1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 3d352ac68SChris Mason * Copyright (C) 2007,2008 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 6a6b6e75eSChris Mason #include <linux/sched.h> 75a0e3ad6STejun Heo #include <linux/slab.h> 8bd989ba3SJan Schmidt #include <linux/rbtree.h> 9adf02123SDavid Sterba #include <linux/mm.h> 10e41d12f5SChristoph Hellwig #include <linux/error-injection.h> 119b569ea0SJosef Bacik #include "messages.h" 12eb60ceacSChris Mason #include "ctree.h" 13eb60ceacSChris Mason #include "disk-io.h" 147f5c1516SChris Mason #include "transaction.h" 155f39d397SChris Mason #include "print-tree.h" 16925baeddSChris Mason #include "locking.h" 17de37aa51SNikolay Borisov #include "volumes.h" 18f616f5cdSQu Wenruo #include "qgroup.h" 19f3a84ccdSFilipe Manana #include "tree-mod-log.h" 2088c602abSQu Wenruo #include "tree-checker.h" 21ec8eb376SJosef Bacik #include "fs.h" 22*ad1ac501SJosef Bacik #include "accessors.h" 239a8dd150SChris Mason 24226463d7SJosef Bacik static struct kmem_cache *btrfs_path_cachep; 25226463d7SJosef Bacik 26e089f05cSChris Mason static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root 27e089f05cSChris Mason *root, struct btrfs_path *path, int level); 28310712b2SOmar Sandoval static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, 29310712b2SOmar Sandoval const struct btrfs_key *ins_key, struct btrfs_path *path, 30310712b2SOmar Sandoval int data_size, int extend); 315f39d397SChris Mason static int push_node_left(struct btrfs_trans_handle *trans, 322ff7e61eSJeff Mahoney struct extent_buffer *dst, 33971a1f66SChris Mason struct extent_buffer *src, int empty); 345f39d397SChris Mason static int balance_node_right(struct btrfs_trans_handle *trans, 355f39d397SChris Mason struct extent_buffer *dst_buf, 365f39d397SChris Mason struct extent_buffer *src_buf); 37afe5fea7STsutomu Itoh static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, 38afe5fea7STsutomu Itoh int level, int slot); 39d97e63b6SChris Mason 40af024ed2SJohannes Thumshirn static const struct btrfs_csums { 41af024ed2SJohannes Thumshirn u16 size; 4259a0fcdbSDavid Sterba const char name[10]; 4359a0fcdbSDavid Sterba const char driver[12]; 44af024ed2SJohannes Thumshirn } btrfs_csums[] = { 45af024ed2SJohannes Thumshirn [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, 463951e7f0SJohannes Thumshirn [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" }, 473831bf00SJohannes Thumshirn [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" }, 48352ae07bSDavid Sterba [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b", 49352ae07bSDavid Sterba .driver = "blake2b-256" }, 50af024ed2SJohannes Thumshirn }; 51af024ed2SJohannes Thumshirn 52af024ed2SJohannes Thumshirn int btrfs_super_csum_size(const struct btrfs_super_block *s) 53af024ed2SJohannes Thumshirn { 54af024ed2SJohannes Thumshirn u16 t = btrfs_super_csum_type(s); 55af024ed2SJohannes Thumshirn /* 56af024ed2SJohannes Thumshirn * csum type is validated at mount time 57af024ed2SJohannes Thumshirn */ 58af024ed2SJohannes Thumshirn return btrfs_csums[t].size; 59af024ed2SJohannes Thumshirn } 60af024ed2SJohannes Thumshirn 61af024ed2SJohannes Thumshirn const char *btrfs_super_csum_name(u16 csum_type) 62af024ed2SJohannes Thumshirn { 63af024ed2SJohannes Thumshirn /* csum type is validated at mount time */ 64af024ed2SJohannes Thumshirn return btrfs_csums[csum_type].name; 65af024ed2SJohannes Thumshirn } 66af024ed2SJohannes Thumshirn 67b4e967beSDavid Sterba /* 68b4e967beSDavid Sterba * Return driver name if defined, otherwise the name that's also a valid driver 69b4e967beSDavid Sterba * name 70b4e967beSDavid Sterba */ 71b4e967beSDavid Sterba const char *btrfs_super_csum_driver(u16 csum_type) 72b4e967beSDavid Sterba { 73b4e967beSDavid Sterba /* csum type is validated at mount time */ 7459a0fcdbSDavid Sterba return btrfs_csums[csum_type].driver[0] ? 7559a0fcdbSDavid Sterba btrfs_csums[csum_type].driver : 76b4e967beSDavid Sterba btrfs_csums[csum_type].name; 77b4e967beSDavid Sterba } 78b4e967beSDavid Sterba 79604997b4SDavid Sterba size_t __attribute_const__ btrfs_get_num_csums(void) 80f7cea56cSDavid Sterba { 81f7cea56cSDavid Sterba return ARRAY_SIZE(btrfs_csums); 82f7cea56cSDavid Sterba } 83f7cea56cSDavid Sterba 842c90e5d6SChris Mason struct btrfs_path *btrfs_alloc_path(void) 852c90e5d6SChris Mason { 86e2c89907SMasahiro Yamada return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); 872c90e5d6SChris Mason } 882c90e5d6SChris Mason 89d352ac68SChris Mason /* this also releases the path */ 902c90e5d6SChris Mason void btrfs_free_path(struct btrfs_path *p) 912c90e5d6SChris Mason { 92ff175d57SJesper Juhl if (!p) 93ff175d57SJesper Juhl return; 94b3b4aa74SDavid Sterba btrfs_release_path(p); 952c90e5d6SChris Mason kmem_cache_free(btrfs_path_cachep, p); 962c90e5d6SChris Mason } 972c90e5d6SChris Mason 98d352ac68SChris Mason /* 99d352ac68SChris Mason * path release drops references on the extent buffers in the path 100d352ac68SChris Mason * and it drops any locks held by this path 101d352ac68SChris Mason * 102d352ac68SChris Mason * It is safe to call this on paths that no locks or extent buffers held. 103d352ac68SChris Mason */ 104b3b4aa74SDavid Sterba noinline void btrfs_release_path(struct btrfs_path *p) 105eb60ceacSChris Mason { 106eb60ceacSChris Mason int i; 107a2135011SChris Mason 108234b63a0SChris Mason for (i = 0; i < BTRFS_MAX_LEVEL; i++) { 1093f157a2fSChris Mason p->slots[i] = 0; 110eb60ceacSChris Mason if (!p->nodes[i]) 111925baeddSChris Mason continue; 112925baeddSChris Mason if (p->locks[i]) { 113bd681513SChris Mason btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); 114925baeddSChris Mason p->locks[i] = 0; 115925baeddSChris Mason } 1165f39d397SChris Mason free_extent_buffer(p->nodes[i]); 1173f157a2fSChris Mason p->nodes[i] = NULL; 118eb60ceacSChris Mason } 119eb60ceacSChris Mason } 120eb60ceacSChris Mason 121d352ac68SChris Mason /* 1228bb808c6SDavid Sterba * We want the transaction abort to print stack trace only for errors where the 1238bb808c6SDavid Sterba * cause could be a bug, eg. due to ENOSPC, and not for common errors that are 1248bb808c6SDavid Sterba * caused by external factors. 1258bb808c6SDavid Sterba */ 1268bb808c6SDavid Sterba bool __cold abort_should_print_stack(int errno) 1278bb808c6SDavid Sterba { 1288bb808c6SDavid Sterba switch (errno) { 1298bb808c6SDavid Sterba case -EIO: 1308bb808c6SDavid Sterba case -EROFS: 1318bb808c6SDavid Sterba case -ENOMEM: 1328bb808c6SDavid Sterba return false; 1338bb808c6SDavid Sterba } 1348bb808c6SDavid Sterba return true; 1358bb808c6SDavid Sterba } 1368bb808c6SDavid Sterba 1378bb808c6SDavid Sterba /* 138d352ac68SChris Mason * safely gets a reference on the root node of a tree. A lock 139d352ac68SChris Mason * is not taken, so a concurrent writer may put a different node 140d352ac68SChris Mason * at the root of the tree. See btrfs_lock_root_node for the 141d352ac68SChris Mason * looping required. 142d352ac68SChris Mason * 143d352ac68SChris Mason * The extent buffer returned by this has a reference taken, so 144d352ac68SChris Mason * it won't disappear. It may stop being the root of the tree 145d352ac68SChris Mason * at any time because there are no locks held. 146d352ac68SChris Mason */ 147925baeddSChris Mason struct extent_buffer *btrfs_root_node(struct btrfs_root *root) 148925baeddSChris Mason { 149925baeddSChris Mason struct extent_buffer *eb; 150240f62c8SChris Mason 1513083ee2eSJosef Bacik while (1) { 152240f62c8SChris Mason rcu_read_lock(); 153240f62c8SChris Mason eb = rcu_dereference(root->node); 1543083ee2eSJosef Bacik 1553083ee2eSJosef Bacik /* 1563083ee2eSJosef Bacik * RCU really hurts here, we could free up the root node because 15701327610SNicholas D Steeves * it was COWed but we may not get the new root node yet so do 1583083ee2eSJosef Bacik * the inc_not_zero dance and if it doesn't work then 1593083ee2eSJosef Bacik * synchronize_rcu and try again. 1603083ee2eSJosef Bacik */ 1613083ee2eSJosef Bacik if (atomic_inc_not_zero(&eb->refs)) { 162240f62c8SChris Mason rcu_read_unlock(); 1633083ee2eSJosef Bacik break; 1643083ee2eSJosef Bacik } 1653083ee2eSJosef Bacik rcu_read_unlock(); 1663083ee2eSJosef Bacik synchronize_rcu(); 1673083ee2eSJosef Bacik } 168925baeddSChris Mason return eb; 169925baeddSChris Mason } 170925baeddSChris Mason 17192a7cc42SQu Wenruo /* 17292a7cc42SQu Wenruo * Cowonly root (not-shareable trees, everything not subvolume or reloc roots), 17392a7cc42SQu Wenruo * just get put onto a simple dirty list. Transaction walks this list to make 17492a7cc42SQu Wenruo * sure they get properly updated on disk. 175d352ac68SChris Mason */ 1760b86a832SChris Mason static void add_root_to_dirty_list(struct btrfs_root *root) 1770b86a832SChris Mason { 1780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1790b246afaSJeff Mahoney 180e7070be1SJosef Bacik if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || 181e7070be1SJosef Bacik !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) 182e7070be1SJosef Bacik return; 183e7070be1SJosef Bacik 1840b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 185e7070be1SJosef Bacik if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { 186e7070be1SJosef Bacik /* Want the extent tree to be the last on the list */ 1874fd786e6SMisono Tomohiro if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) 188e7070be1SJosef Bacik list_move_tail(&root->dirty_list, 1890b246afaSJeff Mahoney &fs_info->dirty_cowonly_roots); 190e7070be1SJosef Bacik else 191e7070be1SJosef Bacik list_move(&root->dirty_list, 1920b246afaSJeff Mahoney &fs_info->dirty_cowonly_roots); 1930b86a832SChris Mason } 1940b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1950b86a832SChris Mason } 1960b86a832SChris Mason 197d352ac68SChris Mason /* 198d352ac68SChris Mason * used by snapshot creation to make a copy of a root for a tree with 199d352ac68SChris Mason * a given objectid. The buffer with the new root node is returned in 200d352ac68SChris Mason * cow_ret, and this func returns zero on success or a negative error code. 201d352ac68SChris Mason */ 202be20aa9dSChris Mason int btrfs_copy_root(struct btrfs_trans_handle *trans, 203be20aa9dSChris Mason struct btrfs_root *root, 204be20aa9dSChris Mason struct extent_buffer *buf, 205be20aa9dSChris Mason struct extent_buffer **cow_ret, u64 new_root_objectid) 206be20aa9dSChris Mason { 2070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 208be20aa9dSChris Mason struct extent_buffer *cow; 209be20aa9dSChris Mason int ret = 0; 210be20aa9dSChris Mason int level; 2115d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 212be20aa9dSChris Mason 21392a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 2140b246afaSJeff Mahoney trans->transid != fs_info->running_transaction->transid); 21592a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 21627cdeb70SMiao Xie trans->transid != root->last_trans); 217be20aa9dSChris Mason 218be20aa9dSChris Mason level = btrfs_header_level(buf); 2195d4f98a2SYan Zheng if (level == 0) 2205d4f98a2SYan Zheng btrfs_item_key(buf, &disk_key, 0); 2215d4f98a2SYan Zheng else 2225d4f98a2SYan Zheng btrfs_node_key(buf, &disk_key, 0); 22331840ae1SZheng Yan 2244d75f8a9SDavid Sterba cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, 225cf6f34aaSJosef Bacik &disk_key, level, buf->start, 0, 226cf6f34aaSJosef Bacik BTRFS_NESTING_NEW_ROOT); 2275d4f98a2SYan Zheng if (IS_ERR(cow)) 228be20aa9dSChris Mason return PTR_ERR(cow); 229be20aa9dSChris Mason 23058e8012cSDavid Sterba copy_extent_buffer_full(cow, buf); 231be20aa9dSChris Mason btrfs_set_header_bytenr(cow, cow->start); 232be20aa9dSChris Mason btrfs_set_header_generation(cow, trans->transid); 2335d4f98a2SYan Zheng btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 2345d4f98a2SYan Zheng btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 2355d4f98a2SYan Zheng BTRFS_HEADER_FLAG_RELOC); 2365d4f98a2SYan Zheng if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 2375d4f98a2SYan Zheng btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 2385d4f98a2SYan Zheng else 239be20aa9dSChris Mason btrfs_set_header_owner(cow, new_root_objectid); 240be20aa9dSChris Mason 241de37aa51SNikolay Borisov write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 2422b82032cSYan Zheng 243be20aa9dSChris Mason WARN_ON(btrfs_header_generation(buf) > trans->transid); 2445d4f98a2SYan Zheng if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 245e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 2465d4f98a2SYan Zheng else 247e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 248867ed321SJosef Bacik if (ret) { 24972c9925fSFilipe Manana btrfs_tree_unlock(cow); 25072c9925fSFilipe Manana free_extent_buffer(cow); 251867ed321SJosef Bacik btrfs_abort_transaction(trans, ret); 252be20aa9dSChris Mason return ret; 253867ed321SJosef Bacik } 254be20aa9dSChris Mason 255be20aa9dSChris Mason btrfs_mark_buffer_dirty(cow); 256be20aa9dSChris Mason *cow_ret = cow; 257be20aa9dSChris Mason return 0; 258be20aa9dSChris Mason } 259be20aa9dSChris Mason 260d352ac68SChris Mason /* 2615d4f98a2SYan Zheng * check if the tree block can be shared by multiple trees 2625d4f98a2SYan Zheng */ 2635d4f98a2SYan Zheng int btrfs_block_can_be_shared(struct btrfs_root *root, 2645d4f98a2SYan Zheng struct extent_buffer *buf) 2655d4f98a2SYan Zheng { 2665d4f98a2SYan Zheng /* 26792a7cc42SQu Wenruo * Tree blocks not in shareable trees and tree roots are never shared. 26892a7cc42SQu Wenruo * If a block was allocated after the last snapshot and the block was 26992a7cc42SQu Wenruo * not allocated by tree relocation, we know the block is not shared. 2705d4f98a2SYan Zheng */ 27192a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 2725d4f98a2SYan Zheng buf != root->node && buf != root->commit_root && 2735d4f98a2SYan Zheng (btrfs_header_generation(buf) <= 2745d4f98a2SYan Zheng btrfs_root_last_snapshot(&root->root_item) || 2755d4f98a2SYan Zheng btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) 2765d4f98a2SYan Zheng return 1; 277a79865c6SNikolay Borisov 2785d4f98a2SYan Zheng return 0; 2795d4f98a2SYan Zheng } 2805d4f98a2SYan Zheng 2815d4f98a2SYan Zheng static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, 2825d4f98a2SYan Zheng struct btrfs_root *root, 2835d4f98a2SYan Zheng struct extent_buffer *buf, 284f0486c68SYan, Zheng struct extent_buffer *cow, 285f0486c68SYan, Zheng int *last_ref) 2865d4f98a2SYan Zheng { 2870b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 2885d4f98a2SYan Zheng u64 refs; 2895d4f98a2SYan Zheng u64 owner; 2905d4f98a2SYan Zheng u64 flags; 2915d4f98a2SYan Zheng u64 new_flags = 0; 2925d4f98a2SYan Zheng int ret; 2935d4f98a2SYan Zheng 2945d4f98a2SYan Zheng /* 2955d4f98a2SYan Zheng * Backrefs update rules: 2965d4f98a2SYan Zheng * 2975d4f98a2SYan Zheng * Always use full backrefs for extent pointers in tree block 2985d4f98a2SYan Zheng * allocated by tree relocation. 2995d4f98a2SYan Zheng * 3005d4f98a2SYan Zheng * If a shared tree block is no longer referenced by its owner 3015d4f98a2SYan Zheng * tree (btrfs_header_owner(buf) == root->root_key.objectid), 3025d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 3035d4f98a2SYan Zheng * 3045d4f98a2SYan Zheng * If a tree block is been relocating 3055d4f98a2SYan Zheng * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), 3065d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 3075d4f98a2SYan Zheng * The reason for this is some operations (such as drop tree) 3085d4f98a2SYan Zheng * are only allowed for blocks use full backrefs. 3095d4f98a2SYan Zheng */ 3105d4f98a2SYan Zheng 3115d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, buf)) { 3122ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, 3133173a18fSJosef Bacik btrfs_header_level(buf), 1, 3143173a18fSJosef Bacik &refs, &flags); 315be1a5564SMark Fasheh if (ret) 316be1a5564SMark Fasheh return ret; 317e5df9573SMark Fasheh if (refs == 0) { 318e5df9573SMark Fasheh ret = -EROFS; 3190b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 320e5df9573SMark Fasheh return ret; 321e5df9573SMark Fasheh } 3225d4f98a2SYan Zheng } else { 3235d4f98a2SYan Zheng refs = 1; 3245d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 3255d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 3265d4f98a2SYan Zheng flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; 3275d4f98a2SYan Zheng else 3285d4f98a2SYan Zheng flags = 0; 3295d4f98a2SYan Zheng } 3305d4f98a2SYan Zheng 3315d4f98a2SYan Zheng owner = btrfs_header_owner(buf); 3325d4f98a2SYan Zheng BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && 3335d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); 3345d4f98a2SYan Zheng 3355d4f98a2SYan Zheng if (refs > 1) { 3365d4f98a2SYan Zheng if ((owner == root->root_key.objectid || 3375d4f98a2SYan Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && 3385d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { 339e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, buf, 1); 340692826b2SJeff Mahoney if (ret) 341692826b2SJeff Mahoney return ret; 3425d4f98a2SYan Zheng 3435d4f98a2SYan Zheng if (root->root_key.objectid == 3445d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) { 345e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 0); 346692826b2SJeff Mahoney if (ret) 347692826b2SJeff Mahoney return ret; 348e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 349692826b2SJeff Mahoney if (ret) 350692826b2SJeff Mahoney return ret; 3515d4f98a2SYan Zheng } 3525d4f98a2SYan Zheng new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; 3535d4f98a2SYan Zheng } else { 3545d4f98a2SYan Zheng 3555d4f98a2SYan Zheng if (root->root_key.objectid == 3565d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 357e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 3585d4f98a2SYan Zheng else 359e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 360692826b2SJeff Mahoney if (ret) 361692826b2SJeff Mahoney return ret; 3625d4f98a2SYan Zheng } 3635d4f98a2SYan Zheng if (new_flags != 0) { 364b1c79e09SJosef Bacik int level = btrfs_header_level(buf); 365b1c79e09SJosef Bacik 36642c9d0b5SDavid Sterba ret = btrfs_set_disk_extent_flags(trans, buf, 3672fe6a5a1SDavid Sterba new_flags, level); 368be1a5564SMark Fasheh if (ret) 369be1a5564SMark Fasheh return ret; 3705d4f98a2SYan Zheng } 3715d4f98a2SYan Zheng } else { 3725d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { 3735d4f98a2SYan Zheng if (root->root_key.objectid == 3745d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 375e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 3765d4f98a2SYan Zheng else 377e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 378692826b2SJeff Mahoney if (ret) 379692826b2SJeff Mahoney return ret; 380e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 1); 381692826b2SJeff Mahoney if (ret) 382692826b2SJeff Mahoney return ret; 3835d4f98a2SYan Zheng } 3846a884d7dSDavid Sterba btrfs_clean_tree_block(buf); 385f0486c68SYan, Zheng *last_ref = 1; 3865d4f98a2SYan Zheng } 3875d4f98a2SYan Zheng return 0; 3885d4f98a2SYan Zheng } 3895d4f98a2SYan Zheng 3905d4f98a2SYan Zheng /* 391d397712bSChris Mason * does the dirty work in cow of a single block. The parent block (if 392d397712bSChris Mason * supplied) is updated to point to the new cow copy. The new buffer is marked 393d397712bSChris Mason * dirty and returned locked. If you modify the block it needs to be marked 394d397712bSChris Mason * dirty again. 395d352ac68SChris Mason * 396d352ac68SChris Mason * search_start -- an allocation hint for the new block 397d352ac68SChris Mason * 398d397712bSChris Mason * empty_size -- a hint that you plan on doing more cow. This is the size in 399d397712bSChris Mason * bytes the allocator should try to find free next to the block it returns. 400d397712bSChris Mason * This is just a hint and may be ignored by the allocator. 401d352ac68SChris Mason */ 402d397712bSChris Mason static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, 4035f39d397SChris Mason struct btrfs_root *root, 4045f39d397SChris Mason struct extent_buffer *buf, 4055f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 4065f39d397SChris Mason struct extent_buffer **cow_ret, 4079631e4ccSJosef Bacik u64 search_start, u64 empty_size, 4089631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 4096702ed49SChris Mason { 4100b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4115d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 4125f39d397SChris Mason struct extent_buffer *cow; 413be1a5564SMark Fasheh int level, ret; 414f0486c68SYan, Zheng int last_ref = 0; 415925baeddSChris Mason int unlock_orig = 0; 4160f5053ebSGoldwyn Rodrigues u64 parent_start = 0; 4176702ed49SChris Mason 418925baeddSChris Mason if (*cow_ret == buf) 419925baeddSChris Mason unlock_orig = 1; 420925baeddSChris Mason 42149d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 422925baeddSChris Mason 42392a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4240b246afaSJeff Mahoney trans->transid != fs_info->running_transaction->transid); 42592a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 42627cdeb70SMiao Xie trans->transid != root->last_trans); 4275f39d397SChris Mason 4287bb86316SChris Mason level = btrfs_header_level(buf); 42931840ae1SZheng Yan 4305d4f98a2SYan Zheng if (level == 0) 4315d4f98a2SYan Zheng btrfs_item_key(buf, &disk_key, 0); 4325d4f98a2SYan Zheng else 4335d4f98a2SYan Zheng btrfs_node_key(buf, &disk_key, 0); 4345d4f98a2SYan Zheng 4350f5053ebSGoldwyn Rodrigues if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) 4365d4f98a2SYan Zheng parent_start = parent->start; 4375d4f98a2SYan Zheng 43879bd3712SFilipe Manana cow = btrfs_alloc_tree_block(trans, root, parent_start, 43979bd3712SFilipe Manana root->root_key.objectid, &disk_key, level, 44079bd3712SFilipe Manana search_start, empty_size, nest); 4416702ed49SChris Mason if (IS_ERR(cow)) 4426702ed49SChris Mason return PTR_ERR(cow); 4436702ed49SChris Mason 444b4ce94deSChris Mason /* cow is set to blocking by btrfs_init_new_buffer */ 445b4ce94deSChris Mason 44658e8012cSDavid Sterba copy_extent_buffer_full(cow, buf); 447db94535dSChris Mason btrfs_set_header_bytenr(cow, cow->start); 4485f39d397SChris Mason btrfs_set_header_generation(cow, trans->transid); 4495d4f98a2SYan Zheng btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 4505d4f98a2SYan Zheng btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 4515d4f98a2SYan Zheng BTRFS_HEADER_FLAG_RELOC); 4525d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 4535d4f98a2SYan Zheng btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 4545d4f98a2SYan Zheng else 4555f39d397SChris Mason btrfs_set_header_owner(cow, root->root_key.objectid); 4566702ed49SChris Mason 457de37aa51SNikolay Borisov write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 4582b82032cSYan Zheng 459be1a5564SMark Fasheh ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); 460b68dc2a9SMark Fasheh if (ret) { 461572c83acSJosef Bacik btrfs_tree_unlock(cow); 462572c83acSJosef Bacik free_extent_buffer(cow); 46366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 464b68dc2a9SMark Fasheh return ret; 465b68dc2a9SMark Fasheh } 4661a40e23bSZheng Yan 46792a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 46883d4cfd4SJosef Bacik ret = btrfs_reloc_cow_block(trans, root, buf, cow); 46993314e3bSZhaolei if (ret) { 470572c83acSJosef Bacik btrfs_tree_unlock(cow); 471572c83acSJosef Bacik free_extent_buffer(cow); 47266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 47383d4cfd4SJosef Bacik return ret; 47483d4cfd4SJosef Bacik } 47593314e3bSZhaolei } 4763fd0a558SYan, Zheng 4776702ed49SChris Mason if (buf == root->node) { 478925baeddSChris Mason WARN_ON(parent && parent != buf); 4795d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 4805d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 4815d4f98a2SYan Zheng parent_start = buf->start; 482925baeddSChris Mason 48367439dadSDavid Sterba atomic_inc(&cow->refs); 484406808abSFilipe Manana ret = btrfs_tree_mod_log_insert_root(root->node, cow, true); 485d9d19a01SDavid Sterba BUG_ON(ret < 0); 486240f62c8SChris Mason rcu_assign_pointer(root->node, cow); 487925baeddSChris Mason 4887a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), buf, 4897a163608SFilipe Manana parent_start, last_ref); 4905f39d397SChris Mason free_extent_buffer(buf); 4910b86a832SChris Mason add_root_to_dirty_list(root); 4926702ed49SChris Mason } else { 4935d4f98a2SYan Zheng WARN_ON(trans->transid != btrfs_header_generation(parent)); 494f3a84ccdSFilipe Manana btrfs_tree_mod_log_insert_key(parent, parent_slot, 49533cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 4965f39d397SChris Mason btrfs_set_node_blockptr(parent, parent_slot, 497db94535dSChris Mason cow->start); 49874493f7aSChris Mason btrfs_set_node_ptr_generation(parent, parent_slot, 49974493f7aSChris Mason trans->transid); 5006702ed49SChris Mason btrfs_mark_buffer_dirty(parent); 5015de865eeSFilipe David Borba Manana if (last_ref) { 502f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_free_eb(buf); 5035de865eeSFilipe David Borba Manana if (ret) { 504572c83acSJosef Bacik btrfs_tree_unlock(cow); 505572c83acSJosef Bacik free_extent_buffer(cow); 50666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 5075de865eeSFilipe David Borba Manana return ret; 5085de865eeSFilipe David Borba Manana } 5095de865eeSFilipe David Borba Manana } 5107a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), buf, 5117a163608SFilipe Manana parent_start, last_ref); 5126702ed49SChris Mason } 513925baeddSChris Mason if (unlock_orig) 514925baeddSChris Mason btrfs_tree_unlock(buf); 5153083ee2eSJosef Bacik free_extent_buffer_stale(buf); 5166702ed49SChris Mason btrfs_mark_buffer_dirty(cow); 5176702ed49SChris Mason *cow_ret = cow; 5186702ed49SChris Mason return 0; 5196702ed49SChris Mason } 5206702ed49SChris Mason 5215d4f98a2SYan Zheng static inline int should_cow_block(struct btrfs_trans_handle *trans, 5225d4f98a2SYan Zheng struct btrfs_root *root, 5235d4f98a2SYan Zheng struct extent_buffer *buf) 5245d4f98a2SYan Zheng { 525f5ee5c9aSJeff Mahoney if (btrfs_is_testing(root->fs_info)) 526faa2dbf0SJosef Bacik return 0; 527fccb84c9SDavid Sterba 528d1980131SDavid Sterba /* Ensure we can see the FORCE_COW bit */ 529d1980131SDavid Sterba smp_mb__before_atomic(); 530f1ebcc74SLiu Bo 531f1ebcc74SLiu Bo /* 532f1ebcc74SLiu Bo * We do not need to cow a block if 533f1ebcc74SLiu Bo * 1) this block is not created or changed in this transaction; 534f1ebcc74SLiu Bo * 2) this block does not belong to TREE_RELOC tree; 535f1ebcc74SLiu Bo * 3) the root is not forced COW. 536f1ebcc74SLiu Bo * 537f1ebcc74SLiu Bo * What is forced COW: 53801327610SNicholas D Steeves * when we create snapshot during committing the transaction, 53952042d8eSAndrea Gelmini * after we've finished copying src root, we must COW the shared 540f1ebcc74SLiu Bo * block to ensure the metadata consistency. 541f1ebcc74SLiu Bo */ 5425d4f98a2SYan Zheng if (btrfs_header_generation(buf) == trans->transid && 5435d4f98a2SYan Zheng !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && 5445d4f98a2SYan Zheng !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 545f1ebcc74SLiu Bo btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && 54627cdeb70SMiao Xie !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) 5475d4f98a2SYan Zheng return 0; 5485d4f98a2SYan Zheng return 1; 5495d4f98a2SYan Zheng } 5505d4f98a2SYan Zheng 551d352ac68SChris Mason /* 552d352ac68SChris Mason * cows a single block, see __btrfs_cow_block for the real work. 55301327610SNicholas D Steeves * This version of it has extra checks so that a block isn't COWed more than 554d352ac68SChris Mason * once per transaction, as long as it hasn't been written yet 555d352ac68SChris Mason */ 556d397712bSChris Mason noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, 5575f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *buf, 5585f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 5599631e4ccSJosef Bacik struct extent_buffer **cow_ret, 5609631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 56102217ed2SChris Mason { 5620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5636702ed49SChris Mason u64 search_start; 564f510cfecSChris Mason int ret; 565dc17ff8fSChris Mason 56683354f07SJosef Bacik if (test_bit(BTRFS_ROOT_DELETING, &root->state)) 56783354f07SJosef Bacik btrfs_err(fs_info, 56883354f07SJosef Bacik "COW'ing blocks on a fs root that's being dropped"); 56983354f07SJosef Bacik 5700b246afaSJeff Mahoney if (trans->transaction != fs_info->running_transaction) 57131b1a2bdSJulia Lawall WARN(1, KERN_CRIT "trans %llu running %llu\n", 572c1c9ff7cSGeert Uytterhoeven trans->transid, 5730b246afaSJeff Mahoney fs_info->running_transaction->transid); 57431b1a2bdSJulia Lawall 5750b246afaSJeff Mahoney if (trans->transid != fs_info->generation) 57631b1a2bdSJulia Lawall WARN(1, KERN_CRIT "trans %llu running %llu\n", 5770b246afaSJeff Mahoney trans->transid, fs_info->generation); 578dc17ff8fSChris Mason 5795d4f98a2SYan Zheng if (!should_cow_block(trans, root, buf)) { 58002217ed2SChris Mason *cow_ret = buf; 58102217ed2SChris Mason return 0; 58202217ed2SChris Mason } 583c487685dSChris Mason 584ee22184bSByongho Lee search_start = buf->start & ~((u64)SZ_1G - 1); 585b4ce94deSChris Mason 586f616f5cdSQu Wenruo /* 587f616f5cdSQu Wenruo * Before CoWing this block for later modification, check if it's 588f616f5cdSQu Wenruo * the subtree root and do the delayed subtree trace if needed. 589f616f5cdSQu Wenruo * 590f616f5cdSQu Wenruo * Also We don't care about the error, as it's handled internally. 591f616f5cdSQu Wenruo */ 592f616f5cdSQu Wenruo btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); 593f510cfecSChris Mason ret = __btrfs_cow_block(trans, root, buf, parent, 5949631e4ccSJosef Bacik parent_slot, cow_ret, search_start, 0, nest); 5951abe9b8aSliubo 5961abe9b8aSliubo trace_btrfs_cow_block(root, buf, *cow_ret); 5971abe9b8aSliubo 598f510cfecSChris Mason return ret; 5992c90e5d6SChris Mason } 600f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); 6016702ed49SChris Mason 602d352ac68SChris Mason /* 603d352ac68SChris Mason * helper function for defrag to decide if two blocks pointed to by a 604d352ac68SChris Mason * node are actually close by 605d352ac68SChris Mason */ 6066b80053dSChris Mason static int close_blocks(u64 blocknr, u64 other, u32 blocksize) 6076702ed49SChris Mason { 6086b80053dSChris Mason if (blocknr < other && other - (blocknr + blocksize) < 32768) 6096702ed49SChris Mason return 1; 6106b80053dSChris Mason if (blocknr > other && blocknr - (other + blocksize) < 32768) 6116702ed49SChris Mason return 1; 61202217ed2SChris Mason return 0; 61302217ed2SChris Mason } 61402217ed2SChris Mason 615ce6ef5abSDavid Sterba #ifdef __LITTLE_ENDIAN 616ce6ef5abSDavid Sterba 617ce6ef5abSDavid Sterba /* 618ce6ef5abSDavid Sterba * Compare two keys, on little-endian the disk order is same as CPU order and 619ce6ef5abSDavid Sterba * we can avoid the conversion. 620ce6ef5abSDavid Sterba */ 621ce6ef5abSDavid Sterba static int comp_keys(const struct btrfs_disk_key *disk_key, 622ce6ef5abSDavid Sterba const struct btrfs_key *k2) 623ce6ef5abSDavid Sterba { 624ce6ef5abSDavid Sterba const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; 625ce6ef5abSDavid Sterba 626ce6ef5abSDavid Sterba return btrfs_comp_cpu_keys(k1, k2); 627ce6ef5abSDavid Sterba } 628ce6ef5abSDavid Sterba 629ce6ef5abSDavid Sterba #else 630ce6ef5abSDavid Sterba 631081e9573SChris Mason /* 632081e9573SChris Mason * compare two keys in a memcmp fashion 633081e9573SChris Mason */ 634310712b2SOmar Sandoval static int comp_keys(const struct btrfs_disk_key *disk, 635310712b2SOmar Sandoval const struct btrfs_key *k2) 636081e9573SChris Mason { 637081e9573SChris Mason struct btrfs_key k1; 638081e9573SChris Mason 639081e9573SChris Mason btrfs_disk_key_to_cpu(&k1, disk); 640081e9573SChris Mason 64120736abaSDiego Calleja return btrfs_comp_cpu_keys(&k1, k2); 642081e9573SChris Mason } 643ce6ef5abSDavid Sterba #endif 644081e9573SChris Mason 645f3465ca4SJosef Bacik /* 646f3465ca4SJosef Bacik * same as comp_keys only with two btrfs_key's 647f3465ca4SJosef Bacik */ 648e1f60a65SDavid Sterba int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) 649f3465ca4SJosef Bacik { 650f3465ca4SJosef Bacik if (k1->objectid > k2->objectid) 651f3465ca4SJosef Bacik return 1; 652f3465ca4SJosef Bacik if (k1->objectid < k2->objectid) 653f3465ca4SJosef Bacik return -1; 654f3465ca4SJosef Bacik if (k1->type > k2->type) 655f3465ca4SJosef Bacik return 1; 656f3465ca4SJosef Bacik if (k1->type < k2->type) 657f3465ca4SJosef Bacik return -1; 658f3465ca4SJosef Bacik if (k1->offset > k2->offset) 659f3465ca4SJosef Bacik return 1; 660f3465ca4SJosef Bacik if (k1->offset < k2->offset) 661f3465ca4SJosef Bacik return -1; 662f3465ca4SJosef Bacik return 0; 663f3465ca4SJosef Bacik } 664081e9573SChris Mason 665d352ac68SChris Mason /* 666d352ac68SChris Mason * this is used by the defrag code to go through all the 667d352ac68SChris Mason * leaves pointed to by a node and reallocate them so that 668d352ac68SChris Mason * disk order is close to key order 669d352ac68SChris Mason */ 6706702ed49SChris Mason int btrfs_realloc_node(struct btrfs_trans_handle *trans, 6715f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *parent, 672de78b51aSEric Sandeen int start_slot, u64 *last_ret, 673a6b6e75eSChris Mason struct btrfs_key *progress) 6746702ed49SChris Mason { 6750b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6766b80053dSChris Mason struct extent_buffer *cur; 6776702ed49SChris Mason u64 blocknr; 678e9d0b13bSChris Mason u64 search_start = *last_ret; 679e9d0b13bSChris Mason u64 last_block = 0; 6806702ed49SChris Mason u64 other; 6816702ed49SChris Mason u32 parent_nritems; 6826702ed49SChris Mason int end_slot; 6836702ed49SChris Mason int i; 6846702ed49SChris Mason int err = 0; 6856b80053dSChris Mason u32 blocksize; 686081e9573SChris Mason int progress_passed = 0; 687081e9573SChris Mason struct btrfs_disk_key disk_key; 6886702ed49SChris Mason 6890b246afaSJeff Mahoney WARN_ON(trans->transaction != fs_info->running_transaction); 6900b246afaSJeff Mahoney WARN_ON(trans->transid != fs_info->generation); 69186479a04SChris Mason 6926b80053dSChris Mason parent_nritems = btrfs_header_nritems(parent); 6930b246afaSJeff Mahoney blocksize = fs_info->nodesize; 6945dfe2be7SFilipe Manana end_slot = parent_nritems - 1; 6956702ed49SChris Mason 6965dfe2be7SFilipe Manana if (parent_nritems <= 1) 6976702ed49SChris Mason return 0; 6986702ed49SChris Mason 6995dfe2be7SFilipe Manana for (i = start_slot; i <= end_slot; i++) { 7006702ed49SChris Mason int close = 1; 701a6b6e75eSChris Mason 702081e9573SChris Mason btrfs_node_key(parent, &disk_key, i); 703081e9573SChris Mason if (!progress_passed && comp_keys(&disk_key, progress) < 0) 704081e9573SChris Mason continue; 705081e9573SChris Mason 706081e9573SChris Mason progress_passed = 1; 7076b80053dSChris Mason blocknr = btrfs_node_blockptr(parent, i); 708e9d0b13bSChris Mason if (last_block == 0) 709e9d0b13bSChris Mason last_block = blocknr; 7105708b959SChris Mason 7116702ed49SChris Mason if (i > 0) { 7126b80053dSChris Mason other = btrfs_node_blockptr(parent, i - 1); 7136b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 7146702ed49SChris Mason } 7155dfe2be7SFilipe Manana if (!close && i < end_slot) { 7166b80053dSChris Mason other = btrfs_node_blockptr(parent, i + 1); 7176b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 7186702ed49SChris Mason } 719e9d0b13bSChris Mason if (close) { 720e9d0b13bSChris Mason last_block = blocknr; 7216702ed49SChris Mason continue; 722e9d0b13bSChris Mason } 7236702ed49SChris Mason 724206983b7SJosef Bacik cur = btrfs_read_node_slot(parent, i); 725206983b7SJosef Bacik if (IS_ERR(cur)) 72664c043deSLiu Bo return PTR_ERR(cur); 727e9d0b13bSChris Mason if (search_start == 0) 7286b80053dSChris Mason search_start = last_block; 729e9d0b13bSChris Mason 730e7a84565SChris Mason btrfs_tree_lock(cur); 7316b80053dSChris Mason err = __btrfs_cow_block(trans, root, cur, parent, i, 732e7a84565SChris Mason &cur, search_start, 7336b80053dSChris Mason min(16 * blocksize, 7349631e4ccSJosef Bacik (end_slot - i) * blocksize), 7359631e4ccSJosef Bacik BTRFS_NESTING_COW); 736252c38f0SYan if (err) { 737e7a84565SChris Mason btrfs_tree_unlock(cur); 7386b80053dSChris Mason free_extent_buffer(cur); 7396702ed49SChris Mason break; 740252c38f0SYan } 741e7a84565SChris Mason search_start = cur->start; 742e7a84565SChris Mason last_block = cur->start; 743f2183bdeSChris Mason *last_ret = search_start; 744e7a84565SChris Mason btrfs_tree_unlock(cur); 745e7a84565SChris Mason free_extent_buffer(cur); 7466702ed49SChris Mason } 7476702ed49SChris Mason return err; 7486702ed49SChris Mason } 7496702ed49SChris Mason 75074123bd7SChris Mason /* 751fb81212cSFilipe Manana * Search for a key in the given extent_buffer. 7525f39d397SChris Mason * 753fb81212cSFilipe Manana * The lower boundary for the search is specified by the slot number @low. Use a 754fb81212cSFilipe Manana * value of 0 to search over the whole extent buffer. 75574123bd7SChris Mason * 756fb81212cSFilipe Manana * The slot in the extent buffer is returned via @slot. If the key exists in the 757fb81212cSFilipe Manana * extent buffer, then @slot will point to the slot where the key is, otherwise 758fb81212cSFilipe Manana * it points to the slot where you would insert the key. 759fb81212cSFilipe Manana * 760fb81212cSFilipe Manana * Slot may point to the total number of items (i.e. one position beyond the last 761fb81212cSFilipe Manana * key) if the key is bigger than the last key in the extent buffer. 76274123bd7SChris Mason */ 763fb81212cSFilipe Manana static noinline int generic_bin_search(struct extent_buffer *eb, int low, 76467d5e289SMarcos Paulo de Souza const struct btrfs_key *key, int *slot) 765be0e5c09SChris Mason { 766fb81212cSFilipe Manana unsigned long p; 767fb81212cSFilipe Manana int item_size; 76867d5e289SMarcos Paulo de Souza int high = btrfs_header_nritems(eb); 769be0e5c09SChris Mason int ret; 7705cd17f34SDavid Sterba const int key_size = sizeof(struct btrfs_disk_key); 771be0e5c09SChris Mason 7725e24e9afSLiu Bo if (low > high) { 7735e24e9afSLiu Bo btrfs_err(eb->fs_info, 7745e24e9afSLiu Bo "%s: low (%d) > high (%d) eb %llu owner %llu level %d", 7755e24e9afSLiu Bo __func__, low, high, eb->start, 7765e24e9afSLiu Bo btrfs_header_owner(eb), btrfs_header_level(eb)); 7775e24e9afSLiu Bo return -EINVAL; 7785e24e9afSLiu Bo } 7795e24e9afSLiu Bo 780fb81212cSFilipe Manana if (btrfs_header_level(eb) == 0) { 781fb81212cSFilipe Manana p = offsetof(struct btrfs_leaf, items); 782fb81212cSFilipe Manana item_size = sizeof(struct btrfs_item); 783fb81212cSFilipe Manana } else { 784fb81212cSFilipe Manana p = offsetof(struct btrfs_node, ptrs); 785fb81212cSFilipe Manana item_size = sizeof(struct btrfs_key_ptr); 786fb81212cSFilipe Manana } 787fb81212cSFilipe Manana 788be0e5c09SChris Mason while (low < high) { 7895cd17f34SDavid Sterba unsigned long oip; 7905cd17f34SDavid Sterba unsigned long offset; 7915cd17f34SDavid Sterba struct btrfs_disk_key *tmp; 7925cd17f34SDavid Sterba struct btrfs_disk_key unaligned; 7935cd17f34SDavid Sterba int mid; 7945cd17f34SDavid Sterba 795be0e5c09SChris Mason mid = (low + high) / 2; 7965f39d397SChris Mason offset = p + mid * item_size; 7975cd17f34SDavid Sterba oip = offset_in_page(offset); 7985f39d397SChris Mason 7995cd17f34SDavid Sterba if (oip + key_size <= PAGE_SIZE) { 800884b07d0SQu Wenruo const unsigned long idx = get_eb_page_index(offset); 8015cd17f34SDavid Sterba char *kaddr = page_address(eb->pages[idx]); 802934d375bSChris Mason 803884b07d0SQu Wenruo oip = get_eb_offset_in_page(eb, offset); 8045cd17f34SDavid Sterba tmp = (struct btrfs_disk_key *)(kaddr + oip); 8055cd17f34SDavid Sterba } else { 8065cd17f34SDavid Sterba read_extent_buffer(eb, &unaligned, offset, key_size); 8075f39d397SChris Mason tmp = &unaligned; 808479965d6SChris Mason } 809479965d6SChris Mason 810be0e5c09SChris Mason ret = comp_keys(tmp, key); 811be0e5c09SChris Mason 812be0e5c09SChris Mason if (ret < 0) 813be0e5c09SChris Mason low = mid + 1; 814be0e5c09SChris Mason else if (ret > 0) 815be0e5c09SChris Mason high = mid; 816be0e5c09SChris Mason else { 817be0e5c09SChris Mason *slot = mid; 818be0e5c09SChris Mason return 0; 819be0e5c09SChris Mason } 820be0e5c09SChris Mason } 821be0e5c09SChris Mason *slot = low; 822be0e5c09SChris Mason return 1; 823be0e5c09SChris Mason } 824be0e5c09SChris Mason 82597571fd0SChris Mason /* 826fb81212cSFilipe Manana * Simple binary search on an extent buffer. Works for both leaves and nodes, and 827fb81212cSFilipe Manana * always searches over the whole range of keys (slot 0 to slot 'nritems - 1'). 82897571fd0SChris Mason */ 829a74b35ecSNikolay Borisov int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, 830e3b83361SQu Wenruo int *slot) 831be0e5c09SChris Mason { 832fb81212cSFilipe Manana return generic_bin_search(eb, 0, key, slot); 833be0e5c09SChris Mason } 834be0e5c09SChris Mason 835f0486c68SYan, Zheng static void root_add_used(struct btrfs_root *root, u32 size) 836f0486c68SYan, Zheng { 837f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 838f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 839f0486c68SYan, Zheng btrfs_root_used(&root->root_item) + size); 840f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 841f0486c68SYan, Zheng } 842f0486c68SYan, Zheng 843f0486c68SYan, Zheng static void root_sub_used(struct btrfs_root *root, u32 size) 844f0486c68SYan, Zheng { 845f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 846f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 847f0486c68SYan, Zheng btrfs_root_used(&root->root_item) - size); 848f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 849f0486c68SYan, Zheng } 850f0486c68SYan, Zheng 851d352ac68SChris Mason /* given a node and slot number, this reads the blocks it points to. The 852d352ac68SChris Mason * extent buffer is returned with a reference taken (but unlocked). 853d352ac68SChris Mason */ 8544b231ae4SDavid Sterba struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, 8554b231ae4SDavid Sterba int slot) 856bb803951SChris Mason { 857ca7a79adSChris Mason int level = btrfs_header_level(parent); 858416bc658SJosef Bacik struct extent_buffer *eb; 859581c1760SQu Wenruo struct btrfs_key first_key; 860416bc658SJosef Bacik 861fb770ae4SLiu Bo if (slot < 0 || slot >= btrfs_header_nritems(parent)) 862fb770ae4SLiu Bo return ERR_PTR(-ENOENT); 863ca7a79adSChris Mason 864ca7a79adSChris Mason BUG_ON(level == 0); 865ca7a79adSChris Mason 866581c1760SQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 867d0d20b0fSDavid Sterba eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), 8681b7ec85eSJosef Bacik btrfs_header_owner(parent), 869581c1760SQu Wenruo btrfs_node_ptr_generation(parent, slot), 870581c1760SQu Wenruo level - 1, &first_key); 8714eb150d6SQu Wenruo if (IS_ERR(eb)) 8724eb150d6SQu Wenruo return eb; 8734eb150d6SQu Wenruo if (!extent_buffer_uptodate(eb)) { 874416bc658SJosef Bacik free_extent_buffer(eb); 8754eb150d6SQu Wenruo return ERR_PTR(-EIO); 876416bc658SJosef Bacik } 877416bc658SJosef Bacik 878416bc658SJosef Bacik return eb; 879bb803951SChris Mason } 880bb803951SChris Mason 881d352ac68SChris Mason /* 882d352ac68SChris Mason * node level balancing, used to make sure nodes are in proper order for 883d352ac68SChris Mason * item deletion. We balance from the top down, so we have to make sure 884d352ac68SChris Mason * that a deletion won't leave an node completely empty later on. 885d352ac68SChris Mason */ 886e02119d5SChris Mason static noinline int balance_level(struct btrfs_trans_handle *trans, 88798ed5174SChris Mason struct btrfs_root *root, 88898ed5174SChris Mason struct btrfs_path *path, int level) 889bb803951SChris Mason { 8900b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8915f39d397SChris Mason struct extent_buffer *right = NULL; 8925f39d397SChris Mason struct extent_buffer *mid; 8935f39d397SChris Mason struct extent_buffer *left = NULL; 8945f39d397SChris Mason struct extent_buffer *parent = NULL; 895bb803951SChris Mason int ret = 0; 896bb803951SChris Mason int wret; 897bb803951SChris Mason int pslot; 898bb803951SChris Mason int orig_slot = path->slots[level]; 89979f95c82SChris Mason u64 orig_ptr; 900bb803951SChris Mason 90198e6b1ebSLiu Bo ASSERT(level > 0); 902bb803951SChris Mason 9035f39d397SChris Mason mid = path->nodes[level]; 904b4ce94deSChris Mason 905ac5887c8SJosef Bacik WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); 9067bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 9077bb86316SChris Mason 9081d4f8a0cSChris Mason orig_ptr = btrfs_node_blockptr(mid, orig_slot); 90979f95c82SChris Mason 910a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 9115f39d397SChris Mason parent = path->nodes[level + 1]; 912bb803951SChris Mason pslot = path->slots[level + 1]; 913a05a9bb1SLi Zefan } 914bb803951SChris Mason 91540689478SChris Mason /* 91640689478SChris Mason * deal with the case where there is only one pointer in the root 91740689478SChris Mason * by promoting the node below to a root 91840689478SChris Mason */ 9195f39d397SChris Mason if (!parent) { 9205f39d397SChris Mason struct extent_buffer *child; 921bb803951SChris Mason 9225f39d397SChris Mason if (btrfs_header_nritems(mid) != 1) 923bb803951SChris Mason return 0; 924bb803951SChris Mason 925bb803951SChris Mason /* promote the child to a root */ 9264b231ae4SDavid Sterba child = btrfs_read_node_slot(mid, 0); 927fb770ae4SLiu Bo if (IS_ERR(child)) { 928fb770ae4SLiu Bo ret = PTR_ERR(child); 9290b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 930305a26afSMark Fasheh goto enospc; 931305a26afSMark Fasheh } 932305a26afSMark Fasheh 933925baeddSChris Mason btrfs_tree_lock(child); 9349631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 9359631e4ccSJosef Bacik BTRFS_NESTING_COW); 936f0486c68SYan, Zheng if (ret) { 937f0486c68SYan, Zheng btrfs_tree_unlock(child); 938f0486c68SYan, Zheng free_extent_buffer(child); 939f0486c68SYan, Zheng goto enospc; 940f0486c68SYan, Zheng } 9412f375ab9SYan 942406808abSFilipe Manana ret = btrfs_tree_mod_log_insert_root(root->node, child, true); 943d9d19a01SDavid Sterba BUG_ON(ret < 0); 944240f62c8SChris Mason rcu_assign_pointer(root->node, child); 945925baeddSChris Mason 9460b86a832SChris Mason add_root_to_dirty_list(root); 947925baeddSChris Mason btrfs_tree_unlock(child); 948b4ce94deSChris Mason 949925baeddSChris Mason path->locks[level] = 0; 950bb803951SChris Mason path->nodes[level] = NULL; 9516a884d7dSDavid Sterba btrfs_clean_tree_block(mid); 952925baeddSChris Mason btrfs_tree_unlock(mid); 953bb803951SChris Mason /* once for the path */ 9545f39d397SChris Mason free_extent_buffer(mid); 955f0486c68SYan, Zheng 956f0486c68SYan, Zheng root_sub_used(root, mid->len); 9577a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); 958bb803951SChris Mason /* once for the root ptr */ 9593083ee2eSJosef Bacik free_extent_buffer_stale(mid); 960f0486c68SYan, Zheng return 0; 961bb803951SChris Mason } 9625f39d397SChris Mason if (btrfs_header_nritems(mid) > 9630b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) 964bb803951SChris Mason return 0; 965bb803951SChris Mason 9664b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 967fb770ae4SLiu Bo if (IS_ERR(left)) 968fb770ae4SLiu Bo left = NULL; 969fb770ae4SLiu Bo 9705f39d397SChris Mason if (left) { 971bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 9725f39d397SChris Mason wret = btrfs_cow_block(trans, root, left, 9739631e4ccSJosef Bacik parent, pslot - 1, &left, 974bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 97554aa1f4dSChris Mason if (wret) { 97654aa1f4dSChris Mason ret = wret; 97754aa1f4dSChris Mason goto enospc; 97854aa1f4dSChris Mason } 9792cc58cf2SChris Mason } 980fb770ae4SLiu Bo 9814b231ae4SDavid Sterba right = btrfs_read_node_slot(parent, pslot + 1); 982fb770ae4SLiu Bo if (IS_ERR(right)) 983fb770ae4SLiu Bo right = NULL; 984fb770ae4SLiu Bo 9855f39d397SChris Mason if (right) { 986bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 9875f39d397SChris Mason wret = btrfs_cow_block(trans, root, right, 9889631e4ccSJosef Bacik parent, pslot + 1, &right, 989bf59a5a2SJosef Bacik BTRFS_NESTING_RIGHT_COW); 9902cc58cf2SChris Mason if (wret) { 9912cc58cf2SChris Mason ret = wret; 9922cc58cf2SChris Mason goto enospc; 9932cc58cf2SChris Mason } 9942cc58cf2SChris Mason } 9952cc58cf2SChris Mason 9962cc58cf2SChris Mason /* first, try to make some room in the middle buffer */ 9975f39d397SChris Mason if (left) { 9985f39d397SChris Mason orig_slot += btrfs_header_nritems(left); 999d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 100079f95c82SChris Mason if (wret < 0) 100179f95c82SChris Mason ret = wret; 1002bb803951SChris Mason } 100379f95c82SChris Mason 100479f95c82SChris Mason /* 100579f95c82SChris Mason * then try to empty the right most buffer into the middle 100679f95c82SChris Mason */ 10075f39d397SChris Mason if (right) { 1008d30a668fSDavid Sterba wret = push_node_left(trans, mid, right, 1); 100954aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 101079f95c82SChris Mason ret = wret; 10115f39d397SChris Mason if (btrfs_header_nritems(right) == 0) { 10126a884d7dSDavid Sterba btrfs_clean_tree_block(right); 1013925baeddSChris Mason btrfs_tree_unlock(right); 1014afe5fea7STsutomu Itoh del_ptr(root, path, level + 1, pslot + 1); 1015f0486c68SYan, Zheng root_sub_used(root, right->len); 10167a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), right, 10177a163608SFilipe Manana 0, 1); 10183083ee2eSJosef Bacik free_extent_buffer_stale(right); 1019f0486c68SYan, Zheng right = NULL; 1020bb803951SChris Mason } else { 10215f39d397SChris Mason struct btrfs_disk_key right_key; 10225f39d397SChris Mason btrfs_node_key(right, &right_key, 0); 1023f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, 102433cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 10250e82bcfeSDavid Sterba BUG_ON(ret < 0); 10265f39d397SChris Mason btrfs_set_node_key(parent, &right_key, pslot + 1); 10275f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 1028bb803951SChris Mason } 1029bb803951SChris Mason } 10305f39d397SChris Mason if (btrfs_header_nritems(mid) == 1) { 103179f95c82SChris Mason /* 103279f95c82SChris Mason * we're not allowed to leave a node with one item in the 103379f95c82SChris Mason * tree during a delete. A deletion from lower in the tree 103479f95c82SChris Mason * could try to delete the only pointer in this node. 103579f95c82SChris Mason * So, pull some keys from the left. 103679f95c82SChris Mason * There has to be a left pointer at this point because 103779f95c82SChris Mason * otherwise we would have pulled some pointers from the 103879f95c82SChris Mason * right 103979f95c82SChris Mason */ 1040305a26afSMark Fasheh if (!left) { 1041305a26afSMark Fasheh ret = -EROFS; 10420b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 1043305a26afSMark Fasheh goto enospc; 1044305a26afSMark Fasheh } 104555d32ed8SDavid Sterba wret = balance_node_right(trans, mid, left); 104654aa1f4dSChris Mason if (wret < 0) { 104779f95c82SChris Mason ret = wret; 104854aa1f4dSChris Mason goto enospc; 104954aa1f4dSChris Mason } 1050bce4eae9SChris Mason if (wret == 1) { 1051d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 1052bce4eae9SChris Mason if (wret < 0) 1053bce4eae9SChris Mason ret = wret; 1054bce4eae9SChris Mason } 105579f95c82SChris Mason BUG_ON(wret == 1); 105679f95c82SChris Mason } 10575f39d397SChris Mason if (btrfs_header_nritems(mid) == 0) { 10586a884d7dSDavid Sterba btrfs_clean_tree_block(mid); 1059925baeddSChris Mason btrfs_tree_unlock(mid); 1060afe5fea7STsutomu Itoh del_ptr(root, path, level + 1, pslot); 1061f0486c68SYan, Zheng root_sub_used(root, mid->len); 10627a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); 10633083ee2eSJosef Bacik free_extent_buffer_stale(mid); 1064f0486c68SYan, Zheng mid = NULL; 106579f95c82SChris Mason } else { 106679f95c82SChris Mason /* update the parent key to reflect our changes */ 10675f39d397SChris Mason struct btrfs_disk_key mid_key; 10685f39d397SChris Mason btrfs_node_key(mid, &mid_key, 0); 1069f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot, 107033cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 10710e82bcfeSDavid Sterba BUG_ON(ret < 0); 10725f39d397SChris Mason btrfs_set_node_key(parent, &mid_key, pslot); 10735f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 107479f95c82SChris Mason } 1075bb803951SChris Mason 107679f95c82SChris Mason /* update the path */ 10775f39d397SChris Mason if (left) { 10785f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 107967439dadSDavid Sterba atomic_inc(&left->refs); 1080925baeddSChris Mason /* left was locked after cow */ 10815f39d397SChris Mason path->nodes[level] = left; 1082bb803951SChris Mason path->slots[level + 1] -= 1; 1083bb803951SChris Mason path->slots[level] = orig_slot; 1084925baeddSChris Mason if (mid) { 1085925baeddSChris Mason btrfs_tree_unlock(mid); 10865f39d397SChris Mason free_extent_buffer(mid); 1087925baeddSChris Mason } 1088bb803951SChris Mason } else { 10895f39d397SChris Mason orig_slot -= btrfs_header_nritems(left); 1090bb803951SChris Mason path->slots[level] = orig_slot; 1091bb803951SChris Mason } 1092bb803951SChris Mason } 109379f95c82SChris Mason /* double check we haven't messed things up */ 1094e20d96d6SChris Mason if (orig_ptr != 10955f39d397SChris Mason btrfs_node_blockptr(path->nodes[level], path->slots[level])) 109679f95c82SChris Mason BUG(); 109754aa1f4dSChris Mason enospc: 1098925baeddSChris Mason if (right) { 1099925baeddSChris Mason btrfs_tree_unlock(right); 11005f39d397SChris Mason free_extent_buffer(right); 1101925baeddSChris Mason } 1102925baeddSChris Mason if (left) { 1103925baeddSChris Mason if (path->nodes[level] != left) 1104925baeddSChris Mason btrfs_tree_unlock(left); 11055f39d397SChris Mason free_extent_buffer(left); 1106925baeddSChris Mason } 1107bb803951SChris Mason return ret; 1108bb803951SChris Mason } 1109bb803951SChris Mason 1110d352ac68SChris Mason /* Node balancing for insertion. Here we only split or push nodes around 1111d352ac68SChris Mason * when they are completely full. This is also done top down, so we 1112d352ac68SChris Mason * have to be pessimistic. 1113d352ac68SChris Mason */ 1114d397712bSChris Mason static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, 1115e66f709bSChris Mason struct btrfs_root *root, 1116e66f709bSChris Mason struct btrfs_path *path, int level) 1117e66f709bSChris Mason { 11180b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 11195f39d397SChris Mason struct extent_buffer *right = NULL; 11205f39d397SChris Mason struct extent_buffer *mid; 11215f39d397SChris Mason struct extent_buffer *left = NULL; 11225f39d397SChris Mason struct extent_buffer *parent = NULL; 1123e66f709bSChris Mason int ret = 0; 1124e66f709bSChris Mason int wret; 1125e66f709bSChris Mason int pslot; 1126e66f709bSChris Mason int orig_slot = path->slots[level]; 1127e66f709bSChris Mason 1128e66f709bSChris Mason if (level == 0) 1129e66f709bSChris Mason return 1; 1130e66f709bSChris Mason 11315f39d397SChris Mason mid = path->nodes[level]; 11327bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 1133e66f709bSChris Mason 1134a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 11355f39d397SChris Mason parent = path->nodes[level + 1]; 1136e66f709bSChris Mason pslot = path->slots[level + 1]; 1137a05a9bb1SLi Zefan } 1138e66f709bSChris Mason 11395f39d397SChris Mason if (!parent) 1140e66f709bSChris Mason return 1; 1141e66f709bSChris Mason 11424b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 1143fb770ae4SLiu Bo if (IS_ERR(left)) 1144fb770ae4SLiu Bo left = NULL; 1145e66f709bSChris Mason 1146e66f709bSChris Mason /* first, try to make some room in the middle buffer */ 11475f39d397SChris Mason if (left) { 1148e66f709bSChris Mason u32 left_nr; 1149925baeddSChris Mason 1150bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 1151b4ce94deSChris Mason 11525f39d397SChris Mason left_nr = btrfs_header_nritems(left); 11530b246afaSJeff Mahoney if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 115433ade1f8SChris Mason wret = 1; 115533ade1f8SChris Mason } else { 11565f39d397SChris Mason ret = btrfs_cow_block(trans, root, left, parent, 11579631e4ccSJosef Bacik pslot - 1, &left, 1158bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 115954aa1f4dSChris Mason if (ret) 116054aa1f4dSChris Mason wret = 1; 116154aa1f4dSChris Mason else { 1162d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 0); 116354aa1f4dSChris Mason } 116433ade1f8SChris Mason } 1165e66f709bSChris Mason if (wret < 0) 1166e66f709bSChris Mason ret = wret; 1167e66f709bSChris Mason if (wret == 0) { 11685f39d397SChris Mason struct btrfs_disk_key disk_key; 1169e66f709bSChris Mason orig_slot += left_nr; 11705f39d397SChris Mason btrfs_node_key(mid, &disk_key, 0); 1171f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot, 117233cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 11730e82bcfeSDavid Sterba BUG_ON(ret < 0); 11745f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot); 11755f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 11765f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 11775f39d397SChris Mason path->nodes[level] = left; 1178e66f709bSChris Mason path->slots[level + 1] -= 1; 1179e66f709bSChris Mason path->slots[level] = orig_slot; 1180925baeddSChris Mason btrfs_tree_unlock(mid); 11815f39d397SChris Mason free_extent_buffer(mid); 1182e66f709bSChris Mason } else { 1183e66f709bSChris Mason orig_slot -= 11845f39d397SChris Mason btrfs_header_nritems(left); 1185e66f709bSChris Mason path->slots[level] = orig_slot; 1186925baeddSChris Mason btrfs_tree_unlock(left); 11875f39d397SChris Mason free_extent_buffer(left); 1188e66f709bSChris Mason } 1189e66f709bSChris Mason return 0; 1190e66f709bSChris Mason } 1191925baeddSChris Mason btrfs_tree_unlock(left); 11925f39d397SChris Mason free_extent_buffer(left); 1193e66f709bSChris Mason } 11944b231ae4SDavid Sterba right = btrfs_read_node_slot(parent, pslot + 1); 1195fb770ae4SLiu Bo if (IS_ERR(right)) 1196fb770ae4SLiu Bo right = NULL; 1197e66f709bSChris Mason 1198e66f709bSChris Mason /* 1199e66f709bSChris Mason * then try to empty the right most buffer into the middle 1200e66f709bSChris Mason */ 12015f39d397SChris Mason if (right) { 120233ade1f8SChris Mason u32 right_nr; 1203b4ce94deSChris Mason 1204bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 1205b4ce94deSChris Mason 12065f39d397SChris Mason right_nr = btrfs_header_nritems(right); 12070b246afaSJeff Mahoney if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 120833ade1f8SChris Mason wret = 1; 120933ade1f8SChris Mason } else { 12105f39d397SChris Mason ret = btrfs_cow_block(trans, root, right, 12115f39d397SChris Mason parent, pslot + 1, 1212bf59a5a2SJosef Bacik &right, BTRFS_NESTING_RIGHT_COW); 121354aa1f4dSChris Mason if (ret) 121454aa1f4dSChris Mason wret = 1; 121554aa1f4dSChris Mason else { 121655d32ed8SDavid Sterba wret = balance_node_right(trans, right, mid); 121733ade1f8SChris Mason } 121854aa1f4dSChris Mason } 1219e66f709bSChris Mason if (wret < 0) 1220e66f709bSChris Mason ret = wret; 1221e66f709bSChris Mason if (wret == 0) { 12225f39d397SChris Mason struct btrfs_disk_key disk_key; 12235f39d397SChris Mason 12245f39d397SChris Mason btrfs_node_key(right, &disk_key, 0); 1225f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, 122633cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 12270e82bcfeSDavid Sterba BUG_ON(ret < 0); 12285f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot + 1); 12295f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 12305f39d397SChris Mason 12315f39d397SChris Mason if (btrfs_header_nritems(mid) <= orig_slot) { 12325f39d397SChris Mason path->nodes[level] = right; 1233e66f709bSChris Mason path->slots[level + 1] += 1; 1234e66f709bSChris Mason path->slots[level] = orig_slot - 12355f39d397SChris Mason btrfs_header_nritems(mid); 1236925baeddSChris Mason btrfs_tree_unlock(mid); 12375f39d397SChris Mason free_extent_buffer(mid); 1238e66f709bSChris Mason } else { 1239925baeddSChris Mason btrfs_tree_unlock(right); 12405f39d397SChris Mason free_extent_buffer(right); 1241e66f709bSChris Mason } 1242e66f709bSChris Mason return 0; 1243e66f709bSChris Mason } 1244925baeddSChris Mason btrfs_tree_unlock(right); 12455f39d397SChris Mason free_extent_buffer(right); 1246e66f709bSChris Mason } 1247e66f709bSChris Mason return 1; 1248e66f709bSChris Mason } 1249e66f709bSChris Mason 125074123bd7SChris Mason /* 1251d352ac68SChris Mason * readahead one full node of leaves, finding things that are close 1252d352ac68SChris Mason * to the block in 'slot', and triggering ra on them. 12533c69faecSChris Mason */ 12542ff7e61eSJeff Mahoney static void reada_for_search(struct btrfs_fs_info *fs_info, 1255e02119d5SChris Mason struct btrfs_path *path, 125601f46658SChris Mason int level, int slot, u64 objectid) 12573c69faecSChris Mason { 12585f39d397SChris Mason struct extent_buffer *node; 125901f46658SChris Mason struct btrfs_disk_key disk_key; 12603c69faecSChris Mason u32 nritems; 12613c69faecSChris Mason u64 search; 1262a7175319SChris Mason u64 target; 12636b80053dSChris Mason u64 nread = 0; 1264ace75066SFilipe Manana u64 nread_max; 12656b80053dSChris Mason u32 nr; 12666b80053dSChris Mason u32 blocksize; 12676b80053dSChris Mason u32 nscan = 0; 1268db94535dSChris Mason 1269ace75066SFilipe Manana if (level != 1 && path->reada != READA_FORWARD_ALWAYS) 12703c69faecSChris Mason return; 12713c69faecSChris Mason 12726702ed49SChris Mason if (!path->nodes[level]) 12736702ed49SChris Mason return; 12746702ed49SChris Mason 12755f39d397SChris Mason node = path->nodes[level]; 1276925baeddSChris Mason 1277ace75066SFilipe Manana /* 1278ace75066SFilipe Manana * Since the time between visiting leaves is much shorter than the time 1279ace75066SFilipe Manana * between visiting nodes, limit read ahead of nodes to 1, to avoid too 1280ace75066SFilipe Manana * much IO at once (possibly random). 1281ace75066SFilipe Manana */ 1282ace75066SFilipe Manana if (path->reada == READA_FORWARD_ALWAYS) { 1283ace75066SFilipe Manana if (level > 1) 1284ace75066SFilipe Manana nread_max = node->fs_info->nodesize; 1285ace75066SFilipe Manana else 1286ace75066SFilipe Manana nread_max = SZ_128K; 1287ace75066SFilipe Manana } else { 1288ace75066SFilipe Manana nread_max = SZ_64K; 1289ace75066SFilipe Manana } 1290ace75066SFilipe Manana 12913c69faecSChris Mason search = btrfs_node_blockptr(node, slot); 12920b246afaSJeff Mahoney blocksize = fs_info->nodesize; 1293069a2e37SFilipe Manana if (path->reada != READA_FORWARD_ALWAYS) { 1294069a2e37SFilipe Manana struct extent_buffer *eb; 1295069a2e37SFilipe Manana 12960b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, search); 12975f39d397SChris Mason if (eb) { 12985f39d397SChris Mason free_extent_buffer(eb); 12993c69faecSChris Mason return; 13003c69faecSChris Mason } 1301069a2e37SFilipe Manana } 13023c69faecSChris Mason 1303a7175319SChris Mason target = search; 13046b80053dSChris Mason 13055f39d397SChris Mason nritems = btrfs_header_nritems(node); 13066b80053dSChris Mason nr = slot; 130725b8b936SJosef Bacik 13083c69faecSChris Mason while (1) { 1309e4058b54SDavid Sterba if (path->reada == READA_BACK) { 13106b80053dSChris Mason if (nr == 0) 13113c69faecSChris Mason break; 13126b80053dSChris Mason nr--; 1313ace75066SFilipe Manana } else if (path->reada == READA_FORWARD || 1314ace75066SFilipe Manana path->reada == READA_FORWARD_ALWAYS) { 13156b80053dSChris Mason nr++; 13166b80053dSChris Mason if (nr >= nritems) 13176b80053dSChris Mason break; 13183c69faecSChris Mason } 1319e4058b54SDavid Sterba if (path->reada == READA_BACK && objectid) { 132001f46658SChris Mason btrfs_node_key(node, &disk_key, nr); 132101f46658SChris Mason if (btrfs_disk_key_objectid(&disk_key) != objectid) 132201f46658SChris Mason break; 132301f46658SChris Mason } 13246b80053dSChris Mason search = btrfs_node_blockptr(node, nr); 1325ace75066SFilipe Manana if (path->reada == READA_FORWARD_ALWAYS || 1326ace75066SFilipe Manana (search <= target && target - search <= 65536) || 1327a7175319SChris Mason (search > target && search - target <= 65536)) { 1328bfb484d9SJosef Bacik btrfs_readahead_node_child(node, nr); 13296b80053dSChris Mason nread += blocksize; 13303c69faecSChris Mason } 13316b80053dSChris Mason nscan++; 1332ace75066SFilipe Manana if (nread > nread_max || nscan > 32) 13336b80053dSChris Mason break; 13343c69faecSChris Mason } 13353c69faecSChris Mason } 1336925baeddSChris Mason 1337bfb484d9SJosef Bacik static noinline void reada_for_balance(struct btrfs_path *path, int level) 1338b4ce94deSChris Mason { 1339bfb484d9SJosef Bacik struct extent_buffer *parent; 1340b4ce94deSChris Mason int slot; 1341b4ce94deSChris Mason int nritems; 1342b4ce94deSChris Mason 13438c594ea8SChris Mason parent = path->nodes[level + 1]; 1344b4ce94deSChris Mason if (!parent) 13450b08851fSJosef Bacik return; 1346b4ce94deSChris Mason 1347b4ce94deSChris Mason nritems = btrfs_header_nritems(parent); 13488c594ea8SChris Mason slot = path->slots[level + 1]; 1349b4ce94deSChris Mason 1350bfb484d9SJosef Bacik if (slot > 0) 1351bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot - 1); 1352bfb484d9SJosef Bacik if (slot + 1 < nritems) 1353bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot + 1); 1354b4ce94deSChris Mason } 1355b4ce94deSChris Mason 1356b4ce94deSChris Mason 1357b4ce94deSChris Mason /* 1358d397712bSChris Mason * when we walk down the tree, it is usually safe to unlock the higher layers 1359d397712bSChris Mason * in the tree. The exceptions are when our path goes through slot 0, because 1360d397712bSChris Mason * operations on the tree might require changing key pointers higher up in the 1361d397712bSChris Mason * tree. 1362d352ac68SChris Mason * 1363d397712bSChris Mason * callers might also have set path->keep_locks, which tells this code to keep 1364d397712bSChris Mason * the lock if the path points to the last slot in the block. This is part of 1365d397712bSChris Mason * walking through the tree, and selecting the next slot in the higher block. 1366d352ac68SChris Mason * 1367d397712bSChris Mason * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so 1368d397712bSChris Mason * if lowest_unlock is 1, level 0 won't be unlocked 1369d352ac68SChris Mason */ 1370e02119d5SChris Mason static noinline void unlock_up(struct btrfs_path *path, int level, 1371f7c79f30SChris Mason int lowest_unlock, int min_write_lock_level, 1372f7c79f30SChris Mason int *write_lock_level) 1373925baeddSChris Mason { 1374925baeddSChris Mason int i; 1375925baeddSChris Mason int skip_level = level; 1376c1227996SNikolay Borisov bool check_skip = true; 1377925baeddSChris Mason 1378925baeddSChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 1379925baeddSChris Mason if (!path->nodes[i]) 1380925baeddSChris Mason break; 1381925baeddSChris Mason if (!path->locks[i]) 1382925baeddSChris Mason break; 1383c1227996SNikolay Borisov 1384c1227996SNikolay Borisov if (check_skip) { 1385c1227996SNikolay Borisov if (path->slots[i] == 0) { 1386925baeddSChris Mason skip_level = i + 1; 1387925baeddSChris Mason continue; 1388925baeddSChris Mason } 1389c1227996SNikolay Borisov 1390c1227996SNikolay Borisov if (path->keep_locks) { 1391925baeddSChris Mason u32 nritems; 1392c1227996SNikolay Borisov 1393c1227996SNikolay Borisov nritems = btrfs_header_nritems(path->nodes[i]); 1394051e1b9fSChris Mason if (nritems < 1 || path->slots[i] >= nritems - 1) { 1395925baeddSChris Mason skip_level = i + 1; 1396925baeddSChris Mason continue; 1397925baeddSChris Mason } 1398925baeddSChris Mason } 1399c1227996SNikolay Borisov } 1400051e1b9fSChris Mason 1401d80bb3f9SLiu Bo if (i >= lowest_unlock && i > skip_level) { 1402c1227996SNikolay Borisov check_skip = false; 1403c1227996SNikolay Borisov btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); 1404925baeddSChris Mason path->locks[i] = 0; 1405f7c79f30SChris Mason if (write_lock_level && 1406f7c79f30SChris Mason i > min_write_lock_level && 1407f7c79f30SChris Mason i <= *write_lock_level) { 1408f7c79f30SChris Mason *write_lock_level = i - 1; 1409f7c79f30SChris Mason } 1410925baeddSChris Mason } 1411925baeddSChris Mason } 1412925baeddSChris Mason } 1413925baeddSChris Mason 14143c69faecSChris Mason /* 1415376a21d7SFilipe Manana * Helper function for btrfs_search_slot() and other functions that do a search 1416376a21d7SFilipe Manana * on a btree. The goal is to find a tree block in the cache (the radix tree at 1417376a21d7SFilipe Manana * fs_info->buffer_radix), but if we can't find it, or it's not up to date, read 1418376a21d7SFilipe Manana * its pages from disk. 1419c8c42864SChris Mason * 1420376a21d7SFilipe Manana * Returns -EAGAIN, with the path unlocked, if the caller needs to repeat the 1421376a21d7SFilipe Manana * whole btree search, starting again from the current root node. 1422c8c42864SChris Mason */ 1423c8c42864SChris Mason static int 1424d07b8528SLiu Bo read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, 1425c8c42864SChris Mason struct extent_buffer **eb_ret, int level, int slot, 1426cda79c54SDavid Sterba const struct btrfs_key *key) 1427c8c42864SChris Mason { 14280b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1429c8c42864SChris Mason u64 blocknr; 1430c8c42864SChris Mason u64 gen; 1431c8c42864SChris Mason struct extent_buffer *tmp; 1432581c1760SQu Wenruo struct btrfs_key first_key; 143376a05b35SChris Mason int ret; 1434581c1760SQu Wenruo int parent_level; 1435b246666eSFilipe Manana bool unlock_up; 1436c8c42864SChris Mason 1437b246666eSFilipe Manana unlock_up = ((level + 1 < BTRFS_MAX_LEVEL) && p->locks[level + 1]); 1438213ff4b7SNikolay Borisov blocknr = btrfs_node_blockptr(*eb_ret, slot); 1439213ff4b7SNikolay Borisov gen = btrfs_node_ptr_generation(*eb_ret, slot); 1440213ff4b7SNikolay Borisov parent_level = btrfs_header_level(*eb_ret); 1441213ff4b7SNikolay Borisov btrfs_node_key_to_cpu(*eb_ret, &first_key, slot); 1442c8c42864SChris Mason 1443b246666eSFilipe Manana /* 1444b246666eSFilipe Manana * If we need to read an extent buffer from disk and we are holding locks 1445b246666eSFilipe Manana * on upper level nodes, we unlock all the upper nodes before reading the 1446b246666eSFilipe Manana * extent buffer, and then return -EAGAIN to the caller as it needs to 1447b246666eSFilipe Manana * restart the search. We don't release the lock on the current level 1448b246666eSFilipe Manana * because we need to walk this node to figure out which blocks to read. 1449b246666eSFilipe Manana */ 14500b246afaSJeff Mahoney tmp = find_extent_buffer(fs_info, blocknr); 1451cb44921aSChris Mason if (tmp) { 1452ace75066SFilipe Manana if (p->reada == READA_FORWARD_ALWAYS) 1453ace75066SFilipe Manana reada_for_search(fs_info, p, level, slot, key->objectid); 1454ace75066SFilipe Manana 1455b9fab919SChris Mason /* first we do an atomic uptodate check */ 1456b9fab919SChris Mason if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { 1457448de471SQu Wenruo /* 1458448de471SQu Wenruo * Do extra check for first_key, eb can be stale due to 1459448de471SQu Wenruo * being cached, read from scrub, or have multiple 1460448de471SQu Wenruo * parents (shared tree blocks). 1461448de471SQu Wenruo */ 1462e064d5e9SDavid Sterba if (btrfs_verify_level_key(tmp, 1463448de471SQu Wenruo parent_level - 1, &first_key, gen)) { 1464448de471SQu Wenruo free_extent_buffer(tmp); 1465448de471SQu Wenruo return -EUCLEAN; 1466448de471SQu Wenruo } 1467c8c42864SChris Mason *eb_ret = tmp; 1468c8c42864SChris Mason return 0; 1469c8c42864SChris Mason } 1470bdf7c00eSJosef Bacik 1471857bc13fSJosef Bacik if (p->nowait) { 1472857bc13fSJosef Bacik free_extent_buffer(tmp); 1473857bc13fSJosef Bacik return -EAGAIN; 1474857bc13fSJosef Bacik } 1475857bc13fSJosef Bacik 1476b246666eSFilipe Manana if (unlock_up) 1477b246666eSFilipe Manana btrfs_unlock_up_safe(p, level + 1); 1478b246666eSFilipe Manana 1479b9fab919SChris Mason /* now we're allowed to do a blocking uptodate check */ 14806a2e9dc4SFilipe Manana ret = btrfs_read_extent_buffer(tmp, gen, parent_level - 1, &first_key); 14819a4ffa1bSQu Wenruo if (ret) { 1482cb44921aSChris Mason free_extent_buffer(tmp); 1483b3b4aa74SDavid Sterba btrfs_release_path(p); 1484cb44921aSChris Mason return -EIO; 1485cb44921aSChris Mason } 148688c602abSQu Wenruo if (btrfs_check_eb_owner(tmp, root->root_key.objectid)) { 148788c602abSQu Wenruo free_extent_buffer(tmp); 148888c602abSQu Wenruo btrfs_release_path(p); 148988c602abSQu Wenruo return -EUCLEAN; 149088c602abSQu Wenruo } 1491b246666eSFilipe Manana 1492b246666eSFilipe Manana if (unlock_up) 1493b246666eSFilipe Manana ret = -EAGAIN; 1494b246666eSFilipe Manana 1495b246666eSFilipe Manana goto out; 1496857bc13fSJosef Bacik } else if (p->nowait) { 1497857bc13fSJosef Bacik return -EAGAIN; 14989a4ffa1bSQu Wenruo } 1499c8c42864SChris Mason 1500b246666eSFilipe Manana if (unlock_up) { 15018c594ea8SChris Mason btrfs_unlock_up_safe(p, level + 1); 15024bb59055SFilipe Manana ret = -EAGAIN; 15034bb59055SFilipe Manana } else { 15044bb59055SFilipe Manana ret = 0; 15054bb59055SFilipe Manana } 15068c594ea8SChris Mason 1507e4058b54SDavid Sterba if (p->reada != READA_NONE) 15082ff7e61eSJeff Mahoney reada_for_search(fs_info, p, level, slot, key->objectid); 1509c8c42864SChris Mason 15101b7ec85eSJosef Bacik tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid, 15111b7ec85eSJosef Bacik gen, parent_level - 1, &first_key); 15124eb150d6SQu Wenruo if (IS_ERR(tmp)) { 15134eb150d6SQu Wenruo btrfs_release_path(p); 15144eb150d6SQu Wenruo return PTR_ERR(tmp); 15154eb150d6SQu Wenruo } 151676a05b35SChris Mason /* 151776a05b35SChris Mason * If the read above didn't mark this buffer up to date, 151876a05b35SChris Mason * it will never end up being up to date. Set ret to EIO now 151976a05b35SChris Mason * and give up so that our caller doesn't loop forever 152076a05b35SChris Mason * on our EAGAINs. 152176a05b35SChris Mason */ 1522e6a1d6fdSLiu Bo if (!extent_buffer_uptodate(tmp)) 152376a05b35SChris Mason ret = -EIO; 152402a3307aSLiu Bo 1525b246666eSFilipe Manana out: 15264bb59055SFilipe Manana if (ret == 0) { 15274bb59055SFilipe Manana *eb_ret = tmp; 15284bb59055SFilipe Manana } else { 15294bb59055SFilipe Manana free_extent_buffer(tmp); 153002a3307aSLiu Bo btrfs_release_path(p); 15314bb59055SFilipe Manana } 15324bb59055SFilipe Manana 153376a05b35SChris Mason return ret; 1534c8c42864SChris Mason } 1535c8c42864SChris Mason 1536c8c42864SChris Mason /* 1537c8c42864SChris Mason * helper function for btrfs_search_slot. This does all of the checks 1538c8c42864SChris Mason * for node-level blocks and does any balancing required based on 1539c8c42864SChris Mason * the ins_len. 1540c8c42864SChris Mason * 1541c8c42864SChris Mason * If no extra work was required, zero is returned. If we had to 1542c8c42864SChris Mason * drop the path, -EAGAIN is returned and btrfs_search_slot must 1543c8c42864SChris Mason * start over 1544c8c42864SChris Mason */ 1545c8c42864SChris Mason static int 1546c8c42864SChris Mason setup_nodes_for_search(struct btrfs_trans_handle *trans, 1547c8c42864SChris Mason struct btrfs_root *root, struct btrfs_path *p, 1548bd681513SChris Mason struct extent_buffer *b, int level, int ins_len, 1549bd681513SChris Mason int *write_lock_level) 1550c8c42864SChris Mason { 15510b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 155295b982deSNikolay Borisov int ret = 0; 15530b246afaSJeff Mahoney 1554c8c42864SChris Mason if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= 15550b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { 1556c8c42864SChris Mason 1557bd681513SChris Mason if (*write_lock_level < level + 1) { 1558bd681513SChris Mason *write_lock_level = level + 1; 1559bd681513SChris Mason btrfs_release_path(p); 156095b982deSNikolay Borisov return -EAGAIN; 1561bd681513SChris Mason } 1562bd681513SChris Mason 1563bfb484d9SJosef Bacik reada_for_balance(p, level); 156495b982deSNikolay Borisov ret = split_node(trans, root, p, level); 1565c8c42864SChris Mason 1566c8c42864SChris Mason b = p->nodes[level]; 1567c8c42864SChris Mason } else if (ins_len < 0 && btrfs_header_nritems(b) < 15680b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { 1569c8c42864SChris Mason 1570bd681513SChris Mason if (*write_lock_level < level + 1) { 1571bd681513SChris Mason *write_lock_level = level + 1; 1572bd681513SChris Mason btrfs_release_path(p); 157395b982deSNikolay Borisov return -EAGAIN; 1574bd681513SChris Mason } 1575bd681513SChris Mason 1576bfb484d9SJosef Bacik reada_for_balance(p, level); 157795b982deSNikolay Borisov ret = balance_level(trans, root, p, level); 157895b982deSNikolay Borisov if (ret) 157995b982deSNikolay Borisov return ret; 1580c8c42864SChris Mason 1581c8c42864SChris Mason b = p->nodes[level]; 1582c8c42864SChris Mason if (!b) { 1583b3b4aa74SDavid Sterba btrfs_release_path(p); 158495b982deSNikolay Borisov return -EAGAIN; 1585c8c42864SChris Mason } 1586c8c42864SChris Mason BUG_ON(btrfs_header_nritems(b) == 1); 1587c8c42864SChris Mason } 1588c8c42864SChris Mason return ret; 1589c8c42864SChris Mason } 1590c8c42864SChris Mason 1591381cf658SDavid Sterba int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, 1592e33d5c3dSKelley Nielsen u64 iobjectid, u64 ioff, u8 key_type, 1593e33d5c3dSKelley Nielsen struct btrfs_key *found_key) 1594e33d5c3dSKelley Nielsen { 1595e33d5c3dSKelley Nielsen int ret; 1596e33d5c3dSKelley Nielsen struct btrfs_key key; 1597e33d5c3dSKelley Nielsen struct extent_buffer *eb; 1598381cf658SDavid Sterba 1599381cf658SDavid Sterba ASSERT(path); 16001d4c08e0SDavid Sterba ASSERT(found_key); 1601e33d5c3dSKelley Nielsen 1602e33d5c3dSKelley Nielsen key.type = key_type; 1603e33d5c3dSKelley Nielsen key.objectid = iobjectid; 1604e33d5c3dSKelley Nielsen key.offset = ioff; 1605e33d5c3dSKelley Nielsen 1606e33d5c3dSKelley Nielsen ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); 16071d4c08e0SDavid Sterba if (ret < 0) 1608e33d5c3dSKelley Nielsen return ret; 1609e33d5c3dSKelley Nielsen 1610e33d5c3dSKelley Nielsen eb = path->nodes[0]; 1611e33d5c3dSKelley Nielsen if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { 1612e33d5c3dSKelley Nielsen ret = btrfs_next_leaf(fs_root, path); 1613e33d5c3dSKelley Nielsen if (ret) 1614e33d5c3dSKelley Nielsen return ret; 1615e33d5c3dSKelley Nielsen eb = path->nodes[0]; 1616e33d5c3dSKelley Nielsen } 1617e33d5c3dSKelley Nielsen 1618e33d5c3dSKelley Nielsen btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); 1619e33d5c3dSKelley Nielsen if (found_key->type != key.type || 1620e33d5c3dSKelley Nielsen found_key->objectid != key.objectid) 1621e33d5c3dSKelley Nielsen return 1; 1622e33d5c3dSKelley Nielsen 1623e33d5c3dSKelley Nielsen return 0; 1624e33d5c3dSKelley Nielsen } 1625e33d5c3dSKelley Nielsen 16261fc28d8eSLiu Bo static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, 16271fc28d8eSLiu Bo struct btrfs_path *p, 16281fc28d8eSLiu Bo int write_lock_level) 16291fc28d8eSLiu Bo { 16301fc28d8eSLiu Bo struct extent_buffer *b; 1631120de408SJosef Bacik int root_lock = 0; 16321fc28d8eSLiu Bo int level = 0; 16331fc28d8eSLiu Bo 16341fc28d8eSLiu Bo if (p->search_commit_root) { 16351fc28d8eSLiu Bo b = root->commit_root; 163667439dadSDavid Sterba atomic_inc(&b->refs); 16371fc28d8eSLiu Bo level = btrfs_header_level(b); 1638f9ddfd05SLiu Bo /* 1639f9ddfd05SLiu Bo * Ensure that all callers have set skip_locking when 1640f9ddfd05SLiu Bo * p->search_commit_root = 1. 1641f9ddfd05SLiu Bo */ 1642f9ddfd05SLiu Bo ASSERT(p->skip_locking == 1); 16431fc28d8eSLiu Bo 16441fc28d8eSLiu Bo goto out; 16451fc28d8eSLiu Bo } 16461fc28d8eSLiu Bo 16471fc28d8eSLiu Bo if (p->skip_locking) { 16481fc28d8eSLiu Bo b = btrfs_root_node(root); 16491fc28d8eSLiu Bo level = btrfs_header_level(b); 16501fc28d8eSLiu Bo goto out; 16511fc28d8eSLiu Bo } 16521fc28d8eSLiu Bo 1653120de408SJosef Bacik /* We try very hard to do read locks on the root */ 1654120de408SJosef Bacik root_lock = BTRFS_READ_LOCK; 1655120de408SJosef Bacik 16561fc28d8eSLiu Bo /* 1657662c653bSLiu Bo * If the level is set to maximum, we can skip trying to get the read 1658662c653bSLiu Bo * lock. 1659662c653bSLiu Bo */ 1660662c653bSLiu Bo if (write_lock_level < BTRFS_MAX_LEVEL) { 1661662c653bSLiu Bo /* 1662662c653bSLiu Bo * We don't know the level of the root node until we actually 1663662c653bSLiu Bo * have it read locked 16641fc28d8eSLiu Bo */ 1665857bc13fSJosef Bacik if (p->nowait) { 1666857bc13fSJosef Bacik b = btrfs_try_read_lock_root_node(root); 1667857bc13fSJosef Bacik if (IS_ERR(b)) 1668857bc13fSJosef Bacik return b; 1669857bc13fSJosef Bacik } else { 16701bb96598SJosef Bacik b = btrfs_read_lock_root_node(root); 1671857bc13fSJosef Bacik } 16721fc28d8eSLiu Bo level = btrfs_header_level(b); 16731fc28d8eSLiu Bo if (level > write_lock_level) 16741fc28d8eSLiu Bo goto out; 16751fc28d8eSLiu Bo 1676662c653bSLiu Bo /* Whoops, must trade for write lock */ 16771fc28d8eSLiu Bo btrfs_tree_read_unlock(b); 16781fc28d8eSLiu Bo free_extent_buffer(b); 1679662c653bSLiu Bo } 1680662c653bSLiu Bo 16811fc28d8eSLiu Bo b = btrfs_lock_root_node(root); 16821fc28d8eSLiu Bo root_lock = BTRFS_WRITE_LOCK; 16831fc28d8eSLiu Bo 16841fc28d8eSLiu Bo /* The level might have changed, check again */ 16851fc28d8eSLiu Bo level = btrfs_header_level(b); 16861fc28d8eSLiu Bo 16871fc28d8eSLiu Bo out: 1688120de408SJosef Bacik /* 1689120de408SJosef Bacik * The root may have failed to write out at some point, and thus is no 1690120de408SJosef Bacik * longer valid, return an error in this case. 1691120de408SJosef Bacik */ 1692120de408SJosef Bacik if (!extent_buffer_uptodate(b)) { 1693120de408SJosef Bacik if (root_lock) 1694120de408SJosef Bacik btrfs_tree_unlock_rw(b, root_lock); 1695120de408SJosef Bacik free_extent_buffer(b); 1696120de408SJosef Bacik return ERR_PTR(-EIO); 1697120de408SJosef Bacik } 1698120de408SJosef Bacik 16991fc28d8eSLiu Bo p->nodes[level] = b; 17001fc28d8eSLiu Bo if (!p->skip_locking) 17011fc28d8eSLiu Bo p->locks[level] = root_lock; 17021fc28d8eSLiu Bo /* 17031fc28d8eSLiu Bo * Callers are responsible for dropping b's references. 17041fc28d8eSLiu Bo */ 17051fc28d8eSLiu Bo return b; 17061fc28d8eSLiu Bo } 17071fc28d8eSLiu Bo 1708d96b3424SFilipe Manana /* 1709d96b3424SFilipe Manana * Replace the extent buffer at the lowest level of the path with a cloned 1710d96b3424SFilipe Manana * version. The purpose is to be able to use it safely, after releasing the 1711d96b3424SFilipe Manana * commit root semaphore, even if relocation is happening in parallel, the 1712d96b3424SFilipe Manana * transaction used for relocation is committed and the extent buffer is 1713d96b3424SFilipe Manana * reallocated in the next transaction. 1714d96b3424SFilipe Manana * 1715d96b3424SFilipe Manana * This is used in a context where the caller does not prevent transaction 1716d96b3424SFilipe Manana * commits from happening, either by holding a transaction handle or holding 1717d96b3424SFilipe Manana * some lock, while it's doing searches through a commit root. 1718d96b3424SFilipe Manana * At the moment it's only used for send operations. 1719d96b3424SFilipe Manana */ 1720d96b3424SFilipe Manana static int finish_need_commit_sem_search(struct btrfs_path *path) 1721d96b3424SFilipe Manana { 1722d96b3424SFilipe Manana const int i = path->lowest_level; 1723d96b3424SFilipe Manana const int slot = path->slots[i]; 1724d96b3424SFilipe Manana struct extent_buffer *lowest = path->nodes[i]; 1725d96b3424SFilipe Manana struct extent_buffer *clone; 1726d96b3424SFilipe Manana 1727d96b3424SFilipe Manana ASSERT(path->need_commit_sem); 1728d96b3424SFilipe Manana 1729d96b3424SFilipe Manana if (!lowest) 1730d96b3424SFilipe Manana return 0; 1731d96b3424SFilipe Manana 1732d96b3424SFilipe Manana lockdep_assert_held_read(&lowest->fs_info->commit_root_sem); 1733d96b3424SFilipe Manana 1734d96b3424SFilipe Manana clone = btrfs_clone_extent_buffer(lowest); 1735d96b3424SFilipe Manana if (!clone) 1736d96b3424SFilipe Manana return -ENOMEM; 1737d96b3424SFilipe Manana 1738d96b3424SFilipe Manana btrfs_release_path(path); 1739d96b3424SFilipe Manana path->nodes[i] = clone; 1740d96b3424SFilipe Manana path->slots[i] = slot; 1741d96b3424SFilipe Manana 1742d96b3424SFilipe Manana return 0; 1743d96b3424SFilipe Manana } 17441fc28d8eSLiu Bo 1745e2e58d0fSFilipe Manana static inline int search_for_key_slot(struct extent_buffer *eb, 1746e2e58d0fSFilipe Manana int search_low_slot, 1747e2e58d0fSFilipe Manana const struct btrfs_key *key, 1748e2e58d0fSFilipe Manana int prev_cmp, 1749e2e58d0fSFilipe Manana int *slot) 1750e2e58d0fSFilipe Manana { 1751e2e58d0fSFilipe Manana /* 1752e2e58d0fSFilipe Manana * If a previous call to btrfs_bin_search() on a parent node returned an 1753e2e58d0fSFilipe Manana * exact match (prev_cmp == 0), we can safely assume the target key will 1754e2e58d0fSFilipe Manana * always be at slot 0 on lower levels, since each key pointer 1755e2e58d0fSFilipe Manana * (struct btrfs_key_ptr) refers to the lowest key accessible from the 1756e2e58d0fSFilipe Manana * subtree it points to. Thus we can skip searching lower levels. 1757e2e58d0fSFilipe Manana */ 1758e2e58d0fSFilipe Manana if (prev_cmp == 0) { 1759e2e58d0fSFilipe Manana *slot = 0; 1760e2e58d0fSFilipe Manana return 0; 1761e2e58d0fSFilipe Manana } 1762e2e58d0fSFilipe Manana 1763e2e58d0fSFilipe Manana return generic_bin_search(eb, search_low_slot, key, slot); 1764e2e58d0fSFilipe Manana } 1765e2e58d0fSFilipe Manana 1766109324cfSFilipe Manana static int search_leaf(struct btrfs_trans_handle *trans, 1767109324cfSFilipe Manana struct btrfs_root *root, 1768109324cfSFilipe Manana const struct btrfs_key *key, 1769109324cfSFilipe Manana struct btrfs_path *path, 1770109324cfSFilipe Manana int ins_len, 1771109324cfSFilipe Manana int prev_cmp) 1772109324cfSFilipe Manana { 1773109324cfSFilipe Manana struct extent_buffer *leaf = path->nodes[0]; 1774109324cfSFilipe Manana int leaf_free_space = -1; 1775109324cfSFilipe Manana int search_low_slot = 0; 1776109324cfSFilipe Manana int ret; 1777109324cfSFilipe Manana bool do_bin_search = true; 1778109324cfSFilipe Manana 1779109324cfSFilipe Manana /* 1780109324cfSFilipe Manana * If we are doing an insertion, the leaf has enough free space and the 1781109324cfSFilipe Manana * destination slot for the key is not slot 0, then we can unlock our 1782109324cfSFilipe Manana * write lock on the parent, and any other upper nodes, before doing the 1783109324cfSFilipe Manana * binary search on the leaf (with search_for_key_slot()), allowing other 1784109324cfSFilipe Manana * tasks to lock the parent and any other upper nodes. 1785109324cfSFilipe Manana */ 1786109324cfSFilipe Manana if (ins_len > 0) { 1787109324cfSFilipe Manana /* 1788109324cfSFilipe Manana * Cache the leaf free space, since we will need it later and it 1789109324cfSFilipe Manana * will not change until then. 1790109324cfSFilipe Manana */ 1791109324cfSFilipe Manana leaf_free_space = btrfs_leaf_free_space(leaf); 1792109324cfSFilipe Manana 1793109324cfSFilipe Manana /* 1794109324cfSFilipe Manana * !path->locks[1] means we have a single node tree, the leaf is 1795109324cfSFilipe Manana * the root of the tree. 1796109324cfSFilipe Manana */ 1797109324cfSFilipe Manana if (path->locks[1] && leaf_free_space >= ins_len) { 1798109324cfSFilipe Manana struct btrfs_disk_key first_key; 1799109324cfSFilipe Manana 1800109324cfSFilipe Manana ASSERT(btrfs_header_nritems(leaf) > 0); 1801109324cfSFilipe Manana btrfs_item_key(leaf, &first_key, 0); 1802109324cfSFilipe Manana 1803109324cfSFilipe Manana /* 1804109324cfSFilipe Manana * Doing the extra comparison with the first key is cheap, 1805109324cfSFilipe Manana * taking into account that the first key is very likely 1806109324cfSFilipe Manana * already in a cache line because it immediately follows 1807109324cfSFilipe Manana * the extent buffer's header and we have recently accessed 1808109324cfSFilipe Manana * the header's level field. 1809109324cfSFilipe Manana */ 1810109324cfSFilipe Manana ret = comp_keys(&first_key, key); 1811109324cfSFilipe Manana if (ret < 0) { 1812109324cfSFilipe Manana /* 1813109324cfSFilipe Manana * The first key is smaller than the key we want 1814109324cfSFilipe Manana * to insert, so we are safe to unlock all upper 1815109324cfSFilipe Manana * nodes and we have to do the binary search. 1816109324cfSFilipe Manana * 1817109324cfSFilipe Manana * We do use btrfs_unlock_up_safe() and not 1818109324cfSFilipe Manana * unlock_up() because the later does not unlock 1819109324cfSFilipe Manana * nodes with a slot of 0 - we can safely unlock 1820109324cfSFilipe Manana * any node even if its slot is 0 since in this 1821109324cfSFilipe Manana * case the key does not end up at slot 0 of the 1822109324cfSFilipe Manana * leaf and there's no need to split the leaf. 1823109324cfSFilipe Manana */ 1824109324cfSFilipe Manana btrfs_unlock_up_safe(path, 1); 1825109324cfSFilipe Manana search_low_slot = 1; 1826109324cfSFilipe Manana } else { 1827109324cfSFilipe Manana /* 1828109324cfSFilipe Manana * The first key is >= then the key we want to 1829109324cfSFilipe Manana * insert, so we can skip the binary search as 1830109324cfSFilipe Manana * the target key will be at slot 0. 1831109324cfSFilipe Manana * 1832109324cfSFilipe Manana * We can not unlock upper nodes when the key is 1833109324cfSFilipe Manana * less than the first key, because we will need 1834109324cfSFilipe Manana * to update the key at slot 0 of the parent node 1835109324cfSFilipe Manana * and possibly of other upper nodes too. 1836109324cfSFilipe Manana * If the key matches the first key, then we can 1837109324cfSFilipe Manana * unlock all the upper nodes, using 1838109324cfSFilipe Manana * btrfs_unlock_up_safe() instead of unlock_up() 1839109324cfSFilipe Manana * as stated above. 1840109324cfSFilipe Manana */ 1841109324cfSFilipe Manana if (ret == 0) 1842109324cfSFilipe Manana btrfs_unlock_up_safe(path, 1); 1843109324cfSFilipe Manana /* 1844109324cfSFilipe Manana * ret is already 0 or 1, matching the result of 1845109324cfSFilipe Manana * a btrfs_bin_search() call, so there is no need 1846109324cfSFilipe Manana * to adjust it. 1847109324cfSFilipe Manana */ 1848109324cfSFilipe Manana do_bin_search = false; 1849109324cfSFilipe Manana path->slots[0] = 0; 1850109324cfSFilipe Manana } 1851109324cfSFilipe Manana } 1852109324cfSFilipe Manana } 1853109324cfSFilipe Manana 1854109324cfSFilipe Manana if (do_bin_search) { 1855109324cfSFilipe Manana ret = search_for_key_slot(leaf, search_low_slot, key, 1856109324cfSFilipe Manana prev_cmp, &path->slots[0]); 1857109324cfSFilipe Manana if (ret < 0) 1858109324cfSFilipe Manana return ret; 1859109324cfSFilipe Manana } 1860109324cfSFilipe Manana 1861109324cfSFilipe Manana if (ins_len > 0) { 1862109324cfSFilipe Manana /* 1863109324cfSFilipe Manana * Item key already exists. In this case, if we are allowed to 1864109324cfSFilipe Manana * insert the item (for example, in dir_item case, item key 1865109324cfSFilipe Manana * collision is allowed), it will be merged with the original 1866109324cfSFilipe Manana * item. Only the item size grows, no new btrfs item will be 1867109324cfSFilipe Manana * added. If search_for_extension is not set, ins_len already 1868109324cfSFilipe Manana * accounts the size btrfs_item, deduct it here so leaf space 1869109324cfSFilipe Manana * check will be correct. 1870109324cfSFilipe Manana */ 1871109324cfSFilipe Manana if (ret == 0 && !path->search_for_extension) { 1872109324cfSFilipe Manana ASSERT(ins_len >= sizeof(struct btrfs_item)); 1873109324cfSFilipe Manana ins_len -= sizeof(struct btrfs_item); 1874109324cfSFilipe Manana } 1875109324cfSFilipe Manana 1876109324cfSFilipe Manana ASSERT(leaf_free_space >= 0); 1877109324cfSFilipe Manana 1878109324cfSFilipe Manana if (leaf_free_space < ins_len) { 1879109324cfSFilipe Manana int err; 1880109324cfSFilipe Manana 1881109324cfSFilipe Manana err = split_leaf(trans, root, key, path, ins_len, 1882109324cfSFilipe Manana (ret == 0)); 1883bb8e9a60SFilipe Manana ASSERT(err <= 0); 1884bb8e9a60SFilipe Manana if (WARN_ON(err > 0)) 1885bb8e9a60SFilipe Manana err = -EUCLEAN; 1886109324cfSFilipe Manana if (err) 1887109324cfSFilipe Manana ret = err; 1888109324cfSFilipe Manana } 1889109324cfSFilipe Manana } 1890109324cfSFilipe Manana 1891109324cfSFilipe Manana return ret; 1892109324cfSFilipe Manana } 1893109324cfSFilipe Manana 1894c8c42864SChris Mason /* 18954271eceaSNikolay Borisov * btrfs_search_slot - look for a key in a tree and perform necessary 18964271eceaSNikolay Borisov * modifications to preserve tree invariants. 189774123bd7SChris Mason * 18984271eceaSNikolay Borisov * @trans: Handle of transaction, used when modifying the tree 18994271eceaSNikolay Borisov * @p: Holds all btree nodes along the search path 19004271eceaSNikolay Borisov * @root: The root node of the tree 19014271eceaSNikolay Borisov * @key: The key we are looking for 19029a664971Sethanwu * @ins_len: Indicates purpose of search: 19039a664971Sethanwu * >0 for inserts it's size of item inserted (*) 19049a664971Sethanwu * <0 for deletions 19059a664971Sethanwu * 0 for plain searches, not modifying the tree 19069a664971Sethanwu * 19079a664971Sethanwu * (*) If size of item inserted doesn't include 19089a664971Sethanwu * sizeof(struct btrfs_item), then p->search_for_extension must 19099a664971Sethanwu * be set. 19104271eceaSNikolay Borisov * @cow: boolean should CoW operations be performed. Must always be 1 19114271eceaSNikolay Borisov * when modifying the tree. 191297571fd0SChris Mason * 19134271eceaSNikolay Borisov * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. 19144271eceaSNikolay Borisov * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) 19154271eceaSNikolay Borisov * 19164271eceaSNikolay Borisov * If @key is found, 0 is returned and you can find the item in the leaf level 19174271eceaSNikolay Borisov * of the path (level 0) 19184271eceaSNikolay Borisov * 19194271eceaSNikolay Borisov * If @key isn't found, 1 is returned and the leaf level of the path (level 0) 19204271eceaSNikolay Borisov * points to the slot where it should be inserted 19214271eceaSNikolay Borisov * 19224271eceaSNikolay Borisov * If an error is encountered while searching the tree a negative error number 19234271eceaSNikolay Borisov * is returned 192474123bd7SChris Mason */ 1925310712b2SOmar Sandoval int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, 1926310712b2SOmar Sandoval const struct btrfs_key *key, struct btrfs_path *p, 1927310712b2SOmar Sandoval int ins_len, int cow) 1928be0e5c09SChris Mason { 1929d96b3424SFilipe Manana struct btrfs_fs_info *fs_info = root->fs_info; 19305f39d397SChris Mason struct extent_buffer *b; 1931be0e5c09SChris Mason int slot; 1932be0e5c09SChris Mason int ret; 193333c66f43SYan Zheng int err; 1934be0e5c09SChris Mason int level; 1935925baeddSChris Mason int lowest_unlock = 1; 1936bd681513SChris Mason /* everything at write_lock_level or lower must be write locked */ 1937bd681513SChris Mason int write_lock_level = 0; 19389f3a7427SChris Mason u8 lowest_level = 0; 1939f7c79f30SChris Mason int min_write_lock_level; 1940d7396f07SFilipe David Borba Manana int prev_cmp; 19419f3a7427SChris Mason 19426702ed49SChris Mason lowest_level = p->lowest_level; 1943323ac95bSChris Mason WARN_ON(lowest_level && ins_len > 0); 194422b0ebdaSChris Mason WARN_ON(p->nodes[0] != NULL); 1945eb653de1SFilipe David Borba Manana BUG_ON(!cow && ins_len); 194625179201SJosef Bacik 1947857bc13fSJosef Bacik /* 1948857bc13fSJosef Bacik * For now only allow nowait for read only operations. There's no 1949857bc13fSJosef Bacik * strict reason why we can't, we just only need it for reads so it's 1950857bc13fSJosef Bacik * only implemented for reads. 1951857bc13fSJosef Bacik */ 1952857bc13fSJosef Bacik ASSERT(!p->nowait || !cow); 1953857bc13fSJosef Bacik 1954bd681513SChris Mason if (ins_len < 0) { 1955925baeddSChris Mason lowest_unlock = 2; 195665b51a00SChris Mason 1957bd681513SChris Mason /* when we are removing items, we might have to go up to level 1958bd681513SChris Mason * two as we update tree pointers Make sure we keep write 1959bd681513SChris Mason * for those levels as well 1960bd681513SChris Mason */ 1961bd681513SChris Mason write_lock_level = 2; 1962bd681513SChris Mason } else if (ins_len > 0) { 1963bd681513SChris Mason /* 1964bd681513SChris Mason * for inserting items, make sure we have a write lock on 1965bd681513SChris Mason * level 1 so we can update keys 1966bd681513SChris Mason */ 1967bd681513SChris Mason write_lock_level = 1; 1968bd681513SChris Mason } 1969bd681513SChris Mason 1970bd681513SChris Mason if (!cow) 1971bd681513SChris Mason write_lock_level = -1; 1972bd681513SChris Mason 197309a2a8f9SJosef Bacik if (cow && (p->keep_locks || p->lowest_level)) 1974bd681513SChris Mason write_lock_level = BTRFS_MAX_LEVEL; 1975bd681513SChris Mason 1976f7c79f30SChris Mason min_write_lock_level = write_lock_level; 1977f7c79f30SChris Mason 1978d96b3424SFilipe Manana if (p->need_commit_sem) { 1979d96b3424SFilipe Manana ASSERT(p->search_commit_root); 1980857bc13fSJosef Bacik if (p->nowait) { 1981857bc13fSJosef Bacik if (!down_read_trylock(&fs_info->commit_root_sem)) 1982857bc13fSJosef Bacik return -EAGAIN; 1983857bc13fSJosef Bacik } else { 1984d96b3424SFilipe Manana down_read(&fs_info->commit_root_sem); 1985d96b3424SFilipe Manana } 1986857bc13fSJosef Bacik } 1987d96b3424SFilipe Manana 1988bb803951SChris Mason again: 1989d7396f07SFilipe David Borba Manana prev_cmp = -1; 19901fc28d8eSLiu Bo b = btrfs_search_slot_get_root(root, p, write_lock_level); 1991be6821f8SFilipe Manana if (IS_ERR(b)) { 1992be6821f8SFilipe Manana ret = PTR_ERR(b); 1993be6821f8SFilipe Manana goto done; 1994be6821f8SFilipe Manana } 1995925baeddSChris Mason 1996eb60ceacSChris Mason while (b) { 1997f624d976SQu Wenruo int dec = 0; 1998f624d976SQu Wenruo 19995f39d397SChris Mason level = btrfs_header_level(b); 200065b51a00SChris Mason 200102217ed2SChris Mason if (cow) { 20029ea2c7c9SNikolay Borisov bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); 20039ea2c7c9SNikolay Borisov 2004c8c42864SChris Mason /* 2005c8c42864SChris Mason * if we don't really need to cow this block 2006c8c42864SChris Mason * then we don't want to set the path blocking, 2007c8c42864SChris Mason * so we test it here 2008c8c42864SChris Mason */ 20095963ffcaSJosef Bacik if (!should_cow_block(trans, root, b)) 201065b51a00SChris Mason goto cow_done; 20115d4f98a2SYan Zheng 2012bd681513SChris Mason /* 2013bd681513SChris Mason * must have write locks on this node and the 2014bd681513SChris Mason * parent 2015bd681513SChris Mason */ 20165124e00eSJosef Bacik if (level > write_lock_level || 20175124e00eSJosef Bacik (level + 1 > write_lock_level && 20185124e00eSJosef Bacik level + 1 < BTRFS_MAX_LEVEL && 20195124e00eSJosef Bacik p->nodes[level + 1])) { 2020bd681513SChris Mason write_lock_level = level + 1; 2021bd681513SChris Mason btrfs_release_path(p); 2022bd681513SChris Mason goto again; 2023bd681513SChris Mason } 2024bd681513SChris Mason 20259ea2c7c9SNikolay Borisov if (last_level) 20269ea2c7c9SNikolay Borisov err = btrfs_cow_block(trans, root, b, NULL, 0, 20279631e4ccSJosef Bacik &b, 20289631e4ccSJosef Bacik BTRFS_NESTING_COW); 20299ea2c7c9SNikolay Borisov else 203033c66f43SYan Zheng err = btrfs_cow_block(trans, root, b, 2031e20d96d6SChris Mason p->nodes[level + 1], 20329631e4ccSJosef Bacik p->slots[level + 1], &b, 20339631e4ccSJosef Bacik BTRFS_NESTING_COW); 203433c66f43SYan Zheng if (err) { 203533c66f43SYan Zheng ret = err; 203665b51a00SChris Mason goto done; 203754aa1f4dSChris Mason } 203802217ed2SChris Mason } 203965b51a00SChris Mason cow_done: 2040eb60ceacSChris Mason p->nodes[level] = b; 2041b4ce94deSChris Mason 2042b4ce94deSChris Mason /* 2043b4ce94deSChris Mason * we have a lock on b and as long as we aren't changing 2044b4ce94deSChris Mason * the tree, there is no way to for the items in b to change. 2045b4ce94deSChris Mason * It is safe to drop the lock on our parent before we 2046b4ce94deSChris Mason * go through the expensive btree search on b. 2047b4ce94deSChris Mason * 2048eb653de1SFilipe David Borba Manana * If we're inserting or deleting (ins_len != 0), then we might 2049eb653de1SFilipe David Borba Manana * be changing slot zero, which may require changing the parent. 2050eb653de1SFilipe David Borba Manana * So, we can't drop the lock until after we know which slot 2051eb653de1SFilipe David Borba Manana * we're operating on. 2052b4ce94deSChris Mason */ 2053eb653de1SFilipe David Borba Manana if (!ins_len && !p->keep_locks) { 2054eb653de1SFilipe David Borba Manana int u = level + 1; 2055eb653de1SFilipe David Borba Manana 2056eb653de1SFilipe David Borba Manana if (u < BTRFS_MAX_LEVEL && p->locks[u]) { 2057eb653de1SFilipe David Borba Manana btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); 2058eb653de1SFilipe David Borba Manana p->locks[u] = 0; 2059eb653de1SFilipe David Borba Manana } 2060eb653de1SFilipe David Borba Manana } 2061b4ce94deSChris Mason 2062e2e58d0fSFilipe Manana if (level == 0) { 2063109324cfSFilipe Manana if (ins_len > 0) 2064e5e1c174SFilipe Manana ASSERT(write_lock_level >= 1); 2065bd681513SChris Mason 2066109324cfSFilipe Manana ret = search_leaf(trans, root, key, p, ins_len, prev_cmp); 2067459931ecSChris Mason if (!p->search_for_split) 2068f7c79f30SChris Mason unlock_up(p, level, lowest_unlock, 20694b6f8e96SLiu Bo min_write_lock_level, NULL); 207065b51a00SChris Mason goto done; 207165b51a00SChris Mason } 2072e2e58d0fSFilipe Manana 2073e2e58d0fSFilipe Manana ret = search_for_key_slot(b, 0, key, prev_cmp, &slot); 2074e2e58d0fSFilipe Manana if (ret < 0) 2075e2e58d0fSFilipe Manana goto done; 2076e2e58d0fSFilipe Manana prev_cmp = ret; 2077e2e58d0fSFilipe Manana 2078f624d976SQu Wenruo if (ret && slot > 0) { 2079f624d976SQu Wenruo dec = 1; 2080f624d976SQu Wenruo slot--; 2081f624d976SQu Wenruo } 2082f624d976SQu Wenruo p->slots[level] = slot; 2083f624d976SQu Wenruo err = setup_nodes_for_search(trans, root, p, b, level, ins_len, 2084f624d976SQu Wenruo &write_lock_level); 2085f624d976SQu Wenruo if (err == -EAGAIN) 2086f624d976SQu Wenruo goto again; 2087f624d976SQu Wenruo if (err) { 2088f624d976SQu Wenruo ret = err; 2089f624d976SQu Wenruo goto done; 2090f624d976SQu Wenruo } 2091f624d976SQu Wenruo b = p->nodes[level]; 2092f624d976SQu Wenruo slot = p->slots[level]; 2093f624d976SQu Wenruo 2094f624d976SQu Wenruo /* 2095f624d976SQu Wenruo * Slot 0 is special, if we change the key we have to update 2096f624d976SQu Wenruo * the parent pointer which means we must have a write lock on 2097f624d976SQu Wenruo * the parent 2098f624d976SQu Wenruo */ 2099f624d976SQu Wenruo if (slot == 0 && ins_len && write_lock_level < level + 1) { 2100f624d976SQu Wenruo write_lock_level = level + 1; 2101f624d976SQu Wenruo btrfs_release_path(p); 2102f624d976SQu Wenruo goto again; 2103f624d976SQu Wenruo } 2104f624d976SQu Wenruo 2105f624d976SQu Wenruo unlock_up(p, level, lowest_unlock, min_write_lock_level, 2106f624d976SQu Wenruo &write_lock_level); 2107f624d976SQu Wenruo 2108f624d976SQu Wenruo if (level == lowest_level) { 2109f624d976SQu Wenruo if (dec) 2110f624d976SQu Wenruo p->slots[level]++; 2111f624d976SQu Wenruo goto done; 2112f624d976SQu Wenruo } 2113f624d976SQu Wenruo 2114f624d976SQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 2115f624d976SQu Wenruo if (err == -EAGAIN) 2116f624d976SQu Wenruo goto again; 2117f624d976SQu Wenruo if (err) { 2118f624d976SQu Wenruo ret = err; 2119f624d976SQu Wenruo goto done; 2120f624d976SQu Wenruo } 2121f624d976SQu Wenruo 2122f624d976SQu Wenruo if (!p->skip_locking) { 2123f624d976SQu Wenruo level = btrfs_header_level(b); 2124b40130b2SJosef Bacik 2125b40130b2SJosef Bacik btrfs_maybe_reset_lockdep_class(root, b); 2126b40130b2SJosef Bacik 2127f624d976SQu Wenruo if (level <= write_lock_level) { 2128f624d976SQu Wenruo btrfs_tree_lock(b); 2129f624d976SQu Wenruo p->locks[level] = BTRFS_WRITE_LOCK; 2130f624d976SQu Wenruo } else { 2131857bc13fSJosef Bacik if (p->nowait) { 2132857bc13fSJosef Bacik if (!btrfs_try_tree_read_lock(b)) { 2133857bc13fSJosef Bacik free_extent_buffer(b); 2134857bc13fSJosef Bacik ret = -EAGAIN; 2135857bc13fSJosef Bacik goto done; 2136857bc13fSJosef Bacik } 2137857bc13fSJosef Bacik } else { 2138fe596ca3SJosef Bacik btrfs_tree_read_lock(b); 2139857bc13fSJosef Bacik } 2140f624d976SQu Wenruo p->locks[level] = BTRFS_READ_LOCK; 2141f624d976SQu Wenruo } 2142f624d976SQu Wenruo p->nodes[level] = b; 2143f624d976SQu Wenruo } 214465b51a00SChris Mason } 214565b51a00SChris Mason ret = 1; 214665b51a00SChris Mason done: 21475f5bc6b1SFilipe Manana if (ret < 0 && !p->skip_release_on_error) 2148b3b4aa74SDavid Sterba btrfs_release_path(p); 2149d96b3424SFilipe Manana 2150d96b3424SFilipe Manana if (p->need_commit_sem) { 2151d96b3424SFilipe Manana int ret2; 2152d96b3424SFilipe Manana 2153d96b3424SFilipe Manana ret2 = finish_need_commit_sem_search(p); 2154d96b3424SFilipe Manana up_read(&fs_info->commit_root_sem); 2155d96b3424SFilipe Manana if (ret2) 2156d96b3424SFilipe Manana ret = ret2; 2157d96b3424SFilipe Manana } 2158d96b3424SFilipe Manana 2159be0e5c09SChris Mason return ret; 2160be0e5c09SChris Mason } 2161f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); 2162be0e5c09SChris Mason 216374123bd7SChris Mason /* 21645d9e75c4SJan Schmidt * Like btrfs_search_slot, this looks for a key in the given tree. It uses the 21655d9e75c4SJan Schmidt * current state of the tree together with the operations recorded in the tree 21665d9e75c4SJan Schmidt * modification log to search for the key in a previous version of this tree, as 21675d9e75c4SJan Schmidt * denoted by the time_seq parameter. 21685d9e75c4SJan Schmidt * 21695d9e75c4SJan Schmidt * Naturally, there is no support for insert, delete or cow operations. 21705d9e75c4SJan Schmidt * 21715d9e75c4SJan Schmidt * The resulting path and return value will be set up as if we called 21725d9e75c4SJan Schmidt * btrfs_search_slot at that point in time with ins_len and cow both set to 0. 21735d9e75c4SJan Schmidt */ 2174310712b2SOmar Sandoval int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, 21755d9e75c4SJan Schmidt struct btrfs_path *p, u64 time_seq) 21765d9e75c4SJan Schmidt { 21770b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 21785d9e75c4SJan Schmidt struct extent_buffer *b; 21795d9e75c4SJan Schmidt int slot; 21805d9e75c4SJan Schmidt int ret; 21815d9e75c4SJan Schmidt int err; 21825d9e75c4SJan Schmidt int level; 21835d9e75c4SJan Schmidt int lowest_unlock = 1; 21845d9e75c4SJan Schmidt u8 lowest_level = 0; 21855d9e75c4SJan Schmidt 21865d9e75c4SJan Schmidt lowest_level = p->lowest_level; 21875d9e75c4SJan Schmidt WARN_ON(p->nodes[0] != NULL); 2188c922b016SStefan Roesch ASSERT(!p->nowait); 21895d9e75c4SJan Schmidt 21905d9e75c4SJan Schmidt if (p->search_commit_root) { 21915d9e75c4SJan Schmidt BUG_ON(time_seq); 21925d9e75c4SJan Schmidt return btrfs_search_slot(NULL, root, key, p, 0, 0); 21935d9e75c4SJan Schmidt } 21945d9e75c4SJan Schmidt 21955d9e75c4SJan Schmidt again: 2196f3a84ccdSFilipe Manana b = btrfs_get_old_root(root, time_seq); 2197315bed43SNikolay Borisov if (!b) { 2198315bed43SNikolay Borisov ret = -EIO; 2199315bed43SNikolay Borisov goto done; 2200315bed43SNikolay Borisov } 22015d9e75c4SJan Schmidt level = btrfs_header_level(b); 22025d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 22035d9e75c4SJan Schmidt 22045d9e75c4SJan Schmidt while (b) { 2205abe9339dSQu Wenruo int dec = 0; 2206abe9339dSQu Wenruo 22075d9e75c4SJan Schmidt level = btrfs_header_level(b); 22085d9e75c4SJan Schmidt p->nodes[level] = b; 22095d9e75c4SJan Schmidt 22105d9e75c4SJan Schmidt /* 22115d9e75c4SJan Schmidt * we have a lock on b and as long as we aren't changing 22125d9e75c4SJan Schmidt * the tree, there is no way to for the items in b to change. 22135d9e75c4SJan Schmidt * It is safe to drop the lock on our parent before we 22145d9e75c4SJan Schmidt * go through the expensive btree search on b. 22155d9e75c4SJan Schmidt */ 22165d9e75c4SJan Schmidt btrfs_unlock_up_safe(p, level + 1); 22175d9e75c4SJan Schmidt 2218995e9a16SNikolay Borisov ret = btrfs_bin_search(b, key, &slot); 2219cbca7d59SFilipe Manana if (ret < 0) 2220cbca7d59SFilipe Manana goto done; 22215d9e75c4SJan Schmidt 2222abe9339dSQu Wenruo if (level == 0) { 2223abe9339dSQu Wenruo p->slots[level] = slot; 2224abe9339dSQu Wenruo unlock_up(p, level, lowest_unlock, 0, NULL); 2225abe9339dSQu Wenruo goto done; 2226abe9339dSQu Wenruo } 2227abe9339dSQu Wenruo 22285d9e75c4SJan Schmidt if (ret && slot > 0) { 22295d9e75c4SJan Schmidt dec = 1; 2230abe9339dSQu Wenruo slot--; 22315d9e75c4SJan Schmidt } 22325d9e75c4SJan Schmidt p->slots[level] = slot; 22335d9e75c4SJan Schmidt unlock_up(p, level, lowest_unlock, 0, NULL); 22345d9e75c4SJan Schmidt 22355d9e75c4SJan Schmidt if (level == lowest_level) { 22365d9e75c4SJan Schmidt if (dec) 22375d9e75c4SJan Schmidt p->slots[level]++; 22385d9e75c4SJan Schmidt goto done; 22395d9e75c4SJan Schmidt } 22405d9e75c4SJan Schmidt 2241abe9339dSQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 22425d9e75c4SJan Schmidt if (err == -EAGAIN) 22435d9e75c4SJan Schmidt goto again; 22445d9e75c4SJan Schmidt if (err) { 22455d9e75c4SJan Schmidt ret = err; 22465d9e75c4SJan Schmidt goto done; 22475d9e75c4SJan Schmidt } 22485d9e75c4SJan Schmidt 22495d9e75c4SJan Schmidt level = btrfs_header_level(b); 22505d9e75c4SJan Schmidt btrfs_tree_read_lock(b); 2251f3a84ccdSFilipe Manana b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq); 2252db7f3436SJosef Bacik if (!b) { 2253db7f3436SJosef Bacik ret = -ENOMEM; 2254db7f3436SJosef Bacik goto done; 2255db7f3436SJosef Bacik } 22565d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 22575d9e75c4SJan Schmidt p->nodes[level] = b; 22585d9e75c4SJan Schmidt } 22595d9e75c4SJan Schmidt ret = 1; 22605d9e75c4SJan Schmidt done: 22615d9e75c4SJan Schmidt if (ret < 0) 22625d9e75c4SJan Schmidt btrfs_release_path(p); 22635d9e75c4SJan Schmidt 22645d9e75c4SJan Schmidt return ret; 22655d9e75c4SJan Schmidt } 22665d9e75c4SJan Schmidt 22675d9e75c4SJan Schmidt /* 22682f38b3e1SArne Jansen * helper to use instead of search slot if no exact match is needed but 22692f38b3e1SArne Jansen * instead the next or previous item should be returned. 22702f38b3e1SArne Jansen * When find_higher is true, the next higher item is returned, the next lower 22712f38b3e1SArne Jansen * otherwise. 22722f38b3e1SArne Jansen * When return_any and find_higher are both true, and no higher item is found, 22732f38b3e1SArne Jansen * return the next lower instead. 22742f38b3e1SArne Jansen * When return_any is true and find_higher is false, and no lower item is found, 22752f38b3e1SArne Jansen * return the next higher instead. 22762f38b3e1SArne Jansen * It returns 0 if any item is found, 1 if none is found (tree empty), and 22772f38b3e1SArne Jansen * < 0 on error 22782f38b3e1SArne Jansen */ 22792f38b3e1SArne Jansen int btrfs_search_slot_for_read(struct btrfs_root *root, 2280310712b2SOmar Sandoval const struct btrfs_key *key, 2281310712b2SOmar Sandoval struct btrfs_path *p, int find_higher, 2282310712b2SOmar Sandoval int return_any) 22832f38b3e1SArne Jansen { 22842f38b3e1SArne Jansen int ret; 22852f38b3e1SArne Jansen struct extent_buffer *leaf; 22862f38b3e1SArne Jansen 22872f38b3e1SArne Jansen again: 22882f38b3e1SArne Jansen ret = btrfs_search_slot(NULL, root, key, p, 0, 0); 22892f38b3e1SArne Jansen if (ret <= 0) 22902f38b3e1SArne Jansen return ret; 22912f38b3e1SArne Jansen /* 22922f38b3e1SArne Jansen * a return value of 1 means the path is at the position where the 22932f38b3e1SArne Jansen * item should be inserted. Normally this is the next bigger item, 22942f38b3e1SArne Jansen * but in case the previous item is the last in a leaf, path points 22952f38b3e1SArne Jansen * to the first free slot in the previous leaf, i.e. at an invalid 22962f38b3e1SArne Jansen * item. 22972f38b3e1SArne Jansen */ 22982f38b3e1SArne Jansen leaf = p->nodes[0]; 22992f38b3e1SArne Jansen 23002f38b3e1SArne Jansen if (find_higher) { 23012f38b3e1SArne Jansen if (p->slots[0] >= btrfs_header_nritems(leaf)) { 23022f38b3e1SArne Jansen ret = btrfs_next_leaf(root, p); 23032f38b3e1SArne Jansen if (ret <= 0) 23042f38b3e1SArne Jansen return ret; 23052f38b3e1SArne Jansen if (!return_any) 23062f38b3e1SArne Jansen return 1; 23072f38b3e1SArne Jansen /* 23082f38b3e1SArne Jansen * no higher item found, return the next 23092f38b3e1SArne Jansen * lower instead 23102f38b3e1SArne Jansen */ 23112f38b3e1SArne Jansen return_any = 0; 23122f38b3e1SArne Jansen find_higher = 0; 23132f38b3e1SArne Jansen btrfs_release_path(p); 23142f38b3e1SArne Jansen goto again; 23152f38b3e1SArne Jansen } 23162f38b3e1SArne Jansen } else { 23172f38b3e1SArne Jansen if (p->slots[0] == 0) { 23182f38b3e1SArne Jansen ret = btrfs_prev_leaf(root, p); 2319e6793769SArne Jansen if (ret < 0) 23202f38b3e1SArne Jansen return ret; 2321e6793769SArne Jansen if (!ret) { 232223c6bf6aSFilipe David Borba Manana leaf = p->nodes[0]; 232323c6bf6aSFilipe David Borba Manana if (p->slots[0] == btrfs_header_nritems(leaf)) 232423c6bf6aSFilipe David Borba Manana p->slots[0]--; 2325e6793769SArne Jansen return 0; 2326e6793769SArne Jansen } 23272f38b3e1SArne Jansen if (!return_any) 23282f38b3e1SArne Jansen return 1; 23292f38b3e1SArne Jansen /* 23302f38b3e1SArne Jansen * no lower item found, return the next 23312f38b3e1SArne Jansen * higher instead 23322f38b3e1SArne Jansen */ 23332f38b3e1SArne Jansen return_any = 0; 23342f38b3e1SArne Jansen find_higher = 1; 23352f38b3e1SArne Jansen btrfs_release_path(p); 23362f38b3e1SArne Jansen goto again; 2337e6793769SArne Jansen } else { 23382f38b3e1SArne Jansen --p->slots[0]; 23392f38b3e1SArne Jansen } 23402f38b3e1SArne Jansen } 23412f38b3e1SArne Jansen return 0; 23422f38b3e1SArne Jansen } 23432f38b3e1SArne Jansen 23442f38b3e1SArne Jansen /* 23450ff40a91SMarcos Paulo de Souza * Execute search and call btrfs_previous_item to traverse backwards if the item 23460ff40a91SMarcos Paulo de Souza * was not found. 23470ff40a91SMarcos Paulo de Souza * 23480ff40a91SMarcos Paulo de Souza * Return 0 if found, 1 if not found and < 0 if error. 23490ff40a91SMarcos Paulo de Souza */ 23500ff40a91SMarcos Paulo de Souza int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key, 23510ff40a91SMarcos Paulo de Souza struct btrfs_path *path) 23520ff40a91SMarcos Paulo de Souza { 23530ff40a91SMarcos Paulo de Souza int ret; 23540ff40a91SMarcos Paulo de Souza 23550ff40a91SMarcos Paulo de Souza ret = btrfs_search_slot(NULL, root, key, path, 0, 0); 23560ff40a91SMarcos Paulo de Souza if (ret > 0) 23570ff40a91SMarcos Paulo de Souza ret = btrfs_previous_item(root, path, key->objectid, key->type); 23580ff40a91SMarcos Paulo de Souza 23590ff40a91SMarcos Paulo de Souza if (ret == 0) 23600ff40a91SMarcos Paulo de Souza btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); 23610ff40a91SMarcos Paulo de Souza 23620ff40a91SMarcos Paulo de Souza return ret; 23630ff40a91SMarcos Paulo de Souza } 23640ff40a91SMarcos Paulo de Souza 236562142be3SGabriel Niebler /** 236662142be3SGabriel Niebler * Search for a valid slot for the given path. 236762142be3SGabriel Niebler * 236862142be3SGabriel Niebler * @root: The root node of the tree. 236962142be3SGabriel Niebler * @key: Will contain a valid item if found. 237062142be3SGabriel Niebler * @path: The starting point to validate the slot. 237162142be3SGabriel Niebler * 237262142be3SGabriel Niebler * Return: 0 if the item is valid 237362142be3SGabriel Niebler * 1 if not found 237462142be3SGabriel Niebler * <0 if error. 237562142be3SGabriel Niebler */ 237662142be3SGabriel Niebler int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key, 237762142be3SGabriel Niebler struct btrfs_path *path) 237862142be3SGabriel Niebler { 237962142be3SGabriel Niebler while (1) { 238062142be3SGabriel Niebler int ret; 238162142be3SGabriel Niebler const int slot = path->slots[0]; 238262142be3SGabriel Niebler const struct extent_buffer *leaf = path->nodes[0]; 238362142be3SGabriel Niebler 238462142be3SGabriel Niebler /* This is where we start walking the path. */ 238562142be3SGabriel Niebler if (slot >= btrfs_header_nritems(leaf)) { 238662142be3SGabriel Niebler /* 238762142be3SGabriel Niebler * If we've reached the last slot in this leaf we need 238862142be3SGabriel Niebler * to go to the next leaf and reset the path. 238962142be3SGabriel Niebler */ 239062142be3SGabriel Niebler ret = btrfs_next_leaf(root, path); 239162142be3SGabriel Niebler if (ret) 239262142be3SGabriel Niebler return ret; 239362142be3SGabriel Niebler continue; 239462142be3SGabriel Niebler } 239562142be3SGabriel Niebler /* Store the found, valid item in @key. */ 239662142be3SGabriel Niebler btrfs_item_key_to_cpu(leaf, key, slot); 239762142be3SGabriel Niebler break; 239862142be3SGabriel Niebler } 239962142be3SGabriel Niebler return 0; 240062142be3SGabriel Niebler } 240162142be3SGabriel Niebler 24020ff40a91SMarcos Paulo de Souza /* 240374123bd7SChris Mason * adjust the pointers going up the tree, starting at level 240474123bd7SChris Mason * making sure the right key of each node is points to 'key'. 240574123bd7SChris Mason * This is used after shifting pointers to the left, so it stops 240674123bd7SChris Mason * fixing up pointers when a given leaf/node is not in slot 0 of the 240774123bd7SChris Mason * higher levels 2408aa5d6bedSChris Mason * 240974123bd7SChris Mason */ 2410b167fa91SNikolay Borisov static void fixup_low_keys(struct btrfs_path *path, 24115f39d397SChris Mason struct btrfs_disk_key *key, int level) 2412be0e5c09SChris Mason { 2413be0e5c09SChris Mason int i; 24145f39d397SChris Mason struct extent_buffer *t; 24150e82bcfeSDavid Sterba int ret; 24165f39d397SChris Mason 2417234b63a0SChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 2418be0e5c09SChris Mason int tslot = path->slots[i]; 24190e82bcfeSDavid Sterba 2420eb60ceacSChris Mason if (!path->nodes[i]) 2421be0e5c09SChris Mason break; 24225f39d397SChris Mason t = path->nodes[i]; 2423f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(t, tslot, 242433cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 24250e82bcfeSDavid Sterba BUG_ON(ret < 0); 24265f39d397SChris Mason btrfs_set_node_key(t, key, tslot); 2427d6025579SChris Mason btrfs_mark_buffer_dirty(path->nodes[i]); 2428be0e5c09SChris Mason if (tslot != 0) 2429be0e5c09SChris Mason break; 2430be0e5c09SChris Mason } 2431be0e5c09SChris Mason } 2432be0e5c09SChris Mason 243374123bd7SChris Mason /* 243431840ae1SZheng Yan * update item key. 243531840ae1SZheng Yan * 243631840ae1SZheng Yan * This function isn't completely safe. It's the caller's responsibility 243731840ae1SZheng Yan * that the new key won't break the order 243831840ae1SZheng Yan */ 2439b7a0365eSDaniel Dressler void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, 2440b7a0365eSDaniel Dressler struct btrfs_path *path, 2441310712b2SOmar Sandoval const struct btrfs_key *new_key) 244231840ae1SZheng Yan { 244331840ae1SZheng Yan struct btrfs_disk_key disk_key; 244431840ae1SZheng Yan struct extent_buffer *eb; 244531840ae1SZheng Yan int slot; 244631840ae1SZheng Yan 244731840ae1SZheng Yan eb = path->nodes[0]; 244831840ae1SZheng Yan slot = path->slots[0]; 244931840ae1SZheng Yan if (slot > 0) { 245031840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot - 1); 24517c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { 24527c15d410SQu Wenruo btrfs_crit(fs_info, 24537c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 24547c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 24557c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 24567c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 24577c15d410SQu Wenruo new_key->objectid, new_key->type, 24587c15d410SQu Wenruo new_key->offset); 24597c15d410SQu Wenruo btrfs_print_leaf(eb); 24607c15d410SQu Wenruo BUG(); 24617c15d410SQu Wenruo } 246231840ae1SZheng Yan } 246331840ae1SZheng Yan if (slot < btrfs_header_nritems(eb) - 1) { 246431840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot + 1); 24657c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { 24667c15d410SQu Wenruo btrfs_crit(fs_info, 24677c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 24687c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 24697c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 24707c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 24717c15d410SQu Wenruo new_key->objectid, new_key->type, 24727c15d410SQu Wenruo new_key->offset); 24737c15d410SQu Wenruo btrfs_print_leaf(eb); 24747c15d410SQu Wenruo BUG(); 24757c15d410SQu Wenruo } 247631840ae1SZheng Yan } 247731840ae1SZheng Yan 247831840ae1SZheng Yan btrfs_cpu_key_to_disk(&disk_key, new_key); 247931840ae1SZheng Yan btrfs_set_item_key(eb, &disk_key, slot); 248031840ae1SZheng Yan btrfs_mark_buffer_dirty(eb); 248131840ae1SZheng Yan if (slot == 0) 2482b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 248331840ae1SZheng Yan } 248431840ae1SZheng Yan 248531840ae1SZheng Yan /* 2486d16c702fSQu Wenruo * Check key order of two sibling extent buffers. 2487d16c702fSQu Wenruo * 2488d16c702fSQu Wenruo * Return true if something is wrong. 2489d16c702fSQu Wenruo * Return false if everything is fine. 2490d16c702fSQu Wenruo * 2491d16c702fSQu Wenruo * Tree-checker only works inside one tree block, thus the following 2492d16c702fSQu Wenruo * corruption can not be detected by tree-checker: 2493d16c702fSQu Wenruo * 2494d16c702fSQu Wenruo * Leaf @left | Leaf @right 2495d16c702fSQu Wenruo * -------------------------------------------------------------- 2496d16c702fSQu Wenruo * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | 2497d16c702fSQu Wenruo * 2498d16c702fSQu Wenruo * Key f6 in leaf @left itself is valid, but not valid when the next 2499d16c702fSQu Wenruo * key in leaf @right is 7. 2500d16c702fSQu Wenruo * This can only be checked at tree block merge time. 2501d16c702fSQu Wenruo * And since tree checker has ensured all key order in each tree block 2502d16c702fSQu Wenruo * is correct, we only need to bother the last key of @left and the first 2503d16c702fSQu Wenruo * key of @right. 2504d16c702fSQu Wenruo */ 2505d16c702fSQu Wenruo static bool check_sibling_keys(struct extent_buffer *left, 2506d16c702fSQu Wenruo struct extent_buffer *right) 2507d16c702fSQu Wenruo { 2508d16c702fSQu Wenruo struct btrfs_key left_last; 2509d16c702fSQu Wenruo struct btrfs_key right_first; 2510d16c702fSQu Wenruo int level = btrfs_header_level(left); 2511d16c702fSQu Wenruo int nr_left = btrfs_header_nritems(left); 2512d16c702fSQu Wenruo int nr_right = btrfs_header_nritems(right); 2513d16c702fSQu Wenruo 2514d16c702fSQu Wenruo /* No key to check in one of the tree blocks */ 2515d16c702fSQu Wenruo if (!nr_left || !nr_right) 2516d16c702fSQu Wenruo return false; 2517d16c702fSQu Wenruo 2518d16c702fSQu Wenruo if (level) { 2519d16c702fSQu Wenruo btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); 2520d16c702fSQu Wenruo btrfs_node_key_to_cpu(right, &right_first, 0); 2521d16c702fSQu Wenruo } else { 2522d16c702fSQu Wenruo btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); 2523d16c702fSQu Wenruo btrfs_item_key_to_cpu(right, &right_first, 0); 2524d16c702fSQu Wenruo } 2525d16c702fSQu Wenruo 2526d16c702fSQu Wenruo if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) { 2527d16c702fSQu Wenruo btrfs_crit(left->fs_info, 2528d16c702fSQu Wenruo "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", 2529d16c702fSQu Wenruo left_last.objectid, left_last.type, 2530d16c702fSQu Wenruo left_last.offset, right_first.objectid, 2531d16c702fSQu Wenruo right_first.type, right_first.offset); 2532d16c702fSQu Wenruo return true; 2533d16c702fSQu Wenruo } 2534d16c702fSQu Wenruo return false; 2535d16c702fSQu Wenruo } 2536d16c702fSQu Wenruo 2537d16c702fSQu Wenruo /* 253874123bd7SChris Mason * try to push data from one node into the next node left in the 253979f95c82SChris Mason * tree. 2540aa5d6bedSChris Mason * 2541aa5d6bedSChris Mason * returns 0 if some ptrs were pushed left, < 0 if there was some horrible 2542aa5d6bedSChris Mason * error, and > 0 if there was no room in the left hand block. 254374123bd7SChris Mason */ 254498ed5174SChris Mason static int push_node_left(struct btrfs_trans_handle *trans, 25452ff7e61eSJeff Mahoney struct extent_buffer *dst, 2546971a1f66SChris Mason struct extent_buffer *src, int empty) 2547be0e5c09SChris Mason { 2548d30a668fSDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 2549be0e5c09SChris Mason int push_items = 0; 2550bb803951SChris Mason int src_nritems; 2551bb803951SChris Mason int dst_nritems; 2552aa5d6bedSChris Mason int ret = 0; 2553be0e5c09SChris Mason 25545f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 25555f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 25560b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 25577bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 25587bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 255954aa1f4dSChris Mason 2560bce4eae9SChris Mason if (!empty && src_nritems <= 8) 2561971a1f66SChris Mason return 1; 2562971a1f66SChris Mason 2563d397712bSChris Mason if (push_items <= 0) 2564be0e5c09SChris Mason return 1; 2565be0e5c09SChris Mason 2566bce4eae9SChris Mason if (empty) { 2567971a1f66SChris Mason push_items = min(src_nritems, push_items); 2568bce4eae9SChris Mason if (push_items < src_nritems) { 2569bce4eae9SChris Mason /* leave at least 8 pointers in the node if 2570bce4eae9SChris Mason * we aren't going to empty it 2571bce4eae9SChris Mason */ 2572bce4eae9SChris Mason if (src_nritems - push_items < 8) { 2573bce4eae9SChris Mason if (push_items <= 8) 2574bce4eae9SChris Mason return 1; 2575bce4eae9SChris Mason push_items -= 8; 2576bce4eae9SChris Mason } 2577bce4eae9SChris Mason } 2578bce4eae9SChris Mason } else 2579bce4eae9SChris Mason push_items = min(src_nritems - 8, push_items); 258079f95c82SChris Mason 2581d16c702fSQu Wenruo /* dst is the left eb, src is the middle eb */ 2582d16c702fSQu Wenruo if (check_sibling_keys(dst, src)) { 2583d16c702fSQu Wenruo ret = -EUCLEAN; 2584d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 2585d16c702fSQu Wenruo return ret; 2586d16c702fSQu Wenruo } 2587f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); 25885de865eeSFilipe David Borba Manana if (ret) { 258966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 25905de865eeSFilipe David Borba Manana return ret; 25915de865eeSFilipe David Borba Manana } 25925f39d397SChris Mason copy_extent_buffer(dst, src, 25935f39d397SChris Mason btrfs_node_key_ptr_offset(dst_nritems), 25945f39d397SChris Mason btrfs_node_key_ptr_offset(0), 2595123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 25965f39d397SChris Mason 2597bb803951SChris Mason if (push_items < src_nritems) { 259857911b8bSJan Schmidt /* 2599f3a84ccdSFilipe Manana * Don't call btrfs_tree_mod_log_insert_move() here, key removal 2600f3a84ccdSFilipe Manana * was already fully logged by btrfs_tree_mod_log_eb_copy() above. 260157911b8bSJan Schmidt */ 26025f39d397SChris Mason memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), 26035f39d397SChris Mason btrfs_node_key_ptr_offset(push_items), 2604e2fa7227SChris Mason (src_nritems - push_items) * 2605123abc88SChris Mason sizeof(struct btrfs_key_ptr)); 2606bb803951SChris Mason } 26075f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 26085f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 26095f39d397SChris Mason btrfs_mark_buffer_dirty(src); 26105f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 261131840ae1SZheng Yan 2612bb803951SChris Mason return ret; 2613be0e5c09SChris Mason } 2614be0e5c09SChris Mason 261597571fd0SChris Mason /* 261679f95c82SChris Mason * try to push data from one node into the next node right in the 261779f95c82SChris Mason * tree. 261879f95c82SChris Mason * 261979f95c82SChris Mason * returns 0 if some ptrs were pushed, < 0 if there was some horrible 262079f95c82SChris Mason * error, and > 0 if there was no room in the right hand block. 262179f95c82SChris Mason * 262279f95c82SChris Mason * this will only push up to 1/2 the contents of the left node over 262379f95c82SChris Mason */ 26245f39d397SChris Mason static int balance_node_right(struct btrfs_trans_handle *trans, 26255f39d397SChris Mason struct extent_buffer *dst, 26265f39d397SChris Mason struct extent_buffer *src) 262779f95c82SChris Mason { 262855d32ed8SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 262979f95c82SChris Mason int push_items = 0; 263079f95c82SChris Mason int max_push; 263179f95c82SChris Mason int src_nritems; 263279f95c82SChris Mason int dst_nritems; 263379f95c82SChris Mason int ret = 0; 263479f95c82SChris Mason 26357bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 26367bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 26377bb86316SChris Mason 26385f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 26395f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 26400b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 2641d397712bSChris Mason if (push_items <= 0) 264279f95c82SChris Mason return 1; 2643bce4eae9SChris Mason 2644d397712bSChris Mason if (src_nritems < 4) 2645bce4eae9SChris Mason return 1; 264679f95c82SChris Mason 264779f95c82SChris Mason max_push = src_nritems / 2 + 1; 264879f95c82SChris Mason /* don't try to empty the node */ 2649d397712bSChris Mason if (max_push >= src_nritems) 265079f95c82SChris Mason return 1; 2651252c38f0SYan 265279f95c82SChris Mason if (max_push < push_items) 265379f95c82SChris Mason push_items = max_push; 265479f95c82SChris Mason 2655d16c702fSQu Wenruo /* dst is the right eb, src is the middle eb */ 2656d16c702fSQu Wenruo if (check_sibling_keys(src, dst)) { 2657d16c702fSQu Wenruo ret = -EUCLEAN; 2658d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 2659d16c702fSQu Wenruo return ret; 2660d16c702fSQu Wenruo } 2661f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); 2662bf1d3425SDavid Sterba BUG_ON(ret < 0); 26635f39d397SChris Mason memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), 26645f39d397SChris Mason btrfs_node_key_ptr_offset(0), 26655f39d397SChris Mason (dst_nritems) * 26665f39d397SChris Mason sizeof(struct btrfs_key_ptr)); 2667d6025579SChris Mason 2668f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, 2669ed874f0dSDavid Sterba push_items); 26705de865eeSFilipe David Borba Manana if (ret) { 267166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 26725de865eeSFilipe David Borba Manana return ret; 26735de865eeSFilipe David Borba Manana } 26745f39d397SChris Mason copy_extent_buffer(dst, src, 26755f39d397SChris Mason btrfs_node_key_ptr_offset(0), 26765f39d397SChris Mason btrfs_node_key_ptr_offset(src_nritems - push_items), 2677123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 267879f95c82SChris Mason 26795f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 26805f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 268179f95c82SChris Mason 26825f39d397SChris Mason btrfs_mark_buffer_dirty(src); 26835f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 268431840ae1SZheng Yan 268579f95c82SChris Mason return ret; 268679f95c82SChris Mason } 268779f95c82SChris Mason 268879f95c82SChris Mason /* 268997571fd0SChris Mason * helper function to insert a new root level in the tree. 269097571fd0SChris Mason * A new node is allocated, and a single item is inserted to 269197571fd0SChris Mason * point to the existing root 2692aa5d6bedSChris Mason * 2693aa5d6bedSChris Mason * returns zero on success or < 0 on failure. 269497571fd0SChris Mason */ 2695d397712bSChris Mason static noinline int insert_new_root(struct btrfs_trans_handle *trans, 26965f39d397SChris Mason struct btrfs_root *root, 2697fdd99c72SLiu Bo struct btrfs_path *path, int level) 269874123bd7SChris Mason { 26990b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 27007bb86316SChris Mason u64 lower_gen; 27015f39d397SChris Mason struct extent_buffer *lower; 27025f39d397SChris Mason struct extent_buffer *c; 2703925baeddSChris Mason struct extent_buffer *old; 27045f39d397SChris Mason struct btrfs_disk_key lower_key; 2705d9d19a01SDavid Sterba int ret; 27065c680ed6SChris Mason 27075c680ed6SChris Mason BUG_ON(path->nodes[level]); 27085c680ed6SChris Mason BUG_ON(path->nodes[level-1] != root->node); 27095c680ed6SChris Mason 27107bb86316SChris Mason lower = path->nodes[level-1]; 27117bb86316SChris Mason if (level == 1) 27127bb86316SChris Mason btrfs_item_key(lower, &lower_key, 0); 27137bb86316SChris Mason else 27147bb86316SChris Mason btrfs_node_key(lower, &lower_key, 0); 27157bb86316SChris Mason 271679bd3712SFilipe Manana c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 271779bd3712SFilipe Manana &lower_key, level, root->node->start, 0, 2718cf6f34aaSJosef Bacik BTRFS_NESTING_NEW_ROOT); 27195f39d397SChris Mason if (IS_ERR(c)) 27205f39d397SChris Mason return PTR_ERR(c); 2721925baeddSChris Mason 27220b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 2723f0486c68SYan, Zheng 27245f39d397SChris Mason btrfs_set_header_nritems(c, 1); 27255f39d397SChris Mason btrfs_set_node_key(c, &lower_key, 0); 2726db94535dSChris Mason btrfs_set_node_blockptr(c, 0, lower->start); 27277bb86316SChris Mason lower_gen = btrfs_header_generation(lower); 272831840ae1SZheng Yan WARN_ON(lower_gen != trans->transid); 27297bb86316SChris Mason 27307bb86316SChris Mason btrfs_set_node_ptr_generation(c, 0, lower_gen); 27315f39d397SChris Mason 27325f39d397SChris Mason btrfs_mark_buffer_dirty(c); 2733d5719762SChris Mason 2734925baeddSChris Mason old = root->node; 2735406808abSFilipe Manana ret = btrfs_tree_mod_log_insert_root(root->node, c, false); 2736d9d19a01SDavid Sterba BUG_ON(ret < 0); 2737240f62c8SChris Mason rcu_assign_pointer(root->node, c); 2738925baeddSChris Mason 2739925baeddSChris Mason /* the super has an extra ref to root->node */ 2740925baeddSChris Mason free_extent_buffer(old); 2741925baeddSChris Mason 27420b86a832SChris Mason add_root_to_dirty_list(root); 274367439dadSDavid Sterba atomic_inc(&c->refs); 27445f39d397SChris Mason path->nodes[level] = c; 2745ac5887c8SJosef Bacik path->locks[level] = BTRFS_WRITE_LOCK; 274674123bd7SChris Mason path->slots[level] = 0; 274774123bd7SChris Mason return 0; 274874123bd7SChris Mason } 27495c680ed6SChris Mason 27505c680ed6SChris Mason /* 27515c680ed6SChris Mason * worker function to insert a single pointer in a node. 27525c680ed6SChris Mason * the node should have enough room for the pointer already 275397571fd0SChris Mason * 27545c680ed6SChris Mason * slot and level indicate where you want the key to go, and 27555c680ed6SChris Mason * blocknr is the block the key points to. 27565c680ed6SChris Mason */ 2757143bede5SJeff Mahoney static void insert_ptr(struct btrfs_trans_handle *trans, 27586ad3cf6dSDavid Sterba struct btrfs_path *path, 2759143bede5SJeff Mahoney struct btrfs_disk_key *key, u64 bytenr, 2760c3e06965SJan Schmidt int slot, int level) 27615c680ed6SChris Mason { 27625f39d397SChris Mason struct extent_buffer *lower; 27635c680ed6SChris Mason int nritems; 2764f3ea38daSJan Schmidt int ret; 27655c680ed6SChris Mason 27665c680ed6SChris Mason BUG_ON(!path->nodes[level]); 276749d0c642SFilipe Manana btrfs_assert_tree_write_locked(path->nodes[level]); 27685f39d397SChris Mason lower = path->nodes[level]; 27695f39d397SChris Mason nritems = btrfs_header_nritems(lower); 2770c293498bSStoyan Gaydarov BUG_ON(slot > nritems); 27716ad3cf6dSDavid Sterba BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); 277274123bd7SChris Mason if (slot != nritems) { 2773bf1d3425SDavid Sterba if (level) { 2774f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_move(lower, slot + 1, 2775f3a84ccdSFilipe Manana slot, nritems - slot); 2776bf1d3425SDavid Sterba BUG_ON(ret < 0); 2777bf1d3425SDavid Sterba } 27785f39d397SChris Mason memmove_extent_buffer(lower, 27795f39d397SChris Mason btrfs_node_key_ptr_offset(slot + 1), 27805f39d397SChris Mason btrfs_node_key_ptr_offset(slot), 2781123abc88SChris Mason (nritems - slot) * sizeof(struct btrfs_key_ptr)); 278274123bd7SChris Mason } 2783c3e06965SJan Schmidt if (level) { 2784f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(lower, slot, 278533cff222SFilipe Manana BTRFS_MOD_LOG_KEY_ADD); 2786f3ea38daSJan Schmidt BUG_ON(ret < 0); 2787f3ea38daSJan Schmidt } 27885f39d397SChris Mason btrfs_set_node_key(lower, key, slot); 2789db94535dSChris Mason btrfs_set_node_blockptr(lower, slot, bytenr); 279074493f7aSChris Mason WARN_ON(trans->transid == 0); 279174493f7aSChris Mason btrfs_set_node_ptr_generation(lower, slot, trans->transid); 27925f39d397SChris Mason btrfs_set_header_nritems(lower, nritems + 1); 27935f39d397SChris Mason btrfs_mark_buffer_dirty(lower); 279474123bd7SChris Mason } 279574123bd7SChris Mason 279697571fd0SChris Mason /* 279797571fd0SChris Mason * split the node at the specified level in path in two. 279897571fd0SChris Mason * The path is corrected to point to the appropriate node after the split 279997571fd0SChris Mason * 280097571fd0SChris Mason * Before splitting this tries to make some room in the node by pushing 280197571fd0SChris Mason * left and right, if either one works, it returns right away. 2802aa5d6bedSChris Mason * 2803aa5d6bedSChris Mason * returns 0 on success and < 0 on failure 280497571fd0SChris Mason */ 2805e02119d5SChris Mason static noinline int split_node(struct btrfs_trans_handle *trans, 2806e02119d5SChris Mason struct btrfs_root *root, 2807e02119d5SChris Mason struct btrfs_path *path, int level) 2808be0e5c09SChris Mason { 28090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 28105f39d397SChris Mason struct extent_buffer *c; 28115f39d397SChris Mason struct extent_buffer *split; 28125f39d397SChris Mason struct btrfs_disk_key disk_key; 2813be0e5c09SChris Mason int mid; 28145c680ed6SChris Mason int ret; 28157518a238SChris Mason u32 c_nritems; 2816be0e5c09SChris Mason 28175f39d397SChris Mason c = path->nodes[level]; 28187bb86316SChris Mason WARN_ON(btrfs_header_generation(c) != trans->transid); 28195f39d397SChris Mason if (c == root->node) { 2820d9abbf1cSJan Schmidt /* 282190f8d62eSJan Schmidt * trying to split the root, lets make a new one 282290f8d62eSJan Schmidt * 2823fdd99c72SLiu Bo * tree mod log: We don't log_removal old root in 282490f8d62eSJan Schmidt * insert_new_root, because that root buffer will be kept as a 282590f8d62eSJan Schmidt * normal node. We are going to log removal of half of the 2826f3a84ccdSFilipe Manana * elements below with btrfs_tree_mod_log_eb_copy(). We're 2827f3a84ccdSFilipe Manana * holding a tree lock on the buffer, which is why we cannot 2828f3a84ccdSFilipe Manana * race with other tree_mod_log users. 2829d9abbf1cSJan Schmidt */ 2830fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, level + 1); 28315c680ed6SChris Mason if (ret) 28325c680ed6SChris Mason return ret; 2833b3612421SChris Mason } else { 2834e66f709bSChris Mason ret = push_nodes_for_insert(trans, root, path, level); 28355f39d397SChris Mason c = path->nodes[level]; 28365f39d397SChris Mason if (!ret && btrfs_header_nritems(c) < 28370b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) 2838e66f709bSChris Mason return 0; 283954aa1f4dSChris Mason if (ret < 0) 284054aa1f4dSChris Mason return ret; 28415c680ed6SChris Mason } 2842e66f709bSChris Mason 28435f39d397SChris Mason c_nritems = btrfs_header_nritems(c); 28445d4f98a2SYan Zheng mid = (c_nritems + 1) / 2; 28455d4f98a2SYan Zheng btrfs_node_key(c, &disk_key, mid); 28467bb86316SChris Mason 284779bd3712SFilipe Manana split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 284879bd3712SFilipe Manana &disk_key, level, c->start, 0, 284979bd3712SFilipe Manana BTRFS_NESTING_SPLIT); 28505f39d397SChris Mason if (IS_ERR(split)) 28515f39d397SChris Mason return PTR_ERR(split); 285254aa1f4dSChris Mason 28530b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 2854bc877d28SNikolay Borisov ASSERT(btrfs_header_level(c) == level); 28555f39d397SChris Mason 2856f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); 28575de865eeSFilipe David Borba Manana if (ret) { 285866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 28595de865eeSFilipe David Borba Manana return ret; 28605de865eeSFilipe David Borba Manana } 28615f39d397SChris Mason copy_extent_buffer(split, c, 28625f39d397SChris Mason btrfs_node_key_ptr_offset(0), 28635f39d397SChris Mason btrfs_node_key_ptr_offset(mid), 2864123abc88SChris Mason (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); 28655f39d397SChris Mason btrfs_set_header_nritems(split, c_nritems - mid); 28665f39d397SChris Mason btrfs_set_header_nritems(c, mid); 2867aa5d6bedSChris Mason 28685f39d397SChris Mason btrfs_mark_buffer_dirty(c); 28695f39d397SChris Mason btrfs_mark_buffer_dirty(split); 28705f39d397SChris Mason 28716ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, split->start, 2872c3e06965SJan Schmidt path->slots[level + 1] + 1, level + 1); 2873aa5d6bedSChris Mason 28745de08d7dSChris Mason if (path->slots[level] >= mid) { 28755c680ed6SChris Mason path->slots[level] -= mid; 2876925baeddSChris Mason btrfs_tree_unlock(c); 28775f39d397SChris Mason free_extent_buffer(c); 28785f39d397SChris Mason path->nodes[level] = split; 28795c680ed6SChris Mason path->slots[level + 1] += 1; 2880eb60ceacSChris Mason } else { 2881925baeddSChris Mason btrfs_tree_unlock(split); 28825f39d397SChris Mason free_extent_buffer(split); 2883be0e5c09SChris Mason } 2884d5286a92SNikolay Borisov return 0; 2885be0e5c09SChris Mason } 2886be0e5c09SChris Mason 288774123bd7SChris Mason /* 288874123bd7SChris Mason * how many bytes are required to store the items in a leaf. start 288974123bd7SChris Mason * and nr indicate which items in the leaf to check. This totals up the 289074123bd7SChris Mason * space used both by the item structs and the item data 289174123bd7SChris Mason */ 28925f39d397SChris Mason static int leaf_space_used(struct extent_buffer *l, int start, int nr) 2893be0e5c09SChris Mason { 2894be0e5c09SChris Mason int data_len; 28955f39d397SChris Mason int nritems = btrfs_header_nritems(l); 2896d4dbff95SChris Mason int end = min(nritems, start + nr) - 1; 2897be0e5c09SChris Mason 2898be0e5c09SChris Mason if (!nr) 2899be0e5c09SChris Mason return 0; 29003212fa14SJosef Bacik data_len = btrfs_item_offset(l, start) + btrfs_item_size(l, start); 29013212fa14SJosef Bacik data_len = data_len - btrfs_item_offset(l, end); 29020783fcfcSChris Mason data_len += sizeof(struct btrfs_item) * nr; 2903d4dbff95SChris Mason WARN_ON(data_len < 0); 2904be0e5c09SChris Mason return data_len; 2905be0e5c09SChris Mason } 2906be0e5c09SChris Mason 290774123bd7SChris Mason /* 2908d4dbff95SChris Mason * The space between the end of the leaf items and 2909d4dbff95SChris Mason * the start of the leaf data. IOW, how much room 2910d4dbff95SChris Mason * the leaf has left for both items and data 2911d4dbff95SChris Mason */ 2912e902baacSDavid Sterba noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) 2913d4dbff95SChris Mason { 2914e902baacSDavid Sterba struct btrfs_fs_info *fs_info = leaf->fs_info; 29155f39d397SChris Mason int nritems = btrfs_header_nritems(leaf); 29165f39d397SChris Mason int ret; 29170b246afaSJeff Mahoney 29180b246afaSJeff Mahoney ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); 29195f39d397SChris Mason if (ret < 0) { 29200b246afaSJeff Mahoney btrfs_crit(fs_info, 2921efe120a0SFrank Holton "leaf free space ret %d, leaf data size %lu, used %d nritems %d", 2922da17066cSJeff Mahoney ret, 29230b246afaSJeff Mahoney (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), 29245f39d397SChris Mason leaf_space_used(leaf, 0, nritems), nritems); 29255f39d397SChris Mason } 29265f39d397SChris Mason return ret; 2927d4dbff95SChris Mason } 2928d4dbff95SChris Mason 292999d8f83cSChris Mason /* 293099d8f83cSChris Mason * min slot controls the lowest index we're willing to push to the 293199d8f83cSChris Mason * right. We'll push up to and including min_slot, but no lower 293299d8f83cSChris Mason */ 2933f72f0010SDavid Sterba static noinline int __push_leaf_right(struct btrfs_path *path, 293444871b1bSChris Mason int data_size, int empty, 293544871b1bSChris Mason struct extent_buffer *right, 293699d8f83cSChris Mason int free_space, u32 left_nritems, 293799d8f83cSChris Mason u32 min_slot) 293800ec4c51SChris Mason { 2939f72f0010SDavid Sterba struct btrfs_fs_info *fs_info = right->fs_info; 29405f39d397SChris Mason struct extent_buffer *left = path->nodes[0]; 294144871b1bSChris Mason struct extent_buffer *upper = path->nodes[1]; 2942cfed81a0SChris Mason struct btrfs_map_token token; 29435f39d397SChris Mason struct btrfs_disk_key disk_key; 294400ec4c51SChris Mason int slot; 294534a38218SChris Mason u32 i; 294600ec4c51SChris Mason int push_space = 0; 294700ec4c51SChris Mason int push_items = 0; 294834a38218SChris Mason u32 nr; 29497518a238SChris Mason u32 right_nritems; 29505f39d397SChris Mason u32 data_end; 2951db94535dSChris Mason u32 this_item_size; 295200ec4c51SChris Mason 295334a38218SChris Mason if (empty) 295434a38218SChris Mason nr = 0; 295534a38218SChris Mason else 295699d8f83cSChris Mason nr = max_t(u32, 1, min_slot); 295734a38218SChris Mason 295831840ae1SZheng Yan if (path->slots[0] >= left_nritems) 295987b29b20SYan Zheng push_space += data_size; 296031840ae1SZheng Yan 296144871b1bSChris Mason slot = path->slots[1]; 296234a38218SChris Mason i = left_nritems - 1; 296334a38218SChris Mason while (i >= nr) { 296431840ae1SZheng Yan if (!empty && push_items > 0) { 296531840ae1SZheng Yan if (path->slots[0] > i) 296631840ae1SZheng Yan break; 296731840ae1SZheng Yan if (path->slots[0] == i) { 2968e902baacSDavid Sterba int space = btrfs_leaf_free_space(left); 2969e902baacSDavid Sterba 297031840ae1SZheng Yan if (space + push_space * 2 > free_space) 297131840ae1SZheng Yan break; 297231840ae1SZheng Yan } 297331840ae1SZheng Yan } 297431840ae1SZheng Yan 297500ec4c51SChris Mason if (path->slots[0] == i) 297687b29b20SYan Zheng push_space += data_size; 2977db94535dSChris Mason 29783212fa14SJosef Bacik this_item_size = btrfs_item_size(left, i); 297974794207SJosef Bacik if (this_item_size + sizeof(struct btrfs_item) + 298074794207SJosef Bacik push_space > free_space) 298100ec4c51SChris Mason break; 298231840ae1SZheng Yan 298300ec4c51SChris Mason push_items++; 298474794207SJosef Bacik push_space += this_item_size + sizeof(struct btrfs_item); 298534a38218SChris Mason if (i == 0) 298634a38218SChris Mason break; 298734a38218SChris Mason i--; 2988db94535dSChris Mason } 29895f39d397SChris Mason 2990925baeddSChris Mason if (push_items == 0) 2991925baeddSChris Mason goto out_unlock; 29925f39d397SChris Mason 29936c1500f2SJulia Lawall WARN_ON(!empty && push_items == left_nritems); 29945f39d397SChris Mason 299500ec4c51SChris Mason /* push left to right */ 29965f39d397SChris Mason right_nritems = btrfs_header_nritems(right); 299734a38218SChris Mason 2998dc2e724eSJosef Bacik push_space = btrfs_item_data_end(left, left_nritems - push_items); 29998f881e8cSDavid Sterba push_space -= leaf_data_end(left); 30005f39d397SChris Mason 300100ec4c51SChris Mason /* make room in the right data area */ 30028f881e8cSDavid Sterba data_end = leaf_data_end(right); 30035f39d397SChris Mason memmove_extent_buffer(right, 30043d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end - push_space, 30053d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end, 30060b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); 30075f39d397SChris Mason 300800ec4c51SChris Mason /* copy from the left data area */ 30093d9ec8c4SNikolay Borisov copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + 30100b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 30118f881e8cSDavid Sterba BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), 3012d6025579SChris Mason push_space); 30135f39d397SChris Mason 30145f39d397SChris Mason memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), 30155f39d397SChris Mason btrfs_item_nr_offset(0), 30160783fcfcSChris Mason right_nritems * sizeof(struct btrfs_item)); 30175f39d397SChris Mason 301800ec4c51SChris Mason /* copy the items from left to right */ 30195f39d397SChris Mason copy_extent_buffer(right, left, btrfs_item_nr_offset(0), 30205f39d397SChris Mason btrfs_item_nr_offset(left_nritems - push_items), 30210783fcfcSChris Mason push_items * sizeof(struct btrfs_item)); 302200ec4c51SChris Mason 302300ec4c51SChris Mason /* update the item pointers */ 3024c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 30257518a238SChris Mason right_nritems += push_items; 30265f39d397SChris Mason btrfs_set_header_nritems(right, right_nritems); 30270b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 30287518a238SChris Mason for (i = 0; i < right_nritems; i++) { 30293212fa14SJosef Bacik push_space -= btrfs_token_item_size(&token, i); 30303212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, push_space); 3031db94535dSChris Mason } 3032db94535dSChris Mason 30337518a238SChris Mason left_nritems -= push_items; 30345f39d397SChris Mason btrfs_set_header_nritems(left, left_nritems); 303500ec4c51SChris Mason 303634a38218SChris Mason if (left_nritems) 30375f39d397SChris Mason btrfs_mark_buffer_dirty(left); 3038f0486c68SYan, Zheng else 30396a884d7dSDavid Sterba btrfs_clean_tree_block(left); 3040f0486c68SYan, Zheng 30415f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3042a429e513SChris Mason 30435f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 30445f39d397SChris Mason btrfs_set_node_key(upper, &disk_key, slot + 1); 3045d6025579SChris Mason btrfs_mark_buffer_dirty(upper); 304602217ed2SChris Mason 304700ec4c51SChris Mason /* then fixup the leaf pointer in the path */ 30487518a238SChris Mason if (path->slots[0] >= left_nritems) { 30497518a238SChris Mason path->slots[0] -= left_nritems; 3050925baeddSChris Mason if (btrfs_header_nritems(path->nodes[0]) == 0) 30516a884d7dSDavid Sterba btrfs_clean_tree_block(path->nodes[0]); 3052925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 30535f39d397SChris Mason free_extent_buffer(path->nodes[0]); 30545f39d397SChris Mason path->nodes[0] = right; 305500ec4c51SChris Mason path->slots[1] += 1; 305600ec4c51SChris Mason } else { 3057925baeddSChris Mason btrfs_tree_unlock(right); 30585f39d397SChris Mason free_extent_buffer(right); 305900ec4c51SChris Mason } 306000ec4c51SChris Mason return 0; 3061925baeddSChris Mason 3062925baeddSChris Mason out_unlock: 3063925baeddSChris Mason btrfs_tree_unlock(right); 3064925baeddSChris Mason free_extent_buffer(right); 3065925baeddSChris Mason return 1; 306600ec4c51SChris Mason } 3067925baeddSChris Mason 306800ec4c51SChris Mason /* 306944871b1bSChris Mason * push some data in the path leaf to the right, trying to free up at 307074123bd7SChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 307144871b1bSChris Mason * 307244871b1bSChris Mason * returns 1 if the push failed because the other node didn't have enough 307344871b1bSChris Mason * room, 0 if everything worked out and < 0 if there were major errors. 307499d8f83cSChris Mason * 307599d8f83cSChris Mason * this will push starting from min_slot to the end of the leaf. It won't 307699d8f83cSChris Mason * push any slot lower than min_slot 307774123bd7SChris Mason */ 307844871b1bSChris Mason static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root 307999d8f83cSChris Mason *root, struct btrfs_path *path, 308099d8f83cSChris Mason int min_data_size, int data_size, 308199d8f83cSChris Mason int empty, u32 min_slot) 3082be0e5c09SChris Mason { 308344871b1bSChris Mason struct extent_buffer *left = path->nodes[0]; 308444871b1bSChris Mason struct extent_buffer *right; 308544871b1bSChris Mason struct extent_buffer *upper; 308644871b1bSChris Mason int slot; 308744871b1bSChris Mason int free_space; 308844871b1bSChris Mason u32 left_nritems; 308944871b1bSChris Mason int ret; 309044871b1bSChris Mason 309144871b1bSChris Mason if (!path->nodes[1]) 309244871b1bSChris Mason return 1; 309344871b1bSChris Mason 309444871b1bSChris Mason slot = path->slots[1]; 309544871b1bSChris Mason upper = path->nodes[1]; 309644871b1bSChris Mason if (slot >= btrfs_header_nritems(upper) - 1) 309744871b1bSChris Mason return 1; 309844871b1bSChris Mason 309949d0c642SFilipe Manana btrfs_assert_tree_write_locked(path->nodes[1]); 310044871b1bSChris Mason 31014b231ae4SDavid Sterba right = btrfs_read_node_slot(upper, slot + 1); 3102fb770ae4SLiu Bo /* 3103fb770ae4SLiu Bo * slot + 1 is not valid or we fail to read the right node, 3104fb770ae4SLiu Bo * no big deal, just return. 3105fb770ae4SLiu Bo */ 3106fb770ae4SLiu Bo if (IS_ERR(right)) 310791ca338dSTsutomu Itoh return 1; 310891ca338dSTsutomu Itoh 3109bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 311044871b1bSChris Mason 3111e902baacSDavid Sterba free_space = btrfs_leaf_free_space(right); 311244871b1bSChris Mason if (free_space < data_size) 311344871b1bSChris Mason goto out_unlock; 311444871b1bSChris Mason 311544871b1bSChris Mason ret = btrfs_cow_block(trans, root, right, upper, 3116bf59a5a2SJosef Bacik slot + 1, &right, BTRFS_NESTING_RIGHT_COW); 311744871b1bSChris Mason if (ret) 311844871b1bSChris Mason goto out_unlock; 311944871b1bSChris Mason 312044871b1bSChris Mason left_nritems = btrfs_header_nritems(left); 312144871b1bSChris Mason if (left_nritems == 0) 312244871b1bSChris Mason goto out_unlock; 312344871b1bSChris Mason 3124d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3125d16c702fSQu Wenruo ret = -EUCLEAN; 3126d16c702fSQu Wenruo btrfs_tree_unlock(right); 3127d16c702fSQu Wenruo free_extent_buffer(right); 3128d16c702fSQu Wenruo return ret; 3129d16c702fSQu Wenruo } 31302ef1fed2SFilipe David Borba Manana if (path->slots[0] == left_nritems && !empty) { 31312ef1fed2SFilipe David Borba Manana /* Key greater than all keys in the leaf, right neighbor has 31322ef1fed2SFilipe David Borba Manana * enough room for it and we're not emptying our leaf to delete 31332ef1fed2SFilipe David Borba Manana * it, therefore use right neighbor to insert the new item and 313452042d8eSAndrea Gelmini * no need to touch/dirty our left leaf. */ 31352ef1fed2SFilipe David Borba Manana btrfs_tree_unlock(left); 31362ef1fed2SFilipe David Borba Manana free_extent_buffer(left); 31372ef1fed2SFilipe David Borba Manana path->nodes[0] = right; 31382ef1fed2SFilipe David Borba Manana path->slots[0] = 0; 31392ef1fed2SFilipe David Borba Manana path->slots[1]++; 31402ef1fed2SFilipe David Borba Manana return 0; 31412ef1fed2SFilipe David Borba Manana } 31422ef1fed2SFilipe David Borba Manana 3143f72f0010SDavid Sterba return __push_leaf_right(path, min_data_size, empty, 314499d8f83cSChris Mason right, free_space, left_nritems, min_slot); 314544871b1bSChris Mason out_unlock: 314644871b1bSChris Mason btrfs_tree_unlock(right); 314744871b1bSChris Mason free_extent_buffer(right); 314844871b1bSChris Mason return 1; 314944871b1bSChris Mason } 315044871b1bSChris Mason 315144871b1bSChris Mason /* 315244871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 315344871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 315499d8f83cSChris Mason * 315599d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 315699d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the 315799d8f83cSChris Mason * items 315844871b1bSChris Mason */ 31598087c193SDavid Sterba static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, 316044871b1bSChris Mason int empty, struct extent_buffer *left, 316199d8f83cSChris Mason int free_space, u32 right_nritems, 316299d8f83cSChris Mason u32 max_slot) 316344871b1bSChris Mason { 31648087c193SDavid Sterba struct btrfs_fs_info *fs_info = left->fs_info; 31655f39d397SChris Mason struct btrfs_disk_key disk_key; 31665f39d397SChris Mason struct extent_buffer *right = path->nodes[0]; 3167be0e5c09SChris Mason int i; 3168be0e5c09SChris Mason int push_space = 0; 3169be0e5c09SChris Mason int push_items = 0; 31707518a238SChris Mason u32 old_left_nritems; 317134a38218SChris Mason u32 nr; 3172aa5d6bedSChris Mason int ret = 0; 3173db94535dSChris Mason u32 this_item_size; 3174db94535dSChris Mason u32 old_left_item_size; 3175cfed81a0SChris Mason struct btrfs_map_token token; 3176cfed81a0SChris Mason 317734a38218SChris Mason if (empty) 317899d8f83cSChris Mason nr = min(right_nritems, max_slot); 317934a38218SChris Mason else 318099d8f83cSChris Mason nr = min(right_nritems - 1, max_slot); 318134a38218SChris Mason 318234a38218SChris Mason for (i = 0; i < nr; i++) { 318331840ae1SZheng Yan if (!empty && push_items > 0) { 318431840ae1SZheng Yan if (path->slots[0] < i) 318531840ae1SZheng Yan break; 318631840ae1SZheng Yan if (path->slots[0] == i) { 3187e902baacSDavid Sterba int space = btrfs_leaf_free_space(right); 3188e902baacSDavid Sterba 318931840ae1SZheng Yan if (space + push_space * 2 > free_space) 319031840ae1SZheng Yan break; 319131840ae1SZheng Yan } 319231840ae1SZheng Yan } 319331840ae1SZheng Yan 3194be0e5c09SChris Mason if (path->slots[0] == i) 319587b29b20SYan Zheng push_space += data_size; 3196db94535dSChris Mason 31973212fa14SJosef Bacik this_item_size = btrfs_item_size(right, i); 319874794207SJosef Bacik if (this_item_size + sizeof(struct btrfs_item) + push_space > 319974794207SJosef Bacik free_space) 3200be0e5c09SChris Mason break; 3201db94535dSChris Mason 3202be0e5c09SChris Mason push_items++; 320374794207SJosef Bacik push_space += this_item_size + sizeof(struct btrfs_item); 3204be0e5c09SChris Mason } 3205db94535dSChris Mason 3206be0e5c09SChris Mason if (push_items == 0) { 3207925baeddSChris Mason ret = 1; 3208925baeddSChris Mason goto out; 3209be0e5c09SChris Mason } 3210fae7f21cSDulshani Gunawardhana WARN_ON(!empty && push_items == btrfs_header_nritems(right)); 32115f39d397SChris Mason 3212be0e5c09SChris Mason /* push data from right to left */ 32135f39d397SChris Mason copy_extent_buffer(left, right, 32145f39d397SChris Mason btrfs_item_nr_offset(btrfs_header_nritems(left)), 32155f39d397SChris Mason btrfs_item_nr_offset(0), 32165f39d397SChris Mason push_items * sizeof(struct btrfs_item)); 32175f39d397SChris Mason 32180b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - 32193212fa14SJosef Bacik btrfs_item_offset(right, push_items - 1); 32205f39d397SChris Mason 32213d9ec8c4SNikolay Borisov copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + 32228f881e8cSDavid Sterba leaf_data_end(left) - push_space, 32233d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + 32243212fa14SJosef Bacik btrfs_item_offset(right, push_items - 1), 3225be0e5c09SChris Mason push_space); 32265f39d397SChris Mason old_left_nritems = btrfs_header_nritems(left); 322787b29b20SYan Zheng BUG_ON(old_left_nritems <= 0); 3228eb60ceacSChris Mason 3229c82f823cSDavid Sterba btrfs_init_map_token(&token, left); 32303212fa14SJosef Bacik old_left_item_size = btrfs_item_offset(left, old_left_nritems - 1); 3231be0e5c09SChris Mason for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { 32325f39d397SChris Mason u32 ioff; 3233db94535dSChris Mason 32343212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 32353212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, 3236cc4c13d5SDavid Sterba ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); 3237be0e5c09SChris Mason } 32385f39d397SChris Mason btrfs_set_header_nritems(left, old_left_nritems + push_items); 3239be0e5c09SChris Mason 3240be0e5c09SChris Mason /* fixup right node */ 324131b1a2bdSJulia Lawall if (push_items > right_nritems) 324231b1a2bdSJulia Lawall WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, 3243d397712bSChris Mason right_nritems); 324434a38218SChris Mason 324534a38218SChris Mason if (push_items < right_nritems) { 32463212fa14SJosef Bacik push_space = btrfs_item_offset(right, push_items - 1) - 32478f881e8cSDavid Sterba leaf_data_end(right); 32483d9ec8c4SNikolay Borisov memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + 32490b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 32503d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + 32518f881e8cSDavid Sterba leaf_data_end(right), push_space); 32525f39d397SChris Mason 32535f39d397SChris Mason memmove_extent_buffer(right, btrfs_item_nr_offset(0), 32545f39d397SChris Mason btrfs_item_nr_offset(push_items), 32555f39d397SChris Mason (btrfs_header_nritems(right) - push_items) * 32560783fcfcSChris Mason sizeof(struct btrfs_item)); 325734a38218SChris Mason } 3258c82f823cSDavid Sterba 3259c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 3260eef1c494SYan right_nritems -= push_items; 3261eef1c494SYan btrfs_set_header_nritems(right, right_nritems); 32620b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 32635f39d397SChris Mason for (i = 0; i < right_nritems; i++) { 32643212fa14SJosef Bacik push_space = push_space - btrfs_token_item_size(&token, i); 32653212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, push_space); 3266db94535dSChris Mason } 3267eb60ceacSChris Mason 32685f39d397SChris Mason btrfs_mark_buffer_dirty(left); 326934a38218SChris Mason if (right_nritems) 32705f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3271f0486c68SYan, Zheng else 32726a884d7dSDavid Sterba btrfs_clean_tree_block(right); 3273098f59c2SChris Mason 32745f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 3275b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 3276be0e5c09SChris Mason 3277be0e5c09SChris Mason /* then fixup the leaf pointer in the path */ 3278be0e5c09SChris Mason if (path->slots[0] < push_items) { 3279be0e5c09SChris Mason path->slots[0] += old_left_nritems; 3280925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 32815f39d397SChris Mason free_extent_buffer(path->nodes[0]); 32825f39d397SChris Mason path->nodes[0] = left; 3283be0e5c09SChris Mason path->slots[1] -= 1; 3284be0e5c09SChris Mason } else { 3285925baeddSChris Mason btrfs_tree_unlock(left); 32865f39d397SChris Mason free_extent_buffer(left); 3287be0e5c09SChris Mason path->slots[0] -= push_items; 3288be0e5c09SChris Mason } 3289eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 3290aa5d6bedSChris Mason return ret; 3291925baeddSChris Mason out: 3292925baeddSChris Mason btrfs_tree_unlock(left); 3293925baeddSChris Mason free_extent_buffer(left); 3294925baeddSChris Mason return ret; 3295be0e5c09SChris Mason } 3296be0e5c09SChris Mason 329774123bd7SChris Mason /* 329844871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 329944871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 330099d8f83cSChris Mason * 330199d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 330299d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the 330399d8f83cSChris Mason * items 330444871b1bSChris Mason */ 330544871b1bSChris Mason static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root 330699d8f83cSChris Mason *root, struct btrfs_path *path, int min_data_size, 330799d8f83cSChris Mason int data_size, int empty, u32 max_slot) 330844871b1bSChris Mason { 330944871b1bSChris Mason struct extent_buffer *right = path->nodes[0]; 331044871b1bSChris Mason struct extent_buffer *left; 331144871b1bSChris Mason int slot; 331244871b1bSChris Mason int free_space; 331344871b1bSChris Mason u32 right_nritems; 331444871b1bSChris Mason int ret = 0; 331544871b1bSChris Mason 331644871b1bSChris Mason slot = path->slots[1]; 331744871b1bSChris Mason if (slot == 0) 331844871b1bSChris Mason return 1; 331944871b1bSChris Mason if (!path->nodes[1]) 332044871b1bSChris Mason return 1; 332144871b1bSChris Mason 332244871b1bSChris Mason right_nritems = btrfs_header_nritems(right); 332344871b1bSChris Mason if (right_nritems == 0) 332444871b1bSChris Mason return 1; 332544871b1bSChris Mason 332649d0c642SFilipe Manana btrfs_assert_tree_write_locked(path->nodes[1]); 332744871b1bSChris Mason 33284b231ae4SDavid Sterba left = btrfs_read_node_slot(path->nodes[1], slot - 1); 3329fb770ae4SLiu Bo /* 3330fb770ae4SLiu Bo * slot - 1 is not valid or we fail to read the left node, 3331fb770ae4SLiu Bo * no big deal, just return. 3332fb770ae4SLiu Bo */ 3333fb770ae4SLiu Bo if (IS_ERR(left)) 333491ca338dSTsutomu Itoh return 1; 333591ca338dSTsutomu Itoh 3336bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 333744871b1bSChris Mason 3338e902baacSDavid Sterba free_space = btrfs_leaf_free_space(left); 333944871b1bSChris Mason if (free_space < data_size) { 334044871b1bSChris Mason ret = 1; 334144871b1bSChris Mason goto out; 334244871b1bSChris Mason } 334344871b1bSChris Mason 334444871b1bSChris Mason ret = btrfs_cow_block(trans, root, left, 33459631e4ccSJosef Bacik path->nodes[1], slot - 1, &left, 3346bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 334744871b1bSChris Mason if (ret) { 334844871b1bSChris Mason /* we hit -ENOSPC, but it isn't fatal here */ 334979787eaaSJeff Mahoney if (ret == -ENOSPC) 335044871b1bSChris Mason ret = 1; 335144871b1bSChris Mason goto out; 335244871b1bSChris Mason } 335344871b1bSChris Mason 3354d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3355d16c702fSQu Wenruo ret = -EUCLEAN; 3356d16c702fSQu Wenruo goto out; 3357d16c702fSQu Wenruo } 33588087c193SDavid Sterba return __push_leaf_left(path, min_data_size, 335999d8f83cSChris Mason empty, left, free_space, right_nritems, 336099d8f83cSChris Mason max_slot); 336144871b1bSChris Mason out: 336244871b1bSChris Mason btrfs_tree_unlock(left); 336344871b1bSChris Mason free_extent_buffer(left); 336444871b1bSChris Mason return ret; 336544871b1bSChris Mason } 336644871b1bSChris Mason 336744871b1bSChris Mason /* 336874123bd7SChris Mason * split the path's leaf in two, making sure there is at least data_size 336974123bd7SChris Mason * available for the resulting leaf level of the path. 337074123bd7SChris Mason */ 3371143bede5SJeff Mahoney static noinline void copy_for_split(struct btrfs_trans_handle *trans, 337244871b1bSChris Mason struct btrfs_path *path, 337344871b1bSChris Mason struct extent_buffer *l, 337444871b1bSChris Mason struct extent_buffer *right, 337544871b1bSChris Mason int slot, int mid, int nritems) 3376be0e5c09SChris Mason { 337794f94ad9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 3378be0e5c09SChris Mason int data_copy_size; 3379be0e5c09SChris Mason int rt_data_off; 3380be0e5c09SChris Mason int i; 3381d4dbff95SChris Mason struct btrfs_disk_key disk_key; 3382cfed81a0SChris Mason struct btrfs_map_token token; 3383cfed81a0SChris Mason 33845f39d397SChris Mason nritems = nritems - mid; 33855f39d397SChris Mason btrfs_set_header_nritems(right, nritems); 3386dc2e724eSJosef Bacik data_copy_size = btrfs_item_data_end(l, mid) - leaf_data_end(l); 33875f39d397SChris Mason 33885f39d397SChris Mason copy_extent_buffer(right, l, btrfs_item_nr_offset(0), 33895f39d397SChris Mason btrfs_item_nr_offset(mid), 33905f39d397SChris Mason nritems * sizeof(struct btrfs_item)); 33915f39d397SChris Mason 33925f39d397SChris Mason copy_extent_buffer(right, l, 33933d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - 33943d9ec8c4SNikolay Borisov data_copy_size, BTRFS_LEAF_DATA_OFFSET + 33958f881e8cSDavid Sterba leaf_data_end(l), data_copy_size); 339674123bd7SChris Mason 3397dc2e724eSJosef Bacik rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_data_end(l, mid); 33985f39d397SChris Mason 3399c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 34005f39d397SChris Mason for (i = 0; i < nritems; i++) { 3401db94535dSChris Mason u32 ioff; 3402db94535dSChris Mason 34033212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 34043212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff + rt_data_off); 34050783fcfcSChris Mason } 340674123bd7SChris Mason 34075f39d397SChris Mason btrfs_set_header_nritems(l, mid); 34085f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 34096ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); 34105f39d397SChris Mason 34115f39d397SChris Mason btrfs_mark_buffer_dirty(right); 34125f39d397SChris Mason btrfs_mark_buffer_dirty(l); 3413eb60ceacSChris Mason BUG_ON(path->slots[0] != slot); 34145f39d397SChris Mason 3415be0e5c09SChris Mason if (mid <= slot) { 3416925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 34175f39d397SChris Mason free_extent_buffer(path->nodes[0]); 34185f39d397SChris Mason path->nodes[0] = right; 3419be0e5c09SChris Mason path->slots[0] -= mid; 3420be0e5c09SChris Mason path->slots[1] += 1; 3421925baeddSChris Mason } else { 3422925baeddSChris Mason btrfs_tree_unlock(right); 34235f39d397SChris Mason free_extent_buffer(right); 3424925baeddSChris Mason } 34255f39d397SChris Mason 3426eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 342744871b1bSChris Mason } 342844871b1bSChris Mason 342944871b1bSChris Mason /* 343099d8f83cSChris Mason * double splits happen when we need to insert a big item in the middle 343199d8f83cSChris Mason * of a leaf. A double split can leave us with 3 mostly empty leaves: 343299d8f83cSChris Mason * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] 343399d8f83cSChris Mason * A B C 343499d8f83cSChris Mason * 343599d8f83cSChris Mason * We avoid this by trying to push the items on either side of our target 343699d8f83cSChris Mason * into the adjacent leaves. If all goes well we can avoid the double split 343799d8f83cSChris Mason * completely. 343899d8f83cSChris Mason */ 343999d8f83cSChris Mason static noinline int push_for_double_split(struct btrfs_trans_handle *trans, 344099d8f83cSChris Mason struct btrfs_root *root, 344199d8f83cSChris Mason struct btrfs_path *path, 344299d8f83cSChris Mason int data_size) 344399d8f83cSChris Mason { 344499d8f83cSChris Mason int ret; 344599d8f83cSChris Mason int progress = 0; 344699d8f83cSChris Mason int slot; 344799d8f83cSChris Mason u32 nritems; 34485a4267caSFilipe David Borba Manana int space_needed = data_size; 344999d8f83cSChris Mason 345099d8f83cSChris Mason slot = path->slots[0]; 34515a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(path->nodes[0])) 3452e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 345399d8f83cSChris Mason 345499d8f83cSChris Mason /* 345599d8f83cSChris Mason * try to push all the items after our slot into the 345699d8f83cSChris Mason * right leaf 345799d8f83cSChris Mason */ 34585a4267caSFilipe David Borba Manana ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); 345999d8f83cSChris Mason if (ret < 0) 346099d8f83cSChris Mason return ret; 346199d8f83cSChris Mason 346299d8f83cSChris Mason if (ret == 0) 346399d8f83cSChris Mason progress++; 346499d8f83cSChris Mason 346599d8f83cSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 346699d8f83cSChris Mason /* 346799d8f83cSChris Mason * our goal is to get our slot at the start or end of a leaf. If 346899d8f83cSChris Mason * we've done so we're done 346999d8f83cSChris Mason */ 347099d8f83cSChris Mason if (path->slots[0] == 0 || path->slots[0] == nritems) 347199d8f83cSChris Mason return 0; 347299d8f83cSChris Mason 3473e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 347499d8f83cSChris Mason return 0; 347599d8f83cSChris Mason 347699d8f83cSChris Mason /* try to push all the items before our slot into the next leaf */ 347799d8f83cSChris Mason slot = path->slots[0]; 3478263d3995SFilipe Manana space_needed = data_size; 3479263d3995SFilipe Manana if (slot > 0) 3480e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 34815a4267caSFilipe David Borba Manana ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); 348299d8f83cSChris Mason if (ret < 0) 348399d8f83cSChris Mason return ret; 348499d8f83cSChris Mason 348599d8f83cSChris Mason if (ret == 0) 348699d8f83cSChris Mason progress++; 348799d8f83cSChris Mason 348899d8f83cSChris Mason if (progress) 348999d8f83cSChris Mason return 0; 349099d8f83cSChris Mason return 1; 349199d8f83cSChris Mason } 349299d8f83cSChris Mason 349399d8f83cSChris Mason /* 349444871b1bSChris Mason * split the path's leaf in two, making sure there is at least data_size 349544871b1bSChris Mason * available for the resulting leaf level of the path. 349644871b1bSChris Mason * 349744871b1bSChris Mason * returns 0 if all went well and < 0 on failure. 349844871b1bSChris Mason */ 349944871b1bSChris Mason static noinline int split_leaf(struct btrfs_trans_handle *trans, 350044871b1bSChris Mason struct btrfs_root *root, 3501310712b2SOmar Sandoval const struct btrfs_key *ins_key, 350244871b1bSChris Mason struct btrfs_path *path, int data_size, 350344871b1bSChris Mason int extend) 350444871b1bSChris Mason { 35055d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 350644871b1bSChris Mason struct extent_buffer *l; 350744871b1bSChris Mason u32 nritems; 350844871b1bSChris Mason int mid; 350944871b1bSChris Mason int slot; 351044871b1bSChris Mason struct extent_buffer *right; 3511b7a0365eSDaniel Dressler struct btrfs_fs_info *fs_info = root->fs_info; 351244871b1bSChris Mason int ret = 0; 351344871b1bSChris Mason int wret; 35145d4f98a2SYan Zheng int split; 351544871b1bSChris Mason int num_doubles = 0; 351699d8f83cSChris Mason int tried_avoid_double = 0; 351744871b1bSChris Mason 3518a5719521SYan, Zheng l = path->nodes[0]; 3519a5719521SYan, Zheng slot = path->slots[0]; 35203212fa14SJosef Bacik if (extend && data_size + btrfs_item_size(l, slot) + 35210b246afaSJeff Mahoney sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) 3522a5719521SYan, Zheng return -EOVERFLOW; 3523a5719521SYan, Zheng 352444871b1bSChris Mason /* first try to make some room by pushing left and right */ 352533157e05SLiu Bo if (data_size && path->nodes[1]) { 35265a4267caSFilipe David Borba Manana int space_needed = data_size; 35275a4267caSFilipe David Borba Manana 35285a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(l)) 3529e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 35305a4267caSFilipe David Borba Manana 35315a4267caSFilipe David Borba Manana wret = push_leaf_right(trans, root, path, space_needed, 35325a4267caSFilipe David Borba Manana space_needed, 0, 0); 353344871b1bSChris Mason if (wret < 0) 353444871b1bSChris Mason return wret; 353544871b1bSChris Mason if (wret) { 3536263d3995SFilipe Manana space_needed = data_size; 3537263d3995SFilipe Manana if (slot > 0) 3538e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 35395a4267caSFilipe David Borba Manana wret = push_leaf_left(trans, root, path, space_needed, 35405a4267caSFilipe David Borba Manana space_needed, 0, (u32)-1); 354144871b1bSChris Mason if (wret < 0) 354244871b1bSChris Mason return wret; 354344871b1bSChris Mason } 354444871b1bSChris Mason l = path->nodes[0]; 354544871b1bSChris Mason 354644871b1bSChris Mason /* did the pushes work? */ 3547e902baacSDavid Sterba if (btrfs_leaf_free_space(l) >= data_size) 354844871b1bSChris Mason return 0; 354944871b1bSChris Mason } 355044871b1bSChris Mason 355144871b1bSChris Mason if (!path->nodes[1]) { 3552fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, 1); 355344871b1bSChris Mason if (ret) 355444871b1bSChris Mason return ret; 355544871b1bSChris Mason } 355644871b1bSChris Mason again: 35575d4f98a2SYan Zheng split = 1; 355844871b1bSChris Mason l = path->nodes[0]; 355944871b1bSChris Mason slot = path->slots[0]; 356044871b1bSChris Mason nritems = btrfs_header_nritems(l); 356144871b1bSChris Mason mid = (nritems + 1) / 2; 356244871b1bSChris Mason 35635d4f98a2SYan Zheng if (mid <= slot) { 35645d4f98a2SYan Zheng if (nritems == 1 || 35655d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + data_size > 35660b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 35675d4f98a2SYan Zheng if (slot >= nritems) { 35685d4f98a2SYan Zheng split = 0; 35695d4f98a2SYan Zheng } else { 35705d4f98a2SYan Zheng mid = slot; 35715d4f98a2SYan Zheng if (mid != nritems && 35725d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 35730b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 357499d8f83cSChris Mason if (data_size && !tried_avoid_double) 357599d8f83cSChris Mason goto push_for_double; 35765d4f98a2SYan Zheng split = 2; 35775d4f98a2SYan Zheng } 35785d4f98a2SYan Zheng } 35795d4f98a2SYan Zheng } 35805d4f98a2SYan Zheng } else { 35815d4f98a2SYan Zheng if (leaf_space_used(l, 0, mid) + data_size > 35820b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 35835d4f98a2SYan Zheng if (!extend && data_size && slot == 0) { 35845d4f98a2SYan Zheng split = 0; 35855d4f98a2SYan Zheng } else if ((extend || !data_size) && slot == 0) { 35865d4f98a2SYan Zheng mid = 1; 35875d4f98a2SYan Zheng } else { 35885d4f98a2SYan Zheng mid = slot; 35895d4f98a2SYan Zheng if (mid != nritems && 35905d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 35910b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 359299d8f83cSChris Mason if (data_size && !tried_avoid_double) 359399d8f83cSChris Mason goto push_for_double; 35945d4f98a2SYan Zheng split = 2; 35955d4f98a2SYan Zheng } 35965d4f98a2SYan Zheng } 35975d4f98a2SYan Zheng } 35985d4f98a2SYan Zheng } 35995d4f98a2SYan Zheng 36005d4f98a2SYan Zheng if (split == 0) 36015d4f98a2SYan Zheng btrfs_cpu_key_to_disk(&disk_key, ins_key); 36025d4f98a2SYan Zheng else 36035d4f98a2SYan Zheng btrfs_item_key(l, &disk_key, mid); 36045d4f98a2SYan Zheng 3605ca9d473aSJosef Bacik /* 3606ca9d473aSJosef Bacik * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double 3607ca9d473aSJosef Bacik * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES 3608ca9d473aSJosef Bacik * subclasses, which is 8 at the time of this patch, and we've maxed it 3609ca9d473aSJosef Bacik * out. In the future we could add a 3610ca9d473aSJosef Bacik * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just 3611ca9d473aSJosef Bacik * use BTRFS_NESTING_NEW_ROOT. 3612ca9d473aSJosef Bacik */ 361379bd3712SFilipe Manana right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 361479bd3712SFilipe Manana &disk_key, 0, l->start, 0, 361579bd3712SFilipe Manana num_doubles ? BTRFS_NESTING_NEW_ROOT : 3616ca9d473aSJosef Bacik BTRFS_NESTING_SPLIT); 3617f0486c68SYan, Zheng if (IS_ERR(right)) 361844871b1bSChris Mason return PTR_ERR(right); 3619f0486c68SYan, Zheng 36200b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 362144871b1bSChris Mason 36225d4f98a2SYan Zheng if (split == 0) { 362344871b1bSChris Mason if (mid <= slot) { 362444871b1bSChris Mason btrfs_set_header_nritems(right, 0); 36256ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, 36262ff7e61eSJeff Mahoney right->start, path->slots[1] + 1, 1); 362744871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 362844871b1bSChris Mason free_extent_buffer(path->nodes[0]); 362944871b1bSChris Mason path->nodes[0] = right; 363044871b1bSChris Mason path->slots[0] = 0; 363144871b1bSChris Mason path->slots[1] += 1; 363244871b1bSChris Mason } else { 363344871b1bSChris Mason btrfs_set_header_nritems(right, 0); 36346ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, 36352ff7e61eSJeff Mahoney right->start, path->slots[1], 1); 363644871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 363744871b1bSChris Mason free_extent_buffer(path->nodes[0]); 363844871b1bSChris Mason path->nodes[0] = right; 363944871b1bSChris Mason path->slots[0] = 0; 3640143bede5SJeff Mahoney if (path->slots[1] == 0) 3641b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 36425d4f98a2SYan Zheng } 3643196e0249SLiu Bo /* 3644196e0249SLiu Bo * We create a new leaf 'right' for the required ins_len and 3645196e0249SLiu Bo * we'll do btrfs_mark_buffer_dirty() on this leaf after copying 3646196e0249SLiu Bo * the content of ins_len to 'right'. 3647196e0249SLiu Bo */ 364844871b1bSChris Mason return ret; 364944871b1bSChris Mason } 365044871b1bSChris Mason 365194f94ad9SDavid Sterba copy_for_split(trans, path, l, right, slot, mid, nritems); 365244871b1bSChris Mason 36535d4f98a2SYan Zheng if (split == 2) { 3654cc0c5538SChris Mason BUG_ON(num_doubles != 0); 3655cc0c5538SChris Mason num_doubles++; 3656cc0c5538SChris Mason goto again; 36573326d1b0SChris Mason } 365844871b1bSChris Mason 3659143bede5SJeff Mahoney return 0; 366099d8f83cSChris Mason 366199d8f83cSChris Mason push_for_double: 366299d8f83cSChris Mason push_for_double_split(trans, root, path, data_size); 366399d8f83cSChris Mason tried_avoid_double = 1; 3664e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 366599d8f83cSChris Mason return 0; 366699d8f83cSChris Mason goto again; 3667be0e5c09SChris Mason } 3668be0e5c09SChris Mason 3669ad48fd75SYan, Zheng static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, 3670ad48fd75SYan, Zheng struct btrfs_root *root, 3671ad48fd75SYan, Zheng struct btrfs_path *path, int ins_len) 3672ad48fd75SYan, Zheng { 3673ad48fd75SYan, Zheng struct btrfs_key key; 3674ad48fd75SYan, Zheng struct extent_buffer *leaf; 3675ad48fd75SYan, Zheng struct btrfs_file_extent_item *fi; 3676ad48fd75SYan, Zheng u64 extent_len = 0; 3677ad48fd75SYan, Zheng u32 item_size; 3678ad48fd75SYan, Zheng int ret; 3679ad48fd75SYan, Zheng 3680ad48fd75SYan, Zheng leaf = path->nodes[0]; 3681ad48fd75SYan, Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3682ad48fd75SYan, Zheng 3683ad48fd75SYan, Zheng BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && 3684ad48fd75SYan, Zheng key.type != BTRFS_EXTENT_CSUM_KEY); 3685ad48fd75SYan, Zheng 3686e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) >= ins_len) 3687ad48fd75SYan, Zheng return 0; 3688ad48fd75SYan, Zheng 36893212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 3690ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 3691ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 3692ad48fd75SYan, Zheng struct btrfs_file_extent_item); 3693ad48fd75SYan, Zheng extent_len = btrfs_file_extent_num_bytes(leaf, fi); 3694ad48fd75SYan, Zheng } 3695b3b4aa74SDavid Sterba btrfs_release_path(path); 3696ad48fd75SYan, Zheng 3697ad48fd75SYan, Zheng path->keep_locks = 1; 3698ad48fd75SYan, Zheng path->search_for_split = 1; 3699ad48fd75SYan, Zheng ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 3700ad48fd75SYan, Zheng path->search_for_split = 0; 3701a8df6fe6SFilipe Manana if (ret > 0) 3702a8df6fe6SFilipe Manana ret = -EAGAIN; 3703ad48fd75SYan, Zheng if (ret < 0) 3704ad48fd75SYan, Zheng goto err; 3705ad48fd75SYan, Zheng 3706ad48fd75SYan, Zheng ret = -EAGAIN; 3707ad48fd75SYan, Zheng leaf = path->nodes[0]; 3708a8df6fe6SFilipe Manana /* if our item isn't there, return now */ 37093212fa14SJosef Bacik if (item_size != btrfs_item_size(leaf, path->slots[0])) 3710ad48fd75SYan, Zheng goto err; 3711ad48fd75SYan, Zheng 3712109f6aefSChris Mason /* the leaf has changed, it now has room. return now */ 3713e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) 3714109f6aefSChris Mason goto err; 3715109f6aefSChris Mason 3716ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 3717ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 3718ad48fd75SYan, Zheng struct btrfs_file_extent_item); 3719ad48fd75SYan, Zheng if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) 3720ad48fd75SYan, Zheng goto err; 3721ad48fd75SYan, Zheng } 3722ad48fd75SYan, Zheng 3723ad48fd75SYan, Zheng ret = split_leaf(trans, root, &key, path, ins_len, 1); 3724f0486c68SYan, Zheng if (ret) 3725f0486c68SYan, Zheng goto err; 3726ad48fd75SYan, Zheng 3727ad48fd75SYan, Zheng path->keep_locks = 0; 3728ad48fd75SYan, Zheng btrfs_unlock_up_safe(path, 1); 3729ad48fd75SYan, Zheng return 0; 3730ad48fd75SYan, Zheng err: 3731ad48fd75SYan, Zheng path->keep_locks = 0; 3732ad48fd75SYan, Zheng return ret; 3733ad48fd75SYan, Zheng } 3734ad48fd75SYan, Zheng 373525263cd7SDavid Sterba static noinline int split_item(struct btrfs_path *path, 3736310712b2SOmar Sandoval const struct btrfs_key *new_key, 3737459931ecSChris Mason unsigned long split_offset) 3738459931ecSChris Mason { 3739459931ecSChris Mason struct extent_buffer *leaf; 3740c91666b1SJosef Bacik int orig_slot, slot; 3741ad48fd75SYan, Zheng char *buf; 3742459931ecSChris Mason u32 nritems; 3743ad48fd75SYan, Zheng u32 item_size; 3744459931ecSChris Mason u32 orig_offset; 3745459931ecSChris Mason struct btrfs_disk_key disk_key; 3746459931ecSChris Mason 3747459931ecSChris Mason leaf = path->nodes[0]; 3748e902baacSDavid Sterba BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); 3749b9473439SChris Mason 3750c91666b1SJosef Bacik orig_slot = path->slots[0]; 37513212fa14SJosef Bacik orig_offset = btrfs_item_offset(leaf, path->slots[0]); 37523212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 3753459931ecSChris Mason 3754459931ecSChris Mason buf = kmalloc(item_size, GFP_NOFS); 3755ad48fd75SYan, Zheng if (!buf) 3756ad48fd75SYan, Zheng return -ENOMEM; 3757ad48fd75SYan, Zheng 3758459931ecSChris Mason read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, 3759459931ecSChris Mason path->slots[0]), item_size); 3760ad48fd75SYan, Zheng 3761459931ecSChris Mason slot = path->slots[0] + 1; 3762459931ecSChris Mason nritems = btrfs_header_nritems(leaf); 3763459931ecSChris Mason if (slot != nritems) { 3764459931ecSChris Mason /* shift the items */ 3765459931ecSChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), 3766459931ecSChris Mason btrfs_item_nr_offset(slot), 3767459931ecSChris Mason (nritems - slot) * sizeof(struct btrfs_item)); 3768459931ecSChris Mason } 3769459931ecSChris Mason 3770459931ecSChris Mason btrfs_cpu_key_to_disk(&disk_key, new_key); 3771459931ecSChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 3772459931ecSChris Mason 37733212fa14SJosef Bacik btrfs_set_item_offset(leaf, slot, orig_offset); 37743212fa14SJosef Bacik btrfs_set_item_size(leaf, slot, item_size - split_offset); 3775459931ecSChris Mason 37763212fa14SJosef Bacik btrfs_set_item_offset(leaf, orig_slot, 3777459931ecSChris Mason orig_offset + item_size - split_offset); 37783212fa14SJosef Bacik btrfs_set_item_size(leaf, orig_slot, split_offset); 3779459931ecSChris Mason 3780459931ecSChris Mason btrfs_set_header_nritems(leaf, nritems + 1); 3781459931ecSChris Mason 3782459931ecSChris Mason /* write the data for the start of the original item */ 3783459931ecSChris Mason write_extent_buffer(leaf, buf, 3784459931ecSChris Mason btrfs_item_ptr_offset(leaf, path->slots[0]), 3785459931ecSChris Mason split_offset); 3786459931ecSChris Mason 3787459931ecSChris Mason /* write the data for the new item */ 3788459931ecSChris Mason write_extent_buffer(leaf, buf + split_offset, 3789459931ecSChris Mason btrfs_item_ptr_offset(leaf, slot), 3790459931ecSChris Mason item_size - split_offset); 3791459931ecSChris Mason btrfs_mark_buffer_dirty(leaf); 3792459931ecSChris Mason 3793e902baacSDavid Sterba BUG_ON(btrfs_leaf_free_space(leaf) < 0); 3794459931ecSChris Mason kfree(buf); 3795ad48fd75SYan, Zheng return 0; 3796ad48fd75SYan, Zheng } 3797ad48fd75SYan, Zheng 3798ad48fd75SYan, Zheng /* 3799ad48fd75SYan, Zheng * This function splits a single item into two items, 3800ad48fd75SYan, Zheng * giving 'new_key' to the new item and splitting the 3801ad48fd75SYan, Zheng * old one at split_offset (from the start of the item). 3802ad48fd75SYan, Zheng * 3803ad48fd75SYan, Zheng * The path may be released by this operation. After 3804ad48fd75SYan, Zheng * the split, the path is pointing to the old item. The 3805ad48fd75SYan, Zheng * new item is going to be in the same node as the old one. 3806ad48fd75SYan, Zheng * 3807ad48fd75SYan, Zheng * Note, the item being split must be smaller enough to live alone on 3808ad48fd75SYan, Zheng * a tree block with room for one extra struct btrfs_item 3809ad48fd75SYan, Zheng * 3810ad48fd75SYan, Zheng * This allows us to split the item in place, keeping a lock on the 3811ad48fd75SYan, Zheng * leaf the entire time. 3812ad48fd75SYan, Zheng */ 3813ad48fd75SYan, Zheng int btrfs_split_item(struct btrfs_trans_handle *trans, 3814ad48fd75SYan, Zheng struct btrfs_root *root, 3815ad48fd75SYan, Zheng struct btrfs_path *path, 3816310712b2SOmar Sandoval const struct btrfs_key *new_key, 3817ad48fd75SYan, Zheng unsigned long split_offset) 3818ad48fd75SYan, Zheng { 3819ad48fd75SYan, Zheng int ret; 3820ad48fd75SYan, Zheng ret = setup_leaf_for_split(trans, root, path, 3821ad48fd75SYan, Zheng sizeof(struct btrfs_item)); 3822ad48fd75SYan, Zheng if (ret) 3823459931ecSChris Mason return ret; 3824ad48fd75SYan, Zheng 382525263cd7SDavid Sterba ret = split_item(path, new_key, split_offset); 3826ad48fd75SYan, Zheng return ret; 3827ad48fd75SYan, Zheng } 3828ad48fd75SYan, Zheng 3829ad48fd75SYan, Zheng /* 3830d352ac68SChris Mason * make the item pointed to by the path smaller. new_size indicates 3831d352ac68SChris Mason * how small to make it, and from_end tells us if we just chop bytes 3832d352ac68SChris Mason * off the end of the item or if we shift the item to chop bytes off 3833d352ac68SChris Mason * the front. 3834d352ac68SChris Mason */ 383578ac4f9eSDavid Sterba void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) 3836b18c6685SChris Mason { 3837b18c6685SChris Mason int slot; 38385f39d397SChris Mason struct extent_buffer *leaf; 3839b18c6685SChris Mason u32 nritems; 3840b18c6685SChris Mason unsigned int data_end; 3841b18c6685SChris Mason unsigned int old_data_start; 3842b18c6685SChris Mason unsigned int old_size; 3843b18c6685SChris Mason unsigned int size_diff; 3844b18c6685SChris Mason int i; 3845cfed81a0SChris Mason struct btrfs_map_token token; 3846cfed81a0SChris Mason 38475f39d397SChris Mason leaf = path->nodes[0]; 3848179e29e4SChris Mason slot = path->slots[0]; 3849179e29e4SChris Mason 38503212fa14SJosef Bacik old_size = btrfs_item_size(leaf, slot); 3851179e29e4SChris Mason if (old_size == new_size) 3852143bede5SJeff Mahoney return; 3853b18c6685SChris Mason 38545f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 38558f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 3856b18c6685SChris Mason 38573212fa14SJosef Bacik old_data_start = btrfs_item_offset(leaf, slot); 3858179e29e4SChris Mason 3859b18c6685SChris Mason size_diff = old_size - new_size; 3860b18c6685SChris Mason 3861b18c6685SChris Mason BUG_ON(slot < 0); 3862b18c6685SChris Mason BUG_ON(slot >= nritems); 3863b18c6685SChris Mason 3864b18c6685SChris Mason /* 3865b18c6685SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 3866b18c6685SChris Mason */ 3867b18c6685SChris Mason /* first correct the data pointers */ 3868c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 3869b18c6685SChris Mason for (i = slot; i < nritems; i++) { 38705f39d397SChris Mason u32 ioff; 3871db94535dSChris Mason 38723212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 38733212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff + size_diff); 3874b18c6685SChris Mason } 3875db94535dSChris Mason 3876b18c6685SChris Mason /* shift the data */ 3877179e29e4SChris Mason if (from_end) { 38783d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 38793d9ec8c4SNikolay Borisov data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + 3880b18c6685SChris Mason data_end, old_data_start + new_size - data_end); 3881179e29e4SChris Mason } else { 3882179e29e4SChris Mason struct btrfs_disk_key disk_key; 3883179e29e4SChris Mason u64 offset; 3884179e29e4SChris Mason 3885179e29e4SChris Mason btrfs_item_key(leaf, &disk_key, slot); 3886179e29e4SChris Mason 3887179e29e4SChris Mason if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { 3888179e29e4SChris Mason unsigned long ptr; 3889179e29e4SChris Mason struct btrfs_file_extent_item *fi; 3890179e29e4SChris Mason 3891179e29e4SChris Mason fi = btrfs_item_ptr(leaf, slot, 3892179e29e4SChris Mason struct btrfs_file_extent_item); 3893179e29e4SChris Mason fi = (struct btrfs_file_extent_item *)( 3894179e29e4SChris Mason (unsigned long)fi - size_diff); 3895179e29e4SChris Mason 3896179e29e4SChris Mason if (btrfs_file_extent_type(leaf, fi) == 3897179e29e4SChris Mason BTRFS_FILE_EXTENT_INLINE) { 3898179e29e4SChris Mason ptr = btrfs_item_ptr_offset(leaf, slot); 3899179e29e4SChris Mason memmove_extent_buffer(leaf, ptr, 3900179e29e4SChris Mason (unsigned long)fi, 39017ec20afbSDavid Sterba BTRFS_FILE_EXTENT_INLINE_DATA_START); 3902179e29e4SChris Mason } 3903179e29e4SChris Mason } 3904179e29e4SChris Mason 39053d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 39063d9ec8c4SNikolay Borisov data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + 3907179e29e4SChris Mason data_end, old_data_start - data_end); 3908179e29e4SChris Mason 3909179e29e4SChris Mason offset = btrfs_disk_key_offset(&disk_key); 3910179e29e4SChris Mason btrfs_set_disk_key_offset(&disk_key, offset + size_diff); 3911179e29e4SChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 3912179e29e4SChris Mason if (slot == 0) 3913b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 3914179e29e4SChris Mason } 39155f39d397SChris Mason 39163212fa14SJosef Bacik btrfs_set_item_size(leaf, slot, new_size); 39175f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 3918b18c6685SChris Mason 3919e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 3920a4f78750SDavid Sterba btrfs_print_leaf(leaf); 3921b18c6685SChris Mason BUG(); 39225f39d397SChris Mason } 3923b18c6685SChris Mason } 3924b18c6685SChris Mason 3925d352ac68SChris Mason /* 39268f69dbd2SStefan Behrens * make the item pointed to by the path bigger, data_size is the added size. 3927d352ac68SChris Mason */ 3928c71dd880SDavid Sterba void btrfs_extend_item(struct btrfs_path *path, u32 data_size) 39296567e837SChris Mason { 39306567e837SChris Mason int slot; 39315f39d397SChris Mason struct extent_buffer *leaf; 39326567e837SChris Mason u32 nritems; 39336567e837SChris Mason unsigned int data_end; 39346567e837SChris Mason unsigned int old_data; 39356567e837SChris Mason unsigned int old_size; 39366567e837SChris Mason int i; 3937cfed81a0SChris Mason struct btrfs_map_token token; 3938cfed81a0SChris Mason 39395f39d397SChris Mason leaf = path->nodes[0]; 39406567e837SChris Mason 39415f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 39428f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 39436567e837SChris Mason 3944e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < data_size) { 3945a4f78750SDavid Sterba btrfs_print_leaf(leaf); 39466567e837SChris Mason BUG(); 39475f39d397SChris Mason } 39486567e837SChris Mason slot = path->slots[0]; 3949dc2e724eSJosef Bacik old_data = btrfs_item_data_end(leaf, slot); 39506567e837SChris Mason 39516567e837SChris Mason BUG_ON(slot < 0); 39523326d1b0SChris Mason if (slot >= nritems) { 3953a4f78750SDavid Sterba btrfs_print_leaf(leaf); 3954c71dd880SDavid Sterba btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", 3955d397712bSChris Mason slot, nritems); 3956290342f6SArnd Bergmann BUG(); 39573326d1b0SChris Mason } 39586567e837SChris Mason 39596567e837SChris Mason /* 39606567e837SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 39616567e837SChris Mason */ 39626567e837SChris Mason /* first correct the data pointers */ 3963c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 39646567e837SChris Mason for (i = slot; i < nritems; i++) { 39655f39d397SChris Mason u32 ioff; 3966db94535dSChris Mason 39673212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 39683212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff - data_size); 39696567e837SChris Mason } 39705f39d397SChris Mason 39716567e837SChris Mason /* shift the data */ 39723d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 39733d9ec8c4SNikolay Borisov data_end - data_size, BTRFS_LEAF_DATA_OFFSET + 39746567e837SChris Mason data_end, old_data - data_end); 39755f39d397SChris Mason 39766567e837SChris Mason data_end = old_data; 39773212fa14SJosef Bacik old_size = btrfs_item_size(leaf, slot); 39783212fa14SJosef Bacik btrfs_set_item_size(leaf, slot, old_size + data_size); 39795f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 39806567e837SChris Mason 3981e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 3982a4f78750SDavid Sterba btrfs_print_leaf(leaf); 39836567e837SChris Mason BUG(); 39845f39d397SChris Mason } 39856567e837SChris Mason } 39866567e837SChris Mason 3987da9ffb24SNikolay Borisov /** 3988da9ffb24SNikolay Borisov * setup_items_for_insert - Helper called before inserting one or more items 3989da9ffb24SNikolay Borisov * to a leaf. Main purpose is to save stack depth by doing the bulk of the work 3990da9ffb24SNikolay Borisov * in a function that doesn't call btrfs_search_slot 3991da9ffb24SNikolay Borisov * 3992da9ffb24SNikolay Borisov * @root: root we are inserting items to 3993da9ffb24SNikolay Borisov * @path: points to the leaf/slot where we are going to insert new items 3994b7ef5f3aSFilipe Manana * @batch: information about the batch of items to insert 399574123bd7SChris Mason */ 3996f0641656SFilipe Manana static void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, 3997b7ef5f3aSFilipe Manana const struct btrfs_item_batch *batch) 3998be0e5c09SChris Mason { 39990b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 40009c58309dSChris Mason int i; 40017518a238SChris Mason u32 nritems; 4002be0e5c09SChris Mason unsigned int data_end; 4003e2fa7227SChris Mason struct btrfs_disk_key disk_key; 400444871b1bSChris Mason struct extent_buffer *leaf; 400544871b1bSChris Mason int slot; 4006cfed81a0SChris Mason struct btrfs_map_token token; 4007fc0d82e1SNikolay Borisov u32 total_size; 4008fc0d82e1SNikolay Borisov 4009b7ef5f3aSFilipe Manana /* 4010b7ef5f3aSFilipe Manana * Before anything else, update keys in the parent and other ancestors 4011b7ef5f3aSFilipe Manana * if needed, then release the write locks on them, so that other tasks 4012b7ef5f3aSFilipe Manana * can use them while we modify the leaf. 4013b7ef5f3aSFilipe Manana */ 401424cdc847SFilipe Manana if (path->slots[0] == 0) { 4015b7ef5f3aSFilipe Manana btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]); 4016b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 401724cdc847SFilipe Manana } 401824cdc847SFilipe Manana btrfs_unlock_up_safe(path, 1); 401924cdc847SFilipe Manana 40205f39d397SChris Mason leaf = path->nodes[0]; 402144871b1bSChris Mason slot = path->slots[0]; 402274123bd7SChris Mason 40235f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 40248f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 4025b7ef5f3aSFilipe Manana total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); 4026eb60ceacSChris Mason 4027e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < total_size) { 4028a4f78750SDavid Sterba btrfs_print_leaf(leaf); 40290b246afaSJeff Mahoney btrfs_crit(fs_info, "not enough freespace need %u have %d", 4030e902baacSDavid Sterba total_size, btrfs_leaf_free_space(leaf)); 4031be0e5c09SChris Mason BUG(); 4032d4dbff95SChris Mason } 40335f39d397SChris Mason 4034c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 4035be0e5c09SChris Mason if (slot != nritems) { 4036dc2e724eSJosef Bacik unsigned int old_data = btrfs_item_data_end(leaf, slot); 4037be0e5c09SChris Mason 40385f39d397SChris Mason if (old_data < data_end) { 4039a4f78750SDavid Sterba btrfs_print_leaf(leaf); 40407269ddd2SNikolay Borisov btrfs_crit(fs_info, 40417269ddd2SNikolay Borisov "item at slot %d with data offset %u beyond data end of leaf %u", 40425f39d397SChris Mason slot, old_data, data_end); 4043290342f6SArnd Bergmann BUG(); 40445f39d397SChris Mason } 4045be0e5c09SChris Mason /* 4046be0e5c09SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 4047be0e5c09SChris Mason */ 4048be0e5c09SChris Mason /* first correct the data pointers */ 40490783fcfcSChris Mason for (i = slot; i < nritems; i++) { 40505f39d397SChris Mason u32 ioff; 4051db94535dSChris Mason 40523212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 40533212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, 4054b7ef5f3aSFilipe Manana ioff - batch->total_data_size); 40550783fcfcSChris Mason } 4056be0e5c09SChris Mason /* shift the items */ 4057b7ef5f3aSFilipe Manana memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + batch->nr), 40585f39d397SChris Mason btrfs_item_nr_offset(slot), 40590783fcfcSChris Mason (nritems - slot) * sizeof(struct btrfs_item)); 4060be0e5c09SChris Mason 4061be0e5c09SChris Mason /* shift the data */ 40623d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 4063b7ef5f3aSFilipe Manana data_end - batch->total_data_size, 4064b7ef5f3aSFilipe Manana BTRFS_LEAF_DATA_OFFSET + data_end, 4065b7ef5f3aSFilipe Manana old_data - data_end); 4066be0e5c09SChris Mason data_end = old_data; 4067be0e5c09SChris Mason } 40685f39d397SChris Mason 406962e2749eSChris Mason /* setup the item for the new data */ 4070b7ef5f3aSFilipe Manana for (i = 0; i < batch->nr; i++) { 4071b7ef5f3aSFilipe Manana btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]); 40729c58309dSChris Mason btrfs_set_item_key(leaf, &disk_key, slot + i); 4073b7ef5f3aSFilipe Manana data_end -= batch->data_sizes[i]; 40743212fa14SJosef Bacik btrfs_set_token_item_offset(&token, slot + i, data_end); 40753212fa14SJosef Bacik btrfs_set_token_item_size(&token, slot + i, batch->data_sizes[i]); 40769c58309dSChris Mason } 407744871b1bSChris Mason 4078b7ef5f3aSFilipe Manana btrfs_set_header_nritems(leaf, nritems + batch->nr); 4079b9473439SChris Mason btrfs_mark_buffer_dirty(leaf); 4080aa5d6bedSChris Mason 4081e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4082a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4083be0e5c09SChris Mason BUG(); 40845f39d397SChris Mason } 408544871b1bSChris Mason } 408644871b1bSChris Mason 408744871b1bSChris Mason /* 4088f0641656SFilipe Manana * Insert a new item into a leaf. 4089f0641656SFilipe Manana * 4090f0641656SFilipe Manana * @root: The root of the btree. 4091f0641656SFilipe Manana * @path: A path pointing to the target leaf and slot. 4092f0641656SFilipe Manana * @key: The key of the new item. 4093f0641656SFilipe Manana * @data_size: The size of the data associated with the new key. 4094f0641656SFilipe Manana */ 4095f0641656SFilipe Manana void btrfs_setup_item_for_insert(struct btrfs_root *root, 4096f0641656SFilipe Manana struct btrfs_path *path, 4097f0641656SFilipe Manana const struct btrfs_key *key, 4098f0641656SFilipe Manana u32 data_size) 4099f0641656SFilipe Manana { 4100f0641656SFilipe Manana struct btrfs_item_batch batch; 4101f0641656SFilipe Manana 4102f0641656SFilipe Manana batch.keys = key; 4103f0641656SFilipe Manana batch.data_sizes = &data_size; 4104f0641656SFilipe Manana batch.total_data_size = data_size; 4105f0641656SFilipe Manana batch.nr = 1; 4106f0641656SFilipe Manana 4107f0641656SFilipe Manana setup_items_for_insert(root, path, &batch); 4108f0641656SFilipe Manana } 4109f0641656SFilipe Manana 4110f0641656SFilipe Manana /* 411144871b1bSChris Mason * Given a key and some data, insert items into the tree. 411244871b1bSChris Mason * This does all the path init required, making room in the tree if needed. 411344871b1bSChris Mason */ 411444871b1bSChris Mason int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, 411544871b1bSChris Mason struct btrfs_root *root, 411644871b1bSChris Mason struct btrfs_path *path, 4117b7ef5f3aSFilipe Manana const struct btrfs_item_batch *batch) 411844871b1bSChris Mason { 411944871b1bSChris Mason int ret = 0; 412044871b1bSChris Mason int slot; 4121b7ef5f3aSFilipe Manana u32 total_size; 412244871b1bSChris Mason 4123b7ef5f3aSFilipe Manana total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); 4124b7ef5f3aSFilipe Manana ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1); 412544871b1bSChris Mason if (ret == 0) 412644871b1bSChris Mason return -EEXIST; 412744871b1bSChris Mason if (ret < 0) 4128143bede5SJeff Mahoney return ret; 412944871b1bSChris Mason 413044871b1bSChris Mason slot = path->slots[0]; 413144871b1bSChris Mason BUG_ON(slot < 0); 413244871b1bSChris Mason 4133b7ef5f3aSFilipe Manana setup_items_for_insert(root, path, batch); 4134143bede5SJeff Mahoney return 0; 413562e2749eSChris Mason } 413662e2749eSChris Mason 413762e2749eSChris Mason /* 413862e2749eSChris Mason * Given a key and some data, insert an item into the tree. 413962e2749eSChris Mason * This does all the path init required, making room in the tree if needed. 414062e2749eSChris Mason */ 4141310712b2SOmar Sandoval int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, 4142310712b2SOmar Sandoval const struct btrfs_key *cpu_key, void *data, 4143310712b2SOmar Sandoval u32 data_size) 414462e2749eSChris Mason { 414562e2749eSChris Mason int ret = 0; 41462c90e5d6SChris Mason struct btrfs_path *path; 41475f39d397SChris Mason struct extent_buffer *leaf; 41485f39d397SChris Mason unsigned long ptr; 414962e2749eSChris Mason 41502c90e5d6SChris Mason path = btrfs_alloc_path(); 4151db5b493aSTsutomu Itoh if (!path) 4152db5b493aSTsutomu Itoh return -ENOMEM; 41532c90e5d6SChris Mason ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); 415462e2749eSChris Mason if (!ret) { 41555f39d397SChris Mason leaf = path->nodes[0]; 41565f39d397SChris Mason ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); 41575f39d397SChris Mason write_extent_buffer(leaf, data, ptr, data_size); 41585f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 415962e2749eSChris Mason } 41602c90e5d6SChris Mason btrfs_free_path(path); 4161aa5d6bedSChris Mason return ret; 4162be0e5c09SChris Mason } 4163be0e5c09SChris Mason 416474123bd7SChris Mason /* 4165f0641656SFilipe Manana * This function duplicates an item, giving 'new_key' to the new item. 4166f0641656SFilipe Manana * It guarantees both items live in the same tree leaf and the new item is 4167f0641656SFilipe Manana * contiguous with the original item. 4168f0641656SFilipe Manana * 4169f0641656SFilipe Manana * This allows us to split a file extent in place, keeping a lock on the leaf 4170f0641656SFilipe Manana * the entire time. 4171f0641656SFilipe Manana */ 4172f0641656SFilipe Manana int btrfs_duplicate_item(struct btrfs_trans_handle *trans, 4173f0641656SFilipe Manana struct btrfs_root *root, 4174f0641656SFilipe Manana struct btrfs_path *path, 4175f0641656SFilipe Manana const struct btrfs_key *new_key) 4176f0641656SFilipe Manana { 4177f0641656SFilipe Manana struct extent_buffer *leaf; 4178f0641656SFilipe Manana int ret; 4179f0641656SFilipe Manana u32 item_size; 4180f0641656SFilipe Manana 4181f0641656SFilipe Manana leaf = path->nodes[0]; 41823212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 4183f0641656SFilipe Manana ret = setup_leaf_for_split(trans, root, path, 4184f0641656SFilipe Manana item_size + sizeof(struct btrfs_item)); 4185f0641656SFilipe Manana if (ret) 4186f0641656SFilipe Manana return ret; 4187f0641656SFilipe Manana 4188f0641656SFilipe Manana path->slots[0]++; 4189f0641656SFilipe Manana btrfs_setup_item_for_insert(root, path, new_key, item_size); 4190f0641656SFilipe Manana leaf = path->nodes[0]; 4191f0641656SFilipe Manana memcpy_extent_buffer(leaf, 4192f0641656SFilipe Manana btrfs_item_ptr_offset(leaf, path->slots[0]), 4193f0641656SFilipe Manana btrfs_item_ptr_offset(leaf, path->slots[0] - 1), 4194f0641656SFilipe Manana item_size); 4195f0641656SFilipe Manana return 0; 4196f0641656SFilipe Manana } 4197f0641656SFilipe Manana 4198f0641656SFilipe Manana /* 41995de08d7dSChris Mason * delete the pointer from a given node. 420074123bd7SChris Mason * 4201d352ac68SChris Mason * the tree should have been previously balanced so the deletion does not 4202d352ac68SChris Mason * empty a node. 420374123bd7SChris Mason */ 4204afe5fea7STsutomu Itoh static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, 4205afe5fea7STsutomu Itoh int level, int slot) 4206be0e5c09SChris Mason { 42075f39d397SChris Mason struct extent_buffer *parent = path->nodes[level]; 42087518a238SChris Mason u32 nritems; 4209f3ea38daSJan Schmidt int ret; 4210be0e5c09SChris Mason 42115f39d397SChris Mason nritems = btrfs_header_nritems(parent); 4212be0e5c09SChris Mason if (slot != nritems - 1) { 4213bf1d3425SDavid Sterba if (level) { 4214f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_move(parent, slot, 4215f3a84ccdSFilipe Manana slot + 1, nritems - slot - 1); 4216bf1d3425SDavid Sterba BUG_ON(ret < 0); 4217bf1d3425SDavid Sterba } 42185f39d397SChris Mason memmove_extent_buffer(parent, 42195f39d397SChris Mason btrfs_node_key_ptr_offset(slot), 42205f39d397SChris Mason btrfs_node_key_ptr_offset(slot + 1), 4221d6025579SChris Mason sizeof(struct btrfs_key_ptr) * 4222d6025579SChris Mason (nritems - slot - 1)); 422357ba86c0SChris Mason } else if (level) { 4224f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, slot, 422533cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REMOVE); 422657ba86c0SChris Mason BUG_ON(ret < 0); 4227be0e5c09SChris Mason } 4228f3ea38daSJan Schmidt 42297518a238SChris Mason nritems--; 42305f39d397SChris Mason btrfs_set_header_nritems(parent, nritems); 42317518a238SChris Mason if (nritems == 0 && parent == root->node) { 42325f39d397SChris Mason BUG_ON(btrfs_header_level(root->node) != 1); 4233eb60ceacSChris Mason /* just turn the root into a leaf and break */ 42345f39d397SChris Mason btrfs_set_header_level(root->node, 0); 4235bb803951SChris Mason } else if (slot == 0) { 42365f39d397SChris Mason struct btrfs_disk_key disk_key; 42375f39d397SChris Mason 42385f39d397SChris Mason btrfs_node_key(parent, &disk_key, 0); 4239b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, level + 1); 4240be0e5c09SChris Mason } 4241d6025579SChris Mason btrfs_mark_buffer_dirty(parent); 4242be0e5c09SChris Mason } 4243be0e5c09SChris Mason 424474123bd7SChris Mason /* 4245323ac95bSChris Mason * a helper function to delete the leaf pointed to by path->slots[1] and 42465d4f98a2SYan Zheng * path->nodes[1]. 4247323ac95bSChris Mason * 4248323ac95bSChris Mason * This deletes the pointer in path->nodes[1] and frees the leaf 4249323ac95bSChris Mason * block extent. zero is returned if it all worked out, < 0 otherwise. 4250323ac95bSChris Mason * 4251323ac95bSChris Mason * The path must have already been setup for deleting the leaf, including 4252323ac95bSChris Mason * all the proper balancing. path->nodes[1] must be locked. 4253323ac95bSChris Mason */ 4254143bede5SJeff Mahoney static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, 4255323ac95bSChris Mason struct btrfs_root *root, 42565d4f98a2SYan Zheng struct btrfs_path *path, 42575d4f98a2SYan Zheng struct extent_buffer *leaf) 4258323ac95bSChris Mason { 42595d4f98a2SYan Zheng WARN_ON(btrfs_header_generation(leaf) != trans->transid); 4260afe5fea7STsutomu Itoh del_ptr(root, path, 1, path->slots[1]); 4261323ac95bSChris Mason 42624d081c41SChris Mason /* 42634d081c41SChris Mason * btrfs_free_extent is expensive, we want to make sure we 42644d081c41SChris Mason * aren't holding any locks when we call it 42654d081c41SChris Mason */ 42664d081c41SChris Mason btrfs_unlock_up_safe(path, 0); 42674d081c41SChris Mason 4268f0486c68SYan, Zheng root_sub_used(root, leaf->len); 4269f0486c68SYan, Zheng 427067439dadSDavid Sterba atomic_inc(&leaf->refs); 42717a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1); 42723083ee2eSJosef Bacik free_extent_buffer_stale(leaf); 4273323ac95bSChris Mason } 4274323ac95bSChris Mason /* 427574123bd7SChris Mason * delete the item at the leaf level in path. If that empties 427674123bd7SChris Mason * the leaf, remove it from the tree 427774123bd7SChris Mason */ 427885e21bacSChris Mason int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, 427985e21bacSChris Mason struct btrfs_path *path, int slot, int nr) 4280be0e5c09SChris Mason { 42810b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 42825f39d397SChris Mason struct extent_buffer *leaf; 4283aa5d6bedSChris Mason int ret = 0; 4284aa5d6bedSChris Mason int wret; 42857518a238SChris Mason u32 nritems; 4286be0e5c09SChris Mason 42875f39d397SChris Mason leaf = path->nodes[0]; 42885f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 4289be0e5c09SChris Mason 429085e21bacSChris Mason if (slot + nr != nritems) { 42910cae23b6SFilipe Manana const u32 last_off = btrfs_item_offset(leaf, slot + nr - 1); 42920cae23b6SFilipe Manana const int data_end = leaf_data_end(leaf); 4293c82f823cSDavid Sterba struct btrfs_map_token token; 42940cae23b6SFilipe Manana u32 dsize = 0; 42950cae23b6SFilipe Manana int i; 42960cae23b6SFilipe Manana 42970cae23b6SFilipe Manana for (i = 0; i < nr; i++) 42980cae23b6SFilipe Manana dsize += btrfs_item_size(leaf, slot + i); 42995f39d397SChris Mason 43003d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 4301d6025579SChris Mason data_end + dsize, 43023d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end, 430385e21bacSChris Mason last_off - data_end); 43045f39d397SChris Mason 4305c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 430685e21bacSChris Mason for (i = slot + nr; i < nritems; i++) { 43075f39d397SChris Mason u32 ioff; 4308db94535dSChris Mason 43093212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 43103212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff + dsize); 43110783fcfcSChris Mason } 4312db94535dSChris Mason 43135f39d397SChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), 431485e21bacSChris Mason btrfs_item_nr_offset(slot + nr), 43150783fcfcSChris Mason sizeof(struct btrfs_item) * 431685e21bacSChris Mason (nritems - slot - nr)); 4317be0e5c09SChris Mason } 431885e21bacSChris Mason btrfs_set_header_nritems(leaf, nritems - nr); 431985e21bacSChris Mason nritems -= nr; 43205f39d397SChris Mason 432174123bd7SChris Mason /* delete the leaf if we've emptied it */ 43227518a238SChris Mason if (nritems == 0) { 43235f39d397SChris Mason if (leaf == root->node) { 43245f39d397SChris Mason btrfs_set_header_level(leaf, 0); 43259a8dd150SChris Mason } else { 43266a884d7dSDavid Sterba btrfs_clean_tree_block(leaf); 4327143bede5SJeff Mahoney btrfs_del_leaf(trans, root, path, leaf); 43289a8dd150SChris Mason } 4329be0e5c09SChris Mason } else { 43307518a238SChris Mason int used = leaf_space_used(leaf, 0, nritems); 4331aa5d6bedSChris Mason if (slot == 0) { 43325f39d397SChris Mason struct btrfs_disk_key disk_key; 43335f39d397SChris Mason 43345f39d397SChris Mason btrfs_item_key(leaf, &disk_key, 0); 4335b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 4336aa5d6bedSChris Mason } 4337aa5d6bedSChris Mason 43387c4063d1SFilipe Manana /* 43397c4063d1SFilipe Manana * Try to delete the leaf if it is mostly empty. We do this by 43407c4063d1SFilipe Manana * trying to move all its items into its left and right neighbours. 43417c4063d1SFilipe Manana * If we can't move all the items, then we don't delete it - it's 43427c4063d1SFilipe Manana * not ideal, but future insertions might fill the leaf with more 43437c4063d1SFilipe Manana * items, or items from other leaves might be moved later into our 43447c4063d1SFilipe Manana * leaf due to deletions on those leaves. 43457c4063d1SFilipe Manana */ 43460b246afaSJeff Mahoney if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { 43477c4063d1SFilipe Manana u32 min_push_space; 43487c4063d1SFilipe Manana 4349be0e5c09SChris Mason /* push_leaf_left fixes the path. 4350be0e5c09SChris Mason * make sure the path still points to our leaf 4351be0e5c09SChris Mason * for possible call to del_ptr below 4352be0e5c09SChris Mason */ 43534920c9acSChris Mason slot = path->slots[1]; 435467439dadSDavid Sterba atomic_inc(&leaf->refs); 43557c4063d1SFilipe Manana /* 43567c4063d1SFilipe Manana * We want to be able to at least push one item to the 43577c4063d1SFilipe Manana * left neighbour leaf, and that's the first item. 43587c4063d1SFilipe Manana */ 43597c4063d1SFilipe Manana min_push_space = sizeof(struct btrfs_item) + 43607c4063d1SFilipe Manana btrfs_item_size(leaf, 0); 43617c4063d1SFilipe Manana wret = push_leaf_left(trans, root, path, 0, 43627c4063d1SFilipe Manana min_push_space, 1, (u32)-1); 436354aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4364aa5d6bedSChris Mason ret = wret; 43655f39d397SChris Mason 43665f39d397SChris Mason if (path->nodes[0] == leaf && 43675f39d397SChris Mason btrfs_header_nritems(leaf)) { 43687c4063d1SFilipe Manana /* 43697c4063d1SFilipe Manana * If we were not able to push all items from our 43707c4063d1SFilipe Manana * leaf to its left neighbour, then attempt to 43717c4063d1SFilipe Manana * either push all the remaining items to the 43727c4063d1SFilipe Manana * right neighbour or none. There's no advantage 43737c4063d1SFilipe Manana * in pushing only some items, instead of all, as 43747c4063d1SFilipe Manana * it's pointless to end up with a leaf having 43757c4063d1SFilipe Manana * too few items while the neighbours can be full 43767c4063d1SFilipe Manana * or nearly full. 43777c4063d1SFilipe Manana */ 43787c4063d1SFilipe Manana nritems = btrfs_header_nritems(leaf); 43797c4063d1SFilipe Manana min_push_space = leaf_space_used(leaf, 0, nritems); 43807c4063d1SFilipe Manana wret = push_leaf_right(trans, root, path, 0, 43817c4063d1SFilipe Manana min_push_space, 1, 0); 438254aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4383aa5d6bedSChris Mason ret = wret; 4384aa5d6bedSChris Mason } 43855f39d397SChris Mason 43865f39d397SChris Mason if (btrfs_header_nritems(leaf) == 0) { 4387323ac95bSChris Mason path->slots[1] = slot; 4388143bede5SJeff Mahoney btrfs_del_leaf(trans, root, path, leaf); 43895f39d397SChris Mason free_extent_buffer(leaf); 4390143bede5SJeff Mahoney ret = 0; 43915de08d7dSChris Mason } else { 4392925baeddSChris Mason /* if we're still in the path, make sure 4393925baeddSChris Mason * we're dirty. Otherwise, one of the 4394925baeddSChris Mason * push_leaf functions must have already 4395925baeddSChris Mason * dirtied this buffer 4396925baeddSChris Mason */ 4397925baeddSChris Mason if (path->nodes[0] == leaf) 43985f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 43995f39d397SChris Mason free_extent_buffer(leaf); 4400be0e5c09SChris Mason } 4401d5719762SChris Mason } else { 44025f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 4403be0e5c09SChris Mason } 44049a8dd150SChris Mason } 4405aa5d6bedSChris Mason return ret; 44069a8dd150SChris Mason } 44079a8dd150SChris Mason 440897571fd0SChris Mason /* 4409925baeddSChris Mason * search the tree again to find a leaf with lesser keys 44107bb86316SChris Mason * returns 0 if it found something or 1 if there are no lesser leaves. 44117bb86316SChris Mason * returns < 0 on io errors. 4412d352ac68SChris Mason * 4413d352ac68SChris Mason * This may release the path, and so you may lose any locks held at the 4414d352ac68SChris Mason * time you call it. 44157bb86316SChris Mason */ 441616e7549fSJosef Bacik int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) 44177bb86316SChris Mason { 4418925baeddSChris Mason struct btrfs_key key; 4419925baeddSChris Mason struct btrfs_disk_key found_key; 4420925baeddSChris Mason int ret; 44217bb86316SChris Mason 4422925baeddSChris Mason btrfs_item_key_to_cpu(path->nodes[0], &key, 0); 4423925baeddSChris Mason 4424e8b0d724SFilipe David Borba Manana if (key.offset > 0) { 4425925baeddSChris Mason key.offset--; 4426e8b0d724SFilipe David Borba Manana } else if (key.type > 0) { 4427925baeddSChris Mason key.type--; 4428e8b0d724SFilipe David Borba Manana key.offset = (u64)-1; 4429e8b0d724SFilipe David Borba Manana } else if (key.objectid > 0) { 4430925baeddSChris Mason key.objectid--; 4431e8b0d724SFilipe David Borba Manana key.type = (u8)-1; 4432e8b0d724SFilipe David Borba Manana key.offset = (u64)-1; 4433e8b0d724SFilipe David Borba Manana } else { 44347bb86316SChris Mason return 1; 4435e8b0d724SFilipe David Borba Manana } 44367bb86316SChris Mason 4437b3b4aa74SDavid Sterba btrfs_release_path(path); 4438925baeddSChris Mason ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4439925baeddSChris Mason if (ret < 0) 4440925baeddSChris Mason return ret; 4441925baeddSChris Mason btrfs_item_key(path->nodes[0], &found_key, 0); 4442925baeddSChris Mason ret = comp_keys(&found_key, &key); 4443337c6f68SFilipe Manana /* 4444337c6f68SFilipe Manana * We might have had an item with the previous key in the tree right 4445337c6f68SFilipe Manana * before we released our path. And after we released our path, that 4446337c6f68SFilipe Manana * item might have been pushed to the first slot (0) of the leaf we 4447337c6f68SFilipe Manana * were holding due to a tree balance. Alternatively, an item with the 4448337c6f68SFilipe Manana * previous key can exist as the only element of a leaf (big fat item). 4449337c6f68SFilipe Manana * Therefore account for these 2 cases, so that our callers (like 4450337c6f68SFilipe Manana * btrfs_previous_item) don't miss an existing item with a key matching 4451337c6f68SFilipe Manana * the previous key we computed above. 4452337c6f68SFilipe Manana */ 4453337c6f68SFilipe Manana if (ret <= 0) 44547bb86316SChris Mason return 0; 4455925baeddSChris Mason return 1; 44567bb86316SChris Mason } 44577bb86316SChris Mason 44583f157a2fSChris Mason /* 44593f157a2fSChris Mason * A helper function to walk down the tree starting at min_key, and looking 4460de78b51aSEric Sandeen * for nodes or leaves that are have a minimum transaction id. 4461de78b51aSEric Sandeen * This is used by the btree defrag code, and tree logging 44623f157a2fSChris Mason * 44633f157a2fSChris Mason * This does not cow, but it does stuff the starting key it finds back 44643f157a2fSChris Mason * into min_key, so you can call btrfs_search_slot with cow=1 on the 44653f157a2fSChris Mason * key and get a writable path. 44663f157a2fSChris Mason * 44673f157a2fSChris Mason * This honors path->lowest_level to prevent descent past a given level 44683f157a2fSChris Mason * of the tree. 44693f157a2fSChris Mason * 4470d352ac68SChris Mason * min_trans indicates the oldest transaction that you are interested 4471d352ac68SChris Mason * in walking through. Any nodes or leaves older than min_trans are 4472d352ac68SChris Mason * skipped over (without reading them). 4473d352ac68SChris Mason * 44743f157a2fSChris Mason * returns zero if something useful was found, < 0 on error and 1 if there 44753f157a2fSChris Mason * was nothing in the tree that matched the search criteria. 44763f157a2fSChris Mason */ 44773f157a2fSChris Mason int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, 4478de78b51aSEric Sandeen struct btrfs_path *path, 44793f157a2fSChris Mason u64 min_trans) 44803f157a2fSChris Mason { 44813f157a2fSChris Mason struct extent_buffer *cur; 44823f157a2fSChris Mason struct btrfs_key found_key; 44833f157a2fSChris Mason int slot; 44849652480bSYan int sret; 44853f157a2fSChris Mason u32 nritems; 44863f157a2fSChris Mason int level; 44873f157a2fSChris Mason int ret = 1; 4488f98de9b9SFilipe Manana int keep_locks = path->keep_locks; 44893f157a2fSChris Mason 4490c922b016SStefan Roesch ASSERT(!path->nowait); 4491f98de9b9SFilipe Manana path->keep_locks = 1; 44923f157a2fSChris Mason again: 4493bd681513SChris Mason cur = btrfs_read_lock_root_node(root); 44943f157a2fSChris Mason level = btrfs_header_level(cur); 4495e02119d5SChris Mason WARN_ON(path->nodes[level]); 44963f157a2fSChris Mason path->nodes[level] = cur; 4497bd681513SChris Mason path->locks[level] = BTRFS_READ_LOCK; 44983f157a2fSChris Mason 44993f157a2fSChris Mason if (btrfs_header_generation(cur) < min_trans) { 45003f157a2fSChris Mason ret = 1; 45013f157a2fSChris Mason goto out; 45023f157a2fSChris Mason } 45033f157a2fSChris Mason while (1) { 45043f157a2fSChris Mason nritems = btrfs_header_nritems(cur); 45053f157a2fSChris Mason level = btrfs_header_level(cur); 4506e3b83361SQu Wenruo sret = btrfs_bin_search(cur, min_key, &slot); 4507cbca7d59SFilipe Manana if (sret < 0) { 4508cbca7d59SFilipe Manana ret = sret; 4509cbca7d59SFilipe Manana goto out; 4510cbca7d59SFilipe Manana } 45113f157a2fSChris Mason 4512323ac95bSChris Mason /* at the lowest level, we're done, setup the path and exit */ 4513323ac95bSChris Mason if (level == path->lowest_level) { 4514e02119d5SChris Mason if (slot >= nritems) 4515e02119d5SChris Mason goto find_next_key; 45163f157a2fSChris Mason ret = 0; 45173f157a2fSChris Mason path->slots[level] = slot; 45183f157a2fSChris Mason btrfs_item_key_to_cpu(cur, &found_key, slot); 45193f157a2fSChris Mason goto out; 45203f157a2fSChris Mason } 45219652480bSYan if (sret && slot > 0) 45229652480bSYan slot--; 45233f157a2fSChris Mason /* 4524de78b51aSEric Sandeen * check this node pointer against the min_trans parameters. 4525260db43cSRandy Dunlap * If it is too old, skip to the next one. 45263f157a2fSChris Mason */ 45273f157a2fSChris Mason while (slot < nritems) { 45283f157a2fSChris Mason u64 gen; 4529e02119d5SChris Mason 45303f157a2fSChris Mason gen = btrfs_node_ptr_generation(cur, slot); 45313f157a2fSChris Mason if (gen < min_trans) { 45323f157a2fSChris Mason slot++; 45333f157a2fSChris Mason continue; 45343f157a2fSChris Mason } 45353f157a2fSChris Mason break; 45363f157a2fSChris Mason } 4537e02119d5SChris Mason find_next_key: 45383f157a2fSChris Mason /* 45393f157a2fSChris Mason * we didn't find a candidate key in this node, walk forward 45403f157a2fSChris Mason * and find another one 45413f157a2fSChris Mason */ 45423f157a2fSChris Mason if (slot >= nritems) { 4543e02119d5SChris Mason path->slots[level] = slot; 4544e02119d5SChris Mason sret = btrfs_find_next_key(root, path, min_key, level, 4545de78b51aSEric Sandeen min_trans); 4546e02119d5SChris Mason if (sret == 0) { 4547b3b4aa74SDavid Sterba btrfs_release_path(path); 45483f157a2fSChris Mason goto again; 45493f157a2fSChris Mason } else { 45503f157a2fSChris Mason goto out; 45513f157a2fSChris Mason } 45523f157a2fSChris Mason } 45533f157a2fSChris Mason /* save our key for returning back */ 45543f157a2fSChris Mason btrfs_node_key_to_cpu(cur, &found_key, slot); 45553f157a2fSChris Mason path->slots[level] = slot; 45563f157a2fSChris Mason if (level == path->lowest_level) { 45573f157a2fSChris Mason ret = 0; 45583f157a2fSChris Mason goto out; 45593f157a2fSChris Mason } 45604b231ae4SDavid Sterba cur = btrfs_read_node_slot(cur, slot); 4561fb770ae4SLiu Bo if (IS_ERR(cur)) { 4562fb770ae4SLiu Bo ret = PTR_ERR(cur); 4563fb770ae4SLiu Bo goto out; 4564fb770ae4SLiu Bo } 45653f157a2fSChris Mason 4566bd681513SChris Mason btrfs_tree_read_lock(cur); 4567b4ce94deSChris Mason 4568bd681513SChris Mason path->locks[level - 1] = BTRFS_READ_LOCK; 45693f157a2fSChris Mason path->nodes[level - 1] = cur; 4570f7c79f30SChris Mason unlock_up(path, level, 1, 0, NULL); 45713f157a2fSChris Mason } 45723f157a2fSChris Mason out: 4573f98de9b9SFilipe Manana path->keep_locks = keep_locks; 4574f98de9b9SFilipe Manana if (ret == 0) { 4575f98de9b9SFilipe Manana btrfs_unlock_up_safe(path, path->lowest_level + 1); 4576f98de9b9SFilipe Manana memcpy(min_key, &found_key, sizeof(found_key)); 4577f98de9b9SFilipe Manana } 45783f157a2fSChris Mason return ret; 45793f157a2fSChris Mason } 45803f157a2fSChris Mason 45813f157a2fSChris Mason /* 45823f157a2fSChris Mason * this is similar to btrfs_next_leaf, but does not try to preserve 45833f157a2fSChris Mason * and fixup the path. It looks for and returns the next key in the 4584de78b51aSEric Sandeen * tree based on the current path and the min_trans parameters. 45853f157a2fSChris Mason * 45863f157a2fSChris Mason * 0 is returned if another key is found, < 0 if there are any errors 45873f157a2fSChris Mason * and 1 is returned if there are no higher keys in the tree 45883f157a2fSChris Mason * 45893f157a2fSChris Mason * path->keep_locks should be set to 1 on the search made before 45903f157a2fSChris Mason * calling this function. 45913f157a2fSChris Mason */ 4592e7a84565SChris Mason int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, 4593de78b51aSEric Sandeen struct btrfs_key *key, int level, u64 min_trans) 4594e7a84565SChris Mason { 4595e7a84565SChris Mason int slot; 4596e7a84565SChris Mason struct extent_buffer *c; 4597e7a84565SChris Mason 45986a9fb468SJosef Bacik WARN_ON(!path->keep_locks && !path->skip_locking); 4599e7a84565SChris Mason while (level < BTRFS_MAX_LEVEL) { 4600e7a84565SChris Mason if (!path->nodes[level]) 4601e7a84565SChris Mason return 1; 4602e7a84565SChris Mason 4603e7a84565SChris Mason slot = path->slots[level] + 1; 4604e7a84565SChris Mason c = path->nodes[level]; 46053f157a2fSChris Mason next: 4606e7a84565SChris Mason if (slot >= btrfs_header_nritems(c)) { 460733c66f43SYan Zheng int ret; 460833c66f43SYan Zheng int orig_lowest; 460933c66f43SYan Zheng struct btrfs_key cur_key; 461033c66f43SYan Zheng if (level + 1 >= BTRFS_MAX_LEVEL || 461133c66f43SYan Zheng !path->nodes[level + 1]) 4612e7a84565SChris Mason return 1; 461333c66f43SYan Zheng 46146a9fb468SJosef Bacik if (path->locks[level + 1] || path->skip_locking) { 461533c66f43SYan Zheng level++; 4616e7a84565SChris Mason continue; 4617e7a84565SChris Mason } 461833c66f43SYan Zheng 461933c66f43SYan Zheng slot = btrfs_header_nritems(c) - 1; 462033c66f43SYan Zheng if (level == 0) 462133c66f43SYan Zheng btrfs_item_key_to_cpu(c, &cur_key, slot); 462233c66f43SYan Zheng else 462333c66f43SYan Zheng btrfs_node_key_to_cpu(c, &cur_key, slot); 462433c66f43SYan Zheng 462533c66f43SYan Zheng orig_lowest = path->lowest_level; 4626b3b4aa74SDavid Sterba btrfs_release_path(path); 462733c66f43SYan Zheng path->lowest_level = level; 462833c66f43SYan Zheng ret = btrfs_search_slot(NULL, root, &cur_key, path, 462933c66f43SYan Zheng 0, 0); 463033c66f43SYan Zheng path->lowest_level = orig_lowest; 463133c66f43SYan Zheng if (ret < 0) 463233c66f43SYan Zheng return ret; 463333c66f43SYan Zheng 463433c66f43SYan Zheng c = path->nodes[level]; 463533c66f43SYan Zheng slot = path->slots[level]; 463633c66f43SYan Zheng if (ret == 0) 463733c66f43SYan Zheng slot++; 463833c66f43SYan Zheng goto next; 463933c66f43SYan Zheng } 464033c66f43SYan Zheng 4641e7a84565SChris Mason if (level == 0) 4642e7a84565SChris Mason btrfs_item_key_to_cpu(c, key, slot); 46433f157a2fSChris Mason else { 46443f157a2fSChris Mason u64 gen = btrfs_node_ptr_generation(c, slot); 46453f157a2fSChris Mason 46463f157a2fSChris Mason if (gen < min_trans) { 46473f157a2fSChris Mason slot++; 46483f157a2fSChris Mason goto next; 46493f157a2fSChris Mason } 4650e7a84565SChris Mason btrfs_node_key_to_cpu(c, key, slot); 46513f157a2fSChris Mason } 4652e7a84565SChris Mason return 0; 4653e7a84565SChris Mason } 4654e7a84565SChris Mason return 1; 4655e7a84565SChris Mason } 4656e7a84565SChris Mason 46573d7806ecSJan Schmidt int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, 46583d7806ecSJan Schmidt u64 time_seq) 46593d7806ecSJan Schmidt { 4660d97e63b6SChris Mason int slot; 46618e73f275SChris Mason int level; 46625f39d397SChris Mason struct extent_buffer *c; 46638e73f275SChris Mason struct extent_buffer *next; 4664d96b3424SFilipe Manana struct btrfs_fs_info *fs_info = root->fs_info; 4665925baeddSChris Mason struct btrfs_key key; 4666d96b3424SFilipe Manana bool need_commit_sem = false; 4667925baeddSChris Mason u32 nritems; 4668925baeddSChris Mason int ret; 46690e46318dSJosef Bacik int i; 4670925baeddSChris Mason 4671bdcdd86cSFilipe Manana /* 4672bdcdd86cSFilipe Manana * The nowait semantics are used only for write paths, where we don't 4673bdcdd86cSFilipe Manana * use the tree mod log and sequence numbers. 4674bdcdd86cSFilipe Manana */ 4675bdcdd86cSFilipe Manana if (time_seq) 4676c922b016SStefan Roesch ASSERT(!path->nowait); 4677c922b016SStefan Roesch 4678925baeddSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 4679d397712bSChris Mason if (nritems == 0) 4680925baeddSChris Mason return 1; 4681925baeddSChris Mason 46828e73f275SChris Mason btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); 46838e73f275SChris Mason again: 46848e73f275SChris Mason level = 1; 46858e73f275SChris Mason next = NULL; 4686b3b4aa74SDavid Sterba btrfs_release_path(path); 46878e73f275SChris Mason 4688a2135011SChris Mason path->keep_locks = 1; 46898e73f275SChris Mason 4690d96b3424SFilipe Manana if (time_seq) { 46913d7806ecSJan Schmidt ret = btrfs_search_old_slot(root, &key, path, time_seq); 4692d96b3424SFilipe Manana } else { 4693d96b3424SFilipe Manana if (path->need_commit_sem) { 4694d96b3424SFilipe Manana path->need_commit_sem = 0; 4695d96b3424SFilipe Manana need_commit_sem = true; 4696bdcdd86cSFilipe Manana if (path->nowait) { 4697bdcdd86cSFilipe Manana if (!down_read_trylock(&fs_info->commit_root_sem)) { 4698bdcdd86cSFilipe Manana ret = -EAGAIN; 4699bdcdd86cSFilipe Manana goto done; 4700bdcdd86cSFilipe Manana } 4701bdcdd86cSFilipe Manana } else { 4702d96b3424SFilipe Manana down_read(&fs_info->commit_root_sem); 4703d96b3424SFilipe Manana } 4704bdcdd86cSFilipe Manana } 4705925baeddSChris Mason ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4706d96b3424SFilipe Manana } 4707925baeddSChris Mason path->keep_locks = 0; 4708925baeddSChris Mason 4709925baeddSChris Mason if (ret < 0) 4710d96b3424SFilipe Manana goto done; 4711925baeddSChris Mason 4712a2135011SChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 4713168fd7d2SChris Mason /* 4714168fd7d2SChris Mason * by releasing the path above we dropped all our locks. A balance 4715168fd7d2SChris Mason * could have added more items next to the key that used to be 4716168fd7d2SChris Mason * at the very end of the block. So, check again here and 4717168fd7d2SChris Mason * advance the path if there are now more items available. 4718168fd7d2SChris Mason */ 4719a2135011SChris Mason if (nritems > 0 && path->slots[0] < nritems - 1) { 4720e457afecSYan Zheng if (ret == 0) 4721168fd7d2SChris Mason path->slots[0]++; 47228e73f275SChris Mason ret = 0; 4723925baeddSChris Mason goto done; 4724925baeddSChris Mason } 47250b43e04fSLiu Bo /* 47260b43e04fSLiu Bo * So the above check misses one case: 47270b43e04fSLiu Bo * - after releasing the path above, someone has removed the item that 47280b43e04fSLiu Bo * used to be at the very end of the block, and balance between leafs 47290b43e04fSLiu Bo * gets another one with bigger key.offset to replace it. 47300b43e04fSLiu Bo * 47310b43e04fSLiu Bo * This one should be returned as well, or we can get leaf corruption 47320b43e04fSLiu Bo * later(esp. in __btrfs_drop_extents()). 47330b43e04fSLiu Bo * 47340b43e04fSLiu Bo * And a bit more explanation about this check, 47350b43e04fSLiu Bo * with ret > 0, the key isn't found, the path points to the slot 47360b43e04fSLiu Bo * where it should be inserted, so the path->slots[0] item must be the 47370b43e04fSLiu Bo * bigger one. 47380b43e04fSLiu Bo */ 47390b43e04fSLiu Bo if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { 47400b43e04fSLiu Bo ret = 0; 47410b43e04fSLiu Bo goto done; 47420b43e04fSLiu Bo } 4743d97e63b6SChris Mason 4744234b63a0SChris Mason while (level < BTRFS_MAX_LEVEL) { 47458e73f275SChris Mason if (!path->nodes[level]) { 47468e73f275SChris Mason ret = 1; 47478e73f275SChris Mason goto done; 47488e73f275SChris Mason } 47495f39d397SChris Mason 4750d97e63b6SChris Mason slot = path->slots[level] + 1; 4751d97e63b6SChris Mason c = path->nodes[level]; 47525f39d397SChris Mason if (slot >= btrfs_header_nritems(c)) { 4753d97e63b6SChris Mason level++; 47548e73f275SChris Mason if (level == BTRFS_MAX_LEVEL) { 47558e73f275SChris Mason ret = 1; 47568e73f275SChris Mason goto done; 47578e73f275SChris Mason } 4758d97e63b6SChris Mason continue; 4759d97e63b6SChris Mason } 47605f39d397SChris Mason 47610e46318dSJosef Bacik 47620e46318dSJosef Bacik /* 47630e46318dSJosef Bacik * Our current level is where we're going to start from, and to 47640e46318dSJosef Bacik * make sure lockdep doesn't complain we need to drop our locks 47650e46318dSJosef Bacik * and nodes from 0 to our current level. 47660e46318dSJosef Bacik */ 47670e46318dSJosef Bacik for (i = 0; i < level; i++) { 47680e46318dSJosef Bacik if (path->locks[level]) { 47690e46318dSJosef Bacik btrfs_tree_read_unlock(path->nodes[i]); 47700e46318dSJosef Bacik path->locks[i] = 0; 47710e46318dSJosef Bacik } 47720e46318dSJosef Bacik free_extent_buffer(path->nodes[i]); 47730e46318dSJosef Bacik path->nodes[i] = NULL; 4774925baeddSChris Mason } 47755f39d397SChris Mason 47768e73f275SChris Mason next = c; 4777d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 4778cda79c54SDavid Sterba slot, &key); 4779bdcdd86cSFilipe Manana if (ret == -EAGAIN && !path->nowait) 47808e73f275SChris Mason goto again; 47815f39d397SChris Mason 478276a05b35SChris Mason if (ret < 0) { 4783b3b4aa74SDavid Sterba btrfs_release_path(path); 478476a05b35SChris Mason goto done; 478576a05b35SChris Mason } 478676a05b35SChris Mason 47875cd57b2cSChris Mason if (!path->skip_locking) { 4788bd681513SChris Mason ret = btrfs_try_tree_read_lock(next); 4789bdcdd86cSFilipe Manana if (!ret && path->nowait) { 4790bdcdd86cSFilipe Manana ret = -EAGAIN; 4791bdcdd86cSFilipe Manana goto done; 4792bdcdd86cSFilipe Manana } 4793d42244a0SJan Schmidt if (!ret && time_seq) { 4794d42244a0SJan Schmidt /* 4795d42244a0SJan Schmidt * If we don't get the lock, we may be racing 4796d42244a0SJan Schmidt * with push_leaf_left, holding that lock while 4797d42244a0SJan Schmidt * itself waiting for the leaf we've currently 4798d42244a0SJan Schmidt * locked. To solve this situation, we give up 4799d42244a0SJan Schmidt * on our lock and cycle. 4800d42244a0SJan Schmidt */ 4801cf538830SJan Schmidt free_extent_buffer(next); 4802d42244a0SJan Schmidt btrfs_release_path(path); 4803d42244a0SJan Schmidt cond_resched(); 4804d42244a0SJan Schmidt goto again; 4805d42244a0SJan Schmidt } 48060e46318dSJosef Bacik if (!ret) 48070e46318dSJosef Bacik btrfs_tree_read_lock(next); 4808bd681513SChris Mason } 4809d97e63b6SChris Mason break; 4810d97e63b6SChris Mason } 4811d97e63b6SChris Mason path->slots[level] = slot; 4812d97e63b6SChris Mason while (1) { 4813d97e63b6SChris Mason level--; 4814d97e63b6SChris Mason path->nodes[level] = next; 4815d97e63b6SChris Mason path->slots[level] = 0; 4816a74a4b97SChris Mason if (!path->skip_locking) 4817ffeb03cfSJosef Bacik path->locks[level] = BTRFS_READ_LOCK; 4818d97e63b6SChris Mason if (!level) 4819d97e63b6SChris Mason break; 4820b4ce94deSChris Mason 4821d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 4822cda79c54SDavid Sterba 0, &key); 4823bdcdd86cSFilipe Manana if (ret == -EAGAIN && !path->nowait) 48248e73f275SChris Mason goto again; 48258e73f275SChris Mason 482676a05b35SChris Mason if (ret < 0) { 4827b3b4aa74SDavid Sterba btrfs_release_path(path); 482876a05b35SChris Mason goto done; 482976a05b35SChris Mason } 483076a05b35SChris Mason 4831bdcdd86cSFilipe Manana if (!path->skip_locking) { 4832bdcdd86cSFilipe Manana if (path->nowait) { 4833bdcdd86cSFilipe Manana if (!btrfs_try_tree_read_lock(next)) { 4834bdcdd86cSFilipe Manana ret = -EAGAIN; 4835bdcdd86cSFilipe Manana goto done; 4836bdcdd86cSFilipe Manana } 4837bdcdd86cSFilipe Manana } else { 48380e46318dSJosef Bacik btrfs_tree_read_lock(next); 4839d97e63b6SChris Mason } 4840bdcdd86cSFilipe Manana } 4841bdcdd86cSFilipe Manana } 48428e73f275SChris Mason ret = 0; 4843925baeddSChris Mason done: 4844f7c79f30SChris Mason unlock_up(path, 0, 1, 0, NULL); 4845d96b3424SFilipe Manana if (need_commit_sem) { 4846d96b3424SFilipe Manana int ret2; 4847d96b3424SFilipe Manana 4848d96b3424SFilipe Manana path->need_commit_sem = 1; 4849d96b3424SFilipe Manana ret2 = finish_need_commit_sem_search(path); 4850d96b3424SFilipe Manana up_read(&fs_info->commit_root_sem); 4851d96b3424SFilipe Manana if (ret2) 4852d96b3424SFilipe Manana ret = ret2; 4853d96b3424SFilipe Manana } 48548e73f275SChris Mason 48558e73f275SChris Mason return ret; 4856d97e63b6SChris Mason } 48570b86a832SChris Mason 4858890d2b1aSJosef Bacik int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq) 4859890d2b1aSJosef Bacik { 4860890d2b1aSJosef Bacik path->slots[0]++; 4861890d2b1aSJosef Bacik if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) 4862890d2b1aSJosef Bacik return btrfs_next_old_leaf(root, path, time_seq); 4863890d2b1aSJosef Bacik return 0; 4864890d2b1aSJosef Bacik } 4865890d2b1aSJosef Bacik 48663f157a2fSChris Mason /* 48673f157a2fSChris Mason * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps 48683f157a2fSChris Mason * searching until it gets past min_objectid or finds an item of 'type' 48693f157a2fSChris Mason * 48703f157a2fSChris Mason * returns 0 if something is found, 1 if nothing was found and < 0 on error 48713f157a2fSChris Mason */ 48720b86a832SChris Mason int btrfs_previous_item(struct btrfs_root *root, 48730b86a832SChris Mason struct btrfs_path *path, u64 min_objectid, 48740b86a832SChris Mason int type) 48750b86a832SChris Mason { 48760b86a832SChris Mason struct btrfs_key found_key; 48770b86a832SChris Mason struct extent_buffer *leaf; 4878e02119d5SChris Mason u32 nritems; 48790b86a832SChris Mason int ret; 48800b86a832SChris Mason 48810b86a832SChris Mason while (1) { 48820b86a832SChris Mason if (path->slots[0] == 0) { 48830b86a832SChris Mason ret = btrfs_prev_leaf(root, path); 48840b86a832SChris Mason if (ret != 0) 48850b86a832SChris Mason return ret; 48860b86a832SChris Mason } else { 48870b86a832SChris Mason path->slots[0]--; 48880b86a832SChris Mason } 48890b86a832SChris Mason leaf = path->nodes[0]; 4890e02119d5SChris Mason nritems = btrfs_header_nritems(leaf); 4891e02119d5SChris Mason if (nritems == 0) 4892e02119d5SChris Mason return 1; 4893e02119d5SChris Mason if (path->slots[0] == nritems) 4894e02119d5SChris Mason path->slots[0]--; 4895e02119d5SChris Mason 48960b86a832SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 4897e02119d5SChris Mason if (found_key.objectid < min_objectid) 4898e02119d5SChris Mason break; 48990a4eefbbSYan Zheng if (found_key.type == type) 49000a4eefbbSYan Zheng return 0; 4901e02119d5SChris Mason if (found_key.objectid == min_objectid && 4902e02119d5SChris Mason found_key.type < type) 4903e02119d5SChris Mason break; 49040b86a832SChris Mason } 49050b86a832SChris Mason return 1; 49060b86a832SChris Mason } 4907ade2e0b3SWang Shilong 4908ade2e0b3SWang Shilong /* 4909ade2e0b3SWang Shilong * search in extent tree to find a previous Metadata/Data extent item with 4910ade2e0b3SWang Shilong * min objecitd. 4911ade2e0b3SWang Shilong * 4912ade2e0b3SWang Shilong * returns 0 if something is found, 1 if nothing was found and < 0 on error 4913ade2e0b3SWang Shilong */ 4914ade2e0b3SWang Shilong int btrfs_previous_extent_item(struct btrfs_root *root, 4915ade2e0b3SWang Shilong struct btrfs_path *path, u64 min_objectid) 4916ade2e0b3SWang Shilong { 4917ade2e0b3SWang Shilong struct btrfs_key found_key; 4918ade2e0b3SWang Shilong struct extent_buffer *leaf; 4919ade2e0b3SWang Shilong u32 nritems; 4920ade2e0b3SWang Shilong int ret; 4921ade2e0b3SWang Shilong 4922ade2e0b3SWang Shilong while (1) { 4923ade2e0b3SWang Shilong if (path->slots[0] == 0) { 4924ade2e0b3SWang Shilong ret = btrfs_prev_leaf(root, path); 4925ade2e0b3SWang Shilong if (ret != 0) 4926ade2e0b3SWang Shilong return ret; 4927ade2e0b3SWang Shilong } else { 4928ade2e0b3SWang Shilong path->slots[0]--; 4929ade2e0b3SWang Shilong } 4930ade2e0b3SWang Shilong leaf = path->nodes[0]; 4931ade2e0b3SWang Shilong nritems = btrfs_header_nritems(leaf); 4932ade2e0b3SWang Shilong if (nritems == 0) 4933ade2e0b3SWang Shilong return 1; 4934ade2e0b3SWang Shilong if (path->slots[0] == nritems) 4935ade2e0b3SWang Shilong path->slots[0]--; 4936ade2e0b3SWang Shilong 4937ade2e0b3SWang Shilong btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 4938ade2e0b3SWang Shilong if (found_key.objectid < min_objectid) 4939ade2e0b3SWang Shilong break; 4940ade2e0b3SWang Shilong if (found_key.type == BTRFS_EXTENT_ITEM_KEY || 4941ade2e0b3SWang Shilong found_key.type == BTRFS_METADATA_ITEM_KEY) 4942ade2e0b3SWang Shilong return 0; 4943ade2e0b3SWang Shilong if (found_key.objectid == min_objectid && 4944ade2e0b3SWang Shilong found_key.type < BTRFS_EXTENT_ITEM_KEY) 4945ade2e0b3SWang Shilong break; 4946ade2e0b3SWang Shilong } 4947ade2e0b3SWang Shilong return 1; 4948ade2e0b3SWang Shilong } 4949226463d7SJosef Bacik 4950226463d7SJosef Bacik int __init btrfs_ctree_init(void) 4951226463d7SJosef Bacik { 4952226463d7SJosef Bacik btrfs_path_cachep = kmem_cache_create("btrfs_path", 4953226463d7SJosef Bacik sizeof(struct btrfs_path), 0, 4954226463d7SJosef Bacik SLAB_MEM_SPREAD, NULL); 4955226463d7SJosef Bacik if (!btrfs_path_cachep) 4956226463d7SJosef Bacik return -ENOMEM; 4957226463d7SJosef Bacik return 0; 4958226463d7SJosef Bacik } 4959226463d7SJosef Bacik 4960226463d7SJosef Bacik void __cold btrfs_ctree_exit(void) 4961226463d7SJosef Bacik { 4962226463d7SJosef Bacik kmem_cache_destroy(btrfs_path_cachep); 4963226463d7SJosef Bacik } 4964