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" 22ad1ac501SJosef Bacik #include "accessors.h" 23a0231804SJosef Bacik #include "extent-tree.h" 249a8dd150SChris Mason 25226463d7SJosef Bacik static struct kmem_cache *btrfs_path_cachep; 26226463d7SJosef Bacik 27e089f05cSChris Mason static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root 28e089f05cSChris Mason *root, struct btrfs_path *path, int level); 29310712b2SOmar Sandoval static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, 30310712b2SOmar Sandoval const struct btrfs_key *ins_key, struct btrfs_path *path, 31310712b2SOmar Sandoval int data_size, int extend); 325f39d397SChris Mason static int push_node_left(struct btrfs_trans_handle *trans, 332ff7e61eSJeff Mahoney struct extent_buffer *dst, 34971a1f66SChris Mason struct extent_buffer *src, int empty); 355f39d397SChris Mason static int balance_node_right(struct btrfs_trans_handle *trans, 365f39d397SChris Mason struct extent_buffer *dst_buf, 375f39d397SChris Mason struct extent_buffer *src_buf); 38afe5fea7STsutomu Itoh static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, 39afe5fea7STsutomu Itoh int level, int slot); 40d97e63b6SChris Mason 41af024ed2SJohannes Thumshirn static const struct btrfs_csums { 42af024ed2SJohannes Thumshirn u16 size; 4359a0fcdbSDavid Sterba const char name[10]; 4459a0fcdbSDavid Sterba const char driver[12]; 45af024ed2SJohannes Thumshirn } btrfs_csums[] = { 46af024ed2SJohannes Thumshirn [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, 473951e7f0SJohannes Thumshirn [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" }, 483831bf00SJohannes Thumshirn [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" }, 49352ae07bSDavid Sterba [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b", 50352ae07bSDavid Sterba .driver = "blake2b-256" }, 51af024ed2SJohannes Thumshirn }; 52af024ed2SJohannes Thumshirn 53af024ed2SJohannes Thumshirn int btrfs_super_csum_size(const struct btrfs_super_block *s) 54af024ed2SJohannes Thumshirn { 55af024ed2SJohannes Thumshirn u16 t = btrfs_super_csum_type(s); 56af024ed2SJohannes Thumshirn /* 57af024ed2SJohannes Thumshirn * csum type is validated at mount time 58af024ed2SJohannes Thumshirn */ 59af024ed2SJohannes Thumshirn return btrfs_csums[t].size; 60af024ed2SJohannes Thumshirn } 61af024ed2SJohannes Thumshirn 62af024ed2SJohannes Thumshirn const char *btrfs_super_csum_name(u16 csum_type) 63af024ed2SJohannes Thumshirn { 64af024ed2SJohannes Thumshirn /* csum type is validated at mount time */ 65af024ed2SJohannes Thumshirn return btrfs_csums[csum_type].name; 66af024ed2SJohannes Thumshirn } 67af024ed2SJohannes Thumshirn 68b4e967beSDavid Sterba /* 69b4e967beSDavid Sterba * Return driver name if defined, otherwise the name that's also a valid driver 70b4e967beSDavid Sterba * name 71b4e967beSDavid Sterba */ 72b4e967beSDavid Sterba const char *btrfs_super_csum_driver(u16 csum_type) 73b4e967beSDavid Sterba { 74b4e967beSDavid Sterba /* csum type is validated at mount time */ 7559a0fcdbSDavid Sterba return btrfs_csums[csum_type].driver[0] ? 7659a0fcdbSDavid Sterba btrfs_csums[csum_type].driver : 77b4e967beSDavid Sterba btrfs_csums[csum_type].name; 78b4e967beSDavid Sterba } 79b4e967beSDavid Sterba 80604997b4SDavid Sterba size_t __attribute_const__ btrfs_get_num_csums(void) 81f7cea56cSDavid Sterba { 82f7cea56cSDavid Sterba return ARRAY_SIZE(btrfs_csums); 83f7cea56cSDavid Sterba } 84f7cea56cSDavid Sterba 852c90e5d6SChris Mason struct btrfs_path *btrfs_alloc_path(void) 862c90e5d6SChris Mason { 87e2c89907SMasahiro Yamada return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); 882c90e5d6SChris Mason } 892c90e5d6SChris Mason 90d352ac68SChris Mason /* this also releases the path */ 912c90e5d6SChris Mason void btrfs_free_path(struct btrfs_path *p) 922c90e5d6SChris Mason { 93ff175d57SJesper Juhl if (!p) 94ff175d57SJesper Juhl return; 95b3b4aa74SDavid Sterba btrfs_release_path(p); 962c90e5d6SChris Mason kmem_cache_free(btrfs_path_cachep, p); 972c90e5d6SChris Mason } 982c90e5d6SChris Mason 99d352ac68SChris Mason /* 100d352ac68SChris Mason * path release drops references on the extent buffers in the path 101d352ac68SChris Mason * and it drops any locks held by this path 102d352ac68SChris Mason * 103d352ac68SChris Mason * It is safe to call this on paths that no locks or extent buffers held. 104d352ac68SChris Mason */ 105b3b4aa74SDavid Sterba noinline void btrfs_release_path(struct btrfs_path *p) 106eb60ceacSChris Mason { 107eb60ceacSChris Mason int i; 108a2135011SChris Mason 109234b63a0SChris Mason for (i = 0; i < BTRFS_MAX_LEVEL; i++) { 1103f157a2fSChris Mason p->slots[i] = 0; 111eb60ceacSChris Mason if (!p->nodes[i]) 112925baeddSChris Mason continue; 113925baeddSChris Mason if (p->locks[i]) { 114bd681513SChris Mason btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); 115925baeddSChris Mason p->locks[i] = 0; 116925baeddSChris Mason } 1175f39d397SChris Mason free_extent_buffer(p->nodes[i]); 1183f157a2fSChris Mason p->nodes[i] = NULL; 119eb60ceacSChris Mason } 120eb60ceacSChris Mason } 121eb60ceacSChris Mason 122d352ac68SChris Mason /* 1238bb808c6SDavid Sterba * We want the transaction abort to print stack trace only for errors where the 1248bb808c6SDavid Sterba * cause could be a bug, eg. due to ENOSPC, and not for common errors that are 1258bb808c6SDavid Sterba * caused by external factors. 1268bb808c6SDavid Sterba */ 1278bb808c6SDavid Sterba bool __cold abort_should_print_stack(int errno) 1288bb808c6SDavid Sterba { 1298bb808c6SDavid Sterba switch (errno) { 1308bb808c6SDavid Sterba case -EIO: 1318bb808c6SDavid Sterba case -EROFS: 1328bb808c6SDavid Sterba case -ENOMEM: 1338bb808c6SDavid Sterba return false; 1348bb808c6SDavid Sterba } 1358bb808c6SDavid Sterba return true; 1368bb808c6SDavid Sterba } 1378bb808c6SDavid Sterba 1388bb808c6SDavid Sterba /* 139d352ac68SChris Mason * safely gets a reference on the root node of a tree. A lock 140d352ac68SChris Mason * is not taken, so a concurrent writer may put a different node 141d352ac68SChris Mason * at the root of the tree. See btrfs_lock_root_node for the 142d352ac68SChris Mason * looping required. 143d352ac68SChris Mason * 144d352ac68SChris Mason * The extent buffer returned by this has a reference taken, so 145d352ac68SChris Mason * it won't disappear. It may stop being the root of the tree 146d352ac68SChris Mason * at any time because there are no locks held. 147d352ac68SChris Mason */ 148925baeddSChris Mason struct extent_buffer *btrfs_root_node(struct btrfs_root *root) 149925baeddSChris Mason { 150925baeddSChris Mason struct extent_buffer *eb; 151240f62c8SChris Mason 1523083ee2eSJosef Bacik while (1) { 153240f62c8SChris Mason rcu_read_lock(); 154240f62c8SChris Mason eb = rcu_dereference(root->node); 1553083ee2eSJosef Bacik 1563083ee2eSJosef Bacik /* 1573083ee2eSJosef Bacik * RCU really hurts here, we could free up the root node because 15801327610SNicholas D Steeves * it was COWed but we may not get the new root node yet so do 1593083ee2eSJosef Bacik * the inc_not_zero dance and if it doesn't work then 1603083ee2eSJosef Bacik * synchronize_rcu and try again. 1613083ee2eSJosef Bacik */ 1623083ee2eSJosef Bacik if (atomic_inc_not_zero(&eb->refs)) { 163240f62c8SChris Mason rcu_read_unlock(); 1643083ee2eSJosef Bacik break; 1653083ee2eSJosef Bacik } 1663083ee2eSJosef Bacik rcu_read_unlock(); 1673083ee2eSJosef Bacik synchronize_rcu(); 1683083ee2eSJosef Bacik } 169925baeddSChris Mason return eb; 170925baeddSChris Mason } 171925baeddSChris Mason 17292a7cc42SQu Wenruo /* 17392a7cc42SQu Wenruo * Cowonly root (not-shareable trees, everything not subvolume or reloc roots), 17492a7cc42SQu Wenruo * just get put onto a simple dirty list. Transaction walks this list to make 17592a7cc42SQu Wenruo * sure they get properly updated on disk. 176d352ac68SChris Mason */ 1770b86a832SChris Mason static void add_root_to_dirty_list(struct btrfs_root *root) 1780b86a832SChris Mason { 1790b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1800b246afaSJeff Mahoney 181e7070be1SJosef Bacik if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || 182e7070be1SJosef Bacik !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) 183e7070be1SJosef Bacik return; 184e7070be1SJosef Bacik 1850b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 186e7070be1SJosef Bacik if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { 187e7070be1SJosef Bacik /* Want the extent tree to be the last on the list */ 1884fd786e6SMisono Tomohiro if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) 189e7070be1SJosef Bacik list_move_tail(&root->dirty_list, 1900b246afaSJeff Mahoney &fs_info->dirty_cowonly_roots); 191e7070be1SJosef Bacik else 192e7070be1SJosef Bacik list_move(&root->dirty_list, 1930b246afaSJeff Mahoney &fs_info->dirty_cowonly_roots); 1940b86a832SChris Mason } 1950b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1960b86a832SChris Mason } 1970b86a832SChris Mason 198d352ac68SChris Mason /* 199d352ac68SChris Mason * used by snapshot creation to make a copy of a root for a tree with 200d352ac68SChris Mason * a given objectid. The buffer with the new root node is returned in 201d352ac68SChris Mason * cow_ret, and this func returns zero on success or a negative error code. 202d352ac68SChris Mason */ 203be20aa9dSChris Mason int btrfs_copy_root(struct btrfs_trans_handle *trans, 204be20aa9dSChris Mason struct btrfs_root *root, 205be20aa9dSChris Mason struct extent_buffer *buf, 206be20aa9dSChris Mason struct extent_buffer **cow_ret, u64 new_root_objectid) 207be20aa9dSChris Mason { 2080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 209be20aa9dSChris Mason struct extent_buffer *cow; 210be20aa9dSChris Mason int ret = 0; 211be20aa9dSChris Mason int level; 2125d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 213be20aa9dSChris Mason 21492a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 2150b246afaSJeff Mahoney trans->transid != fs_info->running_transaction->transid); 21692a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 21727cdeb70SMiao Xie trans->transid != root->last_trans); 218be20aa9dSChris Mason 219be20aa9dSChris Mason level = btrfs_header_level(buf); 2205d4f98a2SYan Zheng if (level == 0) 2215d4f98a2SYan Zheng btrfs_item_key(buf, &disk_key, 0); 2225d4f98a2SYan Zheng else 2235d4f98a2SYan Zheng btrfs_node_key(buf, &disk_key, 0); 22431840ae1SZheng Yan 2254d75f8a9SDavid Sterba cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, 226cf6f34aaSJosef Bacik &disk_key, level, buf->start, 0, 227cf6f34aaSJosef Bacik BTRFS_NESTING_NEW_ROOT); 2285d4f98a2SYan Zheng if (IS_ERR(cow)) 229be20aa9dSChris Mason return PTR_ERR(cow); 230be20aa9dSChris Mason 23158e8012cSDavid Sterba copy_extent_buffer_full(cow, buf); 232be20aa9dSChris Mason btrfs_set_header_bytenr(cow, cow->start); 233be20aa9dSChris Mason btrfs_set_header_generation(cow, trans->transid); 2345d4f98a2SYan Zheng btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 2355d4f98a2SYan Zheng btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 2365d4f98a2SYan Zheng BTRFS_HEADER_FLAG_RELOC); 2375d4f98a2SYan Zheng if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 2385d4f98a2SYan Zheng btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 2395d4f98a2SYan Zheng else 240be20aa9dSChris Mason btrfs_set_header_owner(cow, new_root_objectid); 241be20aa9dSChris Mason 242de37aa51SNikolay Borisov write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 2432b82032cSYan Zheng 244be20aa9dSChris Mason WARN_ON(btrfs_header_generation(buf) > trans->transid); 2455d4f98a2SYan Zheng if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 246e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 2475d4f98a2SYan Zheng else 248e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 249867ed321SJosef Bacik if (ret) { 25072c9925fSFilipe Manana btrfs_tree_unlock(cow); 25172c9925fSFilipe Manana free_extent_buffer(cow); 252867ed321SJosef Bacik btrfs_abort_transaction(trans, ret); 253be20aa9dSChris Mason return ret; 254867ed321SJosef Bacik } 255be20aa9dSChris Mason 256be20aa9dSChris Mason btrfs_mark_buffer_dirty(cow); 257be20aa9dSChris Mason *cow_ret = cow; 258be20aa9dSChris Mason return 0; 259be20aa9dSChris Mason } 260be20aa9dSChris Mason 261d352ac68SChris Mason /* 2625d4f98a2SYan Zheng * check if the tree block can be shared by multiple trees 2635d4f98a2SYan Zheng */ 2645d4f98a2SYan Zheng int btrfs_block_can_be_shared(struct btrfs_root *root, 2655d4f98a2SYan Zheng struct extent_buffer *buf) 2665d4f98a2SYan Zheng { 2675d4f98a2SYan Zheng /* 26892a7cc42SQu Wenruo * Tree blocks not in shareable trees and tree roots are never shared. 26992a7cc42SQu Wenruo * If a block was allocated after the last snapshot and the block was 27092a7cc42SQu Wenruo * not allocated by tree relocation, we know the block is not shared. 2715d4f98a2SYan Zheng */ 27292a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 2735d4f98a2SYan Zheng buf != root->node && buf != root->commit_root && 2745d4f98a2SYan Zheng (btrfs_header_generation(buf) <= 2755d4f98a2SYan Zheng btrfs_root_last_snapshot(&root->root_item) || 2765d4f98a2SYan Zheng btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) 2775d4f98a2SYan Zheng return 1; 278a79865c6SNikolay Borisov 2795d4f98a2SYan Zheng return 0; 2805d4f98a2SYan Zheng } 2815d4f98a2SYan Zheng 2825d4f98a2SYan Zheng static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, 2835d4f98a2SYan Zheng struct btrfs_root *root, 2845d4f98a2SYan Zheng struct extent_buffer *buf, 285f0486c68SYan, Zheng struct extent_buffer *cow, 286f0486c68SYan, Zheng int *last_ref) 2875d4f98a2SYan Zheng { 2880b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 2895d4f98a2SYan Zheng u64 refs; 2905d4f98a2SYan Zheng u64 owner; 2915d4f98a2SYan Zheng u64 flags; 2925d4f98a2SYan Zheng u64 new_flags = 0; 2935d4f98a2SYan Zheng int ret; 2945d4f98a2SYan Zheng 2955d4f98a2SYan Zheng /* 2965d4f98a2SYan Zheng * Backrefs update rules: 2975d4f98a2SYan Zheng * 2985d4f98a2SYan Zheng * Always use full backrefs for extent pointers in tree block 2995d4f98a2SYan Zheng * allocated by tree relocation. 3005d4f98a2SYan Zheng * 3015d4f98a2SYan Zheng * If a shared tree block is no longer referenced by its owner 3025d4f98a2SYan Zheng * tree (btrfs_header_owner(buf) == root->root_key.objectid), 3035d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 3045d4f98a2SYan Zheng * 3055d4f98a2SYan Zheng * If a tree block is been relocating 3065d4f98a2SYan Zheng * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), 3075d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 3085d4f98a2SYan Zheng * The reason for this is some operations (such as drop tree) 3095d4f98a2SYan Zheng * are only allowed for blocks use full backrefs. 3105d4f98a2SYan Zheng */ 3115d4f98a2SYan Zheng 3125d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, buf)) { 3132ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, 3143173a18fSJosef Bacik btrfs_header_level(buf), 1, 3153173a18fSJosef Bacik &refs, &flags); 316be1a5564SMark Fasheh if (ret) 317be1a5564SMark Fasheh return ret; 318e5df9573SMark Fasheh if (refs == 0) { 319e5df9573SMark Fasheh ret = -EROFS; 3200b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 321e5df9573SMark Fasheh return ret; 322e5df9573SMark Fasheh } 3235d4f98a2SYan Zheng } else { 3245d4f98a2SYan Zheng refs = 1; 3255d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 3265d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 3275d4f98a2SYan Zheng flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; 3285d4f98a2SYan Zheng else 3295d4f98a2SYan Zheng flags = 0; 3305d4f98a2SYan Zheng } 3315d4f98a2SYan Zheng 3325d4f98a2SYan Zheng owner = btrfs_header_owner(buf); 3335d4f98a2SYan Zheng BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && 3345d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); 3355d4f98a2SYan Zheng 3365d4f98a2SYan Zheng if (refs > 1) { 3375d4f98a2SYan Zheng if ((owner == root->root_key.objectid || 3385d4f98a2SYan Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && 3395d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { 340e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, buf, 1); 341692826b2SJeff Mahoney if (ret) 342692826b2SJeff Mahoney return ret; 3435d4f98a2SYan Zheng 3445d4f98a2SYan Zheng if (root->root_key.objectid == 3455d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) { 346e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 0); 347692826b2SJeff Mahoney if (ret) 348692826b2SJeff Mahoney return ret; 349e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 350692826b2SJeff Mahoney if (ret) 351692826b2SJeff Mahoney return ret; 3525d4f98a2SYan Zheng } 3535d4f98a2SYan Zheng new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; 3545d4f98a2SYan Zheng } else { 3555d4f98a2SYan Zheng 3565d4f98a2SYan Zheng if (root->root_key.objectid == 3575d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 358e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 3595d4f98a2SYan Zheng else 360e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 361692826b2SJeff Mahoney if (ret) 362692826b2SJeff Mahoney return ret; 3635d4f98a2SYan Zheng } 3645d4f98a2SYan Zheng if (new_flags != 0) { 365b1c79e09SJosef Bacik int level = btrfs_header_level(buf); 366b1c79e09SJosef Bacik 36742c9d0b5SDavid Sterba ret = btrfs_set_disk_extent_flags(trans, buf, 3682fe6a5a1SDavid Sterba new_flags, level); 369be1a5564SMark Fasheh if (ret) 370be1a5564SMark Fasheh return ret; 3715d4f98a2SYan Zheng } 3725d4f98a2SYan Zheng } else { 3735d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { 3745d4f98a2SYan Zheng if (root->root_key.objectid == 3755d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 376e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 3775d4f98a2SYan Zheng else 378e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 379692826b2SJeff Mahoney if (ret) 380692826b2SJeff Mahoney return ret; 381e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 1); 382692826b2SJeff Mahoney if (ret) 383692826b2SJeff Mahoney return ret; 3845d4f98a2SYan Zheng } 3856a884d7dSDavid Sterba btrfs_clean_tree_block(buf); 386f0486c68SYan, Zheng *last_ref = 1; 3875d4f98a2SYan Zheng } 3885d4f98a2SYan Zheng return 0; 3895d4f98a2SYan Zheng } 3905d4f98a2SYan Zheng 3915d4f98a2SYan Zheng /* 392d397712bSChris Mason * does the dirty work in cow of a single block. The parent block (if 393d397712bSChris Mason * supplied) is updated to point to the new cow copy. The new buffer is marked 394d397712bSChris Mason * dirty and returned locked. If you modify the block it needs to be marked 395d397712bSChris Mason * dirty again. 396d352ac68SChris Mason * 397d352ac68SChris Mason * search_start -- an allocation hint for the new block 398d352ac68SChris Mason * 399d397712bSChris Mason * empty_size -- a hint that you plan on doing more cow. This is the size in 400d397712bSChris Mason * bytes the allocator should try to find free next to the block it returns. 401d397712bSChris Mason * This is just a hint and may be ignored by the allocator. 402d352ac68SChris Mason */ 403d397712bSChris Mason static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, 4045f39d397SChris Mason struct btrfs_root *root, 4055f39d397SChris Mason struct extent_buffer *buf, 4065f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 4075f39d397SChris Mason struct extent_buffer **cow_ret, 4089631e4ccSJosef Bacik u64 search_start, u64 empty_size, 4099631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 4106702ed49SChris Mason { 4110b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4125d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 4135f39d397SChris Mason struct extent_buffer *cow; 414be1a5564SMark Fasheh int level, ret; 415f0486c68SYan, Zheng int last_ref = 0; 416925baeddSChris Mason int unlock_orig = 0; 4170f5053ebSGoldwyn Rodrigues u64 parent_start = 0; 4186702ed49SChris Mason 419925baeddSChris Mason if (*cow_ret == buf) 420925baeddSChris Mason unlock_orig = 1; 421925baeddSChris Mason 42249d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 423925baeddSChris Mason 42492a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4250b246afaSJeff Mahoney trans->transid != fs_info->running_transaction->transid); 42692a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 42727cdeb70SMiao Xie trans->transid != root->last_trans); 4285f39d397SChris Mason 4297bb86316SChris Mason level = btrfs_header_level(buf); 43031840ae1SZheng Yan 4315d4f98a2SYan Zheng if (level == 0) 4325d4f98a2SYan Zheng btrfs_item_key(buf, &disk_key, 0); 4335d4f98a2SYan Zheng else 4345d4f98a2SYan Zheng btrfs_node_key(buf, &disk_key, 0); 4355d4f98a2SYan Zheng 4360f5053ebSGoldwyn Rodrigues if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) 4375d4f98a2SYan Zheng parent_start = parent->start; 4385d4f98a2SYan Zheng 43979bd3712SFilipe Manana cow = btrfs_alloc_tree_block(trans, root, parent_start, 44079bd3712SFilipe Manana root->root_key.objectid, &disk_key, level, 44179bd3712SFilipe Manana search_start, empty_size, nest); 4426702ed49SChris Mason if (IS_ERR(cow)) 4436702ed49SChris Mason return PTR_ERR(cow); 4446702ed49SChris Mason 445b4ce94deSChris Mason /* cow is set to blocking by btrfs_init_new_buffer */ 446b4ce94deSChris Mason 44758e8012cSDavid Sterba copy_extent_buffer_full(cow, buf); 448db94535dSChris Mason btrfs_set_header_bytenr(cow, cow->start); 4495f39d397SChris Mason btrfs_set_header_generation(cow, trans->transid); 4505d4f98a2SYan Zheng btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 4515d4f98a2SYan Zheng btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 4525d4f98a2SYan Zheng BTRFS_HEADER_FLAG_RELOC); 4535d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 4545d4f98a2SYan Zheng btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 4555d4f98a2SYan Zheng else 4565f39d397SChris Mason btrfs_set_header_owner(cow, root->root_key.objectid); 4576702ed49SChris Mason 458de37aa51SNikolay Borisov write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 4592b82032cSYan Zheng 460be1a5564SMark Fasheh ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); 461b68dc2a9SMark Fasheh if (ret) { 462572c83acSJosef Bacik btrfs_tree_unlock(cow); 463572c83acSJosef Bacik free_extent_buffer(cow); 46466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 465b68dc2a9SMark Fasheh return ret; 466b68dc2a9SMark Fasheh } 4671a40e23bSZheng Yan 46892a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 46983d4cfd4SJosef Bacik ret = btrfs_reloc_cow_block(trans, root, buf, cow); 47093314e3bSZhaolei if (ret) { 471572c83acSJosef Bacik btrfs_tree_unlock(cow); 472572c83acSJosef Bacik free_extent_buffer(cow); 47366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 47483d4cfd4SJosef Bacik return ret; 47583d4cfd4SJosef Bacik } 47693314e3bSZhaolei } 4773fd0a558SYan, Zheng 4786702ed49SChris Mason if (buf == root->node) { 479925baeddSChris Mason WARN_ON(parent && parent != buf); 4805d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 4815d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 4825d4f98a2SYan Zheng parent_start = buf->start; 483925baeddSChris Mason 48467439dadSDavid Sterba atomic_inc(&cow->refs); 485406808abSFilipe Manana ret = btrfs_tree_mod_log_insert_root(root->node, cow, true); 486d9d19a01SDavid Sterba BUG_ON(ret < 0); 487240f62c8SChris Mason rcu_assign_pointer(root->node, cow); 488925baeddSChris Mason 4897a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), buf, 4907a163608SFilipe Manana parent_start, last_ref); 4915f39d397SChris Mason free_extent_buffer(buf); 4920b86a832SChris Mason add_root_to_dirty_list(root); 4936702ed49SChris Mason } else { 4945d4f98a2SYan Zheng WARN_ON(trans->transid != btrfs_header_generation(parent)); 495f3a84ccdSFilipe Manana btrfs_tree_mod_log_insert_key(parent, parent_slot, 49633cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 4975f39d397SChris Mason btrfs_set_node_blockptr(parent, parent_slot, 498db94535dSChris Mason cow->start); 49974493f7aSChris Mason btrfs_set_node_ptr_generation(parent, parent_slot, 50074493f7aSChris Mason trans->transid); 5016702ed49SChris Mason btrfs_mark_buffer_dirty(parent); 5025de865eeSFilipe David Borba Manana if (last_ref) { 503f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_free_eb(buf); 5045de865eeSFilipe David Borba Manana if (ret) { 505572c83acSJosef Bacik btrfs_tree_unlock(cow); 506572c83acSJosef Bacik free_extent_buffer(cow); 50766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 5085de865eeSFilipe David Borba Manana return ret; 5095de865eeSFilipe David Borba Manana } 5105de865eeSFilipe David Borba Manana } 5117a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), buf, 5127a163608SFilipe Manana parent_start, last_ref); 5136702ed49SChris Mason } 514925baeddSChris Mason if (unlock_orig) 515925baeddSChris Mason btrfs_tree_unlock(buf); 5163083ee2eSJosef Bacik free_extent_buffer_stale(buf); 5176702ed49SChris Mason btrfs_mark_buffer_dirty(cow); 5186702ed49SChris Mason *cow_ret = cow; 5196702ed49SChris Mason return 0; 5206702ed49SChris Mason } 5216702ed49SChris Mason 5225d4f98a2SYan Zheng static inline int should_cow_block(struct btrfs_trans_handle *trans, 5235d4f98a2SYan Zheng struct btrfs_root *root, 5245d4f98a2SYan Zheng struct extent_buffer *buf) 5255d4f98a2SYan Zheng { 526f5ee5c9aSJeff Mahoney if (btrfs_is_testing(root->fs_info)) 527faa2dbf0SJosef Bacik return 0; 528fccb84c9SDavid Sterba 529d1980131SDavid Sterba /* Ensure we can see the FORCE_COW bit */ 530d1980131SDavid Sterba smp_mb__before_atomic(); 531f1ebcc74SLiu Bo 532f1ebcc74SLiu Bo /* 533f1ebcc74SLiu Bo * We do not need to cow a block if 534f1ebcc74SLiu Bo * 1) this block is not created or changed in this transaction; 535f1ebcc74SLiu Bo * 2) this block does not belong to TREE_RELOC tree; 536f1ebcc74SLiu Bo * 3) the root is not forced COW. 537f1ebcc74SLiu Bo * 538f1ebcc74SLiu Bo * What is forced COW: 53901327610SNicholas D Steeves * when we create snapshot during committing the transaction, 54052042d8eSAndrea Gelmini * after we've finished copying src root, we must COW the shared 541f1ebcc74SLiu Bo * block to ensure the metadata consistency. 542f1ebcc74SLiu Bo */ 5435d4f98a2SYan Zheng if (btrfs_header_generation(buf) == trans->transid && 5445d4f98a2SYan Zheng !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && 5455d4f98a2SYan Zheng !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 546f1ebcc74SLiu Bo btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && 54727cdeb70SMiao Xie !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) 5485d4f98a2SYan Zheng return 0; 5495d4f98a2SYan Zheng return 1; 5505d4f98a2SYan Zheng } 5515d4f98a2SYan Zheng 552d352ac68SChris Mason /* 553d352ac68SChris Mason * cows a single block, see __btrfs_cow_block for the real work. 55401327610SNicholas D Steeves * This version of it has extra checks so that a block isn't COWed more than 555d352ac68SChris Mason * once per transaction, as long as it hasn't been written yet 556d352ac68SChris Mason */ 557d397712bSChris Mason noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, 5585f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *buf, 5595f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 5609631e4ccSJosef Bacik struct extent_buffer **cow_ret, 5619631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 56202217ed2SChris Mason { 5630b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5646702ed49SChris Mason u64 search_start; 565f510cfecSChris Mason int ret; 566dc17ff8fSChris Mason 56783354f07SJosef Bacik if (test_bit(BTRFS_ROOT_DELETING, &root->state)) 56883354f07SJosef Bacik btrfs_err(fs_info, 56983354f07SJosef Bacik "COW'ing blocks on a fs root that's being dropped"); 57083354f07SJosef Bacik 5710b246afaSJeff Mahoney if (trans->transaction != fs_info->running_transaction) 57231b1a2bdSJulia Lawall WARN(1, KERN_CRIT "trans %llu running %llu\n", 573c1c9ff7cSGeert Uytterhoeven trans->transid, 5740b246afaSJeff Mahoney fs_info->running_transaction->transid); 57531b1a2bdSJulia Lawall 5760b246afaSJeff Mahoney if (trans->transid != fs_info->generation) 57731b1a2bdSJulia Lawall WARN(1, KERN_CRIT "trans %llu running %llu\n", 5780b246afaSJeff Mahoney trans->transid, fs_info->generation); 579dc17ff8fSChris Mason 5805d4f98a2SYan Zheng if (!should_cow_block(trans, root, buf)) { 58102217ed2SChris Mason *cow_ret = buf; 58202217ed2SChris Mason return 0; 58302217ed2SChris Mason } 584c487685dSChris Mason 585ee22184bSByongho Lee search_start = buf->start & ~((u64)SZ_1G - 1); 586b4ce94deSChris Mason 587f616f5cdSQu Wenruo /* 588f616f5cdSQu Wenruo * Before CoWing this block for later modification, check if it's 589f616f5cdSQu Wenruo * the subtree root and do the delayed subtree trace if needed. 590f616f5cdSQu Wenruo * 591f616f5cdSQu Wenruo * Also We don't care about the error, as it's handled internally. 592f616f5cdSQu Wenruo */ 593f616f5cdSQu Wenruo btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); 594f510cfecSChris Mason ret = __btrfs_cow_block(trans, root, buf, parent, 5959631e4ccSJosef Bacik parent_slot, cow_ret, search_start, 0, nest); 5961abe9b8aSliubo 5971abe9b8aSliubo trace_btrfs_cow_block(root, buf, *cow_ret); 5981abe9b8aSliubo 599f510cfecSChris Mason return ret; 6002c90e5d6SChris Mason } 601f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); 6026702ed49SChris Mason 603d352ac68SChris Mason /* 604d352ac68SChris Mason * helper function for defrag to decide if two blocks pointed to by a 605d352ac68SChris Mason * node are actually close by 606d352ac68SChris Mason */ 6076b80053dSChris Mason static int close_blocks(u64 blocknr, u64 other, u32 blocksize) 6086702ed49SChris Mason { 6096b80053dSChris Mason if (blocknr < other && other - (blocknr + blocksize) < 32768) 6106702ed49SChris Mason return 1; 6116b80053dSChris Mason if (blocknr > other && blocknr - (other + blocksize) < 32768) 6126702ed49SChris Mason return 1; 61302217ed2SChris Mason return 0; 61402217ed2SChris Mason } 61502217ed2SChris Mason 616ce6ef5abSDavid Sterba #ifdef __LITTLE_ENDIAN 617ce6ef5abSDavid Sterba 618ce6ef5abSDavid Sterba /* 619ce6ef5abSDavid Sterba * Compare two keys, on little-endian the disk order is same as CPU order and 620ce6ef5abSDavid Sterba * we can avoid the conversion. 621ce6ef5abSDavid Sterba */ 622ce6ef5abSDavid Sterba static int comp_keys(const struct btrfs_disk_key *disk_key, 623ce6ef5abSDavid Sterba const struct btrfs_key *k2) 624ce6ef5abSDavid Sterba { 625ce6ef5abSDavid Sterba const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; 626ce6ef5abSDavid Sterba 627ce6ef5abSDavid Sterba return btrfs_comp_cpu_keys(k1, k2); 628ce6ef5abSDavid Sterba } 629ce6ef5abSDavid Sterba 630ce6ef5abSDavid Sterba #else 631ce6ef5abSDavid Sterba 632081e9573SChris Mason /* 633081e9573SChris Mason * compare two keys in a memcmp fashion 634081e9573SChris Mason */ 635310712b2SOmar Sandoval static int comp_keys(const struct btrfs_disk_key *disk, 636310712b2SOmar Sandoval const struct btrfs_key *k2) 637081e9573SChris Mason { 638081e9573SChris Mason struct btrfs_key k1; 639081e9573SChris Mason 640081e9573SChris Mason btrfs_disk_key_to_cpu(&k1, disk); 641081e9573SChris Mason 64220736abaSDiego Calleja return btrfs_comp_cpu_keys(&k1, k2); 643081e9573SChris Mason } 644ce6ef5abSDavid Sterba #endif 645081e9573SChris Mason 646f3465ca4SJosef Bacik /* 647f3465ca4SJosef Bacik * same as comp_keys only with two btrfs_key's 648f3465ca4SJosef Bacik */ 649e1f60a65SDavid Sterba int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) 650f3465ca4SJosef Bacik { 651f3465ca4SJosef Bacik if (k1->objectid > k2->objectid) 652f3465ca4SJosef Bacik return 1; 653f3465ca4SJosef Bacik if (k1->objectid < k2->objectid) 654f3465ca4SJosef Bacik return -1; 655f3465ca4SJosef Bacik if (k1->type > k2->type) 656f3465ca4SJosef Bacik return 1; 657f3465ca4SJosef Bacik if (k1->type < k2->type) 658f3465ca4SJosef Bacik return -1; 659f3465ca4SJosef Bacik if (k1->offset > k2->offset) 660f3465ca4SJosef Bacik return 1; 661f3465ca4SJosef Bacik if (k1->offset < k2->offset) 662f3465ca4SJosef Bacik return -1; 663f3465ca4SJosef Bacik return 0; 664f3465ca4SJosef Bacik } 665081e9573SChris Mason 666d352ac68SChris Mason /* 667d352ac68SChris Mason * this is used by the defrag code to go through all the 668d352ac68SChris Mason * leaves pointed to by a node and reallocate them so that 669d352ac68SChris Mason * disk order is close to key order 670d352ac68SChris Mason */ 6716702ed49SChris Mason int btrfs_realloc_node(struct btrfs_trans_handle *trans, 6725f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *parent, 673de78b51aSEric Sandeen int start_slot, u64 *last_ret, 674a6b6e75eSChris Mason struct btrfs_key *progress) 6756702ed49SChris Mason { 6760b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6776b80053dSChris Mason struct extent_buffer *cur; 6786702ed49SChris Mason u64 blocknr; 679e9d0b13bSChris Mason u64 search_start = *last_ret; 680e9d0b13bSChris Mason u64 last_block = 0; 6816702ed49SChris Mason u64 other; 6826702ed49SChris Mason u32 parent_nritems; 6836702ed49SChris Mason int end_slot; 6846702ed49SChris Mason int i; 6856702ed49SChris Mason int err = 0; 6866b80053dSChris Mason u32 blocksize; 687081e9573SChris Mason int progress_passed = 0; 688081e9573SChris Mason struct btrfs_disk_key disk_key; 6896702ed49SChris Mason 6900b246afaSJeff Mahoney WARN_ON(trans->transaction != fs_info->running_transaction); 6910b246afaSJeff Mahoney WARN_ON(trans->transid != fs_info->generation); 69286479a04SChris Mason 6936b80053dSChris Mason parent_nritems = btrfs_header_nritems(parent); 6940b246afaSJeff Mahoney blocksize = fs_info->nodesize; 6955dfe2be7SFilipe Manana end_slot = parent_nritems - 1; 6966702ed49SChris Mason 6975dfe2be7SFilipe Manana if (parent_nritems <= 1) 6986702ed49SChris Mason return 0; 6996702ed49SChris Mason 7005dfe2be7SFilipe Manana for (i = start_slot; i <= end_slot; i++) { 7016702ed49SChris Mason int close = 1; 702a6b6e75eSChris Mason 703081e9573SChris Mason btrfs_node_key(parent, &disk_key, i); 704081e9573SChris Mason if (!progress_passed && comp_keys(&disk_key, progress) < 0) 705081e9573SChris Mason continue; 706081e9573SChris Mason 707081e9573SChris Mason progress_passed = 1; 7086b80053dSChris Mason blocknr = btrfs_node_blockptr(parent, i); 709e9d0b13bSChris Mason if (last_block == 0) 710e9d0b13bSChris Mason last_block = blocknr; 7115708b959SChris Mason 7126702ed49SChris Mason if (i > 0) { 7136b80053dSChris Mason other = btrfs_node_blockptr(parent, i - 1); 7146b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 7156702ed49SChris Mason } 7165dfe2be7SFilipe Manana if (!close && i < end_slot) { 7176b80053dSChris Mason other = btrfs_node_blockptr(parent, i + 1); 7186b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 7196702ed49SChris Mason } 720e9d0b13bSChris Mason if (close) { 721e9d0b13bSChris Mason last_block = blocknr; 7226702ed49SChris Mason continue; 723e9d0b13bSChris Mason } 7246702ed49SChris Mason 725206983b7SJosef Bacik cur = btrfs_read_node_slot(parent, i); 726206983b7SJosef Bacik if (IS_ERR(cur)) 72764c043deSLiu Bo return PTR_ERR(cur); 728e9d0b13bSChris Mason if (search_start == 0) 7296b80053dSChris Mason search_start = last_block; 730e9d0b13bSChris Mason 731e7a84565SChris Mason btrfs_tree_lock(cur); 7326b80053dSChris Mason err = __btrfs_cow_block(trans, root, cur, parent, i, 733e7a84565SChris Mason &cur, search_start, 7346b80053dSChris Mason min(16 * blocksize, 7359631e4ccSJosef Bacik (end_slot - i) * blocksize), 7369631e4ccSJosef Bacik BTRFS_NESTING_COW); 737252c38f0SYan if (err) { 738e7a84565SChris Mason btrfs_tree_unlock(cur); 7396b80053dSChris Mason free_extent_buffer(cur); 7406702ed49SChris Mason break; 741252c38f0SYan } 742e7a84565SChris Mason search_start = cur->start; 743e7a84565SChris Mason last_block = cur->start; 744f2183bdeSChris Mason *last_ret = search_start; 745e7a84565SChris Mason btrfs_tree_unlock(cur); 746e7a84565SChris Mason free_extent_buffer(cur); 7476702ed49SChris Mason } 7486702ed49SChris Mason return err; 7496702ed49SChris Mason } 7506702ed49SChris Mason 75174123bd7SChris Mason /* 752fb81212cSFilipe Manana * Search for a key in the given extent_buffer. 7535f39d397SChris Mason * 754fb81212cSFilipe Manana * The lower boundary for the search is specified by the slot number @low. Use a 755fb81212cSFilipe Manana * value of 0 to search over the whole extent buffer. 75674123bd7SChris Mason * 757fb81212cSFilipe Manana * The slot in the extent buffer is returned via @slot. If the key exists in the 758fb81212cSFilipe Manana * extent buffer, then @slot will point to the slot where the key is, otherwise 759fb81212cSFilipe Manana * it points to the slot where you would insert the key. 760fb81212cSFilipe Manana * 761fb81212cSFilipe Manana * Slot may point to the total number of items (i.e. one position beyond the last 762fb81212cSFilipe Manana * key) if the key is bigger than the last key in the extent buffer. 76374123bd7SChris Mason */ 764fb81212cSFilipe Manana static noinline int generic_bin_search(struct extent_buffer *eb, int low, 76567d5e289SMarcos Paulo de Souza const struct btrfs_key *key, int *slot) 766be0e5c09SChris Mason { 767fb81212cSFilipe Manana unsigned long p; 768fb81212cSFilipe Manana int item_size; 76967d5e289SMarcos Paulo de Souza int high = btrfs_header_nritems(eb); 770be0e5c09SChris Mason int ret; 7715cd17f34SDavid Sterba const int key_size = sizeof(struct btrfs_disk_key); 772be0e5c09SChris Mason 7735e24e9afSLiu Bo if (low > high) { 7745e24e9afSLiu Bo btrfs_err(eb->fs_info, 7755e24e9afSLiu Bo "%s: low (%d) > high (%d) eb %llu owner %llu level %d", 7765e24e9afSLiu Bo __func__, low, high, eb->start, 7775e24e9afSLiu Bo btrfs_header_owner(eb), btrfs_header_level(eb)); 7785e24e9afSLiu Bo return -EINVAL; 7795e24e9afSLiu Bo } 7805e24e9afSLiu Bo 781fb81212cSFilipe Manana if (btrfs_header_level(eb) == 0) { 782fb81212cSFilipe Manana p = offsetof(struct btrfs_leaf, items); 783fb81212cSFilipe Manana item_size = sizeof(struct btrfs_item); 784fb81212cSFilipe Manana } else { 785fb81212cSFilipe Manana p = offsetof(struct btrfs_node, ptrs); 786fb81212cSFilipe Manana item_size = sizeof(struct btrfs_key_ptr); 787fb81212cSFilipe Manana } 788fb81212cSFilipe Manana 789be0e5c09SChris Mason while (low < high) { 7905cd17f34SDavid Sterba unsigned long oip; 7915cd17f34SDavid Sterba unsigned long offset; 7925cd17f34SDavid Sterba struct btrfs_disk_key *tmp; 7935cd17f34SDavid Sterba struct btrfs_disk_key unaligned; 7945cd17f34SDavid Sterba int mid; 7955cd17f34SDavid Sterba 796be0e5c09SChris Mason mid = (low + high) / 2; 7975f39d397SChris Mason offset = p + mid * item_size; 7985cd17f34SDavid Sterba oip = offset_in_page(offset); 7995f39d397SChris Mason 8005cd17f34SDavid Sterba if (oip + key_size <= PAGE_SIZE) { 801884b07d0SQu Wenruo const unsigned long idx = get_eb_page_index(offset); 8025cd17f34SDavid Sterba char *kaddr = page_address(eb->pages[idx]); 803934d375bSChris Mason 804884b07d0SQu Wenruo oip = get_eb_offset_in_page(eb, offset); 8055cd17f34SDavid Sterba tmp = (struct btrfs_disk_key *)(kaddr + oip); 8065cd17f34SDavid Sterba } else { 8075cd17f34SDavid Sterba read_extent_buffer(eb, &unaligned, offset, key_size); 8085f39d397SChris Mason tmp = &unaligned; 809479965d6SChris Mason } 810479965d6SChris Mason 811be0e5c09SChris Mason ret = comp_keys(tmp, key); 812be0e5c09SChris Mason 813be0e5c09SChris Mason if (ret < 0) 814be0e5c09SChris Mason low = mid + 1; 815be0e5c09SChris Mason else if (ret > 0) 816be0e5c09SChris Mason high = mid; 817be0e5c09SChris Mason else { 818be0e5c09SChris Mason *slot = mid; 819be0e5c09SChris Mason return 0; 820be0e5c09SChris Mason } 821be0e5c09SChris Mason } 822be0e5c09SChris Mason *slot = low; 823be0e5c09SChris Mason return 1; 824be0e5c09SChris Mason } 825be0e5c09SChris Mason 82697571fd0SChris Mason /* 827fb81212cSFilipe Manana * Simple binary search on an extent buffer. Works for both leaves and nodes, and 828fb81212cSFilipe Manana * always searches over the whole range of keys (slot 0 to slot 'nritems - 1'). 82997571fd0SChris Mason */ 830a74b35ecSNikolay Borisov int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, 831e3b83361SQu Wenruo int *slot) 832be0e5c09SChris Mason { 833fb81212cSFilipe Manana return generic_bin_search(eb, 0, key, slot); 834be0e5c09SChris Mason } 835be0e5c09SChris Mason 836f0486c68SYan, Zheng static void root_add_used(struct btrfs_root *root, u32 size) 837f0486c68SYan, Zheng { 838f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 839f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 840f0486c68SYan, Zheng btrfs_root_used(&root->root_item) + size); 841f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 842f0486c68SYan, Zheng } 843f0486c68SYan, Zheng 844f0486c68SYan, Zheng static void root_sub_used(struct btrfs_root *root, u32 size) 845f0486c68SYan, Zheng { 846f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 847f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 848f0486c68SYan, Zheng btrfs_root_used(&root->root_item) - size); 849f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 850f0486c68SYan, Zheng } 851f0486c68SYan, Zheng 852d352ac68SChris Mason /* given a node and slot number, this reads the blocks it points to. The 853d352ac68SChris Mason * extent buffer is returned with a reference taken (but unlocked). 854d352ac68SChris Mason */ 8554b231ae4SDavid Sterba struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, 8564b231ae4SDavid Sterba int slot) 857bb803951SChris Mason { 858ca7a79adSChris Mason int level = btrfs_header_level(parent); 859416bc658SJosef Bacik struct extent_buffer *eb; 860581c1760SQu Wenruo struct btrfs_key first_key; 861416bc658SJosef Bacik 862fb770ae4SLiu Bo if (slot < 0 || slot >= btrfs_header_nritems(parent)) 863fb770ae4SLiu Bo return ERR_PTR(-ENOENT); 864ca7a79adSChris Mason 865ca7a79adSChris Mason BUG_ON(level == 0); 866ca7a79adSChris Mason 867581c1760SQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 868d0d20b0fSDavid Sterba eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), 8691b7ec85eSJosef Bacik btrfs_header_owner(parent), 870581c1760SQu Wenruo btrfs_node_ptr_generation(parent, slot), 871581c1760SQu Wenruo level - 1, &first_key); 8724eb150d6SQu Wenruo if (IS_ERR(eb)) 8734eb150d6SQu Wenruo return eb; 8744eb150d6SQu Wenruo if (!extent_buffer_uptodate(eb)) { 875416bc658SJosef Bacik free_extent_buffer(eb); 8764eb150d6SQu Wenruo return ERR_PTR(-EIO); 877416bc658SJosef Bacik } 878416bc658SJosef Bacik 879416bc658SJosef Bacik return eb; 880bb803951SChris Mason } 881bb803951SChris Mason 882d352ac68SChris Mason /* 883d352ac68SChris Mason * node level balancing, used to make sure nodes are in proper order for 884d352ac68SChris Mason * item deletion. We balance from the top down, so we have to make sure 885d352ac68SChris Mason * that a deletion won't leave an node completely empty later on. 886d352ac68SChris Mason */ 887e02119d5SChris Mason static noinline int balance_level(struct btrfs_trans_handle *trans, 88898ed5174SChris Mason struct btrfs_root *root, 88998ed5174SChris Mason struct btrfs_path *path, int level) 890bb803951SChris Mason { 8910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8925f39d397SChris Mason struct extent_buffer *right = NULL; 8935f39d397SChris Mason struct extent_buffer *mid; 8945f39d397SChris Mason struct extent_buffer *left = NULL; 8955f39d397SChris Mason struct extent_buffer *parent = NULL; 896bb803951SChris Mason int ret = 0; 897bb803951SChris Mason int wret; 898bb803951SChris Mason int pslot; 899bb803951SChris Mason int orig_slot = path->slots[level]; 90079f95c82SChris Mason u64 orig_ptr; 901bb803951SChris Mason 90298e6b1ebSLiu Bo ASSERT(level > 0); 903bb803951SChris Mason 9045f39d397SChris Mason mid = path->nodes[level]; 905b4ce94deSChris Mason 906ac5887c8SJosef Bacik WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); 9077bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 9087bb86316SChris Mason 9091d4f8a0cSChris Mason orig_ptr = btrfs_node_blockptr(mid, orig_slot); 91079f95c82SChris Mason 911a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 9125f39d397SChris Mason parent = path->nodes[level + 1]; 913bb803951SChris Mason pslot = path->slots[level + 1]; 914a05a9bb1SLi Zefan } 915bb803951SChris Mason 91640689478SChris Mason /* 91740689478SChris Mason * deal with the case where there is only one pointer in the root 91840689478SChris Mason * by promoting the node below to a root 91940689478SChris Mason */ 9205f39d397SChris Mason if (!parent) { 9215f39d397SChris Mason struct extent_buffer *child; 922bb803951SChris Mason 9235f39d397SChris Mason if (btrfs_header_nritems(mid) != 1) 924bb803951SChris Mason return 0; 925bb803951SChris Mason 926bb803951SChris Mason /* promote the child to a root */ 9274b231ae4SDavid Sterba child = btrfs_read_node_slot(mid, 0); 928fb770ae4SLiu Bo if (IS_ERR(child)) { 929fb770ae4SLiu Bo ret = PTR_ERR(child); 9300b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 931305a26afSMark Fasheh goto enospc; 932305a26afSMark Fasheh } 933305a26afSMark Fasheh 934925baeddSChris Mason btrfs_tree_lock(child); 9359631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 9369631e4ccSJosef Bacik BTRFS_NESTING_COW); 937f0486c68SYan, Zheng if (ret) { 938f0486c68SYan, Zheng btrfs_tree_unlock(child); 939f0486c68SYan, Zheng free_extent_buffer(child); 940f0486c68SYan, Zheng goto enospc; 941f0486c68SYan, Zheng } 9422f375ab9SYan 943406808abSFilipe Manana ret = btrfs_tree_mod_log_insert_root(root->node, child, true); 944d9d19a01SDavid Sterba BUG_ON(ret < 0); 945240f62c8SChris Mason rcu_assign_pointer(root->node, child); 946925baeddSChris Mason 9470b86a832SChris Mason add_root_to_dirty_list(root); 948925baeddSChris Mason btrfs_tree_unlock(child); 949b4ce94deSChris Mason 950925baeddSChris Mason path->locks[level] = 0; 951bb803951SChris Mason path->nodes[level] = NULL; 9526a884d7dSDavid Sterba btrfs_clean_tree_block(mid); 953925baeddSChris Mason btrfs_tree_unlock(mid); 954bb803951SChris Mason /* once for the path */ 9555f39d397SChris Mason free_extent_buffer(mid); 956f0486c68SYan, Zheng 957f0486c68SYan, Zheng root_sub_used(root, mid->len); 9587a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); 959bb803951SChris Mason /* once for the root ptr */ 9603083ee2eSJosef Bacik free_extent_buffer_stale(mid); 961f0486c68SYan, Zheng return 0; 962bb803951SChris Mason } 9635f39d397SChris Mason if (btrfs_header_nritems(mid) > 9640b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) 965bb803951SChris Mason return 0; 966bb803951SChris Mason 9674b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 968fb770ae4SLiu Bo if (IS_ERR(left)) 969fb770ae4SLiu Bo left = NULL; 970fb770ae4SLiu Bo 9715f39d397SChris Mason if (left) { 972bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 9735f39d397SChris Mason wret = btrfs_cow_block(trans, root, left, 9749631e4ccSJosef Bacik parent, pslot - 1, &left, 975bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 97654aa1f4dSChris Mason if (wret) { 97754aa1f4dSChris Mason ret = wret; 97854aa1f4dSChris Mason goto enospc; 97954aa1f4dSChris Mason } 9802cc58cf2SChris Mason } 981fb770ae4SLiu Bo 9824b231ae4SDavid Sterba right = btrfs_read_node_slot(parent, pslot + 1); 983fb770ae4SLiu Bo if (IS_ERR(right)) 984fb770ae4SLiu Bo right = NULL; 985fb770ae4SLiu Bo 9865f39d397SChris Mason if (right) { 987bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 9885f39d397SChris Mason wret = btrfs_cow_block(trans, root, right, 9899631e4ccSJosef Bacik parent, pslot + 1, &right, 990bf59a5a2SJosef Bacik BTRFS_NESTING_RIGHT_COW); 9912cc58cf2SChris Mason if (wret) { 9922cc58cf2SChris Mason ret = wret; 9932cc58cf2SChris Mason goto enospc; 9942cc58cf2SChris Mason } 9952cc58cf2SChris Mason } 9962cc58cf2SChris Mason 9972cc58cf2SChris Mason /* first, try to make some room in the middle buffer */ 9985f39d397SChris Mason if (left) { 9995f39d397SChris Mason orig_slot += btrfs_header_nritems(left); 1000d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 100179f95c82SChris Mason if (wret < 0) 100279f95c82SChris Mason ret = wret; 1003bb803951SChris Mason } 100479f95c82SChris Mason 100579f95c82SChris Mason /* 100679f95c82SChris Mason * then try to empty the right most buffer into the middle 100779f95c82SChris Mason */ 10085f39d397SChris Mason if (right) { 1009d30a668fSDavid Sterba wret = push_node_left(trans, mid, right, 1); 101054aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 101179f95c82SChris Mason ret = wret; 10125f39d397SChris Mason if (btrfs_header_nritems(right) == 0) { 10136a884d7dSDavid Sterba btrfs_clean_tree_block(right); 1014925baeddSChris Mason btrfs_tree_unlock(right); 1015afe5fea7STsutomu Itoh del_ptr(root, path, level + 1, pslot + 1); 1016f0486c68SYan, Zheng root_sub_used(root, right->len); 10177a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), right, 10187a163608SFilipe Manana 0, 1); 10193083ee2eSJosef Bacik free_extent_buffer_stale(right); 1020f0486c68SYan, Zheng right = NULL; 1021bb803951SChris Mason } else { 10225f39d397SChris Mason struct btrfs_disk_key right_key; 10235f39d397SChris Mason btrfs_node_key(right, &right_key, 0); 1024f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, 102533cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 10260e82bcfeSDavid Sterba BUG_ON(ret < 0); 10275f39d397SChris Mason btrfs_set_node_key(parent, &right_key, pslot + 1); 10285f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 1029bb803951SChris Mason } 1030bb803951SChris Mason } 10315f39d397SChris Mason if (btrfs_header_nritems(mid) == 1) { 103279f95c82SChris Mason /* 103379f95c82SChris Mason * we're not allowed to leave a node with one item in the 103479f95c82SChris Mason * tree during a delete. A deletion from lower in the tree 103579f95c82SChris Mason * could try to delete the only pointer in this node. 103679f95c82SChris Mason * So, pull some keys from the left. 103779f95c82SChris Mason * There has to be a left pointer at this point because 103879f95c82SChris Mason * otherwise we would have pulled some pointers from the 103979f95c82SChris Mason * right 104079f95c82SChris Mason */ 1041305a26afSMark Fasheh if (!left) { 1042305a26afSMark Fasheh ret = -EROFS; 10430b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 1044305a26afSMark Fasheh goto enospc; 1045305a26afSMark Fasheh } 104655d32ed8SDavid Sterba wret = balance_node_right(trans, mid, left); 104754aa1f4dSChris Mason if (wret < 0) { 104879f95c82SChris Mason ret = wret; 104954aa1f4dSChris Mason goto enospc; 105054aa1f4dSChris Mason } 1051bce4eae9SChris Mason if (wret == 1) { 1052d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 1053bce4eae9SChris Mason if (wret < 0) 1054bce4eae9SChris Mason ret = wret; 1055bce4eae9SChris Mason } 105679f95c82SChris Mason BUG_ON(wret == 1); 105779f95c82SChris Mason } 10585f39d397SChris Mason if (btrfs_header_nritems(mid) == 0) { 10596a884d7dSDavid Sterba btrfs_clean_tree_block(mid); 1060925baeddSChris Mason btrfs_tree_unlock(mid); 1061afe5fea7STsutomu Itoh del_ptr(root, path, level + 1, pslot); 1062f0486c68SYan, Zheng root_sub_used(root, mid->len); 10637a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); 10643083ee2eSJosef Bacik free_extent_buffer_stale(mid); 1065f0486c68SYan, Zheng mid = NULL; 106679f95c82SChris Mason } else { 106779f95c82SChris Mason /* update the parent key to reflect our changes */ 10685f39d397SChris Mason struct btrfs_disk_key mid_key; 10695f39d397SChris Mason btrfs_node_key(mid, &mid_key, 0); 1070f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot, 107133cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 10720e82bcfeSDavid Sterba BUG_ON(ret < 0); 10735f39d397SChris Mason btrfs_set_node_key(parent, &mid_key, pslot); 10745f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 107579f95c82SChris Mason } 1076bb803951SChris Mason 107779f95c82SChris Mason /* update the path */ 10785f39d397SChris Mason if (left) { 10795f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 108067439dadSDavid Sterba atomic_inc(&left->refs); 1081925baeddSChris Mason /* left was locked after cow */ 10825f39d397SChris Mason path->nodes[level] = left; 1083bb803951SChris Mason path->slots[level + 1] -= 1; 1084bb803951SChris Mason path->slots[level] = orig_slot; 1085925baeddSChris Mason if (mid) { 1086925baeddSChris Mason btrfs_tree_unlock(mid); 10875f39d397SChris Mason free_extent_buffer(mid); 1088925baeddSChris Mason } 1089bb803951SChris Mason } else { 10905f39d397SChris Mason orig_slot -= btrfs_header_nritems(left); 1091bb803951SChris Mason path->slots[level] = orig_slot; 1092bb803951SChris Mason } 1093bb803951SChris Mason } 109479f95c82SChris Mason /* double check we haven't messed things up */ 1095e20d96d6SChris Mason if (orig_ptr != 10965f39d397SChris Mason btrfs_node_blockptr(path->nodes[level], path->slots[level])) 109779f95c82SChris Mason BUG(); 109854aa1f4dSChris Mason enospc: 1099925baeddSChris Mason if (right) { 1100925baeddSChris Mason btrfs_tree_unlock(right); 11015f39d397SChris Mason free_extent_buffer(right); 1102925baeddSChris Mason } 1103925baeddSChris Mason if (left) { 1104925baeddSChris Mason if (path->nodes[level] != left) 1105925baeddSChris Mason btrfs_tree_unlock(left); 11065f39d397SChris Mason free_extent_buffer(left); 1107925baeddSChris Mason } 1108bb803951SChris Mason return ret; 1109bb803951SChris Mason } 1110bb803951SChris Mason 1111d352ac68SChris Mason /* Node balancing for insertion. Here we only split or push nodes around 1112d352ac68SChris Mason * when they are completely full. This is also done top down, so we 1113d352ac68SChris Mason * have to be pessimistic. 1114d352ac68SChris Mason */ 1115d397712bSChris Mason static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, 1116e66f709bSChris Mason struct btrfs_root *root, 1117e66f709bSChris Mason struct btrfs_path *path, int level) 1118e66f709bSChris Mason { 11190b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 11205f39d397SChris Mason struct extent_buffer *right = NULL; 11215f39d397SChris Mason struct extent_buffer *mid; 11225f39d397SChris Mason struct extent_buffer *left = NULL; 11235f39d397SChris Mason struct extent_buffer *parent = NULL; 1124e66f709bSChris Mason int ret = 0; 1125e66f709bSChris Mason int wret; 1126e66f709bSChris Mason int pslot; 1127e66f709bSChris Mason int orig_slot = path->slots[level]; 1128e66f709bSChris Mason 1129e66f709bSChris Mason if (level == 0) 1130e66f709bSChris Mason return 1; 1131e66f709bSChris Mason 11325f39d397SChris Mason mid = path->nodes[level]; 11337bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 1134e66f709bSChris Mason 1135a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 11365f39d397SChris Mason parent = path->nodes[level + 1]; 1137e66f709bSChris Mason pslot = path->slots[level + 1]; 1138a05a9bb1SLi Zefan } 1139e66f709bSChris Mason 11405f39d397SChris Mason if (!parent) 1141e66f709bSChris Mason return 1; 1142e66f709bSChris Mason 11434b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 1144fb770ae4SLiu Bo if (IS_ERR(left)) 1145fb770ae4SLiu Bo left = NULL; 1146e66f709bSChris Mason 1147e66f709bSChris Mason /* first, try to make some room in the middle buffer */ 11485f39d397SChris Mason if (left) { 1149e66f709bSChris Mason u32 left_nr; 1150925baeddSChris Mason 1151bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 1152b4ce94deSChris Mason 11535f39d397SChris Mason left_nr = btrfs_header_nritems(left); 11540b246afaSJeff Mahoney if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 115533ade1f8SChris Mason wret = 1; 115633ade1f8SChris Mason } else { 11575f39d397SChris Mason ret = btrfs_cow_block(trans, root, left, parent, 11589631e4ccSJosef Bacik pslot - 1, &left, 1159bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 116054aa1f4dSChris Mason if (ret) 116154aa1f4dSChris Mason wret = 1; 116254aa1f4dSChris Mason else { 1163d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 0); 116454aa1f4dSChris Mason } 116533ade1f8SChris Mason } 1166e66f709bSChris Mason if (wret < 0) 1167e66f709bSChris Mason ret = wret; 1168e66f709bSChris Mason if (wret == 0) { 11695f39d397SChris Mason struct btrfs_disk_key disk_key; 1170e66f709bSChris Mason orig_slot += left_nr; 11715f39d397SChris Mason btrfs_node_key(mid, &disk_key, 0); 1172f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot, 117333cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 11740e82bcfeSDavid Sterba BUG_ON(ret < 0); 11755f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot); 11765f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 11775f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 11785f39d397SChris Mason path->nodes[level] = left; 1179e66f709bSChris Mason path->slots[level + 1] -= 1; 1180e66f709bSChris Mason path->slots[level] = orig_slot; 1181925baeddSChris Mason btrfs_tree_unlock(mid); 11825f39d397SChris Mason free_extent_buffer(mid); 1183e66f709bSChris Mason } else { 1184e66f709bSChris Mason orig_slot -= 11855f39d397SChris Mason btrfs_header_nritems(left); 1186e66f709bSChris Mason path->slots[level] = orig_slot; 1187925baeddSChris Mason btrfs_tree_unlock(left); 11885f39d397SChris Mason free_extent_buffer(left); 1189e66f709bSChris Mason } 1190e66f709bSChris Mason return 0; 1191e66f709bSChris Mason } 1192925baeddSChris Mason btrfs_tree_unlock(left); 11935f39d397SChris Mason free_extent_buffer(left); 1194e66f709bSChris Mason } 11954b231ae4SDavid Sterba right = btrfs_read_node_slot(parent, pslot + 1); 1196fb770ae4SLiu Bo if (IS_ERR(right)) 1197fb770ae4SLiu Bo right = NULL; 1198e66f709bSChris Mason 1199e66f709bSChris Mason /* 1200e66f709bSChris Mason * then try to empty the right most buffer into the middle 1201e66f709bSChris Mason */ 12025f39d397SChris Mason if (right) { 120333ade1f8SChris Mason u32 right_nr; 1204b4ce94deSChris Mason 1205bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 1206b4ce94deSChris Mason 12075f39d397SChris Mason right_nr = btrfs_header_nritems(right); 12080b246afaSJeff Mahoney if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 120933ade1f8SChris Mason wret = 1; 121033ade1f8SChris Mason } else { 12115f39d397SChris Mason ret = btrfs_cow_block(trans, root, right, 12125f39d397SChris Mason parent, pslot + 1, 1213bf59a5a2SJosef Bacik &right, BTRFS_NESTING_RIGHT_COW); 121454aa1f4dSChris Mason if (ret) 121554aa1f4dSChris Mason wret = 1; 121654aa1f4dSChris Mason else { 121755d32ed8SDavid Sterba wret = balance_node_right(trans, right, mid); 121833ade1f8SChris Mason } 121954aa1f4dSChris Mason } 1220e66f709bSChris Mason if (wret < 0) 1221e66f709bSChris Mason ret = wret; 1222e66f709bSChris Mason if (wret == 0) { 12235f39d397SChris Mason struct btrfs_disk_key disk_key; 12245f39d397SChris Mason 12255f39d397SChris Mason btrfs_node_key(right, &disk_key, 0); 1226f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, 122733cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 12280e82bcfeSDavid Sterba BUG_ON(ret < 0); 12295f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot + 1); 12305f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 12315f39d397SChris Mason 12325f39d397SChris Mason if (btrfs_header_nritems(mid) <= orig_slot) { 12335f39d397SChris Mason path->nodes[level] = right; 1234e66f709bSChris Mason path->slots[level + 1] += 1; 1235e66f709bSChris Mason path->slots[level] = orig_slot - 12365f39d397SChris Mason btrfs_header_nritems(mid); 1237925baeddSChris Mason btrfs_tree_unlock(mid); 12385f39d397SChris Mason free_extent_buffer(mid); 1239e66f709bSChris Mason } else { 1240925baeddSChris Mason btrfs_tree_unlock(right); 12415f39d397SChris Mason free_extent_buffer(right); 1242e66f709bSChris Mason } 1243e66f709bSChris Mason return 0; 1244e66f709bSChris Mason } 1245925baeddSChris Mason btrfs_tree_unlock(right); 12465f39d397SChris Mason free_extent_buffer(right); 1247e66f709bSChris Mason } 1248e66f709bSChris Mason return 1; 1249e66f709bSChris Mason } 1250e66f709bSChris Mason 125174123bd7SChris Mason /* 1252d352ac68SChris Mason * readahead one full node of leaves, finding things that are close 1253d352ac68SChris Mason * to the block in 'slot', and triggering ra on them. 12543c69faecSChris Mason */ 12552ff7e61eSJeff Mahoney static void reada_for_search(struct btrfs_fs_info *fs_info, 1256e02119d5SChris Mason struct btrfs_path *path, 125701f46658SChris Mason int level, int slot, u64 objectid) 12583c69faecSChris Mason { 12595f39d397SChris Mason struct extent_buffer *node; 126001f46658SChris Mason struct btrfs_disk_key disk_key; 12613c69faecSChris Mason u32 nritems; 12623c69faecSChris Mason u64 search; 1263a7175319SChris Mason u64 target; 12646b80053dSChris Mason u64 nread = 0; 1265ace75066SFilipe Manana u64 nread_max; 12666b80053dSChris Mason u32 nr; 12676b80053dSChris Mason u32 blocksize; 12686b80053dSChris Mason u32 nscan = 0; 1269db94535dSChris Mason 1270ace75066SFilipe Manana if (level != 1 && path->reada != READA_FORWARD_ALWAYS) 12713c69faecSChris Mason return; 12723c69faecSChris Mason 12736702ed49SChris Mason if (!path->nodes[level]) 12746702ed49SChris Mason return; 12756702ed49SChris Mason 12765f39d397SChris Mason node = path->nodes[level]; 1277925baeddSChris Mason 1278ace75066SFilipe Manana /* 1279ace75066SFilipe Manana * Since the time between visiting leaves is much shorter than the time 1280ace75066SFilipe Manana * between visiting nodes, limit read ahead of nodes to 1, to avoid too 1281ace75066SFilipe Manana * much IO at once (possibly random). 1282ace75066SFilipe Manana */ 1283ace75066SFilipe Manana if (path->reada == READA_FORWARD_ALWAYS) { 1284ace75066SFilipe Manana if (level > 1) 1285ace75066SFilipe Manana nread_max = node->fs_info->nodesize; 1286ace75066SFilipe Manana else 1287ace75066SFilipe Manana nread_max = SZ_128K; 1288ace75066SFilipe Manana } else { 1289ace75066SFilipe Manana nread_max = SZ_64K; 1290ace75066SFilipe Manana } 1291ace75066SFilipe Manana 12923c69faecSChris Mason search = btrfs_node_blockptr(node, slot); 12930b246afaSJeff Mahoney blocksize = fs_info->nodesize; 1294069a2e37SFilipe Manana if (path->reada != READA_FORWARD_ALWAYS) { 1295069a2e37SFilipe Manana struct extent_buffer *eb; 1296069a2e37SFilipe Manana 12970b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, search); 12985f39d397SChris Mason if (eb) { 12995f39d397SChris Mason free_extent_buffer(eb); 13003c69faecSChris Mason return; 13013c69faecSChris Mason } 1302069a2e37SFilipe Manana } 13033c69faecSChris Mason 1304a7175319SChris Mason target = search; 13056b80053dSChris Mason 13065f39d397SChris Mason nritems = btrfs_header_nritems(node); 13076b80053dSChris Mason nr = slot; 130825b8b936SJosef Bacik 13093c69faecSChris Mason while (1) { 1310e4058b54SDavid Sterba if (path->reada == READA_BACK) { 13116b80053dSChris Mason if (nr == 0) 13123c69faecSChris Mason break; 13136b80053dSChris Mason nr--; 1314ace75066SFilipe Manana } else if (path->reada == READA_FORWARD || 1315ace75066SFilipe Manana path->reada == READA_FORWARD_ALWAYS) { 13166b80053dSChris Mason nr++; 13176b80053dSChris Mason if (nr >= nritems) 13186b80053dSChris Mason break; 13193c69faecSChris Mason } 1320e4058b54SDavid Sterba if (path->reada == READA_BACK && objectid) { 132101f46658SChris Mason btrfs_node_key(node, &disk_key, nr); 132201f46658SChris Mason if (btrfs_disk_key_objectid(&disk_key) != objectid) 132301f46658SChris Mason break; 132401f46658SChris Mason } 13256b80053dSChris Mason search = btrfs_node_blockptr(node, nr); 1326ace75066SFilipe Manana if (path->reada == READA_FORWARD_ALWAYS || 1327ace75066SFilipe Manana (search <= target && target - search <= 65536) || 1328a7175319SChris Mason (search > target && search - target <= 65536)) { 1329bfb484d9SJosef Bacik btrfs_readahead_node_child(node, nr); 13306b80053dSChris Mason nread += blocksize; 13313c69faecSChris Mason } 13326b80053dSChris Mason nscan++; 1333ace75066SFilipe Manana if (nread > nread_max || nscan > 32) 13346b80053dSChris Mason break; 13353c69faecSChris Mason } 13363c69faecSChris Mason } 1337925baeddSChris Mason 1338bfb484d9SJosef Bacik static noinline void reada_for_balance(struct btrfs_path *path, int level) 1339b4ce94deSChris Mason { 1340bfb484d9SJosef Bacik struct extent_buffer *parent; 1341b4ce94deSChris Mason int slot; 1342b4ce94deSChris Mason int nritems; 1343b4ce94deSChris Mason 13448c594ea8SChris Mason parent = path->nodes[level + 1]; 1345b4ce94deSChris Mason if (!parent) 13460b08851fSJosef Bacik return; 1347b4ce94deSChris Mason 1348b4ce94deSChris Mason nritems = btrfs_header_nritems(parent); 13498c594ea8SChris Mason slot = path->slots[level + 1]; 1350b4ce94deSChris Mason 1351bfb484d9SJosef Bacik if (slot > 0) 1352bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot - 1); 1353bfb484d9SJosef Bacik if (slot + 1 < nritems) 1354bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot + 1); 1355b4ce94deSChris Mason } 1356b4ce94deSChris Mason 1357b4ce94deSChris Mason 1358b4ce94deSChris Mason /* 1359d397712bSChris Mason * when we walk down the tree, it is usually safe to unlock the higher layers 1360d397712bSChris Mason * in the tree. The exceptions are when our path goes through slot 0, because 1361d397712bSChris Mason * operations on the tree might require changing key pointers higher up in the 1362d397712bSChris Mason * tree. 1363d352ac68SChris Mason * 1364d397712bSChris Mason * callers might also have set path->keep_locks, which tells this code to keep 1365d397712bSChris Mason * the lock if the path points to the last slot in the block. This is part of 1366d397712bSChris Mason * walking through the tree, and selecting the next slot in the higher block. 1367d352ac68SChris Mason * 1368d397712bSChris Mason * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so 1369d397712bSChris Mason * if lowest_unlock is 1, level 0 won't be unlocked 1370d352ac68SChris Mason */ 1371e02119d5SChris Mason static noinline void unlock_up(struct btrfs_path *path, int level, 1372f7c79f30SChris Mason int lowest_unlock, int min_write_lock_level, 1373f7c79f30SChris Mason int *write_lock_level) 1374925baeddSChris Mason { 1375925baeddSChris Mason int i; 1376925baeddSChris Mason int skip_level = level; 1377c1227996SNikolay Borisov bool check_skip = true; 1378925baeddSChris Mason 1379925baeddSChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 1380925baeddSChris Mason if (!path->nodes[i]) 1381925baeddSChris Mason break; 1382925baeddSChris Mason if (!path->locks[i]) 1383925baeddSChris Mason break; 1384c1227996SNikolay Borisov 1385c1227996SNikolay Borisov if (check_skip) { 1386c1227996SNikolay Borisov if (path->slots[i] == 0) { 1387925baeddSChris Mason skip_level = i + 1; 1388925baeddSChris Mason continue; 1389925baeddSChris Mason } 1390c1227996SNikolay Borisov 1391c1227996SNikolay Borisov if (path->keep_locks) { 1392925baeddSChris Mason u32 nritems; 1393c1227996SNikolay Borisov 1394c1227996SNikolay Borisov nritems = btrfs_header_nritems(path->nodes[i]); 1395051e1b9fSChris Mason if (nritems < 1 || path->slots[i] >= nritems - 1) { 1396925baeddSChris Mason skip_level = i + 1; 1397925baeddSChris Mason continue; 1398925baeddSChris Mason } 1399925baeddSChris Mason } 1400c1227996SNikolay Borisov } 1401051e1b9fSChris Mason 1402d80bb3f9SLiu Bo if (i >= lowest_unlock && i > skip_level) { 1403c1227996SNikolay Borisov check_skip = false; 1404c1227996SNikolay Borisov btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); 1405925baeddSChris Mason path->locks[i] = 0; 1406f7c79f30SChris Mason if (write_lock_level && 1407f7c79f30SChris Mason i > min_write_lock_level && 1408f7c79f30SChris Mason i <= *write_lock_level) { 1409f7c79f30SChris Mason *write_lock_level = i - 1; 1410f7c79f30SChris Mason } 1411925baeddSChris Mason } 1412925baeddSChris Mason } 1413925baeddSChris Mason } 1414925baeddSChris Mason 14153c69faecSChris Mason /* 1416376a21d7SFilipe Manana * Helper function for btrfs_search_slot() and other functions that do a search 1417376a21d7SFilipe Manana * on a btree. The goal is to find a tree block in the cache (the radix tree at 1418376a21d7SFilipe Manana * fs_info->buffer_radix), but if we can't find it, or it's not up to date, read 1419376a21d7SFilipe Manana * its pages from disk. 1420c8c42864SChris Mason * 1421376a21d7SFilipe Manana * Returns -EAGAIN, with the path unlocked, if the caller needs to repeat the 1422376a21d7SFilipe Manana * whole btree search, starting again from the current root node. 1423c8c42864SChris Mason */ 1424c8c42864SChris Mason static int 1425d07b8528SLiu Bo read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, 1426c8c42864SChris Mason struct extent_buffer **eb_ret, int level, int slot, 1427cda79c54SDavid Sterba const struct btrfs_key *key) 1428c8c42864SChris Mason { 14290b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1430c8c42864SChris Mason u64 blocknr; 1431c8c42864SChris Mason u64 gen; 1432c8c42864SChris Mason struct extent_buffer *tmp; 1433581c1760SQu Wenruo struct btrfs_key first_key; 143476a05b35SChris Mason int ret; 1435581c1760SQu Wenruo int parent_level; 1436b246666eSFilipe Manana bool unlock_up; 1437c8c42864SChris Mason 1438b246666eSFilipe Manana unlock_up = ((level + 1 < BTRFS_MAX_LEVEL) && p->locks[level + 1]); 1439213ff4b7SNikolay Borisov blocknr = btrfs_node_blockptr(*eb_ret, slot); 1440213ff4b7SNikolay Borisov gen = btrfs_node_ptr_generation(*eb_ret, slot); 1441213ff4b7SNikolay Borisov parent_level = btrfs_header_level(*eb_ret); 1442213ff4b7SNikolay Borisov btrfs_node_key_to_cpu(*eb_ret, &first_key, slot); 1443c8c42864SChris Mason 1444b246666eSFilipe Manana /* 1445b246666eSFilipe Manana * If we need to read an extent buffer from disk and we are holding locks 1446b246666eSFilipe Manana * on upper level nodes, we unlock all the upper nodes before reading the 1447b246666eSFilipe Manana * extent buffer, and then return -EAGAIN to the caller as it needs to 1448b246666eSFilipe Manana * restart the search. We don't release the lock on the current level 1449b246666eSFilipe Manana * because we need to walk this node to figure out which blocks to read. 1450b246666eSFilipe Manana */ 14510b246afaSJeff Mahoney tmp = find_extent_buffer(fs_info, blocknr); 1452cb44921aSChris Mason if (tmp) { 1453ace75066SFilipe Manana if (p->reada == READA_FORWARD_ALWAYS) 1454ace75066SFilipe Manana reada_for_search(fs_info, p, level, slot, key->objectid); 1455ace75066SFilipe Manana 1456b9fab919SChris Mason /* first we do an atomic uptodate check */ 1457b9fab919SChris Mason if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { 1458448de471SQu Wenruo /* 1459448de471SQu Wenruo * Do extra check for first_key, eb can be stale due to 1460448de471SQu Wenruo * being cached, read from scrub, or have multiple 1461448de471SQu Wenruo * parents (shared tree blocks). 1462448de471SQu Wenruo */ 1463e064d5e9SDavid Sterba if (btrfs_verify_level_key(tmp, 1464448de471SQu Wenruo parent_level - 1, &first_key, gen)) { 1465448de471SQu Wenruo free_extent_buffer(tmp); 1466448de471SQu Wenruo return -EUCLEAN; 1467448de471SQu Wenruo } 1468c8c42864SChris Mason *eb_ret = tmp; 1469c8c42864SChris Mason return 0; 1470c8c42864SChris Mason } 1471bdf7c00eSJosef Bacik 1472857bc13fSJosef Bacik if (p->nowait) { 1473857bc13fSJosef Bacik free_extent_buffer(tmp); 1474857bc13fSJosef Bacik return -EAGAIN; 1475857bc13fSJosef Bacik } 1476857bc13fSJosef Bacik 1477b246666eSFilipe Manana if (unlock_up) 1478b246666eSFilipe Manana btrfs_unlock_up_safe(p, level + 1); 1479b246666eSFilipe Manana 1480b9fab919SChris Mason /* now we're allowed to do a blocking uptodate check */ 14816a2e9dc4SFilipe Manana ret = btrfs_read_extent_buffer(tmp, gen, parent_level - 1, &first_key); 14829a4ffa1bSQu Wenruo if (ret) { 1483cb44921aSChris Mason free_extent_buffer(tmp); 1484b3b4aa74SDavid Sterba btrfs_release_path(p); 1485cb44921aSChris Mason return -EIO; 1486cb44921aSChris Mason } 148788c602abSQu Wenruo if (btrfs_check_eb_owner(tmp, root->root_key.objectid)) { 148888c602abSQu Wenruo free_extent_buffer(tmp); 148988c602abSQu Wenruo btrfs_release_path(p); 149088c602abSQu Wenruo return -EUCLEAN; 149188c602abSQu Wenruo } 1492b246666eSFilipe Manana 1493b246666eSFilipe Manana if (unlock_up) 1494b246666eSFilipe Manana ret = -EAGAIN; 1495b246666eSFilipe Manana 1496b246666eSFilipe Manana goto out; 1497857bc13fSJosef Bacik } else if (p->nowait) { 1498857bc13fSJosef Bacik return -EAGAIN; 14999a4ffa1bSQu Wenruo } 1500c8c42864SChris Mason 1501b246666eSFilipe Manana if (unlock_up) { 15028c594ea8SChris Mason btrfs_unlock_up_safe(p, level + 1); 15034bb59055SFilipe Manana ret = -EAGAIN; 15044bb59055SFilipe Manana } else { 15054bb59055SFilipe Manana ret = 0; 15064bb59055SFilipe Manana } 15078c594ea8SChris Mason 1508e4058b54SDavid Sterba if (p->reada != READA_NONE) 15092ff7e61eSJeff Mahoney reada_for_search(fs_info, p, level, slot, key->objectid); 1510c8c42864SChris Mason 15111b7ec85eSJosef Bacik tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid, 15121b7ec85eSJosef Bacik gen, parent_level - 1, &first_key); 15134eb150d6SQu Wenruo if (IS_ERR(tmp)) { 15144eb150d6SQu Wenruo btrfs_release_path(p); 15154eb150d6SQu Wenruo return PTR_ERR(tmp); 15164eb150d6SQu Wenruo } 151776a05b35SChris Mason /* 151876a05b35SChris Mason * If the read above didn't mark this buffer up to date, 151976a05b35SChris Mason * it will never end up being up to date. Set ret to EIO now 152076a05b35SChris Mason * and give up so that our caller doesn't loop forever 152176a05b35SChris Mason * on our EAGAINs. 152276a05b35SChris Mason */ 1523e6a1d6fdSLiu Bo if (!extent_buffer_uptodate(tmp)) 152476a05b35SChris Mason ret = -EIO; 152502a3307aSLiu Bo 1526b246666eSFilipe Manana out: 15274bb59055SFilipe Manana if (ret == 0) { 15284bb59055SFilipe Manana *eb_ret = tmp; 15294bb59055SFilipe Manana } else { 15304bb59055SFilipe Manana free_extent_buffer(tmp); 153102a3307aSLiu Bo btrfs_release_path(p); 15324bb59055SFilipe Manana } 15334bb59055SFilipe Manana 153476a05b35SChris Mason return ret; 1535c8c42864SChris Mason } 1536c8c42864SChris Mason 1537c8c42864SChris Mason /* 1538c8c42864SChris Mason * helper function for btrfs_search_slot. This does all of the checks 1539c8c42864SChris Mason * for node-level blocks and does any balancing required based on 1540c8c42864SChris Mason * the ins_len. 1541c8c42864SChris Mason * 1542c8c42864SChris Mason * If no extra work was required, zero is returned. If we had to 1543c8c42864SChris Mason * drop the path, -EAGAIN is returned and btrfs_search_slot must 1544c8c42864SChris Mason * start over 1545c8c42864SChris Mason */ 1546c8c42864SChris Mason static int 1547c8c42864SChris Mason setup_nodes_for_search(struct btrfs_trans_handle *trans, 1548c8c42864SChris Mason struct btrfs_root *root, struct btrfs_path *p, 1549bd681513SChris Mason struct extent_buffer *b, int level, int ins_len, 1550bd681513SChris Mason int *write_lock_level) 1551c8c42864SChris Mason { 15520b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 155395b982deSNikolay Borisov int ret = 0; 15540b246afaSJeff Mahoney 1555c8c42864SChris Mason if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= 15560b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { 1557c8c42864SChris Mason 1558bd681513SChris Mason if (*write_lock_level < level + 1) { 1559bd681513SChris Mason *write_lock_level = level + 1; 1560bd681513SChris Mason btrfs_release_path(p); 156195b982deSNikolay Borisov return -EAGAIN; 1562bd681513SChris Mason } 1563bd681513SChris Mason 1564bfb484d9SJosef Bacik reada_for_balance(p, level); 156595b982deSNikolay Borisov ret = split_node(trans, root, p, level); 1566c8c42864SChris Mason 1567c8c42864SChris Mason b = p->nodes[level]; 1568c8c42864SChris Mason } else if (ins_len < 0 && btrfs_header_nritems(b) < 15690b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { 1570c8c42864SChris Mason 1571bd681513SChris Mason if (*write_lock_level < level + 1) { 1572bd681513SChris Mason *write_lock_level = level + 1; 1573bd681513SChris Mason btrfs_release_path(p); 157495b982deSNikolay Borisov return -EAGAIN; 1575bd681513SChris Mason } 1576bd681513SChris Mason 1577bfb484d9SJosef Bacik reada_for_balance(p, level); 157895b982deSNikolay Borisov ret = balance_level(trans, root, p, level); 157995b982deSNikolay Borisov if (ret) 158095b982deSNikolay Borisov return ret; 1581c8c42864SChris Mason 1582c8c42864SChris Mason b = p->nodes[level]; 1583c8c42864SChris Mason if (!b) { 1584b3b4aa74SDavid Sterba btrfs_release_path(p); 158595b982deSNikolay Borisov return -EAGAIN; 1586c8c42864SChris Mason } 1587c8c42864SChris Mason BUG_ON(btrfs_header_nritems(b) == 1); 1588c8c42864SChris Mason } 1589c8c42864SChris Mason return ret; 1590c8c42864SChris Mason } 1591c8c42864SChris Mason 1592381cf658SDavid Sterba int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, 1593e33d5c3dSKelley Nielsen u64 iobjectid, u64 ioff, u8 key_type, 1594e33d5c3dSKelley Nielsen struct btrfs_key *found_key) 1595e33d5c3dSKelley Nielsen { 1596e33d5c3dSKelley Nielsen int ret; 1597e33d5c3dSKelley Nielsen struct btrfs_key key; 1598e33d5c3dSKelley Nielsen struct extent_buffer *eb; 1599381cf658SDavid Sterba 1600381cf658SDavid Sterba ASSERT(path); 16011d4c08e0SDavid Sterba ASSERT(found_key); 1602e33d5c3dSKelley Nielsen 1603e33d5c3dSKelley Nielsen key.type = key_type; 1604e33d5c3dSKelley Nielsen key.objectid = iobjectid; 1605e33d5c3dSKelley Nielsen key.offset = ioff; 1606e33d5c3dSKelley Nielsen 1607e33d5c3dSKelley Nielsen ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); 16081d4c08e0SDavid Sterba if (ret < 0) 1609e33d5c3dSKelley Nielsen return ret; 1610e33d5c3dSKelley Nielsen 1611e33d5c3dSKelley Nielsen eb = path->nodes[0]; 1612e33d5c3dSKelley Nielsen if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { 1613e33d5c3dSKelley Nielsen ret = btrfs_next_leaf(fs_root, path); 1614e33d5c3dSKelley Nielsen if (ret) 1615e33d5c3dSKelley Nielsen return ret; 1616e33d5c3dSKelley Nielsen eb = path->nodes[0]; 1617e33d5c3dSKelley Nielsen } 1618e33d5c3dSKelley Nielsen 1619e33d5c3dSKelley Nielsen btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); 1620e33d5c3dSKelley Nielsen if (found_key->type != key.type || 1621e33d5c3dSKelley Nielsen found_key->objectid != key.objectid) 1622e33d5c3dSKelley Nielsen return 1; 1623e33d5c3dSKelley Nielsen 1624e33d5c3dSKelley Nielsen return 0; 1625e33d5c3dSKelley Nielsen } 1626e33d5c3dSKelley Nielsen 16271fc28d8eSLiu Bo static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, 16281fc28d8eSLiu Bo struct btrfs_path *p, 16291fc28d8eSLiu Bo int write_lock_level) 16301fc28d8eSLiu Bo { 16311fc28d8eSLiu Bo struct extent_buffer *b; 1632120de408SJosef Bacik int root_lock = 0; 16331fc28d8eSLiu Bo int level = 0; 16341fc28d8eSLiu Bo 16351fc28d8eSLiu Bo if (p->search_commit_root) { 16361fc28d8eSLiu Bo b = root->commit_root; 163767439dadSDavid Sterba atomic_inc(&b->refs); 16381fc28d8eSLiu Bo level = btrfs_header_level(b); 1639f9ddfd05SLiu Bo /* 1640f9ddfd05SLiu Bo * Ensure that all callers have set skip_locking when 1641f9ddfd05SLiu Bo * p->search_commit_root = 1. 1642f9ddfd05SLiu Bo */ 1643f9ddfd05SLiu Bo ASSERT(p->skip_locking == 1); 16441fc28d8eSLiu Bo 16451fc28d8eSLiu Bo goto out; 16461fc28d8eSLiu Bo } 16471fc28d8eSLiu Bo 16481fc28d8eSLiu Bo if (p->skip_locking) { 16491fc28d8eSLiu Bo b = btrfs_root_node(root); 16501fc28d8eSLiu Bo level = btrfs_header_level(b); 16511fc28d8eSLiu Bo goto out; 16521fc28d8eSLiu Bo } 16531fc28d8eSLiu Bo 1654120de408SJosef Bacik /* We try very hard to do read locks on the root */ 1655120de408SJosef Bacik root_lock = BTRFS_READ_LOCK; 1656120de408SJosef Bacik 16571fc28d8eSLiu Bo /* 1658662c653bSLiu Bo * If the level is set to maximum, we can skip trying to get the read 1659662c653bSLiu Bo * lock. 1660662c653bSLiu Bo */ 1661662c653bSLiu Bo if (write_lock_level < BTRFS_MAX_LEVEL) { 1662662c653bSLiu Bo /* 1663662c653bSLiu Bo * We don't know the level of the root node until we actually 1664662c653bSLiu Bo * have it read locked 16651fc28d8eSLiu Bo */ 1666857bc13fSJosef Bacik if (p->nowait) { 1667857bc13fSJosef Bacik b = btrfs_try_read_lock_root_node(root); 1668857bc13fSJosef Bacik if (IS_ERR(b)) 1669857bc13fSJosef Bacik return b; 1670857bc13fSJosef Bacik } else { 16711bb96598SJosef Bacik b = btrfs_read_lock_root_node(root); 1672857bc13fSJosef Bacik } 16731fc28d8eSLiu Bo level = btrfs_header_level(b); 16741fc28d8eSLiu Bo if (level > write_lock_level) 16751fc28d8eSLiu Bo goto out; 16761fc28d8eSLiu Bo 1677662c653bSLiu Bo /* Whoops, must trade for write lock */ 16781fc28d8eSLiu Bo btrfs_tree_read_unlock(b); 16791fc28d8eSLiu Bo free_extent_buffer(b); 1680662c653bSLiu Bo } 1681662c653bSLiu Bo 16821fc28d8eSLiu Bo b = btrfs_lock_root_node(root); 16831fc28d8eSLiu Bo root_lock = BTRFS_WRITE_LOCK; 16841fc28d8eSLiu Bo 16851fc28d8eSLiu Bo /* The level might have changed, check again */ 16861fc28d8eSLiu Bo level = btrfs_header_level(b); 16871fc28d8eSLiu Bo 16881fc28d8eSLiu Bo out: 1689120de408SJosef Bacik /* 1690120de408SJosef Bacik * The root may have failed to write out at some point, and thus is no 1691120de408SJosef Bacik * longer valid, return an error in this case. 1692120de408SJosef Bacik */ 1693120de408SJosef Bacik if (!extent_buffer_uptodate(b)) { 1694120de408SJosef Bacik if (root_lock) 1695120de408SJosef Bacik btrfs_tree_unlock_rw(b, root_lock); 1696120de408SJosef Bacik free_extent_buffer(b); 1697120de408SJosef Bacik return ERR_PTR(-EIO); 1698120de408SJosef Bacik } 1699120de408SJosef Bacik 17001fc28d8eSLiu Bo p->nodes[level] = b; 17011fc28d8eSLiu Bo if (!p->skip_locking) 17021fc28d8eSLiu Bo p->locks[level] = root_lock; 17031fc28d8eSLiu Bo /* 17041fc28d8eSLiu Bo * Callers are responsible for dropping b's references. 17051fc28d8eSLiu Bo */ 17061fc28d8eSLiu Bo return b; 17071fc28d8eSLiu Bo } 17081fc28d8eSLiu Bo 1709d96b3424SFilipe Manana /* 1710d96b3424SFilipe Manana * Replace the extent buffer at the lowest level of the path with a cloned 1711d96b3424SFilipe Manana * version. The purpose is to be able to use it safely, after releasing the 1712d96b3424SFilipe Manana * commit root semaphore, even if relocation is happening in parallel, the 1713d96b3424SFilipe Manana * transaction used for relocation is committed and the extent buffer is 1714d96b3424SFilipe Manana * reallocated in the next transaction. 1715d96b3424SFilipe Manana * 1716d96b3424SFilipe Manana * This is used in a context where the caller does not prevent transaction 1717d96b3424SFilipe Manana * commits from happening, either by holding a transaction handle or holding 1718d96b3424SFilipe Manana * some lock, while it's doing searches through a commit root. 1719d96b3424SFilipe Manana * At the moment it's only used for send operations. 1720d96b3424SFilipe Manana */ 1721d96b3424SFilipe Manana static int finish_need_commit_sem_search(struct btrfs_path *path) 1722d96b3424SFilipe Manana { 1723d96b3424SFilipe Manana const int i = path->lowest_level; 1724d96b3424SFilipe Manana const int slot = path->slots[i]; 1725d96b3424SFilipe Manana struct extent_buffer *lowest = path->nodes[i]; 1726d96b3424SFilipe Manana struct extent_buffer *clone; 1727d96b3424SFilipe Manana 1728d96b3424SFilipe Manana ASSERT(path->need_commit_sem); 1729d96b3424SFilipe Manana 1730d96b3424SFilipe Manana if (!lowest) 1731d96b3424SFilipe Manana return 0; 1732d96b3424SFilipe Manana 1733d96b3424SFilipe Manana lockdep_assert_held_read(&lowest->fs_info->commit_root_sem); 1734d96b3424SFilipe Manana 1735d96b3424SFilipe Manana clone = btrfs_clone_extent_buffer(lowest); 1736d96b3424SFilipe Manana if (!clone) 1737d96b3424SFilipe Manana return -ENOMEM; 1738d96b3424SFilipe Manana 1739d96b3424SFilipe Manana btrfs_release_path(path); 1740d96b3424SFilipe Manana path->nodes[i] = clone; 1741d96b3424SFilipe Manana path->slots[i] = slot; 1742d96b3424SFilipe Manana 1743d96b3424SFilipe Manana return 0; 1744d96b3424SFilipe Manana } 17451fc28d8eSLiu Bo 1746e2e58d0fSFilipe Manana static inline int search_for_key_slot(struct extent_buffer *eb, 1747e2e58d0fSFilipe Manana int search_low_slot, 1748e2e58d0fSFilipe Manana const struct btrfs_key *key, 1749e2e58d0fSFilipe Manana int prev_cmp, 1750e2e58d0fSFilipe Manana int *slot) 1751e2e58d0fSFilipe Manana { 1752e2e58d0fSFilipe Manana /* 1753e2e58d0fSFilipe Manana * If a previous call to btrfs_bin_search() on a parent node returned an 1754e2e58d0fSFilipe Manana * exact match (prev_cmp == 0), we can safely assume the target key will 1755e2e58d0fSFilipe Manana * always be at slot 0 on lower levels, since each key pointer 1756e2e58d0fSFilipe Manana * (struct btrfs_key_ptr) refers to the lowest key accessible from the 1757e2e58d0fSFilipe Manana * subtree it points to. Thus we can skip searching lower levels. 1758e2e58d0fSFilipe Manana */ 1759e2e58d0fSFilipe Manana if (prev_cmp == 0) { 1760e2e58d0fSFilipe Manana *slot = 0; 1761e2e58d0fSFilipe Manana return 0; 1762e2e58d0fSFilipe Manana } 1763e2e58d0fSFilipe Manana 1764e2e58d0fSFilipe Manana return generic_bin_search(eb, search_low_slot, key, slot); 1765e2e58d0fSFilipe Manana } 1766e2e58d0fSFilipe Manana 1767109324cfSFilipe Manana static int search_leaf(struct btrfs_trans_handle *trans, 1768109324cfSFilipe Manana struct btrfs_root *root, 1769109324cfSFilipe Manana const struct btrfs_key *key, 1770109324cfSFilipe Manana struct btrfs_path *path, 1771109324cfSFilipe Manana int ins_len, 1772109324cfSFilipe Manana int prev_cmp) 1773109324cfSFilipe Manana { 1774109324cfSFilipe Manana struct extent_buffer *leaf = path->nodes[0]; 1775109324cfSFilipe Manana int leaf_free_space = -1; 1776109324cfSFilipe Manana int search_low_slot = 0; 1777109324cfSFilipe Manana int ret; 1778109324cfSFilipe Manana bool do_bin_search = true; 1779109324cfSFilipe Manana 1780109324cfSFilipe Manana /* 1781109324cfSFilipe Manana * If we are doing an insertion, the leaf has enough free space and the 1782109324cfSFilipe Manana * destination slot for the key is not slot 0, then we can unlock our 1783109324cfSFilipe Manana * write lock on the parent, and any other upper nodes, before doing the 1784109324cfSFilipe Manana * binary search on the leaf (with search_for_key_slot()), allowing other 1785109324cfSFilipe Manana * tasks to lock the parent and any other upper nodes. 1786109324cfSFilipe Manana */ 1787109324cfSFilipe Manana if (ins_len > 0) { 1788109324cfSFilipe Manana /* 1789109324cfSFilipe Manana * Cache the leaf free space, since we will need it later and it 1790109324cfSFilipe Manana * will not change until then. 1791109324cfSFilipe Manana */ 1792109324cfSFilipe Manana leaf_free_space = btrfs_leaf_free_space(leaf); 1793109324cfSFilipe Manana 1794109324cfSFilipe Manana /* 1795109324cfSFilipe Manana * !path->locks[1] means we have a single node tree, the leaf is 1796109324cfSFilipe Manana * the root of the tree. 1797109324cfSFilipe Manana */ 1798109324cfSFilipe Manana if (path->locks[1] && leaf_free_space >= ins_len) { 1799109324cfSFilipe Manana struct btrfs_disk_key first_key; 1800109324cfSFilipe Manana 1801109324cfSFilipe Manana ASSERT(btrfs_header_nritems(leaf) > 0); 1802109324cfSFilipe Manana btrfs_item_key(leaf, &first_key, 0); 1803109324cfSFilipe Manana 1804109324cfSFilipe Manana /* 1805109324cfSFilipe Manana * Doing the extra comparison with the first key is cheap, 1806109324cfSFilipe Manana * taking into account that the first key is very likely 1807109324cfSFilipe Manana * already in a cache line because it immediately follows 1808109324cfSFilipe Manana * the extent buffer's header and we have recently accessed 1809109324cfSFilipe Manana * the header's level field. 1810109324cfSFilipe Manana */ 1811109324cfSFilipe Manana ret = comp_keys(&first_key, key); 1812109324cfSFilipe Manana if (ret < 0) { 1813109324cfSFilipe Manana /* 1814109324cfSFilipe Manana * The first key is smaller than the key we want 1815109324cfSFilipe Manana * to insert, so we are safe to unlock all upper 1816109324cfSFilipe Manana * nodes and we have to do the binary search. 1817109324cfSFilipe Manana * 1818109324cfSFilipe Manana * We do use btrfs_unlock_up_safe() and not 1819109324cfSFilipe Manana * unlock_up() because the later does not unlock 1820109324cfSFilipe Manana * nodes with a slot of 0 - we can safely unlock 1821109324cfSFilipe Manana * any node even if its slot is 0 since in this 1822109324cfSFilipe Manana * case the key does not end up at slot 0 of the 1823109324cfSFilipe Manana * leaf and there's no need to split the leaf. 1824109324cfSFilipe Manana */ 1825109324cfSFilipe Manana btrfs_unlock_up_safe(path, 1); 1826109324cfSFilipe Manana search_low_slot = 1; 1827109324cfSFilipe Manana } else { 1828109324cfSFilipe Manana /* 1829109324cfSFilipe Manana * The first key is >= then the key we want to 1830109324cfSFilipe Manana * insert, so we can skip the binary search as 1831109324cfSFilipe Manana * the target key will be at slot 0. 1832109324cfSFilipe Manana * 1833109324cfSFilipe Manana * We can not unlock upper nodes when the key is 1834109324cfSFilipe Manana * less than the first key, because we will need 1835109324cfSFilipe Manana * to update the key at slot 0 of the parent node 1836109324cfSFilipe Manana * and possibly of other upper nodes too. 1837109324cfSFilipe Manana * If the key matches the first key, then we can 1838109324cfSFilipe Manana * unlock all the upper nodes, using 1839109324cfSFilipe Manana * btrfs_unlock_up_safe() instead of unlock_up() 1840109324cfSFilipe Manana * as stated above. 1841109324cfSFilipe Manana */ 1842109324cfSFilipe Manana if (ret == 0) 1843109324cfSFilipe Manana btrfs_unlock_up_safe(path, 1); 1844109324cfSFilipe Manana /* 1845109324cfSFilipe Manana * ret is already 0 or 1, matching the result of 1846109324cfSFilipe Manana * a btrfs_bin_search() call, so there is no need 1847109324cfSFilipe Manana * to adjust it. 1848109324cfSFilipe Manana */ 1849109324cfSFilipe Manana do_bin_search = false; 1850109324cfSFilipe Manana path->slots[0] = 0; 1851109324cfSFilipe Manana } 1852109324cfSFilipe Manana } 1853109324cfSFilipe Manana } 1854109324cfSFilipe Manana 1855109324cfSFilipe Manana if (do_bin_search) { 1856109324cfSFilipe Manana ret = search_for_key_slot(leaf, search_low_slot, key, 1857109324cfSFilipe Manana prev_cmp, &path->slots[0]); 1858109324cfSFilipe Manana if (ret < 0) 1859109324cfSFilipe Manana return ret; 1860109324cfSFilipe Manana } 1861109324cfSFilipe Manana 1862109324cfSFilipe Manana if (ins_len > 0) { 1863109324cfSFilipe Manana /* 1864109324cfSFilipe Manana * Item key already exists. In this case, if we are allowed to 1865109324cfSFilipe Manana * insert the item (for example, in dir_item case, item key 1866109324cfSFilipe Manana * collision is allowed), it will be merged with the original 1867109324cfSFilipe Manana * item. Only the item size grows, no new btrfs item will be 1868109324cfSFilipe Manana * added. If search_for_extension is not set, ins_len already 1869109324cfSFilipe Manana * accounts the size btrfs_item, deduct it here so leaf space 1870109324cfSFilipe Manana * check will be correct. 1871109324cfSFilipe Manana */ 1872109324cfSFilipe Manana if (ret == 0 && !path->search_for_extension) { 1873109324cfSFilipe Manana ASSERT(ins_len >= sizeof(struct btrfs_item)); 1874109324cfSFilipe Manana ins_len -= sizeof(struct btrfs_item); 1875109324cfSFilipe Manana } 1876109324cfSFilipe Manana 1877109324cfSFilipe Manana ASSERT(leaf_free_space >= 0); 1878109324cfSFilipe Manana 1879109324cfSFilipe Manana if (leaf_free_space < ins_len) { 1880109324cfSFilipe Manana int err; 1881109324cfSFilipe Manana 1882109324cfSFilipe Manana err = split_leaf(trans, root, key, path, ins_len, 1883109324cfSFilipe Manana (ret == 0)); 1884bb8e9a60SFilipe Manana ASSERT(err <= 0); 1885bb8e9a60SFilipe Manana if (WARN_ON(err > 0)) 1886bb8e9a60SFilipe Manana err = -EUCLEAN; 1887109324cfSFilipe Manana if (err) 1888109324cfSFilipe Manana ret = err; 1889109324cfSFilipe Manana } 1890109324cfSFilipe Manana } 1891109324cfSFilipe Manana 1892109324cfSFilipe Manana return ret; 1893109324cfSFilipe Manana } 1894109324cfSFilipe Manana 1895c8c42864SChris Mason /* 18964271eceaSNikolay Borisov * btrfs_search_slot - look for a key in a tree and perform necessary 18974271eceaSNikolay Borisov * modifications to preserve tree invariants. 189874123bd7SChris Mason * 18994271eceaSNikolay Borisov * @trans: Handle of transaction, used when modifying the tree 19004271eceaSNikolay Borisov * @p: Holds all btree nodes along the search path 19014271eceaSNikolay Borisov * @root: The root node of the tree 19024271eceaSNikolay Borisov * @key: The key we are looking for 19039a664971Sethanwu * @ins_len: Indicates purpose of search: 19049a664971Sethanwu * >0 for inserts it's size of item inserted (*) 19059a664971Sethanwu * <0 for deletions 19069a664971Sethanwu * 0 for plain searches, not modifying the tree 19079a664971Sethanwu * 19089a664971Sethanwu * (*) If size of item inserted doesn't include 19099a664971Sethanwu * sizeof(struct btrfs_item), then p->search_for_extension must 19109a664971Sethanwu * be set. 19114271eceaSNikolay Borisov * @cow: boolean should CoW operations be performed. Must always be 1 19124271eceaSNikolay Borisov * when modifying the tree. 191397571fd0SChris Mason * 19144271eceaSNikolay Borisov * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. 19154271eceaSNikolay Borisov * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) 19164271eceaSNikolay Borisov * 19174271eceaSNikolay Borisov * If @key is found, 0 is returned and you can find the item in the leaf level 19184271eceaSNikolay Borisov * of the path (level 0) 19194271eceaSNikolay Borisov * 19204271eceaSNikolay Borisov * If @key isn't found, 1 is returned and the leaf level of the path (level 0) 19214271eceaSNikolay Borisov * points to the slot where it should be inserted 19224271eceaSNikolay Borisov * 19234271eceaSNikolay Borisov * If an error is encountered while searching the tree a negative error number 19244271eceaSNikolay Borisov * is returned 192574123bd7SChris Mason */ 1926310712b2SOmar Sandoval int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, 1927310712b2SOmar Sandoval const struct btrfs_key *key, struct btrfs_path *p, 1928310712b2SOmar Sandoval int ins_len, int cow) 1929be0e5c09SChris Mason { 1930d96b3424SFilipe Manana struct btrfs_fs_info *fs_info = root->fs_info; 19315f39d397SChris Mason struct extent_buffer *b; 1932be0e5c09SChris Mason int slot; 1933be0e5c09SChris Mason int ret; 193433c66f43SYan Zheng int err; 1935be0e5c09SChris Mason int level; 1936925baeddSChris Mason int lowest_unlock = 1; 1937bd681513SChris Mason /* everything at write_lock_level or lower must be write locked */ 1938bd681513SChris Mason int write_lock_level = 0; 19399f3a7427SChris Mason u8 lowest_level = 0; 1940f7c79f30SChris Mason int min_write_lock_level; 1941d7396f07SFilipe David Borba Manana int prev_cmp; 19429f3a7427SChris Mason 19436702ed49SChris Mason lowest_level = p->lowest_level; 1944323ac95bSChris Mason WARN_ON(lowest_level && ins_len > 0); 194522b0ebdaSChris Mason WARN_ON(p->nodes[0] != NULL); 1946eb653de1SFilipe David Borba Manana BUG_ON(!cow && ins_len); 194725179201SJosef Bacik 1948857bc13fSJosef Bacik /* 1949857bc13fSJosef Bacik * For now only allow nowait for read only operations. There's no 1950857bc13fSJosef Bacik * strict reason why we can't, we just only need it for reads so it's 1951857bc13fSJosef Bacik * only implemented for reads. 1952857bc13fSJosef Bacik */ 1953857bc13fSJosef Bacik ASSERT(!p->nowait || !cow); 1954857bc13fSJosef Bacik 1955bd681513SChris Mason if (ins_len < 0) { 1956925baeddSChris Mason lowest_unlock = 2; 195765b51a00SChris Mason 1958bd681513SChris Mason /* when we are removing items, we might have to go up to level 1959bd681513SChris Mason * two as we update tree pointers Make sure we keep write 1960bd681513SChris Mason * for those levels as well 1961bd681513SChris Mason */ 1962bd681513SChris Mason write_lock_level = 2; 1963bd681513SChris Mason } else if (ins_len > 0) { 1964bd681513SChris Mason /* 1965bd681513SChris Mason * for inserting items, make sure we have a write lock on 1966bd681513SChris Mason * level 1 so we can update keys 1967bd681513SChris Mason */ 1968bd681513SChris Mason write_lock_level = 1; 1969bd681513SChris Mason } 1970bd681513SChris Mason 1971bd681513SChris Mason if (!cow) 1972bd681513SChris Mason write_lock_level = -1; 1973bd681513SChris Mason 197409a2a8f9SJosef Bacik if (cow && (p->keep_locks || p->lowest_level)) 1975bd681513SChris Mason write_lock_level = BTRFS_MAX_LEVEL; 1976bd681513SChris Mason 1977f7c79f30SChris Mason min_write_lock_level = write_lock_level; 1978f7c79f30SChris Mason 1979d96b3424SFilipe Manana if (p->need_commit_sem) { 1980d96b3424SFilipe Manana ASSERT(p->search_commit_root); 1981857bc13fSJosef Bacik if (p->nowait) { 1982857bc13fSJosef Bacik if (!down_read_trylock(&fs_info->commit_root_sem)) 1983857bc13fSJosef Bacik return -EAGAIN; 1984857bc13fSJosef Bacik } else { 1985d96b3424SFilipe Manana down_read(&fs_info->commit_root_sem); 1986d96b3424SFilipe Manana } 1987857bc13fSJosef Bacik } 1988d96b3424SFilipe Manana 1989bb803951SChris Mason again: 1990d7396f07SFilipe David Borba Manana prev_cmp = -1; 19911fc28d8eSLiu Bo b = btrfs_search_slot_get_root(root, p, write_lock_level); 1992be6821f8SFilipe Manana if (IS_ERR(b)) { 1993be6821f8SFilipe Manana ret = PTR_ERR(b); 1994be6821f8SFilipe Manana goto done; 1995be6821f8SFilipe Manana } 1996925baeddSChris Mason 1997eb60ceacSChris Mason while (b) { 1998f624d976SQu Wenruo int dec = 0; 1999f624d976SQu Wenruo 20005f39d397SChris Mason level = btrfs_header_level(b); 200165b51a00SChris Mason 200202217ed2SChris Mason if (cow) { 20039ea2c7c9SNikolay Borisov bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); 20049ea2c7c9SNikolay Borisov 2005c8c42864SChris Mason /* 2006c8c42864SChris Mason * if we don't really need to cow this block 2007c8c42864SChris Mason * then we don't want to set the path blocking, 2008c8c42864SChris Mason * so we test it here 2009c8c42864SChris Mason */ 20105963ffcaSJosef Bacik if (!should_cow_block(trans, root, b)) 201165b51a00SChris Mason goto cow_done; 20125d4f98a2SYan Zheng 2013bd681513SChris Mason /* 2014bd681513SChris Mason * must have write locks on this node and the 2015bd681513SChris Mason * parent 2016bd681513SChris Mason */ 20175124e00eSJosef Bacik if (level > write_lock_level || 20185124e00eSJosef Bacik (level + 1 > write_lock_level && 20195124e00eSJosef Bacik level + 1 < BTRFS_MAX_LEVEL && 20205124e00eSJosef Bacik p->nodes[level + 1])) { 2021bd681513SChris Mason write_lock_level = level + 1; 2022bd681513SChris Mason btrfs_release_path(p); 2023bd681513SChris Mason goto again; 2024bd681513SChris Mason } 2025bd681513SChris Mason 20269ea2c7c9SNikolay Borisov if (last_level) 20279ea2c7c9SNikolay Borisov err = btrfs_cow_block(trans, root, b, NULL, 0, 20289631e4ccSJosef Bacik &b, 20299631e4ccSJosef Bacik BTRFS_NESTING_COW); 20309ea2c7c9SNikolay Borisov else 203133c66f43SYan Zheng err = btrfs_cow_block(trans, root, b, 2032e20d96d6SChris Mason p->nodes[level + 1], 20339631e4ccSJosef Bacik p->slots[level + 1], &b, 20349631e4ccSJosef Bacik BTRFS_NESTING_COW); 203533c66f43SYan Zheng if (err) { 203633c66f43SYan Zheng ret = err; 203765b51a00SChris Mason goto done; 203854aa1f4dSChris Mason } 203902217ed2SChris Mason } 204065b51a00SChris Mason cow_done: 2041eb60ceacSChris Mason p->nodes[level] = b; 2042b4ce94deSChris Mason 2043b4ce94deSChris Mason /* 2044b4ce94deSChris Mason * we have a lock on b and as long as we aren't changing 2045b4ce94deSChris Mason * the tree, there is no way to for the items in b to change. 2046b4ce94deSChris Mason * It is safe to drop the lock on our parent before we 2047b4ce94deSChris Mason * go through the expensive btree search on b. 2048b4ce94deSChris Mason * 2049eb653de1SFilipe David Borba Manana * If we're inserting or deleting (ins_len != 0), then we might 2050eb653de1SFilipe David Borba Manana * be changing slot zero, which may require changing the parent. 2051eb653de1SFilipe David Borba Manana * So, we can't drop the lock until after we know which slot 2052eb653de1SFilipe David Borba Manana * we're operating on. 2053b4ce94deSChris Mason */ 2054eb653de1SFilipe David Borba Manana if (!ins_len && !p->keep_locks) { 2055eb653de1SFilipe David Borba Manana int u = level + 1; 2056eb653de1SFilipe David Borba Manana 2057eb653de1SFilipe David Borba Manana if (u < BTRFS_MAX_LEVEL && p->locks[u]) { 2058eb653de1SFilipe David Borba Manana btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); 2059eb653de1SFilipe David Borba Manana p->locks[u] = 0; 2060eb653de1SFilipe David Borba Manana } 2061eb653de1SFilipe David Borba Manana } 2062b4ce94deSChris Mason 2063e2e58d0fSFilipe Manana if (level == 0) { 2064109324cfSFilipe Manana if (ins_len > 0) 2065e5e1c174SFilipe Manana ASSERT(write_lock_level >= 1); 2066bd681513SChris Mason 2067109324cfSFilipe Manana ret = search_leaf(trans, root, key, p, ins_len, prev_cmp); 2068459931ecSChris Mason if (!p->search_for_split) 2069f7c79f30SChris Mason unlock_up(p, level, lowest_unlock, 20704b6f8e96SLiu Bo min_write_lock_level, NULL); 207165b51a00SChris Mason goto done; 207265b51a00SChris Mason } 2073e2e58d0fSFilipe Manana 2074e2e58d0fSFilipe Manana ret = search_for_key_slot(b, 0, key, prev_cmp, &slot); 2075e2e58d0fSFilipe Manana if (ret < 0) 2076e2e58d0fSFilipe Manana goto done; 2077e2e58d0fSFilipe Manana prev_cmp = ret; 2078e2e58d0fSFilipe Manana 2079f624d976SQu Wenruo if (ret && slot > 0) { 2080f624d976SQu Wenruo dec = 1; 2081f624d976SQu Wenruo slot--; 2082f624d976SQu Wenruo } 2083f624d976SQu Wenruo p->slots[level] = slot; 2084f624d976SQu Wenruo err = setup_nodes_for_search(trans, root, p, b, level, ins_len, 2085f624d976SQu Wenruo &write_lock_level); 2086f624d976SQu Wenruo if (err == -EAGAIN) 2087f624d976SQu Wenruo goto again; 2088f624d976SQu Wenruo if (err) { 2089f624d976SQu Wenruo ret = err; 2090f624d976SQu Wenruo goto done; 2091f624d976SQu Wenruo } 2092f624d976SQu Wenruo b = p->nodes[level]; 2093f624d976SQu Wenruo slot = p->slots[level]; 2094f624d976SQu Wenruo 2095f624d976SQu Wenruo /* 2096f624d976SQu Wenruo * Slot 0 is special, if we change the key we have to update 2097f624d976SQu Wenruo * the parent pointer which means we must have a write lock on 2098f624d976SQu Wenruo * the parent 2099f624d976SQu Wenruo */ 2100f624d976SQu Wenruo if (slot == 0 && ins_len && write_lock_level < level + 1) { 2101f624d976SQu Wenruo write_lock_level = level + 1; 2102f624d976SQu Wenruo btrfs_release_path(p); 2103f624d976SQu Wenruo goto again; 2104f624d976SQu Wenruo } 2105f624d976SQu Wenruo 2106f624d976SQu Wenruo unlock_up(p, level, lowest_unlock, min_write_lock_level, 2107f624d976SQu Wenruo &write_lock_level); 2108f624d976SQu Wenruo 2109f624d976SQu Wenruo if (level == lowest_level) { 2110f624d976SQu Wenruo if (dec) 2111f624d976SQu Wenruo p->slots[level]++; 2112f624d976SQu Wenruo goto done; 2113f624d976SQu Wenruo } 2114f624d976SQu Wenruo 2115f624d976SQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 2116f624d976SQu Wenruo if (err == -EAGAIN) 2117f624d976SQu Wenruo goto again; 2118f624d976SQu Wenruo if (err) { 2119f624d976SQu Wenruo ret = err; 2120f624d976SQu Wenruo goto done; 2121f624d976SQu Wenruo } 2122f624d976SQu Wenruo 2123f624d976SQu Wenruo if (!p->skip_locking) { 2124f624d976SQu Wenruo level = btrfs_header_level(b); 2125b40130b2SJosef Bacik 2126b40130b2SJosef Bacik btrfs_maybe_reset_lockdep_class(root, b); 2127b40130b2SJosef Bacik 2128f624d976SQu Wenruo if (level <= write_lock_level) { 2129f624d976SQu Wenruo btrfs_tree_lock(b); 2130f624d976SQu Wenruo p->locks[level] = BTRFS_WRITE_LOCK; 2131f624d976SQu Wenruo } else { 2132857bc13fSJosef Bacik if (p->nowait) { 2133857bc13fSJosef Bacik if (!btrfs_try_tree_read_lock(b)) { 2134857bc13fSJosef Bacik free_extent_buffer(b); 2135857bc13fSJosef Bacik ret = -EAGAIN; 2136857bc13fSJosef Bacik goto done; 2137857bc13fSJosef Bacik } 2138857bc13fSJosef Bacik } else { 2139fe596ca3SJosef Bacik btrfs_tree_read_lock(b); 2140857bc13fSJosef Bacik } 2141f624d976SQu Wenruo p->locks[level] = BTRFS_READ_LOCK; 2142f624d976SQu Wenruo } 2143f624d976SQu Wenruo p->nodes[level] = b; 2144f624d976SQu Wenruo } 214565b51a00SChris Mason } 214665b51a00SChris Mason ret = 1; 214765b51a00SChris Mason done: 21485f5bc6b1SFilipe Manana if (ret < 0 && !p->skip_release_on_error) 2149b3b4aa74SDavid Sterba btrfs_release_path(p); 2150d96b3424SFilipe Manana 2151d96b3424SFilipe Manana if (p->need_commit_sem) { 2152d96b3424SFilipe Manana int ret2; 2153d96b3424SFilipe Manana 2154d96b3424SFilipe Manana ret2 = finish_need_commit_sem_search(p); 2155d96b3424SFilipe Manana up_read(&fs_info->commit_root_sem); 2156d96b3424SFilipe Manana if (ret2) 2157d96b3424SFilipe Manana ret = ret2; 2158d96b3424SFilipe Manana } 2159d96b3424SFilipe Manana 2160be0e5c09SChris Mason return ret; 2161be0e5c09SChris Mason } 2162f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); 2163be0e5c09SChris Mason 216474123bd7SChris Mason /* 21655d9e75c4SJan Schmidt * Like btrfs_search_slot, this looks for a key in the given tree. It uses the 21665d9e75c4SJan Schmidt * current state of the tree together with the operations recorded in the tree 21675d9e75c4SJan Schmidt * modification log to search for the key in a previous version of this tree, as 21685d9e75c4SJan Schmidt * denoted by the time_seq parameter. 21695d9e75c4SJan Schmidt * 21705d9e75c4SJan Schmidt * Naturally, there is no support for insert, delete or cow operations. 21715d9e75c4SJan Schmidt * 21725d9e75c4SJan Schmidt * The resulting path and return value will be set up as if we called 21735d9e75c4SJan Schmidt * btrfs_search_slot at that point in time with ins_len and cow both set to 0. 21745d9e75c4SJan Schmidt */ 2175310712b2SOmar Sandoval int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, 21765d9e75c4SJan Schmidt struct btrfs_path *p, u64 time_seq) 21775d9e75c4SJan Schmidt { 21780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 21795d9e75c4SJan Schmidt struct extent_buffer *b; 21805d9e75c4SJan Schmidt int slot; 21815d9e75c4SJan Schmidt int ret; 21825d9e75c4SJan Schmidt int err; 21835d9e75c4SJan Schmidt int level; 21845d9e75c4SJan Schmidt int lowest_unlock = 1; 21855d9e75c4SJan Schmidt u8 lowest_level = 0; 21865d9e75c4SJan Schmidt 21875d9e75c4SJan Schmidt lowest_level = p->lowest_level; 21885d9e75c4SJan Schmidt WARN_ON(p->nodes[0] != NULL); 2189c922b016SStefan Roesch ASSERT(!p->nowait); 21905d9e75c4SJan Schmidt 21915d9e75c4SJan Schmidt if (p->search_commit_root) { 21925d9e75c4SJan Schmidt BUG_ON(time_seq); 21935d9e75c4SJan Schmidt return btrfs_search_slot(NULL, root, key, p, 0, 0); 21945d9e75c4SJan Schmidt } 21955d9e75c4SJan Schmidt 21965d9e75c4SJan Schmidt again: 2197f3a84ccdSFilipe Manana b = btrfs_get_old_root(root, time_seq); 2198315bed43SNikolay Borisov if (!b) { 2199315bed43SNikolay Borisov ret = -EIO; 2200315bed43SNikolay Borisov goto done; 2201315bed43SNikolay Borisov } 22025d9e75c4SJan Schmidt level = btrfs_header_level(b); 22035d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 22045d9e75c4SJan Schmidt 22055d9e75c4SJan Schmidt while (b) { 2206abe9339dSQu Wenruo int dec = 0; 2207abe9339dSQu Wenruo 22085d9e75c4SJan Schmidt level = btrfs_header_level(b); 22095d9e75c4SJan Schmidt p->nodes[level] = b; 22105d9e75c4SJan Schmidt 22115d9e75c4SJan Schmidt /* 22125d9e75c4SJan Schmidt * we have a lock on b and as long as we aren't changing 22135d9e75c4SJan Schmidt * the tree, there is no way to for the items in b to change. 22145d9e75c4SJan Schmidt * It is safe to drop the lock on our parent before we 22155d9e75c4SJan Schmidt * go through the expensive btree search on b. 22165d9e75c4SJan Schmidt */ 22175d9e75c4SJan Schmidt btrfs_unlock_up_safe(p, level + 1); 22185d9e75c4SJan Schmidt 2219995e9a16SNikolay Borisov ret = btrfs_bin_search(b, key, &slot); 2220cbca7d59SFilipe Manana if (ret < 0) 2221cbca7d59SFilipe Manana goto done; 22225d9e75c4SJan Schmidt 2223abe9339dSQu Wenruo if (level == 0) { 2224abe9339dSQu Wenruo p->slots[level] = slot; 2225abe9339dSQu Wenruo unlock_up(p, level, lowest_unlock, 0, NULL); 2226abe9339dSQu Wenruo goto done; 2227abe9339dSQu Wenruo } 2228abe9339dSQu Wenruo 22295d9e75c4SJan Schmidt if (ret && slot > 0) { 22305d9e75c4SJan Schmidt dec = 1; 2231abe9339dSQu Wenruo slot--; 22325d9e75c4SJan Schmidt } 22335d9e75c4SJan Schmidt p->slots[level] = slot; 22345d9e75c4SJan Schmidt unlock_up(p, level, lowest_unlock, 0, NULL); 22355d9e75c4SJan Schmidt 22365d9e75c4SJan Schmidt if (level == lowest_level) { 22375d9e75c4SJan Schmidt if (dec) 22385d9e75c4SJan Schmidt p->slots[level]++; 22395d9e75c4SJan Schmidt goto done; 22405d9e75c4SJan Schmidt } 22415d9e75c4SJan Schmidt 2242abe9339dSQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 22435d9e75c4SJan Schmidt if (err == -EAGAIN) 22445d9e75c4SJan Schmidt goto again; 22455d9e75c4SJan Schmidt if (err) { 22465d9e75c4SJan Schmidt ret = err; 22475d9e75c4SJan Schmidt goto done; 22485d9e75c4SJan Schmidt } 22495d9e75c4SJan Schmidt 22505d9e75c4SJan Schmidt level = btrfs_header_level(b); 22515d9e75c4SJan Schmidt btrfs_tree_read_lock(b); 2252f3a84ccdSFilipe Manana b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq); 2253db7f3436SJosef Bacik if (!b) { 2254db7f3436SJosef Bacik ret = -ENOMEM; 2255db7f3436SJosef Bacik goto done; 2256db7f3436SJosef Bacik } 22575d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 22585d9e75c4SJan Schmidt p->nodes[level] = b; 22595d9e75c4SJan Schmidt } 22605d9e75c4SJan Schmidt ret = 1; 22615d9e75c4SJan Schmidt done: 22625d9e75c4SJan Schmidt if (ret < 0) 22635d9e75c4SJan Schmidt btrfs_release_path(p); 22645d9e75c4SJan Schmidt 22655d9e75c4SJan Schmidt return ret; 22665d9e75c4SJan Schmidt } 22675d9e75c4SJan Schmidt 22685d9e75c4SJan Schmidt /* 22692f38b3e1SArne Jansen * helper to use instead of search slot if no exact match is needed but 22702f38b3e1SArne Jansen * instead the next or previous item should be returned. 22712f38b3e1SArne Jansen * When find_higher is true, the next higher item is returned, the next lower 22722f38b3e1SArne Jansen * otherwise. 22732f38b3e1SArne Jansen * When return_any and find_higher are both true, and no higher item is found, 22742f38b3e1SArne Jansen * return the next lower instead. 22752f38b3e1SArne Jansen * When return_any is true and find_higher is false, and no lower item is found, 22762f38b3e1SArne Jansen * return the next higher instead. 22772f38b3e1SArne Jansen * It returns 0 if any item is found, 1 if none is found (tree empty), and 22782f38b3e1SArne Jansen * < 0 on error 22792f38b3e1SArne Jansen */ 22802f38b3e1SArne Jansen int btrfs_search_slot_for_read(struct btrfs_root *root, 2281310712b2SOmar Sandoval const struct btrfs_key *key, 2282310712b2SOmar Sandoval struct btrfs_path *p, int find_higher, 2283310712b2SOmar Sandoval int return_any) 22842f38b3e1SArne Jansen { 22852f38b3e1SArne Jansen int ret; 22862f38b3e1SArne Jansen struct extent_buffer *leaf; 22872f38b3e1SArne Jansen 22882f38b3e1SArne Jansen again: 22892f38b3e1SArne Jansen ret = btrfs_search_slot(NULL, root, key, p, 0, 0); 22902f38b3e1SArne Jansen if (ret <= 0) 22912f38b3e1SArne Jansen return ret; 22922f38b3e1SArne Jansen /* 22932f38b3e1SArne Jansen * a return value of 1 means the path is at the position where the 22942f38b3e1SArne Jansen * item should be inserted. Normally this is the next bigger item, 22952f38b3e1SArne Jansen * but in case the previous item is the last in a leaf, path points 22962f38b3e1SArne Jansen * to the first free slot in the previous leaf, i.e. at an invalid 22972f38b3e1SArne Jansen * item. 22982f38b3e1SArne Jansen */ 22992f38b3e1SArne Jansen leaf = p->nodes[0]; 23002f38b3e1SArne Jansen 23012f38b3e1SArne Jansen if (find_higher) { 23022f38b3e1SArne Jansen if (p->slots[0] >= btrfs_header_nritems(leaf)) { 23032f38b3e1SArne Jansen ret = btrfs_next_leaf(root, p); 23042f38b3e1SArne Jansen if (ret <= 0) 23052f38b3e1SArne Jansen return ret; 23062f38b3e1SArne Jansen if (!return_any) 23072f38b3e1SArne Jansen return 1; 23082f38b3e1SArne Jansen /* 23092f38b3e1SArne Jansen * no higher item found, return the next 23102f38b3e1SArne Jansen * lower instead 23112f38b3e1SArne Jansen */ 23122f38b3e1SArne Jansen return_any = 0; 23132f38b3e1SArne Jansen find_higher = 0; 23142f38b3e1SArne Jansen btrfs_release_path(p); 23152f38b3e1SArne Jansen goto again; 23162f38b3e1SArne Jansen } 23172f38b3e1SArne Jansen } else { 23182f38b3e1SArne Jansen if (p->slots[0] == 0) { 23192f38b3e1SArne Jansen ret = btrfs_prev_leaf(root, p); 2320e6793769SArne Jansen if (ret < 0) 23212f38b3e1SArne Jansen return ret; 2322e6793769SArne Jansen if (!ret) { 232323c6bf6aSFilipe David Borba Manana leaf = p->nodes[0]; 232423c6bf6aSFilipe David Borba Manana if (p->slots[0] == btrfs_header_nritems(leaf)) 232523c6bf6aSFilipe David Borba Manana p->slots[0]--; 2326e6793769SArne Jansen return 0; 2327e6793769SArne Jansen } 23282f38b3e1SArne Jansen if (!return_any) 23292f38b3e1SArne Jansen return 1; 23302f38b3e1SArne Jansen /* 23312f38b3e1SArne Jansen * no lower item found, return the next 23322f38b3e1SArne Jansen * higher instead 23332f38b3e1SArne Jansen */ 23342f38b3e1SArne Jansen return_any = 0; 23352f38b3e1SArne Jansen find_higher = 1; 23362f38b3e1SArne Jansen btrfs_release_path(p); 23372f38b3e1SArne Jansen goto again; 2338e6793769SArne Jansen } else { 23392f38b3e1SArne Jansen --p->slots[0]; 23402f38b3e1SArne Jansen } 23412f38b3e1SArne Jansen } 23422f38b3e1SArne Jansen return 0; 23432f38b3e1SArne Jansen } 23442f38b3e1SArne Jansen 23452f38b3e1SArne Jansen /* 23460ff40a91SMarcos Paulo de Souza * Execute search and call btrfs_previous_item to traverse backwards if the item 23470ff40a91SMarcos Paulo de Souza * was not found. 23480ff40a91SMarcos Paulo de Souza * 23490ff40a91SMarcos Paulo de Souza * Return 0 if found, 1 if not found and < 0 if error. 23500ff40a91SMarcos Paulo de Souza */ 23510ff40a91SMarcos Paulo de Souza int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key, 23520ff40a91SMarcos Paulo de Souza struct btrfs_path *path) 23530ff40a91SMarcos Paulo de Souza { 23540ff40a91SMarcos Paulo de Souza int ret; 23550ff40a91SMarcos Paulo de Souza 23560ff40a91SMarcos Paulo de Souza ret = btrfs_search_slot(NULL, root, key, path, 0, 0); 23570ff40a91SMarcos Paulo de Souza if (ret > 0) 23580ff40a91SMarcos Paulo de Souza ret = btrfs_previous_item(root, path, key->objectid, key->type); 23590ff40a91SMarcos Paulo de Souza 23600ff40a91SMarcos Paulo de Souza if (ret == 0) 23610ff40a91SMarcos Paulo de Souza btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); 23620ff40a91SMarcos Paulo de Souza 23630ff40a91SMarcos Paulo de Souza return ret; 23640ff40a91SMarcos Paulo de Souza } 23650ff40a91SMarcos Paulo de Souza 2366*43dd529aSDavid Sterba /* 236762142be3SGabriel Niebler * Search for a valid slot for the given path. 236862142be3SGabriel Niebler * 236962142be3SGabriel Niebler * @root: The root node of the tree. 237062142be3SGabriel Niebler * @key: Will contain a valid item if found. 237162142be3SGabriel Niebler * @path: The starting point to validate the slot. 237262142be3SGabriel Niebler * 237362142be3SGabriel Niebler * Return: 0 if the item is valid 237462142be3SGabriel Niebler * 1 if not found 237562142be3SGabriel Niebler * <0 if error. 237662142be3SGabriel Niebler */ 237762142be3SGabriel Niebler int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key, 237862142be3SGabriel Niebler struct btrfs_path *path) 237962142be3SGabriel Niebler { 238062142be3SGabriel Niebler while (1) { 238162142be3SGabriel Niebler int ret; 238262142be3SGabriel Niebler const int slot = path->slots[0]; 238362142be3SGabriel Niebler const struct extent_buffer *leaf = path->nodes[0]; 238462142be3SGabriel Niebler 238562142be3SGabriel Niebler /* This is where we start walking the path. */ 238662142be3SGabriel Niebler if (slot >= btrfs_header_nritems(leaf)) { 238762142be3SGabriel Niebler /* 238862142be3SGabriel Niebler * If we've reached the last slot in this leaf we need 238962142be3SGabriel Niebler * to go to the next leaf and reset the path. 239062142be3SGabriel Niebler */ 239162142be3SGabriel Niebler ret = btrfs_next_leaf(root, path); 239262142be3SGabriel Niebler if (ret) 239362142be3SGabriel Niebler return ret; 239462142be3SGabriel Niebler continue; 239562142be3SGabriel Niebler } 239662142be3SGabriel Niebler /* Store the found, valid item in @key. */ 239762142be3SGabriel Niebler btrfs_item_key_to_cpu(leaf, key, slot); 239862142be3SGabriel Niebler break; 239962142be3SGabriel Niebler } 240062142be3SGabriel Niebler return 0; 240162142be3SGabriel Niebler } 240262142be3SGabriel Niebler 24030ff40a91SMarcos Paulo de Souza /* 240474123bd7SChris Mason * adjust the pointers going up the tree, starting at level 240574123bd7SChris Mason * making sure the right key of each node is points to 'key'. 240674123bd7SChris Mason * This is used after shifting pointers to the left, so it stops 240774123bd7SChris Mason * fixing up pointers when a given leaf/node is not in slot 0 of the 240874123bd7SChris Mason * higher levels 2409aa5d6bedSChris Mason * 241074123bd7SChris Mason */ 2411b167fa91SNikolay Borisov static void fixup_low_keys(struct btrfs_path *path, 24125f39d397SChris Mason struct btrfs_disk_key *key, int level) 2413be0e5c09SChris Mason { 2414be0e5c09SChris Mason int i; 24155f39d397SChris Mason struct extent_buffer *t; 24160e82bcfeSDavid Sterba int ret; 24175f39d397SChris Mason 2418234b63a0SChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 2419be0e5c09SChris Mason int tslot = path->slots[i]; 24200e82bcfeSDavid Sterba 2421eb60ceacSChris Mason if (!path->nodes[i]) 2422be0e5c09SChris Mason break; 24235f39d397SChris Mason t = path->nodes[i]; 2424f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(t, tslot, 242533cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 24260e82bcfeSDavid Sterba BUG_ON(ret < 0); 24275f39d397SChris Mason btrfs_set_node_key(t, key, tslot); 2428d6025579SChris Mason btrfs_mark_buffer_dirty(path->nodes[i]); 2429be0e5c09SChris Mason if (tslot != 0) 2430be0e5c09SChris Mason break; 2431be0e5c09SChris Mason } 2432be0e5c09SChris Mason } 2433be0e5c09SChris Mason 243474123bd7SChris Mason /* 243531840ae1SZheng Yan * update item key. 243631840ae1SZheng Yan * 243731840ae1SZheng Yan * This function isn't completely safe. It's the caller's responsibility 243831840ae1SZheng Yan * that the new key won't break the order 243931840ae1SZheng Yan */ 2440b7a0365eSDaniel Dressler void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, 2441b7a0365eSDaniel Dressler struct btrfs_path *path, 2442310712b2SOmar Sandoval const struct btrfs_key *new_key) 244331840ae1SZheng Yan { 244431840ae1SZheng Yan struct btrfs_disk_key disk_key; 244531840ae1SZheng Yan struct extent_buffer *eb; 244631840ae1SZheng Yan int slot; 244731840ae1SZheng Yan 244831840ae1SZheng Yan eb = path->nodes[0]; 244931840ae1SZheng Yan slot = path->slots[0]; 245031840ae1SZheng Yan if (slot > 0) { 245131840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot - 1); 24527c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { 24537c15d410SQu Wenruo btrfs_crit(fs_info, 24547c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 24557c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 24567c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 24577c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 24587c15d410SQu Wenruo new_key->objectid, new_key->type, 24597c15d410SQu Wenruo new_key->offset); 24607c15d410SQu Wenruo btrfs_print_leaf(eb); 24617c15d410SQu Wenruo BUG(); 24627c15d410SQu Wenruo } 246331840ae1SZheng Yan } 246431840ae1SZheng Yan if (slot < btrfs_header_nritems(eb) - 1) { 246531840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot + 1); 24667c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { 24677c15d410SQu Wenruo btrfs_crit(fs_info, 24687c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 24697c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 24707c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 24717c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 24727c15d410SQu Wenruo new_key->objectid, new_key->type, 24737c15d410SQu Wenruo new_key->offset); 24747c15d410SQu Wenruo btrfs_print_leaf(eb); 24757c15d410SQu Wenruo BUG(); 24767c15d410SQu Wenruo } 247731840ae1SZheng Yan } 247831840ae1SZheng Yan 247931840ae1SZheng Yan btrfs_cpu_key_to_disk(&disk_key, new_key); 248031840ae1SZheng Yan btrfs_set_item_key(eb, &disk_key, slot); 248131840ae1SZheng Yan btrfs_mark_buffer_dirty(eb); 248231840ae1SZheng Yan if (slot == 0) 2483b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 248431840ae1SZheng Yan } 248531840ae1SZheng Yan 248631840ae1SZheng Yan /* 2487d16c702fSQu Wenruo * Check key order of two sibling extent buffers. 2488d16c702fSQu Wenruo * 2489d16c702fSQu Wenruo * Return true if something is wrong. 2490d16c702fSQu Wenruo * Return false if everything is fine. 2491d16c702fSQu Wenruo * 2492d16c702fSQu Wenruo * Tree-checker only works inside one tree block, thus the following 2493d16c702fSQu Wenruo * corruption can not be detected by tree-checker: 2494d16c702fSQu Wenruo * 2495d16c702fSQu Wenruo * Leaf @left | Leaf @right 2496d16c702fSQu Wenruo * -------------------------------------------------------------- 2497d16c702fSQu Wenruo * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | 2498d16c702fSQu Wenruo * 2499d16c702fSQu Wenruo * Key f6 in leaf @left itself is valid, but not valid when the next 2500d16c702fSQu Wenruo * key in leaf @right is 7. 2501d16c702fSQu Wenruo * This can only be checked at tree block merge time. 2502d16c702fSQu Wenruo * And since tree checker has ensured all key order in each tree block 2503d16c702fSQu Wenruo * is correct, we only need to bother the last key of @left and the first 2504d16c702fSQu Wenruo * key of @right. 2505d16c702fSQu Wenruo */ 2506d16c702fSQu Wenruo static bool check_sibling_keys(struct extent_buffer *left, 2507d16c702fSQu Wenruo struct extent_buffer *right) 2508d16c702fSQu Wenruo { 2509d16c702fSQu Wenruo struct btrfs_key left_last; 2510d16c702fSQu Wenruo struct btrfs_key right_first; 2511d16c702fSQu Wenruo int level = btrfs_header_level(left); 2512d16c702fSQu Wenruo int nr_left = btrfs_header_nritems(left); 2513d16c702fSQu Wenruo int nr_right = btrfs_header_nritems(right); 2514d16c702fSQu Wenruo 2515d16c702fSQu Wenruo /* No key to check in one of the tree blocks */ 2516d16c702fSQu Wenruo if (!nr_left || !nr_right) 2517d16c702fSQu Wenruo return false; 2518d16c702fSQu Wenruo 2519d16c702fSQu Wenruo if (level) { 2520d16c702fSQu Wenruo btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); 2521d16c702fSQu Wenruo btrfs_node_key_to_cpu(right, &right_first, 0); 2522d16c702fSQu Wenruo } else { 2523d16c702fSQu Wenruo btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); 2524d16c702fSQu Wenruo btrfs_item_key_to_cpu(right, &right_first, 0); 2525d16c702fSQu Wenruo } 2526d16c702fSQu Wenruo 2527d16c702fSQu Wenruo if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) { 2528d16c702fSQu Wenruo btrfs_crit(left->fs_info, 2529d16c702fSQu Wenruo "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", 2530d16c702fSQu Wenruo left_last.objectid, left_last.type, 2531d16c702fSQu Wenruo left_last.offset, right_first.objectid, 2532d16c702fSQu Wenruo right_first.type, right_first.offset); 2533d16c702fSQu Wenruo return true; 2534d16c702fSQu Wenruo } 2535d16c702fSQu Wenruo return false; 2536d16c702fSQu Wenruo } 2537d16c702fSQu Wenruo 2538d16c702fSQu Wenruo /* 253974123bd7SChris Mason * try to push data from one node into the next node left in the 254079f95c82SChris Mason * tree. 2541aa5d6bedSChris Mason * 2542aa5d6bedSChris Mason * returns 0 if some ptrs were pushed left, < 0 if there was some horrible 2543aa5d6bedSChris Mason * error, and > 0 if there was no room in the left hand block. 254474123bd7SChris Mason */ 254598ed5174SChris Mason static int push_node_left(struct btrfs_trans_handle *trans, 25462ff7e61eSJeff Mahoney struct extent_buffer *dst, 2547971a1f66SChris Mason struct extent_buffer *src, int empty) 2548be0e5c09SChris Mason { 2549d30a668fSDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 2550be0e5c09SChris Mason int push_items = 0; 2551bb803951SChris Mason int src_nritems; 2552bb803951SChris Mason int dst_nritems; 2553aa5d6bedSChris Mason int ret = 0; 2554be0e5c09SChris Mason 25555f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 25565f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 25570b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 25587bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 25597bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 256054aa1f4dSChris Mason 2561bce4eae9SChris Mason if (!empty && src_nritems <= 8) 2562971a1f66SChris Mason return 1; 2563971a1f66SChris Mason 2564d397712bSChris Mason if (push_items <= 0) 2565be0e5c09SChris Mason return 1; 2566be0e5c09SChris Mason 2567bce4eae9SChris Mason if (empty) { 2568971a1f66SChris Mason push_items = min(src_nritems, push_items); 2569bce4eae9SChris Mason if (push_items < src_nritems) { 2570bce4eae9SChris Mason /* leave at least 8 pointers in the node if 2571bce4eae9SChris Mason * we aren't going to empty it 2572bce4eae9SChris Mason */ 2573bce4eae9SChris Mason if (src_nritems - push_items < 8) { 2574bce4eae9SChris Mason if (push_items <= 8) 2575bce4eae9SChris Mason return 1; 2576bce4eae9SChris Mason push_items -= 8; 2577bce4eae9SChris Mason } 2578bce4eae9SChris Mason } 2579bce4eae9SChris Mason } else 2580bce4eae9SChris Mason push_items = min(src_nritems - 8, push_items); 258179f95c82SChris Mason 2582d16c702fSQu Wenruo /* dst is the left eb, src is the middle eb */ 2583d16c702fSQu Wenruo if (check_sibling_keys(dst, src)) { 2584d16c702fSQu Wenruo ret = -EUCLEAN; 2585d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 2586d16c702fSQu Wenruo return ret; 2587d16c702fSQu Wenruo } 2588f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); 25895de865eeSFilipe David Borba Manana if (ret) { 259066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 25915de865eeSFilipe David Borba Manana return ret; 25925de865eeSFilipe David Borba Manana } 25935f39d397SChris Mason copy_extent_buffer(dst, src, 25945f39d397SChris Mason btrfs_node_key_ptr_offset(dst_nritems), 25955f39d397SChris Mason btrfs_node_key_ptr_offset(0), 2596123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 25975f39d397SChris Mason 2598bb803951SChris Mason if (push_items < src_nritems) { 259957911b8bSJan Schmidt /* 2600f3a84ccdSFilipe Manana * Don't call btrfs_tree_mod_log_insert_move() here, key removal 2601f3a84ccdSFilipe Manana * was already fully logged by btrfs_tree_mod_log_eb_copy() above. 260257911b8bSJan Schmidt */ 26035f39d397SChris Mason memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), 26045f39d397SChris Mason btrfs_node_key_ptr_offset(push_items), 2605e2fa7227SChris Mason (src_nritems - push_items) * 2606123abc88SChris Mason sizeof(struct btrfs_key_ptr)); 2607bb803951SChris Mason } 26085f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 26095f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 26105f39d397SChris Mason btrfs_mark_buffer_dirty(src); 26115f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 261231840ae1SZheng Yan 2613bb803951SChris Mason return ret; 2614be0e5c09SChris Mason } 2615be0e5c09SChris Mason 261697571fd0SChris Mason /* 261779f95c82SChris Mason * try to push data from one node into the next node right in the 261879f95c82SChris Mason * tree. 261979f95c82SChris Mason * 262079f95c82SChris Mason * returns 0 if some ptrs were pushed, < 0 if there was some horrible 262179f95c82SChris Mason * error, and > 0 if there was no room in the right hand block. 262279f95c82SChris Mason * 262379f95c82SChris Mason * this will only push up to 1/2 the contents of the left node over 262479f95c82SChris Mason */ 26255f39d397SChris Mason static int balance_node_right(struct btrfs_trans_handle *trans, 26265f39d397SChris Mason struct extent_buffer *dst, 26275f39d397SChris Mason struct extent_buffer *src) 262879f95c82SChris Mason { 262955d32ed8SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 263079f95c82SChris Mason int push_items = 0; 263179f95c82SChris Mason int max_push; 263279f95c82SChris Mason int src_nritems; 263379f95c82SChris Mason int dst_nritems; 263479f95c82SChris Mason int ret = 0; 263579f95c82SChris Mason 26367bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 26377bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 26387bb86316SChris Mason 26395f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 26405f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 26410b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 2642d397712bSChris Mason if (push_items <= 0) 264379f95c82SChris Mason return 1; 2644bce4eae9SChris Mason 2645d397712bSChris Mason if (src_nritems < 4) 2646bce4eae9SChris Mason return 1; 264779f95c82SChris Mason 264879f95c82SChris Mason max_push = src_nritems / 2 + 1; 264979f95c82SChris Mason /* don't try to empty the node */ 2650d397712bSChris Mason if (max_push >= src_nritems) 265179f95c82SChris Mason return 1; 2652252c38f0SYan 265379f95c82SChris Mason if (max_push < push_items) 265479f95c82SChris Mason push_items = max_push; 265579f95c82SChris Mason 2656d16c702fSQu Wenruo /* dst is the right eb, src is the middle eb */ 2657d16c702fSQu Wenruo if (check_sibling_keys(src, dst)) { 2658d16c702fSQu Wenruo ret = -EUCLEAN; 2659d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 2660d16c702fSQu Wenruo return ret; 2661d16c702fSQu Wenruo } 2662f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); 2663bf1d3425SDavid Sterba BUG_ON(ret < 0); 26645f39d397SChris Mason memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), 26655f39d397SChris Mason btrfs_node_key_ptr_offset(0), 26665f39d397SChris Mason (dst_nritems) * 26675f39d397SChris Mason sizeof(struct btrfs_key_ptr)); 2668d6025579SChris Mason 2669f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, 2670ed874f0dSDavid Sterba push_items); 26715de865eeSFilipe David Borba Manana if (ret) { 267266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 26735de865eeSFilipe David Borba Manana return ret; 26745de865eeSFilipe David Borba Manana } 26755f39d397SChris Mason copy_extent_buffer(dst, src, 26765f39d397SChris Mason btrfs_node_key_ptr_offset(0), 26775f39d397SChris Mason btrfs_node_key_ptr_offset(src_nritems - push_items), 2678123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 267979f95c82SChris Mason 26805f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 26815f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 268279f95c82SChris Mason 26835f39d397SChris Mason btrfs_mark_buffer_dirty(src); 26845f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 268531840ae1SZheng Yan 268679f95c82SChris Mason return ret; 268779f95c82SChris Mason } 268879f95c82SChris Mason 268979f95c82SChris Mason /* 269097571fd0SChris Mason * helper function to insert a new root level in the tree. 269197571fd0SChris Mason * A new node is allocated, and a single item is inserted to 269297571fd0SChris Mason * point to the existing root 2693aa5d6bedSChris Mason * 2694aa5d6bedSChris Mason * returns zero on success or < 0 on failure. 269597571fd0SChris Mason */ 2696d397712bSChris Mason static noinline int insert_new_root(struct btrfs_trans_handle *trans, 26975f39d397SChris Mason struct btrfs_root *root, 2698fdd99c72SLiu Bo struct btrfs_path *path, int level) 269974123bd7SChris Mason { 27000b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 27017bb86316SChris Mason u64 lower_gen; 27025f39d397SChris Mason struct extent_buffer *lower; 27035f39d397SChris Mason struct extent_buffer *c; 2704925baeddSChris Mason struct extent_buffer *old; 27055f39d397SChris Mason struct btrfs_disk_key lower_key; 2706d9d19a01SDavid Sterba int ret; 27075c680ed6SChris Mason 27085c680ed6SChris Mason BUG_ON(path->nodes[level]); 27095c680ed6SChris Mason BUG_ON(path->nodes[level-1] != root->node); 27105c680ed6SChris Mason 27117bb86316SChris Mason lower = path->nodes[level-1]; 27127bb86316SChris Mason if (level == 1) 27137bb86316SChris Mason btrfs_item_key(lower, &lower_key, 0); 27147bb86316SChris Mason else 27157bb86316SChris Mason btrfs_node_key(lower, &lower_key, 0); 27167bb86316SChris Mason 271779bd3712SFilipe Manana c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 271879bd3712SFilipe Manana &lower_key, level, root->node->start, 0, 2719cf6f34aaSJosef Bacik BTRFS_NESTING_NEW_ROOT); 27205f39d397SChris Mason if (IS_ERR(c)) 27215f39d397SChris Mason return PTR_ERR(c); 2722925baeddSChris Mason 27230b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 2724f0486c68SYan, Zheng 27255f39d397SChris Mason btrfs_set_header_nritems(c, 1); 27265f39d397SChris Mason btrfs_set_node_key(c, &lower_key, 0); 2727db94535dSChris Mason btrfs_set_node_blockptr(c, 0, lower->start); 27287bb86316SChris Mason lower_gen = btrfs_header_generation(lower); 272931840ae1SZheng Yan WARN_ON(lower_gen != trans->transid); 27307bb86316SChris Mason 27317bb86316SChris Mason btrfs_set_node_ptr_generation(c, 0, lower_gen); 27325f39d397SChris Mason 27335f39d397SChris Mason btrfs_mark_buffer_dirty(c); 2734d5719762SChris Mason 2735925baeddSChris Mason old = root->node; 2736406808abSFilipe Manana ret = btrfs_tree_mod_log_insert_root(root->node, c, false); 2737d9d19a01SDavid Sterba BUG_ON(ret < 0); 2738240f62c8SChris Mason rcu_assign_pointer(root->node, c); 2739925baeddSChris Mason 2740925baeddSChris Mason /* the super has an extra ref to root->node */ 2741925baeddSChris Mason free_extent_buffer(old); 2742925baeddSChris Mason 27430b86a832SChris Mason add_root_to_dirty_list(root); 274467439dadSDavid Sterba atomic_inc(&c->refs); 27455f39d397SChris Mason path->nodes[level] = c; 2746ac5887c8SJosef Bacik path->locks[level] = BTRFS_WRITE_LOCK; 274774123bd7SChris Mason path->slots[level] = 0; 274874123bd7SChris Mason return 0; 274974123bd7SChris Mason } 27505c680ed6SChris Mason 27515c680ed6SChris Mason /* 27525c680ed6SChris Mason * worker function to insert a single pointer in a node. 27535c680ed6SChris Mason * the node should have enough room for the pointer already 275497571fd0SChris Mason * 27555c680ed6SChris Mason * slot and level indicate where you want the key to go, and 27565c680ed6SChris Mason * blocknr is the block the key points to. 27575c680ed6SChris Mason */ 2758143bede5SJeff Mahoney static void insert_ptr(struct btrfs_trans_handle *trans, 27596ad3cf6dSDavid Sterba struct btrfs_path *path, 2760143bede5SJeff Mahoney struct btrfs_disk_key *key, u64 bytenr, 2761c3e06965SJan Schmidt int slot, int level) 27625c680ed6SChris Mason { 27635f39d397SChris Mason struct extent_buffer *lower; 27645c680ed6SChris Mason int nritems; 2765f3ea38daSJan Schmidt int ret; 27665c680ed6SChris Mason 27675c680ed6SChris Mason BUG_ON(!path->nodes[level]); 276849d0c642SFilipe Manana btrfs_assert_tree_write_locked(path->nodes[level]); 27695f39d397SChris Mason lower = path->nodes[level]; 27705f39d397SChris Mason nritems = btrfs_header_nritems(lower); 2771c293498bSStoyan Gaydarov BUG_ON(slot > nritems); 27726ad3cf6dSDavid Sterba BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); 277374123bd7SChris Mason if (slot != nritems) { 2774bf1d3425SDavid Sterba if (level) { 2775f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_move(lower, slot + 1, 2776f3a84ccdSFilipe Manana slot, nritems - slot); 2777bf1d3425SDavid Sterba BUG_ON(ret < 0); 2778bf1d3425SDavid Sterba } 27795f39d397SChris Mason memmove_extent_buffer(lower, 27805f39d397SChris Mason btrfs_node_key_ptr_offset(slot + 1), 27815f39d397SChris Mason btrfs_node_key_ptr_offset(slot), 2782123abc88SChris Mason (nritems - slot) * sizeof(struct btrfs_key_ptr)); 278374123bd7SChris Mason } 2784c3e06965SJan Schmidt if (level) { 2785f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(lower, slot, 278633cff222SFilipe Manana BTRFS_MOD_LOG_KEY_ADD); 2787f3ea38daSJan Schmidt BUG_ON(ret < 0); 2788f3ea38daSJan Schmidt } 27895f39d397SChris Mason btrfs_set_node_key(lower, key, slot); 2790db94535dSChris Mason btrfs_set_node_blockptr(lower, slot, bytenr); 279174493f7aSChris Mason WARN_ON(trans->transid == 0); 279274493f7aSChris Mason btrfs_set_node_ptr_generation(lower, slot, trans->transid); 27935f39d397SChris Mason btrfs_set_header_nritems(lower, nritems + 1); 27945f39d397SChris Mason btrfs_mark_buffer_dirty(lower); 279574123bd7SChris Mason } 279674123bd7SChris Mason 279797571fd0SChris Mason /* 279897571fd0SChris Mason * split the node at the specified level in path in two. 279997571fd0SChris Mason * The path is corrected to point to the appropriate node after the split 280097571fd0SChris Mason * 280197571fd0SChris Mason * Before splitting this tries to make some room in the node by pushing 280297571fd0SChris Mason * left and right, if either one works, it returns right away. 2803aa5d6bedSChris Mason * 2804aa5d6bedSChris Mason * returns 0 on success and < 0 on failure 280597571fd0SChris Mason */ 2806e02119d5SChris Mason static noinline int split_node(struct btrfs_trans_handle *trans, 2807e02119d5SChris Mason struct btrfs_root *root, 2808e02119d5SChris Mason struct btrfs_path *path, int level) 2809be0e5c09SChris Mason { 28100b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 28115f39d397SChris Mason struct extent_buffer *c; 28125f39d397SChris Mason struct extent_buffer *split; 28135f39d397SChris Mason struct btrfs_disk_key disk_key; 2814be0e5c09SChris Mason int mid; 28155c680ed6SChris Mason int ret; 28167518a238SChris Mason u32 c_nritems; 2817be0e5c09SChris Mason 28185f39d397SChris Mason c = path->nodes[level]; 28197bb86316SChris Mason WARN_ON(btrfs_header_generation(c) != trans->transid); 28205f39d397SChris Mason if (c == root->node) { 2821d9abbf1cSJan Schmidt /* 282290f8d62eSJan Schmidt * trying to split the root, lets make a new one 282390f8d62eSJan Schmidt * 2824fdd99c72SLiu Bo * tree mod log: We don't log_removal old root in 282590f8d62eSJan Schmidt * insert_new_root, because that root buffer will be kept as a 282690f8d62eSJan Schmidt * normal node. We are going to log removal of half of the 2827f3a84ccdSFilipe Manana * elements below with btrfs_tree_mod_log_eb_copy(). We're 2828f3a84ccdSFilipe Manana * holding a tree lock on the buffer, which is why we cannot 2829f3a84ccdSFilipe Manana * race with other tree_mod_log users. 2830d9abbf1cSJan Schmidt */ 2831fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, level + 1); 28325c680ed6SChris Mason if (ret) 28335c680ed6SChris Mason return ret; 2834b3612421SChris Mason } else { 2835e66f709bSChris Mason ret = push_nodes_for_insert(trans, root, path, level); 28365f39d397SChris Mason c = path->nodes[level]; 28375f39d397SChris Mason if (!ret && btrfs_header_nritems(c) < 28380b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) 2839e66f709bSChris Mason return 0; 284054aa1f4dSChris Mason if (ret < 0) 284154aa1f4dSChris Mason return ret; 28425c680ed6SChris Mason } 2843e66f709bSChris Mason 28445f39d397SChris Mason c_nritems = btrfs_header_nritems(c); 28455d4f98a2SYan Zheng mid = (c_nritems + 1) / 2; 28465d4f98a2SYan Zheng btrfs_node_key(c, &disk_key, mid); 28477bb86316SChris Mason 284879bd3712SFilipe Manana split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 284979bd3712SFilipe Manana &disk_key, level, c->start, 0, 285079bd3712SFilipe Manana BTRFS_NESTING_SPLIT); 28515f39d397SChris Mason if (IS_ERR(split)) 28525f39d397SChris Mason return PTR_ERR(split); 285354aa1f4dSChris Mason 28540b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 2855bc877d28SNikolay Borisov ASSERT(btrfs_header_level(c) == level); 28565f39d397SChris Mason 2857f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); 28585de865eeSFilipe David Borba Manana if (ret) { 285966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 28605de865eeSFilipe David Borba Manana return ret; 28615de865eeSFilipe David Borba Manana } 28625f39d397SChris Mason copy_extent_buffer(split, c, 28635f39d397SChris Mason btrfs_node_key_ptr_offset(0), 28645f39d397SChris Mason btrfs_node_key_ptr_offset(mid), 2865123abc88SChris Mason (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); 28665f39d397SChris Mason btrfs_set_header_nritems(split, c_nritems - mid); 28675f39d397SChris Mason btrfs_set_header_nritems(c, mid); 2868aa5d6bedSChris Mason 28695f39d397SChris Mason btrfs_mark_buffer_dirty(c); 28705f39d397SChris Mason btrfs_mark_buffer_dirty(split); 28715f39d397SChris Mason 28726ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, split->start, 2873c3e06965SJan Schmidt path->slots[level + 1] + 1, level + 1); 2874aa5d6bedSChris Mason 28755de08d7dSChris Mason if (path->slots[level] >= mid) { 28765c680ed6SChris Mason path->slots[level] -= mid; 2877925baeddSChris Mason btrfs_tree_unlock(c); 28785f39d397SChris Mason free_extent_buffer(c); 28795f39d397SChris Mason path->nodes[level] = split; 28805c680ed6SChris Mason path->slots[level + 1] += 1; 2881eb60ceacSChris Mason } else { 2882925baeddSChris Mason btrfs_tree_unlock(split); 28835f39d397SChris Mason free_extent_buffer(split); 2884be0e5c09SChris Mason } 2885d5286a92SNikolay Borisov return 0; 2886be0e5c09SChris Mason } 2887be0e5c09SChris Mason 288874123bd7SChris Mason /* 288974123bd7SChris Mason * how many bytes are required to store the items in a leaf. start 289074123bd7SChris Mason * and nr indicate which items in the leaf to check. This totals up the 289174123bd7SChris Mason * space used both by the item structs and the item data 289274123bd7SChris Mason */ 28935f39d397SChris Mason static int leaf_space_used(struct extent_buffer *l, int start, int nr) 2894be0e5c09SChris Mason { 2895be0e5c09SChris Mason int data_len; 28965f39d397SChris Mason int nritems = btrfs_header_nritems(l); 2897d4dbff95SChris Mason int end = min(nritems, start + nr) - 1; 2898be0e5c09SChris Mason 2899be0e5c09SChris Mason if (!nr) 2900be0e5c09SChris Mason return 0; 29013212fa14SJosef Bacik data_len = btrfs_item_offset(l, start) + btrfs_item_size(l, start); 29023212fa14SJosef Bacik data_len = data_len - btrfs_item_offset(l, end); 29030783fcfcSChris Mason data_len += sizeof(struct btrfs_item) * nr; 2904d4dbff95SChris Mason WARN_ON(data_len < 0); 2905be0e5c09SChris Mason return data_len; 2906be0e5c09SChris Mason } 2907be0e5c09SChris Mason 290874123bd7SChris Mason /* 2909d4dbff95SChris Mason * The space between the end of the leaf items and 2910d4dbff95SChris Mason * the start of the leaf data. IOW, how much room 2911d4dbff95SChris Mason * the leaf has left for both items and data 2912d4dbff95SChris Mason */ 2913e902baacSDavid Sterba noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) 2914d4dbff95SChris Mason { 2915e902baacSDavid Sterba struct btrfs_fs_info *fs_info = leaf->fs_info; 29165f39d397SChris Mason int nritems = btrfs_header_nritems(leaf); 29175f39d397SChris Mason int ret; 29180b246afaSJeff Mahoney 29190b246afaSJeff Mahoney ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); 29205f39d397SChris Mason if (ret < 0) { 29210b246afaSJeff Mahoney btrfs_crit(fs_info, 2922efe120a0SFrank Holton "leaf free space ret %d, leaf data size %lu, used %d nritems %d", 2923da17066cSJeff Mahoney ret, 29240b246afaSJeff Mahoney (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), 29255f39d397SChris Mason leaf_space_used(leaf, 0, nritems), nritems); 29265f39d397SChris Mason } 29275f39d397SChris Mason return ret; 2928d4dbff95SChris Mason } 2929d4dbff95SChris Mason 293099d8f83cSChris Mason /* 293199d8f83cSChris Mason * min slot controls the lowest index we're willing to push to the 293299d8f83cSChris Mason * right. We'll push up to and including min_slot, but no lower 293399d8f83cSChris Mason */ 2934f72f0010SDavid Sterba static noinline int __push_leaf_right(struct btrfs_path *path, 293544871b1bSChris Mason int data_size, int empty, 293644871b1bSChris Mason struct extent_buffer *right, 293799d8f83cSChris Mason int free_space, u32 left_nritems, 293899d8f83cSChris Mason u32 min_slot) 293900ec4c51SChris Mason { 2940f72f0010SDavid Sterba struct btrfs_fs_info *fs_info = right->fs_info; 29415f39d397SChris Mason struct extent_buffer *left = path->nodes[0]; 294244871b1bSChris Mason struct extent_buffer *upper = path->nodes[1]; 2943cfed81a0SChris Mason struct btrfs_map_token token; 29445f39d397SChris Mason struct btrfs_disk_key disk_key; 294500ec4c51SChris Mason int slot; 294634a38218SChris Mason u32 i; 294700ec4c51SChris Mason int push_space = 0; 294800ec4c51SChris Mason int push_items = 0; 294934a38218SChris Mason u32 nr; 29507518a238SChris Mason u32 right_nritems; 29515f39d397SChris Mason u32 data_end; 2952db94535dSChris Mason u32 this_item_size; 295300ec4c51SChris Mason 295434a38218SChris Mason if (empty) 295534a38218SChris Mason nr = 0; 295634a38218SChris Mason else 295799d8f83cSChris Mason nr = max_t(u32, 1, min_slot); 295834a38218SChris Mason 295931840ae1SZheng Yan if (path->slots[0] >= left_nritems) 296087b29b20SYan Zheng push_space += data_size; 296131840ae1SZheng Yan 296244871b1bSChris Mason slot = path->slots[1]; 296334a38218SChris Mason i = left_nritems - 1; 296434a38218SChris Mason while (i >= nr) { 296531840ae1SZheng Yan if (!empty && push_items > 0) { 296631840ae1SZheng Yan if (path->slots[0] > i) 296731840ae1SZheng Yan break; 296831840ae1SZheng Yan if (path->slots[0] == i) { 2969e902baacSDavid Sterba int space = btrfs_leaf_free_space(left); 2970e902baacSDavid Sterba 297131840ae1SZheng Yan if (space + push_space * 2 > free_space) 297231840ae1SZheng Yan break; 297331840ae1SZheng Yan } 297431840ae1SZheng Yan } 297531840ae1SZheng Yan 297600ec4c51SChris Mason if (path->slots[0] == i) 297787b29b20SYan Zheng push_space += data_size; 2978db94535dSChris Mason 29793212fa14SJosef Bacik this_item_size = btrfs_item_size(left, i); 298074794207SJosef Bacik if (this_item_size + sizeof(struct btrfs_item) + 298174794207SJosef Bacik push_space > free_space) 298200ec4c51SChris Mason break; 298331840ae1SZheng Yan 298400ec4c51SChris Mason push_items++; 298574794207SJosef Bacik push_space += this_item_size + sizeof(struct btrfs_item); 298634a38218SChris Mason if (i == 0) 298734a38218SChris Mason break; 298834a38218SChris Mason i--; 2989db94535dSChris Mason } 29905f39d397SChris Mason 2991925baeddSChris Mason if (push_items == 0) 2992925baeddSChris Mason goto out_unlock; 29935f39d397SChris Mason 29946c1500f2SJulia Lawall WARN_ON(!empty && push_items == left_nritems); 29955f39d397SChris Mason 299600ec4c51SChris Mason /* push left to right */ 29975f39d397SChris Mason right_nritems = btrfs_header_nritems(right); 299834a38218SChris Mason 2999dc2e724eSJosef Bacik push_space = btrfs_item_data_end(left, left_nritems - push_items); 30008f881e8cSDavid Sterba push_space -= leaf_data_end(left); 30015f39d397SChris Mason 300200ec4c51SChris Mason /* make room in the right data area */ 30038f881e8cSDavid Sterba data_end = leaf_data_end(right); 30045f39d397SChris Mason memmove_extent_buffer(right, 30053d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end - push_space, 30063d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end, 30070b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); 30085f39d397SChris Mason 300900ec4c51SChris Mason /* copy from the left data area */ 30103d9ec8c4SNikolay Borisov copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + 30110b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 30128f881e8cSDavid Sterba BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), 3013d6025579SChris Mason push_space); 30145f39d397SChris Mason 30155f39d397SChris Mason memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), 30165f39d397SChris Mason btrfs_item_nr_offset(0), 30170783fcfcSChris Mason right_nritems * sizeof(struct btrfs_item)); 30185f39d397SChris Mason 301900ec4c51SChris Mason /* copy the items from left to right */ 30205f39d397SChris Mason copy_extent_buffer(right, left, btrfs_item_nr_offset(0), 30215f39d397SChris Mason btrfs_item_nr_offset(left_nritems - push_items), 30220783fcfcSChris Mason push_items * sizeof(struct btrfs_item)); 302300ec4c51SChris Mason 302400ec4c51SChris Mason /* update the item pointers */ 3025c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 30267518a238SChris Mason right_nritems += push_items; 30275f39d397SChris Mason btrfs_set_header_nritems(right, right_nritems); 30280b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 30297518a238SChris Mason for (i = 0; i < right_nritems; i++) { 30303212fa14SJosef Bacik push_space -= btrfs_token_item_size(&token, i); 30313212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, push_space); 3032db94535dSChris Mason } 3033db94535dSChris Mason 30347518a238SChris Mason left_nritems -= push_items; 30355f39d397SChris Mason btrfs_set_header_nritems(left, left_nritems); 303600ec4c51SChris Mason 303734a38218SChris Mason if (left_nritems) 30385f39d397SChris Mason btrfs_mark_buffer_dirty(left); 3039f0486c68SYan, Zheng else 30406a884d7dSDavid Sterba btrfs_clean_tree_block(left); 3041f0486c68SYan, Zheng 30425f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3043a429e513SChris Mason 30445f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 30455f39d397SChris Mason btrfs_set_node_key(upper, &disk_key, slot + 1); 3046d6025579SChris Mason btrfs_mark_buffer_dirty(upper); 304702217ed2SChris Mason 304800ec4c51SChris Mason /* then fixup the leaf pointer in the path */ 30497518a238SChris Mason if (path->slots[0] >= left_nritems) { 30507518a238SChris Mason path->slots[0] -= left_nritems; 3051925baeddSChris Mason if (btrfs_header_nritems(path->nodes[0]) == 0) 30526a884d7dSDavid Sterba btrfs_clean_tree_block(path->nodes[0]); 3053925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 30545f39d397SChris Mason free_extent_buffer(path->nodes[0]); 30555f39d397SChris Mason path->nodes[0] = right; 305600ec4c51SChris Mason path->slots[1] += 1; 305700ec4c51SChris Mason } else { 3058925baeddSChris Mason btrfs_tree_unlock(right); 30595f39d397SChris Mason free_extent_buffer(right); 306000ec4c51SChris Mason } 306100ec4c51SChris Mason return 0; 3062925baeddSChris Mason 3063925baeddSChris Mason out_unlock: 3064925baeddSChris Mason btrfs_tree_unlock(right); 3065925baeddSChris Mason free_extent_buffer(right); 3066925baeddSChris Mason return 1; 306700ec4c51SChris Mason } 3068925baeddSChris Mason 306900ec4c51SChris Mason /* 307044871b1bSChris Mason * push some data in the path leaf to the right, trying to free up at 307174123bd7SChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 307244871b1bSChris Mason * 307344871b1bSChris Mason * returns 1 if the push failed because the other node didn't have enough 307444871b1bSChris Mason * room, 0 if everything worked out and < 0 if there were major errors. 307599d8f83cSChris Mason * 307699d8f83cSChris Mason * this will push starting from min_slot to the end of the leaf. It won't 307799d8f83cSChris Mason * push any slot lower than min_slot 307874123bd7SChris Mason */ 307944871b1bSChris Mason static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root 308099d8f83cSChris Mason *root, struct btrfs_path *path, 308199d8f83cSChris Mason int min_data_size, int data_size, 308299d8f83cSChris Mason int empty, u32 min_slot) 3083be0e5c09SChris Mason { 308444871b1bSChris Mason struct extent_buffer *left = path->nodes[0]; 308544871b1bSChris Mason struct extent_buffer *right; 308644871b1bSChris Mason struct extent_buffer *upper; 308744871b1bSChris Mason int slot; 308844871b1bSChris Mason int free_space; 308944871b1bSChris Mason u32 left_nritems; 309044871b1bSChris Mason int ret; 309144871b1bSChris Mason 309244871b1bSChris Mason if (!path->nodes[1]) 309344871b1bSChris Mason return 1; 309444871b1bSChris Mason 309544871b1bSChris Mason slot = path->slots[1]; 309644871b1bSChris Mason upper = path->nodes[1]; 309744871b1bSChris Mason if (slot >= btrfs_header_nritems(upper) - 1) 309844871b1bSChris Mason return 1; 309944871b1bSChris Mason 310049d0c642SFilipe Manana btrfs_assert_tree_write_locked(path->nodes[1]); 310144871b1bSChris Mason 31024b231ae4SDavid Sterba right = btrfs_read_node_slot(upper, slot + 1); 3103fb770ae4SLiu Bo /* 3104fb770ae4SLiu Bo * slot + 1 is not valid or we fail to read the right node, 3105fb770ae4SLiu Bo * no big deal, just return. 3106fb770ae4SLiu Bo */ 3107fb770ae4SLiu Bo if (IS_ERR(right)) 310891ca338dSTsutomu Itoh return 1; 310991ca338dSTsutomu Itoh 3110bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 311144871b1bSChris Mason 3112e902baacSDavid Sterba free_space = btrfs_leaf_free_space(right); 311344871b1bSChris Mason if (free_space < data_size) 311444871b1bSChris Mason goto out_unlock; 311544871b1bSChris Mason 311644871b1bSChris Mason ret = btrfs_cow_block(trans, root, right, upper, 3117bf59a5a2SJosef Bacik slot + 1, &right, BTRFS_NESTING_RIGHT_COW); 311844871b1bSChris Mason if (ret) 311944871b1bSChris Mason goto out_unlock; 312044871b1bSChris Mason 312144871b1bSChris Mason left_nritems = btrfs_header_nritems(left); 312244871b1bSChris Mason if (left_nritems == 0) 312344871b1bSChris Mason goto out_unlock; 312444871b1bSChris Mason 3125d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3126d16c702fSQu Wenruo ret = -EUCLEAN; 3127d16c702fSQu Wenruo btrfs_tree_unlock(right); 3128d16c702fSQu Wenruo free_extent_buffer(right); 3129d16c702fSQu Wenruo return ret; 3130d16c702fSQu Wenruo } 31312ef1fed2SFilipe David Borba Manana if (path->slots[0] == left_nritems && !empty) { 31322ef1fed2SFilipe David Borba Manana /* Key greater than all keys in the leaf, right neighbor has 31332ef1fed2SFilipe David Borba Manana * enough room for it and we're not emptying our leaf to delete 31342ef1fed2SFilipe David Borba Manana * it, therefore use right neighbor to insert the new item and 313552042d8eSAndrea Gelmini * no need to touch/dirty our left leaf. */ 31362ef1fed2SFilipe David Borba Manana btrfs_tree_unlock(left); 31372ef1fed2SFilipe David Borba Manana free_extent_buffer(left); 31382ef1fed2SFilipe David Borba Manana path->nodes[0] = right; 31392ef1fed2SFilipe David Borba Manana path->slots[0] = 0; 31402ef1fed2SFilipe David Borba Manana path->slots[1]++; 31412ef1fed2SFilipe David Borba Manana return 0; 31422ef1fed2SFilipe David Borba Manana } 31432ef1fed2SFilipe David Borba Manana 3144f72f0010SDavid Sterba return __push_leaf_right(path, min_data_size, empty, 314599d8f83cSChris Mason right, free_space, left_nritems, min_slot); 314644871b1bSChris Mason out_unlock: 314744871b1bSChris Mason btrfs_tree_unlock(right); 314844871b1bSChris Mason free_extent_buffer(right); 314944871b1bSChris Mason return 1; 315044871b1bSChris Mason } 315144871b1bSChris Mason 315244871b1bSChris Mason /* 315344871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 315444871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 315599d8f83cSChris Mason * 315699d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 315799d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the 315899d8f83cSChris Mason * items 315944871b1bSChris Mason */ 31608087c193SDavid Sterba static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, 316144871b1bSChris Mason int empty, struct extent_buffer *left, 316299d8f83cSChris Mason int free_space, u32 right_nritems, 316399d8f83cSChris Mason u32 max_slot) 316444871b1bSChris Mason { 31658087c193SDavid Sterba struct btrfs_fs_info *fs_info = left->fs_info; 31665f39d397SChris Mason struct btrfs_disk_key disk_key; 31675f39d397SChris Mason struct extent_buffer *right = path->nodes[0]; 3168be0e5c09SChris Mason int i; 3169be0e5c09SChris Mason int push_space = 0; 3170be0e5c09SChris Mason int push_items = 0; 31717518a238SChris Mason u32 old_left_nritems; 317234a38218SChris Mason u32 nr; 3173aa5d6bedSChris Mason int ret = 0; 3174db94535dSChris Mason u32 this_item_size; 3175db94535dSChris Mason u32 old_left_item_size; 3176cfed81a0SChris Mason struct btrfs_map_token token; 3177cfed81a0SChris Mason 317834a38218SChris Mason if (empty) 317999d8f83cSChris Mason nr = min(right_nritems, max_slot); 318034a38218SChris Mason else 318199d8f83cSChris Mason nr = min(right_nritems - 1, max_slot); 318234a38218SChris Mason 318334a38218SChris Mason for (i = 0; i < nr; i++) { 318431840ae1SZheng Yan if (!empty && push_items > 0) { 318531840ae1SZheng Yan if (path->slots[0] < i) 318631840ae1SZheng Yan break; 318731840ae1SZheng Yan if (path->slots[0] == i) { 3188e902baacSDavid Sterba int space = btrfs_leaf_free_space(right); 3189e902baacSDavid Sterba 319031840ae1SZheng Yan if (space + push_space * 2 > free_space) 319131840ae1SZheng Yan break; 319231840ae1SZheng Yan } 319331840ae1SZheng Yan } 319431840ae1SZheng Yan 3195be0e5c09SChris Mason if (path->slots[0] == i) 319687b29b20SYan Zheng push_space += data_size; 3197db94535dSChris Mason 31983212fa14SJosef Bacik this_item_size = btrfs_item_size(right, i); 319974794207SJosef Bacik if (this_item_size + sizeof(struct btrfs_item) + push_space > 320074794207SJosef Bacik free_space) 3201be0e5c09SChris Mason break; 3202db94535dSChris Mason 3203be0e5c09SChris Mason push_items++; 320474794207SJosef Bacik push_space += this_item_size + sizeof(struct btrfs_item); 3205be0e5c09SChris Mason } 3206db94535dSChris Mason 3207be0e5c09SChris Mason if (push_items == 0) { 3208925baeddSChris Mason ret = 1; 3209925baeddSChris Mason goto out; 3210be0e5c09SChris Mason } 3211fae7f21cSDulshani Gunawardhana WARN_ON(!empty && push_items == btrfs_header_nritems(right)); 32125f39d397SChris Mason 3213be0e5c09SChris Mason /* push data from right to left */ 32145f39d397SChris Mason copy_extent_buffer(left, right, 32155f39d397SChris Mason btrfs_item_nr_offset(btrfs_header_nritems(left)), 32165f39d397SChris Mason btrfs_item_nr_offset(0), 32175f39d397SChris Mason push_items * sizeof(struct btrfs_item)); 32185f39d397SChris Mason 32190b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - 32203212fa14SJosef Bacik btrfs_item_offset(right, push_items - 1); 32215f39d397SChris Mason 32223d9ec8c4SNikolay Borisov copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + 32238f881e8cSDavid Sterba leaf_data_end(left) - push_space, 32243d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + 32253212fa14SJosef Bacik btrfs_item_offset(right, push_items - 1), 3226be0e5c09SChris Mason push_space); 32275f39d397SChris Mason old_left_nritems = btrfs_header_nritems(left); 322887b29b20SYan Zheng BUG_ON(old_left_nritems <= 0); 3229eb60ceacSChris Mason 3230c82f823cSDavid Sterba btrfs_init_map_token(&token, left); 32313212fa14SJosef Bacik old_left_item_size = btrfs_item_offset(left, old_left_nritems - 1); 3232be0e5c09SChris Mason for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { 32335f39d397SChris Mason u32 ioff; 3234db94535dSChris Mason 32353212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 32363212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, 3237cc4c13d5SDavid Sterba ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); 3238be0e5c09SChris Mason } 32395f39d397SChris Mason btrfs_set_header_nritems(left, old_left_nritems + push_items); 3240be0e5c09SChris Mason 3241be0e5c09SChris Mason /* fixup right node */ 324231b1a2bdSJulia Lawall if (push_items > right_nritems) 324331b1a2bdSJulia Lawall WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, 3244d397712bSChris Mason right_nritems); 324534a38218SChris Mason 324634a38218SChris Mason if (push_items < right_nritems) { 32473212fa14SJosef Bacik push_space = btrfs_item_offset(right, push_items - 1) - 32488f881e8cSDavid Sterba leaf_data_end(right); 32493d9ec8c4SNikolay Borisov memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + 32500b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 32513d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + 32528f881e8cSDavid Sterba leaf_data_end(right), push_space); 32535f39d397SChris Mason 32545f39d397SChris Mason memmove_extent_buffer(right, btrfs_item_nr_offset(0), 32555f39d397SChris Mason btrfs_item_nr_offset(push_items), 32565f39d397SChris Mason (btrfs_header_nritems(right) - push_items) * 32570783fcfcSChris Mason sizeof(struct btrfs_item)); 325834a38218SChris Mason } 3259c82f823cSDavid Sterba 3260c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 3261eef1c494SYan right_nritems -= push_items; 3262eef1c494SYan btrfs_set_header_nritems(right, right_nritems); 32630b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 32645f39d397SChris Mason for (i = 0; i < right_nritems; i++) { 32653212fa14SJosef Bacik push_space = push_space - btrfs_token_item_size(&token, i); 32663212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, push_space); 3267db94535dSChris Mason } 3268eb60ceacSChris Mason 32695f39d397SChris Mason btrfs_mark_buffer_dirty(left); 327034a38218SChris Mason if (right_nritems) 32715f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3272f0486c68SYan, Zheng else 32736a884d7dSDavid Sterba btrfs_clean_tree_block(right); 3274098f59c2SChris Mason 32755f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 3276b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 3277be0e5c09SChris Mason 3278be0e5c09SChris Mason /* then fixup the leaf pointer in the path */ 3279be0e5c09SChris Mason if (path->slots[0] < push_items) { 3280be0e5c09SChris Mason path->slots[0] += old_left_nritems; 3281925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 32825f39d397SChris Mason free_extent_buffer(path->nodes[0]); 32835f39d397SChris Mason path->nodes[0] = left; 3284be0e5c09SChris Mason path->slots[1] -= 1; 3285be0e5c09SChris Mason } else { 3286925baeddSChris Mason btrfs_tree_unlock(left); 32875f39d397SChris Mason free_extent_buffer(left); 3288be0e5c09SChris Mason path->slots[0] -= push_items; 3289be0e5c09SChris Mason } 3290eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 3291aa5d6bedSChris Mason return ret; 3292925baeddSChris Mason out: 3293925baeddSChris Mason btrfs_tree_unlock(left); 3294925baeddSChris Mason free_extent_buffer(left); 3295925baeddSChris Mason return ret; 3296be0e5c09SChris Mason } 3297be0e5c09SChris Mason 329874123bd7SChris Mason /* 329944871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 330044871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 330199d8f83cSChris Mason * 330299d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 330399d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the 330499d8f83cSChris Mason * items 330544871b1bSChris Mason */ 330644871b1bSChris Mason static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root 330799d8f83cSChris Mason *root, struct btrfs_path *path, int min_data_size, 330899d8f83cSChris Mason int data_size, int empty, u32 max_slot) 330944871b1bSChris Mason { 331044871b1bSChris Mason struct extent_buffer *right = path->nodes[0]; 331144871b1bSChris Mason struct extent_buffer *left; 331244871b1bSChris Mason int slot; 331344871b1bSChris Mason int free_space; 331444871b1bSChris Mason u32 right_nritems; 331544871b1bSChris Mason int ret = 0; 331644871b1bSChris Mason 331744871b1bSChris Mason slot = path->slots[1]; 331844871b1bSChris Mason if (slot == 0) 331944871b1bSChris Mason return 1; 332044871b1bSChris Mason if (!path->nodes[1]) 332144871b1bSChris Mason return 1; 332244871b1bSChris Mason 332344871b1bSChris Mason right_nritems = btrfs_header_nritems(right); 332444871b1bSChris Mason if (right_nritems == 0) 332544871b1bSChris Mason return 1; 332644871b1bSChris Mason 332749d0c642SFilipe Manana btrfs_assert_tree_write_locked(path->nodes[1]); 332844871b1bSChris Mason 33294b231ae4SDavid Sterba left = btrfs_read_node_slot(path->nodes[1], slot - 1); 3330fb770ae4SLiu Bo /* 3331fb770ae4SLiu Bo * slot - 1 is not valid or we fail to read the left node, 3332fb770ae4SLiu Bo * no big deal, just return. 3333fb770ae4SLiu Bo */ 3334fb770ae4SLiu Bo if (IS_ERR(left)) 333591ca338dSTsutomu Itoh return 1; 333691ca338dSTsutomu Itoh 3337bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 333844871b1bSChris Mason 3339e902baacSDavid Sterba free_space = btrfs_leaf_free_space(left); 334044871b1bSChris Mason if (free_space < data_size) { 334144871b1bSChris Mason ret = 1; 334244871b1bSChris Mason goto out; 334344871b1bSChris Mason } 334444871b1bSChris Mason 334544871b1bSChris Mason ret = btrfs_cow_block(trans, root, left, 33469631e4ccSJosef Bacik path->nodes[1], slot - 1, &left, 3347bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 334844871b1bSChris Mason if (ret) { 334944871b1bSChris Mason /* we hit -ENOSPC, but it isn't fatal here */ 335079787eaaSJeff Mahoney if (ret == -ENOSPC) 335144871b1bSChris Mason ret = 1; 335244871b1bSChris Mason goto out; 335344871b1bSChris Mason } 335444871b1bSChris Mason 3355d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3356d16c702fSQu Wenruo ret = -EUCLEAN; 3357d16c702fSQu Wenruo goto out; 3358d16c702fSQu Wenruo } 33598087c193SDavid Sterba return __push_leaf_left(path, min_data_size, 336099d8f83cSChris Mason empty, left, free_space, right_nritems, 336199d8f83cSChris Mason max_slot); 336244871b1bSChris Mason out: 336344871b1bSChris Mason btrfs_tree_unlock(left); 336444871b1bSChris Mason free_extent_buffer(left); 336544871b1bSChris Mason return ret; 336644871b1bSChris Mason } 336744871b1bSChris Mason 336844871b1bSChris Mason /* 336974123bd7SChris Mason * split the path's leaf in two, making sure there is at least data_size 337074123bd7SChris Mason * available for the resulting leaf level of the path. 337174123bd7SChris Mason */ 3372143bede5SJeff Mahoney static noinline void copy_for_split(struct btrfs_trans_handle *trans, 337344871b1bSChris Mason struct btrfs_path *path, 337444871b1bSChris Mason struct extent_buffer *l, 337544871b1bSChris Mason struct extent_buffer *right, 337644871b1bSChris Mason int slot, int mid, int nritems) 3377be0e5c09SChris Mason { 337894f94ad9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 3379be0e5c09SChris Mason int data_copy_size; 3380be0e5c09SChris Mason int rt_data_off; 3381be0e5c09SChris Mason int i; 3382d4dbff95SChris Mason struct btrfs_disk_key disk_key; 3383cfed81a0SChris Mason struct btrfs_map_token token; 3384cfed81a0SChris Mason 33855f39d397SChris Mason nritems = nritems - mid; 33865f39d397SChris Mason btrfs_set_header_nritems(right, nritems); 3387dc2e724eSJosef Bacik data_copy_size = btrfs_item_data_end(l, mid) - leaf_data_end(l); 33885f39d397SChris Mason 33895f39d397SChris Mason copy_extent_buffer(right, l, btrfs_item_nr_offset(0), 33905f39d397SChris Mason btrfs_item_nr_offset(mid), 33915f39d397SChris Mason nritems * sizeof(struct btrfs_item)); 33925f39d397SChris Mason 33935f39d397SChris Mason copy_extent_buffer(right, l, 33943d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - 33953d9ec8c4SNikolay Borisov data_copy_size, BTRFS_LEAF_DATA_OFFSET + 33968f881e8cSDavid Sterba leaf_data_end(l), data_copy_size); 339774123bd7SChris Mason 3398dc2e724eSJosef Bacik rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_data_end(l, mid); 33995f39d397SChris Mason 3400c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 34015f39d397SChris Mason for (i = 0; i < nritems; i++) { 3402db94535dSChris Mason u32 ioff; 3403db94535dSChris Mason 34043212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 34053212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff + rt_data_off); 34060783fcfcSChris Mason } 340774123bd7SChris Mason 34085f39d397SChris Mason btrfs_set_header_nritems(l, mid); 34095f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 34106ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); 34115f39d397SChris Mason 34125f39d397SChris Mason btrfs_mark_buffer_dirty(right); 34135f39d397SChris Mason btrfs_mark_buffer_dirty(l); 3414eb60ceacSChris Mason BUG_ON(path->slots[0] != slot); 34155f39d397SChris Mason 3416be0e5c09SChris Mason if (mid <= slot) { 3417925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 34185f39d397SChris Mason free_extent_buffer(path->nodes[0]); 34195f39d397SChris Mason path->nodes[0] = right; 3420be0e5c09SChris Mason path->slots[0] -= mid; 3421be0e5c09SChris Mason path->slots[1] += 1; 3422925baeddSChris Mason } else { 3423925baeddSChris Mason btrfs_tree_unlock(right); 34245f39d397SChris Mason free_extent_buffer(right); 3425925baeddSChris Mason } 34265f39d397SChris Mason 3427eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 342844871b1bSChris Mason } 342944871b1bSChris Mason 343044871b1bSChris Mason /* 343199d8f83cSChris Mason * double splits happen when we need to insert a big item in the middle 343299d8f83cSChris Mason * of a leaf. A double split can leave us with 3 mostly empty leaves: 343399d8f83cSChris Mason * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] 343499d8f83cSChris Mason * A B C 343599d8f83cSChris Mason * 343699d8f83cSChris Mason * We avoid this by trying to push the items on either side of our target 343799d8f83cSChris Mason * into the adjacent leaves. If all goes well we can avoid the double split 343899d8f83cSChris Mason * completely. 343999d8f83cSChris Mason */ 344099d8f83cSChris Mason static noinline int push_for_double_split(struct btrfs_trans_handle *trans, 344199d8f83cSChris Mason struct btrfs_root *root, 344299d8f83cSChris Mason struct btrfs_path *path, 344399d8f83cSChris Mason int data_size) 344499d8f83cSChris Mason { 344599d8f83cSChris Mason int ret; 344699d8f83cSChris Mason int progress = 0; 344799d8f83cSChris Mason int slot; 344899d8f83cSChris Mason u32 nritems; 34495a4267caSFilipe David Borba Manana int space_needed = data_size; 345099d8f83cSChris Mason 345199d8f83cSChris Mason slot = path->slots[0]; 34525a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(path->nodes[0])) 3453e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 345499d8f83cSChris Mason 345599d8f83cSChris Mason /* 345699d8f83cSChris Mason * try to push all the items after our slot into the 345799d8f83cSChris Mason * right leaf 345899d8f83cSChris Mason */ 34595a4267caSFilipe David Borba Manana ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); 346099d8f83cSChris Mason if (ret < 0) 346199d8f83cSChris Mason return ret; 346299d8f83cSChris Mason 346399d8f83cSChris Mason if (ret == 0) 346499d8f83cSChris Mason progress++; 346599d8f83cSChris Mason 346699d8f83cSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 346799d8f83cSChris Mason /* 346899d8f83cSChris Mason * our goal is to get our slot at the start or end of a leaf. If 346999d8f83cSChris Mason * we've done so we're done 347099d8f83cSChris Mason */ 347199d8f83cSChris Mason if (path->slots[0] == 0 || path->slots[0] == nritems) 347299d8f83cSChris Mason return 0; 347399d8f83cSChris Mason 3474e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 347599d8f83cSChris Mason return 0; 347699d8f83cSChris Mason 347799d8f83cSChris Mason /* try to push all the items before our slot into the next leaf */ 347899d8f83cSChris Mason slot = path->slots[0]; 3479263d3995SFilipe Manana space_needed = data_size; 3480263d3995SFilipe Manana if (slot > 0) 3481e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 34825a4267caSFilipe David Borba Manana ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); 348399d8f83cSChris Mason if (ret < 0) 348499d8f83cSChris Mason return ret; 348599d8f83cSChris Mason 348699d8f83cSChris Mason if (ret == 0) 348799d8f83cSChris Mason progress++; 348899d8f83cSChris Mason 348999d8f83cSChris Mason if (progress) 349099d8f83cSChris Mason return 0; 349199d8f83cSChris Mason return 1; 349299d8f83cSChris Mason } 349399d8f83cSChris Mason 349499d8f83cSChris Mason /* 349544871b1bSChris Mason * split the path's leaf in two, making sure there is at least data_size 349644871b1bSChris Mason * available for the resulting leaf level of the path. 349744871b1bSChris Mason * 349844871b1bSChris Mason * returns 0 if all went well and < 0 on failure. 349944871b1bSChris Mason */ 350044871b1bSChris Mason static noinline int split_leaf(struct btrfs_trans_handle *trans, 350144871b1bSChris Mason struct btrfs_root *root, 3502310712b2SOmar Sandoval const struct btrfs_key *ins_key, 350344871b1bSChris Mason struct btrfs_path *path, int data_size, 350444871b1bSChris Mason int extend) 350544871b1bSChris Mason { 35065d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 350744871b1bSChris Mason struct extent_buffer *l; 350844871b1bSChris Mason u32 nritems; 350944871b1bSChris Mason int mid; 351044871b1bSChris Mason int slot; 351144871b1bSChris Mason struct extent_buffer *right; 3512b7a0365eSDaniel Dressler struct btrfs_fs_info *fs_info = root->fs_info; 351344871b1bSChris Mason int ret = 0; 351444871b1bSChris Mason int wret; 35155d4f98a2SYan Zheng int split; 351644871b1bSChris Mason int num_doubles = 0; 351799d8f83cSChris Mason int tried_avoid_double = 0; 351844871b1bSChris Mason 3519a5719521SYan, Zheng l = path->nodes[0]; 3520a5719521SYan, Zheng slot = path->slots[0]; 35213212fa14SJosef Bacik if (extend && data_size + btrfs_item_size(l, slot) + 35220b246afaSJeff Mahoney sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) 3523a5719521SYan, Zheng return -EOVERFLOW; 3524a5719521SYan, Zheng 352544871b1bSChris Mason /* first try to make some room by pushing left and right */ 352633157e05SLiu Bo if (data_size && path->nodes[1]) { 35275a4267caSFilipe David Borba Manana int space_needed = data_size; 35285a4267caSFilipe David Borba Manana 35295a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(l)) 3530e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 35315a4267caSFilipe David Borba Manana 35325a4267caSFilipe David Borba Manana wret = push_leaf_right(trans, root, path, space_needed, 35335a4267caSFilipe David Borba Manana space_needed, 0, 0); 353444871b1bSChris Mason if (wret < 0) 353544871b1bSChris Mason return wret; 353644871b1bSChris Mason if (wret) { 3537263d3995SFilipe Manana space_needed = data_size; 3538263d3995SFilipe Manana if (slot > 0) 3539e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 35405a4267caSFilipe David Borba Manana wret = push_leaf_left(trans, root, path, space_needed, 35415a4267caSFilipe David Borba Manana space_needed, 0, (u32)-1); 354244871b1bSChris Mason if (wret < 0) 354344871b1bSChris Mason return wret; 354444871b1bSChris Mason } 354544871b1bSChris Mason l = path->nodes[0]; 354644871b1bSChris Mason 354744871b1bSChris Mason /* did the pushes work? */ 3548e902baacSDavid Sterba if (btrfs_leaf_free_space(l) >= data_size) 354944871b1bSChris Mason return 0; 355044871b1bSChris Mason } 355144871b1bSChris Mason 355244871b1bSChris Mason if (!path->nodes[1]) { 3553fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, 1); 355444871b1bSChris Mason if (ret) 355544871b1bSChris Mason return ret; 355644871b1bSChris Mason } 355744871b1bSChris Mason again: 35585d4f98a2SYan Zheng split = 1; 355944871b1bSChris Mason l = path->nodes[0]; 356044871b1bSChris Mason slot = path->slots[0]; 356144871b1bSChris Mason nritems = btrfs_header_nritems(l); 356244871b1bSChris Mason mid = (nritems + 1) / 2; 356344871b1bSChris Mason 35645d4f98a2SYan Zheng if (mid <= slot) { 35655d4f98a2SYan Zheng if (nritems == 1 || 35665d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + data_size > 35670b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 35685d4f98a2SYan Zheng if (slot >= nritems) { 35695d4f98a2SYan Zheng split = 0; 35705d4f98a2SYan Zheng } else { 35715d4f98a2SYan Zheng mid = slot; 35725d4f98a2SYan Zheng if (mid != nritems && 35735d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 35740b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 357599d8f83cSChris Mason if (data_size && !tried_avoid_double) 357699d8f83cSChris Mason goto push_for_double; 35775d4f98a2SYan Zheng split = 2; 35785d4f98a2SYan Zheng } 35795d4f98a2SYan Zheng } 35805d4f98a2SYan Zheng } 35815d4f98a2SYan Zheng } else { 35825d4f98a2SYan Zheng if (leaf_space_used(l, 0, mid) + data_size > 35830b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 35845d4f98a2SYan Zheng if (!extend && data_size && slot == 0) { 35855d4f98a2SYan Zheng split = 0; 35865d4f98a2SYan Zheng } else if ((extend || !data_size) && slot == 0) { 35875d4f98a2SYan Zheng mid = 1; 35885d4f98a2SYan Zheng } else { 35895d4f98a2SYan Zheng mid = slot; 35905d4f98a2SYan Zheng if (mid != nritems && 35915d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 35920b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 359399d8f83cSChris Mason if (data_size && !tried_avoid_double) 359499d8f83cSChris Mason goto push_for_double; 35955d4f98a2SYan Zheng split = 2; 35965d4f98a2SYan Zheng } 35975d4f98a2SYan Zheng } 35985d4f98a2SYan Zheng } 35995d4f98a2SYan Zheng } 36005d4f98a2SYan Zheng 36015d4f98a2SYan Zheng if (split == 0) 36025d4f98a2SYan Zheng btrfs_cpu_key_to_disk(&disk_key, ins_key); 36035d4f98a2SYan Zheng else 36045d4f98a2SYan Zheng btrfs_item_key(l, &disk_key, mid); 36055d4f98a2SYan Zheng 3606ca9d473aSJosef Bacik /* 3607ca9d473aSJosef Bacik * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double 3608ca9d473aSJosef Bacik * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES 3609ca9d473aSJosef Bacik * subclasses, which is 8 at the time of this patch, and we've maxed it 3610ca9d473aSJosef Bacik * out. In the future we could add a 3611ca9d473aSJosef Bacik * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just 3612ca9d473aSJosef Bacik * use BTRFS_NESTING_NEW_ROOT. 3613ca9d473aSJosef Bacik */ 361479bd3712SFilipe Manana right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 361579bd3712SFilipe Manana &disk_key, 0, l->start, 0, 361679bd3712SFilipe Manana num_doubles ? BTRFS_NESTING_NEW_ROOT : 3617ca9d473aSJosef Bacik BTRFS_NESTING_SPLIT); 3618f0486c68SYan, Zheng if (IS_ERR(right)) 361944871b1bSChris Mason return PTR_ERR(right); 3620f0486c68SYan, Zheng 36210b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 362244871b1bSChris Mason 36235d4f98a2SYan Zheng if (split == 0) { 362444871b1bSChris Mason if (mid <= slot) { 362544871b1bSChris Mason btrfs_set_header_nritems(right, 0); 36266ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, 36272ff7e61eSJeff Mahoney right->start, path->slots[1] + 1, 1); 362844871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 362944871b1bSChris Mason free_extent_buffer(path->nodes[0]); 363044871b1bSChris Mason path->nodes[0] = right; 363144871b1bSChris Mason path->slots[0] = 0; 363244871b1bSChris Mason path->slots[1] += 1; 363344871b1bSChris Mason } else { 363444871b1bSChris Mason btrfs_set_header_nritems(right, 0); 36356ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, 36362ff7e61eSJeff Mahoney right->start, path->slots[1], 1); 363744871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 363844871b1bSChris Mason free_extent_buffer(path->nodes[0]); 363944871b1bSChris Mason path->nodes[0] = right; 364044871b1bSChris Mason path->slots[0] = 0; 3641143bede5SJeff Mahoney if (path->slots[1] == 0) 3642b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 36435d4f98a2SYan Zheng } 3644196e0249SLiu Bo /* 3645196e0249SLiu Bo * We create a new leaf 'right' for the required ins_len and 3646196e0249SLiu Bo * we'll do btrfs_mark_buffer_dirty() on this leaf after copying 3647196e0249SLiu Bo * the content of ins_len to 'right'. 3648196e0249SLiu Bo */ 364944871b1bSChris Mason return ret; 365044871b1bSChris Mason } 365144871b1bSChris Mason 365294f94ad9SDavid Sterba copy_for_split(trans, path, l, right, slot, mid, nritems); 365344871b1bSChris Mason 36545d4f98a2SYan Zheng if (split == 2) { 3655cc0c5538SChris Mason BUG_ON(num_doubles != 0); 3656cc0c5538SChris Mason num_doubles++; 3657cc0c5538SChris Mason goto again; 36583326d1b0SChris Mason } 365944871b1bSChris Mason 3660143bede5SJeff Mahoney return 0; 366199d8f83cSChris Mason 366299d8f83cSChris Mason push_for_double: 366399d8f83cSChris Mason push_for_double_split(trans, root, path, data_size); 366499d8f83cSChris Mason tried_avoid_double = 1; 3665e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 366699d8f83cSChris Mason return 0; 366799d8f83cSChris Mason goto again; 3668be0e5c09SChris Mason } 3669be0e5c09SChris Mason 3670ad48fd75SYan, Zheng static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, 3671ad48fd75SYan, Zheng struct btrfs_root *root, 3672ad48fd75SYan, Zheng struct btrfs_path *path, int ins_len) 3673ad48fd75SYan, Zheng { 3674ad48fd75SYan, Zheng struct btrfs_key key; 3675ad48fd75SYan, Zheng struct extent_buffer *leaf; 3676ad48fd75SYan, Zheng struct btrfs_file_extent_item *fi; 3677ad48fd75SYan, Zheng u64 extent_len = 0; 3678ad48fd75SYan, Zheng u32 item_size; 3679ad48fd75SYan, Zheng int ret; 3680ad48fd75SYan, Zheng 3681ad48fd75SYan, Zheng leaf = path->nodes[0]; 3682ad48fd75SYan, Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3683ad48fd75SYan, Zheng 3684ad48fd75SYan, Zheng BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && 3685ad48fd75SYan, Zheng key.type != BTRFS_EXTENT_CSUM_KEY); 3686ad48fd75SYan, Zheng 3687e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) >= ins_len) 3688ad48fd75SYan, Zheng return 0; 3689ad48fd75SYan, Zheng 36903212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 3691ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 3692ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 3693ad48fd75SYan, Zheng struct btrfs_file_extent_item); 3694ad48fd75SYan, Zheng extent_len = btrfs_file_extent_num_bytes(leaf, fi); 3695ad48fd75SYan, Zheng } 3696b3b4aa74SDavid Sterba btrfs_release_path(path); 3697ad48fd75SYan, Zheng 3698ad48fd75SYan, Zheng path->keep_locks = 1; 3699ad48fd75SYan, Zheng path->search_for_split = 1; 3700ad48fd75SYan, Zheng ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 3701ad48fd75SYan, Zheng path->search_for_split = 0; 3702a8df6fe6SFilipe Manana if (ret > 0) 3703a8df6fe6SFilipe Manana ret = -EAGAIN; 3704ad48fd75SYan, Zheng if (ret < 0) 3705ad48fd75SYan, Zheng goto err; 3706ad48fd75SYan, Zheng 3707ad48fd75SYan, Zheng ret = -EAGAIN; 3708ad48fd75SYan, Zheng leaf = path->nodes[0]; 3709a8df6fe6SFilipe Manana /* if our item isn't there, return now */ 37103212fa14SJosef Bacik if (item_size != btrfs_item_size(leaf, path->slots[0])) 3711ad48fd75SYan, Zheng goto err; 3712ad48fd75SYan, Zheng 3713109f6aefSChris Mason /* the leaf has changed, it now has room. return now */ 3714e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) 3715109f6aefSChris Mason goto err; 3716109f6aefSChris Mason 3717ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 3718ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 3719ad48fd75SYan, Zheng struct btrfs_file_extent_item); 3720ad48fd75SYan, Zheng if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) 3721ad48fd75SYan, Zheng goto err; 3722ad48fd75SYan, Zheng } 3723ad48fd75SYan, Zheng 3724ad48fd75SYan, Zheng ret = split_leaf(trans, root, &key, path, ins_len, 1); 3725f0486c68SYan, Zheng if (ret) 3726f0486c68SYan, Zheng goto err; 3727ad48fd75SYan, Zheng 3728ad48fd75SYan, Zheng path->keep_locks = 0; 3729ad48fd75SYan, Zheng btrfs_unlock_up_safe(path, 1); 3730ad48fd75SYan, Zheng return 0; 3731ad48fd75SYan, Zheng err: 3732ad48fd75SYan, Zheng path->keep_locks = 0; 3733ad48fd75SYan, Zheng return ret; 3734ad48fd75SYan, Zheng } 3735ad48fd75SYan, Zheng 373625263cd7SDavid Sterba static noinline int split_item(struct btrfs_path *path, 3737310712b2SOmar Sandoval const struct btrfs_key *new_key, 3738459931ecSChris Mason unsigned long split_offset) 3739459931ecSChris Mason { 3740459931ecSChris Mason struct extent_buffer *leaf; 3741c91666b1SJosef Bacik int orig_slot, slot; 3742ad48fd75SYan, Zheng char *buf; 3743459931ecSChris Mason u32 nritems; 3744ad48fd75SYan, Zheng u32 item_size; 3745459931ecSChris Mason u32 orig_offset; 3746459931ecSChris Mason struct btrfs_disk_key disk_key; 3747459931ecSChris Mason 3748459931ecSChris Mason leaf = path->nodes[0]; 3749e902baacSDavid Sterba BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); 3750b9473439SChris Mason 3751c91666b1SJosef Bacik orig_slot = path->slots[0]; 37523212fa14SJosef Bacik orig_offset = btrfs_item_offset(leaf, path->slots[0]); 37533212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 3754459931ecSChris Mason 3755459931ecSChris Mason buf = kmalloc(item_size, GFP_NOFS); 3756ad48fd75SYan, Zheng if (!buf) 3757ad48fd75SYan, Zheng return -ENOMEM; 3758ad48fd75SYan, Zheng 3759459931ecSChris Mason read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, 3760459931ecSChris Mason path->slots[0]), item_size); 3761ad48fd75SYan, Zheng 3762459931ecSChris Mason slot = path->slots[0] + 1; 3763459931ecSChris Mason nritems = btrfs_header_nritems(leaf); 3764459931ecSChris Mason if (slot != nritems) { 3765459931ecSChris Mason /* shift the items */ 3766459931ecSChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), 3767459931ecSChris Mason btrfs_item_nr_offset(slot), 3768459931ecSChris Mason (nritems - slot) * sizeof(struct btrfs_item)); 3769459931ecSChris Mason } 3770459931ecSChris Mason 3771459931ecSChris Mason btrfs_cpu_key_to_disk(&disk_key, new_key); 3772459931ecSChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 3773459931ecSChris Mason 37743212fa14SJosef Bacik btrfs_set_item_offset(leaf, slot, orig_offset); 37753212fa14SJosef Bacik btrfs_set_item_size(leaf, slot, item_size - split_offset); 3776459931ecSChris Mason 37773212fa14SJosef Bacik btrfs_set_item_offset(leaf, orig_slot, 3778459931ecSChris Mason orig_offset + item_size - split_offset); 37793212fa14SJosef Bacik btrfs_set_item_size(leaf, orig_slot, split_offset); 3780459931ecSChris Mason 3781459931ecSChris Mason btrfs_set_header_nritems(leaf, nritems + 1); 3782459931ecSChris Mason 3783459931ecSChris Mason /* write the data for the start of the original item */ 3784459931ecSChris Mason write_extent_buffer(leaf, buf, 3785459931ecSChris Mason btrfs_item_ptr_offset(leaf, path->slots[0]), 3786459931ecSChris Mason split_offset); 3787459931ecSChris Mason 3788459931ecSChris Mason /* write the data for the new item */ 3789459931ecSChris Mason write_extent_buffer(leaf, buf + split_offset, 3790459931ecSChris Mason btrfs_item_ptr_offset(leaf, slot), 3791459931ecSChris Mason item_size - split_offset); 3792459931ecSChris Mason btrfs_mark_buffer_dirty(leaf); 3793459931ecSChris Mason 3794e902baacSDavid Sterba BUG_ON(btrfs_leaf_free_space(leaf) < 0); 3795459931ecSChris Mason kfree(buf); 3796ad48fd75SYan, Zheng return 0; 3797ad48fd75SYan, Zheng } 3798ad48fd75SYan, Zheng 3799ad48fd75SYan, Zheng /* 3800ad48fd75SYan, Zheng * This function splits a single item into two items, 3801ad48fd75SYan, Zheng * giving 'new_key' to the new item and splitting the 3802ad48fd75SYan, Zheng * old one at split_offset (from the start of the item). 3803ad48fd75SYan, Zheng * 3804ad48fd75SYan, Zheng * The path may be released by this operation. After 3805ad48fd75SYan, Zheng * the split, the path is pointing to the old item. The 3806ad48fd75SYan, Zheng * new item is going to be in the same node as the old one. 3807ad48fd75SYan, Zheng * 3808ad48fd75SYan, Zheng * Note, the item being split must be smaller enough to live alone on 3809ad48fd75SYan, Zheng * a tree block with room for one extra struct btrfs_item 3810ad48fd75SYan, Zheng * 3811ad48fd75SYan, Zheng * This allows us to split the item in place, keeping a lock on the 3812ad48fd75SYan, Zheng * leaf the entire time. 3813ad48fd75SYan, Zheng */ 3814ad48fd75SYan, Zheng int btrfs_split_item(struct btrfs_trans_handle *trans, 3815ad48fd75SYan, Zheng struct btrfs_root *root, 3816ad48fd75SYan, Zheng struct btrfs_path *path, 3817310712b2SOmar Sandoval const struct btrfs_key *new_key, 3818ad48fd75SYan, Zheng unsigned long split_offset) 3819ad48fd75SYan, Zheng { 3820ad48fd75SYan, Zheng int ret; 3821ad48fd75SYan, Zheng ret = setup_leaf_for_split(trans, root, path, 3822ad48fd75SYan, Zheng sizeof(struct btrfs_item)); 3823ad48fd75SYan, Zheng if (ret) 3824459931ecSChris Mason return ret; 3825ad48fd75SYan, Zheng 382625263cd7SDavid Sterba ret = split_item(path, new_key, split_offset); 3827ad48fd75SYan, Zheng return ret; 3828ad48fd75SYan, Zheng } 3829ad48fd75SYan, Zheng 3830ad48fd75SYan, Zheng /* 3831d352ac68SChris Mason * make the item pointed to by the path smaller. new_size indicates 3832d352ac68SChris Mason * how small to make it, and from_end tells us if we just chop bytes 3833d352ac68SChris Mason * off the end of the item or if we shift the item to chop bytes off 3834d352ac68SChris Mason * the front. 3835d352ac68SChris Mason */ 383678ac4f9eSDavid Sterba void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) 3837b18c6685SChris Mason { 3838b18c6685SChris Mason int slot; 38395f39d397SChris Mason struct extent_buffer *leaf; 3840b18c6685SChris Mason u32 nritems; 3841b18c6685SChris Mason unsigned int data_end; 3842b18c6685SChris Mason unsigned int old_data_start; 3843b18c6685SChris Mason unsigned int old_size; 3844b18c6685SChris Mason unsigned int size_diff; 3845b18c6685SChris Mason int i; 3846cfed81a0SChris Mason struct btrfs_map_token token; 3847cfed81a0SChris Mason 38485f39d397SChris Mason leaf = path->nodes[0]; 3849179e29e4SChris Mason slot = path->slots[0]; 3850179e29e4SChris Mason 38513212fa14SJosef Bacik old_size = btrfs_item_size(leaf, slot); 3852179e29e4SChris Mason if (old_size == new_size) 3853143bede5SJeff Mahoney return; 3854b18c6685SChris Mason 38555f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 38568f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 3857b18c6685SChris Mason 38583212fa14SJosef Bacik old_data_start = btrfs_item_offset(leaf, slot); 3859179e29e4SChris Mason 3860b18c6685SChris Mason size_diff = old_size - new_size; 3861b18c6685SChris Mason 3862b18c6685SChris Mason BUG_ON(slot < 0); 3863b18c6685SChris Mason BUG_ON(slot >= nritems); 3864b18c6685SChris Mason 3865b18c6685SChris Mason /* 3866b18c6685SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 3867b18c6685SChris Mason */ 3868b18c6685SChris Mason /* first correct the data pointers */ 3869c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 3870b18c6685SChris Mason for (i = slot; i < nritems; i++) { 38715f39d397SChris Mason u32 ioff; 3872db94535dSChris Mason 38733212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 38743212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff + size_diff); 3875b18c6685SChris Mason } 3876db94535dSChris Mason 3877b18c6685SChris Mason /* shift the data */ 3878179e29e4SChris Mason if (from_end) { 38793d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 38803d9ec8c4SNikolay Borisov data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + 3881b18c6685SChris Mason data_end, old_data_start + new_size - data_end); 3882179e29e4SChris Mason } else { 3883179e29e4SChris Mason struct btrfs_disk_key disk_key; 3884179e29e4SChris Mason u64 offset; 3885179e29e4SChris Mason 3886179e29e4SChris Mason btrfs_item_key(leaf, &disk_key, slot); 3887179e29e4SChris Mason 3888179e29e4SChris Mason if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { 3889179e29e4SChris Mason unsigned long ptr; 3890179e29e4SChris Mason struct btrfs_file_extent_item *fi; 3891179e29e4SChris Mason 3892179e29e4SChris Mason fi = btrfs_item_ptr(leaf, slot, 3893179e29e4SChris Mason struct btrfs_file_extent_item); 3894179e29e4SChris Mason fi = (struct btrfs_file_extent_item *)( 3895179e29e4SChris Mason (unsigned long)fi - size_diff); 3896179e29e4SChris Mason 3897179e29e4SChris Mason if (btrfs_file_extent_type(leaf, fi) == 3898179e29e4SChris Mason BTRFS_FILE_EXTENT_INLINE) { 3899179e29e4SChris Mason ptr = btrfs_item_ptr_offset(leaf, slot); 3900179e29e4SChris Mason memmove_extent_buffer(leaf, ptr, 3901179e29e4SChris Mason (unsigned long)fi, 39027ec20afbSDavid Sterba BTRFS_FILE_EXTENT_INLINE_DATA_START); 3903179e29e4SChris Mason } 3904179e29e4SChris Mason } 3905179e29e4SChris Mason 39063d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 39073d9ec8c4SNikolay Borisov data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + 3908179e29e4SChris Mason data_end, old_data_start - data_end); 3909179e29e4SChris Mason 3910179e29e4SChris Mason offset = btrfs_disk_key_offset(&disk_key); 3911179e29e4SChris Mason btrfs_set_disk_key_offset(&disk_key, offset + size_diff); 3912179e29e4SChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 3913179e29e4SChris Mason if (slot == 0) 3914b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 3915179e29e4SChris Mason } 39165f39d397SChris Mason 39173212fa14SJosef Bacik btrfs_set_item_size(leaf, slot, new_size); 39185f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 3919b18c6685SChris Mason 3920e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 3921a4f78750SDavid Sterba btrfs_print_leaf(leaf); 3922b18c6685SChris Mason BUG(); 39235f39d397SChris Mason } 3924b18c6685SChris Mason } 3925b18c6685SChris Mason 3926d352ac68SChris Mason /* 39278f69dbd2SStefan Behrens * make the item pointed to by the path bigger, data_size is the added size. 3928d352ac68SChris Mason */ 3929c71dd880SDavid Sterba void btrfs_extend_item(struct btrfs_path *path, u32 data_size) 39306567e837SChris Mason { 39316567e837SChris Mason int slot; 39325f39d397SChris Mason struct extent_buffer *leaf; 39336567e837SChris Mason u32 nritems; 39346567e837SChris Mason unsigned int data_end; 39356567e837SChris Mason unsigned int old_data; 39366567e837SChris Mason unsigned int old_size; 39376567e837SChris Mason int i; 3938cfed81a0SChris Mason struct btrfs_map_token token; 3939cfed81a0SChris Mason 39405f39d397SChris Mason leaf = path->nodes[0]; 39416567e837SChris Mason 39425f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 39438f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 39446567e837SChris Mason 3945e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < data_size) { 3946a4f78750SDavid Sterba btrfs_print_leaf(leaf); 39476567e837SChris Mason BUG(); 39485f39d397SChris Mason } 39496567e837SChris Mason slot = path->slots[0]; 3950dc2e724eSJosef Bacik old_data = btrfs_item_data_end(leaf, slot); 39516567e837SChris Mason 39526567e837SChris Mason BUG_ON(slot < 0); 39533326d1b0SChris Mason if (slot >= nritems) { 3954a4f78750SDavid Sterba btrfs_print_leaf(leaf); 3955c71dd880SDavid Sterba btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", 3956d397712bSChris Mason slot, nritems); 3957290342f6SArnd Bergmann BUG(); 39583326d1b0SChris Mason } 39596567e837SChris Mason 39606567e837SChris Mason /* 39616567e837SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 39626567e837SChris Mason */ 39636567e837SChris Mason /* first correct the data pointers */ 3964c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 39656567e837SChris Mason for (i = slot; i < nritems; i++) { 39665f39d397SChris Mason u32 ioff; 3967db94535dSChris Mason 39683212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 39693212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff - data_size); 39706567e837SChris Mason } 39715f39d397SChris Mason 39726567e837SChris Mason /* shift the data */ 39733d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 39743d9ec8c4SNikolay Borisov data_end - data_size, BTRFS_LEAF_DATA_OFFSET + 39756567e837SChris Mason data_end, old_data - data_end); 39765f39d397SChris Mason 39776567e837SChris Mason data_end = old_data; 39783212fa14SJosef Bacik old_size = btrfs_item_size(leaf, slot); 39793212fa14SJosef Bacik btrfs_set_item_size(leaf, slot, old_size + data_size); 39805f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 39816567e837SChris Mason 3982e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 3983a4f78750SDavid Sterba btrfs_print_leaf(leaf); 39846567e837SChris Mason BUG(); 39855f39d397SChris Mason } 39866567e837SChris Mason } 39876567e837SChris Mason 3988*43dd529aSDavid Sterba /* 3989*43dd529aSDavid Sterba * Make space in the node before inserting one or more items. 3990da9ffb24SNikolay Borisov * 3991da9ffb24SNikolay Borisov * @root: root we are inserting items to 3992da9ffb24SNikolay Borisov * @path: points to the leaf/slot where we are going to insert new items 3993b7ef5f3aSFilipe Manana * @batch: information about the batch of items to insert 3994*43dd529aSDavid Sterba * 3995*43dd529aSDavid Sterba * Main purpose is to save stack depth by doing the bulk of the work in a 3996*43dd529aSDavid Sterba * function that doesn't call btrfs_search_slot 399774123bd7SChris Mason */ 3998f0641656SFilipe Manana static void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, 3999b7ef5f3aSFilipe Manana const struct btrfs_item_batch *batch) 4000be0e5c09SChris Mason { 40010b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 40029c58309dSChris Mason int i; 40037518a238SChris Mason u32 nritems; 4004be0e5c09SChris Mason unsigned int data_end; 4005e2fa7227SChris Mason struct btrfs_disk_key disk_key; 400644871b1bSChris Mason struct extent_buffer *leaf; 400744871b1bSChris Mason int slot; 4008cfed81a0SChris Mason struct btrfs_map_token token; 4009fc0d82e1SNikolay Borisov u32 total_size; 4010fc0d82e1SNikolay Borisov 4011b7ef5f3aSFilipe Manana /* 4012b7ef5f3aSFilipe Manana * Before anything else, update keys in the parent and other ancestors 4013b7ef5f3aSFilipe Manana * if needed, then release the write locks on them, so that other tasks 4014b7ef5f3aSFilipe Manana * can use them while we modify the leaf. 4015b7ef5f3aSFilipe Manana */ 401624cdc847SFilipe Manana if (path->slots[0] == 0) { 4017b7ef5f3aSFilipe Manana btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]); 4018b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 401924cdc847SFilipe Manana } 402024cdc847SFilipe Manana btrfs_unlock_up_safe(path, 1); 402124cdc847SFilipe Manana 40225f39d397SChris Mason leaf = path->nodes[0]; 402344871b1bSChris Mason slot = path->slots[0]; 402474123bd7SChris Mason 40255f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 40268f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 4027b7ef5f3aSFilipe Manana total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); 4028eb60ceacSChris Mason 4029e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < total_size) { 4030a4f78750SDavid Sterba btrfs_print_leaf(leaf); 40310b246afaSJeff Mahoney btrfs_crit(fs_info, "not enough freespace need %u have %d", 4032e902baacSDavid Sterba total_size, btrfs_leaf_free_space(leaf)); 4033be0e5c09SChris Mason BUG(); 4034d4dbff95SChris Mason } 40355f39d397SChris Mason 4036c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 4037be0e5c09SChris Mason if (slot != nritems) { 4038dc2e724eSJosef Bacik unsigned int old_data = btrfs_item_data_end(leaf, slot); 4039be0e5c09SChris Mason 40405f39d397SChris Mason if (old_data < data_end) { 4041a4f78750SDavid Sterba btrfs_print_leaf(leaf); 40427269ddd2SNikolay Borisov btrfs_crit(fs_info, 40437269ddd2SNikolay Borisov "item at slot %d with data offset %u beyond data end of leaf %u", 40445f39d397SChris Mason slot, old_data, data_end); 4045290342f6SArnd Bergmann BUG(); 40465f39d397SChris Mason } 4047be0e5c09SChris Mason /* 4048be0e5c09SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 4049be0e5c09SChris Mason */ 4050be0e5c09SChris Mason /* first correct the data pointers */ 40510783fcfcSChris Mason for (i = slot; i < nritems; i++) { 40525f39d397SChris Mason u32 ioff; 4053db94535dSChris Mason 40543212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 40553212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, 4056b7ef5f3aSFilipe Manana ioff - batch->total_data_size); 40570783fcfcSChris Mason } 4058be0e5c09SChris Mason /* shift the items */ 4059b7ef5f3aSFilipe Manana memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + batch->nr), 40605f39d397SChris Mason btrfs_item_nr_offset(slot), 40610783fcfcSChris Mason (nritems - slot) * sizeof(struct btrfs_item)); 4062be0e5c09SChris Mason 4063be0e5c09SChris Mason /* shift the data */ 40643d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 4065b7ef5f3aSFilipe Manana data_end - batch->total_data_size, 4066b7ef5f3aSFilipe Manana BTRFS_LEAF_DATA_OFFSET + data_end, 4067b7ef5f3aSFilipe Manana old_data - data_end); 4068be0e5c09SChris Mason data_end = old_data; 4069be0e5c09SChris Mason } 40705f39d397SChris Mason 407162e2749eSChris Mason /* setup the item for the new data */ 4072b7ef5f3aSFilipe Manana for (i = 0; i < batch->nr; i++) { 4073b7ef5f3aSFilipe Manana btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]); 40749c58309dSChris Mason btrfs_set_item_key(leaf, &disk_key, slot + i); 4075b7ef5f3aSFilipe Manana data_end -= batch->data_sizes[i]; 40763212fa14SJosef Bacik btrfs_set_token_item_offset(&token, slot + i, data_end); 40773212fa14SJosef Bacik btrfs_set_token_item_size(&token, slot + i, batch->data_sizes[i]); 40789c58309dSChris Mason } 407944871b1bSChris Mason 4080b7ef5f3aSFilipe Manana btrfs_set_header_nritems(leaf, nritems + batch->nr); 4081b9473439SChris Mason btrfs_mark_buffer_dirty(leaf); 4082aa5d6bedSChris Mason 4083e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4084a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4085be0e5c09SChris Mason BUG(); 40865f39d397SChris Mason } 408744871b1bSChris Mason } 408844871b1bSChris Mason 408944871b1bSChris Mason /* 4090f0641656SFilipe Manana * Insert a new item into a leaf. 4091f0641656SFilipe Manana * 4092f0641656SFilipe Manana * @root: The root of the btree. 4093f0641656SFilipe Manana * @path: A path pointing to the target leaf and slot. 4094f0641656SFilipe Manana * @key: The key of the new item. 4095f0641656SFilipe Manana * @data_size: The size of the data associated with the new key. 4096f0641656SFilipe Manana */ 4097f0641656SFilipe Manana void btrfs_setup_item_for_insert(struct btrfs_root *root, 4098f0641656SFilipe Manana struct btrfs_path *path, 4099f0641656SFilipe Manana const struct btrfs_key *key, 4100f0641656SFilipe Manana u32 data_size) 4101f0641656SFilipe Manana { 4102f0641656SFilipe Manana struct btrfs_item_batch batch; 4103f0641656SFilipe Manana 4104f0641656SFilipe Manana batch.keys = key; 4105f0641656SFilipe Manana batch.data_sizes = &data_size; 4106f0641656SFilipe Manana batch.total_data_size = data_size; 4107f0641656SFilipe Manana batch.nr = 1; 4108f0641656SFilipe Manana 4109f0641656SFilipe Manana setup_items_for_insert(root, path, &batch); 4110f0641656SFilipe Manana } 4111f0641656SFilipe Manana 4112f0641656SFilipe Manana /* 411344871b1bSChris Mason * Given a key and some data, insert items into the tree. 411444871b1bSChris Mason * This does all the path init required, making room in the tree if needed. 411544871b1bSChris Mason */ 411644871b1bSChris Mason int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, 411744871b1bSChris Mason struct btrfs_root *root, 411844871b1bSChris Mason struct btrfs_path *path, 4119b7ef5f3aSFilipe Manana const struct btrfs_item_batch *batch) 412044871b1bSChris Mason { 412144871b1bSChris Mason int ret = 0; 412244871b1bSChris Mason int slot; 4123b7ef5f3aSFilipe Manana u32 total_size; 412444871b1bSChris Mason 4125b7ef5f3aSFilipe Manana total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); 4126b7ef5f3aSFilipe Manana ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1); 412744871b1bSChris Mason if (ret == 0) 412844871b1bSChris Mason return -EEXIST; 412944871b1bSChris Mason if (ret < 0) 4130143bede5SJeff Mahoney return ret; 413144871b1bSChris Mason 413244871b1bSChris Mason slot = path->slots[0]; 413344871b1bSChris Mason BUG_ON(slot < 0); 413444871b1bSChris Mason 4135b7ef5f3aSFilipe Manana setup_items_for_insert(root, path, batch); 4136143bede5SJeff Mahoney return 0; 413762e2749eSChris Mason } 413862e2749eSChris Mason 413962e2749eSChris Mason /* 414062e2749eSChris Mason * Given a key and some data, insert an item into the tree. 414162e2749eSChris Mason * This does all the path init required, making room in the tree if needed. 414262e2749eSChris Mason */ 4143310712b2SOmar Sandoval int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, 4144310712b2SOmar Sandoval const struct btrfs_key *cpu_key, void *data, 4145310712b2SOmar Sandoval u32 data_size) 414662e2749eSChris Mason { 414762e2749eSChris Mason int ret = 0; 41482c90e5d6SChris Mason struct btrfs_path *path; 41495f39d397SChris Mason struct extent_buffer *leaf; 41505f39d397SChris Mason unsigned long ptr; 415162e2749eSChris Mason 41522c90e5d6SChris Mason path = btrfs_alloc_path(); 4153db5b493aSTsutomu Itoh if (!path) 4154db5b493aSTsutomu Itoh return -ENOMEM; 41552c90e5d6SChris Mason ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); 415662e2749eSChris Mason if (!ret) { 41575f39d397SChris Mason leaf = path->nodes[0]; 41585f39d397SChris Mason ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); 41595f39d397SChris Mason write_extent_buffer(leaf, data, ptr, data_size); 41605f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 416162e2749eSChris Mason } 41622c90e5d6SChris Mason btrfs_free_path(path); 4163aa5d6bedSChris Mason return ret; 4164be0e5c09SChris Mason } 4165be0e5c09SChris Mason 416674123bd7SChris Mason /* 4167f0641656SFilipe Manana * This function duplicates an item, giving 'new_key' to the new item. 4168f0641656SFilipe Manana * It guarantees both items live in the same tree leaf and the new item is 4169f0641656SFilipe Manana * contiguous with the original item. 4170f0641656SFilipe Manana * 4171f0641656SFilipe Manana * This allows us to split a file extent in place, keeping a lock on the leaf 4172f0641656SFilipe Manana * the entire time. 4173f0641656SFilipe Manana */ 4174f0641656SFilipe Manana int btrfs_duplicate_item(struct btrfs_trans_handle *trans, 4175f0641656SFilipe Manana struct btrfs_root *root, 4176f0641656SFilipe Manana struct btrfs_path *path, 4177f0641656SFilipe Manana const struct btrfs_key *new_key) 4178f0641656SFilipe Manana { 4179f0641656SFilipe Manana struct extent_buffer *leaf; 4180f0641656SFilipe Manana int ret; 4181f0641656SFilipe Manana u32 item_size; 4182f0641656SFilipe Manana 4183f0641656SFilipe Manana leaf = path->nodes[0]; 41843212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 4185f0641656SFilipe Manana ret = setup_leaf_for_split(trans, root, path, 4186f0641656SFilipe Manana item_size + sizeof(struct btrfs_item)); 4187f0641656SFilipe Manana if (ret) 4188f0641656SFilipe Manana return ret; 4189f0641656SFilipe Manana 4190f0641656SFilipe Manana path->slots[0]++; 4191f0641656SFilipe Manana btrfs_setup_item_for_insert(root, path, new_key, item_size); 4192f0641656SFilipe Manana leaf = path->nodes[0]; 4193f0641656SFilipe Manana memcpy_extent_buffer(leaf, 4194f0641656SFilipe Manana btrfs_item_ptr_offset(leaf, path->slots[0]), 4195f0641656SFilipe Manana btrfs_item_ptr_offset(leaf, path->slots[0] - 1), 4196f0641656SFilipe Manana item_size); 4197f0641656SFilipe Manana return 0; 4198f0641656SFilipe Manana } 4199f0641656SFilipe Manana 4200f0641656SFilipe Manana /* 42015de08d7dSChris Mason * delete the pointer from a given node. 420274123bd7SChris Mason * 4203d352ac68SChris Mason * the tree should have been previously balanced so the deletion does not 4204d352ac68SChris Mason * empty a node. 420574123bd7SChris Mason */ 4206afe5fea7STsutomu Itoh static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, 4207afe5fea7STsutomu Itoh int level, int slot) 4208be0e5c09SChris Mason { 42095f39d397SChris Mason struct extent_buffer *parent = path->nodes[level]; 42107518a238SChris Mason u32 nritems; 4211f3ea38daSJan Schmidt int ret; 4212be0e5c09SChris Mason 42135f39d397SChris Mason nritems = btrfs_header_nritems(parent); 4214be0e5c09SChris Mason if (slot != nritems - 1) { 4215bf1d3425SDavid Sterba if (level) { 4216f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_move(parent, slot, 4217f3a84ccdSFilipe Manana slot + 1, nritems - slot - 1); 4218bf1d3425SDavid Sterba BUG_ON(ret < 0); 4219bf1d3425SDavid Sterba } 42205f39d397SChris Mason memmove_extent_buffer(parent, 42215f39d397SChris Mason btrfs_node_key_ptr_offset(slot), 42225f39d397SChris Mason btrfs_node_key_ptr_offset(slot + 1), 4223d6025579SChris Mason sizeof(struct btrfs_key_ptr) * 4224d6025579SChris Mason (nritems - slot - 1)); 422557ba86c0SChris Mason } else if (level) { 4226f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, slot, 422733cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REMOVE); 422857ba86c0SChris Mason BUG_ON(ret < 0); 4229be0e5c09SChris Mason } 4230f3ea38daSJan Schmidt 42317518a238SChris Mason nritems--; 42325f39d397SChris Mason btrfs_set_header_nritems(parent, nritems); 42337518a238SChris Mason if (nritems == 0 && parent == root->node) { 42345f39d397SChris Mason BUG_ON(btrfs_header_level(root->node) != 1); 4235eb60ceacSChris Mason /* just turn the root into a leaf and break */ 42365f39d397SChris Mason btrfs_set_header_level(root->node, 0); 4237bb803951SChris Mason } else if (slot == 0) { 42385f39d397SChris Mason struct btrfs_disk_key disk_key; 42395f39d397SChris Mason 42405f39d397SChris Mason btrfs_node_key(parent, &disk_key, 0); 4241b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, level + 1); 4242be0e5c09SChris Mason } 4243d6025579SChris Mason btrfs_mark_buffer_dirty(parent); 4244be0e5c09SChris Mason } 4245be0e5c09SChris Mason 424674123bd7SChris Mason /* 4247323ac95bSChris Mason * a helper function to delete the leaf pointed to by path->slots[1] and 42485d4f98a2SYan Zheng * path->nodes[1]. 4249323ac95bSChris Mason * 4250323ac95bSChris Mason * This deletes the pointer in path->nodes[1] and frees the leaf 4251323ac95bSChris Mason * block extent. zero is returned if it all worked out, < 0 otherwise. 4252323ac95bSChris Mason * 4253323ac95bSChris Mason * The path must have already been setup for deleting the leaf, including 4254323ac95bSChris Mason * all the proper balancing. path->nodes[1] must be locked. 4255323ac95bSChris Mason */ 4256143bede5SJeff Mahoney static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, 4257323ac95bSChris Mason struct btrfs_root *root, 42585d4f98a2SYan Zheng struct btrfs_path *path, 42595d4f98a2SYan Zheng struct extent_buffer *leaf) 4260323ac95bSChris Mason { 42615d4f98a2SYan Zheng WARN_ON(btrfs_header_generation(leaf) != trans->transid); 4262afe5fea7STsutomu Itoh del_ptr(root, path, 1, path->slots[1]); 4263323ac95bSChris Mason 42644d081c41SChris Mason /* 42654d081c41SChris Mason * btrfs_free_extent is expensive, we want to make sure we 42664d081c41SChris Mason * aren't holding any locks when we call it 42674d081c41SChris Mason */ 42684d081c41SChris Mason btrfs_unlock_up_safe(path, 0); 42694d081c41SChris Mason 4270f0486c68SYan, Zheng root_sub_used(root, leaf->len); 4271f0486c68SYan, Zheng 427267439dadSDavid Sterba atomic_inc(&leaf->refs); 42737a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1); 42743083ee2eSJosef Bacik free_extent_buffer_stale(leaf); 4275323ac95bSChris Mason } 4276323ac95bSChris Mason /* 427774123bd7SChris Mason * delete the item at the leaf level in path. If that empties 427874123bd7SChris Mason * the leaf, remove it from the tree 427974123bd7SChris Mason */ 428085e21bacSChris Mason int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, 428185e21bacSChris Mason struct btrfs_path *path, int slot, int nr) 4282be0e5c09SChris Mason { 42830b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 42845f39d397SChris Mason struct extent_buffer *leaf; 4285aa5d6bedSChris Mason int ret = 0; 4286aa5d6bedSChris Mason int wret; 42877518a238SChris Mason u32 nritems; 4288be0e5c09SChris Mason 42895f39d397SChris Mason leaf = path->nodes[0]; 42905f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 4291be0e5c09SChris Mason 429285e21bacSChris Mason if (slot + nr != nritems) { 42930cae23b6SFilipe Manana const u32 last_off = btrfs_item_offset(leaf, slot + nr - 1); 42940cae23b6SFilipe Manana const int data_end = leaf_data_end(leaf); 4295c82f823cSDavid Sterba struct btrfs_map_token token; 42960cae23b6SFilipe Manana u32 dsize = 0; 42970cae23b6SFilipe Manana int i; 42980cae23b6SFilipe Manana 42990cae23b6SFilipe Manana for (i = 0; i < nr; i++) 43000cae23b6SFilipe Manana dsize += btrfs_item_size(leaf, slot + i); 43015f39d397SChris Mason 43023d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 4303d6025579SChris Mason data_end + dsize, 43043d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end, 430585e21bacSChris Mason last_off - data_end); 43065f39d397SChris Mason 4307c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 430885e21bacSChris Mason for (i = slot + nr; i < nritems; i++) { 43095f39d397SChris Mason u32 ioff; 4310db94535dSChris Mason 43113212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 43123212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff + dsize); 43130783fcfcSChris Mason } 4314db94535dSChris Mason 43155f39d397SChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), 431685e21bacSChris Mason btrfs_item_nr_offset(slot + nr), 43170783fcfcSChris Mason sizeof(struct btrfs_item) * 431885e21bacSChris Mason (nritems - slot - nr)); 4319be0e5c09SChris Mason } 432085e21bacSChris Mason btrfs_set_header_nritems(leaf, nritems - nr); 432185e21bacSChris Mason nritems -= nr; 43225f39d397SChris Mason 432374123bd7SChris Mason /* delete the leaf if we've emptied it */ 43247518a238SChris Mason if (nritems == 0) { 43255f39d397SChris Mason if (leaf == root->node) { 43265f39d397SChris Mason btrfs_set_header_level(leaf, 0); 43279a8dd150SChris Mason } else { 43286a884d7dSDavid Sterba btrfs_clean_tree_block(leaf); 4329143bede5SJeff Mahoney btrfs_del_leaf(trans, root, path, leaf); 43309a8dd150SChris Mason } 4331be0e5c09SChris Mason } else { 43327518a238SChris Mason int used = leaf_space_used(leaf, 0, nritems); 4333aa5d6bedSChris Mason if (slot == 0) { 43345f39d397SChris Mason struct btrfs_disk_key disk_key; 43355f39d397SChris Mason 43365f39d397SChris Mason btrfs_item_key(leaf, &disk_key, 0); 4337b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 4338aa5d6bedSChris Mason } 4339aa5d6bedSChris Mason 43407c4063d1SFilipe Manana /* 43417c4063d1SFilipe Manana * Try to delete the leaf if it is mostly empty. We do this by 43427c4063d1SFilipe Manana * trying to move all its items into its left and right neighbours. 43437c4063d1SFilipe Manana * If we can't move all the items, then we don't delete it - it's 43447c4063d1SFilipe Manana * not ideal, but future insertions might fill the leaf with more 43457c4063d1SFilipe Manana * items, or items from other leaves might be moved later into our 43467c4063d1SFilipe Manana * leaf due to deletions on those leaves. 43477c4063d1SFilipe Manana */ 43480b246afaSJeff Mahoney if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { 43497c4063d1SFilipe Manana u32 min_push_space; 43507c4063d1SFilipe Manana 4351be0e5c09SChris Mason /* push_leaf_left fixes the path. 4352be0e5c09SChris Mason * make sure the path still points to our leaf 4353be0e5c09SChris Mason * for possible call to del_ptr below 4354be0e5c09SChris Mason */ 43554920c9acSChris Mason slot = path->slots[1]; 435667439dadSDavid Sterba atomic_inc(&leaf->refs); 43577c4063d1SFilipe Manana /* 43587c4063d1SFilipe Manana * We want to be able to at least push one item to the 43597c4063d1SFilipe Manana * left neighbour leaf, and that's the first item. 43607c4063d1SFilipe Manana */ 43617c4063d1SFilipe Manana min_push_space = sizeof(struct btrfs_item) + 43627c4063d1SFilipe Manana btrfs_item_size(leaf, 0); 43637c4063d1SFilipe Manana wret = push_leaf_left(trans, root, path, 0, 43647c4063d1SFilipe Manana min_push_space, 1, (u32)-1); 436554aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4366aa5d6bedSChris Mason ret = wret; 43675f39d397SChris Mason 43685f39d397SChris Mason if (path->nodes[0] == leaf && 43695f39d397SChris Mason btrfs_header_nritems(leaf)) { 43707c4063d1SFilipe Manana /* 43717c4063d1SFilipe Manana * If we were not able to push all items from our 43727c4063d1SFilipe Manana * leaf to its left neighbour, then attempt to 43737c4063d1SFilipe Manana * either push all the remaining items to the 43747c4063d1SFilipe Manana * right neighbour or none. There's no advantage 43757c4063d1SFilipe Manana * in pushing only some items, instead of all, as 43767c4063d1SFilipe Manana * it's pointless to end up with a leaf having 43777c4063d1SFilipe Manana * too few items while the neighbours can be full 43787c4063d1SFilipe Manana * or nearly full. 43797c4063d1SFilipe Manana */ 43807c4063d1SFilipe Manana nritems = btrfs_header_nritems(leaf); 43817c4063d1SFilipe Manana min_push_space = leaf_space_used(leaf, 0, nritems); 43827c4063d1SFilipe Manana wret = push_leaf_right(trans, root, path, 0, 43837c4063d1SFilipe Manana min_push_space, 1, 0); 438454aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4385aa5d6bedSChris Mason ret = wret; 4386aa5d6bedSChris Mason } 43875f39d397SChris Mason 43885f39d397SChris Mason if (btrfs_header_nritems(leaf) == 0) { 4389323ac95bSChris Mason path->slots[1] = slot; 4390143bede5SJeff Mahoney btrfs_del_leaf(trans, root, path, leaf); 43915f39d397SChris Mason free_extent_buffer(leaf); 4392143bede5SJeff Mahoney ret = 0; 43935de08d7dSChris Mason } else { 4394925baeddSChris Mason /* if we're still in the path, make sure 4395925baeddSChris Mason * we're dirty. Otherwise, one of the 4396925baeddSChris Mason * push_leaf functions must have already 4397925baeddSChris Mason * dirtied this buffer 4398925baeddSChris Mason */ 4399925baeddSChris Mason if (path->nodes[0] == leaf) 44005f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 44015f39d397SChris Mason free_extent_buffer(leaf); 4402be0e5c09SChris Mason } 4403d5719762SChris Mason } else { 44045f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 4405be0e5c09SChris Mason } 44069a8dd150SChris Mason } 4407aa5d6bedSChris Mason return ret; 44089a8dd150SChris Mason } 44099a8dd150SChris Mason 441097571fd0SChris Mason /* 4411925baeddSChris Mason * search the tree again to find a leaf with lesser keys 44127bb86316SChris Mason * returns 0 if it found something or 1 if there are no lesser leaves. 44137bb86316SChris Mason * returns < 0 on io errors. 4414d352ac68SChris Mason * 4415d352ac68SChris Mason * This may release the path, and so you may lose any locks held at the 4416d352ac68SChris Mason * time you call it. 44177bb86316SChris Mason */ 441816e7549fSJosef Bacik int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) 44197bb86316SChris Mason { 4420925baeddSChris Mason struct btrfs_key key; 4421925baeddSChris Mason struct btrfs_disk_key found_key; 4422925baeddSChris Mason int ret; 44237bb86316SChris Mason 4424925baeddSChris Mason btrfs_item_key_to_cpu(path->nodes[0], &key, 0); 4425925baeddSChris Mason 4426e8b0d724SFilipe David Borba Manana if (key.offset > 0) { 4427925baeddSChris Mason key.offset--; 4428e8b0d724SFilipe David Borba Manana } else if (key.type > 0) { 4429925baeddSChris Mason key.type--; 4430e8b0d724SFilipe David Borba Manana key.offset = (u64)-1; 4431e8b0d724SFilipe David Borba Manana } else if (key.objectid > 0) { 4432925baeddSChris Mason key.objectid--; 4433e8b0d724SFilipe David Borba Manana key.type = (u8)-1; 4434e8b0d724SFilipe David Borba Manana key.offset = (u64)-1; 4435e8b0d724SFilipe David Borba Manana } else { 44367bb86316SChris Mason return 1; 4437e8b0d724SFilipe David Borba Manana } 44387bb86316SChris Mason 4439b3b4aa74SDavid Sterba btrfs_release_path(path); 4440925baeddSChris Mason ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4441925baeddSChris Mason if (ret < 0) 4442925baeddSChris Mason return ret; 4443925baeddSChris Mason btrfs_item_key(path->nodes[0], &found_key, 0); 4444925baeddSChris Mason ret = comp_keys(&found_key, &key); 4445337c6f68SFilipe Manana /* 4446337c6f68SFilipe Manana * We might have had an item with the previous key in the tree right 4447337c6f68SFilipe Manana * before we released our path. And after we released our path, that 4448337c6f68SFilipe Manana * item might have been pushed to the first slot (0) of the leaf we 4449337c6f68SFilipe Manana * were holding due to a tree balance. Alternatively, an item with the 4450337c6f68SFilipe Manana * previous key can exist as the only element of a leaf (big fat item). 4451337c6f68SFilipe Manana * Therefore account for these 2 cases, so that our callers (like 4452337c6f68SFilipe Manana * btrfs_previous_item) don't miss an existing item with a key matching 4453337c6f68SFilipe Manana * the previous key we computed above. 4454337c6f68SFilipe Manana */ 4455337c6f68SFilipe Manana if (ret <= 0) 44567bb86316SChris Mason return 0; 4457925baeddSChris Mason return 1; 44587bb86316SChris Mason } 44597bb86316SChris Mason 44603f157a2fSChris Mason /* 44613f157a2fSChris Mason * A helper function to walk down the tree starting at min_key, and looking 4462de78b51aSEric Sandeen * for nodes or leaves that are have a minimum transaction id. 4463de78b51aSEric Sandeen * This is used by the btree defrag code, and tree logging 44643f157a2fSChris Mason * 44653f157a2fSChris Mason * This does not cow, but it does stuff the starting key it finds back 44663f157a2fSChris Mason * into min_key, so you can call btrfs_search_slot with cow=1 on the 44673f157a2fSChris Mason * key and get a writable path. 44683f157a2fSChris Mason * 44693f157a2fSChris Mason * This honors path->lowest_level to prevent descent past a given level 44703f157a2fSChris Mason * of the tree. 44713f157a2fSChris Mason * 4472d352ac68SChris Mason * min_trans indicates the oldest transaction that you are interested 4473d352ac68SChris Mason * in walking through. Any nodes or leaves older than min_trans are 4474d352ac68SChris Mason * skipped over (without reading them). 4475d352ac68SChris Mason * 44763f157a2fSChris Mason * returns zero if something useful was found, < 0 on error and 1 if there 44773f157a2fSChris Mason * was nothing in the tree that matched the search criteria. 44783f157a2fSChris Mason */ 44793f157a2fSChris Mason int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, 4480de78b51aSEric Sandeen struct btrfs_path *path, 44813f157a2fSChris Mason u64 min_trans) 44823f157a2fSChris Mason { 44833f157a2fSChris Mason struct extent_buffer *cur; 44843f157a2fSChris Mason struct btrfs_key found_key; 44853f157a2fSChris Mason int slot; 44869652480bSYan int sret; 44873f157a2fSChris Mason u32 nritems; 44883f157a2fSChris Mason int level; 44893f157a2fSChris Mason int ret = 1; 4490f98de9b9SFilipe Manana int keep_locks = path->keep_locks; 44913f157a2fSChris Mason 4492c922b016SStefan Roesch ASSERT(!path->nowait); 4493f98de9b9SFilipe Manana path->keep_locks = 1; 44943f157a2fSChris Mason again: 4495bd681513SChris Mason cur = btrfs_read_lock_root_node(root); 44963f157a2fSChris Mason level = btrfs_header_level(cur); 4497e02119d5SChris Mason WARN_ON(path->nodes[level]); 44983f157a2fSChris Mason path->nodes[level] = cur; 4499bd681513SChris Mason path->locks[level] = BTRFS_READ_LOCK; 45003f157a2fSChris Mason 45013f157a2fSChris Mason if (btrfs_header_generation(cur) < min_trans) { 45023f157a2fSChris Mason ret = 1; 45033f157a2fSChris Mason goto out; 45043f157a2fSChris Mason } 45053f157a2fSChris Mason while (1) { 45063f157a2fSChris Mason nritems = btrfs_header_nritems(cur); 45073f157a2fSChris Mason level = btrfs_header_level(cur); 4508e3b83361SQu Wenruo sret = btrfs_bin_search(cur, min_key, &slot); 4509cbca7d59SFilipe Manana if (sret < 0) { 4510cbca7d59SFilipe Manana ret = sret; 4511cbca7d59SFilipe Manana goto out; 4512cbca7d59SFilipe Manana } 45133f157a2fSChris Mason 4514323ac95bSChris Mason /* at the lowest level, we're done, setup the path and exit */ 4515323ac95bSChris Mason if (level == path->lowest_level) { 4516e02119d5SChris Mason if (slot >= nritems) 4517e02119d5SChris Mason goto find_next_key; 45183f157a2fSChris Mason ret = 0; 45193f157a2fSChris Mason path->slots[level] = slot; 45203f157a2fSChris Mason btrfs_item_key_to_cpu(cur, &found_key, slot); 45213f157a2fSChris Mason goto out; 45223f157a2fSChris Mason } 45239652480bSYan if (sret && slot > 0) 45249652480bSYan slot--; 45253f157a2fSChris Mason /* 4526de78b51aSEric Sandeen * check this node pointer against the min_trans parameters. 4527260db43cSRandy Dunlap * If it is too old, skip to the next one. 45283f157a2fSChris Mason */ 45293f157a2fSChris Mason while (slot < nritems) { 45303f157a2fSChris Mason u64 gen; 4531e02119d5SChris Mason 45323f157a2fSChris Mason gen = btrfs_node_ptr_generation(cur, slot); 45333f157a2fSChris Mason if (gen < min_trans) { 45343f157a2fSChris Mason slot++; 45353f157a2fSChris Mason continue; 45363f157a2fSChris Mason } 45373f157a2fSChris Mason break; 45383f157a2fSChris Mason } 4539e02119d5SChris Mason find_next_key: 45403f157a2fSChris Mason /* 45413f157a2fSChris Mason * we didn't find a candidate key in this node, walk forward 45423f157a2fSChris Mason * and find another one 45433f157a2fSChris Mason */ 45443f157a2fSChris Mason if (slot >= nritems) { 4545e02119d5SChris Mason path->slots[level] = slot; 4546e02119d5SChris Mason sret = btrfs_find_next_key(root, path, min_key, level, 4547de78b51aSEric Sandeen min_trans); 4548e02119d5SChris Mason if (sret == 0) { 4549b3b4aa74SDavid Sterba btrfs_release_path(path); 45503f157a2fSChris Mason goto again; 45513f157a2fSChris Mason } else { 45523f157a2fSChris Mason goto out; 45533f157a2fSChris Mason } 45543f157a2fSChris Mason } 45553f157a2fSChris Mason /* save our key for returning back */ 45563f157a2fSChris Mason btrfs_node_key_to_cpu(cur, &found_key, slot); 45573f157a2fSChris Mason path->slots[level] = slot; 45583f157a2fSChris Mason if (level == path->lowest_level) { 45593f157a2fSChris Mason ret = 0; 45603f157a2fSChris Mason goto out; 45613f157a2fSChris Mason } 45624b231ae4SDavid Sterba cur = btrfs_read_node_slot(cur, slot); 4563fb770ae4SLiu Bo if (IS_ERR(cur)) { 4564fb770ae4SLiu Bo ret = PTR_ERR(cur); 4565fb770ae4SLiu Bo goto out; 4566fb770ae4SLiu Bo } 45673f157a2fSChris Mason 4568bd681513SChris Mason btrfs_tree_read_lock(cur); 4569b4ce94deSChris Mason 4570bd681513SChris Mason path->locks[level - 1] = BTRFS_READ_LOCK; 45713f157a2fSChris Mason path->nodes[level - 1] = cur; 4572f7c79f30SChris Mason unlock_up(path, level, 1, 0, NULL); 45733f157a2fSChris Mason } 45743f157a2fSChris Mason out: 4575f98de9b9SFilipe Manana path->keep_locks = keep_locks; 4576f98de9b9SFilipe Manana if (ret == 0) { 4577f98de9b9SFilipe Manana btrfs_unlock_up_safe(path, path->lowest_level + 1); 4578f98de9b9SFilipe Manana memcpy(min_key, &found_key, sizeof(found_key)); 4579f98de9b9SFilipe Manana } 45803f157a2fSChris Mason return ret; 45813f157a2fSChris Mason } 45823f157a2fSChris Mason 45833f157a2fSChris Mason /* 45843f157a2fSChris Mason * this is similar to btrfs_next_leaf, but does not try to preserve 45853f157a2fSChris Mason * and fixup the path. It looks for and returns the next key in the 4586de78b51aSEric Sandeen * tree based on the current path and the min_trans parameters. 45873f157a2fSChris Mason * 45883f157a2fSChris Mason * 0 is returned if another key is found, < 0 if there are any errors 45893f157a2fSChris Mason * and 1 is returned if there are no higher keys in the tree 45903f157a2fSChris Mason * 45913f157a2fSChris Mason * path->keep_locks should be set to 1 on the search made before 45923f157a2fSChris Mason * calling this function. 45933f157a2fSChris Mason */ 4594e7a84565SChris Mason int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, 4595de78b51aSEric Sandeen struct btrfs_key *key, int level, u64 min_trans) 4596e7a84565SChris Mason { 4597e7a84565SChris Mason int slot; 4598e7a84565SChris Mason struct extent_buffer *c; 4599e7a84565SChris Mason 46006a9fb468SJosef Bacik WARN_ON(!path->keep_locks && !path->skip_locking); 4601e7a84565SChris Mason while (level < BTRFS_MAX_LEVEL) { 4602e7a84565SChris Mason if (!path->nodes[level]) 4603e7a84565SChris Mason return 1; 4604e7a84565SChris Mason 4605e7a84565SChris Mason slot = path->slots[level] + 1; 4606e7a84565SChris Mason c = path->nodes[level]; 46073f157a2fSChris Mason next: 4608e7a84565SChris Mason if (slot >= btrfs_header_nritems(c)) { 460933c66f43SYan Zheng int ret; 461033c66f43SYan Zheng int orig_lowest; 461133c66f43SYan Zheng struct btrfs_key cur_key; 461233c66f43SYan Zheng if (level + 1 >= BTRFS_MAX_LEVEL || 461333c66f43SYan Zheng !path->nodes[level + 1]) 4614e7a84565SChris Mason return 1; 461533c66f43SYan Zheng 46166a9fb468SJosef Bacik if (path->locks[level + 1] || path->skip_locking) { 461733c66f43SYan Zheng level++; 4618e7a84565SChris Mason continue; 4619e7a84565SChris Mason } 462033c66f43SYan Zheng 462133c66f43SYan Zheng slot = btrfs_header_nritems(c) - 1; 462233c66f43SYan Zheng if (level == 0) 462333c66f43SYan Zheng btrfs_item_key_to_cpu(c, &cur_key, slot); 462433c66f43SYan Zheng else 462533c66f43SYan Zheng btrfs_node_key_to_cpu(c, &cur_key, slot); 462633c66f43SYan Zheng 462733c66f43SYan Zheng orig_lowest = path->lowest_level; 4628b3b4aa74SDavid Sterba btrfs_release_path(path); 462933c66f43SYan Zheng path->lowest_level = level; 463033c66f43SYan Zheng ret = btrfs_search_slot(NULL, root, &cur_key, path, 463133c66f43SYan Zheng 0, 0); 463233c66f43SYan Zheng path->lowest_level = orig_lowest; 463333c66f43SYan Zheng if (ret < 0) 463433c66f43SYan Zheng return ret; 463533c66f43SYan Zheng 463633c66f43SYan Zheng c = path->nodes[level]; 463733c66f43SYan Zheng slot = path->slots[level]; 463833c66f43SYan Zheng if (ret == 0) 463933c66f43SYan Zheng slot++; 464033c66f43SYan Zheng goto next; 464133c66f43SYan Zheng } 464233c66f43SYan Zheng 4643e7a84565SChris Mason if (level == 0) 4644e7a84565SChris Mason btrfs_item_key_to_cpu(c, key, slot); 46453f157a2fSChris Mason else { 46463f157a2fSChris Mason u64 gen = btrfs_node_ptr_generation(c, slot); 46473f157a2fSChris Mason 46483f157a2fSChris Mason if (gen < min_trans) { 46493f157a2fSChris Mason slot++; 46503f157a2fSChris Mason goto next; 46513f157a2fSChris Mason } 4652e7a84565SChris Mason btrfs_node_key_to_cpu(c, key, slot); 46533f157a2fSChris Mason } 4654e7a84565SChris Mason return 0; 4655e7a84565SChris Mason } 4656e7a84565SChris Mason return 1; 4657e7a84565SChris Mason } 4658e7a84565SChris Mason 46593d7806ecSJan Schmidt int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, 46603d7806ecSJan Schmidt u64 time_seq) 46613d7806ecSJan Schmidt { 4662d97e63b6SChris Mason int slot; 46638e73f275SChris Mason int level; 46645f39d397SChris Mason struct extent_buffer *c; 46658e73f275SChris Mason struct extent_buffer *next; 4666d96b3424SFilipe Manana struct btrfs_fs_info *fs_info = root->fs_info; 4667925baeddSChris Mason struct btrfs_key key; 4668d96b3424SFilipe Manana bool need_commit_sem = false; 4669925baeddSChris Mason u32 nritems; 4670925baeddSChris Mason int ret; 46710e46318dSJosef Bacik int i; 4672925baeddSChris Mason 4673bdcdd86cSFilipe Manana /* 4674bdcdd86cSFilipe Manana * The nowait semantics are used only for write paths, where we don't 4675bdcdd86cSFilipe Manana * use the tree mod log and sequence numbers. 4676bdcdd86cSFilipe Manana */ 4677bdcdd86cSFilipe Manana if (time_seq) 4678c922b016SStefan Roesch ASSERT(!path->nowait); 4679c922b016SStefan Roesch 4680925baeddSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 4681d397712bSChris Mason if (nritems == 0) 4682925baeddSChris Mason return 1; 4683925baeddSChris Mason 46848e73f275SChris Mason btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); 46858e73f275SChris Mason again: 46868e73f275SChris Mason level = 1; 46878e73f275SChris Mason next = NULL; 4688b3b4aa74SDavid Sterba btrfs_release_path(path); 46898e73f275SChris Mason 4690a2135011SChris Mason path->keep_locks = 1; 46918e73f275SChris Mason 4692d96b3424SFilipe Manana if (time_seq) { 46933d7806ecSJan Schmidt ret = btrfs_search_old_slot(root, &key, path, time_seq); 4694d96b3424SFilipe Manana } else { 4695d96b3424SFilipe Manana if (path->need_commit_sem) { 4696d96b3424SFilipe Manana path->need_commit_sem = 0; 4697d96b3424SFilipe Manana need_commit_sem = true; 4698bdcdd86cSFilipe Manana if (path->nowait) { 4699bdcdd86cSFilipe Manana if (!down_read_trylock(&fs_info->commit_root_sem)) { 4700bdcdd86cSFilipe Manana ret = -EAGAIN; 4701bdcdd86cSFilipe Manana goto done; 4702bdcdd86cSFilipe Manana } 4703bdcdd86cSFilipe Manana } else { 4704d96b3424SFilipe Manana down_read(&fs_info->commit_root_sem); 4705d96b3424SFilipe Manana } 4706bdcdd86cSFilipe Manana } 4707925baeddSChris Mason ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4708d96b3424SFilipe Manana } 4709925baeddSChris Mason path->keep_locks = 0; 4710925baeddSChris Mason 4711925baeddSChris Mason if (ret < 0) 4712d96b3424SFilipe Manana goto done; 4713925baeddSChris Mason 4714a2135011SChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 4715168fd7d2SChris Mason /* 4716168fd7d2SChris Mason * by releasing the path above we dropped all our locks. A balance 4717168fd7d2SChris Mason * could have added more items next to the key that used to be 4718168fd7d2SChris Mason * at the very end of the block. So, check again here and 4719168fd7d2SChris Mason * advance the path if there are now more items available. 4720168fd7d2SChris Mason */ 4721a2135011SChris Mason if (nritems > 0 && path->slots[0] < nritems - 1) { 4722e457afecSYan Zheng if (ret == 0) 4723168fd7d2SChris Mason path->slots[0]++; 47248e73f275SChris Mason ret = 0; 4725925baeddSChris Mason goto done; 4726925baeddSChris Mason } 47270b43e04fSLiu Bo /* 47280b43e04fSLiu Bo * So the above check misses one case: 47290b43e04fSLiu Bo * - after releasing the path above, someone has removed the item that 47300b43e04fSLiu Bo * used to be at the very end of the block, and balance between leafs 47310b43e04fSLiu Bo * gets another one with bigger key.offset to replace it. 47320b43e04fSLiu Bo * 47330b43e04fSLiu Bo * This one should be returned as well, or we can get leaf corruption 47340b43e04fSLiu Bo * later(esp. in __btrfs_drop_extents()). 47350b43e04fSLiu Bo * 47360b43e04fSLiu Bo * And a bit more explanation about this check, 47370b43e04fSLiu Bo * with ret > 0, the key isn't found, the path points to the slot 47380b43e04fSLiu Bo * where it should be inserted, so the path->slots[0] item must be the 47390b43e04fSLiu Bo * bigger one. 47400b43e04fSLiu Bo */ 47410b43e04fSLiu Bo if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { 47420b43e04fSLiu Bo ret = 0; 47430b43e04fSLiu Bo goto done; 47440b43e04fSLiu Bo } 4745d97e63b6SChris Mason 4746234b63a0SChris Mason while (level < BTRFS_MAX_LEVEL) { 47478e73f275SChris Mason if (!path->nodes[level]) { 47488e73f275SChris Mason ret = 1; 47498e73f275SChris Mason goto done; 47508e73f275SChris Mason } 47515f39d397SChris Mason 4752d97e63b6SChris Mason slot = path->slots[level] + 1; 4753d97e63b6SChris Mason c = path->nodes[level]; 47545f39d397SChris Mason if (slot >= btrfs_header_nritems(c)) { 4755d97e63b6SChris Mason level++; 47568e73f275SChris Mason if (level == BTRFS_MAX_LEVEL) { 47578e73f275SChris Mason ret = 1; 47588e73f275SChris Mason goto done; 47598e73f275SChris Mason } 4760d97e63b6SChris Mason continue; 4761d97e63b6SChris Mason } 47625f39d397SChris Mason 47630e46318dSJosef Bacik 47640e46318dSJosef Bacik /* 47650e46318dSJosef Bacik * Our current level is where we're going to start from, and to 47660e46318dSJosef Bacik * make sure lockdep doesn't complain we need to drop our locks 47670e46318dSJosef Bacik * and nodes from 0 to our current level. 47680e46318dSJosef Bacik */ 47690e46318dSJosef Bacik for (i = 0; i < level; i++) { 47700e46318dSJosef Bacik if (path->locks[level]) { 47710e46318dSJosef Bacik btrfs_tree_read_unlock(path->nodes[i]); 47720e46318dSJosef Bacik path->locks[i] = 0; 47730e46318dSJosef Bacik } 47740e46318dSJosef Bacik free_extent_buffer(path->nodes[i]); 47750e46318dSJosef Bacik path->nodes[i] = NULL; 4776925baeddSChris Mason } 47775f39d397SChris Mason 47788e73f275SChris Mason next = c; 4779d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 4780cda79c54SDavid Sterba slot, &key); 4781bdcdd86cSFilipe Manana if (ret == -EAGAIN && !path->nowait) 47828e73f275SChris Mason goto again; 47835f39d397SChris Mason 478476a05b35SChris Mason if (ret < 0) { 4785b3b4aa74SDavid Sterba btrfs_release_path(path); 478676a05b35SChris Mason goto done; 478776a05b35SChris Mason } 478876a05b35SChris Mason 47895cd57b2cSChris Mason if (!path->skip_locking) { 4790bd681513SChris Mason ret = btrfs_try_tree_read_lock(next); 4791bdcdd86cSFilipe Manana if (!ret && path->nowait) { 4792bdcdd86cSFilipe Manana ret = -EAGAIN; 4793bdcdd86cSFilipe Manana goto done; 4794bdcdd86cSFilipe Manana } 4795d42244a0SJan Schmidt if (!ret && time_seq) { 4796d42244a0SJan Schmidt /* 4797d42244a0SJan Schmidt * If we don't get the lock, we may be racing 4798d42244a0SJan Schmidt * with push_leaf_left, holding that lock while 4799d42244a0SJan Schmidt * itself waiting for the leaf we've currently 4800d42244a0SJan Schmidt * locked. To solve this situation, we give up 4801d42244a0SJan Schmidt * on our lock and cycle. 4802d42244a0SJan Schmidt */ 4803cf538830SJan Schmidt free_extent_buffer(next); 4804d42244a0SJan Schmidt btrfs_release_path(path); 4805d42244a0SJan Schmidt cond_resched(); 4806d42244a0SJan Schmidt goto again; 4807d42244a0SJan Schmidt } 48080e46318dSJosef Bacik if (!ret) 48090e46318dSJosef Bacik btrfs_tree_read_lock(next); 4810bd681513SChris Mason } 4811d97e63b6SChris Mason break; 4812d97e63b6SChris Mason } 4813d97e63b6SChris Mason path->slots[level] = slot; 4814d97e63b6SChris Mason while (1) { 4815d97e63b6SChris Mason level--; 4816d97e63b6SChris Mason path->nodes[level] = next; 4817d97e63b6SChris Mason path->slots[level] = 0; 4818a74a4b97SChris Mason if (!path->skip_locking) 4819ffeb03cfSJosef Bacik path->locks[level] = BTRFS_READ_LOCK; 4820d97e63b6SChris Mason if (!level) 4821d97e63b6SChris Mason break; 4822b4ce94deSChris Mason 4823d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 4824cda79c54SDavid Sterba 0, &key); 4825bdcdd86cSFilipe Manana if (ret == -EAGAIN && !path->nowait) 48268e73f275SChris Mason goto again; 48278e73f275SChris Mason 482876a05b35SChris Mason if (ret < 0) { 4829b3b4aa74SDavid Sterba btrfs_release_path(path); 483076a05b35SChris Mason goto done; 483176a05b35SChris Mason } 483276a05b35SChris Mason 4833bdcdd86cSFilipe Manana if (!path->skip_locking) { 4834bdcdd86cSFilipe Manana if (path->nowait) { 4835bdcdd86cSFilipe Manana if (!btrfs_try_tree_read_lock(next)) { 4836bdcdd86cSFilipe Manana ret = -EAGAIN; 4837bdcdd86cSFilipe Manana goto done; 4838bdcdd86cSFilipe Manana } 4839bdcdd86cSFilipe Manana } else { 48400e46318dSJosef Bacik btrfs_tree_read_lock(next); 4841d97e63b6SChris Mason } 4842bdcdd86cSFilipe Manana } 4843bdcdd86cSFilipe Manana } 48448e73f275SChris Mason ret = 0; 4845925baeddSChris Mason done: 4846f7c79f30SChris Mason unlock_up(path, 0, 1, 0, NULL); 4847d96b3424SFilipe Manana if (need_commit_sem) { 4848d96b3424SFilipe Manana int ret2; 4849d96b3424SFilipe Manana 4850d96b3424SFilipe Manana path->need_commit_sem = 1; 4851d96b3424SFilipe Manana ret2 = finish_need_commit_sem_search(path); 4852d96b3424SFilipe Manana up_read(&fs_info->commit_root_sem); 4853d96b3424SFilipe Manana if (ret2) 4854d96b3424SFilipe Manana ret = ret2; 4855d96b3424SFilipe Manana } 48568e73f275SChris Mason 48578e73f275SChris Mason return ret; 4858d97e63b6SChris Mason } 48590b86a832SChris Mason 4860890d2b1aSJosef Bacik int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq) 4861890d2b1aSJosef Bacik { 4862890d2b1aSJosef Bacik path->slots[0]++; 4863890d2b1aSJosef Bacik if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) 4864890d2b1aSJosef Bacik return btrfs_next_old_leaf(root, path, time_seq); 4865890d2b1aSJosef Bacik return 0; 4866890d2b1aSJosef Bacik } 4867890d2b1aSJosef Bacik 48683f157a2fSChris Mason /* 48693f157a2fSChris Mason * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps 48703f157a2fSChris Mason * searching until it gets past min_objectid or finds an item of 'type' 48713f157a2fSChris Mason * 48723f157a2fSChris Mason * returns 0 if something is found, 1 if nothing was found and < 0 on error 48733f157a2fSChris Mason */ 48740b86a832SChris Mason int btrfs_previous_item(struct btrfs_root *root, 48750b86a832SChris Mason struct btrfs_path *path, u64 min_objectid, 48760b86a832SChris Mason int type) 48770b86a832SChris Mason { 48780b86a832SChris Mason struct btrfs_key found_key; 48790b86a832SChris Mason struct extent_buffer *leaf; 4880e02119d5SChris Mason u32 nritems; 48810b86a832SChris Mason int ret; 48820b86a832SChris Mason 48830b86a832SChris Mason while (1) { 48840b86a832SChris Mason if (path->slots[0] == 0) { 48850b86a832SChris Mason ret = btrfs_prev_leaf(root, path); 48860b86a832SChris Mason if (ret != 0) 48870b86a832SChris Mason return ret; 48880b86a832SChris Mason } else { 48890b86a832SChris Mason path->slots[0]--; 48900b86a832SChris Mason } 48910b86a832SChris Mason leaf = path->nodes[0]; 4892e02119d5SChris Mason nritems = btrfs_header_nritems(leaf); 4893e02119d5SChris Mason if (nritems == 0) 4894e02119d5SChris Mason return 1; 4895e02119d5SChris Mason if (path->slots[0] == nritems) 4896e02119d5SChris Mason path->slots[0]--; 4897e02119d5SChris Mason 48980b86a832SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 4899e02119d5SChris Mason if (found_key.objectid < min_objectid) 4900e02119d5SChris Mason break; 49010a4eefbbSYan Zheng if (found_key.type == type) 49020a4eefbbSYan Zheng return 0; 4903e02119d5SChris Mason if (found_key.objectid == min_objectid && 4904e02119d5SChris Mason found_key.type < type) 4905e02119d5SChris Mason break; 49060b86a832SChris Mason } 49070b86a832SChris Mason return 1; 49080b86a832SChris Mason } 4909ade2e0b3SWang Shilong 4910ade2e0b3SWang Shilong /* 4911ade2e0b3SWang Shilong * search in extent tree to find a previous Metadata/Data extent item with 4912ade2e0b3SWang Shilong * min objecitd. 4913ade2e0b3SWang Shilong * 4914ade2e0b3SWang Shilong * returns 0 if something is found, 1 if nothing was found and < 0 on error 4915ade2e0b3SWang Shilong */ 4916ade2e0b3SWang Shilong int btrfs_previous_extent_item(struct btrfs_root *root, 4917ade2e0b3SWang Shilong struct btrfs_path *path, u64 min_objectid) 4918ade2e0b3SWang Shilong { 4919ade2e0b3SWang Shilong struct btrfs_key found_key; 4920ade2e0b3SWang Shilong struct extent_buffer *leaf; 4921ade2e0b3SWang Shilong u32 nritems; 4922ade2e0b3SWang Shilong int ret; 4923ade2e0b3SWang Shilong 4924ade2e0b3SWang Shilong while (1) { 4925ade2e0b3SWang Shilong if (path->slots[0] == 0) { 4926ade2e0b3SWang Shilong ret = btrfs_prev_leaf(root, path); 4927ade2e0b3SWang Shilong if (ret != 0) 4928ade2e0b3SWang Shilong return ret; 4929ade2e0b3SWang Shilong } else { 4930ade2e0b3SWang Shilong path->slots[0]--; 4931ade2e0b3SWang Shilong } 4932ade2e0b3SWang Shilong leaf = path->nodes[0]; 4933ade2e0b3SWang Shilong nritems = btrfs_header_nritems(leaf); 4934ade2e0b3SWang Shilong if (nritems == 0) 4935ade2e0b3SWang Shilong return 1; 4936ade2e0b3SWang Shilong if (path->slots[0] == nritems) 4937ade2e0b3SWang Shilong path->slots[0]--; 4938ade2e0b3SWang Shilong 4939ade2e0b3SWang Shilong btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 4940ade2e0b3SWang Shilong if (found_key.objectid < min_objectid) 4941ade2e0b3SWang Shilong break; 4942ade2e0b3SWang Shilong if (found_key.type == BTRFS_EXTENT_ITEM_KEY || 4943ade2e0b3SWang Shilong found_key.type == BTRFS_METADATA_ITEM_KEY) 4944ade2e0b3SWang Shilong return 0; 4945ade2e0b3SWang Shilong if (found_key.objectid == min_objectid && 4946ade2e0b3SWang Shilong found_key.type < BTRFS_EXTENT_ITEM_KEY) 4947ade2e0b3SWang Shilong break; 4948ade2e0b3SWang Shilong } 4949ade2e0b3SWang Shilong return 1; 4950ade2e0b3SWang Shilong } 4951226463d7SJosef Bacik 4952226463d7SJosef Bacik int __init btrfs_ctree_init(void) 4953226463d7SJosef Bacik { 4954226463d7SJosef Bacik btrfs_path_cachep = kmem_cache_create("btrfs_path", 4955226463d7SJosef Bacik sizeof(struct btrfs_path), 0, 4956226463d7SJosef Bacik SLAB_MEM_SPREAD, NULL); 4957226463d7SJosef Bacik if (!btrfs_path_cachep) 4958226463d7SJosef Bacik return -ENOMEM; 4959226463d7SJosef Bacik return 0; 4960226463d7SJosef Bacik } 4961226463d7SJosef Bacik 4962226463d7SJosef Bacik void __cold btrfs_ctree_exit(void) 4963226463d7SJosef Bacik { 4964226463d7SJosef Bacik kmem_cache_destroy(btrfs_path_cachep); 4965226463d7SJosef Bacik } 4966