9334d81ea8bSQu Wenruo need_check = false; 9344d81ea8bSQu Wenruo list_add_tail(&edge->list[UPPER], 9354d81ea8bSQu Wenruo &cache->pending_edge); 9364d81ea8bSQu Wenruo } else { 9374d81ea8bSQu Wenruo if (upper->checked) 9384d81ea8bSQu Wenruo need_check = true; 9394d81ea8bSQu Wenruo INIT_LIST_HEAD(&edge->list[UPPER]); 9404d81ea8bSQu Wenruo } 9414d81ea8bSQu Wenruo } else { 942a26195a5SQu Wenruo upper = rb_entry(rb_node, struct btrfs_backref_node, 943a26195a5SQu Wenruo rb_node); 9444d81ea8bSQu Wenruo ASSERT(upper->checked); 9454d81ea8bSQu Wenruo INIT_LIST_HEAD(&edge->list[UPPER]); 9464d81ea8bSQu Wenruo if (!upper->owner) 9474d81ea8bSQu Wenruo upper->owner = btrfs_header_owner(eb); 9484d81ea8bSQu Wenruo } 9492a979612SQu Wenruo link_backref_edge(edge, lower, upper, LINK_LOWER); 9504d81ea8bSQu Wenruo 9514d81ea8bSQu Wenruo if (rb_node) { 9524d81ea8bSQu Wenruo btrfs_put_root(root); 9534d81ea8bSQu Wenruo break; 9544d81ea8bSQu Wenruo } 9554d81ea8bSQu Wenruo lower = upper; 9564d81ea8bSQu Wenruo upper = NULL; 9574d81ea8bSQu Wenruo } 9584d81ea8bSQu Wenruo out: 9594d81ea8bSQu Wenruo btrfs_release_path(path); 9604d81ea8bSQu Wenruo return ret; 9614d81ea8bSQu Wenruo } 9624d81ea8bSQu Wenruo 963a26195a5SQu Wenruo static int handle_one_tree_block(struct btrfs_backref_cache *cache, 964e7d571c7SQu Wenruo struct btrfs_path *path, 965e7d571c7SQu Wenruo struct btrfs_backref_iter *iter, 9665d4f98a2SYan Zheng struct btrfs_key *node_key, 967a26195a5SQu Wenruo struct btrfs_backref_node *cur) 9685d4f98a2SYan Zheng { 969e7d571c7SQu Wenruo struct btrfs_fs_info *fs_info = cache->fs_info; 970a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 971a26195a5SQu Wenruo struct btrfs_backref_node *exist; 9725d4f98a2SYan Zheng int ret; 9735d4f98a2SYan Zheng 97471f572a9SQu Wenruo ret = btrfs_backref_iter_start(iter, cur->bytenr); 975e7d571c7SQu Wenruo if (ret < 0) 976e7d571c7SQu Wenruo return ret; 97771f572a9SQu Wenruo /* 97871f572a9SQu Wenruo * We skip the first btrfs_tree_block_info, as we don't use the key 97971f572a9SQu Wenruo * stored in it, but fetch it from the tree block 98071f572a9SQu Wenruo */ 98171f572a9SQu Wenruo if (btrfs_backref_has_tree_block_info(iter)) { 98271f572a9SQu Wenruo ret = btrfs_backref_iter_next(iter); 983e7d571c7SQu Wenruo if (ret < 0) 98471f572a9SQu Wenruo goto out; 98571f572a9SQu Wenruo /* No extra backref? This means the tree block is corrupted */ 98671f572a9SQu Wenruo if (ret > 0) { 987e7d571c7SQu Wenruo ret = -EUCLEAN; 98871f572a9SQu Wenruo goto out; 98971f572a9SQu Wenruo } 99071f572a9SQu Wenruo } 9915d4f98a2SYan Zheng WARN_ON(cur->checked); 9925d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 9935d4f98a2SYan Zheng /* 99470f23fd6SJustin P. Mattock * the backref was added previously when processing 9955d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 9965d4f98a2SYan Zheng */ 99775bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 998a26195a5SQu Wenruo edge = list_entry(cur->upper.next, struct btrfs_backref_edge, 9995d4f98a2SYan Zheng list[LOWER]); 100075bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 10015d4f98a2SYan Zheng exist = edge->node[UPPER]; 10025d4f98a2SYan Zheng /* 10035d4f98a2SYan Zheng * add the upper level block to pending list if we need 10045d4f98a2SYan Zheng * check its backrefs 10055d4f98a2SYan Zheng */ 10065d4f98a2SYan Zheng if (!exist->checked) 100784780289SQu Wenruo list_add_tail(&edge->list[UPPER], &cache->pending_edge); 10085d4f98a2SYan Zheng } else { 10095d4f98a2SYan Zheng exist = NULL; 10105d4f98a2SYan Zheng } 10115d4f98a2SYan Zheng 101271f572a9SQu Wenruo for (; ret == 0; ret = btrfs_backref_iter_next(iter)) { 101371f572a9SQu Wenruo struct extent_buffer *eb; 101471f572a9SQu Wenruo struct btrfs_key key; 10153de28d57SLiu Bo int type; 101671f572a9SQu Wenruo 101771f572a9SQu Wenruo cond_resched(); 101871f572a9SQu Wenruo eb = btrfs_backref_get_eb(iter); 101971f572a9SQu Wenruo 102071f572a9SQu Wenruo key.objectid = iter->bytenr; 102171f572a9SQu Wenruo if (btrfs_backref_iter_is_inline_ref(iter)) { 102271f572a9SQu Wenruo struct btrfs_extent_inline_ref *iref; 102371f572a9SQu Wenruo 102471f572a9SQu Wenruo /* update key for inline back ref */ 102571f572a9SQu Wenruo iref = (struct btrfs_extent_inline_ref *) 102671f572a9SQu Wenruo ((unsigned long)iter->cur_ptr); 10273de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 10283de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 10293de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 1030e7d571c7SQu Wenruo ret = -EUCLEAN; 10313de28d57SLiu Bo goto out; 10323de28d57SLiu Bo } 10333de28d57SLiu Bo key.type = type; 10345d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 103571f572a9SQu Wenruo } else { 103671f572a9SQu Wenruo key.type = iter->cur_key.type; 103771f572a9SQu Wenruo key.offset = iter->cur_key.offset; 10385d4f98a2SYan Zheng } 10395d4f98a2SYan Zheng 1040fa6ac715SQu Wenruo /* 1041fa6ac715SQu Wenruo * Parent node found and matches current inline ref, no need to 1042fa6ac715SQu Wenruo * rebuild this node for this inline ref. 1043fa6ac715SQu Wenruo */ 10445d4f98a2SYan Zheng if (exist && 10455d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 10465d4f98a2SYan Zheng exist->owner == key.offset) || 10475d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 10485d4f98a2SYan Zheng exist->bytenr == key.offset))) { 10495d4f98a2SYan Zheng exist = NULL; 105071f572a9SQu Wenruo continue; 10515d4f98a2SYan Zheng } 10525d4f98a2SYan Zheng 1053fa6ac715SQu Wenruo /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ 10545d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 10554007ea87SQu Wenruo ret = handle_direct_tree_backref(cache, &key, cur); 1056e7d571c7SQu Wenruo if (ret < 0) 10572433bea5SQu Wenruo goto out; 105871f572a9SQu Wenruo continue; 10596d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 1060e7d571c7SQu Wenruo ret = -EINVAL; 1061e7d571c7SQu Wenruo btrfs_print_v0_err(fs_info); 1062e7d571c7SQu Wenruo btrfs_handle_fs_error(fs_info, ret, NULL); 1063ba3c2b19SNikolay Borisov goto out; 10645d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 106571f572a9SQu Wenruo continue; 10665d4f98a2SYan Zheng } 10675d4f98a2SYan Zheng 1068fa6ac715SQu Wenruo /* 1069fa6ac715SQu Wenruo * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset 1070fa6ac715SQu Wenruo * means the root objectid. We need to search the tree to get 1071fa6ac715SQu Wenruo * its parent bytenr. 1072fa6ac715SQu Wenruo */ 10734d81ea8bSQu Wenruo ret = handle_indirect_tree_backref(cache, path, &key, node_key, 10744d81ea8bSQu Wenruo cur); 1075e7d571c7SQu Wenruo if (ret < 0) 107671f572a9SQu Wenruo goto out; 10775d4f98a2SYan Zheng } 107871f572a9SQu Wenruo ret = 0; 10795d4f98a2SYan Zheng cur->checked = 1; 10805d4f98a2SYan Zheng WARN_ON(exist); 1081e7d571c7SQu Wenruo out: 1082e7d571c7SQu Wenruo btrfs_backref_iter_release(iter); 1083e7d571c7SQu Wenruo return ret; 1084e7d571c7SQu Wenruo } 10855d4f98a2SYan Zheng 1086e7d571c7SQu Wenruo /* 10871f872924SQu Wenruo * In handle_one_tree_backref(), we have only linked the lower node to the edge, 10881f872924SQu Wenruo * but the upper node hasn't been linked to the edge. 1089a26195a5SQu Wenruo * This means we can only iterate through btrfs_backref_node::upper to reach 1090a26195a5SQu Wenruo * parent edges, but not through btrfs_backref_node::lower to reach children 1091a26195a5SQu Wenruo * edges. 10921f872924SQu Wenruo * 1093a26195a5SQu Wenruo * This function will finish the btrfs_backref_node::lower to related edges, 1094a26195a5SQu Wenruo * so that backref cache can be bi-directionally iterated. 10951f872924SQu Wenruo * 10961f872924SQu Wenruo * Also, this will add the nodes to backref cache for the next run. 10971f872924SQu Wenruo */ 1098a26195a5SQu Wenruo static int finish_upper_links(struct btrfs_backref_cache *cache, 1099a26195a5SQu Wenruo struct btrfs_backref_node *start) 11001f872924SQu Wenruo { 11011f872924SQu Wenruo struct list_head *useless_node = &cache->useless_node; 1102a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 11031f872924SQu Wenruo struct rb_node *rb_node; 11041f872924SQu Wenruo LIST_HEAD(pending_edge); 11051f872924SQu Wenruo 11061f872924SQu Wenruo ASSERT(start->checked); 11071f872924SQu Wenruo 11081f872924SQu Wenruo /* Insert this node to cache if it's not COW-only */ 11091f872924SQu Wenruo if (!start->cowonly) { 11101f872924SQu Wenruo rb_node = tree_insert(&cache->rb_root, start->bytenr, 11111f872924SQu Wenruo &start->rb_node); 11121f872924SQu Wenruo if (rb_node) 11131f872924SQu Wenruo backref_tree_panic(rb_node, -EEXIST, start->bytenr); 11141f872924SQu Wenruo list_add_tail(&start->lower, &cache->leaves); 11151f872924SQu Wenruo } 11161f872924SQu Wenruo 11171f872924SQu Wenruo /* 11181f872924SQu Wenruo * Use breadth first search to iterate all related edges. 11191f872924SQu Wenruo * 11201f872924SQu Wenruo * The starting points are all the edges of this node 11211f872924SQu Wenruo */ 11221f872924SQu Wenruo list_for_each_entry(edge, &start->upper, list[LOWER]) 11231f872924SQu Wenruo list_add_tail(&edge->list[UPPER], &pending_edge); 11241f872924SQu Wenruo 11251f872924SQu Wenruo while (!list_empty(&pending_edge)) { 1126a26195a5SQu Wenruo struct btrfs_backref_node *upper; 1127a26195a5SQu Wenruo struct btrfs_backref_node *lower; 11281f872924SQu Wenruo struct rb_node *rb_node; 11291f872924SQu Wenruo 1130a26195a5SQu Wenruo edge = list_first_entry(&pending_edge, 1131a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 11321f872924SQu Wenruo list_del_init(&edge->list[UPPER]); 11331f872924SQu Wenruo upper = edge->node[UPPER]; 11341f872924SQu Wenruo lower = edge->node[LOWER]; 11351f872924SQu Wenruo 11361f872924SQu Wenruo /* Parent is detached, no need to keep any edges */ 11371f872924SQu Wenruo if (upper->detached) { 11381f872924SQu Wenruo list_del(&edge->list[LOWER]); 11391f872924SQu Wenruo free_backref_edge(cache, edge); 11401f872924SQu Wenruo 11411f872924SQu Wenruo /* Lower node is orphan, queue for cleanup */ 11421f872924SQu Wenruo if (list_empty(&lower->upper)) 11431f872924SQu Wenruo list_add(&lower->list, useless_node); 11441f872924SQu Wenruo continue; 11451f872924SQu Wenruo } 11461f872924SQu Wenruo 11471f872924SQu Wenruo /* 11481f872924SQu Wenruo * All new nodes added in current build_backref_tree() haven't 11491f872924SQu Wenruo * been linked to the cache rb tree. 11501f872924SQu Wenruo * So if we have upper->rb_node populated, this means a cache 11511f872924SQu Wenruo * hit. We only need to link the edge, as @upper and all its 11521f872924SQu Wenruo * parent have already been linked. 11531f872924SQu Wenruo */ 11541f872924SQu Wenruo if (!RB_EMPTY_NODE(&upper->rb_node)) { 11551f872924SQu Wenruo if (upper->lowest) { 11561f872924SQu Wenruo list_del_init(&upper->lower); 11571f872924SQu Wenruo upper->lowest = 0; 11581f872924SQu Wenruo } 11591f872924SQu Wenruo 11601f872924SQu Wenruo list_add_tail(&edge->list[UPPER], &upper->lower); 11611f872924SQu Wenruo continue; 11621f872924SQu Wenruo } 11631f872924SQu Wenruo 11641f872924SQu Wenruo /* Sanity check, we shouldn't have any unchecked nodes */ 11651f872924SQu Wenruo if (!upper->checked) { 11661f872924SQu Wenruo ASSERT(0); 11671f872924SQu Wenruo return -EUCLEAN; 11681f872924SQu Wenruo } 11691f872924SQu Wenruo 11701f872924SQu Wenruo /* Sanity check, COW-only node has non-COW-only parent */ 11711f872924SQu Wenruo if (start->cowonly != upper->cowonly) { 11721f872924SQu Wenruo ASSERT(0); 11731f872924SQu Wenruo return -EUCLEAN; 11741f872924SQu Wenruo } 11751f872924SQu Wenruo 11761f872924SQu Wenruo /* Only cache non-COW-only (subvolume trees) tree blocks */ 11771f872924SQu Wenruo if (!upper->cowonly) { 11781f872924SQu Wenruo rb_node = tree_insert(&cache->rb_root, upper->bytenr, 11791f872924SQu Wenruo &upper->rb_node); 11801f872924SQu Wenruo if (rb_node) { 11811f872924SQu Wenruo backref_tree_panic(rb_node, -EEXIST, 11821f872924SQu Wenruo upper->bytenr); 11831f872924SQu Wenruo return -EUCLEAN; 11841f872924SQu Wenruo } 11851f872924SQu Wenruo } 11861f872924SQu Wenruo 11871f872924SQu Wenruo list_add_tail(&edge->list[UPPER], &upper->lower); 11881f872924SQu Wenruo 11891f872924SQu Wenruo /* 11901f872924SQu Wenruo * Also queue all the parent edges of this uncached node to 11911f872924SQu Wenruo * finish the upper linkage 11921f872924SQu Wenruo */ 11931f872924SQu Wenruo list_for_each_entry(edge, &upper->upper, list[LOWER]) 11941f872924SQu Wenruo list_add_tail(&edge->list[UPPER], &pending_edge); 11951f872924SQu Wenruo } 11961f872924SQu Wenruo return 0; 11971f872924SQu Wenruo } 11981f872924SQu Wenruo 11991f872924SQu Wenruo /* 120029db137bSQu Wenruo * For useless nodes, do two major clean ups: 120129db137bSQu Wenruo * 120229db137bSQu Wenruo * - Cleanup the children edges and nodes 120329db137bSQu Wenruo * If child node is also orphan (no parent) during cleanup, then the child 120429db137bSQu Wenruo * node will also be cleaned up. 120529db137bSQu Wenruo * 120629db137bSQu Wenruo * - Freeing up leaves (level 0), keeps nodes detached 120729db137bSQu Wenruo * For nodes, the node is still cached as "detached" 120829db137bSQu Wenruo * 120929db137bSQu Wenruo * Return false if @node is not in the @useless_nodes list. 121029db137bSQu Wenruo * Return true if @node is in the @useless_nodes list. 121129db137bSQu Wenruo */ 121229db137bSQu Wenruo static bool handle_useless_nodes(struct reloc_control *rc, 1213a26195a5SQu Wenruo struct btrfs_backref_node *node) 121429db137bSQu Wenruo { 1215a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 121629db137bSQu Wenruo struct list_head *useless_node = &cache->useless_node; 121729db137bSQu Wenruo bool ret = false; 121829db137bSQu Wenruo 121929db137bSQu Wenruo while (!list_empty(useless_node)) { 1220a26195a5SQu Wenruo struct btrfs_backref_node *cur; 122129db137bSQu Wenruo 1222a26195a5SQu Wenruo cur = list_first_entry(useless_node, struct btrfs_backref_node, 122329db137bSQu Wenruo list); 122429db137bSQu Wenruo list_del_init(&cur->list); 122529db137bSQu Wenruo 122629db137bSQu Wenruo /* Only tree root nodes can be added to @useless_nodes */ 122729db137bSQu Wenruo ASSERT(list_empty(&cur->upper)); 122829db137bSQu Wenruo 122929db137bSQu Wenruo if (cur == node) 123029db137bSQu Wenruo ret = true; 123129db137bSQu Wenruo 123229db137bSQu Wenruo /* The node is the lowest node */ 123329db137bSQu Wenruo if (cur->lowest) { 123429db137bSQu Wenruo list_del_init(&cur->lower); 123529db137bSQu Wenruo cur->lowest = 0; 123629db137bSQu Wenruo } 123729db137bSQu Wenruo 123829db137bSQu Wenruo /* Cleanup the lower edges */ 123929db137bSQu Wenruo while (!list_empty(&cur->lower)) { 1240a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 1241a26195a5SQu Wenruo struct btrfs_backref_node *lower; 124229db137bSQu Wenruo 124329db137bSQu Wenruo edge = list_entry(cur->lower.next, 1244a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 124529db137bSQu Wenruo list_del(&edge->list[UPPER]); 124629db137bSQu Wenruo list_del(&edge->list[LOWER]); 124729db137bSQu Wenruo lower = edge->node[LOWER]; 124829db137bSQu Wenruo free_backref_edge(cache, edge); 124929db137bSQu Wenruo 125029db137bSQu Wenruo /* Child node is also orphan, queue for cleanup */ 125129db137bSQu Wenruo if (list_empty(&lower->upper)) 125229db137bSQu Wenruo list_add(&lower->list, useless_node); 125329db137bSQu Wenruo } 125429db137bSQu Wenruo /* Mark this block processed for relocation */ 125529db137bSQu Wenruo mark_block_processed(rc, cur); 125629db137bSQu Wenruo 125729db137bSQu Wenruo /* 125829db137bSQu Wenruo * Backref nodes for tree leaves are deleted from the cache. 125929db137bSQu Wenruo * Backref nodes for upper level tree blocks are left in the 126029db137bSQu Wenruo * cache to avoid unnecessary backref lookup. 126129db137bSQu Wenruo */ 126229db137bSQu Wenruo if (cur->level > 0) { 126329db137bSQu Wenruo list_add(&cur->list, &cache->detached); 126429db137bSQu Wenruo cur->detached = 1; 126529db137bSQu Wenruo } else { 126629db137bSQu Wenruo rb_erase(&cur->rb_node, &cache->rb_root); 126729db137bSQu Wenruo free_backref_node(cache, cur); 126829db137bSQu Wenruo } 126929db137bSQu Wenruo } 127029db137bSQu Wenruo return ret; 127129db137bSQu Wenruo } 127229db137bSQu Wenruo 127329db137bSQu Wenruo /* 1274e7d571c7SQu Wenruo * Build backref tree for a given tree block. Root of the backref tree 1275e7d571c7SQu Wenruo * corresponds the tree block, leaves of the backref tree correspond roots of 1276e7d571c7SQu Wenruo * b-trees that reference the tree block. 1277e7d571c7SQu Wenruo * 1278e7d571c7SQu Wenruo * The basic idea of this function is check backrefs of a given block to find 1279e7d571c7SQu Wenruo * upper level blocks that reference the block, and then check backrefs of 1280e7d571c7SQu Wenruo * these upper level blocks recursively. The recursion stops when tree root is 1281e7d571c7SQu Wenruo * reached or backrefs for the block is cached. 1282e7d571c7SQu Wenruo * 1283e7d571c7SQu Wenruo * NOTE: if we find that backrefs for a block are cached, we know backrefs for 1284e7d571c7SQu Wenruo * all upper level blocks that directly/indirectly reference the block are also 1285e7d571c7SQu Wenruo * cached. 1286e7d571c7SQu Wenruo */ 1287a26195a5SQu Wenruo static noinline_for_stack struct btrfs_backref_node *build_backref_tree( 1288e7d571c7SQu Wenruo struct reloc_control *rc, struct btrfs_key *node_key, 1289e7d571c7SQu Wenruo int level, u64 bytenr) 1290e7d571c7SQu Wenruo { 1291e7d571c7SQu Wenruo struct btrfs_backref_iter *iter; 1292a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 1293e7d571c7SQu Wenruo /* For searching parent of TREE_BLOCK_REF */ 1294e7d571c7SQu Wenruo struct btrfs_path *path; 1295a26195a5SQu Wenruo struct btrfs_backref_node *cur; 1296a26195a5SQu Wenruo struct btrfs_backref_node *upper; 1297a26195a5SQu Wenruo struct btrfs_backref_node *lower; 1298a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 1299a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 1300e7d571c7SQu Wenruo int ret; 1301e7d571c7SQu Wenruo int err = 0; 1302e7d571c7SQu Wenruo 1303e7d571c7SQu Wenruo iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS); 1304e7d571c7SQu Wenruo if (!iter) 1305e7d571c7SQu Wenruo return ERR_PTR(-ENOMEM); 1306e7d571c7SQu Wenruo path = btrfs_alloc_path(); 1307e7d571c7SQu Wenruo if (!path) { 1308e7d571c7SQu Wenruo err = -ENOMEM; 1309e7d571c7SQu Wenruo goto out; 1310e7d571c7SQu Wenruo } 1311e7d571c7SQu Wenruo 1312e7d571c7SQu Wenruo node = alloc_backref_node(cache, bytenr, level); 1313e7d571c7SQu Wenruo if (!node) { 1314e7d571c7SQu Wenruo err = -ENOMEM; 1315e7d571c7SQu Wenruo goto out; 1316e7d571c7SQu Wenruo } 1317e7d571c7SQu Wenruo 1318e7d571c7SQu Wenruo node->lowest = 1; 1319e7d571c7SQu Wenruo cur = node; 1320e7d571c7SQu Wenruo 1321e7d571c7SQu Wenruo /* Breadth-first search to build backref cache */ 1322e7d571c7SQu Wenruo do { 1323e7d571c7SQu Wenruo ret = handle_one_tree_block(cache, path, iter, node_key, cur); 1324e7d571c7SQu Wenruo if (ret < 0) { 1325e7d571c7SQu Wenruo err = ret; 1326e7d571c7SQu Wenruo goto out; 1327e7d571c7SQu Wenruo } 1328e7d571c7SQu Wenruo edge = list_first_entry_or_null(&cache->pending_edge, 1329a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 1330e7d571c7SQu Wenruo /* 1331e7d571c7SQu Wenruo * The pending list isn't empty, take the first block to 1332e7d571c7SQu Wenruo * process 1333e7d571c7SQu Wenruo */ 1334e7d571c7SQu Wenruo if (edge) { 13355d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 13365d4f98a2SYan Zheng cur = edge->node[UPPER]; 13375d4f98a2SYan Zheng } 1338e7d571c7SQu Wenruo } while (edge); 13395d4f98a2SYan Zheng 13401f872924SQu Wenruo /* Finish the upper linkage of newly added edges/nodes */ 13411f872924SQu Wenruo ret = finish_upper_links(cache, node); 13421f872924SQu Wenruo if (ret < 0) { 13431f872924SQu Wenruo err = ret; 134475bfb9afSJosef Bacik goto out; 134575bfb9afSJosef Bacik } 134675bfb9afSJosef Bacik 134729db137bSQu Wenruo if (handle_useless_nodes(rc, node)) 13483fd0a558SYan, Zheng node = NULL; 13495d4f98a2SYan Zheng out: 135071f572a9SQu Wenruo btrfs_backref_iter_free(iter); 135171f572a9SQu Wenruo btrfs_free_path(path); 13525d4f98a2SYan Zheng if (err) { 135384780289SQu Wenruo while (!list_empty(&cache->useless_node)) { 135484780289SQu Wenruo lower = list_first_entry(&cache->useless_node, 1355a26195a5SQu Wenruo struct btrfs_backref_node, list); 135675bfb9afSJosef Bacik list_del_init(&lower->list); 13573fd0a558SYan, Zheng } 135884780289SQu Wenruo while (!list_empty(&cache->pending_edge)) { 135984780289SQu Wenruo edge = list_first_entry(&cache->pending_edge, 1360a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 136175bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 13623fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 136375bfb9afSJosef Bacik lower = edge->node[LOWER]; 13645d4f98a2SYan Zheng upper = edge->node[UPPER]; 13653fd0a558SYan, Zheng free_backref_edge(cache, edge); 136675bfb9afSJosef Bacik 136775bfb9afSJosef Bacik /* 136875bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 136975bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 137075bfb9afSJosef Bacik */ 137175bfb9afSJosef Bacik if (list_empty(&lower->upper) && 137275bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 137384780289SQu Wenruo list_add(&lower->list, &cache->useless_node); 137475bfb9afSJosef Bacik 137575bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 137675bfb9afSJosef Bacik continue; 137775bfb9afSJosef Bacik 137801327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 137975bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 138084780289SQu Wenruo list_add_tail(&edge->list[UPPER], 138184780289SQu Wenruo &cache->pending_edge); 138275bfb9afSJosef Bacik if (list_empty(&upper->upper)) 138384780289SQu Wenruo list_add(&upper->list, &cache->useless_node); 138475bfb9afSJosef Bacik } 138575bfb9afSJosef Bacik 138684780289SQu Wenruo while (!list_empty(&cache->useless_node)) { 138784780289SQu Wenruo lower = list_first_entry(&cache->useless_node, 1388a26195a5SQu Wenruo struct btrfs_backref_node, list); 138975bfb9afSJosef Bacik list_del_init(&lower->list); 13900fd8c3daSLiu Bo if (lower == node) 13910fd8c3daSLiu Bo node = NULL; 139275bfb9afSJosef Bacik free_backref_node(cache, lower); 13935d4f98a2SYan Zheng } 13940fd8c3daSLiu Bo 13958e19c973SJosef Bacik remove_backref_node(cache, node); 139684780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 139784780289SQu Wenruo list_empty(&cache->pending_edge)); 13985d4f98a2SYan Zheng return ERR_PTR(err); 13995d4f98a2SYan Zheng } 140075bfb9afSJosef Bacik ASSERT(!node || !node->detached); 140184780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 140284780289SQu Wenruo list_empty(&cache->pending_edge)); 14035d4f98a2SYan Zheng return node; 14045d4f98a2SYan Zheng } 14055d4f98a2SYan Zheng 14065d4f98a2SYan Zheng /* 14073fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 14083fd0a558SYan, Zheng * the backref node is created by cloning backref node that 14093fd0a558SYan, Zheng * corresponds to root of source tree 14103fd0a558SYan, Zheng */ 14113fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 14123fd0a558SYan, Zheng struct reloc_control *rc, 14133fd0a558SYan, Zheng struct btrfs_root *src, 14143fd0a558SYan, Zheng struct btrfs_root *dest) 14153fd0a558SYan, Zheng { 14163fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 1417a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 1418a26195a5SQu Wenruo struct btrfs_backref_node *node = NULL; 1419a26195a5SQu Wenruo struct btrfs_backref_node *new_node; 1420a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 1421a26195a5SQu Wenruo struct btrfs_backref_edge *new_edge; 14223fd0a558SYan, Zheng struct rb_node *rb_node; 14233fd0a558SYan, Zheng 14243fd0a558SYan, Zheng if (cache->last_trans > 0) 14253fd0a558SYan, Zheng update_backref_cache(trans, cache); 14263fd0a558SYan, Zheng 14273fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 14283fd0a558SYan, Zheng if (rb_node) { 1429a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, rb_node); 14303fd0a558SYan, Zheng if (node->detached) 14313fd0a558SYan, Zheng node = NULL; 14323fd0a558SYan, Zheng else 14333fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 14343fd0a558SYan, Zheng } 14353fd0a558SYan, Zheng 14363fd0a558SYan, Zheng if (!node) { 14373fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 14383fd0a558SYan, Zheng reloc_root->commit_root->start); 14393fd0a558SYan, Zheng if (rb_node) { 1440a26195a5SQu Wenruo node = rb_entry(rb_node, struct btrfs_backref_node, 14413fd0a558SYan, Zheng rb_node); 14423fd0a558SYan, Zheng BUG_ON(node->detached); 14433fd0a558SYan, Zheng } 14443fd0a558SYan, Zheng } 14453fd0a558SYan, Zheng 14463fd0a558SYan, Zheng if (!node) 14473fd0a558SYan, Zheng return 0; 14483fd0a558SYan, Zheng 14490304f2d8SQu Wenruo new_node = alloc_backref_node(cache, dest->node->start, node->level); 14503fd0a558SYan, Zheng if (!new_node) 14513fd0a558SYan, Zheng return -ENOMEM; 14523fd0a558SYan, Zheng 14533fd0a558SYan, Zheng new_node->lowest = node->lowest; 14546848ad64SYan, Zheng new_node->checked = 1; 145500246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 14560b530bc5SJosef Bacik ASSERT(new_node->root); 14573fd0a558SYan, Zheng 14583fd0a558SYan, Zheng if (!node->lowest) { 14593fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 14603fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 14613fd0a558SYan, Zheng if (!new_edge) 14623fd0a558SYan, Zheng goto fail; 14633fd0a558SYan, Zheng 14642a979612SQu Wenruo link_backref_edge(new_edge, edge->node[LOWER], new_node, 14652a979612SQu Wenruo LINK_UPPER); 14663fd0a558SYan, Zheng } 146776b9e23dSMiao Xie } else { 146876b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 14693fd0a558SYan, Zheng } 14703fd0a558SYan, Zheng 14713fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 14723fd0a558SYan, Zheng &new_node->rb_node); 147343c04fb1SJeff Mahoney if (rb_node) 147443c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 14753fd0a558SYan, Zheng 14763fd0a558SYan, Zheng if (!new_node->lowest) { 14773fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 14783fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 14793fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 14803fd0a558SYan, Zheng } 14813fd0a558SYan, Zheng } 14823fd0a558SYan, Zheng return 0; 14833fd0a558SYan, Zheng fail: 14843fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 14853fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 1486a26195a5SQu Wenruo struct btrfs_backref_edge, list[UPPER]); 14873fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 14883fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 14893fd0a558SYan, Zheng } 14903fd0a558SYan, Zheng free_backref_node(cache, new_node); 14913fd0a558SYan, Zheng return -ENOMEM; 14923fd0a558SYan, Zheng } 14933fd0a558SYan, Zheng 14943fd0a558SYan, Zheng /* 14955d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 14965d4f98a2SYan Zheng */ 1497ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 14985d4f98a2SYan Zheng { 14990b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15005d4f98a2SYan Zheng struct rb_node *rb_node; 15015d4f98a2SYan Zheng struct mapping_node *node; 15020b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 15035d4f98a2SYan Zheng 15045d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1505ffd7b339SJeff Mahoney if (!node) 1506ffd7b339SJeff Mahoney return -ENOMEM; 15075d4f98a2SYan Zheng 1508ea287ab1SJosef Bacik node->bytenr = root->commit_root->start; 15095d4f98a2SYan Zheng node->data = root; 15105d4f98a2SYan Zheng 15115d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 15125d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 15135d4f98a2SYan Zheng node->bytenr, &node->rb_node); 15145d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1515ffd7b339SJeff Mahoney if (rb_node) { 15160b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 15175d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 15185d163e0eSJeff Mahoney node->bytenr); 1519ffd7b339SJeff Mahoney } 15205d4f98a2SYan Zheng 15215d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 15225d4f98a2SYan Zheng return 0; 15235d4f98a2SYan Zheng } 15245d4f98a2SYan Zheng 15255d4f98a2SYan Zheng /* 1526c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 15275d4f98a2SYan Zheng * mapping 15285d4f98a2SYan Zheng */ 1529c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 15305d4f98a2SYan Zheng { 15310b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15325d4f98a2SYan Zheng struct rb_node *rb_node; 15335d4f98a2SYan Zheng struct mapping_node *node = NULL; 15340b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1535f44deb74SJosef Bacik bool put_ref = false; 15365d4f98a2SYan Zheng 153765c6e82bSQu Wenruo if (rc && root->node) { 15385d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 15395d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1540ea287ab1SJosef Bacik root->commit_root->start); 1541c974c464SWang Shilong if (rb_node) { 1542c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1543c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1544ea287ab1SJosef Bacik RB_CLEAR_NODE(&node->rb_node); 1545c974c464SWang Shilong } 1546c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1547c974c464SWang Shilong if (!node) 1548c974c464SWang Shilong return; 1549c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1550389305b2SQu Wenruo } 1551c974c464SWang Shilong 1552f44deb74SJosef Bacik /* 1553f44deb74SJosef Bacik * We only put the reloc root here if it's on the list. There's a lot 1554f44deb74SJosef Bacik * of places where the pattern is to splice the rc->reloc_roots, process 1555f44deb74SJosef Bacik * the reloc roots, and then add the reloc root back onto 1556f44deb74SJosef Bacik * rc->reloc_roots. If we call __del_reloc_root while it's off of the 1557f44deb74SJosef Bacik * list we don't want the reference being dropped, because the guy 1558f44deb74SJosef Bacik * messing with the list is in charge of the reference. 1559f44deb74SJosef Bacik */ 15600b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1561f44deb74SJosef Bacik if (!list_empty(&root->root_list)) { 1562f44deb74SJosef Bacik put_ref = true; 1563c974c464SWang Shilong list_del_init(&root->root_list); 1564f44deb74SJosef Bacik } 15650b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1566f44deb74SJosef Bacik if (put_ref) 1567f44deb74SJosef Bacik btrfs_put_root(root); 1568c974c464SWang Shilong kfree(node); 1569c974c464SWang Shilong } 1570c974c464SWang Shilong 1571c974c464SWang Shilong /* 1572c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1573c974c464SWang Shilong * mapping 1574c974c464SWang Shilong */ 1575ea287ab1SJosef Bacik static int __update_reloc_root(struct btrfs_root *root) 1576c974c464SWang Shilong { 15770b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1578c974c464SWang Shilong struct rb_node *rb_node; 1579c974c464SWang Shilong struct mapping_node *node = NULL; 15800b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1581c974c464SWang Shilong 1582c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1583c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1584ea287ab1SJosef Bacik root->commit_root->start); 15855d4f98a2SYan Zheng if (rb_node) { 15865d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 15875d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 15885d4f98a2SYan Zheng } 15895d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 15905d4f98a2SYan Zheng 15918f71f3e0SLiu Bo if (!node) 15928f71f3e0SLiu Bo return 0; 15935d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 15945d4f98a2SYan Zheng 15955d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1596ea287ab1SJosef Bacik node->bytenr = root->node->start; 15975d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 15985d4f98a2SYan Zheng node->bytenr, &node->rb_node); 15995d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 160043c04fb1SJeff Mahoney if (rb_node) 160143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 16025d4f98a2SYan Zheng return 0; 16035d4f98a2SYan Zheng } 16045d4f98a2SYan Zheng 16053fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 16063fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 16075d4f98a2SYan Zheng { 16080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 16095d4f98a2SYan Zheng struct btrfs_root *reloc_root; 16105d4f98a2SYan Zheng struct extent_buffer *eb; 16115d4f98a2SYan Zheng struct btrfs_root_item *root_item; 16125d4f98a2SYan Zheng struct btrfs_key root_key; 16135d4f98a2SYan Zheng int ret; 16145d4f98a2SYan Zheng 16155d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 16165d4f98a2SYan Zheng BUG_ON(!root_item); 16175d4f98a2SYan Zheng 16185d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 16195d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 16203fd0a558SYan, Zheng root_key.offset = objectid; 16215d4f98a2SYan Zheng 16223fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1623054570a1SFilipe Manana u64 commit_root_gen; 1624054570a1SFilipe Manana 16253fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 16265d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 16275d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 16285d4f98a2SYan Zheng BUG_ON(ret); 1629054570a1SFilipe Manana /* 1630054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1631054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1632054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1633054570a1SFilipe Manana * either inside the critical section of a transaction commit 1634054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1635054570a1SFilipe Manana * it's created before the transaction commit is started. 1636054570a1SFilipe Manana */ 1637054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1638054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 16393fd0a558SYan, Zheng } else { 16403fd0a558SYan, Zheng /* 16413fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 16423fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 16433fd0a558SYan, Zheng * modified after it was created have RELOC flag 16443fd0a558SYan, Zheng * set in their headers. so it's OK to not update 16453fd0a558SYan, Zheng * the 'last_snapshot'. 16463fd0a558SYan, Zheng */ 16473fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 16483fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 16493fd0a558SYan, Zheng BUG_ON(ret); 16503fd0a558SYan, Zheng } 16513fd0a558SYan, Zheng 16525d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 16535d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 16545d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 16555d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 16563fd0a558SYan, Zheng 16573fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 16583fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 16593fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 16603fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 16615d4f98a2SYan Zheng root_item->drop_level = 0; 16623fd0a558SYan, Zheng } 16635d4f98a2SYan Zheng 16645d4f98a2SYan Zheng btrfs_tree_unlock(eb); 16655d4f98a2SYan Zheng free_extent_buffer(eb); 16665d4f98a2SYan Zheng 16670b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 16685d4f98a2SYan Zheng &root_key, root_item); 16695d4f98a2SYan Zheng BUG_ON(ret); 16705d4f98a2SYan Zheng kfree(root_item); 16715d4f98a2SYan Zheng 16723dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 16735d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 16743dbf1738SJosef Bacik set_bit(BTRFS_ROOT_REF_COWS, &reloc_root->state); 16755d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 16763fd0a558SYan, Zheng return reloc_root; 16773fd0a558SYan, Zheng } 16783fd0a558SYan, Zheng 16793fd0a558SYan, Zheng /* 16803fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 16813fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 1682f44deb74SJosef Bacik * 1683f44deb74SJosef Bacik * The reloc_root comes out of here with two references, one for 1684f44deb74SJosef Bacik * root->reloc_root, and another for being on the rc->reloc_roots list. 16853fd0a558SYan, Zheng */ 16863fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 16873fd0a558SYan, Zheng struct btrfs_root *root) 16883fd0a558SYan, Zheng { 16890b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 16903fd0a558SYan, Zheng struct btrfs_root *reloc_root; 16910b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 169220dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 16933fd0a558SYan, Zheng int clear_rsv = 0; 1694ffd7b339SJeff Mahoney int ret; 16953fd0a558SYan, Zheng 1696aec7db3bSJosef Bacik if (!rc) 16972abc726aSJosef Bacik return 0; 16982abc726aSJosef Bacik 16991fac4a54SQu Wenruo /* 17001fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 17011fac4a54SQu Wenruo * create/update the dead reloc tree 17021fac4a54SQu Wenruo */ 17036282675eSQu Wenruo if (reloc_root_is_dead(root)) 17041fac4a54SQu Wenruo return 0; 17051fac4a54SQu Wenruo 1706aec7db3bSJosef Bacik /* 1707aec7db3bSJosef Bacik * This is subtle but important. We do not do 1708aec7db3bSJosef Bacik * record_root_in_transaction for reloc roots, instead we record their 1709aec7db3bSJosef Bacik * corresponding fs root, and then here we update the last trans for the 1710aec7db3bSJosef Bacik * reloc root. This means that we have to do this for the entire life 1711aec7db3bSJosef Bacik * of the reloc root, regardless of which stage of the relocation we are 1712aec7db3bSJosef Bacik * in. 1713aec7db3bSJosef Bacik */ 17143fd0a558SYan, Zheng if (root->reloc_root) { 17153fd0a558SYan, Zheng reloc_root = root->reloc_root; 17163fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 17173fd0a558SYan, Zheng return 0; 17183fd0a558SYan, Zheng } 17193fd0a558SYan, Zheng 1720aec7db3bSJosef Bacik /* 1721aec7db3bSJosef Bacik * We are merging reloc roots, we do not need new reloc trees. Also 1722aec7db3bSJosef Bacik * reloc trees never need their own reloc tree. 1723aec7db3bSJosef Bacik */ 1724aec7db3bSJosef Bacik if (!rc->create_reloc_tree || 1725aec7db3bSJosef Bacik root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 1726aec7db3bSJosef Bacik return 0; 1727aec7db3bSJosef Bacik 172820dd2cbfSMiao Xie if (!trans->reloc_reserved) { 172920dd2cbfSMiao Xie rsv = trans->block_rsv; 17303fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 17313fd0a558SYan, Zheng clear_rsv = 1; 17323fd0a558SYan, Zheng } 17333fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 17343fd0a558SYan, Zheng if (clear_rsv) 173520dd2cbfSMiao Xie trans->block_rsv = rsv; 17365d4f98a2SYan Zheng 1737ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1738ffd7b339SJeff Mahoney BUG_ON(ret < 0); 1739f44deb74SJosef Bacik root->reloc_root = btrfs_grab_root(reloc_root); 17405d4f98a2SYan Zheng return 0; 17415d4f98a2SYan Zheng } 17425d4f98a2SYan Zheng 17435d4f98a2SYan Zheng /* 17445d4f98a2SYan Zheng * update root item of reloc tree 17455d4f98a2SYan Zheng */ 17465d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 17475d4f98a2SYan Zheng struct btrfs_root *root) 17485d4f98a2SYan Zheng { 17490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 17505d4f98a2SYan Zheng struct btrfs_root *reloc_root; 17515d4f98a2SYan Zheng struct btrfs_root_item *root_item; 17525d4f98a2SYan Zheng int ret; 17535d4f98a2SYan Zheng 17546282675eSQu Wenruo if (!have_reloc_root(root)) 17557585717fSChris Mason goto out; 17565d4f98a2SYan Zheng 17575d4f98a2SYan Zheng reloc_root = root->reloc_root; 17585d4f98a2SYan Zheng root_item = &reloc_root->root_item; 17595d4f98a2SYan Zheng 1760f44deb74SJosef Bacik /* 1761f44deb74SJosef Bacik * We are probably ok here, but __del_reloc_root() will drop its ref of 1762f44deb74SJosef Bacik * the root. We have the ref for root->reloc_root, but just in case 1763f44deb74SJosef Bacik * hold it while we update the reloc root. 1764f44deb74SJosef Bacik */ 1765f44deb74SJosef Bacik btrfs_grab_root(reloc_root); 1766f44deb74SJosef Bacik 1767d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 17680b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 17693fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 1770d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 17716282675eSQu Wenruo /* 17726282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 17736282675eSQu Wenruo * have_reloc_root will not touch it from now on. 17746282675eSQu Wenruo */ 17756282675eSQu Wenruo smp_wmb(); 1776c974c464SWang Shilong __del_reloc_root(reloc_root); 17775d4f98a2SYan Zheng } 17785d4f98a2SYan Zheng 17795d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 1780ea287ab1SJosef Bacik __update_reloc_root(reloc_root); 17815d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 17825d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 17835d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 17845d4f98a2SYan Zheng } 17855d4f98a2SYan Zheng 17860b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 17875d4f98a2SYan Zheng &reloc_root->root_key, root_item); 17885d4f98a2SYan Zheng BUG_ON(ret); 1789f44deb74SJosef Bacik btrfs_put_root(reloc_root); 17907585717fSChris Mason out: 17915d4f98a2SYan Zheng return 0; 17925d4f98a2SYan Zheng } 17935d4f98a2SYan Zheng 17945d4f98a2SYan Zheng /* 17955d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 17965d4f98a2SYan Zheng * in a subvolume 17975d4f98a2SYan Zheng */ 17985d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 17995d4f98a2SYan Zheng { 18005d4f98a2SYan Zheng struct rb_node *node; 18015d4f98a2SYan Zheng struct rb_node *prev; 18025d4f98a2SYan Zheng struct btrfs_inode *entry; 18035d4f98a2SYan Zheng struct inode *inode; 18045d4f98a2SYan Zheng 18055d4f98a2SYan Zheng spin_lock(&root->inode_lock); 18065d4f98a2SYan Zheng again: 18075d4f98a2SYan Zheng node = root->inode_tree.rb_node; 18085d4f98a2SYan Zheng prev = NULL; 18095d4f98a2SYan Zheng while (node) { 18105d4f98a2SYan Zheng prev = node; 18115d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 18125d4f98a2SYan Zheng 18134a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 18145d4f98a2SYan Zheng node = node->rb_left; 18154a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 18165d4f98a2SYan Zheng node = node->rb_right; 18175d4f98a2SYan Zheng else 18185d4f98a2SYan Zheng break; 18195d4f98a2SYan Zheng } 18205d4f98a2SYan Zheng if (!node) { 18215d4f98a2SYan Zheng while (prev) { 18225d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 18234a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 18245d4f98a2SYan Zheng node = prev; 18255d4f98a2SYan Zheng break; 18265d4f98a2SYan Zheng } 18275d4f98a2SYan Zheng prev = rb_next(prev); 18285d4f98a2SYan Zheng } 18295d4f98a2SYan Zheng } 18305d4f98a2SYan Zheng while (node) { 18315d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 18325d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 18335d4f98a2SYan Zheng if (inode) { 18345d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 18355d4f98a2SYan Zheng return inode; 18365d4f98a2SYan Zheng } 18375d4f98a2SYan Zheng 18384a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 18395d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 18405d4f98a2SYan Zheng goto again; 18415d4f98a2SYan Zheng 18425d4f98a2SYan Zheng node = rb_next(node); 18435d4f98a2SYan Zheng } 18445d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 18455d4f98a2SYan Zheng return NULL; 18465d4f98a2SYan Zheng } 18475d4f98a2SYan Zheng 18485d4f98a2SYan Zheng /* 18495d4f98a2SYan Zheng * get new location of data 18505d4f98a2SYan Zheng */ 18515d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 18525d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 18535d4f98a2SYan Zheng { 18545d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 18555d4f98a2SYan Zheng struct btrfs_path *path; 18565d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 18575d4f98a2SYan Zheng struct extent_buffer *leaf; 18585d4f98a2SYan Zheng int ret; 18595d4f98a2SYan Zheng 18605d4f98a2SYan Zheng path = btrfs_alloc_path(); 18615d4f98a2SYan Zheng if (!path) 18625d4f98a2SYan Zheng return -ENOMEM; 18635d4f98a2SYan Zheng 18645d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1865f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1866f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 18675d4f98a2SYan Zheng if (ret < 0) 18685d4f98a2SYan Zheng goto out; 18695d4f98a2SYan Zheng if (ret > 0) { 18705d4f98a2SYan Zheng ret = -ENOENT; 18715d4f98a2SYan Zheng goto out; 18725d4f98a2SYan Zheng } 18735d4f98a2SYan Zheng 18745d4f98a2SYan Zheng leaf = path->nodes[0]; 18755d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 18765d4f98a2SYan Zheng struct btrfs_file_extent_item); 18775d4f98a2SYan Zheng 18785d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 18795d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 18805d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 18815d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 18825d4f98a2SYan Zheng 18835d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 188483d4cfd4SJosef Bacik ret = -EINVAL; 18855d4f98a2SYan Zheng goto out; 18865d4f98a2SYan Zheng } 18875d4f98a2SYan Zheng 18885d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 18895d4f98a2SYan Zheng ret = 0; 18905d4f98a2SYan Zheng out: 18915d4f98a2SYan Zheng btrfs_free_path(path); 18925d4f98a2SYan Zheng return ret; 18935d4f98a2SYan Zheng } 18945d4f98a2SYan Zheng 18955d4f98a2SYan Zheng /* 18965d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 18975d4f98a2SYan Zheng * the new locations. 18985d4f98a2SYan Zheng */ 18993fd0a558SYan, Zheng static noinline_for_stack 19003fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 19015d4f98a2SYan Zheng struct reloc_control *rc, 19025d4f98a2SYan Zheng struct btrfs_root *root, 19033fd0a558SYan, Zheng struct extent_buffer *leaf) 19045d4f98a2SYan Zheng { 19050b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 19065d4f98a2SYan Zheng struct btrfs_key key; 19075d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 19085d4f98a2SYan Zheng struct inode *inode = NULL; 19095d4f98a2SYan Zheng u64 parent; 19105d4f98a2SYan Zheng u64 bytenr; 19113fd0a558SYan, Zheng u64 new_bytenr = 0; 19125d4f98a2SYan Zheng u64 num_bytes; 19135d4f98a2SYan Zheng u64 end; 19145d4f98a2SYan Zheng u32 nritems; 19155d4f98a2SYan Zheng u32 i; 191683d4cfd4SJosef Bacik int ret = 0; 19175d4f98a2SYan Zheng int first = 1; 19185d4f98a2SYan Zheng int dirty = 0; 19195d4f98a2SYan Zheng 19205d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 19215d4f98a2SYan Zheng return 0; 19225d4f98a2SYan Zheng 19235d4f98a2SYan Zheng /* reloc trees always use full backref */ 19245d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 19255d4f98a2SYan Zheng parent = leaf->start; 19265d4f98a2SYan Zheng else 19275d4f98a2SYan Zheng parent = 0; 19285d4f98a2SYan Zheng 19295d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 19305d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 193182fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 193282fa113fSQu Wenruo 19335d4f98a2SYan Zheng cond_resched(); 19345d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 19355d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 19365d4f98a2SYan Zheng continue; 19375d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 19385d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 19395d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 19405d4f98a2SYan Zheng continue; 19415d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 19425d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 19435d4f98a2SYan Zheng if (bytenr == 0) 19445d4f98a2SYan Zheng continue; 19459569cc20SQu Wenruo if (!in_range(bytenr, rc->block_group->start, 19469569cc20SQu Wenruo rc->block_group->length)) 19475d4f98a2SYan Zheng continue; 19485d4f98a2SYan Zheng 19495d4f98a2SYan Zheng /* 19505d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 19515d4f98a2SYan Zheng * to complete and drop the extent cache 19525d4f98a2SYan Zheng */ 19535d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 19545d4f98a2SYan Zheng if (first) { 19555d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 19565d4f98a2SYan Zheng first = 0; 19574a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 19583fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 19595d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 19605d4f98a2SYan Zheng } 19614a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 19625d4f98a2SYan Zheng end = key.offset + 19635d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 19645d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 19650b246afaSJeff Mahoney fs_info->sectorsize)); 19660b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 19675d4f98a2SYan Zheng end--; 19685d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1969d0082371SJeff Mahoney key.offset, end); 19705d4f98a2SYan Zheng if (!ret) 19715d4f98a2SYan Zheng continue; 19725d4f98a2SYan Zheng 1973dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1974dcdbc059SNikolay Borisov key.offset, end, 1); 19755d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1976d0082371SJeff Mahoney key.offset, end); 19775d4f98a2SYan Zheng } 19785d4f98a2SYan Zheng } 19795d4f98a2SYan Zheng 19805d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 19815d4f98a2SYan Zheng bytenr, num_bytes); 198283d4cfd4SJosef Bacik if (ret) { 198383d4cfd4SJosef Bacik /* 198483d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 198583d4cfd4SJosef Bacik * in the file extent yet. 198683d4cfd4SJosef Bacik */ 198783d4cfd4SJosef Bacik break; 19883fd0a558SYan, Zheng } 19895d4f98a2SYan Zheng 19905d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 19915d4f98a2SYan Zheng dirty = 1; 19925d4f98a2SYan Zheng 19935d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 199482fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 199582fa113fSQu Wenruo num_bytes, parent); 199682fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 199782fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1998b06c4bf5SFilipe Manana key.objectid, key.offset); 199982fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 200083d4cfd4SJosef Bacik if (ret) { 200166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 200283d4cfd4SJosef Bacik break; 200383d4cfd4SJosef Bacik } 20045d4f98a2SYan Zheng 2005ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 2006ffd4bb2aSQu Wenruo num_bytes, parent); 2007ffd4bb2aSQu Wenruo ref.real_root = root->root_key.objectid; 2008ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 2009b06c4bf5SFilipe Manana key.objectid, key.offset); 2010ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 201183d4cfd4SJosef Bacik if (ret) { 201266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 201383d4cfd4SJosef Bacik break; 201483d4cfd4SJosef Bacik } 20155d4f98a2SYan Zheng } 20165d4f98a2SYan Zheng if (dirty) 20175d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 20183fd0a558SYan, Zheng if (inode) 20193fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 202083d4cfd4SJosef Bacik return ret; 20215d4f98a2SYan Zheng } 20225d4f98a2SYan Zheng 20235d4f98a2SYan Zheng static noinline_for_stack 20245d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 20255d4f98a2SYan Zheng struct btrfs_path *path, int level) 20265d4f98a2SYan Zheng { 20275d4f98a2SYan Zheng struct btrfs_disk_key key1; 20285d4f98a2SYan Zheng struct btrfs_disk_key key2; 20295d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 20305d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 20315d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 20325d4f98a2SYan Zheng } 20335d4f98a2SYan Zheng 20345d4f98a2SYan Zheng /* 20355d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 20365d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 20375d4f98a2SYan Zheng * reloc tree was create can be replaced. 20385d4f98a2SYan Zheng * 20395d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 20405d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 20415d4f98a2SYan Zheng * errors, a negative error number is returned. 20425d4f98a2SYan Zheng */ 20433fd0a558SYan, Zheng static noinline_for_stack 20443d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 20455d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 20465d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 20475d4f98a2SYan Zheng int lowest_level, int max_level) 20485d4f98a2SYan Zheng { 20490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 20505d4f98a2SYan Zheng struct extent_buffer *eb; 20515d4f98a2SYan Zheng struct extent_buffer *parent; 205282fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 20535d4f98a2SYan Zheng struct btrfs_key key; 20545d4f98a2SYan Zheng u64 old_bytenr; 20555d4f98a2SYan Zheng u64 new_bytenr; 20565d4f98a2SYan Zheng u64 old_ptr_gen; 20575d4f98a2SYan Zheng u64 new_ptr_gen; 20585d4f98a2SYan Zheng u64 last_snapshot; 20595d4f98a2SYan Zheng u32 blocksize; 20603fd0a558SYan, Zheng int cow = 0; 20615d4f98a2SYan Zheng int level; 20625d4f98a2SYan Zheng int ret; 20635d4f98a2SYan Zheng int slot; 20645d4f98a2SYan Zheng 20655d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 20665d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 20675d4f98a2SYan Zheng 20685d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 20693fd0a558SYan, Zheng again: 20705d4f98a2SYan Zheng slot = path->slots[lowest_level]; 20715d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 20725d4f98a2SYan Zheng 20735d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 20748bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 20755d4f98a2SYan Zheng level = btrfs_header_level(eb); 20765d4f98a2SYan Zheng 20775d4f98a2SYan Zheng if (level < lowest_level) { 20785d4f98a2SYan Zheng btrfs_tree_unlock(eb); 20795d4f98a2SYan Zheng free_extent_buffer(eb); 20805d4f98a2SYan Zheng return 0; 20815d4f98a2SYan Zheng } 20825d4f98a2SYan Zheng 20833fd0a558SYan, Zheng if (cow) { 20845d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 20855d4f98a2SYan Zheng BUG_ON(ret); 20863fd0a558SYan, Zheng } 20878bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 20885d4f98a2SYan Zheng 20895d4f98a2SYan Zheng if (next_key) { 20905d4f98a2SYan Zheng next_key->objectid = (u64)-1; 20915d4f98a2SYan Zheng next_key->type = (u8)-1; 20925d4f98a2SYan Zheng next_key->offset = (u64)-1; 20935d4f98a2SYan Zheng } 20945d4f98a2SYan Zheng 20955d4f98a2SYan Zheng parent = eb; 20965d4f98a2SYan Zheng while (1) { 2097581c1760SQu Wenruo struct btrfs_key first_key; 2098581c1760SQu Wenruo 20995d4f98a2SYan Zheng level = btrfs_header_level(parent); 21005d4f98a2SYan Zheng BUG_ON(level < lowest_level); 21015d4f98a2SYan Zheng 21025d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 2103cbca7d59SFilipe Manana if (ret < 0) 2104cbca7d59SFilipe Manana break; 21055d4f98a2SYan Zheng if (ret && slot > 0) 21065d4f98a2SYan Zheng slot--; 21075d4f98a2SYan Zheng 21085d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 21095d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 21105d4f98a2SYan Zheng 21115d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 21120b246afaSJeff Mahoney blocksize = fs_info->nodesize; 21135d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 211417515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 21155d4f98a2SYan Zheng 21165d4f98a2SYan Zheng if (level <= max_level) { 21175d4f98a2SYan Zheng eb = path->nodes[level]; 21185d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 21195d4f98a2SYan Zheng path->slots[level]); 21205d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 21215d4f98a2SYan Zheng path->slots[level]); 21225d4f98a2SYan Zheng } else { 21235d4f98a2SYan Zheng new_bytenr = 0; 21245d4f98a2SYan Zheng new_ptr_gen = 0; 21255d4f98a2SYan Zheng } 21265d4f98a2SYan Zheng 2127fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 21285d4f98a2SYan Zheng ret = level; 21295d4f98a2SYan Zheng break; 21305d4f98a2SYan Zheng } 21315d4f98a2SYan Zheng 21325d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 21335d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 21343fd0a558SYan, Zheng if (level <= lowest_level) { 21355d4f98a2SYan Zheng ret = 0; 21365d4f98a2SYan Zheng break; 21375d4f98a2SYan Zheng } 21385d4f98a2SYan Zheng 2139581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 2140581c1760SQu Wenruo level - 1, &first_key); 214164c043deSLiu Bo if (IS_ERR(eb)) { 214264c043deSLiu Bo ret = PTR_ERR(eb); 2143264813acSLiu Bo break; 214464c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 214564c043deSLiu Bo ret = -EIO; 2146416bc658SJosef Bacik free_extent_buffer(eb); 2147379cde74SStefan Behrens break; 2148416bc658SJosef Bacik } 21495d4f98a2SYan Zheng btrfs_tree_lock(eb); 21503fd0a558SYan, Zheng if (cow) { 21515d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 21525d4f98a2SYan Zheng slot, &eb); 21535d4f98a2SYan Zheng BUG_ON(ret); 21545d4f98a2SYan Zheng } 21558bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 21565d4f98a2SYan Zheng 21575d4f98a2SYan Zheng btrfs_tree_unlock(parent); 21585d4f98a2SYan Zheng free_extent_buffer(parent); 21595d4f98a2SYan Zheng 21605d4f98a2SYan Zheng parent = eb; 21615d4f98a2SYan Zheng continue; 21625d4f98a2SYan Zheng } 21635d4f98a2SYan Zheng 21643fd0a558SYan, Zheng if (!cow) { 21653fd0a558SYan, Zheng btrfs_tree_unlock(parent); 21663fd0a558SYan, Zheng free_extent_buffer(parent); 21673fd0a558SYan, Zheng cow = 1; 21683fd0a558SYan, Zheng goto again; 21693fd0a558SYan, Zheng } 21703fd0a558SYan, Zheng 21715d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 21725d4f98a2SYan Zheng path->slots[level]); 2173b3b4aa74SDavid Sterba btrfs_release_path(path); 21745d4f98a2SYan Zheng 21755d4f98a2SYan Zheng path->lowest_level = level; 21765d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 21775d4f98a2SYan Zheng path->lowest_level = 0; 21785d4f98a2SYan Zheng BUG_ON(ret); 21795d4f98a2SYan Zheng 21805d4f98a2SYan Zheng /* 2181824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 2182824d8dffSQu Wenruo * 2183824d8dffSQu Wenruo * We must trace both trees. 2184824d8dffSQu Wenruo * 1) Tree reloc subtree 2185824d8dffSQu Wenruo * If not traced, we will leak data numbers 2186824d8dffSQu Wenruo * 2) Fs subtree 2187824d8dffSQu Wenruo * If not traced, we will double count old data 2188f616f5cdSQu Wenruo * 2189f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 2190f616f5cdSQu Wenruo * the swapped tree blocks. 2191f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 2192f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 2193824d8dffSQu Wenruo */ 2194370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 2195370a11b8SQu Wenruo rc->block_group, parent, slot, 2196370a11b8SQu Wenruo path->nodes[level], path->slots[level], 2197370a11b8SQu Wenruo last_snapshot); 2198370a11b8SQu Wenruo if (ret < 0) 2199370a11b8SQu Wenruo break; 2200824d8dffSQu Wenruo /* 22015d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 22025d4f98a2SYan Zheng */ 22035d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 22045d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 22055d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 22065d4f98a2SYan Zheng 22075d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 22085d4f98a2SYan Zheng path->slots[level], old_bytenr); 22095d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 22105d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 22115d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 22125d4f98a2SYan Zheng 221382fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 221482fa113fSQu Wenruo blocksize, path->nodes[level]->start); 221582fa113fSQu Wenruo ref.skip_qgroup = true; 221682fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 221782fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 22185d4f98a2SYan Zheng BUG_ON(ret); 221982fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 222082fa113fSQu Wenruo blocksize, 0); 222182fa113fSQu Wenruo ref.skip_qgroup = true; 222282fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 222382fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 22245d4f98a2SYan Zheng BUG_ON(ret); 22255d4f98a2SYan Zheng 2226ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 2227ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 2228ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 2229ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 2230ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 22315d4f98a2SYan Zheng BUG_ON(ret); 22325d4f98a2SYan Zheng 2233ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 2234ffd4bb2aSQu Wenruo blocksize, 0); 2235ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 2236ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 2237ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 22385d4f98a2SYan Zheng BUG_ON(ret); 22395d4f98a2SYan Zheng 22405d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 22415d4f98a2SYan Zheng 22425d4f98a2SYan Zheng ret = level; 22435d4f98a2SYan Zheng break; 22445d4f98a2SYan Zheng } 22455d4f98a2SYan Zheng btrfs_tree_unlock(parent); 22465d4f98a2SYan Zheng free_extent_buffer(parent); 22475d4f98a2SYan Zheng return ret; 22485d4f98a2SYan Zheng } 22495d4f98a2SYan Zheng 22505d4f98a2SYan Zheng /* 22515d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 22525d4f98a2SYan Zheng */ 22535d4f98a2SYan Zheng static noinline_for_stack 22545d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 22555d4f98a2SYan Zheng int *level) 22565d4f98a2SYan Zheng { 22575d4f98a2SYan Zheng struct extent_buffer *eb; 22585d4f98a2SYan Zheng int i; 22595d4f98a2SYan Zheng u64 last_snapshot; 22605d4f98a2SYan Zheng u32 nritems; 22615d4f98a2SYan Zheng 22625d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 22635d4f98a2SYan Zheng 22645d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 22655d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 22665d4f98a2SYan Zheng path->nodes[i] = NULL; 22675d4f98a2SYan Zheng } 22685d4f98a2SYan Zheng 22695d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 22705d4f98a2SYan Zheng eb = path->nodes[i]; 22715d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 22725d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 22735d4f98a2SYan Zheng path->slots[i]++; 22745d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 22755d4f98a2SYan Zheng last_snapshot) 22765d4f98a2SYan Zheng continue; 22775d4f98a2SYan Zheng 22785d4f98a2SYan Zheng *level = i; 22795d4f98a2SYan Zheng return 0; 22805d4f98a2SYan Zheng } 22815d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 22825d4f98a2SYan Zheng path->nodes[i] = NULL; 22835d4f98a2SYan Zheng } 22845d4f98a2SYan Zheng return 1; 22855d4f98a2SYan Zheng } 22865d4f98a2SYan Zheng 22875d4f98a2SYan Zheng /* 22885d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 22895d4f98a2SYan Zheng */ 22905d4f98a2SYan Zheng static noinline_for_stack 22915d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 22925d4f98a2SYan Zheng int *level) 22935d4f98a2SYan Zheng { 22942ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 22955d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 22965d4f98a2SYan Zheng int i; 22975d4f98a2SYan Zheng u64 bytenr; 22985d4f98a2SYan Zheng u64 ptr_gen = 0; 22995d4f98a2SYan Zheng u64 last_snapshot; 23005d4f98a2SYan Zheng u32 nritems; 23015d4f98a2SYan Zheng 23025d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 23035d4f98a2SYan Zheng 23045d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2305581c1760SQu Wenruo struct btrfs_key first_key; 2306581c1760SQu Wenruo 23075d4f98a2SYan Zheng eb = path->nodes[i]; 23085d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 23095d4f98a2SYan Zheng while (path->slots[i] < nritems) { 23105d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 23115d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 23125d4f98a2SYan Zheng break; 23135d4f98a2SYan Zheng path->slots[i]++; 23145d4f98a2SYan Zheng } 23155d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 23165d4f98a2SYan Zheng if (i == *level) 23175d4f98a2SYan Zheng break; 23185d4f98a2SYan Zheng *level = i + 1; 23195d4f98a2SYan Zheng return 0; 23205d4f98a2SYan Zheng } 23215d4f98a2SYan Zheng if (i == 1) { 23225d4f98a2SYan Zheng *level = i; 23235d4f98a2SYan Zheng return 0; 23245d4f98a2SYan Zheng } 23255d4f98a2SYan Zheng 23265d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2327581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2328581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2329581c1760SQu Wenruo &first_key); 233064c043deSLiu Bo if (IS_ERR(eb)) { 233164c043deSLiu Bo return PTR_ERR(eb); 233264c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2333416bc658SJosef Bacik free_extent_buffer(eb); 2334416bc658SJosef Bacik return -EIO; 2335416bc658SJosef Bacik } 23365d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 23375d4f98a2SYan Zheng path->nodes[i - 1] = eb; 23385d4f98a2SYan Zheng path->slots[i - 1] = 0; 23395d4f98a2SYan Zheng } 23405d4f98a2SYan Zheng return 1; 23415d4f98a2SYan Zheng } 23425d4f98a2SYan Zheng 23435d4f98a2SYan Zheng /* 23445d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 23455d4f98a2SYan Zheng * [min_key, max_key) 23465d4f98a2SYan Zheng */ 23475d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 23485d4f98a2SYan Zheng struct btrfs_key *min_key, 23495d4f98a2SYan Zheng struct btrfs_key *max_key) 23505d4f98a2SYan Zheng { 23510b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23525d4f98a2SYan Zheng struct inode *inode = NULL; 23535d4f98a2SYan Zheng u64 objectid; 23545d4f98a2SYan Zheng u64 start, end; 235533345d01SLi Zefan u64 ino; 23565d4f98a2SYan Zheng 23575d4f98a2SYan Zheng objectid = min_key->objectid; 23585d4f98a2SYan Zheng while (1) { 23595d4f98a2SYan Zheng cond_resched(); 23605d4f98a2SYan Zheng iput(inode); 23615d4f98a2SYan Zheng 23625d4f98a2SYan Zheng if (objectid > max_key->objectid) 23635d4f98a2SYan Zheng break; 23645d4f98a2SYan Zheng 23655d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 23665d4f98a2SYan Zheng if (!inode) 23675d4f98a2SYan Zheng break; 23684a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 23695d4f98a2SYan Zheng 237033345d01SLi Zefan if (ino > max_key->objectid) { 23715d4f98a2SYan Zheng iput(inode); 23725d4f98a2SYan Zheng break; 23735d4f98a2SYan Zheng } 23745d4f98a2SYan Zheng 237533345d01SLi Zefan objectid = ino + 1; 23765d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 23775d4f98a2SYan Zheng continue; 23785d4f98a2SYan Zheng 237933345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 23805d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 23815d4f98a2SYan Zheng continue; 23825d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 23835d4f98a2SYan Zheng start = 0; 23845d4f98a2SYan Zheng else { 23855d4f98a2SYan Zheng start = min_key->offset; 23860b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 23875d4f98a2SYan Zheng } 23885d4f98a2SYan Zheng } else { 23895d4f98a2SYan Zheng start = 0; 23905d4f98a2SYan Zheng } 23915d4f98a2SYan Zheng 239233345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 23935d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 23945d4f98a2SYan Zheng continue; 23955d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 23965d4f98a2SYan Zheng end = (u64)-1; 23975d4f98a2SYan Zheng } else { 23985d4f98a2SYan Zheng if (max_key->offset == 0) 23995d4f98a2SYan Zheng continue; 24005d4f98a2SYan Zheng end = max_key->offset; 24010b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 24025d4f98a2SYan Zheng end--; 24035d4f98a2SYan Zheng } 24045d4f98a2SYan Zheng } else { 24055d4f98a2SYan Zheng end = (u64)-1; 24065d4f98a2SYan Zheng } 24075d4f98a2SYan Zheng 24085d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2409d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2410dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2411d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 24125d4f98a2SYan Zheng } 24135d4f98a2SYan Zheng return 0; 24145d4f98a2SYan Zheng } 24155d4f98a2SYan Zheng 24165d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 24175d4f98a2SYan Zheng struct btrfs_key *key) 24185d4f98a2SYan Zheng 24195d4f98a2SYan Zheng { 24205d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 24215d4f98a2SYan Zheng if (!path->nodes[level]) 24225d4f98a2SYan Zheng break; 24235d4f98a2SYan Zheng if (path->slots[level] + 1 < 24245d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 24255d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 24265d4f98a2SYan Zheng path->slots[level] + 1); 24275d4f98a2SYan Zheng return 0; 24285d4f98a2SYan Zheng } 24295d4f98a2SYan Zheng level++; 24305d4f98a2SYan Zheng } 24315d4f98a2SYan Zheng return 1; 24325d4f98a2SYan Zheng } 24335d4f98a2SYan Zheng 24345d4f98a2SYan Zheng /* 2435d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 2436d2311e69SQu Wenruo */ 2437d2311e69SQu Wenruo static void insert_dirty_subvol(struct btrfs_trans_handle *trans, 2438d2311e69SQu Wenruo struct reloc_control *rc, 2439d2311e69SQu Wenruo struct btrfs_root *root) 2440d2311e69SQu Wenruo { 2441d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2442d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 2443d2311e69SQu Wenruo 2444d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 2445d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 2446d2311e69SQu Wenruo ASSERT(reloc_root); 2447d2311e69SQu Wenruo 2448d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 2449d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 2450d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 2451d2311e69SQu Wenruo reloc_root_item->drop_level = 0; 2452d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 2453d2311e69SQu Wenruo btrfs_update_reloc_root(trans, root); 2454d2311e69SQu Wenruo 2455d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 245600246528SJosef Bacik btrfs_grab_root(root); 2457d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 2458d2311e69SQu Wenruo } 2459d2311e69SQu Wenruo } 2460d2311e69SQu Wenruo 2461d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 2462d2311e69SQu Wenruo { 2463d2311e69SQu Wenruo struct btrfs_root *root; 2464d2311e69SQu Wenruo struct btrfs_root *next; 2465d2311e69SQu Wenruo int ret = 0; 246630d40577SQu Wenruo int ret2; 2467d2311e69SQu Wenruo 2468d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 2469d2311e69SQu Wenruo reloc_dirty_list) { 247030d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 247130d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 2472d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2473d2311e69SQu Wenruo 2474d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 2475d2311e69SQu Wenruo root->reloc_root = NULL; 24766282675eSQu Wenruo /* 24776282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 24786282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 24796282675eSQu Wenruo */ 24806282675eSQu Wenruo smp_wmb(); 24811fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 2482f28de8d8SJosef Bacik if (reloc_root) { 2483f44deb74SJosef Bacik /* 2484f44deb74SJosef Bacik * btrfs_drop_snapshot drops our ref we hold for 2485f44deb74SJosef Bacik * ->reloc_root. If it fails however we must 2486f44deb74SJosef Bacik * drop the ref ourselves. 2487f44deb74SJosef Bacik */ 2488f28de8d8SJosef Bacik ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); 2489f44deb74SJosef Bacik if (ret2 < 0) { 2490f44deb74SJosef Bacik btrfs_put_root(reloc_root); 2491f44deb74SJosef Bacik if (!ret) 2492f28de8d8SJosef Bacik ret = ret2; 2493f28de8d8SJosef Bacik } 2494f44deb74SJosef Bacik } 249500246528SJosef Bacik btrfs_put_root(root); 249630d40577SQu Wenruo } else { 249730d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 24980078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(root, 0, 1); 2499f44deb74SJosef Bacik if (ret2 < 0) { 2500f44deb74SJosef Bacik btrfs_put_root(root); 2501f44deb74SJosef Bacik if (!ret) 250230d40577SQu Wenruo ret = ret2; 250330d40577SQu Wenruo } 2504d2311e69SQu Wenruo } 2505f44deb74SJosef Bacik } 2506d2311e69SQu Wenruo return ret; 2507d2311e69SQu Wenruo } 2508d2311e69SQu Wenruo 2509d2311e69SQu Wenruo /* 25105d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 25115d4f98a2SYan Zheng * fs tree. 25125d4f98a2SYan Zheng */ 25135d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 25145d4f98a2SYan Zheng struct btrfs_root *root) 25155d4f98a2SYan Zheng { 25160b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 25175d4f98a2SYan Zheng struct btrfs_key key; 25185d4f98a2SYan Zheng struct btrfs_key next_key; 25199e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 25205d4f98a2SYan Zheng struct btrfs_root *reloc_root; 25215d4f98a2SYan Zheng struct btrfs_root_item *root_item; 25225d4f98a2SYan Zheng struct btrfs_path *path; 25233fd0a558SYan, Zheng struct extent_buffer *leaf; 25245d4f98a2SYan Zheng int level; 25255d4f98a2SYan Zheng int max_level; 25265d4f98a2SYan Zheng int replaced = 0; 25275d4f98a2SYan Zheng int ret; 25285d4f98a2SYan Zheng int err = 0; 25293fd0a558SYan, Zheng u32 min_reserved; 25305d4f98a2SYan Zheng 25315d4f98a2SYan Zheng path = btrfs_alloc_path(); 25325d4f98a2SYan Zheng if (!path) 25335d4f98a2SYan Zheng return -ENOMEM; 2534e4058b54SDavid Sterba path->reada = READA_FORWARD; 25355d4f98a2SYan Zheng 25365d4f98a2SYan Zheng reloc_root = root->reloc_root; 25375d4f98a2SYan Zheng root_item = &reloc_root->root_item; 25385d4f98a2SYan Zheng 25395d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 25405d4f98a2SYan Zheng level = btrfs_root_level(root_item); 254167439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 25425d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 25435d4f98a2SYan Zheng path->slots[level] = 0; 25445d4f98a2SYan Zheng } else { 25455d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 25465d4f98a2SYan Zheng 25475d4f98a2SYan Zheng level = root_item->drop_level; 25485d4f98a2SYan Zheng BUG_ON(level == 0); 25495d4f98a2SYan Zheng path->lowest_level = level; 25505d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 255133c66f43SYan Zheng path->lowest_level = 0; 25525d4f98a2SYan Zheng if (ret < 0) { 25535d4f98a2SYan Zheng btrfs_free_path(path); 25545d4f98a2SYan Zheng return ret; 25555d4f98a2SYan Zheng } 25565d4f98a2SYan Zheng 25575d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 25585d4f98a2SYan Zheng path->slots[level]); 25595d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 25605d4f98a2SYan Zheng 25615d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 25625d4f98a2SYan Zheng } 25635d4f98a2SYan Zheng 25640b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 25655d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 25665d4f98a2SYan Zheng 25675d4f98a2SYan Zheng while (1) { 256808e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 256908e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 25703fd0a558SYan, Zheng if (ret) { 25719e6a0c52SJosef Bacik err = ret; 25729e6a0c52SJosef Bacik goto out; 25733fd0a558SYan, Zheng } 25749e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 25759e6a0c52SJosef Bacik if (IS_ERR(trans)) { 25769e6a0c52SJosef Bacik err = PTR_ERR(trans); 25779e6a0c52SJosef Bacik trans = NULL; 25789e6a0c52SJosef Bacik goto out; 25799e6a0c52SJosef Bacik } 25802abc726aSJosef Bacik 25812abc726aSJosef Bacik /* 25822abc726aSJosef Bacik * At this point we no longer have a reloc_control, so we can't 25832abc726aSJosef Bacik * depend on btrfs_init_reloc_root to update our last_trans. 25842abc726aSJosef Bacik * 25852abc726aSJosef Bacik * But that's ok, we started the trans handle on our 25862abc726aSJosef Bacik * corresponding fs_root, which means it's been added to the 25872abc726aSJosef Bacik * dirty list. At commit time we'll still call 25882abc726aSJosef Bacik * btrfs_update_reloc_root() and update our root item 25892abc726aSJosef Bacik * appropriately. 25902abc726aSJosef Bacik */ 25912abc726aSJosef Bacik reloc_root->last_trans = trans->transid; 25929e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 25933fd0a558SYan, Zheng 25943fd0a558SYan, Zheng replaced = 0; 25955d4f98a2SYan Zheng max_level = level; 25965d4f98a2SYan Zheng 25975d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 25985d4f98a2SYan Zheng if (ret < 0) { 25995d4f98a2SYan Zheng err = ret; 26005d4f98a2SYan Zheng goto out; 26015d4f98a2SYan Zheng } 26025d4f98a2SYan Zheng if (ret > 0) 26035d4f98a2SYan Zheng break; 26045d4f98a2SYan Zheng 26055d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 26065d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 26075d4f98a2SYan Zheng ret = 0; 26085d4f98a2SYan Zheng } else { 26093d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 26103fd0a558SYan, Zheng &next_key, level, max_level); 26115d4f98a2SYan Zheng } 26125d4f98a2SYan Zheng if (ret < 0) { 26135d4f98a2SYan Zheng err = ret; 26145d4f98a2SYan Zheng goto out; 26155d4f98a2SYan Zheng } 26165d4f98a2SYan Zheng 26175d4f98a2SYan Zheng if (ret > 0) { 26185d4f98a2SYan Zheng level = ret; 26195d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 26205d4f98a2SYan Zheng path->slots[level]); 26215d4f98a2SYan Zheng replaced = 1; 26225d4f98a2SYan Zheng } 26235d4f98a2SYan Zheng 26245d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 26255d4f98a2SYan Zheng if (ret > 0) 26265d4f98a2SYan Zheng break; 26275d4f98a2SYan Zheng 26285d4f98a2SYan Zheng BUG_ON(level == 0); 26295d4f98a2SYan Zheng /* 26305d4f98a2SYan Zheng * save the merging progress in the drop_progress. 26315d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 26325d4f98a2SYan Zheng */ 26335d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 26345d4f98a2SYan Zheng path->slots[level]); 26355d4f98a2SYan Zheng root_item->drop_level = level; 26365d4f98a2SYan Zheng 26373a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 26389e6a0c52SJosef Bacik trans = NULL; 26395d4f98a2SYan Zheng 26402ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 26415d4f98a2SYan Zheng 26425d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 26435d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 26445d4f98a2SYan Zheng } 26455d4f98a2SYan Zheng 26465d4f98a2SYan Zheng /* 26475d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 26485d4f98a2SYan Zheng * relocated and the block is tree root. 26495d4f98a2SYan Zheng */ 26505d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 26515d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 26525d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 26535d4f98a2SYan Zheng free_extent_buffer(leaf); 26545d4f98a2SYan Zheng if (ret < 0) 26555d4f98a2SYan Zheng err = ret; 26565d4f98a2SYan Zheng out: 26575d4f98a2SYan Zheng btrfs_free_path(path); 26585d4f98a2SYan Zheng 2659d2311e69SQu Wenruo if (err == 0) 2660d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 26615d4f98a2SYan Zheng 26629e6a0c52SJosef Bacik if (trans) 26633a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 26645d4f98a2SYan Zheng 26652ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 26665d4f98a2SYan Zheng 26675d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 26685d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 26695d4f98a2SYan Zheng 26705d4f98a2SYan Zheng return err; 26715d4f98a2SYan Zheng } 26725d4f98a2SYan Zheng 26733fd0a558SYan, Zheng static noinline_for_stack 26743fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 26755d4f98a2SYan Zheng { 26763fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 26770b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 26783fd0a558SYan, Zheng struct btrfs_root *reloc_root; 26795d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 26803fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 26813fd0a558SYan, Zheng u64 num_bytes = 0; 26823fd0a558SYan, Zheng int ret; 26833fd0a558SYan, Zheng 26840b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 26850b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 26863fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 26870b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 26887585717fSChris Mason 26893fd0a558SYan, Zheng again: 26903fd0a558SYan, Zheng if (!err) { 26913fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 269208e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 269308e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 26943fd0a558SYan, Zheng if (ret) 26953fd0a558SYan, Zheng err = ret; 26963fd0a558SYan, Zheng } 26973fd0a558SYan, Zheng 26987a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 26993612b495STsutomu Itoh if (IS_ERR(trans)) { 27003612b495STsutomu Itoh if (!err) 27012ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 270263f018beSNikolay Borisov num_bytes, NULL); 27033612b495STsutomu Itoh return PTR_ERR(trans); 27043612b495STsutomu Itoh } 27053fd0a558SYan, Zheng 27063fd0a558SYan, Zheng if (!err) { 27073fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 27083a45bb20SJeff Mahoney btrfs_end_transaction(trans); 27092ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 271063f018beSNikolay Borisov num_bytes, NULL); 27113fd0a558SYan, Zheng goto again; 27123fd0a558SYan, Zheng } 27133fd0a558SYan, Zheng } 27143fd0a558SYan, Zheng 27153fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 27163fd0a558SYan, Zheng 27173fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 27183fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 27193fd0a558SYan, Zheng struct btrfs_root, root_list); 27203fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 27213fd0a558SYan, Zheng 27220b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 27233fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 27243fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 27253fd0a558SYan, Zheng 27263fd0a558SYan, Zheng /* 27273fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 27283fd0a558SYan, Zheng * knows it should resumes merging 27293fd0a558SYan, Zheng */ 27303fd0a558SYan, Zheng if (!err) 27313fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 27323fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 27333fd0a558SYan, Zheng 27343fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 273500246528SJosef Bacik btrfs_put_root(root); 27363fd0a558SYan, Zheng } 27373fd0a558SYan, Zheng 27383fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 27393fd0a558SYan, Zheng 27403fd0a558SYan, Zheng if (!err) 27413a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 27423fd0a558SYan, Zheng else 27433a45bb20SJeff Mahoney btrfs_end_transaction(trans); 27443fd0a558SYan, Zheng return err; 27453fd0a558SYan, Zheng } 27463fd0a558SYan, Zheng 27473fd0a558SYan, Zheng static noinline_for_stack 2748aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2749aca1bba6SLiu Bo { 2750aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2751aca1bba6SLiu Bo 2752aca1bba6SLiu Bo while (!list_empty(list)) { 2753aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2754aca1bba6SLiu Bo root_list); 2755bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 2756aca1bba6SLiu Bo } 2757aca1bba6SLiu Bo } 2758aca1bba6SLiu Bo 2759aca1bba6SLiu Bo static noinline_for_stack 276094404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 27613fd0a558SYan, Zheng { 27620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27635d4f98a2SYan Zheng struct btrfs_root *root; 27645d4f98a2SYan Zheng struct btrfs_root *reloc_root; 27653fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 27663fd0a558SYan, Zheng int found = 0; 2767aca1bba6SLiu Bo int ret = 0; 27683fd0a558SYan, Zheng again: 27693fd0a558SYan, Zheng root = rc->extent_root; 27707585717fSChris Mason 27717585717fSChris Mason /* 27727585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 27737585717fSChris Mason * we have to make sure nobody is in the middle of 27747585717fSChris Mason * adding their roots to the list while we are 27757585717fSChris Mason * doing this splice 27767585717fSChris Mason */ 27770b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 27783fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 27790b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 27805d4f98a2SYan Zheng 27813fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 27823fd0a558SYan, Zheng found = 1; 27833fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 27843fd0a558SYan, Zheng struct btrfs_root, root_list); 27855d4f98a2SYan Zheng 27865d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 27870b246afaSJeff Mahoney root = read_fs_root(fs_info, 27885d4f98a2SYan Zheng reloc_root->root_key.offset); 27895d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 27905d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 27915d4f98a2SYan Zheng 27923fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 279300246528SJosef Bacik btrfs_put_root(root); 2794b37b39cdSJosef Bacik if (ret) { 279525e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 279625e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 279725e293c2SWang Shilong &reloc_roots); 2798aca1bba6SLiu Bo goto out; 2799b37b39cdSJosef Bacik } 28003fd0a558SYan, Zheng } else { 28013fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 280230d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 280330d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 280430d40577SQu Wenruo &rc->dirty_subvol_roots); 28053fd0a558SYan, Zheng } 28065d4f98a2SYan Zheng } 28075d4f98a2SYan Zheng 28083fd0a558SYan, Zheng if (found) { 28093fd0a558SYan, Zheng found = 0; 28103fd0a558SYan, Zheng goto again; 28115d4f98a2SYan Zheng } 2812aca1bba6SLiu Bo out: 2813aca1bba6SLiu Bo if (ret) { 28140b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2815aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2816aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2817467bb1d2SWang Shilong 2818467bb1d2SWang Shilong /* new reloc root may be added */ 28190b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2820467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 28210b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2822467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2823467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2824aca1bba6SLiu Bo } 2825aca1bba6SLiu Bo 28267b7b7431SJosef Bacik /* 28277b7b7431SJosef Bacik * We used to have 28287b7b7431SJosef Bacik * 28297b7b7431SJosef Bacik * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 28307b7b7431SJosef Bacik * 28317b7b7431SJosef Bacik * here, but it's wrong. If we fail to start the transaction in 28327b7b7431SJosef Bacik * prepare_to_merge() we will have only 0 ref reloc roots, none of which 28337b7b7431SJosef Bacik * have actually been removed from the reloc_root_tree rb tree. This is 28347b7b7431SJosef Bacik * fine because we're bailing here, and we hold a reference on the root 28357b7b7431SJosef Bacik * for the list that holds it, so these roots will be cleaned up when we 28367b7b7431SJosef Bacik * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root 28377b7b7431SJosef Bacik * will be cleaned up on unmount. 28387b7b7431SJosef Bacik * 28397b7b7431SJosef Bacik * The remaining nodes will be cleaned up by free_reloc_control. 28407b7b7431SJosef Bacik */ 28415d4f98a2SYan Zheng } 28425d4f98a2SYan Zheng 28435d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 28445d4f98a2SYan Zheng { 28455d4f98a2SYan Zheng struct tree_block *block; 28465d4f98a2SYan Zheng struct rb_node *rb_node; 28475d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 28485d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 28495d4f98a2SYan Zheng rb_erase(rb_node, blocks); 28505d4f98a2SYan Zheng kfree(block); 28515d4f98a2SYan Zheng } 28525d4f98a2SYan Zheng } 28535d4f98a2SYan Zheng 28545d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 28555d4f98a2SYan Zheng struct btrfs_root *reloc_root) 28565d4f98a2SYan Zheng { 28570b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 28585d4f98a2SYan Zheng struct btrfs_root *root; 2859442b1ac5SJosef Bacik int ret; 28605d4f98a2SYan Zheng 28615d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 28625d4f98a2SYan Zheng return 0; 28635d4f98a2SYan Zheng 28640b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 28655d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 28665d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 2867442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 286800246528SJosef Bacik btrfs_put_root(root); 28695d4f98a2SYan Zheng 2870442b1ac5SJosef Bacik return ret; 28715d4f98a2SYan Zheng } 28725d4f98a2SYan Zheng 28733fd0a558SYan, Zheng static noinline_for_stack 28743fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 28753fd0a558SYan, Zheng struct reloc_control *rc, 2876a26195a5SQu Wenruo struct btrfs_backref_node *node, 2877a26195a5SQu Wenruo struct btrfs_backref_edge *edges[]) 28785d4f98a2SYan Zheng { 2879a26195a5SQu Wenruo struct btrfs_backref_node *next; 28805d4f98a2SYan Zheng struct btrfs_root *root; 28813fd0a558SYan, Zheng int index = 0; 28823fd0a558SYan, Zheng 28835d4f98a2SYan Zheng next = node; 28845d4f98a2SYan Zheng while (1) { 28855d4f98a2SYan Zheng cond_resched(); 28865d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 28875d4f98a2SYan Zheng root = next->root; 28883fd0a558SYan, Zheng BUG_ON(!root); 288927cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 28905d4f98a2SYan Zheng 28915d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 28925d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 28935d4f98a2SYan Zheng break; 28945d4f98a2SYan Zheng } 28955d4f98a2SYan Zheng 28965d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 28973fd0a558SYan, Zheng root = root->reloc_root; 28983fd0a558SYan, Zheng 28993fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 29003fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 29013fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 29023fd0a558SYan, Zheng next->new_bytenr = root->node->start; 290300246528SJosef Bacik btrfs_put_root(next->root); 290400246528SJosef Bacik next->root = btrfs_grab_root(root); 29050b530bc5SJosef Bacik ASSERT(next->root); 29063fd0a558SYan, Zheng list_add_tail(&next->list, 29073fd0a558SYan, Zheng &rc->backref_cache.changed); 29089569cc20SQu Wenruo mark_block_processed(rc, next); 29095d4f98a2SYan Zheng break; 29105d4f98a2SYan Zheng } 29115d4f98a2SYan Zheng 29123fd0a558SYan, Zheng WARN_ON(1); 29135d4f98a2SYan Zheng root = NULL; 29145d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 29155d4f98a2SYan Zheng if (!next || next->level <= node->level) 29165d4f98a2SYan Zheng break; 29175d4f98a2SYan Zheng } 29183fd0a558SYan, Zheng if (!root) 29193fd0a558SYan, Zheng return NULL; 29205d4f98a2SYan Zheng 29213fd0a558SYan, Zheng next = node; 29223fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 29233fd0a558SYan, Zheng while (1) { 29243fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 29253fd0a558SYan, Zheng if (--index < 0) 29263fd0a558SYan, Zheng break; 29273fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 29283fd0a558SYan, Zheng } 29295d4f98a2SYan Zheng return root; 29305d4f98a2SYan Zheng } 29315d4f98a2SYan Zheng 29323fd0a558SYan, Zheng /* 29333fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 29343fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 29353fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 29363fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 29373fd0a558SYan, Zheng */ 29385d4f98a2SYan Zheng static noinline_for_stack 2939a26195a5SQu Wenruo struct btrfs_root *select_one_root(struct btrfs_backref_node *node) 29405d4f98a2SYan Zheng { 2941a26195a5SQu Wenruo struct btrfs_backref_node *next; 29423fd0a558SYan, Zheng struct btrfs_root *root; 29433fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 2944a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 29453fd0a558SYan, Zheng int index = 0; 29463fd0a558SYan, Zheng 29473fd0a558SYan, Zheng next = node; 29483fd0a558SYan, Zheng while (1) { 29493fd0a558SYan, Zheng cond_resched(); 29503fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 29513fd0a558SYan, Zheng root = next->root; 29523fd0a558SYan, Zheng BUG_ON(!root); 29533fd0a558SYan, Zheng 295425985edcSLucas De Marchi /* no other choice for non-references counted tree */ 295527cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 29563fd0a558SYan, Zheng return root; 29573fd0a558SYan, Zheng 29583fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 29593fd0a558SYan, Zheng fs_root = root; 29603fd0a558SYan, Zheng 29613fd0a558SYan, Zheng if (next != node) 29623fd0a558SYan, Zheng return NULL; 29633fd0a558SYan, Zheng 29643fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 29653fd0a558SYan, Zheng if (!next || next->level <= node->level) 29663fd0a558SYan, Zheng break; 29673fd0a558SYan, Zheng } 29683fd0a558SYan, Zheng 29693fd0a558SYan, Zheng if (!fs_root) 29703fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 29713fd0a558SYan, Zheng return fs_root; 29725d4f98a2SYan Zheng } 29735d4f98a2SYan Zheng 29745d4f98a2SYan Zheng static noinline_for_stack 29753fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 2976a26195a5SQu Wenruo struct btrfs_backref_node *node, int reserve) 29775d4f98a2SYan Zheng { 29780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 2979a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 2980a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 2981a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 29823fd0a558SYan, Zheng u64 num_bytes = 0; 29833fd0a558SYan, Zheng int index = 0; 29845d4f98a2SYan Zheng 29853fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 29863fd0a558SYan, Zheng 29873fd0a558SYan, Zheng while (next) { 29883fd0a558SYan, Zheng cond_resched(); 29895d4f98a2SYan Zheng while (1) { 29903fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 29915d4f98a2SYan Zheng break; 29925d4f98a2SYan Zheng 29930b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 29943fd0a558SYan, Zheng 29953fd0a558SYan, Zheng if (list_empty(&next->upper)) 29963fd0a558SYan, Zheng break; 29973fd0a558SYan, Zheng 29983fd0a558SYan, Zheng edge = list_entry(next->upper.next, 2999a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 30003fd0a558SYan, Zheng edges[index++] = edge; 30013fd0a558SYan, Zheng next = edge->node[UPPER]; 30025d4f98a2SYan Zheng } 30033fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 30043fd0a558SYan, Zheng } 30053fd0a558SYan, Zheng return num_bytes; 30063fd0a558SYan, Zheng } 30073fd0a558SYan, Zheng 30083fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 30093fd0a558SYan, Zheng struct reloc_control *rc, 3010a26195a5SQu Wenruo struct btrfs_backref_node *node) 30113fd0a558SYan, Zheng { 30123fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 3013da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 30143fd0a558SYan, Zheng u64 num_bytes; 30153fd0a558SYan, Zheng int ret; 30160647bf56SWang Shilong u64 tmp; 30173fd0a558SYan, Zheng 30183fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 30193fd0a558SYan, Zheng 30203fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 30210647bf56SWang Shilong rc->reserved_bytes += num_bytes; 30228ca17f0fSJosef Bacik 30238ca17f0fSJosef Bacik /* 30248ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 30258ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 30268ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 30278ca17f0fSJosef Bacik */ 30280647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 30298ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 30303fd0a558SYan, Zheng if (ret) { 3031da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 30320647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 30330647bf56SWang Shilong tmp <<= 1; 30340647bf56SWang Shilong /* 30350647bf56SWang Shilong * only one thread can access block_rsv at this point, 30360647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 30370647bf56SWang Shilong * we expand more reservation size here to allow enough 303852042d8eSAndrea Gelmini * space for relocation and we will return earlier in 30390647bf56SWang Shilong * enospc case. 30400647bf56SWang Shilong */ 3041da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 30420647bf56SWang Shilong RELOCATION_RESERVED_NODES; 30438ca17f0fSJosef Bacik return -EAGAIN; 30443fd0a558SYan, Zheng } 30453fd0a558SYan, Zheng 30463fd0a558SYan, Zheng return 0; 30473fd0a558SYan, Zheng } 30483fd0a558SYan, Zheng 30495d4f98a2SYan Zheng /* 30505d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 30515d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 30525d4f98a2SYan Zheng * 30535d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 30545d4f98a2SYan Zheng * in that case this function just updates pointers. 30555d4f98a2SYan Zheng */ 30565d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 30573fd0a558SYan, Zheng struct reloc_control *rc, 3058a26195a5SQu Wenruo struct btrfs_backref_node *node, 30595d4f98a2SYan Zheng struct btrfs_key *key, 30605d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 30615d4f98a2SYan Zheng { 30622ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 3063a26195a5SQu Wenruo struct btrfs_backref_node *upper; 3064a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 3065a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 30665d4f98a2SYan Zheng struct btrfs_root *root; 30675d4f98a2SYan Zheng struct extent_buffer *eb; 30685d4f98a2SYan Zheng u32 blocksize; 30695d4f98a2SYan Zheng u64 bytenr; 30705d4f98a2SYan Zheng u64 generation; 30715d4f98a2SYan Zheng int slot; 30725d4f98a2SYan Zheng int ret; 30735d4f98a2SYan Zheng int err = 0; 30745d4f98a2SYan Zheng 30755d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 30765d4f98a2SYan Zheng 30775d4f98a2SYan Zheng path->lowest_level = node->level + 1; 30783fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 30795d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 3080581c1760SQu Wenruo struct btrfs_key first_key; 308182fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 3082581c1760SQu Wenruo 30835d4f98a2SYan Zheng cond_resched(); 30845d4f98a2SYan Zheng 30855d4f98a2SYan Zheng upper = edge->node[UPPER]; 3086dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 30873fd0a558SYan, Zheng BUG_ON(!root); 30885d4f98a2SYan Zheng 30893fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 30903fd0a558SYan, Zheng if (!lowest) { 30913fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 30923fd0a558SYan, Zheng upper->level, &slot); 3093cbca7d59SFilipe Manana if (ret < 0) { 3094cbca7d59SFilipe Manana err = ret; 3095cbca7d59SFilipe Manana goto next; 3096cbca7d59SFilipe Manana } 30973fd0a558SYan, Zheng BUG_ON(ret); 30983fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 30993fd0a558SYan, Zheng if (node->eb->start == bytenr) 31003fd0a558SYan, Zheng goto next; 31013fd0a558SYan, Zheng } 31025d4f98a2SYan Zheng drop_node_buffer(upper); 31033fd0a558SYan, Zheng } 31045d4f98a2SYan Zheng 31055d4f98a2SYan Zheng if (!upper->eb) { 31065d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 31073561b9dbSLiu Bo if (ret) { 31083561b9dbSLiu Bo if (ret < 0) 31095d4f98a2SYan Zheng err = ret; 31103561b9dbSLiu Bo else 31113561b9dbSLiu Bo err = -ENOENT; 31123561b9dbSLiu Bo 31133561b9dbSLiu Bo btrfs_release_path(path); 31145d4f98a2SYan Zheng break; 31155d4f98a2SYan Zheng } 31165d4f98a2SYan Zheng 31173fd0a558SYan, Zheng if (!upper->eb) { 31183fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 31193fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 31203fd0a558SYan, Zheng } else { 31213fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 31223fd0a558SYan, Zheng } 31233fd0a558SYan, Zheng 31243fd0a558SYan, Zheng upper->locked = 1; 31253fd0a558SYan, Zheng path->locks[upper->level] = 0; 31263fd0a558SYan, Zheng 31275d4f98a2SYan Zheng slot = path->slots[upper->level]; 3128b3b4aa74SDavid Sterba btrfs_release_path(path); 31295d4f98a2SYan Zheng } else { 31305d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 31315d4f98a2SYan Zheng &slot); 3132cbca7d59SFilipe Manana if (ret < 0) { 3133cbca7d59SFilipe Manana err = ret; 3134cbca7d59SFilipe Manana goto next; 3135cbca7d59SFilipe Manana } 31365d4f98a2SYan Zheng BUG_ON(ret); 31375d4f98a2SYan Zheng } 31385d4f98a2SYan Zheng 31395d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 31403fd0a558SYan, Zheng if (lowest) { 31414547f4d8SLiu Bo if (bytenr != node->bytenr) { 31424547f4d8SLiu Bo btrfs_err(root->fs_info, 31434547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 31444547f4d8SLiu Bo bytenr, node->bytenr, slot, 31454547f4d8SLiu Bo upper->eb->start); 31464547f4d8SLiu Bo err = -EIO; 31474547f4d8SLiu Bo goto next; 31484547f4d8SLiu Bo } 31495d4f98a2SYan Zheng } else { 31503fd0a558SYan, Zheng if (node->eb->start == bytenr) 31513fd0a558SYan, Zheng goto next; 31525d4f98a2SYan Zheng } 31535d4f98a2SYan Zheng 3154da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 31555d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 3156581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 3157581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 3158581c1760SQu Wenruo upper->level - 1, &first_key); 315964c043deSLiu Bo if (IS_ERR(eb)) { 316064c043deSLiu Bo err = PTR_ERR(eb); 316164c043deSLiu Bo goto next; 316264c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3163416bc658SJosef Bacik free_extent_buffer(eb); 316497d9a8a4STsutomu Itoh err = -EIO; 316597d9a8a4STsutomu Itoh goto next; 316697d9a8a4STsutomu Itoh } 31675d4f98a2SYan Zheng btrfs_tree_lock(eb); 31688bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 31695d4f98a2SYan Zheng 31705d4f98a2SYan Zheng if (!node->eb) { 31715d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 31725d4f98a2SYan Zheng slot, &eb); 31733fd0a558SYan, Zheng btrfs_tree_unlock(eb); 31743fd0a558SYan, Zheng free_extent_buffer(eb); 31755d4f98a2SYan Zheng if (ret < 0) { 31765d4f98a2SYan Zheng err = ret; 31773fd0a558SYan, Zheng goto next; 31785d4f98a2SYan Zheng } 31793fd0a558SYan, Zheng BUG_ON(node->eb != eb); 31805d4f98a2SYan Zheng } else { 31815d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 31825d4f98a2SYan Zheng node->eb->start); 31835d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 31845d4f98a2SYan Zheng trans->transid); 31855d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 31865d4f98a2SYan Zheng 318782fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 31885d4f98a2SYan Zheng node->eb->start, blocksize, 318982fa113fSQu Wenruo upper->eb->start); 319082fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 319182fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 319282fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 319382fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 31945d4f98a2SYan Zheng BUG_ON(ret); 31955d4f98a2SYan Zheng 31965d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 31975d4f98a2SYan Zheng BUG_ON(ret); 31985d4f98a2SYan Zheng } 31993fd0a558SYan, Zheng next: 32003fd0a558SYan, Zheng if (!upper->pending) 32013fd0a558SYan, Zheng drop_node_buffer(upper); 32023fd0a558SYan, Zheng else 32033fd0a558SYan, Zheng unlock_node_buffer(upper); 32043fd0a558SYan, Zheng if (err) 32053fd0a558SYan, Zheng break; 32065d4f98a2SYan Zheng } 32073fd0a558SYan, Zheng 32083fd0a558SYan, Zheng if (!err && node->pending) { 32093fd0a558SYan, Zheng drop_node_buffer(node); 32103fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 32113fd0a558SYan, Zheng node->pending = 0; 32125d4f98a2SYan Zheng } 32133fd0a558SYan, Zheng 32145d4f98a2SYan Zheng path->lowest_level = 0; 32153fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 32165d4f98a2SYan Zheng return err; 32175d4f98a2SYan Zheng } 32185d4f98a2SYan Zheng 32195d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 32203fd0a558SYan, Zheng struct reloc_control *rc, 3221a26195a5SQu Wenruo struct btrfs_backref_node *node, 32225d4f98a2SYan Zheng struct btrfs_path *path) 32235d4f98a2SYan Zheng { 32245d4f98a2SYan Zheng struct btrfs_key key; 32255d4f98a2SYan Zheng 32265d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 32273fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 32285d4f98a2SYan Zheng } 32295d4f98a2SYan Zheng 32305d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 32313fd0a558SYan, Zheng struct reloc_control *rc, 32323fd0a558SYan, Zheng struct btrfs_path *path, int err) 32335d4f98a2SYan Zheng { 32343fd0a558SYan, Zheng LIST_HEAD(list); 3235a26195a5SQu Wenruo struct btrfs_backref_cache *cache = &rc->backref_cache; 3236a26195a5SQu Wenruo struct btrfs_backref_node *node; 32375d4f98a2SYan Zheng int level; 32385d4f98a2SYan Zheng int ret; 32395d4f98a2SYan Zheng 32405d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 32415d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 32425d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 3243a26195a5SQu Wenruo struct btrfs_backref_node, list); 32443fd0a558SYan, Zheng list_move_tail(&node->list, &list); 32453fd0a558SYan, Zheng BUG_ON(!node->pending); 32465d4f98a2SYan Zheng 32473fd0a558SYan, Zheng if (!err) { 32483fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 32495d4f98a2SYan Zheng if (ret < 0) 32505d4f98a2SYan Zheng err = ret; 32515d4f98a2SYan Zheng } 32525d4f98a2SYan Zheng } 32533fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 32543fd0a558SYan, Zheng } 32555d4f98a2SYan Zheng return err; 32565d4f98a2SYan Zheng } 32575d4f98a2SYan Zheng 32585d4f98a2SYan Zheng /* 32595d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 32605d4f98a2SYan Zheng * as processed. 32615d4f98a2SYan Zheng */ 32625d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 3263a26195a5SQu Wenruo struct btrfs_backref_node *node) 32645d4f98a2SYan Zheng { 3265a26195a5SQu Wenruo struct btrfs_backref_node *next = node; 3266a26195a5SQu Wenruo struct btrfs_backref_edge *edge; 3267a26195a5SQu Wenruo struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 32685d4f98a2SYan Zheng int index = 0; 32695d4f98a2SYan Zheng 32705d4f98a2SYan Zheng while (next) { 32715d4f98a2SYan Zheng cond_resched(); 32725d4f98a2SYan Zheng while (1) { 32735d4f98a2SYan Zheng if (next->processed) 32745d4f98a2SYan Zheng break; 32755d4f98a2SYan Zheng 32769569cc20SQu Wenruo mark_block_processed(rc, next); 32775d4f98a2SYan Zheng 32785d4f98a2SYan Zheng if (list_empty(&next->upper)) 32795d4f98a2SYan Zheng break; 32805d4f98a2SYan Zheng 32815d4f98a2SYan Zheng edge = list_entry(next->upper.next, 3282a26195a5SQu Wenruo struct btrfs_backref_edge, list[LOWER]); 32835d4f98a2SYan Zheng edges[index++] = edge; 32845d4f98a2SYan Zheng next = edge->node[UPPER]; 32855d4f98a2SYan Zheng } 32865d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 32875d4f98a2SYan Zheng } 32885d4f98a2SYan Zheng } 32895d4f98a2SYan Zheng 32907476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 32915d4f98a2SYan Zheng { 3292da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 32937476dfdaSDavid Sterba 32945d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 32959655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 32965d4f98a2SYan Zheng return 1; 32975d4f98a2SYan Zheng return 0; 32985d4f98a2SYan Zheng } 32995d4f98a2SYan Zheng 33002ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 33015d4f98a2SYan Zheng struct tree_block *block) 33025d4f98a2SYan Zheng { 33035d4f98a2SYan Zheng struct extent_buffer *eb; 33045d4f98a2SYan Zheng 3305581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 3306581c1760SQu Wenruo block->level, NULL); 330764c043deSLiu Bo if (IS_ERR(eb)) { 330864c043deSLiu Bo return PTR_ERR(eb); 330964c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3310416bc658SJosef Bacik free_extent_buffer(eb); 3311416bc658SJosef Bacik return -EIO; 3312416bc658SJosef Bacik } 33135d4f98a2SYan Zheng if (block->level == 0) 33145d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 33155d4f98a2SYan Zheng else 33165d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 33175d4f98a2SYan Zheng free_extent_buffer(eb); 33185d4f98a2SYan Zheng block->key_ready = 1; 33195d4f98a2SYan Zheng return 0; 33205d4f98a2SYan Zheng } 33215d4f98a2SYan Zheng 33225d4f98a2SYan Zheng /* 33235d4f98a2SYan Zheng * helper function to relocate a tree block 33245d4f98a2SYan Zheng */ 33255d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 33265d4f98a2SYan Zheng struct reloc_control *rc, 3327a26195a5SQu Wenruo struct btrfs_backref_node *node, 33285d4f98a2SYan Zheng struct btrfs_key *key, 33295d4f98a2SYan Zheng struct btrfs_path *path) 33305d4f98a2SYan Zheng { 33315d4f98a2SYan Zheng struct btrfs_root *root; 33323fd0a558SYan, Zheng int ret = 0; 33335d4f98a2SYan Zheng 33343fd0a558SYan, Zheng if (!node) 33355d4f98a2SYan Zheng return 0; 33363fd0a558SYan, Zheng 33375f6b2e5cSJosef Bacik /* 33385f6b2e5cSJosef Bacik * If we fail here we want to drop our backref_node because we are going 33395f6b2e5cSJosef Bacik * to start over and regenerate the tree for it. 33405f6b2e5cSJosef Bacik */ 33415f6b2e5cSJosef Bacik ret = reserve_metadata_space(trans, rc, node); 33425f6b2e5cSJosef Bacik if (ret) 33435f6b2e5cSJosef Bacik goto out; 33445f6b2e5cSJosef Bacik 33453fd0a558SYan, Zheng BUG_ON(node->processed); 3346147d256eSZhaolei root = select_one_root(node); 33473fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 33483fd0a558SYan, Zheng update_processed_blocks(rc, node); 33493fd0a558SYan, Zheng goto out; 33505d4f98a2SYan Zheng } 33515d4f98a2SYan Zheng 33523fd0a558SYan, Zheng if (root) { 335327cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 33543fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 33553fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 33563fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 33573fd0a558SYan, Zheng root = root->reloc_root; 33583fd0a558SYan, Zheng node->new_bytenr = root->node->start; 335900246528SJosef Bacik btrfs_put_root(node->root); 336000246528SJosef Bacik node->root = btrfs_grab_root(root); 33610b530bc5SJosef Bacik ASSERT(node->root); 33623fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 33633fd0a558SYan, Zheng } else { 33645d4f98a2SYan Zheng path->lowest_level = node->level; 33655d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3366b3b4aa74SDavid Sterba btrfs_release_path(path); 33673fd0a558SYan, Zheng if (ret > 0) 33685d4f98a2SYan Zheng ret = 0; 33693fd0a558SYan, Zheng } 33703fd0a558SYan, Zheng if (!ret) 33713fd0a558SYan, Zheng update_processed_blocks(rc, node); 33723fd0a558SYan, Zheng } else { 33733fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 33743fd0a558SYan, Zheng } 33755d4f98a2SYan Zheng out: 33760647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 33773fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 33785d4f98a2SYan Zheng return ret; 33795d4f98a2SYan Zheng } 33805d4f98a2SYan Zheng 33815d4f98a2SYan Zheng /* 33825d4f98a2SYan Zheng * relocate a list of blocks 33835d4f98a2SYan Zheng */ 33845d4f98a2SYan Zheng static noinline_for_stack 33855d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 33865d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 33875d4f98a2SYan Zheng { 33882ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 3389a26195a5SQu Wenruo struct btrfs_backref_node *node; 33905d4f98a2SYan Zheng struct btrfs_path *path; 33915d4f98a2SYan Zheng struct tree_block *block; 339298ff7b94SQu Wenruo struct tree_block *next; 33935d4f98a2SYan Zheng int ret; 33945d4f98a2SYan Zheng int err = 0; 33955d4f98a2SYan Zheng 33965d4f98a2SYan Zheng path = btrfs_alloc_path(); 3397e1a12670SLiu Bo if (!path) { 3398e1a12670SLiu Bo err = -ENOMEM; 339934c2b290SDavid Sterba goto out_free_blocks; 3400e1a12670SLiu Bo } 34015d4f98a2SYan Zheng 340298ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 340398ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 34045d4f98a2SYan Zheng if (!block->key_ready) 34052ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 34065d4f98a2SYan Zheng } 34075d4f98a2SYan Zheng 340898ff7b94SQu Wenruo /* Get first keys */ 340998ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 341034c2b290SDavid Sterba if (!block->key_ready) { 34112ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 341234c2b290SDavid Sterba if (err) 341334c2b290SDavid Sterba goto out_free_path; 341434c2b290SDavid Sterba } 34155d4f98a2SYan Zheng } 34165d4f98a2SYan Zheng 341798ff7b94SQu Wenruo /* Do tree relocation */ 341898ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 34193fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 34205d4f98a2SYan Zheng block->level, block->bytenr); 34215d4f98a2SYan Zheng if (IS_ERR(node)) { 34225d4f98a2SYan Zheng err = PTR_ERR(node); 34235d4f98a2SYan Zheng goto out; 34245d4f98a2SYan Zheng } 34255d4f98a2SYan Zheng 34265d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 34275d4f98a2SYan Zheng path); 34285d4f98a2SYan Zheng if (ret < 0) { 34295d4f98a2SYan Zheng err = ret; 343050dbbb71SJosef Bacik break; 34315d4f98a2SYan Zheng } 34325d4f98a2SYan Zheng } 34335d4f98a2SYan Zheng out: 34343fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 34355d4f98a2SYan Zheng 343634c2b290SDavid Sterba out_free_path: 34375d4f98a2SYan Zheng btrfs_free_path(path); 343834c2b290SDavid Sterba out_free_blocks: 3439e1a12670SLiu Bo free_block_list(blocks); 34405d4f98a2SYan Zheng return err; 34415d4f98a2SYan Zheng } 34425d4f98a2SYan Zheng 34435d4f98a2SYan Zheng static noinline_for_stack 3444efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3445efa56464SYan, Zheng struct file_extent_cluster *cluster) 3446efa56464SYan, Zheng { 3447efa56464SYan, Zheng u64 alloc_hint = 0; 3448efa56464SYan, Zheng u64 start; 3449efa56464SYan, Zheng u64 end; 3450efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3451efa56464SYan, Zheng u64 num_bytes; 3452efa56464SYan, Zheng int nr = 0; 3453efa56464SYan, Zheng int ret = 0; 3454dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3455dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 345618513091SWang Xiaoguang u64 cur_offset; 3457364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3458efa56464SYan, Zheng 3459efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 34605955102cSAl Viro inode_lock(inode); 3461efa56464SYan, Zheng 3462364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3463dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3464efa56464SYan, Zheng if (ret) 3465efa56464SYan, Zheng goto out; 3466efa56464SYan, Zheng 346718513091SWang Xiaoguang cur_offset = prealloc_start; 3468efa56464SYan, Zheng while (nr < cluster->nr) { 3469efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3470efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3471efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3472efa56464SYan, Zheng else 3473efa56464SYan, Zheng end = cluster->end - offset; 3474efa56464SYan, Zheng 3475d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3476efa56464SYan, Zheng num_bytes = end + 1 - start; 347718513091SWang Xiaoguang if (cur_offset < start) 3478bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3479bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3480efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3481efa56464SYan, Zheng num_bytes, num_bytes, 3482efa56464SYan, Zheng end + 1, &alloc_hint); 348318513091SWang Xiaoguang cur_offset = end + 1; 3484d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3485efa56464SYan, Zheng if (ret) 3486efa56464SYan, Zheng break; 3487efa56464SYan, Zheng nr++; 3488efa56464SYan, Zheng } 348918513091SWang Xiaoguang if (cur_offset < prealloc_end) 3490bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3491bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3492efa56464SYan, Zheng out: 34935955102cSAl Viro inode_unlock(inode); 3494364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3495efa56464SYan, Zheng return ret; 3496efa56464SYan, Zheng } 3497efa56464SYan, Zheng 3498efa56464SYan, Zheng static noinline_for_stack 34990257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 35000257bb82SYan, Zheng u64 block_start) 35015d4f98a2SYan Zheng { 35025d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 35035d4f98a2SYan Zheng struct extent_map *em; 35040257bb82SYan, Zheng int ret = 0; 35055d4f98a2SYan Zheng 3506172ddd60SDavid Sterba em = alloc_extent_map(); 35070257bb82SYan, Zheng if (!em) 35080257bb82SYan, Zheng return -ENOMEM; 35090257bb82SYan, Zheng 35105d4f98a2SYan Zheng em->start = start; 35110257bb82SYan, Zheng em->len = end + 1 - start; 35120257bb82SYan, Zheng em->block_len = em->len; 35130257bb82SYan, Zheng em->block_start = block_start; 35145d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 35155d4f98a2SYan Zheng 3516d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 35175d4f98a2SYan Zheng while (1) { 3518890871beSChris Mason write_lock(&em_tree->lock); 351909a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3520890871beSChris Mason write_unlock(&em_tree->lock); 35215d4f98a2SYan Zheng if (ret != -EEXIST) { 35225d4f98a2SYan Zheng free_extent_map(em); 35235d4f98a2SYan Zheng break; 35245d4f98a2SYan Zheng } 3525dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 35265d4f98a2SYan Zheng } 3527d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 35280257bb82SYan, Zheng return ret; 35290257bb82SYan, Zheng } 35305d4f98a2SYan Zheng 3531726a3421SQu Wenruo /* 3532726a3421SQu Wenruo * Allow error injection to test balance cancellation 3533726a3421SQu Wenruo */ 3534726a3421SQu Wenruo int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 3535726a3421SQu Wenruo { 3536726a3421SQu Wenruo return atomic_read(&fs_info->balance_cancel_req); 3537726a3421SQu Wenruo } 3538726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 3539726a3421SQu Wenruo 35400257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 35410257bb82SYan, Zheng struct file_extent_cluster *cluster) 35420257bb82SYan, Zheng { 35432ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 35440257bb82SYan, Zheng u64 page_start; 35450257bb82SYan, Zheng u64 page_end; 35460257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 35470257bb82SYan, Zheng unsigned long index; 35480257bb82SYan, Zheng unsigned long last_index; 35490257bb82SYan, Zheng struct page *page; 35500257bb82SYan, Zheng struct file_ra_state *ra; 35513b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 35520257bb82SYan, Zheng int nr = 0; 35530257bb82SYan, Zheng int ret = 0; 35540257bb82SYan, Zheng 35550257bb82SYan, Zheng if (!cluster->nr) 35560257bb82SYan, Zheng return 0; 35570257bb82SYan, Zheng 35580257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 35590257bb82SYan, Zheng if (!ra) 35600257bb82SYan, Zheng return -ENOMEM; 35610257bb82SYan, Zheng 3562efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 35630257bb82SYan, Zheng if (ret) 3564efa56464SYan, Zheng goto out; 35650257bb82SYan, Zheng 35660257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 35670257bb82SYan, Zheng 3568efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3569efa56464SYan, Zheng cluster->end - offset, cluster->start); 3570efa56464SYan, Zheng if (ret) 3571efa56464SYan, Zheng goto out; 3572efa56464SYan, Zheng 357309cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 357409cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 35750257bb82SYan, Zheng while (index <= last_index) { 35769f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 35779f3db423SNikolay Borisov PAGE_SIZE); 3578efa56464SYan, Zheng if (ret) 3579efa56464SYan, Zheng goto out; 3580efa56464SYan, Zheng 35810257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 35820257bb82SYan, Zheng if (!page) { 35830257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 35840257bb82SYan, Zheng ra, NULL, index, 35850257bb82SYan, Zheng last_index + 1 - index); 3586a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 35873b16a4e3SJosef Bacik mask); 35880257bb82SYan, Zheng if (!page) { 3589691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 359043b18595SQu Wenruo PAGE_SIZE, true); 359144db1216SFilipe Manana btrfs_delalloc_release_extents(BTRFS_I(inode), 35928702ba93SQu Wenruo PAGE_SIZE); 35930257bb82SYan, Zheng ret = -ENOMEM; 3594efa56464SYan, Zheng goto out; 35950257bb82SYan, Zheng } 35960257bb82SYan, Zheng } 35970257bb82SYan, Zheng 35980257bb82SYan, Zheng if (PageReadahead(page)) { 35990257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 36000257bb82SYan, Zheng ra, NULL, page, index, 36010257bb82SYan, Zheng last_index + 1 - index); 36020257bb82SYan, Zheng } 36030257bb82SYan, Zheng 36040257bb82SYan, Zheng if (!PageUptodate(page)) { 36050257bb82SYan, Zheng btrfs_readpage(NULL, page); 36060257bb82SYan, Zheng lock_page(page); 36070257bb82SYan, Zheng if (!PageUptodate(page)) { 36080257bb82SYan, Zheng unlock_page(page); 360909cbfeafSKirill A. Shutemov put_page(page); 3610691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 361143b18595SQu Wenruo PAGE_SIZE, true); 36128b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 36138702ba93SQu Wenruo PAGE_SIZE); 36140257bb82SYan, Zheng ret = -EIO; 3615efa56464SYan, Zheng goto out; 36160257bb82SYan, Zheng } 36170257bb82SYan, Zheng } 36180257bb82SYan, Zheng 36194eee4fa4SMiao Xie page_start = page_offset(page); 362009cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 36210257bb82SYan, Zheng 3622d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 36230257bb82SYan, Zheng 36240257bb82SYan, Zheng set_page_extent_mapped(page); 36250257bb82SYan, Zheng 36260257bb82SYan, Zheng if (nr < cluster->nr && 36270257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 36280257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 36290257bb82SYan, Zheng page_start, page_end, 3630ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 36310257bb82SYan, Zheng nr++; 36320257bb82SYan, Zheng } 36330257bb82SYan, Zheng 3634765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3635330a5827SNikolay Borisov NULL); 3636765f3cebSNikolay Borisov if (ret) { 3637765f3cebSNikolay Borisov unlock_page(page); 3638765f3cebSNikolay Borisov put_page(page); 3639765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 364043b18595SQu Wenruo PAGE_SIZE, true); 3641765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 36428702ba93SQu Wenruo PAGE_SIZE); 3643765f3cebSNikolay Borisov 3644765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3645765f3cebSNikolay Borisov page_start, page_end, 3646765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3647765f3cebSNikolay Borisov goto out; 3648765f3cebSNikolay Borisov 3649765f3cebSNikolay Borisov } 36500257bb82SYan, Zheng set_page_dirty(page); 36510257bb82SYan, Zheng 36520257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3653d0082371SJeff Mahoney page_start, page_end); 36540257bb82SYan, Zheng unlock_page(page); 365509cbfeafSKirill A. Shutemov put_page(page); 36560257bb82SYan, Zheng 36570257bb82SYan, Zheng index++; 36588702ba93SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); 3659efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 36602ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 36617f913c7cSQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 36627f913c7cSQu Wenruo ret = -ECANCELED; 36637f913c7cSQu Wenruo goto out; 36647f913c7cSQu Wenruo } 36650257bb82SYan, Zheng } 36660257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3667efa56464SYan, Zheng out: 36680257bb82SYan, Zheng kfree(ra); 36690257bb82SYan, Zheng return ret; 36700257bb82SYan, Zheng } 36710257bb82SYan, Zheng 36720257bb82SYan, Zheng static noinline_for_stack 36730257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 36740257bb82SYan, Zheng struct file_extent_cluster *cluster) 36750257bb82SYan, Zheng { 36760257bb82SYan, Zheng int ret; 36770257bb82SYan, Zheng 36780257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 36790257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 36800257bb82SYan, Zheng if (ret) 36810257bb82SYan, Zheng return ret; 36820257bb82SYan, Zheng cluster->nr = 0; 36830257bb82SYan, Zheng } 36840257bb82SYan, Zheng 36850257bb82SYan, Zheng if (!cluster->nr) 36860257bb82SYan, Zheng cluster->start = extent_key->objectid; 36870257bb82SYan, Zheng else 36880257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 36890257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 36900257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 36910257bb82SYan, Zheng cluster->nr++; 36920257bb82SYan, Zheng 36930257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 36940257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 36950257bb82SYan, Zheng if (ret) 36960257bb82SYan, Zheng return ret; 36970257bb82SYan, Zheng cluster->nr = 0; 36980257bb82SYan, Zheng } 36990257bb82SYan, Zheng return 0; 37005d4f98a2SYan Zheng } 37015d4f98a2SYan Zheng 37025d4f98a2SYan Zheng /* 37035d4f98a2SYan Zheng * helper to add a tree block to the list. 37045d4f98a2SYan Zheng * the major work is getting the generation and level of the block 37055d4f98a2SYan Zheng */ 37065d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 37075d4f98a2SYan Zheng struct btrfs_key *extent_key, 37085d4f98a2SYan Zheng struct btrfs_path *path, 37095d4f98a2SYan Zheng struct rb_root *blocks) 37105d4f98a2SYan Zheng { 37115d4f98a2SYan Zheng struct extent_buffer *eb; 37125d4f98a2SYan Zheng struct btrfs_extent_item *ei; 37135d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 37145d4f98a2SYan Zheng struct tree_block *block; 37155d4f98a2SYan Zheng struct rb_node *rb_node; 37165d4f98a2SYan Zheng u32 item_size; 37175d4f98a2SYan Zheng int level = -1; 37187fdf4b60SWang Shilong u64 generation; 37195d4f98a2SYan Zheng 37205d4f98a2SYan Zheng eb = path->nodes[0]; 37215d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 37225d4f98a2SYan Zheng 37233173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 37243173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 37255d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 37265d4f98a2SYan Zheng struct btrfs_extent_item); 37273173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 37285d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 37295d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 37305d4f98a2SYan Zheng } else { 37313173a18fSJosef Bacik level = (int)extent_key->offset; 37323173a18fSJosef Bacik } 37333173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 37346d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3735ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3736ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3737ba3c2b19SNikolay Borisov return -EINVAL; 37383173a18fSJosef Bacik } else { 37395d4f98a2SYan Zheng BUG(); 37405d4f98a2SYan Zheng } 37415d4f98a2SYan Zheng 3742b3b4aa74SDavid Sterba btrfs_release_path(path); 37435d4f98a2SYan Zheng 37445d4f98a2SYan Zheng BUG_ON(level == -1); 37455d4f98a2SYan Zheng 37465d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 37475d4f98a2SYan Zheng if (!block) 37485d4f98a2SYan Zheng return -ENOMEM; 37495d4f98a2SYan Zheng 37505d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3751da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 37525d4f98a2SYan Zheng block->key.offset = generation; 37535d4f98a2SYan Zheng block->level = level; 37545d4f98a2SYan Zheng block->key_ready = 0; 37555d4f98a2SYan Zheng 37565d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 375743c04fb1SJeff Mahoney if (rb_node) 375843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 37595d4f98a2SYan Zheng 37605d4f98a2SYan Zheng return 0; 37615d4f98a2SYan Zheng } 37625d4f98a2SYan Zheng 37635d4f98a2SYan Zheng /* 37645d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 37655d4f98a2SYan Zheng */ 37665d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 37675d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 37685d4f98a2SYan Zheng struct rb_root *blocks) 37695d4f98a2SYan Zheng { 37700b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 37715d4f98a2SYan Zheng struct btrfs_path *path; 37725d4f98a2SYan Zheng struct btrfs_key key; 37735d4f98a2SYan Zheng int ret; 37740b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 37755d4f98a2SYan Zheng 37767476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 37775d4f98a2SYan Zheng return 0; 37785d4f98a2SYan Zheng 37795d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 37805d4f98a2SYan Zheng return 0; 37815d4f98a2SYan Zheng 37825d4f98a2SYan Zheng path = btrfs_alloc_path(); 37835d4f98a2SYan Zheng if (!path) 37845d4f98a2SYan Zheng return -ENOMEM; 3785aee68ee5SJosef Bacik again: 37865d4f98a2SYan Zheng key.objectid = bytenr; 3787aee68ee5SJosef Bacik if (skinny) { 3788aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3789aee68ee5SJosef Bacik key.offset = (u64)-1; 3790aee68ee5SJosef Bacik } else { 37915d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 37925d4f98a2SYan Zheng key.offset = blocksize; 3793aee68ee5SJosef Bacik } 37945d4f98a2SYan Zheng 37955d4f98a2SYan Zheng path->search_commit_root = 1; 37965d4f98a2SYan Zheng path->skip_locking = 1; 37975d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 37985d4f98a2SYan Zheng if (ret < 0) 37995d4f98a2SYan Zheng goto out; 38005d4f98a2SYan Zheng 3801aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3802aee68ee5SJosef Bacik if (path->slots[0]) { 3803aee68ee5SJosef Bacik path->slots[0]--; 3804aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3805aee68ee5SJosef Bacik path->slots[0]); 38063173a18fSJosef Bacik if (key.objectid == bytenr && 3807aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3808aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3809aee68ee5SJosef Bacik key.offset == blocksize))) 38103173a18fSJosef Bacik ret = 0; 38113173a18fSJosef Bacik } 3812aee68ee5SJosef Bacik 3813aee68ee5SJosef Bacik if (ret) { 3814aee68ee5SJosef Bacik skinny = false; 3815aee68ee5SJosef Bacik btrfs_release_path(path); 3816aee68ee5SJosef Bacik goto again; 3817aee68ee5SJosef Bacik } 3818aee68ee5SJosef Bacik } 3819cdccee99SLiu Bo if (ret) { 3820cdccee99SLiu Bo ASSERT(ret == 1); 3821cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3822cdccee99SLiu Bo btrfs_err(fs_info, 3823cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3824cdccee99SLiu Bo bytenr); 3825cdccee99SLiu Bo WARN_ON(1); 3826cdccee99SLiu Bo ret = -EINVAL; 3827cdccee99SLiu Bo goto out; 3828cdccee99SLiu Bo } 38293173a18fSJosef Bacik 38305d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 38315d4f98a2SYan Zheng out: 38325d4f98a2SYan Zheng btrfs_free_path(path); 38335d4f98a2SYan Zheng return ret; 38345d4f98a2SYan Zheng } 38355d4f98a2SYan Zheng 38360af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 383732da5386SDavid Sterba struct btrfs_block_group *block_group, 38381bbc621eSChris Mason struct inode *inode, 38391bbc621eSChris Mason u64 ino) 38400af3d00bSJosef Bacik { 38410af3d00bSJosef Bacik struct btrfs_key key; 38420af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 38430af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 38440af3d00bSJosef Bacik int ret = 0; 38450af3d00bSJosef Bacik 38460af3d00bSJosef Bacik if (inode) 38470af3d00bSJosef Bacik goto truncate; 38480af3d00bSJosef Bacik 38490af3d00bSJosef Bacik key.objectid = ino; 38500af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 38510af3d00bSJosef Bacik key.offset = 0; 38520af3d00bSJosef Bacik 38534c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 38542e19f1f9SAl Viro if (IS_ERR(inode)) 38550af3d00bSJosef Bacik return -ENOENT; 38560af3d00bSJosef Bacik 38570af3d00bSJosef Bacik truncate: 38582ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 38597b61cd92SMiao Xie &fs_info->global_block_rsv); 38607b61cd92SMiao Xie if (ret) 38617b61cd92SMiao Xie goto out; 38627b61cd92SMiao Xie 38637a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 38640af3d00bSJosef Bacik if (IS_ERR(trans)) { 38653612b495STsutomu Itoh ret = PTR_ERR(trans); 38660af3d00bSJosef Bacik goto out; 38670af3d00bSJosef Bacik } 38680af3d00bSJosef Bacik 386977ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 38700af3d00bSJosef Bacik 38713a45bb20SJeff Mahoney btrfs_end_transaction(trans); 38722ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 38730af3d00bSJosef Bacik out: 38740af3d00bSJosef Bacik iput(inode); 38750af3d00bSJosef Bacik return ret; 38760af3d00bSJosef Bacik } 38770af3d00bSJosef Bacik 38785d4f98a2SYan Zheng /* 387919b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 388019b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 38815d4f98a2SYan Zheng */ 388219b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 388319b546d7SQu Wenruo struct btrfs_block_group *block_group, 388419b546d7SQu Wenruo u64 data_bytenr) 38855d4f98a2SYan Zheng { 388619b546d7SQu Wenruo u64 space_cache_ino; 388719b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 38885d4f98a2SYan Zheng struct btrfs_key key; 388919b546d7SQu Wenruo bool found = false; 389019b546d7SQu Wenruo int i; 38915d4f98a2SYan Zheng int ret; 38925d4f98a2SYan Zheng 389319b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 389419b546d7SQu Wenruo return 0; 38955d4f98a2SYan Zheng 389619b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 389719b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 389819b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 389919b546d7SQu Wenruo continue; 390019b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 390119b546d7SQu Wenruo if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_REG && 390219b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 390319b546d7SQu Wenruo found = true; 390419b546d7SQu Wenruo space_cache_ino = key.objectid; 390519b546d7SQu Wenruo break; 390619b546d7SQu Wenruo } 390719b546d7SQu Wenruo } 390819b546d7SQu Wenruo if (!found) 390919b546d7SQu Wenruo return -ENOENT; 391019b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 391119b546d7SQu Wenruo space_cache_ino); 39120af3d00bSJosef Bacik return ret; 39135d4f98a2SYan Zheng } 39145d4f98a2SYan Zheng 39155d4f98a2SYan Zheng /* 39162c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 39175d4f98a2SYan Zheng */ 39185d4f98a2SYan Zheng static noinline_for_stack 39195d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 39205d4f98a2SYan Zheng struct btrfs_key *extent_key, 39215d4f98a2SYan Zheng struct btrfs_path *path, 39225d4f98a2SYan Zheng struct rb_root *blocks) 39235d4f98a2SYan Zheng { 392419b546d7SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 392519b546d7SQu Wenruo struct ulist *leaves = NULL; 392619b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 392719b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 392819b546d7SQu Wenruo const u32 blocksize = fs_info->nodesize; 3929647f63bdSFilipe David Borba Manana int ret = 0; 39305d4f98a2SYan Zheng 3931b3b4aa74SDavid Sterba btrfs_release_path(path); 393219b546d7SQu Wenruo ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid, 393319b546d7SQu Wenruo 0, &leaves, NULL, true); 393419b546d7SQu Wenruo if (ret < 0) 393519b546d7SQu Wenruo return ret; 393619b546d7SQu Wenruo 393719b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 393819b546d7SQu Wenruo while ((ref_node = ulist_next(leaves, &leaf_uiter))) { 393919b546d7SQu Wenruo struct extent_buffer *eb; 394019b546d7SQu Wenruo 394119b546d7SQu Wenruo eb = read_tree_block(fs_info, ref_node->val, 0, 0, NULL); 394219b546d7SQu Wenruo if (IS_ERR(eb)) { 394319b546d7SQu Wenruo ret = PTR_ERR(eb); 394419b546d7SQu Wenruo break; 394519b546d7SQu Wenruo } 394619b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 394719b546d7SQu Wenruo extent_key->objectid); 394819b546d7SQu Wenruo free_extent_buffer(eb); 394919b546d7SQu Wenruo if (ret < 0) 395019b546d7SQu Wenruo break; 395119b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 395219b546d7SQu Wenruo if (ret < 0) 395319b546d7SQu Wenruo break; 395419b546d7SQu Wenruo } 395519b546d7SQu Wenruo if (ret < 0) 39565d4f98a2SYan Zheng free_block_list(blocks); 395719b546d7SQu Wenruo ulist_free(leaves); 395819b546d7SQu Wenruo return ret; 39595d4f98a2SYan Zheng } 39605d4f98a2SYan Zheng 39615d4f98a2SYan Zheng /* 39622c016dc2SLiu Bo * helper to find next unprocessed extent 39635d4f98a2SYan Zheng */ 39645d4f98a2SYan Zheng static noinline_for_stack 3965147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 39663fd0a558SYan, Zheng struct btrfs_key *extent_key) 39675d4f98a2SYan Zheng { 39680b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39695d4f98a2SYan Zheng struct btrfs_key key; 39705d4f98a2SYan Zheng struct extent_buffer *leaf; 39715d4f98a2SYan Zheng u64 start, end, last; 39725d4f98a2SYan Zheng int ret; 39735d4f98a2SYan Zheng 3974b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 39755d4f98a2SYan Zheng while (1) { 39765d4f98a2SYan Zheng cond_resched(); 39775d4f98a2SYan Zheng if (rc->search_start >= last) { 39785d4f98a2SYan Zheng ret = 1; 39795d4f98a2SYan Zheng break; 39805d4f98a2SYan Zheng } 39815d4f98a2SYan Zheng 39825d4f98a2SYan Zheng key.objectid = rc->search_start; 39835d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 39845d4f98a2SYan Zheng key.offset = 0; 39855d4f98a2SYan Zheng 39865d4f98a2SYan Zheng path->search_commit_root = 1; 39875d4f98a2SYan Zheng path->skip_locking = 1; 39885d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 39895d4f98a2SYan Zheng 0, 0); 39905d4f98a2SYan Zheng if (ret < 0) 39915d4f98a2SYan Zheng break; 39925d4f98a2SYan Zheng next: 39935d4f98a2SYan Zheng leaf = path->nodes[0]; 39945d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 39955d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 39965d4f98a2SYan Zheng if (ret != 0) 39975d4f98a2SYan Zheng break; 39985d4f98a2SYan Zheng leaf = path->nodes[0]; 39995d4f98a2SYan Zheng } 40005d4f98a2SYan Zheng 40015d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 40025d4f98a2SYan Zheng if (key.objectid >= last) { 40035d4f98a2SYan Zheng ret = 1; 40045d4f98a2SYan Zheng break; 40055d4f98a2SYan Zheng } 40065d4f98a2SYan Zheng 40073173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 40083173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 40093173a18fSJosef Bacik path->slots[0]++; 40103173a18fSJosef Bacik goto next; 40113173a18fSJosef Bacik } 40123173a18fSJosef Bacik 40133173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 40145d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 40155d4f98a2SYan Zheng path->slots[0]++; 40165d4f98a2SYan Zheng goto next; 40175d4f98a2SYan Zheng } 40185d4f98a2SYan Zheng 40193173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 40200b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 40213173a18fSJosef Bacik rc->search_start) { 40223173a18fSJosef Bacik path->slots[0]++; 40233173a18fSJosef Bacik goto next; 40243173a18fSJosef Bacik } 40253173a18fSJosef Bacik 40265d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 40275d4f98a2SYan Zheng key.objectid, &start, &end, 4028e6138876SJosef Bacik EXTENT_DIRTY, NULL); 40295d4f98a2SYan Zheng 40305d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 4031b3b4aa74SDavid Sterba btrfs_release_path(path); 40325d4f98a2SYan Zheng rc->search_start = end + 1; 40335d4f98a2SYan Zheng } else { 40343173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 40355d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 40363173a18fSJosef Bacik else 40373173a18fSJosef Bacik rc->search_start = key.objectid + 40380b246afaSJeff Mahoney fs_info->nodesize; 40393fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 40405d4f98a2SYan Zheng return 0; 40415d4f98a2SYan Zheng } 40425d4f98a2SYan Zheng } 4043b3b4aa74SDavid Sterba btrfs_release_path(path); 40445d4f98a2SYan Zheng return ret; 40455d4f98a2SYan Zheng } 40465d4f98a2SYan Zheng 40475d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 40485d4f98a2SYan Zheng { 40495d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40507585717fSChris Mason 40517585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 40525d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 40537585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 40545d4f98a2SYan Zheng } 40555d4f98a2SYan Zheng 40565d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 40575d4f98a2SYan Zheng { 40585d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40597585717fSChris Mason 40607585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 40615d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 40627585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 40635d4f98a2SYan Zheng } 40645d4f98a2SYan Zheng 40655d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 40665d4f98a2SYan Zheng { 40675d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 40685d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 40695d4f98a2SYan Zheng return 1; 40705d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 40715d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 40725d4f98a2SYan Zheng return 1; 40735d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 40745d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 40755d4f98a2SYan Zheng return 1; 40765d4f98a2SYan Zheng return 0; 40775d4f98a2SYan Zheng } 40785d4f98a2SYan Zheng 40793fd0a558SYan, Zheng static noinline_for_stack 40803fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 40813fd0a558SYan, Zheng { 40823fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 4083ac2fabacSJosef Bacik int ret; 40843fd0a558SYan, Zheng 40852ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 408666d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 40873fd0a558SYan, Zheng if (!rc->block_rsv) 40883fd0a558SYan, Zheng return -ENOMEM; 40893fd0a558SYan, Zheng 40903fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 4091b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 40923fd0a558SYan, Zheng rc->extents_found = 0; 40933fd0a558SYan, Zheng rc->nodes_relocated = 0; 40943fd0a558SYan, Zheng rc->merging_rsv_size = 0; 40950647bf56SWang Shilong rc->reserved_bytes = 0; 4096da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 40970647bf56SWang Shilong RELOCATION_RESERVED_NODES; 4098ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 4099ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 4100ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 4101ac2fabacSJosef Bacik if (ret) 4102ac2fabacSJosef Bacik return ret; 41033fd0a558SYan, Zheng 41043fd0a558SYan, Zheng rc->create_reloc_tree = 1; 41053fd0a558SYan, Zheng set_reloc_control(rc); 41063fd0a558SYan, Zheng 41077a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 410828818947SLiu Bo if (IS_ERR(trans)) { 410928818947SLiu Bo unset_reloc_control(rc); 411028818947SLiu Bo /* 411128818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 411228818947SLiu Bo * cleanup. And callers are responsible to free the above 411328818947SLiu Bo * block rsv. 411428818947SLiu Bo */ 411528818947SLiu Bo return PTR_ERR(trans); 411628818947SLiu Bo } 41173a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 41183fd0a558SYan, Zheng return 0; 41193fd0a558SYan, Zheng } 412076dda93cSYan, Zheng 41215d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 41225d4f98a2SYan Zheng { 41232ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 41245d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 41255d4f98a2SYan Zheng struct btrfs_key key; 41265d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 41275d4f98a2SYan Zheng struct btrfs_path *path; 41285d4f98a2SYan Zheng struct btrfs_extent_item *ei; 41295d4f98a2SYan Zheng u64 flags; 41305d4f98a2SYan Zheng u32 item_size; 41315d4f98a2SYan Zheng int ret; 41325d4f98a2SYan Zheng int err = 0; 4133c87f08caSChris Mason int progress = 0; 41345d4f98a2SYan Zheng 41355d4f98a2SYan Zheng path = btrfs_alloc_path(); 41363fd0a558SYan, Zheng if (!path) 41375d4f98a2SYan Zheng return -ENOMEM; 4138e4058b54SDavid Sterba path->reada = READA_FORWARD; 41393fd0a558SYan, Zheng 41403fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 41413fd0a558SYan, Zheng if (ret) { 41423fd0a558SYan, Zheng err = ret; 41433fd0a558SYan, Zheng goto out_free; 41442423fdfbSJiri Slaby } 41455d4f98a2SYan Zheng 41465d4f98a2SYan Zheng while (1) { 41470647bf56SWang Shilong rc->reserved_bytes = 0; 41480647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 41490647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 41500647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 41510647bf56SWang Shilong if (ret) { 41520647bf56SWang Shilong err = ret; 41530647bf56SWang Shilong break; 41540647bf56SWang Shilong } 4155c87f08caSChris Mason progress++; 4156a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 41570f788c58SLiu Bo if (IS_ERR(trans)) { 41580f788c58SLiu Bo err = PTR_ERR(trans); 41590f788c58SLiu Bo trans = NULL; 41600f788c58SLiu Bo break; 41610f788c58SLiu Bo } 4162c87f08caSChris Mason restart: 41633fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 41643a45bb20SJeff Mahoney btrfs_end_transaction(trans); 416542a657f5SPan Bian trans = NULL; 41663fd0a558SYan, Zheng continue; 41673fd0a558SYan, Zheng } 41683fd0a558SYan, Zheng 4169147d256eSZhaolei ret = find_next_extent(rc, path, &key); 41705d4f98a2SYan Zheng if (ret < 0) 41715d4f98a2SYan Zheng err = ret; 41725d4f98a2SYan Zheng if (ret != 0) 41735d4f98a2SYan Zheng break; 41745d4f98a2SYan Zheng 41755d4f98a2SYan Zheng rc->extents_found++; 41765d4f98a2SYan Zheng 41775d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 41785d4f98a2SYan Zheng struct btrfs_extent_item); 41793fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 41805d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 41815d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 41825d4f98a2SYan Zheng ret = check_extent_flags(flags); 41835d4f98a2SYan Zheng BUG_ON(ret); 41846d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 4185ba3c2b19SNikolay Borisov err = -EINVAL; 4186ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 4187ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 4188ba3c2b19SNikolay Borisov break; 41895d4f98a2SYan Zheng } else { 41905d4f98a2SYan Zheng BUG(); 41915d4f98a2SYan Zheng } 41925d4f98a2SYan Zheng 41935d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 41945d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 41955d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 41965d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41975d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 41985d4f98a2SYan Zheng } else { 4199b3b4aa74SDavid Sterba btrfs_release_path(path); 42005d4f98a2SYan Zheng ret = 0; 42015d4f98a2SYan Zheng } 42025d4f98a2SYan Zheng if (ret < 0) { 42033fd0a558SYan, Zheng err = ret; 42045d4f98a2SYan Zheng break; 42055d4f98a2SYan Zheng } 42065d4f98a2SYan Zheng 42075d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 42085d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 42095d4f98a2SYan Zheng if (ret < 0) { 42103fd0a558SYan, Zheng if (ret != -EAGAIN) { 42115d4f98a2SYan Zheng err = ret; 42125d4f98a2SYan Zheng break; 42135d4f98a2SYan Zheng } 42143fd0a558SYan, Zheng rc->extents_found--; 42153fd0a558SYan, Zheng rc->search_start = key.objectid; 42163fd0a558SYan, Zheng } 42175d4f98a2SYan Zheng } 42185d4f98a2SYan Zheng 42193a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 42202ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42213fd0a558SYan, Zheng trans = NULL; 42225d4f98a2SYan Zheng 42235d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 42245d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 42255d4f98a2SYan Zheng rc->found_file_extent = 1; 42260257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 42273fd0a558SYan, Zheng &key, &rc->cluster); 42285d4f98a2SYan Zheng if (ret < 0) { 42295d4f98a2SYan Zheng err = ret; 42305d4f98a2SYan Zheng break; 42315d4f98a2SYan Zheng } 42325d4f98a2SYan Zheng } 4233f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 4234f31ea088SQu Wenruo err = -ECANCELED; 4235f31ea088SQu Wenruo break; 4236f31ea088SQu Wenruo } 42375d4f98a2SYan Zheng } 4238c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 423943a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 42409689457bSShilong Wang if (ret == 1) { 4241c87f08caSChris Mason err = 0; 4242c87f08caSChris Mason progress = 0; 4243c87f08caSChris Mason goto restart; 4244c87f08caSChris Mason } 4245c87f08caSChris Mason } 42463fd0a558SYan, Zheng 4247b3b4aa74SDavid Sterba btrfs_release_path(path); 424891166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 42495d4f98a2SYan Zheng 42505d4f98a2SYan Zheng if (trans) { 42513a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 42522ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42535d4f98a2SYan Zheng } 42545d4f98a2SYan Zheng 42550257bb82SYan, Zheng if (!err) { 42563fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 42573fd0a558SYan, Zheng &rc->cluster); 42580257bb82SYan, Zheng if (ret < 0) 42590257bb82SYan, Zheng err = ret; 42600257bb82SYan, Zheng } 42610257bb82SYan, Zheng 42623fd0a558SYan, Zheng rc->create_reloc_tree = 0; 42633fd0a558SYan, Zheng set_reloc_control(rc); 42640257bb82SYan, Zheng 42653fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 426663f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 42675d4f98a2SYan Zheng 42687f913c7cSQu Wenruo /* 42697f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 42707f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 42717f913c7cSQu Wenruo * 42727f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 42737f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 42747f913c7cSQu Wenruo * merge_reloc_roots() 42757f913c7cSQu Wenruo */ 42763fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 42775d4f98a2SYan Zheng 42785d4f98a2SYan Zheng merge_reloc_roots(rc); 42795d4f98a2SYan Zheng 42803fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 42815d4f98a2SYan Zheng unset_reloc_control(rc); 428263f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 42835d4f98a2SYan Zheng 42845d4f98a2SYan Zheng /* get rid of pinned extents */ 42857a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 428662b99540SQu Wenruo if (IS_ERR(trans)) { 42873612b495STsutomu Itoh err = PTR_ERR(trans); 428862b99540SQu Wenruo goto out_free; 428962b99540SQu Wenruo } 42903a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 42916217b0faSJosef Bacik out_free: 4292d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4293d2311e69SQu Wenruo if (ret < 0 && !err) 4294d2311e69SQu Wenruo err = ret; 42952ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 42963fd0a558SYan, Zheng btrfs_free_path(path); 42975d4f98a2SYan Zheng return err; 42985d4f98a2SYan Zheng } 42995d4f98a2SYan Zheng 43005d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 43010257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 43025d4f98a2SYan Zheng { 43035d4f98a2SYan Zheng struct btrfs_path *path; 43045d4f98a2SYan Zheng struct btrfs_inode_item *item; 43055d4f98a2SYan Zheng struct extent_buffer *leaf; 43065d4f98a2SYan Zheng int ret; 43075d4f98a2SYan Zheng 43085d4f98a2SYan Zheng path = btrfs_alloc_path(); 43095d4f98a2SYan Zheng if (!path) 43105d4f98a2SYan Zheng return -ENOMEM; 43115d4f98a2SYan Zheng 43125d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 43135d4f98a2SYan Zheng if (ret) 43145d4f98a2SYan Zheng goto out; 43155d4f98a2SYan Zheng 43165d4f98a2SYan Zheng leaf = path->nodes[0]; 43175d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4318b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 43195d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 43200257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 43215d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 43223fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 43233fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 43245d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 43255d4f98a2SYan Zheng out: 43265d4f98a2SYan Zheng btrfs_free_path(path); 43275d4f98a2SYan Zheng return ret; 43285d4f98a2SYan Zheng } 43295d4f98a2SYan Zheng 43305d4f98a2SYan Zheng /* 43315d4f98a2SYan Zheng * helper to create inode for data relocation. 43325d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 43335d4f98a2SYan Zheng */ 43343fd0a558SYan, Zheng static noinline_for_stack 43353fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 433632da5386SDavid Sterba struct btrfs_block_group *group) 43375d4f98a2SYan Zheng { 43385d4f98a2SYan Zheng struct inode *inode = NULL; 43395d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 43405d4f98a2SYan Zheng struct btrfs_root *root; 43415d4f98a2SYan Zheng struct btrfs_key key; 43424624900dSZhaolei u64 objectid; 43435d4f98a2SYan Zheng int err = 0; 43445d4f98a2SYan Zheng 43455d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 43465d4f98a2SYan Zheng if (IS_ERR(root)) 43475d4f98a2SYan Zheng return ERR_CAST(root); 43485d4f98a2SYan Zheng 4349a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 435076deacf0SJosef Bacik if (IS_ERR(trans)) { 435100246528SJosef Bacik btrfs_put_root(root); 43523fd0a558SYan, Zheng return ERR_CAST(trans); 435376deacf0SJosef Bacik } 43545d4f98a2SYan Zheng 4355581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 43565d4f98a2SYan Zheng if (err) 43575d4f98a2SYan Zheng goto out; 43585d4f98a2SYan Zheng 43590257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 43605d4f98a2SYan Zheng BUG_ON(err); 43615d4f98a2SYan Zheng 43625d4f98a2SYan Zheng key.objectid = objectid; 43635d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 43645d4f98a2SYan Zheng key.offset = 0; 43654c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 43662e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 4367b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 43685d4f98a2SYan Zheng 436973f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 43705d4f98a2SYan Zheng out: 437100246528SJosef Bacik btrfs_put_root(root); 43723a45bb20SJeff Mahoney btrfs_end_transaction(trans); 43732ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 43745d4f98a2SYan Zheng if (err) { 43755d4f98a2SYan Zheng if (inode) 43765d4f98a2SYan Zheng iput(inode); 43775d4f98a2SYan Zheng inode = ERR_PTR(err); 43785d4f98a2SYan Zheng } 43795d4f98a2SYan Zheng return inode; 43805d4f98a2SYan Zheng } 43815d4f98a2SYan Zheng 4382c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 43833fd0a558SYan, Zheng { 43843fd0a558SYan, Zheng struct reloc_control *rc; 43853fd0a558SYan, Zheng 43863fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 43873fd0a558SYan, Zheng if (!rc) 43883fd0a558SYan, Zheng return NULL; 43893fd0a558SYan, Zheng 43903fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4391d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 43922433bea5SQu Wenruo backref_cache_init(fs_info, &rc->backref_cache, 1); 43933fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 439443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 439543eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 43963fd0a558SYan, Zheng return rc; 43973fd0a558SYan, Zheng } 43983fd0a558SYan, Zheng 43991a0afa0eSJosef Bacik static void free_reloc_control(struct reloc_control *rc) 44001a0afa0eSJosef Bacik { 44011a0afa0eSJosef Bacik struct mapping_node *node, *tmp; 44021a0afa0eSJosef Bacik 44031a0afa0eSJosef Bacik free_reloc_roots(&rc->reloc_roots); 44041a0afa0eSJosef Bacik rbtree_postorder_for_each_entry_safe(node, tmp, 44051a0afa0eSJosef Bacik &rc->reloc_root_tree.rb_root, rb_node) 44061a0afa0eSJosef Bacik kfree(node); 44071a0afa0eSJosef Bacik 44081a0afa0eSJosef Bacik kfree(rc); 44091a0afa0eSJosef Bacik } 44101a0afa0eSJosef Bacik 44115d4f98a2SYan Zheng /* 4412ebce0e01SAdam Borowski * Print the block group being relocated 4413ebce0e01SAdam Borowski */ 4414ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 441532da5386SDavid Sterba struct btrfs_block_group *block_group) 4416ebce0e01SAdam Borowski { 4417f89e09cfSAnand Jain char buf[128] = {'\0'}; 4418ebce0e01SAdam Borowski 4419f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4420ebce0e01SAdam Borowski 4421ebce0e01SAdam Borowski btrfs_info(fs_info, 4422ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4423b3470b5dSDavid Sterba block_group->start, buf); 4424ebce0e01SAdam Borowski } 4425ebce0e01SAdam Borowski 4426430640e3SQu Wenruo static const char *stage_to_string(int stage) 4427430640e3SQu Wenruo { 4428430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 4429430640e3SQu Wenruo return "move data extents"; 4430430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 4431430640e3SQu Wenruo return "update data pointers"; 4432430640e3SQu Wenruo return "unknown"; 4433430640e3SQu Wenruo } 4434430640e3SQu Wenruo 4435ebce0e01SAdam Borowski /* 44365d4f98a2SYan Zheng * function to relocate all extents in a block group. 44375d4f98a2SYan Zheng */ 44386bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 44395d4f98a2SYan Zheng { 444032da5386SDavid Sterba struct btrfs_block_group *bg; 44416bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 44425d4f98a2SYan Zheng struct reloc_control *rc; 44430af3d00bSJosef Bacik struct inode *inode; 44440af3d00bSJosef Bacik struct btrfs_path *path; 44455d4f98a2SYan Zheng int ret; 4446f0486c68SYan, Zheng int rw = 0; 44475d4f98a2SYan Zheng int err = 0; 44485d4f98a2SYan Zheng 4449eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4450eede2bf3SOmar Sandoval if (!bg) 4451eede2bf3SOmar Sandoval return -ENOENT; 4452eede2bf3SOmar Sandoval 4453eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4454eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4455eede2bf3SOmar Sandoval return -ETXTBSY; 4456eede2bf3SOmar Sandoval } 4457eede2bf3SOmar Sandoval 4458c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4459eede2bf3SOmar Sandoval if (!rc) { 4460eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 44615d4f98a2SYan Zheng return -ENOMEM; 4462eede2bf3SOmar Sandoval } 44635d4f98a2SYan Zheng 4464f0486c68SYan, Zheng rc->extent_root = extent_root; 4465eede2bf3SOmar Sandoval rc->block_group = bg; 44665d4f98a2SYan Zheng 4467b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4468f0486c68SYan, Zheng if (ret) { 4469f0486c68SYan, Zheng err = ret; 4470f0486c68SYan, Zheng goto out; 4471f0486c68SYan, Zheng } 4472f0486c68SYan, Zheng rw = 1; 4473f0486c68SYan, Zheng 44740af3d00bSJosef Bacik path = btrfs_alloc_path(); 44750af3d00bSJosef Bacik if (!path) { 44760af3d00bSJosef Bacik err = -ENOMEM; 44770af3d00bSJosef Bacik goto out; 44780af3d00bSJosef Bacik } 44790af3d00bSJosef Bacik 44807949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 44810af3d00bSJosef Bacik btrfs_free_path(path); 44820af3d00bSJosef Bacik 44830af3d00bSJosef Bacik if (!IS_ERR(inode)) 44841bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 44850af3d00bSJosef Bacik else 44860af3d00bSJosef Bacik ret = PTR_ERR(inode); 44870af3d00bSJosef Bacik 44880af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 44890af3d00bSJosef Bacik err = ret; 44900af3d00bSJosef Bacik goto out; 44910af3d00bSJosef Bacik } 44920af3d00bSJosef Bacik 44935d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 44945d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 44955d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 44965d4f98a2SYan Zheng rc->data_inode = NULL; 44975d4f98a2SYan Zheng goto out; 44985d4f98a2SYan Zheng } 44995d4f98a2SYan Zheng 45000b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 45015d4f98a2SYan Zheng 45029cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4503f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 45046374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4505b3470b5dSDavid Sterba rc->block_group->start, 4506b3470b5dSDavid Sterba rc->block_group->length); 45075d4f98a2SYan Zheng 45085d4f98a2SYan Zheng while (1) { 4509430640e3SQu Wenruo int finishes_stage; 4510430640e3SQu Wenruo 451176dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 45125d4f98a2SYan Zheng ret = relocate_block_group(rc); 451376dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4514ff612ba7SJosef Bacik if (ret < 0) 45155d4f98a2SYan Zheng err = ret; 4516ff612ba7SJosef Bacik 4517430640e3SQu Wenruo finishes_stage = rc->stage; 4518ff612ba7SJosef Bacik /* 4519ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4520ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4521ff612ba7SJosef Bacik * different block group we could end up hitting the 4522ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4523ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4524ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4525ff612ba7SJosef Bacik * out of the loop if we hit an error. 4526ff612ba7SJosef Bacik */ 4527ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4528ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4529ff612ba7SJosef Bacik (u64)-1); 4530ff612ba7SJosef Bacik if (ret) 4531ff612ba7SJosef Bacik err = ret; 4532ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4533ff612ba7SJosef Bacik 0, -1); 4534ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 45355d4f98a2SYan Zheng } 45365d4f98a2SYan Zheng 4537ff612ba7SJosef Bacik if (err < 0) 4538ff612ba7SJosef Bacik goto out; 4539ff612ba7SJosef Bacik 45405d4f98a2SYan Zheng if (rc->extents_found == 0) 45415d4f98a2SYan Zheng break; 45425d4f98a2SYan Zheng 4543430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4544430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 45455d4f98a2SYan Zheng } 45465d4f98a2SYan Zheng 45475d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 45485d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4549bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 45505d4f98a2SYan Zheng out: 4551f0486c68SYan, Zheng if (err && rw) 45522ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 45535d4f98a2SYan Zheng iput(rc->data_inode); 45545d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 45551a0afa0eSJosef Bacik free_reloc_control(rc); 45565d4f98a2SYan Zheng return err; 45575d4f98a2SYan Zheng } 45585d4f98a2SYan Zheng 455976dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 456076dda93cSYan, Zheng { 45610b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 456276dda93cSYan, Zheng struct btrfs_trans_handle *trans; 456379787eaaSJeff Mahoney int ret, err; 456476dda93cSYan, Zheng 45650b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 456679787eaaSJeff Mahoney if (IS_ERR(trans)) 456779787eaaSJeff Mahoney return PTR_ERR(trans); 456876dda93cSYan, Zheng 456976dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 457076dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 457176dda93cSYan, Zheng root->root_item.drop_level = 0; 457276dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 45730b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 457476dda93cSYan, Zheng &root->root_key, &root->root_item); 457576dda93cSYan, Zheng 45763a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 457779787eaaSJeff Mahoney if (err) 457879787eaaSJeff Mahoney return err; 457979787eaaSJeff Mahoney return ret; 458076dda93cSYan, Zheng } 458176dda93cSYan, Zheng 45825d4f98a2SYan Zheng /* 45835d4f98a2SYan Zheng * recover relocation interrupted by system crash. 45845d4f98a2SYan Zheng * 45855d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 45865d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 45875d4f98a2SYan Zheng */ 45885d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 45895d4f98a2SYan Zheng { 45900b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 45915d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 45925d4f98a2SYan Zheng struct btrfs_key key; 45935d4f98a2SYan Zheng struct btrfs_root *fs_root; 45945d4f98a2SYan Zheng struct btrfs_root *reloc_root; 45955d4f98a2SYan Zheng struct btrfs_path *path; 45965d4f98a2SYan Zheng struct extent_buffer *leaf; 45975d4f98a2SYan Zheng struct reloc_control *rc = NULL; 45985d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 45995d4f98a2SYan Zheng int ret; 46005d4f98a2SYan Zheng int err = 0; 46015d4f98a2SYan Zheng 46025d4f98a2SYan Zheng path = btrfs_alloc_path(); 46035d4f98a2SYan Zheng if (!path) 46045d4f98a2SYan Zheng return -ENOMEM; 4605e4058b54SDavid Sterba path->reada = READA_BACK; 46065d4f98a2SYan Zheng 46075d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 46085d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 46095d4f98a2SYan Zheng key.offset = (u64)-1; 46105d4f98a2SYan Zheng 46115d4f98a2SYan Zheng while (1) { 46120b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 46135d4f98a2SYan Zheng path, 0, 0); 46145d4f98a2SYan Zheng if (ret < 0) { 46155d4f98a2SYan Zheng err = ret; 46165d4f98a2SYan Zheng goto out; 46175d4f98a2SYan Zheng } 46185d4f98a2SYan Zheng if (ret > 0) { 46195d4f98a2SYan Zheng if (path->slots[0] == 0) 46205d4f98a2SYan Zheng break; 46215d4f98a2SYan Zheng path->slots[0]--; 46225d4f98a2SYan Zheng } 46235d4f98a2SYan Zheng leaf = path->nodes[0]; 46245d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4625b3b4aa74SDavid Sterba btrfs_release_path(path); 46265d4f98a2SYan Zheng 46275d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 46285d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 46295d4f98a2SYan Zheng break; 46305d4f98a2SYan Zheng 46313dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(root, &key); 46325d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 46335d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 46345d4f98a2SYan Zheng goto out; 46355d4f98a2SYan Zheng } 46365d4f98a2SYan Zheng 46373dbf1738SJosef Bacik set_bit(BTRFS_ROOT_REF_COWS, &reloc_root->state); 46385d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 46395d4f98a2SYan Zheng 46405d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 46410b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 46425d4f98a2SYan Zheng reloc_root->root_key.offset); 46435d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 464476dda93cSYan, Zheng ret = PTR_ERR(fs_root); 464576dda93cSYan, Zheng if (ret != -ENOENT) { 464676dda93cSYan, Zheng err = ret; 46475d4f98a2SYan Zheng goto out; 46485d4f98a2SYan Zheng } 464979787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 465079787eaaSJeff Mahoney if (ret < 0) { 465179787eaaSJeff Mahoney err = ret; 465279787eaaSJeff Mahoney goto out; 465379787eaaSJeff Mahoney } 4654932fd26dSJosef Bacik } else { 465500246528SJosef Bacik btrfs_put_root(fs_root); 465676dda93cSYan, Zheng } 46575d4f98a2SYan Zheng } 46585d4f98a2SYan Zheng 46595d4f98a2SYan Zheng if (key.offset == 0) 46605d4f98a2SYan Zheng break; 46615d4f98a2SYan Zheng 46625d4f98a2SYan Zheng key.offset--; 46635d4f98a2SYan Zheng } 4664b3b4aa74SDavid Sterba btrfs_release_path(path); 46655d4f98a2SYan Zheng 46665d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 46675d4f98a2SYan Zheng goto out; 46685d4f98a2SYan Zheng 4669c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 46705d4f98a2SYan Zheng if (!rc) { 46715d4f98a2SYan Zheng err = -ENOMEM; 46725d4f98a2SYan Zheng goto out; 46735d4f98a2SYan Zheng } 46745d4f98a2SYan Zheng 46750b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 46765d4f98a2SYan Zheng 46775d4f98a2SYan Zheng set_reloc_control(rc); 46785d4f98a2SYan Zheng 46797a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 46803612b495STsutomu Itoh if (IS_ERR(trans)) { 46813612b495STsutomu Itoh err = PTR_ERR(trans); 4682fb2d83eeSJosef Bacik goto out_unset; 46833612b495STsutomu Itoh } 46843fd0a558SYan, Zheng 46853fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 46863fd0a558SYan, Zheng 46875d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 46885d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 46895d4f98a2SYan Zheng struct btrfs_root, root_list); 46905d4f98a2SYan Zheng list_del(&reloc_root->root_list); 46915d4f98a2SYan Zheng 46925d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 46935d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 46945d4f98a2SYan Zheng &rc->reloc_roots); 46955d4f98a2SYan Zheng continue; 46965d4f98a2SYan Zheng } 46975d4f98a2SYan Zheng 46980b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 469979787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 470079787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4701ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 47021402d17dSXiyu Yang btrfs_end_transaction(trans); 4703fb2d83eeSJosef Bacik goto out_unset; 470479787eaaSJeff Mahoney } 47055d4f98a2SYan Zheng 4706ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 470779787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 4708f44deb74SJosef Bacik fs_root->reloc_root = btrfs_grab_root(reloc_root); 470900246528SJosef Bacik btrfs_put_root(fs_root); 47105d4f98a2SYan Zheng } 47115d4f98a2SYan Zheng 47123a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 471379787eaaSJeff Mahoney if (err) 4714fb2d83eeSJosef Bacik goto out_unset; 47155d4f98a2SYan Zheng 47165d4f98a2SYan Zheng merge_reloc_roots(rc); 47175d4f98a2SYan Zheng 47185d4f98a2SYan Zheng unset_reloc_control(rc); 47195d4f98a2SYan Zheng 47207a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 472162b99540SQu Wenruo if (IS_ERR(trans)) { 47223612b495STsutomu Itoh err = PTR_ERR(trans); 47236217b0faSJosef Bacik goto out_clean; 472462b99540SQu Wenruo } 47253a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 47266217b0faSJosef Bacik out_clean: 4727d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4728d2311e69SQu Wenruo if (ret < 0 && !err) 4729d2311e69SQu Wenruo err = ret; 4730fb2d83eeSJosef Bacik out_unset: 4731fb2d83eeSJosef Bacik unset_reloc_control(rc); 47321a0afa0eSJosef Bacik free_reloc_control(rc); 47333612b495STsutomu Itoh out: 4734aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4735aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4736aca1bba6SLiu Bo 47375d4f98a2SYan Zheng btrfs_free_path(path); 47385d4f98a2SYan Zheng 47395d4f98a2SYan Zheng if (err == 0) { 47405d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 47410b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 4742932fd26dSJosef Bacik if (IS_ERR(fs_root)) { 47435d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4744932fd26dSJosef Bacik } else { 474566b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 474600246528SJosef Bacik btrfs_put_root(fs_root); 4747932fd26dSJosef Bacik } 4748932fd26dSJosef Bacik } 47495d4f98a2SYan Zheng return err; 47505d4f98a2SYan Zheng } 47515d4f98a2SYan Zheng 47525d4f98a2SYan Zheng /* 47535d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 47545d4f98a2SYan Zheng * 47555d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 47565d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 47575d4f98a2SYan Zheng */ 47585d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 47595d4f98a2SYan Zheng { 47600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 47615d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 47625d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 47635d4f98a2SYan Zheng int ret; 47645d4f98a2SYan Zheng u64 disk_bytenr; 47654577b014SJosef Bacik u64 new_bytenr; 47665d4f98a2SYan Zheng LIST_HEAD(list); 47675d4f98a2SYan Zheng 47685d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4769bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 47705d4f98a2SYan Zheng 47715d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 47720b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4773a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 477479787eaaSJeff Mahoney if (ret) 477579787eaaSJeff Mahoney goto out; 47765d4f98a2SYan Zheng 47775d4f98a2SYan Zheng while (!list_empty(&list)) { 47785d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 47795d4f98a2SYan Zheng list_del_init(&sums->list); 47805d4f98a2SYan Zheng 47814577b014SJosef Bacik /* 47824577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 47834577b014SJosef Bacik * We need to do this because we will read in entire prealloc 47844577b014SJosef Bacik * extents but we may have written to say the middle of the 47854577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 47864577b014SJosef Bacik * the right disk offset. 47874577b014SJosef Bacik * 47884577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 47894577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 47904577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 47914577b014SJosef Bacik * disk length. 47924577b014SJosef Bacik */ 4793bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 47944577b014SJosef Bacik sums->bytenr = new_bytenr; 47955d4f98a2SYan Zheng 4796f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 47975d4f98a2SYan Zheng } 479879787eaaSJeff Mahoney out: 47995d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4800411fc6bcSAndi Kleen return ret; 48015d4f98a2SYan Zheng } 48023fd0a558SYan, Zheng 480383d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 48043fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 48053fd0a558SYan, Zheng struct extent_buffer *cow) 48063fd0a558SYan, Zheng { 48070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 48083fd0a558SYan, Zheng struct reloc_control *rc; 4809a26195a5SQu Wenruo struct btrfs_backref_node *node; 48103fd0a558SYan, Zheng int first_cow = 0; 48113fd0a558SYan, Zheng int level; 481283d4cfd4SJosef Bacik int ret = 0; 48133fd0a558SYan, Zheng 48140b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 48153fd0a558SYan, Zheng if (!rc) 481683d4cfd4SJosef Bacik return 0; 48173fd0a558SYan, Zheng 48183fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 48193fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 48203fd0a558SYan, Zheng 48213fd0a558SYan, Zheng level = btrfs_header_level(buf); 48223fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 48233fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 48243fd0a558SYan, Zheng first_cow = 1; 48253fd0a558SYan, Zheng 48263fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 48273fd0a558SYan, Zheng rc->create_reloc_tree) { 48283fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 48293fd0a558SYan, Zheng 48303fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 48313fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 48323fd0a558SYan, Zheng node->new_bytenr != buf->start); 48333fd0a558SYan, Zheng 48343fd0a558SYan, Zheng drop_node_buffer(node); 483567439dadSDavid Sterba atomic_inc(&cow->refs); 48363fd0a558SYan, Zheng node->eb = cow; 48373fd0a558SYan, Zheng node->new_bytenr = cow->start; 48383fd0a558SYan, Zheng 48393fd0a558SYan, Zheng if (!node->pending) { 48403fd0a558SYan, Zheng list_move_tail(&node->list, 48413fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 48423fd0a558SYan, Zheng node->pending = 1; 48433fd0a558SYan, Zheng } 48443fd0a558SYan, Zheng 48453fd0a558SYan, Zheng if (first_cow) 48469569cc20SQu Wenruo mark_block_processed(rc, node); 48473fd0a558SYan, Zheng 48483fd0a558SYan, Zheng if (first_cow && level > 0) 48493fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 48503fd0a558SYan, Zheng } 48513fd0a558SYan, Zheng 485283d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 48533fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 485483d4cfd4SJosef Bacik return ret; 48553fd0a558SYan, Zheng } 48563fd0a558SYan, Zheng 48573fd0a558SYan, Zheng /* 48583fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 485901327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 48603fd0a558SYan, Zheng */ 4861147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 48623fd0a558SYan, Zheng u64 *bytes_to_reserve) 48633fd0a558SYan, Zheng { 486410995c04SQu Wenruo struct btrfs_root *root = pending->root; 486510995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 48663fd0a558SYan, Zheng 48676282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 48683fd0a558SYan, Zheng return; 48693fd0a558SYan, Zheng 48703fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 48713fd0a558SYan, Zheng return; 48723fd0a558SYan, Zheng 48733fd0a558SYan, Zheng root = root->reloc_root; 48743fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 48753fd0a558SYan, Zheng /* 48763fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 48773fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 48783fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 48793fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 48803fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 48813fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 48823fd0a558SYan, Zheng * tree may use more space than it frees. so we need 48833fd0a558SYan, Zheng * reserve extra space. 48843fd0a558SYan, Zheng */ 48853fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 48863fd0a558SYan, Zheng } 48873fd0a558SYan, Zheng 48883fd0a558SYan, Zheng /* 48893fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 48903fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 4891f44deb74SJosef Bacik * 4892f44deb74SJosef Bacik * This is similar to btrfs_init_reloc_root(), we come out of here with two 4893f44deb74SJosef Bacik * references held on the reloc_root, one for root->reloc_root and one for 4894f44deb74SJosef Bacik * rc->reloc_roots. 48953fd0a558SYan, Zheng */ 489649b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 48973fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 48983fd0a558SYan, Zheng { 48993fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 49003fd0a558SYan, Zheng struct btrfs_root *reloc_root; 49013fd0a558SYan, Zheng struct btrfs_root *new_root; 490210995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 49033fd0a558SYan, Zheng int ret; 49043fd0a558SYan, Zheng 49056282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 490649b25e05SJeff Mahoney return 0; 49073fd0a558SYan, Zheng 49083fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 49093fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 49103fd0a558SYan, Zheng 49113fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 49123fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 49133fd0a558SYan, Zheng rc->block_rsv, 49143a584174SLu Fengqi rc->nodes_relocated, true); 491549b25e05SJeff Mahoney if (ret) 491649b25e05SJeff Mahoney return ret; 49173fd0a558SYan, Zheng } 49183fd0a558SYan, Zheng 49193fd0a558SYan, Zheng new_root = pending->snap; 49203fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 49213fd0a558SYan, Zheng new_root->root_key.objectid); 492249b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 492349b25e05SJeff Mahoney return PTR_ERR(reloc_root); 49243fd0a558SYan, Zheng 4925ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4926ffd7b339SJeff Mahoney BUG_ON(ret < 0); 4927f44deb74SJosef Bacik new_root->reloc_root = btrfs_grab_root(reloc_root); 49283fd0a558SYan, Zheng 492949b25e05SJeff Mahoney if (rc->create_reloc_tree) 49303fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 493149b25e05SJeff Mahoney return ret; 49323fd0a558SYan, Zheng } 4933