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" 2467707479SJosef Bacik #include "relocation.h" 256bfd0ffaSJosef Bacik #include "file-item.h" 269a8dd150SChris Mason 27226463d7SJosef Bacik static struct kmem_cache *btrfs_path_cachep; 28226463d7SJosef Bacik 29e089f05cSChris Mason static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root 30e089f05cSChris Mason *root, struct btrfs_path *path, int level); 31310712b2SOmar Sandoval static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, 32310712b2SOmar Sandoval const struct btrfs_key *ins_key, struct btrfs_path *path, 33310712b2SOmar Sandoval int data_size, int extend); 345f39d397SChris Mason static int push_node_left(struct btrfs_trans_handle *trans, 352ff7e61eSJeff Mahoney struct extent_buffer *dst, 36971a1f66SChris Mason struct extent_buffer *src, int empty); 375f39d397SChris Mason static int balance_node_right(struct btrfs_trans_handle *trans, 385f39d397SChris Mason struct extent_buffer *dst_buf, 395f39d397SChris Mason struct extent_buffer *src_buf); 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 533a3178c7SJosef Bacik /* 543a3178c7SJosef Bacik * The leaf data grows from end-to-front in the node. this returns the address 553a3178c7SJosef Bacik * of the start of the last item, which is the stop of the leaf data stack. 563a3178c7SJosef Bacik */ 573a3178c7SJosef Bacik static unsigned int leaf_data_end(const struct extent_buffer *leaf) 583a3178c7SJosef Bacik { 593a3178c7SJosef Bacik u32 nr = btrfs_header_nritems(leaf); 603a3178c7SJosef Bacik 613a3178c7SJosef Bacik if (nr == 0) 623a3178c7SJosef Bacik return BTRFS_LEAF_DATA_SIZE(leaf->fs_info); 633a3178c7SJosef Bacik return btrfs_item_offset(leaf, nr - 1); 643a3178c7SJosef Bacik } 653a3178c7SJosef Bacik 66637e3b48SJosef Bacik /* 67637e3b48SJosef Bacik * Move data in a @leaf (using memmove, safe for overlapping ranges). 68637e3b48SJosef Bacik * 69637e3b48SJosef Bacik * @leaf: leaf that we're doing a memmove on 70637e3b48SJosef Bacik * @dst_offset: item data offset we're moving to 71637e3b48SJosef Bacik * @src_offset: item data offset were' moving from 72637e3b48SJosef Bacik * @len: length of the data we're moving 73637e3b48SJosef Bacik * 74637e3b48SJosef Bacik * Wrapper around memmove_extent_buffer() that takes into account the header on 75637e3b48SJosef Bacik * the leaf. The btrfs_item offset's start directly after the header, so we 76637e3b48SJosef Bacik * have to adjust any offsets to account for the header in the leaf. This 77637e3b48SJosef Bacik * handles that math to simplify the callers. 78637e3b48SJosef Bacik */ 79637e3b48SJosef Bacik static inline void memmove_leaf_data(const struct extent_buffer *leaf, 80637e3b48SJosef Bacik unsigned long dst_offset, 81637e3b48SJosef Bacik unsigned long src_offset, 82637e3b48SJosef Bacik unsigned long len) 83637e3b48SJosef Bacik { 848009adf3SJosef Bacik memmove_extent_buffer(leaf, btrfs_item_nr_offset(leaf, 0) + dst_offset, 858009adf3SJosef Bacik btrfs_item_nr_offset(leaf, 0) + src_offset, len); 86637e3b48SJosef Bacik } 87637e3b48SJosef Bacik 88637e3b48SJosef Bacik /* 89637e3b48SJosef Bacik * Copy item data from @src into @dst at the given @offset. 90637e3b48SJosef Bacik * 91637e3b48SJosef Bacik * @dst: destination leaf that we're copying into 92637e3b48SJosef Bacik * @src: source leaf that we're copying from 93637e3b48SJosef Bacik * @dst_offset: item data offset we're copying to 94637e3b48SJosef Bacik * @src_offset: item data offset were' copying from 95637e3b48SJosef Bacik * @len: length of the data we're copying 96637e3b48SJosef Bacik * 97637e3b48SJosef Bacik * Wrapper around copy_extent_buffer() that takes into account the header on 98637e3b48SJosef Bacik * the leaf. The btrfs_item offset's start directly after the header, so we 99637e3b48SJosef Bacik * have to adjust any offsets to account for the header in the leaf. This 100637e3b48SJosef Bacik * handles that math to simplify the callers. 101637e3b48SJosef Bacik */ 102637e3b48SJosef Bacik static inline void copy_leaf_data(const struct extent_buffer *dst, 103637e3b48SJosef Bacik const struct extent_buffer *src, 104637e3b48SJosef Bacik unsigned long dst_offset, 105637e3b48SJosef Bacik unsigned long src_offset, unsigned long len) 106637e3b48SJosef Bacik { 1078009adf3SJosef Bacik copy_extent_buffer(dst, src, btrfs_item_nr_offset(dst, 0) + dst_offset, 1088009adf3SJosef Bacik btrfs_item_nr_offset(src, 0) + src_offset, len); 109637e3b48SJosef Bacik } 110637e3b48SJosef Bacik 111637e3b48SJosef Bacik /* 112637e3b48SJosef Bacik * Move items in a @leaf (using memmove). 113637e3b48SJosef Bacik * 114637e3b48SJosef Bacik * @dst: destination leaf for the items 115637e3b48SJosef Bacik * @dst_item: the item nr we're copying into 116637e3b48SJosef Bacik * @src_item: the item nr we're copying from 117637e3b48SJosef Bacik * @nr_items: the number of items to copy 118637e3b48SJosef Bacik * 119637e3b48SJosef Bacik * Wrapper around memmove_extent_buffer() that does the math to get the 120637e3b48SJosef Bacik * appropriate offsets into the leaf from the item numbers. 121637e3b48SJosef Bacik */ 122637e3b48SJosef Bacik static inline void memmove_leaf_items(const struct extent_buffer *leaf, 123637e3b48SJosef Bacik int dst_item, int src_item, int nr_items) 124637e3b48SJosef Bacik { 125637e3b48SJosef Bacik memmove_extent_buffer(leaf, btrfs_item_nr_offset(leaf, dst_item), 126637e3b48SJosef Bacik btrfs_item_nr_offset(leaf, src_item), 127637e3b48SJosef Bacik nr_items * sizeof(struct btrfs_item)); 128637e3b48SJosef Bacik } 129637e3b48SJosef Bacik 130637e3b48SJosef Bacik /* 131637e3b48SJosef Bacik * Copy items from @src into @dst at the given @offset. 132637e3b48SJosef Bacik * 133637e3b48SJosef Bacik * @dst: destination leaf for the items 134637e3b48SJosef Bacik * @src: source leaf for the items 135637e3b48SJosef Bacik * @dst_item: the item nr we're copying into 136637e3b48SJosef Bacik * @src_item: the item nr we're copying from 137637e3b48SJosef Bacik * @nr_items: the number of items to copy 138637e3b48SJosef Bacik * 139637e3b48SJosef Bacik * Wrapper around copy_extent_buffer() that does the math to get the 140637e3b48SJosef Bacik * appropriate offsets into the leaf from the item numbers. 141637e3b48SJosef Bacik */ 142637e3b48SJosef Bacik static inline void copy_leaf_items(const struct extent_buffer *dst, 143637e3b48SJosef Bacik const struct extent_buffer *src, 144637e3b48SJosef Bacik int dst_item, int src_item, int nr_items) 145637e3b48SJosef Bacik { 146637e3b48SJosef Bacik copy_extent_buffer(dst, src, btrfs_item_nr_offset(dst, dst_item), 147637e3b48SJosef Bacik btrfs_item_nr_offset(src, src_item), 148637e3b48SJosef Bacik nr_items * sizeof(struct btrfs_item)); 149637e3b48SJosef Bacik } 150637e3b48SJosef Bacik 151b3cbfb0dSJosef Bacik /* This exists for btrfs-progs usages. */ 152b3cbfb0dSJosef Bacik u16 btrfs_csum_type_size(u16 type) 153b3cbfb0dSJosef Bacik { 154b3cbfb0dSJosef Bacik return btrfs_csums[type].size; 155b3cbfb0dSJosef Bacik } 156b3cbfb0dSJosef Bacik 157af024ed2SJohannes Thumshirn int btrfs_super_csum_size(const struct btrfs_super_block *s) 158af024ed2SJohannes Thumshirn { 159af024ed2SJohannes Thumshirn u16 t = btrfs_super_csum_type(s); 160af024ed2SJohannes Thumshirn /* 161af024ed2SJohannes Thumshirn * csum type is validated at mount time 162af024ed2SJohannes Thumshirn */ 163b3cbfb0dSJosef Bacik return btrfs_csum_type_size(t); 164af024ed2SJohannes Thumshirn } 165af024ed2SJohannes Thumshirn 166af024ed2SJohannes Thumshirn const char *btrfs_super_csum_name(u16 csum_type) 167af024ed2SJohannes Thumshirn { 168af024ed2SJohannes Thumshirn /* csum type is validated at mount time */ 169af024ed2SJohannes Thumshirn return btrfs_csums[csum_type].name; 170af024ed2SJohannes Thumshirn } 171af024ed2SJohannes Thumshirn 172b4e967beSDavid Sterba /* 173b4e967beSDavid Sterba * Return driver name if defined, otherwise the name that's also a valid driver 174b4e967beSDavid Sterba * name 175b4e967beSDavid Sterba */ 176b4e967beSDavid Sterba const char *btrfs_super_csum_driver(u16 csum_type) 177b4e967beSDavid Sterba { 178b4e967beSDavid Sterba /* csum type is validated at mount time */ 17959a0fcdbSDavid Sterba return btrfs_csums[csum_type].driver[0] ? 18059a0fcdbSDavid Sterba btrfs_csums[csum_type].driver : 181b4e967beSDavid Sterba btrfs_csums[csum_type].name; 182b4e967beSDavid Sterba } 183b4e967beSDavid Sterba 184604997b4SDavid Sterba size_t __attribute_const__ btrfs_get_num_csums(void) 185f7cea56cSDavid Sterba { 186f7cea56cSDavid Sterba return ARRAY_SIZE(btrfs_csums); 187f7cea56cSDavid Sterba } 188f7cea56cSDavid Sterba 1892c90e5d6SChris Mason struct btrfs_path *btrfs_alloc_path(void) 1902c90e5d6SChris Mason { 191a4c853afSChenXiaoSong might_sleep(); 192a4c853afSChenXiaoSong 193e2c89907SMasahiro Yamada return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); 1942c90e5d6SChris Mason } 1952c90e5d6SChris Mason 196d352ac68SChris Mason /* this also releases the path */ 1972c90e5d6SChris Mason void btrfs_free_path(struct btrfs_path *p) 1982c90e5d6SChris Mason { 199ff175d57SJesper Juhl if (!p) 200ff175d57SJesper Juhl return; 201b3b4aa74SDavid Sterba btrfs_release_path(p); 2022c90e5d6SChris Mason kmem_cache_free(btrfs_path_cachep, p); 2032c90e5d6SChris Mason } 2042c90e5d6SChris Mason 205d352ac68SChris Mason /* 206d352ac68SChris Mason * path release drops references on the extent buffers in the path 207d352ac68SChris Mason * and it drops any locks held by this path 208d352ac68SChris Mason * 209d352ac68SChris Mason * It is safe to call this on paths that no locks or extent buffers held. 210d352ac68SChris Mason */ 211b3b4aa74SDavid Sterba noinline void btrfs_release_path(struct btrfs_path *p) 212eb60ceacSChris Mason { 213eb60ceacSChris Mason int i; 214a2135011SChris Mason 215234b63a0SChris Mason for (i = 0; i < BTRFS_MAX_LEVEL; i++) { 2163f157a2fSChris Mason p->slots[i] = 0; 217eb60ceacSChris Mason if (!p->nodes[i]) 218925baeddSChris Mason continue; 219925baeddSChris Mason if (p->locks[i]) { 220bd681513SChris Mason btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); 221925baeddSChris Mason p->locks[i] = 0; 222925baeddSChris Mason } 2235f39d397SChris Mason free_extent_buffer(p->nodes[i]); 2243f157a2fSChris Mason p->nodes[i] = NULL; 225eb60ceacSChris Mason } 226eb60ceacSChris Mason } 227eb60ceacSChris Mason 228d352ac68SChris Mason /* 2298bb808c6SDavid Sterba * We want the transaction abort to print stack trace only for errors where the 2308bb808c6SDavid Sterba * cause could be a bug, eg. due to ENOSPC, and not for common errors that are 2318bb808c6SDavid Sterba * caused by external factors. 2328bb808c6SDavid Sterba */ 2338bb808c6SDavid Sterba bool __cold abort_should_print_stack(int errno) 2348bb808c6SDavid Sterba { 2358bb808c6SDavid Sterba switch (errno) { 2368bb808c6SDavid Sterba case -EIO: 2378bb808c6SDavid Sterba case -EROFS: 2388bb808c6SDavid Sterba case -ENOMEM: 2398bb808c6SDavid Sterba return false; 2408bb808c6SDavid Sterba } 2418bb808c6SDavid Sterba return true; 2428bb808c6SDavid Sterba } 2438bb808c6SDavid Sterba 2448bb808c6SDavid Sterba /* 245d352ac68SChris Mason * safely gets a reference on the root node of a tree. A lock 246d352ac68SChris Mason * is not taken, so a concurrent writer may put a different node 247d352ac68SChris Mason * at the root of the tree. See btrfs_lock_root_node for the 248d352ac68SChris Mason * looping required. 249d352ac68SChris Mason * 250d352ac68SChris Mason * The extent buffer returned by this has a reference taken, so 251d352ac68SChris Mason * it won't disappear. It may stop being the root of the tree 252d352ac68SChris Mason * at any time because there are no locks held. 253d352ac68SChris Mason */ 254925baeddSChris Mason struct extent_buffer *btrfs_root_node(struct btrfs_root *root) 255925baeddSChris Mason { 256925baeddSChris Mason struct extent_buffer *eb; 257240f62c8SChris Mason 2583083ee2eSJosef Bacik while (1) { 259240f62c8SChris Mason rcu_read_lock(); 260240f62c8SChris Mason eb = rcu_dereference(root->node); 2613083ee2eSJosef Bacik 2623083ee2eSJosef Bacik /* 2633083ee2eSJosef Bacik * RCU really hurts here, we could free up the root node because 26401327610SNicholas D Steeves * it was COWed but we may not get the new root node yet so do 2653083ee2eSJosef Bacik * the inc_not_zero dance and if it doesn't work then 2663083ee2eSJosef Bacik * synchronize_rcu and try again. 2673083ee2eSJosef Bacik */ 2683083ee2eSJosef Bacik if (atomic_inc_not_zero(&eb->refs)) { 269240f62c8SChris Mason rcu_read_unlock(); 2703083ee2eSJosef Bacik break; 2713083ee2eSJosef Bacik } 2723083ee2eSJosef Bacik rcu_read_unlock(); 2733083ee2eSJosef Bacik synchronize_rcu(); 2743083ee2eSJosef Bacik } 275925baeddSChris Mason return eb; 276925baeddSChris Mason } 277925baeddSChris Mason 27892a7cc42SQu Wenruo /* 27992a7cc42SQu Wenruo * Cowonly root (not-shareable trees, everything not subvolume or reloc roots), 28092a7cc42SQu Wenruo * just get put onto a simple dirty list. Transaction walks this list to make 28192a7cc42SQu Wenruo * sure they get properly updated on disk. 282d352ac68SChris Mason */ 2830b86a832SChris Mason static void add_root_to_dirty_list(struct btrfs_root *root) 2840b86a832SChris Mason { 2850b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 2860b246afaSJeff Mahoney 287e7070be1SJosef Bacik if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || 288e7070be1SJosef Bacik !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) 289e7070be1SJosef Bacik return; 290e7070be1SJosef Bacik 2910b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 292e7070be1SJosef Bacik if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { 293e7070be1SJosef Bacik /* Want the extent tree to be the last on the list */ 2944fd786e6SMisono Tomohiro if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) 295e7070be1SJosef Bacik list_move_tail(&root->dirty_list, 2960b246afaSJeff Mahoney &fs_info->dirty_cowonly_roots); 297e7070be1SJosef Bacik else 298e7070be1SJosef Bacik list_move(&root->dirty_list, 2990b246afaSJeff Mahoney &fs_info->dirty_cowonly_roots); 3000b86a832SChris Mason } 3010b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 3020b86a832SChris Mason } 3030b86a832SChris Mason 304d352ac68SChris Mason /* 305d352ac68SChris Mason * used by snapshot creation to make a copy of a root for a tree with 306d352ac68SChris Mason * a given objectid. The buffer with the new root node is returned in 307d352ac68SChris Mason * cow_ret, and this func returns zero on success or a negative error code. 308d352ac68SChris Mason */ 309be20aa9dSChris Mason int btrfs_copy_root(struct btrfs_trans_handle *trans, 310be20aa9dSChris Mason struct btrfs_root *root, 311be20aa9dSChris Mason struct extent_buffer *buf, 312be20aa9dSChris Mason struct extent_buffer **cow_ret, u64 new_root_objectid) 313be20aa9dSChris Mason { 3140b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 315be20aa9dSChris Mason struct extent_buffer *cow; 316be20aa9dSChris Mason int ret = 0; 317be20aa9dSChris Mason int level; 3185d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 319be20aa9dSChris Mason 32092a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 3210b246afaSJeff Mahoney trans->transid != fs_info->running_transaction->transid); 32292a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 32327cdeb70SMiao Xie trans->transid != root->last_trans); 324be20aa9dSChris Mason 325be20aa9dSChris Mason level = btrfs_header_level(buf); 3265d4f98a2SYan Zheng if (level == 0) 3275d4f98a2SYan Zheng btrfs_item_key(buf, &disk_key, 0); 3285d4f98a2SYan Zheng else 3295d4f98a2SYan Zheng btrfs_node_key(buf, &disk_key, 0); 33031840ae1SZheng Yan 3314d75f8a9SDavid Sterba cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, 332cf6f34aaSJosef Bacik &disk_key, level, buf->start, 0, 333cf6f34aaSJosef Bacik BTRFS_NESTING_NEW_ROOT); 3345d4f98a2SYan Zheng if (IS_ERR(cow)) 335be20aa9dSChris Mason return PTR_ERR(cow); 336be20aa9dSChris Mason 33758e8012cSDavid Sterba copy_extent_buffer_full(cow, buf); 338be20aa9dSChris Mason btrfs_set_header_bytenr(cow, cow->start); 339be20aa9dSChris Mason btrfs_set_header_generation(cow, trans->transid); 3405d4f98a2SYan Zheng btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 3415d4f98a2SYan Zheng btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 3425d4f98a2SYan Zheng BTRFS_HEADER_FLAG_RELOC); 3435d4f98a2SYan Zheng if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 3445d4f98a2SYan Zheng btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 3455d4f98a2SYan Zheng else 346be20aa9dSChris Mason btrfs_set_header_owner(cow, new_root_objectid); 347be20aa9dSChris Mason 348de37aa51SNikolay Borisov write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 3492b82032cSYan Zheng 350be20aa9dSChris Mason WARN_ON(btrfs_header_generation(buf) > trans->transid); 3515d4f98a2SYan Zheng if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 352e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 3535d4f98a2SYan Zheng else 354e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 355867ed321SJosef Bacik if (ret) { 35672c9925fSFilipe Manana btrfs_tree_unlock(cow); 35772c9925fSFilipe Manana free_extent_buffer(cow); 358867ed321SJosef Bacik btrfs_abort_transaction(trans, ret); 359be20aa9dSChris Mason return ret; 360867ed321SJosef Bacik } 361be20aa9dSChris Mason 362be20aa9dSChris Mason btrfs_mark_buffer_dirty(cow); 363be20aa9dSChris Mason *cow_ret = cow; 364be20aa9dSChris Mason return 0; 365be20aa9dSChris Mason } 366be20aa9dSChris Mason 367d352ac68SChris Mason /* 3685d4f98a2SYan Zheng * check if the tree block can be shared by multiple trees 3695d4f98a2SYan Zheng */ 3705d4f98a2SYan Zheng int btrfs_block_can_be_shared(struct btrfs_root *root, 3715d4f98a2SYan Zheng struct extent_buffer *buf) 3725d4f98a2SYan Zheng { 3735d4f98a2SYan Zheng /* 37492a7cc42SQu Wenruo * Tree blocks not in shareable trees and tree roots are never shared. 37592a7cc42SQu Wenruo * If a block was allocated after the last snapshot and the block was 37692a7cc42SQu Wenruo * not allocated by tree relocation, we know the block is not shared. 3775d4f98a2SYan Zheng */ 37892a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 3795d4f98a2SYan Zheng buf != root->node && buf != root->commit_root && 3805d4f98a2SYan Zheng (btrfs_header_generation(buf) <= 3815d4f98a2SYan Zheng btrfs_root_last_snapshot(&root->root_item) || 3825d4f98a2SYan Zheng btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) 3835d4f98a2SYan Zheng return 1; 384a79865c6SNikolay Borisov 3855d4f98a2SYan Zheng return 0; 3865d4f98a2SYan Zheng } 3875d4f98a2SYan Zheng 3885d4f98a2SYan Zheng static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, 3895d4f98a2SYan Zheng struct btrfs_root *root, 3905d4f98a2SYan Zheng struct extent_buffer *buf, 391f0486c68SYan, Zheng struct extent_buffer *cow, 392f0486c68SYan, Zheng int *last_ref) 3935d4f98a2SYan Zheng { 3940b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 3955d4f98a2SYan Zheng u64 refs; 3965d4f98a2SYan Zheng u64 owner; 3975d4f98a2SYan Zheng u64 flags; 3985d4f98a2SYan Zheng u64 new_flags = 0; 3995d4f98a2SYan Zheng int ret; 4005d4f98a2SYan Zheng 4015d4f98a2SYan Zheng /* 4025d4f98a2SYan Zheng * Backrefs update rules: 4035d4f98a2SYan Zheng * 4045d4f98a2SYan Zheng * Always use full backrefs for extent pointers in tree block 4055d4f98a2SYan Zheng * allocated by tree relocation. 4065d4f98a2SYan Zheng * 4075d4f98a2SYan Zheng * If a shared tree block is no longer referenced by its owner 4085d4f98a2SYan Zheng * tree (btrfs_header_owner(buf) == root->root_key.objectid), 4095d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 4105d4f98a2SYan Zheng * 4115d4f98a2SYan Zheng * If a tree block is been relocating 4125d4f98a2SYan Zheng * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), 4135d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 4145d4f98a2SYan Zheng * The reason for this is some operations (such as drop tree) 4155d4f98a2SYan Zheng * are only allowed for blocks use full backrefs. 4165d4f98a2SYan Zheng */ 4175d4f98a2SYan Zheng 4185d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, buf)) { 4192ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, 4203173a18fSJosef Bacik btrfs_header_level(buf), 1, 4213173a18fSJosef Bacik &refs, &flags); 422be1a5564SMark Fasheh if (ret) 423be1a5564SMark Fasheh return ret; 424eced687eSFilipe Manana if (unlikely(refs == 0)) { 425eced687eSFilipe Manana btrfs_crit(fs_info, 426eced687eSFilipe Manana "found 0 references for tree block at bytenr %llu level %d root %llu", 427eced687eSFilipe Manana buf->start, btrfs_header_level(buf), 428eced687eSFilipe Manana btrfs_root_id(root)); 429eced687eSFilipe Manana ret = -EUCLEAN; 430eced687eSFilipe Manana btrfs_abort_transaction(trans, ret); 431e5df9573SMark Fasheh return ret; 432e5df9573SMark Fasheh } 4335d4f98a2SYan Zheng } else { 4345d4f98a2SYan Zheng refs = 1; 4355d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 4365d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 4375d4f98a2SYan Zheng flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; 4385d4f98a2SYan Zheng else 4395d4f98a2SYan Zheng flags = 0; 4405d4f98a2SYan Zheng } 4415d4f98a2SYan Zheng 4425d4f98a2SYan Zheng owner = btrfs_header_owner(buf); 4435d4f98a2SYan Zheng BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && 4445d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); 4455d4f98a2SYan Zheng 4465d4f98a2SYan Zheng if (refs > 1) { 4475d4f98a2SYan Zheng if ((owner == root->root_key.objectid || 4485d4f98a2SYan Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && 4495d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { 450e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, buf, 1); 451692826b2SJeff Mahoney if (ret) 452692826b2SJeff Mahoney return ret; 4535d4f98a2SYan Zheng 4545d4f98a2SYan Zheng if (root->root_key.objectid == 4555d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) { 456e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 0); 457692826b2SJeff Mahoney if (ret) 458692826b2SJeff Mahoney return ret; 459e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 460692826b2SJeff Mahoney if (ret) 461692826b2SJeff Mahoney return ret; 4625d4f98a2SYan Zheng } 4635d4f98a2SYan Zheng new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; 4645d4f98a2SYan Zheng } else { 4655d4f98a2SYan Zheng 4665d4f98a2SYan Zheng if (root->root_key.objectid == 4675d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 468e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 4695d4f98a2SYan Zheng else 470e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 471692826b2SJeff Mahoney if (ret) 472692826b2SJeff Mahoney return ret; 4735d4f98a2SYan Zheng } 4745d4f98a2SYan Zheng if (new_flags != 0) { 4754aec05faSJosef Bacik ret = btrfs_set_disk_extent_flags(trans, buf, new_flags); 476be1a5564SMark Fasheh if (ret) 477be1a5564SMark Fasheh return ret; 4785d4f98a2SYan Zheng } 4795d4f98a2SYan Zheng } else { 4805d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { 4815d4f98a2SYan Zheng if (root->root_key.objectid == 4825d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 483e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 4845d4f98a2SYan Zheng else 485e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 486692826b2SJeff Mahoney if (ret) 487692826b2SJeff Mahoney return ret; 488e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 1); 489692826b2SJeff Mahoney if (ret) 490692826b2SJeff Mahoney return ret; 4915d4f98a2SYan Zheng } 492190a8339SJosef Bacik btrfs_clear_buffer_dirty(trans, buf); 493f0486c68SYan, Zheng *last_ref = 1; 4945d4f98a2SYan Zheng } 4955d4f98a2SYan Zheng return 0; 4965d4f98a2SYan Zheng } 4975d4f98a2SYan Zheng 4985d4f98a2SYan Zheng /* 499d397712bSChris Mason * does the dirty work in cow of a single block. The parent block (if 500d397712bSChris Mason * supplied) is updated to point to the new cow copy. The new buffer is marked 501d397712bSChris Mason * dirty and returned locked. If you modify the block it needs to be marked 502d397712bSChris Mason * dirty again. 503d352ac68SChris Mason * 504d352ac68SChris Mason * search_start -- an allocation hint for the new block 505d352ac68SChris Mason * 506d397712bSChris Mason * empty_size -- a hint that you plan on doing more cow. This is the size in 507d397712bSChris Mason * bytes the allocator should try to find free next to the block it returns. 508d397712bSChris Mason * This is just a hint and may be ignored by the allocator. 509d352ac68SChris Mason */ 510d397712bSChris Mason static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, 5115f39d397SChris Mason struct btrfs_root *root, 5125f39d397SChris Mason struct extent_buffer *buf, 5135f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 5145f39d397SChris Mason struct extent_buffer **cow_ret, 5159631e4ccSJosef Bacik u64 search_start, u64 empty_size, 5169631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 5176702ed49SChris Mason { 5180b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5195d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 5205f39d397SChris Mason struct extent_buffer *cow; 521be1a5564SMark Fasheh int level, ret; 522f0486c68SYan, Zheng int last_ref = 0; 523925baeddSChris Mason int unlock_orig = 0; 5240f5053ebSGoldwyn Rodrigues u64 parent_start = 0; 5256702ed49SChris Mason 526925baeddSChris Mason if (*cow_ret == buf) 527925baeddSChris Mason unlock_orig = 1; 528925baeddSChris Mason 52949d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 530925baeddSChris Mason 53192a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 5320b246afaSJeff Mahoney trans->transid != fs_info->running_transaction->transid); 53392a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 53427cdeb70SMiao Xie trans->transid != root->last_trans); 5355f39d397SChris Mason 5367bb86316SChris Mason level = btrfs_header_level(buf); 53731840ae1SZheng Yan 5385d4f98a2SYan Zheng if (level == 0) 5395d4f98a2SYan Zheng btrfs_item_key(buf, &disk_key, 0); 5405d4f98a2SYan Zheng else 5415d4f98a2SYan Zheng btrfs_node_key(buf, &disk_key, 0); 5425d4f98a2SYan Zheng 5430f5053ebSGoldwyn Rodrigues if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) 5445d4f98a2SYan Zheng parent_start = parent->start; 5455d4f98a2SYan Zheng 54679bd3712SFilipe Manana cow = btrfs_alloc_tree_block(trans, root, parent_start, 54779bd3712SFilipe Manana root->root_key.objectid, &disk_key, level, 54879bd3712SFilipe Manana search_start, empty_size, nest); 5496702ed49SChris Mason if (IS_ERR(cow)) 5506702ed49SChris Mason return PTR_ERR(cow); 5516702ed49SChris Mason 552b4ce94deSChris Mason /* cow is set to blocking by btrfs_init_new_buffer */ 553b4ce94deSChris Mason 55458e8012cSDavid Sterba copy_extent_buffer_full(cow, buf); 555db94535dSChris Mason btrfs_set_header_bytenr(cow, cow->start); 5565f39d397SChris Mason btrfs_set_header_generation(cow, trans->transid); 5575d4f98a2SYan Zheng btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 5585d4f98a2SYan Zheng btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 5595d4f98a2SYan Zheng BTRFS_HEADER_FLAG_RELOC); 5605d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 5615d4f98a2SYan Zheng btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 5625d4f98a2SYan Zheng else 5635f39d397SChris Mason btrfs_set_header_owner(cow, root->root_key.objectid); 5646702ed49SChris Mason 565de37aa51SNikolay Borisov write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 5662b82032cSYan Zheng 567be1a5564SMark Fasheh ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); 568b68dc2a9SMark Fasheh if (ret) { 569572c83acSJosef Bacik btrfs_tree_unlock(cow); 570572c83acSJosef Bacik free_extent_buffer(cow); 57166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 572b68dc2a9SMark Fasheh return ret; 573b68dc2a9SMark Fasheh } 5741a40e23bSZheng Yan 57592a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 57683d4cfd4SJosef Bacik ret = btrfs_reloc_cow_block(trans, root, buf, cow); 57793314e3bSZhaolei if (ret) { 578572c83acSJosef Bacik btrfs_tree_unlock(cow); 579572c83acSJosef Bacik free_extent_buffer(cow); 58066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 58183d4cfd4SJosef Bacik return ret; 58283d4cfd4SJosef Bacik } 58393314e3bSZhaolei } 5843fd0a558SYan, Zheng 5856702ed49SChris Mason if (buf == root->node) { 586925baeddSChris Mason WARN_ON(parent && parent != buf); 5875d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 5885d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 5895d4f98a2SYan Zheng parent_start = buf->start; 590925baeddSChris Mason 591406808abSFilipe Manana ret = btrfs_tree_mod_log_insert_root(root->node, cow, true); 59240b0a749SFilipe Manana if (ret < 0) { 59340b0a749SFilipe Manana btrfs_tree_unlock(cow); 59440b0a749SFilipe Manana free_extent_buffer(cow); 59540b0a749SFilipe Manana btrfs_abort_transaction(trans, ret); 59640b0a749SFilipe Manana return ret; 59740b0a749SFilipe Manana } 59840b0a749SFilipe Manana atomic_inc(&cow->refs); 599240f62c8SChris Mason rcu_assign_pointer(root->node, cow); 600925baeddSChris Mason 6017a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), buf, 6027a163608SFilipe Manana parent_start, last_ref); 6035f39d397SChris Mason free_extent_buffer(buf); 6040b86a832SChris Mason add_root_to_dirty_list(root); 6056702ed49SChris Mason } else { 6065d4f98a2SYan Zheng WARN_ON(trans->transid != btrfs_header_generation(parent)); 607d09c5152SFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, parent_slot, 60833cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 609d09c5152SFilipe Manana if (ret) { 610d09c5152SFilipe Manana btrfs_tree_unlock(cow); 611d09c5152SFilipe Manana free_extent_buffer(cow); 612d09c5152SFilipe Manana btrfs_abort_transaction(trans, ret); 613d09c5152SFilipe Manana return ret; 614d09c5152SFilipe Manana } 6155f39d397SChris Mason btrfs_set_node_blockptr(parent, parent_slot, 616db94535dSChris Mason cow->start); 61774493f7aSChris Mason btrfs_set_node_ptr_generation(parent, parent_slot, 61874493f7aSChris Mason trans->transid); 6196702ed49SChris Mason btrfs_mark_buffer_dirty(parent); 6205de865eeSFilipe David Borba Manana if (last_ref) { 621f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_free_eb(buf); 6225de865eeSFilipe David Borba Manana if (ret) { 623572c83acSJosef Bacik btrfs_tree_unlock(cow); 624572c83acSJosef Bacik free_extent_buffer(cow); 62566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 6265de865eeSFilipe David Borba Manana return ret; 6275de865eeSFilipe David Borba Manana } 6285de865eeSFilipe David Borba Manana } 6297a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), buf, 6307a163608SFilipe Manana parent_start, last_ref); 6316702ed49SChris Mason } 632925baeddSChris Mason if (unlock_orig) 633925baeddSChris Mason btrfs_tree_unlock(buf); 6343083ee2eSJosef Bacik free_extent_buffer_stale(buf); 6356702ed49SChris Mason btrfs_mark_buffer_dirty(cow); 6366702ed49SChris Mason *cow_ret = cow; 6376702ed49SChris Mason return 0; 6386702ed49SChris Mason } 6396702ed49SChris Mason 6405d4f98a2SYan Zheng static inline int should_cow_block(struct btrfs_trans_handle *trans, 6415d4f98a2SYan Zheng struct btrfs_root *root, 6425d4f98a2SYan Zheng struct extent_buffer *buf) 6435d4f98a2SYan Zheng { 644f5ee5c9aSJeff Mahoney if (btrfs_is_testing(root->fs_info)) 645faa2dbf0SJosef Bacik return 0; 646fccb84c9SDavid Sterba 647d1980131SDavid Sterba /* Ensure we can see the FORCE_COW bit */ 648d1980131SDavid Sterba smp_mb__before_atomic(); 649f1ebcc74SLiu Bo 650f1ebcc74SLiu Bo /* 651f1ebcc74SLiu Bo * We do not need to cow a block if 652f1ebcc74SLiu Bo * 1) this block is not created or changed in this transaction; 653f1ebcc74SLiu Bo * 2) this block does not belong to TREE_RELOC tree; 654f1ebcc74SLiu Bo * 3) the root is not forced COW. 655f1ebcc74SLiu Bo * 656f1ebcc74SLiu Bo * What is forced COW: 65701327610SNicholas D Steeves * when we create snapshot during committing the transaction, 65852042d8eSAndrea Gelmini * after we've finished copying src root, we must COW the shared 659f1ebcc74SLiu Bo * block to ensure the metadata consistency. 660f1ebcc74SLiu Bo */ 6615d4f98a2SYan Zheng if (btrfs_header_generation(buf) == trans->transid && 6625d4f98a2SYan Zheng !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && 6635d4f98a2SYan Zheng !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 664f1ebcc74SLiu Bo btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && 66527cdeb70SMiao Xie !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) 6665d4f98a2SYan Zheng return 0; 6675d4f98a2SYan Zheng return 1; 6685d4f98a2SYan Zheng } 6695d4f98a2SYan Zheng 670d352ac68SChris Mason /* 671d352ac68SChris Mason * cows a single block, see __btrfs_cow_block for the real work. 67201327610SNicholas D Steeves * This version of it has extra checks so that a block isn't COWed more than 673d352ac68SChris Mason * once per transaction, as long as it hasn't been written yet 674d352ac68SChris Mason */ 675d397712bSChris Mason noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, 6765f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *buf, 6775f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 6789631e4ccSJosef Bacik struct extent_buffer **cow_ret, 6799631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 68002217ed2SChris Mason { 6810b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 6826702ed49SChris Mason u64 search_start; 683f510cfecSChris Mason int ret; 684dc17ff8fSChris Mason 68583354f07SJosef Bacik if (test_bit(BTRFS_ROOT_DELETING, &root->state)) 68683354f07SJosef Bacik btrfs_err(fs_info, 68783354f07SJosef Bacik "COW'ing blocks on a fs root that's being dropped"); 68883354f07SJosef Bacik 689*48774f3bSFilipe Manana /* 690*48774f3bSFilipe Manana * COWing must happen through a running transaction, which always 691*48774f3bSFilipe Manana * matches the current fs generation (it's a transaction with a state 692*48774f3bSFilipe Manana * less than TRANS_STATE_UNBLOCKED). If it doesn't, then turn the fs 693*48774f3bSFilipe Manana * into error state to prevent the commit of any transaction. 694*48774f3bSFilipe Manana */ 695*48774f3bSFilipe Manana if (unlikely(trans->transaction != fs_info->running_transaction || 696*48774f3bSFilipe Manana trans->transid != fs_info->generation)) { 697*48774f3bSFilipe Manana btrfs_abort_transaction(trans, -EUCLEAN); 698*48774f3bSFilipe Manana btrfs_crit(fs_info, 699*48774f3bSFilipe Manana "unexpected transaction when attempting to COW block %llu on root %llu, transaction %llu running transaction %llu fs generation %llu", 700*48774f3bSFilipe Manana buf->start, btrfs_root_id(root), trans->transid, 701*48774f3bSFilipe Manana fs_info->running_transaction->transid, 702*48774f3bSFilipe Manana fs_info->generation); 703*48774f3bSFilipe Manana return -EUCLEAN; 704*48774f3bSFilipe Manana } 705dc17ff8fSChris Mason 7065d4f98a2SYan Zheng if (!should_cow_block(trans, root, buf)) { 70702217ed2SChris Mason *cow_ret = buf; 70802217ed2SChris Mason return 0; 70902217ed2SChris Mason } 710c487685dSChris Mason 711ee22184bSByongho Lee search_start = buf->start & ~((u64)SZ_1G - 1); 712b4ce94deSChris Mason 713f616f5cdSQu Wenruo /* 714f616f5cdSQu Wenruo * Before CoWing this block for later modification, check if it's 715f616f5cdSQu Wenruo * the subtree root and do the delayed subtree trace if needed. 716f616f5cdSQu Wenruo * 717f616f5cdSQu Wenruo * Also We don't care about the error, as it's handled internally. 718f616f5cdSQu Wenruo */ 719f616f5cdSQu Wenruo btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); 720f510cfecSChris Mason ret = __btrfs_cow_block(trans, root, buf, parent, 7219631e4ccSJosef Bacik parent_slot, cow_ret, search_start, 0, nest); 7221abe9b8aSliubo 7231abe9b8aSliubo trace_btrfs_cow_block(root, buf, *cow_ret); 7241abe9b8aSliubo 725f510cfecSChris Mason return ret; 7262c90e5d6SChris Mason } 727f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); 7286702ed49SChris Mason 729d352ac68SChris Mason /* 730d352ac68SChris Mason * helper function for defrag to decide if two blocks pointed to by a 731d352ac68SChris Mason * node are actually close by 732d352ac68SChris Mason */ 7336b80053dSChris Mason static int close_blocks(u64 blocknr, u64 other, u32 blocksize) 7346702ed49SChris Mason { 7356b80053dSChris Mason if (blocknr < other && other - (blocknr + blocksize) < 32768) 7366702ed49SChris Mason return 1; 7376b80053dSChris Mason if (blocknr > other && blocknr - (other + blocksize) < 32768) 7386702ed49SChris Mason return 1; 73902217ed2SChris Mason return 0; 74002217ed2SChris Mason } 74102217ed2SChris Mason 742ce6ef5abSDavid Sterba #ifdef __LITTLE_ENDIAN 743ce6ef5abSDavid Sterba 744ce6ef5abSDavid Sterba /* 745ce6ef5abSDavid Sterba * Compare two keys, on little-endian the disk order is same as CPU order and 746ce6ef5abSDavid Sterba * we can avoid the conversion. 747ce6ef5abSDavid Sterba */ 748ce6ef5abSDavid Sterba static int comp_keys(const struct btrfs_disk_key *disk_key, 749ce6ef5abSDavid Sterba const struct btrfs_key *k2) 750ce6ef5abSDavid Sterba { 751ce6ef5abSDavid Sterba const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; 752ce6ef5abSDavid Sterba 753ce6ef5abSDavid Sterba return btrfs_comp_cpu_keys(k1, k2); 754ce6ef5abSDavid Sterba } 755ce6ef5abSDavid Sterba 756ce6ef5abSDavid Sterba #else 757ce6ef5abSDavid Sterba 758081e9573SChris Mason /* 759081e9573SChris Mason * compare two keys in a memcmp fashion 760081e9573SChris Mason */ 761310712b2SOmar Sandoval static int comp_keys(const struct btrfs_disk_key *disk, 762310712b2SOmar Sandoval const struct btrfs_key *k2) 763081e9573SChris Mason { 764081e9573SChris Mason struct btrfs_key k1; 765081e9573SChris Mason 766081e9573SChris Mason btrfs_disk_key_to_cpu(&k1, disk); 767081e9573SChris Mason 76820736abaSDiego Calleja return btrfs_comp_cpu_keys(&k1, k2); 769081e9573SChris Mason } 770ce6ef5abSDavid Sterba #endif 771081e9573SChris Mason 772f3465ca4SJosef Bacik /* 773f3465ca4SJosef Bacik * same as comp_keys only with two btrfs_key's 774f3465ca4SJosef Bacik */ 775e1f60a65SDavid Sterba int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) 776f3465ca4SJosef Bacik { 777f3465ca4SJosef Bacik if (k1->objectid > k2->objectid) 778f3465ca4SJosef Bacik return 1; 779f3465ca4SJosef Bacik if (k1->objectid < k2->objectid) 780f3465ca4SJosef Bacik return -1; 781f3465ca4SJosef Bacik if (k1->type > k2->type) 782f3465ca4SJosef Bacik return 1; 783f3465ca4SJosef Bacik if (k1->type < k2->type) 784f3465ca4SJosef Bacik return -1; 785f3465ca4SJosef Bacik if (k1->offset > k2->offset) 786f3465ca4SJosef Bacik return 1; 787f3465ca4SJosef Bacik if (k1->offset < k2->offset) 788f3465ca4SJosef Bacik return -1; 789f3465ca4SJosef Bacik return 0; 790f3465ca4SJosef Bacik } 791081e9573SChris Mason 792d352ac68SChris Mason /* 793d352ac68SChris Mason * this is used by the defrag code to go through all the 794d352ac68SChris Mason * leaves pointed to by a node and reallocate them so that 795d352ac68SChris Mason * disk order is close to key order 796d352ac68SChris Mason */ 7976702ed49SChris Mason int btrfs_realloc_node(struct btrfs_trans_handle *trans, 7985f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *parent, 799de78b51aSEric Sandeen int start_slot, u64 *last_ret, 800a6b6e75eSChris Mason struct btrfs_key *progress) 8016702ed49SChris Mason { 8020b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8036b80053dSChris Mason struct extent_buffer *cur; 8046702ed49SChris Mason u64 blocknr; 805e9d0b13bSChris Mason u64 search_start = *last_ret; 806e9d0b13bSChris Mason u64 last_block = 0; 8076702ed49SChris Mason u64 other; 8086702ed49SChris Mason u32 parent_nritems; 8096702ed49SChris Mason int end_slot; 8106702ed49SChris Mason int i; 8116702ed49SChris Mason int err = 0; 8126b80053dSChris Mason u32 blocksize; 813081e9573SChris Mason int progress_passed = 0; 814081e9573SChris Mason struct btrfs_disk_key disk_key; 8156702ed49SChris Mason 8160b246afaSJeff Mahoney WARN_ON(trans->transaction != fs_info->running_transaction); 8170b246afaSJeff Mahoney WARN_ON(trans->transid != fs_info->generation); 81886479a04SChris Mason 8196b80053dSChris Mason parent_nritems = btrfs_header_nritems(parent); 8200b246afaSJeff Mahoney blocksize = fs_info->nodesize; 8215dfe2be7SFilipe Manana end_slot = parent_nritems - 1; 8226702ed49SChris Mason 8235dfe2be7SFilipe Manana if (parent_nritems <= 1) 8246702ed49SChris Mason return 0; 8256702ed49SChris Mason 8265dfe2be7SFilipe Manana for (i = start_slot; i <= end_slot; i++) { 8276702ed49SChris Mason int close = 1; 828a6b6e75eSChris Mason 829081e9573SChris Mason btrfs_node_key(parent, &disk_key, i); 830081e9573SChris Mason if (!progress_passed && comp_keys(&disk_key, progress) < 0) 831081e9573SChris Mason continue; 832081e9573SChris Mason 833081e9573SChris Mason progress_passed = 1; 8346b80053dSChris Mason blocknr = btrfs_node_blockptr(parent, i); 835e9d0b13bSChris Mason if (last_block == 0) 836e9d0b13bSChris Mason last_block = blocknr; 8375708b959SChris Mason 8386702ed49SChris Mason if (i > 0) { 8396b80053dSChris Mason other = btrfs_node_blockptr(parent, i - 1); 8406b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 8416702ed49SChris Mason } 8425dfe2be7SFilipe Manana if (!close && i < end_slot) { 8436b80053dSChris Mason other = btrfs_node_blockptr(parent, i + 1); 8446b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 8456702ed49SChris Mason } 846e9d0b13bSChris Mason if (close) { 847e9d0b13bSChris Mason last_block = blocknr; 8486702ed49SChris Mason continue; 849e9d0b13bSChris Mason } 8506702ed49SChris Mason 851206983b7SJosef Bacik cur = btrfs_read_node_slot(parent, i); 852206983b7SJosef Bacik if (IS_ERR(cur)) 85364c043deSLiu Bo return PTR_ERR(cur); 854e9d0b13bSChris Mason if (search_start == 0) 8556b80053dSChris Mason search_start = last_block; 856e9d0b13bSChris Mason 857e7a84565SChris Mason btrfs_tree_lock(cur); 8586b80053dSChris Mason err = __btrfs_cow_block(trans, root, cur, parent, i, 859e7a84565SChris Mason &cur, search_start, 8606b80053dSChris Mason min(16 * blocksize, 8619631e4ccSJosef Bacik (end_slot - i) * blocksize), 8629631e4ccSJosef Bacik BTRFS_NESTING_COW); 863252c38f0SYan if (err) { 864e7a84565SChris Mason btrfs_tree_unlock(cur); 8656b80053dSChris Mason free_extent_buffer(cur); 8666702ed49SChris Mason break; 867252c38f0SYan } 868e7a84565SChris Mason search_start = cur->start; 869e7a84565SChris Mason last_block = cur->start; 870f2183bdeSChris Mason *last_ret = search_start; 871e7a84565SChris Mason btrfs_tree_unlock(cur); 872e7a84565SChris Mason free_extent_buffer(cur); 8736702ed49SChris Mason } 8746702ed49SChris Mason return err; 8756702ed49SChris Mason } 8766702ed49SChris Mason 87774123bd7SChris Mason /* 878fb81212cSFilipe Manana * Search for a key in the given extent_buffer. 8795f39d397SChris Mason * 880a724f313SFilipe Manana * The lower boundary for the search is specified by the slot number @first_slot. 881fdf8d595SAnand Jain * Use a value of 0 to search over the whole extent buffer. Works for both 882fdf8d595SAnand Jain * leaves and nodes. 88374123bd7SChris Mason * 884fb81212cSFilipe Manana * The slot in the extent buffer is returned via @slot. If the key exists in the 885fb81212cSFilipe Manana * extent buffer, then @slot will point to the slot where the key is, otherwise 886fb81212cSFilipe Manana * it points to the slot where you would insert the key. 887fb81212cSFilipe Manana * 888fb81212cSFilipe Manana * Slot may point to the total number of items (i.e. one position beyond the last 889fb81212cSFilipe Manana * key) if the key is bigger than the last key in the extent buffer. 89074123bd7SChris Mason */ 891fdf8d595SAnand Jain int btrfs_bin_search(struct extent_buffer *eb, int first_slot, 89267d5e289SMarcos Paulo de Souza const struct btrfs_key *key, int *slot) 893be0e5c09SChris Mason { 894fb81212cSFilipe Manana unsigned long p; 895fb81212cSFilipe Manana int item_size; 896a724f313SFilipe Manana /* 897a724f313SFilipe Manana * Use unsigned types for the low and high slots, so that we get a more 898a724f313SFilipe Manana * efficient division in the search loop below. 899a724f313SFilipe Manana */ 900a724f313SFilipe Manana u32 low = first_slot; 901a724f313SFilipe Manana u32 high = btrfs_header_nritems(eb); 902be0e5c09SChris Mason int ret; 9035cd17f34SDavid Sterba const int key_size = sizeof(struct btrfs_disk_key); 904be0e5c09SChris Mason 905a724f313SFilipe Manana if (unlikely(low > high)) { 9065e24e9afSLiu Bo btrfs_err(eb->fs_info, 907a724f313SFilipe Manana "%s: low (%u) > high (%u) eb %llu owner %llu level %d", 9085e24e9afSLiu Bo __func__, low, high, eb->start, 9095e24e9afSLiu Bo btrfs_header_owner(eb), btrfs_header_level(eb)); 9105e24e9afSLiu Bo return -EINVAL; 9115e24e9afSLiu Bo } 9125e24e9afSLiu Bo 913fb81212cSFilipe Manana if (btrfs_header_level(eb) == 0) { 914fb81212cSFilipe Manana p = offsetof(struct btrfs_leaf, items); 915fb81212cSFilipe Manana item_size = sizeof(struct btrfs_item); 916fb81212cSFilipe Manana } else { 917fb81212cSFilipe Manana p = offsetof(struct btrfs_node, ptrs); 918fb81212cSFilipe Manana item_size = sizeof(struct btrfs_key_ptr); 919fb81212cSFilipe Manana } 920fb81212cSFilipe Manana 921be0e5c09SChris Mason while (low < high) { 9225cd17f34SDavid Sterba unsigned long oip; 9235cd17f34SDavid Sterba unsigned long offset; 9245cd17f34SDavid Sterba struct btrfs_disk_key *tmp; 9255cd17f34SDavid Sterba struct btrfs_disk_key unaligned; 9265cd17f34SDavid Sterba int mid; 9275cd17f34SDavid Sterba 928be0e5c09SChris Mason mid = (low + high) / 2; 9295f39d397SChris Mason offset = p + mid * item_size; 9305cd17f34SDavid Sterba oip = offset_in_page(offset); 9315f39d397SChris Mason 9325cd17f34SDavid Sterba if (oip + key_size <= PAGE_SIZE) { 933884b07d0SQu Wenruo const unsigned long idx = get_eb_page_index(offset); 9345cd17f34SDavid Sterba char *kaddr = page_address(eb->pages[idx]); 935934d375bSChris Mason 936884b07d0SQu Wenruo oip = get_eb_offset_in_page(eb, offset); 9375cd17f34SDavid Sterba tmp = (struct btrfs_disk_key *)(kaddr + oip); 9385cd17f34SDavid Sterba } else { 9395cd17f34SDavid Sterba read_extent_buffer(eb, &unaligned, offset, key_size); 9405f39d397SChris Mason tmp = &unaligned; 941479965d6SChris Mason } 942479965d6SChris Mason 943be0e5c09SChris Mason ret = comp_keys(tmp, key); 944be0e5c09SChris Mason 945be0e5c09SChris Mason if (ret < 0) 946be0e5c09SChris Mason low = mid + 1; 947be0e5c09SChris Mason else if (ret > 0) 948be0e5c09SChris Mason high = mid; 949be0e5c09SChris Mason else { 950be0e5c09SChris Mason *slot = mid; 951be0e5c09SChris Mason return 0; 952be0e5c09SChris Mason } 953be0e5c09SChris Mason } 954be0e5c09SChris Mason *slot = low; 955be0e5c09SChris Mason return 1; 956be0e5c09SChris Mason } 957be0e5c09SChris Mason 958f0486c68SYan, Zheng static void root_add_used(struct btrfs_root *root, u32 size) 959f0486c68SYan, Zheng { 960f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 961f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 962f0486c68SYan, Zheng btrfs_root_used(&root->root_item) + size); 963f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 964f0486c68SYan, Zheng } 965f0486c68SYan, Zheng 966f0486c68SYan, Zheng static void root_sub_used(struct btrfs_root *root, u32 size) 967f0486c68SYan, Zheng { 968f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 969f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 970f0486c68SYan, Zheng btrfs_root_used(&root->root_item) - size); 971f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 972f0486c68SYan, Zheng } 973f0486c68SYan, Zheng 974d352ac68SChris Mason /* given a node and slot number, this reads the blocks it points to. The 975d352ac68SChris Mason * extent buffer is returned with a reference taken (but unlocked). 976d352ac68SChris Mason */ 9774b231ae4SDavid Sterba struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, 9784b231ae4SDavid Sterba int slot) 979bb803951SChris Mason { 980ca7a79adSChris Mason int level = btrfs_header_level(parent); 981789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 0 }; 982416bc658SJosef Bacik struct extent_buffer *eb; 983416bc658SJosef Bacik 984fb770ae4SLiu Bo if (slot < 0 || slot >= btrfs_header_nritems(parent)) 985fb770ae4SLiu Bo return ERR_PTR(-ENOENT); 986ca7a79adSChris Mason 987d4694728SJosef Bacik ASSERT(level); 988ca7a79adSChris Mason 989789d6a3aSQu Wenruo check.level = level - 1; 990789d6a3aSQu Wenruo check.transid = btrfs_node_ptr_generation(parent, slot); 991789d6a3aSQu Wenruo check.owner_root = btrfs_header_owner(parent); 992789d6a3aSQu Wenruo check.has_first_key = true; 993789d6a3aSQu Wenruo btrfs_node_key_to_cpu(parent, &check.first_key, slot); 994789d6a3aSQu Wenruo 995d0d20b0fSDavid Sterba eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), 996789d6a3aSQu Wenruo &check); 9974eb150d6SQu Wenruo if (IS_ERR(eb)) 9984eb150d6SQu Wenruo return eb; 9994eb150d6SQu Wenruo if (!extent_buffer_uptodate(eb)) { 1000416bc658SJosef Bacik free_extent_buffer(eb); 10014eb150d6SQu Wenruo return ERR_PTR(-EIO); 1002416bc658SJosef Bacik } 1003416bc658SJosef Bacik 1004416bc658SJosef Bacik return eb; 1005bb803951SChris Mason } 1006bb803951SChris Mason 1007d352ac68SChris Mason /* 1008d352ac68SChris Mason * node level balancing, used to make sure nodes are in proper order for 1009d352ac68SChris Mason * item deletion. We balance from the top down, so we have to make sure 1010d352ac68SChris Mason * that a deletion won't leave an node completely empty later on. 1011d352ac68SChris Mason */ 1012e02119d5SChris Mason static noinline int balance_level(struct btrfs_trans_handle *trans, 101398ed5174SChris Mason struct btrfs_root *root, 101498ed5174SChris Mason struct btrfs_path *path, int level) 1015bb803951SChris Mason { 10160b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 10175f39d397SChris Mason struct extent_buffer *right = NULL; 10185f39d397SChris Mason struct extent_buffer *mid; 10195f39d397SChris Mason struct extent_buffer *left = NULL; 10205f39d397SChris Mason struct extent_buffer *parent = NULL; 1021bb803951SChris Mason int ret = 0; 1022bb803951SChris Mason int wret; 1023bb803951SChris Mason int pslot; 1024bb803951SChris Mason int orig_slot = path->slots[level]; 102579f95c82SChris Mason u64 orig_ptr; 1026bb803951SChris Mason 102798e6b1ebSLiu Bo ASSERT(level > 0); 1028bb803951SChris Mason 10295f39d397SChris Mason mid = path->nodes[level]; 1030b4ce94deSChris Mason 1031ac5887c8SJosef Bacik WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); 10327bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 10337bb86316SChris Mason 10341d4f8a0cSChris Mason orig_ptr = btrfs_node_blockptr(mid, orig_slot); 103579f95c82SChris Mason 1036a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 10375f39d397SChris Mason parent = path->nodes[level + 1]; 1038bb803951SChris Mason pslot = path->slots[level + 1]; 1039a05a9bb1SLi Zefan } 1040bb803951SChris Mason 104140689478SChris Mason /* 104240689478SChris Mason * deal with the case where there is only one pointer in the root 104340689478SChris Mason * by promoting the node below to a root 104440689478SChris Mason */ 10455f39d397SChris Mason if (!parent) { 10465f39d397SChris Mason struct extent_buffer *child; 1047bb803951SChris Mason 10485f39d397SChris Mason if (btrfs_header_nritems(mid) != 1) 1049bb803951SChris Mason return 0; 1050bb803951SChris Mason 1051bb803951SChris Mason /* promote the child to a root */ 10524b231ae4SDavid Sterba child = btrfs_read_node_slot(mid, 0); 1053fb770ae4SLiu Bo if (IS_ERR(child)) { 1054fb770ae4SLiu Bo ret = PTR_ERR(child); 1055daefe4d4SFilipe Manana goto out; 1056305a26afSMark Fasheh } 1057305a26afSMark Fasheh 1058925baeddSChris Mason btrfs_tree_lock(child); 10599631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 10609631e4ccSJosef Bacik BTRFS_NESTING_COW); 1061f0486c68SYan, Zheng if (ret) { 1062f0486c68SYan, Zheng btrfs_tree_unlock(child); 1063f0486c68SYan, Zheng free_extent_buffer(child); 1064daefe4d4SFilipe Manana goto out; 1065f0486c68SYan, Zheng } 10662f375ab9SYan 1067406808abSFilipe Manana ret = btrfs_tree_mod_log_insert_root(root->node, child, true); 106839020d8aSFilipe Manana if (ret < 0) { 106939020d8aSFilipe Manana btrfs_tree_unlock(child); 107039020d8aSFilipe Manana free_extent_buffer(child); 107139020d8aSFilipe Manana btrfs_abort_transaction(trans, ret); 1072daefe4d4SFilipe Manana goto out; 107339020d8aSFilipe Manana } 1074240f62c8SChris Mason rcu_assign_pointer(root->node, child); 1075925baeddSChris Mason 10760b86a832SChris Mason add_root_to_dirty_list(root); 1077925baeddSChris Mason btrfs_tree_unlock(child); 1078b4ce94deSChris Mason 1079925baeddSChris Mason path->locks[level] = 0; 1080bb803951SChris Mason path->nodes[level] = NULL; 1081190a8339SJosef Bacik btrfs_clear_buffer_dirty(trans, mid); 1082925baeddSChris Mason btrfs_tree_unlock(mid); 1083bb803951SChris Mason /* once for the path */ 10845f39d397SChris Mason free_extent_buffer(mid); 1085f0486c68SYan, Zheng 1086f0486c68SYan, Zheng root_sub_used(root, mid->len); 10877a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); 1088bb803951SChris Mason /* once for the root ptr */ 10893083ee2eSJosef Bacik free_extent_buffer_stale(mid); 1090f0486c68SYan, Zheng return 0; 1091bb803951SChris Mason } 10925f39d397SChris Mason if (btrfs_header_nritems(mid) > 10930b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) 1094bb803951SChris Mason return 0; 1095bb803951SChris Mason 10969cf14029SJosef Bacik if (pslot) { 10974b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 10989cf14029SJosef Bacik if (IS_ERR(left)) { 10999cf14029SJosef Bacik ret = PTR_ERR(left); 1100fb770ae4SLiu Bo left = NULL; 1101daefe4d4SFilipe Manana goto out; 11029cf14029SJosef Bacik } 1103fb770ae4SLiu Bo 1104bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 11055f39d397SChris Mason wret = btrfs_cow_block(trans, root, left, 11069631e4ccSJosef Bacik parent, pslot - 1, &left, 1107bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 110854aa1f4dSChris Mason if (wret) { 110954aa1f4dSChris Mason ret = wret; 1110daefe4d4SFilipe Manana goto out; 111154aa1f4dSChris Mason } 11122cc58cf2SChris Mason } 1113fb770ae4SLiu Bo 11149cf14029SJosef Bacik if (pslot + 1 < btrfs_header_nritems(parent)) { 11154b231ae4SDavid Sterba right = btrfs_read_node_slot(parent, pslot + 1); 11169cf14029SJosef Bacik if (IS_ERR(right)) { 11179cf14029SJosef Bacik ret = PTR_ERR(right); 1118fb770ae4SLiu Bo right = NULL; 1119daefe4d4SFilipe Manana goto out; 11209cf14029SJosef Bacik } 1121fb770ae4SLiu Bo 1122bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 11235f39d397SChris Mason wret = btrfs_cow_block(trans, root, right, 11249631e4ccSJosef Bacik parent, pslot + 1, &right, 1125bf59a5a2SJosef Bacik BTRFS_NESTING_RIGHT_COW); 11262cc58cf2SChris Mason if (wret) { 11272cc58cf2SChris Mason ret = wret; 1128daefe4d4SFilipe Manana goto out; 11292cc58cf2SChris Mason } 11302cc58cf2SChris Mason } 11312cc58cf2SChris Mason 11322cc58cf2SChris Mason /* first, try to make some room in the middle buffer */ 11335f39d397SChris Mason if (left) { 11345f39d397SChris Mason orig_slot += btrfs_header_nritems(left); 1135d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 113679f95c82SChris Mason if (wret < 0) 113779f95c82SChris Mason ret = wret; 1138bb803951SChris Mason } 113979f95c82SChris Mason 114079f95c82SChris Mason /* 114179f95c82SChris Mason * then try to empty the right most buffer into the middle 114279f95c82SChris Mason */ 11435f39d397SChris Mason if (right) { 1144d30a668fSDavid Sterba wret = push_node_left(trans, mid, right, 1); 114554aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 114679f95c82SChris Mason ret = wret; 11475f39d397SChris Mason if (btrfs_header_nritems(right) == 0) { 1148190a8339SJosef Bacik btrfs_clear_buffer_dirty(trans, right); 1149925baeddSChris Mason btrfs_tree_unlock(right); 1150751a2761SFilipe Manana ret = btrfs_del_ptr(trans, root, path, level + 1, pslot + 1); 1151751a2761SFilipe Manana if (ret < 0) { 1152751a2761SFilipe Manana free_extent_buffer_stale(right); 1153751a2761SFilipe Manana right = NULL; 1154751a2761SFilipe Manana goto out; 1155751a2761SFilipe Manana } 1156f0486c68SYan, Zheng root_sub_used(root, right->len); 11577a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), right, 11587a163608SFilipe Manana 0, 1); 11593083ee2eSJosef Bacik free_extent_buffer_stale(right); 1160f0486c68SYan, Zheng right = NULL; 1161bb803951SChris Mason } else { 11625f39d397SChris Mason struct btrfs_disk_key right_key; 11635f39d397SChris Mason btrfs_node_key(right, &right_key, 0); 1164f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, 116533cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 116639020d8aSFilipe Manana if (ret < 0) { 116739020d8aSFilipe Manana btrfs_abort_transaction(trans, ret); 1168daefe4d4SFilipe Manana goto out; 116939020d8aSFilipe Manana } 11705f39d397SChris Mason btrfs_set_node_key(parent, &right_key, pslot + 1); 11715f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 1172bb803951SChris Mason } 1173bb803951SChris Mason } 11745f39d397SChris Mason if (btrfs_header_nritems(mid) == 1) { 117579f95c82SChris Mason /* 117679f95c82SChris Mason * we're not allowed to leave a node with one item in the 117779f95c82SChris Mason * tree during a delete. A deletion from lower in the tree 117879f95c82SChris Mason * could try to delete the only pointer in this node. 117979f95c82SChris Mason * So, pull some keys from the left. 118079f95c82SChris Mason * There has to be a left pointer at this point because 118179f95c82SChris Mason * otherwise we would have pulled some pointers from the 118279f95c82SChris Mason * right 118379f95c82SChris Mason */ 1184725026edSFilipe Manana if (unlikely(!left)) { 1185725026edSFilipe Manana btrfs_crit(fs_info, 1186725026edSFilipe Manana "missing left child when middle child only has 1 item, parent bytenr %llu level %d mid bytenr %llu root %llu", 1187725026edSFilipe Manana parent->start, btrfs_header_level(parent), 1188725026edSFilipe Manana mid->start, btrfs_root_id(root)); 1189725026edSFilipe Manana ret = -EUCLEAN; 1190725026edSFilipe Manana btrfs_abort_transaction(trans, ret); 1191daefe4d4SFilipe Manana goto out; 1192305a26afSMark Fasheh } 119355d32ed8SDavid Sterba wret = balance_node_right(trans, mid, left); 119454aa1f4dSChris Mason if (wret < 0) { 119579f95c82SChris Mason ret = wret; 1196daefe4d4SFilipe Manana goto out; 119754aa1f4dSChris Mason } 1198bce4eae9SChris Mason if (wret == 1) { 1199d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 1200bce4eae9SChris Mason if (wret < 0) 1201bce4eae9SChris Mason ret = wret; 1202bce4eae9SChris Mason } 120379f95c82SChris Mason BUG_ON(wret == 1); 120479f95c82SChris Mason } 12055f39d397SChris Mason if (btrfs_header_nritems(mid) == 0) { 1206190a8339SJosef Bacik btrfs_clear_buffer_dirty(trans, mid); 1207925baeddSChris Mason btrfs_tree_unlock(mid); 1208751a2761SFilipe Manana ret = btrfs_del_ptr(trans, root, path, level + 1, pslot); 1209751a2761SFilipe Manana if (ret < 0) { 1210751a2761SFilipe Manana free_extent_buffer_stale(mid); 1211751a2761SFilipe Manana mid = NULL; 1212751a2761SFilipe Manana goto out; 1213751a2761SFilipe Manana } 1214f0486c68SYan, Zheng root_sub_used(root, mid->len); 12157a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); 12163083ee2eSJosef Bacik free_extent_buffer_stale(mid); 1217f0486c68SYan, Zheng mid = NULL; 121879f95c82SChris Mason } else { 121979f95c82SChris Mason /* update the parent key to reflect our changes */ 12205f39d397SChris Mason struct btrfs_disk_key mid_key; 12215f39d397SChris Mason btrfs_node_key(mid, &mid_key, 0); 1222f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot, 122333cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 122439020d8aSFilipe Manana if (ret < 0) { 122539020d8aSFilipe Manana btrfs_abort_transaction(trans, ret); 1226daefe4d4SFilipe Manana goto out; 122739020d8aSFilipe Manana } 12285f39d397SChris Mason btrfs_set_node_key(parent, &mid_key, pslot); 12295f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 123079f95c82SChris Mason } 1231bb803951SChris Mason 123279f95c82SChris Mason /* update the path */ 12335f39d397SChris Mason if (left) { 12345f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 123567439dadSDavid Sterba atomic_inc(&left->refs); 1236925baeddSChris Mason /* left was locked after cow */ 12375f39d397SChris Mason path->nodes[level] = left; 1238bb803951SChris Mason path->slots[level + 1] -= 1; 1239bb803951SChris Mason path->slots[level] = orig_slot; 1240925baeddSChris Mason if (mid) { 1241925baeddSChris Mason btrfs_tree_unlock(mid); 12425f39d397SChris Mason free_extent_buffer(mid); 1243925baeddSChris Mason } 1244bb803951SChris Mason } else { 12455f39d397SChris Mason orig_slot -= btrfs_header_nritems(left); 1246bb803951SChris Mason path->slots[level] = orig_slot; 1247bb803951SChris Mason } 1248bb803951SChris Mason } 124979f95c82SChris Mason /* double check we haven't messed things up */ 1250e20d96d6SChris Mason if (orig_ptr != 12515f39d397SChris Mason btrfs_node_blockptr(path->nodes[level], path->slots[level])) 125279f95c82SChris Mason BUG(); 1253daefe4d4SFilipe Manana out: 1254925baeddSChris Mason if (right) { 1255925baeddSChris Mason btrfs_tree_unlock(right); 12565f39d397SChris Mason free_extent_buffer(right); 1257925baeddSChris Mason } 1258925baeddSChris Mason if (left) { 1259925baeddSChris Mason if (path->nodes[level] != left) 1260925baeddSChris Mason btrfs_tree_unlock(left); 12615f39d397SChris Mason free_extent_buffer(left); 1262925baeddSChris Mason } 1263bb803951SChris Mason return ret; 1264bb803951SChris Mason } 1265bb803951SChris Mason 1266d352ac68SChris Mason /* Node balancing for insertion. Here we only split or push nodes around 1267d352ac68SChris Mason * when they are completely full. This is also done top down, so we 1268d352ac68SChris Mason * have to be pessimistic. 1269d352ac68SChris Mason */ 1270d397712bSChris Mason static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, 1271e66f709bSChris Mason struct btrfs_root *root, 1272e66f709bSChris Mason struct btrfs_path *path, int level) 1273e66f709bSChris Mason { 12740b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12755f39d397SChris Mason struct extent_buffer *right = NULL; 12765f39d397SChris Mason struct extent_buffer *mid; 12775f39d397SChris Mason struct extent_buffer *left = NULL; 12785f39d397SChris Mason struct extent_buffer *parent = NULL; 1279e66f709bSChris Mason int ret = 0; 1280e66f709bSChris Mason int wret; 1281e66f709bSChris Mason int pslot; 1282e66f709bSChris Mason int orig_slot = path->slots[level]; 1283e66f709bSChris Mason 1284e66f709bSChris Mason if (level == 0) 1285e66f709bSChris Mason return 1; 1286e66f709bSChris Mason 12875f39d397SChris Mason mid = path->nodes[level]; 12887bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 1289e66f709bSChris Mason 1290a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 12915f39d397SChris Mason parent = path->nodes[level + 1]; 1292e66f709bSChris Mason pslot = path->slots[level + 1]; 1293a05a9bb1SLi Zefan } 1294e66f709bSChris Mason 12955f39d397SChris Mason if (!parent) 1296e66f709bSChris Mason return 1; 1297e66f709bSChris Mason 12989cf14029SJosef Bacik /* first, try to make some room in the middle buffer */ 12999cf14029SJosef Bacik if (pslot) { 13009cf14029SJosef Bacik u32 left_nr; 13019cf14029SJosef Bacik 13024b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 1303fb770ae4SLiu Bo if (IS_ERR(left)) 13049cf14029SJosef Bacik return PTR_ERR(left); 1305925baeddSChris Mason 1306bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 1307b4ce94deSChris Mason 13085f39d397SChris Mason left_nr = btrfs_header_nritems(left); 13090b246afaSJeff Mahoney if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 131033ade1f8SChris Mason wret = 1; 131133ade1f8SChris Mason } else { 13125f39d397SChris Mason ret = btrfs_cow_block(trans, root, left, parent, 13139631e4ccSJosef Bacik pslot - 1, &left, 1314bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 131554aa1f4dSChris Mason if (ret) 131654aa1f4dSChris Mason wret = 1; 131754aa1f4dSChris Mason else { 1318d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 0); 131954aa1f4dSChris Mason } 132033ade1f8SChris Mason } 1321e66f709bSChris Mason if (wret < 0) 1322e66f709bSChris Mason ret = wret; 1323e66f709bSChris Mason if (wret == 0) { 13245f39d397SChris Mason struct btrfs_disk_key disk_key; 1325e66f709bSChris Mason orig_slot += left_nr; 13265f39d397SChris Mason btrfs_node_key(mid, &disk_key, 0); 1327f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot, 132833cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 132911d6ae03SFilipe Manana if (ret < 0) { 133011d6ae03SFilipe Manana btrfs_tree_unlock(left); 133111d6ae03SFilipe Manana free_extent_buffer(left); 133211d6ae03SFilipe Manana btrfs_abort_transaction(trans, ret); 133311d6ae03SFilipe Manana return ret; 133411d6ae03SFilipe Manana } 13355f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot); 13365f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 13375f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 13385f39d397SChris Mason path->nodes[level] = left; 1339e66f709bSChris Mason path->slots[level + 1] -= 1; 1340e66f709bSChris Mason path->slots[level] = orig_slot; 1341925baeddSChris Mason btrfs_tree_unlock(mid); 13425f39d397SChris Mason free_extent_buffer(mid); 1343e66f709bSChris Mason } else { 1344e66f709bSChris Mason orig_slot -= 13455f39d397SChris Mason btrfs_header_nritems(left); 1346e66f709bSChris Mason path->slots[level] = orig_slot; 1347925baeddSChris Mason btrfs_tree_unlock(left); 13485f39d397SChris Mason free_extent_buffer(left); 1349e66f709bSChris Mason } 1350e66f709bSChris Mason return 0; 1351e66f709bSChris Mason } 1352925baeddSChris Mason btrfs_tree_unlock(left); 13535f39d397SChris Mason free_extent_buffer(left); 1354e66f709bSChris Mason } 1355e66f709bSChris Mason 1356e66f709bSChris Mason /* 1357e66f709bSChris Mason * then try to empty the right most buffer into the middle 1358e66f709bSChris Mason */ 13599cf14029SJosef Bacik if (pslot + 1 < btrfs_header_nritems(parent)) { 136033ade1f8SChris Mason u32 right_nr; 1361b4ce94deSChris Mason 13629cf14029SJosef Bacik right = btrfs_read_node_slot(parent, pslot + 1); 13639cf14029SJosef Bacik if (IS_ERR(right)) 13649cf14029SJosef Bacik return PTR_ERR(right); 13659cf14029SJosef Bacik 1366bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 1367b4ce94deSChris Mason 13685f39d397SChris Mason right_nr = btrfs_header_nritems(right); 13690b246afaSJeff Mahoney if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 137033ade1f8SChris Mason wret = 1; 137133ade1f8SChris Mason } else { 13725f39d397SChris Mason ret = btrfs_cow_block(trans, root, right, 13735f39d397SChris Mason parent, pslot + 1, 1374bf59a5a2SJosef Bacik &right, BTRFS_NESTING_RIGHT_COW); 137554aa1f4dSChris Mason if (ret) 137654aa1f4dSChris Mason wret = 1; 137754aa1f4dSChris Mason else { 137855d32ed8SDavid Sterba wret = balance_node_right(trans, right, mid); 137933ade1f8SChris Mason } 138054aa1f4dSChris Mason } 1381e66f709bSChris Mason if (wret < 0) 1382e66f709bSChris Mason ret = wret; 1383e66f709bSChris Mason if (wret == 0) { 13845f39d397SChris Mason struct btrfs_disk_key disk_key; 13855f39d397SChris Mason 13865f39d397SChris Mason btrfs_node_key(right, &disk_key, 0); 1387f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, 138833cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 138911d6ae03SFilipe Manana if (ret < 0) { 139011d6ae03SFilipe Manana btrfs_tree_unlock(right); 139111d6ae03SFilipe Manana free_extent_buffer(right); 139211d6ae03SFilipe Manana btrfs_abort_transaction(trans, ret); 139311d6ae03SFilipe Manana return ret; 139411d6ae03SFilipe Manana } 13955f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot + 1); 13965f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 13975f39d397SChris Mason 13985f39d397SChris Mason if (btrfs_header_nritems(mid) <= orig_slot) { 13995f39d397SChris Mason path->nodes[level] = right; 1400e66f709bSChris Mason path->slots[level + 1] += 1; 1401e66f709bSChris Mason path->slots[level] = orig_slot - 14025f39d397SChris Mason btrfs_header_nritems(mid); 1403925baeddSChris Mason btrfs_tree_unlock(mid); 14045f39d397SChris Mason free_extent_buffer(mid); 1405e66f709bSChris Mason } else { 1406925baeddSChris Mason btrfs_tree_unlock(right); 14075f39d397SChris Mason free_extent_buffer(right); 1408e66f709bSChris Mason } 1409e66f709bSChris Mason return 0; 1410e66f709bSChris Mason } 1411925baeddSChris Mason btrfs_tree_unlock(right); 14125f39d397SChris Mason free_extent_buffer(right); 1413e66f709bSChris Mason } 1414e66f709bSChris Mason return 1; 1415e66f709bSChris Mason } 1416e66f709bSChris Mason 141774123bd7SChris Mason /* 1418d352ac68SChris Mason * readahead one full node of leaves, finding things that are close 1419d352ac68SChris Mason * to the block in 'slot', and triggering ra on them. 14203c69faecSChris Mason */ 14212ff7e61eSJeff Mahoney static void reada_for_search(struct btrfs_fs_info *fs_info, 1422e02119d5SChris Mason struct btrfs_path *path, 142301f46658SChris Mason int level, int slot, u64 objectid) 14243c69faecSChris Mason { 14255f39d397SChris Mason struct extent_buffer *node; 142601f46658SChris Mason struct btrfs_disk_key disk_key; 14273c69faecSChris Mason u32 nritems; 14283c69faecSChris Mason u64 search; 1429a7175319SChris Mason u64 target; 14306b80053dSChris Mason u64 nread = 0; 1431ace75066SFilipe Manana u64 nread_max; 14326b80053dSChris Mason u32 nr; 14336b80053dSChris Mason u32 blocksize; 14346b80053dSChris Mason u32 nscan = 0; 1435db94535dSChris Mason 1436ace75066SFilipe Manana if (level != 1 && path->reada != READA_FORWARD_ALWAYS) 14373c69faecSChris Mason return; 14383c69faecSChris Mason 14396702ed49SChris Mason if (!path->nodes[level]) 14406702ed49SChris Mason return; 14416702ed49SChris Mason 14425f39d397SChris Mason node = path->nodes[level]; 1443925baeddSChris Mason 1444ace75066SFilipe Manana /* 1445ace75066SFilipe Manana * Since the time between visiting leaves is much shorter than the time 1446ace75066SFilipe Manana * between visiting nodes, limit read ahead of nodes to 1, to avoid too 1447ace75066SFilipe Manana * much IO at once (possibly random). 1448ace75066SFilipe Manana */ 1449ace75066SFilipe Manana if (path->reada == READA_FORWARD_ALWAYS) { 1450ace75066SFilipe Manana if (level > 1) 1451ace75066SFilipe Manana nread_max = node->fs_info->nodesize; 1452ace75066SFilipe Manana else 1453ace75066SFilipe Manana nread_max = SZ_128K; 1454ace75066SFilipe Manana } else { 1455ace75066SFilipe Manana nread_max = SZ_64K; 1456ace75066SFilipe Manana } 1457ace75066SFilipe Manana 14583c69faecSChris Mason search = btrfs_node_blockptr(node, slot); 14590b246afaSJeff Mahoney blocksize = fs_info->nodesize; 1460069a2e37SFilipe Manana if (path->reada != READA_FORWARD_ALWAYS) { 1461069a2e37SFilipe Manana struct extent_buffer *eb; 1462069a2e37SFilipe Manana 14630b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, search); 14645f39d397SChris Mason if (eb) { 14655f39d397SChris Mason free_extent_buffer(eb); 14663c69faecSChris Mason return; 14673c69faecSChris Mason } 1468069a2e37SFilipe Manana } 14693c69faecSChris Mason 1470a7175319SChris Mason target = search; 14716b80053dSChris Mason 14725f39d397SChris Mason nritems = btrfs_header_nritems(node); 14736b80053dSChris Mason nr = slot; 147425b8b936SJosef Bacik 14753c69faecSChris Mason while (1) { 1476e4058b54SDavid Sterba if (path->reada == READA_BACK) { 14776b80053dSChris Mason if (nr == 0) 14783c69faecSChris Mason break; 14796b80053dSChris Mason nr--; 1480ace75066SFilipe Manana } else if (path->reada == READA_FORWARD || 1481ace75066SFilipe Manana path->reada == READA_FORWARD_ALWAYS) { 14826b80053dSChris Mason nr++; 14836b80053dSChris Mason if (nr >= nritems) 14846b80053dSChris Mason break; 14853c69faecSChris Mason } 1486e4058b54SDavid Sterba if (path->reada == READA_BACK && objectid) { 148701f46658SChris Mason btrfs_node_key(node, &disk_key, nr); 148801f46658SChris Mason if (btrfs_disk_key_objectid(&disk_key) != objectid) 148901f46658SChris Mason break; 149001f46658SChris Mason } 14916b80053dSChris Mason search = btrfs_node_blockptr(node, nr); 1492ace75066SFilipe Manana if (path->reada == READA_FORWARD_ALWAYS || 1493ace75066SFilipe Manana (search <= target && target - search <= 65536) || 1494a7175319SChris Mason (search > target && search - target <= 65536)) { 1495bfb484d9SJosef Bacik btrfs_readahead_node_child(node, nr); 14966b80053dSChris Mason nread += blocksize; 14973c69faecSChris Mason } 14986b80053dSChris Mason nscan++; 1499ace75066SFilipe Manana if (nread > nread_max || nscan > 32) 15006b80053dSChris Mason break; 15013c69faecSChris Mason } 15023c69faecSChris Mason } 1503925baeddSChris Mason 1504bfb484d9SJosef Bacik static noinline void reada_for_balance(struct btrfs_path *path, int level) 1505b4ce94deSChris Mason { 1506bfb484d9SJosef Bacik struct extent_buffer *parent; 1507b4ce94deSChris Mason int slot; 1508b4ce94deSChris Mason int nritems; 1509b4ce94deSChris Mason 15108c594ea8SChris Mason parent = path->nodes[level + 1]; 1511b4ce94deSChris Mason if (!parent) 15120b08851fSJosef Bacik return; 1513b4ce94deSChris Mason 1514b4ce94deSChris Mason nritems = btrfs_header_nritems(parent); 15158c594ea8SChris Mason slot = path->slots[level + 1]; 1516b4ce94deSChris Mason 1517bfb484d9SJosef Bacik if (slot > 0) 1518bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot - 1); 1519bfb484d9SJosef Bacik if (slot + 1 < nritems) 1520bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot + 1); 1521b4ce94deSChris Mason } 1522b4ce94deSChris Mason 1523b4ce94deSChris Mason 1524b4ce94deSChris Mason /* 1525d397712bSChris Mason * when we walk down the tree, it is usually safe to unlock the higher layers 1526d397712bSChris Mason * in the tree. The exceptions are when our path goes through slot 0, because 1527d397712bSChris Mason * operations on the tree might require changing key pointers higher up in the 1528d397712bSChris Mason * tree. 1529d352ac68SChris Mason * 1530d397712bSChris Mason * callers might also have set path->keep_locks, which tells this code to keep 1531d397712bSChris Mason * the lock if the path points to the last slot in the block. This is part of 1532d397712bSChris Mason * walking through the tree, and selecting the next slot in the higher block. 1533d352ac68SChris Mason * 1534d397712bSChris Mason * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so 1535d397712bSChris Mason * if lowest_unlock is 1, level 0 won't be unlocked 1536d352ac68SChris Mason */ 1537e02119d5SChris Mason static noinline void unlock_up(struct btrfs_path *path, int level, 1538f7c79f30SChris Mason int lowest_unlock, int min_write_lock_level, 1539f7c79f30SChris Mason int *write_lock_level) 1540925baeddSChris Mason { 1541925baeddSChris Mason int i; 1542925baeddSChris Mason int skip_level = level; 1543c1227996SNikolay Borisov bool check_skip = true; 1544925baeddSChris Mason 1545925baeddSChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 1546925baeddSChris Mason if (!path->nodes[i]) 1547925baeddSChris Mason break; 1548925baeddSChris Mason if (!path->locks[i]) 1549925baeddSChris Mason break; 1550c1227996SNikolay Borisov 1551c1227996SNikolay Borisov if (check_skip) { 1552c1227996SNikolay Borisov if (path->slots[i] == 0) { 1553925baeddSChris Mason skip_level = i + 1; 1554925baeddSChris Mason continue; 1555925baeddSChris Mason } 1556c1227996SNikolay Borisov 1557c1227996SNikolay Borisov if (path->keep_locks) { 1558925baeddSChris Mason u32 nritems; 1559c1227996SNikolay Borisov 1560c1227996SNikolay Borisov nritems = btrfs_header_nritems(path->nodes[i]); 1561051e1b9fSChris Mason if (nritems < 1 || path->slots[i] >= nritems - 1) { 1562925baeddSChris Mason skip_level = i + 1; 1563925baeddSChris Mason continue; 1564925baeddSChris Mason } 1565925baeddSChris Mason } 1566c1227996SNikolay Borisov } 1567051e1b9fSChris Mason 1568d80bb3f9SLiu Bo if (i >= lowest_unlock && i > skip_level) { 1569c1227996SNikolay Borisov check_skip = false; 1570c1227996SNikolay Borisov btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); 1571925baeddSChris Mason path->locks[i] = 0; 1572f7c79f30SChris Mason if (write_lock_level && 1573f7c79f30SChris Mason i > min_write_lock_level && 1574f7c79f30SChris Mason i <= *write_lock_level) { 1575f7c79f30SChris Mason *write_lock_level = i - 1; 1576f7c79f30SChris Mason } 1577925baeddSChris Mason } 1578925baeddSChris Mason } 1579925baeddSChris Mason } 1580925baeddSChris Mason 15813c69faecSChris Mason /* 1582376a21d7SFilipe Manana * Helper function for btrfs_search_slot() and other functions that do a search 1583376a21d7SFilipe Manana * on a btree. The goal is to find a tree block in the cache (the radix tree at 1584376a21d7SFilipe Manana * fs_info->buffer_radix), but if we can't find it, or it's not up to date, read 1585376a21d7SFilipe Manana * its pages from disk. 1586c8c42864SChris Mason * 1587376a21d7SFilipe Manana * Returns -EAGAIN, with the path unlocked, if the caller needs to repeat the 1588376a21d7SFilipe Manana * whole btree search, starting again from the current root node. 1589c8c42864SChris Mason */ 1590c8c42864SChris Mason static int 1591d07b8528SLiu Bo read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, 1592c8c42864SChris Mason struct extent_buffer **eb_ret, int level, int slot, 1593cda79c54SDavid Sterba const struct btrfs_key *key) 1594c8c42864SChris Mason { 15950b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1596789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 0 }; 1597c8c42864SChris Mason u64 blocknr; 1598c8c42864SChris Mason u64 gen; 1599c8c42864SChris Mason struct extent_buffer *tmp; 160076a05b35SChris Mason int ret; 1601581c1760SQu Wenruo int parent_level; 1602b246666eSFilipe Manana bool unlock_up; 1603c8c42864SChris Mason 1604b246666eSFilipe Manana unlock_up = ((level + 1 < BTRFS_MAX_LEVEL) && p->locks[level + 1]); 1605213ff4b7SNikolay Borisov blocknr = btrfs_node_blockptr(*eb_ret, slot); 1606213ff4b7SNikolay Borisov gen = btrfs_node_ptr_generation(*eb_ret, slot); 1607213ff4b7SNikolay Borisov parent_level = btrfs_header_level(*eb_ret); 1608789d6a3aSQu Wenruo btrfs_node_key_to_cpu(*eb_ret, &check.first_key, slot); 1609789d6a3aSQu Wenruo check.has_first_key = true; 1610789d6a3aSQu Wenruo check.level = parent_level - 1; 1611789d6a3aSQu Wenruo check.transid = gen; 1612789d6a3aSQu Wenruo check.owner_root = root->root_key.objectid; 1613c8c42864SChris Mason 1614b246666eSFilipe Manana /* 1615b246666eSFilipe Manana * If we need to read an extent buffer from disk and we are holding locks 1616b246666eSFilipe Manana * on upper level nodes, we unlock all the upper nodes before reading the 1617b246666eSFilipe Manana * extent buffer, and then return -EAGAIN to the caller as it needs to 1618b246666eSFilipe Manana * restart the search. We don't release the lock on the current level 1619b246666eSFilipe Manana * because we need to walk this node to figure out which blocks to read. 1620b246666eSFilipe Manana */ 16210b246afaSJeff Mahoney tmp = find_extent_buffer(fs_info, blocknr); 1622cb44921aSChris Mason if (tmp) { 1623ace75066SFilipe Manana if (p->reada == READA_FORWARD_ALWAYS) 1624ace75066SFilipe Manana reada_for_search(fs_info, p, level, slot, key->objectid); 1625ace75066SFilipe Manana 1626b9fab919SChris Mason /* first we do an atomic uptodate check */ 1627b9fab919SChris Mason if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { 1628448de471SQu Wenruo /* 1629448de471SQu Wenruo * Do extra check for first_key, eb can be stale due to 1630448de471SQu Wenruo * being cached, read from scrub, or have multiple 1631448de471SQu Wenruo * parents (shared tree blocks). 1632448de471SQu Wenruo */ 1633e064d5e9SDavid Sterba if (btrfs_verify_level_key(tmp, 1634789d6a3aSQu Wenruo parent_level - 1, &check.first_key, gen)) { 1635448de471SQu Wenruo free_extent_buffer(tmp); 1636448de471SQu Wenruo return -EUCLEAN; 1637448de471SQu Wenruo } 1638c8c42864SChris Mason *eb_ret = tmp; 1639c8c42864SChris Mason return 0; 1640c8c42864SChris Mason } 1641bdf7c00eSJosef Bacik 1642857bc13fSJosef Bacik if (p->nowait) { 1643857bc13fSJosef Bacik free_extent_buffer(tmp); 1644857bc13fSJosef Bacik return -EAGAIN; 1645857bc13fSJosef Bacik } 1646857bc13fSJosef Bacik 1647b246666eSFilipe Manana if (unlock_up) 1648b246666eSFilipe Manana btrfs_unlock_up_safe(p, level + 1); 1649b246666eSFilipe Manana 1650b9fab919SChris Mason /* now we're allowed to do a blocking uptodate check */ 1651789d6a3aSQu Wenruo ret = btrfs_read_extent_buffer(tmp, &check); 16529a4ffa1bSQu Wenruo if (ret) { 1653cb44921aSChris Mason free_extent_buffer(tmp); 1654b3b4aa74SDavid Sterba btrfs_release_path(p); 1655cb44921aSChris Mason return -EIO; 1656cb44921aSChris Mason } 165788c602abSQu Wenruo if (btrfs_check_eb_owner(tmp, root->root_key.objectid)) { 165888c602abSQu Wenruo free_extent_buffer(tmp); 165988c602abSQu Wenruo btrfs_release_path(p); 166088c602abSQu Wenruo return -EUCLEAN; 166188c602abSQu Wenruo } 1662b246666eSFilipe Manana 1663b246666eSFilipe Manana if (unlock_up) 1664b246666eSFilipe Manana ret = -EAGAIN; 1665b246666eSFilipe Manana 1666b246666eSFilipe Manana goto out; 1667857bc13fSJosef Bacik } else if (p->nowait) { 1668857bc13fSJosef Bacik return -EAGAIN; 16699a4ffa1bSQu Wenruo } 1670c8c42864SChris Mason 1671b246666eSFilipe Manana if (unlock_up) { 16728c594ea8SChris Mason btrfs_unlock_up_safe(p, level + 1); 16734bb59055SFilipe Manana ret = -EAGAIN; 16744bb59055SFilipe Manana } else { 16754bb59055SFilipe Manana ret = 0; 16764bb59055SFilipe Manana } 16778c594ea8SChris Mason 1678e4058b54SDavid Sterba if (p->reada != READA_NONE) 16792ff7e61eSJeff Mahoney reada_for_search(fs_info, p, level, slot, key->objectid); 1680c8c42864SChris Mason 1681789d6a3aSQu Wenruo tmp = read_tree_block(fs_info, blocknr, &check); 16824eb150d6SQu Wenruo if (IS_ERR(tmp)) { 16834eb150d6SQu Wenruo btrfs_release_path(p); 16844eb150d6SQu Wenruo return PTR_ERR(tmp); 16854eb150d6SQu Wenruo } 168676a05b35SChris Mason /* 168776a05b35SChris Mason * If the read above didn't mark this buffer up to date, 168876a05b35SChris Mason * it will never end up being up to date. Set ret to EIO now 168976a05b35SChris Mason * and give up so that our caller doesn't loop forever 169076a05b35SChris Mason * on our EAGAINs. 169176a05b35SChris Mason */ 1692e6a1d6fdSLiu Bo if (!extent_buffer_uptodate(tmp)) 169376a05b35SChris Mason ret = -EIO; 169402a3307aSLiu Bo 1695b246666eSFilipe Manana out: 16964bb59055SFilipe Manana if (ret == 0) { 16974bb59055SFilipe Manana *eb_ret = tmp; 16984bb59055SFilipe Manana } else { 16994bb59055SFilipe Manana free_extent_buffer(tmp); 170002a3307aSLiu Bo btrfs_release_path(p); 17014bb59055SFilipe Manana } 17024bb59055SFilipe Manana 170376a05b35SChris Mason return ret; 1704c8c42864SChris Mason } 1705c8c42864SChris Mason 1706c8c42864SChris Mason /* 1707c8c42864SChris Mason * helper function for btrfs_search_slot. This does all of the checks 1708c8c42864SChris Mason * for node-level blocks and does any balancing required based on 1709c8c42864SChris Mason * the ins_len. 1710c8c42864SChris Mason * 1711c8c42864SChris Mason * If no extra work was required, zero is returned. If we had to 1712c8c42864SChris Mason * drop the path, -EAGAIN is returned and btrfs_search_slot must 1713c8c42864SChris Mason * start over 1714c8c42864SChris Mason */ 1715c8c42864SChris Mason static int 1716c8c42864SChris Mason setup_nodes_for_search(struct btrfs_trans_handle *trans, 1717c8c42864SChris Mason struct btrfs_root *root, struct btrfs_path *p, 1718bd681513SChris Mason struct extent_buffer *b, int level, int ins_len, 1719bd681513SChris Mason int *write_lock_level) 1720c8c42864SChris Mason { 17210b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 172295b982deSNikolay Borisov int ret = 0; 17230b246afaSJeff Mahoney 1724c8c42864SChris Mason if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= 17250b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { 1726c8c42864SChris Mason 1727bd681513SChris Mason if (*write_lock_level < level + 1) { 1728bd681513SChris Mason *write_lock_level = level + 1; 1729bd681513SChris Mason btrfs_release_path(p); 173095b982deSNikolay Borisov return -EAGAIN; 1731bd681513SChris Mason } 1732bd681513SChris Mason 1733bfb484d9SJosef Bacik reada_for_balance(p, level); 173495b982deSNikolay Borisov ret = split_node(trans, root, p, level); 1735c8c42864SChris Mason 1736c8c42864SChris Mason b = p->nodes[level]; 1737c8c42864SChris Mason } else if (ins_len < 0 && btrfs_header_nritems(b) < 17380b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { 1739c8c42864SChris Mason 1740bd681513SChris Mason if (*write_lock_level < level + 1) { 1741bd681513SChris Mason *write_lock_level = level + 1; 1742bd681513SChris Mason btrfs_release_path(p); 174395b982deSNikolay Borisov return -EAGAIN; 1744bd681513SChris Mason } 1745bd681513SChris Mason 1746bfb484d9SJosef Bacik reada_for_balance(p, level); 174795b982deSNikolay Borisov ret = balance_level(trans, root, p, level); 174895b982deSNikolay Borisov if (ret) 174995b982deSNikolay Borisov return ret; 1750c8c42864SChris Mason 1751c8c42864SChris Mason b = p->nodes[level]; 1752c8c42864SChris Mason if (!b) { 1753b3b4aa74SDavid Sterba btrfs_release_path(p); 175495b982deSNikolay Borisov return -EAGAIN; 1755c8c42864SChris Mason } 1756c8c42864SChris Mason BUG_ON(btrfs_header_nritems(b) == 1); 1757c8c42864SChris Mason } 1758c8c42864SChris Mason return ret; 1759c8c42864SChris Mason } 1760c8c42864SChris Mason 1761381cf658SDavid Sterba int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, 1762e33d5c3dSKelley Nielsen u64 iobjectid, u64 ioff, u8 key_type, 1763e33d5c3dSKelley Nielsen struct btrfs_key *found_key) 1764e33d5c3dSKelley Nielsen { 1765e33d5c3dSKelley Nielsen int ret; 1766e33d5c3dSKelley Nielsen struct btrfs_key key; 1767e33d5c3dSKelley Nielsen struct extent_buffer *eb; 1768381cf658SDavid Sterba 1769381cf658SDavid Sterba ASSERT(path); 17701d4c08e0SDavid Sterba ASSERT(found_key); 1771e33d5c3dSKelley Nielsen 1772e33d5c3dSKelley Nielsen key.type = key_type; 1773e33d5c3dSKelley Nielsen key.objectid = iobjectid; 1774e33d5c3dSKelley Nielsen key.offset = ioff; 1775e33d5c3dSKelley Nielsen 1776e33d5c3dSKelley Nielsen ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); 17771d4c08e0SDavid Sterba if (ret < 0) 1778e33d5c3dSKelley Nielsen return ret; 1779e33d5c3dSKelley Nielsen 1780e33d5c3dSKelley Nielsen eb = path->nodes[0]; 1781e33d5c3dSKelley Nielsen if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { 1782e33d5c3dSKelley Nielsen ret = btrfs_next_leaf(fs_root, path); 1783e33d5c3dSKelley Nielsen if (ret) 1784e33d5c3dSKelley Nielsen return ret; 1785e33d5c3dSKelley Nielsen eb = path->nodes[0]; 1786e33d5c3dSKelley Nielsen } 1787e33d5c3dSKelley Nielsen 1788e33d5c3dSKelley Nielsen btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); 1789e33d5c3dSKelley Nielsen if (found_key->type != key.type || 1790e33d5c3dSKelley Nielsen found_key->objectid != key.objectid) 1791e33d5c3dSKelley Nielsen return 1; 1792e33d5c3dSKelley Nielsen 1793e33d5c3dSKelley Nielsen return 0; 1794e33d5c3dSKelley Nielsen } 1795e33d5c3dSKelley Nielsen 17961fc28d8eSLiu Bo static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, 17971fc28d8eSLiu Bo struct btrfs_path *p, 17981fc28d8eSLiu Bo int write_lock_level) 17991fc28d8eSLiu Bo { 18001fc28d8eSLiu Bo struct extent_buffer *b; 1801120de408SJosef Bacik int root_lock = 0; 18021fc28d8eSLiu Bo int level = 0; 18031fc28d8eSLiu Bo 18041fc28d8eSLiu Bo if (p->search_commit_root) { 18051fc28d8eSLiu Bo b = root->commit_root; 180667439dadSDavid Sterba atomic_inc(&b->refs); 18071fc28d8eSLiu Bo level = btrfs_header_level(b); 1808f9ddfd05SLiu Bo /* 1809f9ddfd05SLiu Bo * Ensure that all callers have set skip_locking when 1810f9ddfd05SLiu Bo * p->search_commit_root = 1. 1811f9ddfd05SLiu Bo */ 1812f9ddfd05SLiu Bo ASSERT(p->skip_locking == 1); 18131fc28d8eSLiu Bo 18141fc28d8eSLiu Bo goto out; 18151fc28d8eSLiu Bo } 18161fc28d8eSLiu Bo 18171fc28d8eSLiu Bo if (p->skip_locking) { 18181fc28d8eSLiu Bo b = btrfs_root_node(root); 18191fc28d8eSLiu Bo level = btrfs_header_level(b); 18201fc28d8eSLiu Bo goto out; 18211fc28d8eSLiu Bo } 18221fc28d8eSLiu Bo 1823120de408SJosef Bacik /* We try very hard to do read locks on the root */ 1824120de408SJosef Bacik root_lock = BTRFS_READ_LOCK; 1825120de408SJosef Bacik 18261fc28d8eSLiu Bo /* 1827662c653bSLiu Bo * If the level is set to maximum, we can skip trying to get the read 1828662c653bSLiu Bo * lock. 1829662c653bSLiu Bo */ 1830662c653bSLiu Bo if (write_lock_level < BTRFS_MAX_LEVEL) { 1831662c653bSLiu Bo /* 1832662c653bSLiu Bo * We don't know the level of the root node until we actually 1833662c653bSLiu Bo * have it read locked 18341fc28d8eSLiu Bo */ 1835857bc13fSJosef Bacik if (p->nowait) { 1836857bc13fSJosef Bacik b = btrfs_try_read_lock_root_node(root); 1837857bc13fSJosef Bacik if (IS_ERR(b)) 1838857bc13fSJosef Bacik return b; 1839857bc13fSJosef Bacik } else { 18401bb96598SJosef Bacik b = btrfs_read_lock_root_node(root); 1841857bc13fSJosef Bacik } 18421fc28d8eSLiu Bo level = btrfs_header_level(b); 18431fc28d8eSLiu Bo if (level > write_lock_level) 18441fc28d8eSLiu Bo goto out; 18451fc28d8eSLiu Bo 1846662c653bSLiu Bo /* Whoops, must trade for write lock */ 18471fc28d8eSLiu Bo btrfs_tree_read_unlock(b); 18481fc28d8eSLiu Bo free_extent_buffer(b); 1849662c653bSLiu Bo } 1850662c653bSLiu Bo 18511fc28d8eSLiu Bo b = btrfs_lock_root_node(root); 18521fc28d8eSLiu Bo root_lock = BTRFS_WRITE_LOCK; 18531fc28d8eSLiu Bo 18541fc28d8eSLiu Bo /* The level might have changed, check again */ 18551fc28d8eSLiu Bo level = btrfs_header_level(b); 18561fc28d8eSLiu Bo 18571fc28d8eSLiu Bo out: 1858120de408SJosef Bacik /* 1859120de408SJosef Bacik * The root may have failed to write out at some point, and thus is no 1860120de408SJosef Bacik * longer valid, return an error in this case. 1861120de408SJosef Bacik */ 1862120de408SJosef Bacik if (!extent_buffer_uptodate(b)) { 1863120de408SJosef Bacik if (root_lock) 1864120de408SJosef Bacik btrfs_tree_unlock_rw(b, root_lock); 1865120de408SJosef Bacik free_extent_buffer(b); 1866120de408SJosef Bacik return ERR_PTR(-EIO); 1867120de408SJosef Bacik } 1868120de408SJosef Bacik 18691fc28d8eSLiu Bo p->nodes[level] = b; 18701fc28d8eSLiu Bo if (!p->skip_locking) 18711fc28d8eSLiu Bo p->locks[level] = root_lock; 18721fc28d8eSLiu Bo /* 18731fc28d8eSLiu Bo * Callers are responsible for dropping b's references. 18741fc28d8eSLiu Bo */ 18751fc28d8eSLiu Bo return b; 18761fc28d8eSLiu Bo } 18771fc28d8eSLiu Bo 1878d96b3424SFilipe Manana /* 1879d96b3424SFilipe Manana * Replace the extent buffer at the lowest level of the path with a cloned 1880d96b3424SFilipe Manana * version. The purpose is to be able to use it safely, after releasing the 1881d96b3424SFilipe Manana * commit root semaphore, even if relocation is happening in parallel, the 1882d96b3424SFilipe Manana * transaction used for relocation is committed and the extent buffer is 1883d96b3424SFilipe Manana * reallocated in the next transaction. 1884d96b3424SFilipe Manana * 1885d96b3424SFilipe Manana * This is used in a context where the caller does not prevent transaction 1886d96b3424SFilipe Manana * commits from happening, either by holding a transaction handle or holding 1887d96b3424SFilipe Manana * some lock, while it's doing searches through a commit root. 1888d96b3424SFilipe Manana * At the moment it's only used for send operations. 1889d96b3424SFilipe Manana */ 1890d96b3424SFilipe Manana static int finish_need_commit_sem_search(struct btrfs_path *path) 1891d96b3424SFilipe Manana { 1892d96b3424SFilipe Manana const int i = path->lowest_level; 1893d96b3424SFilipe Manana const int slot = path->slots[i]; 1894d96b3424SFilipe Manana struct extent_buffer *lowest = path->nodes[i]; 1895d96b3424SFilipe Manana struct extent_buffer *clone; 1896d96b3424SFilipe Manana 1897d96b3424SFilipe Manana ASSERT(path->need_commit_sem); 1898d96b3424SFilipe Manana 1899d96b3424SFilipe Manana if (!lowest) 1900d96b3424SFilipe Manana return 0; 1901d96b3424SFilipe Manana 1902d96b3424SFilipe Manana lockdep_assert_held_read(&lowest->fs_info->commit_root_sem); 1903d96b3424SFilipe Manana 1904d96b3424SFilipe Manana clone = btrfs_clone_extent_buffer(lowest); 1905d96b3424SFilipe Manana if (!clone) 1906d96b3424SFilipe Manana return -ENOMEM; 1907d96b3424SFilipe Manana 1908d96b3424SFilipe Manana btrfs_release_path(path); 1909d96b3424SFilipe Manana path->nodes[i] = clone; 1910d96b3424SFilipe Manana path->slots[i] = slot; 1911d96b3424SFilipe Manana 1912d96b3424SFilipe Manana return 0; 1913d96b3424SFilipe Manana } 19141fc28d8eSLiu Bo 1915e2e58d0fSFilipe Manana static inline int search_for_key_slot(struct extent_buffer *eb, 1916e2e58d0fSFilipe Manana int search_low_slot, 1917e2e58d0fSFilipe Manana const struct btrfs_key *key, 1918e2e58d0fSFilipe Manana int prev_cmp, 1919e2e58d0fSFilipe Manana int *slot) 1920e2e58d0fSFilipe Manana { 1921e2e58d0fSFilipe Manana /* 1922e2e58d0fSFilipe Manana * If a previous call to btrfs_bin_search() on a parent node returned an 1923e2e58d0fSFilipe Manana * exact match (prev_cmp == 0), we can safely assume the target key will 1924e2e58d0fSFilipe Manana * always be at slot 0 on lower levels, since each key pointer 1925e2e58d0fSFilipe Manana * (struct btrfs_key_ptr) refers to the lowest key accessible from the 1926e2e58d0fSFilipe Manana * subtree it points to. Thus we can skip searching lower levels. 1927e2e58d0fSFilipe Manana */ 1928e2e58d0fSFilipe Manana if (prev_cmp == 0) { 1929e2e58d0fSFilipe Manana *slot = 0; 1930e2e58d0fSFilipe Manana return 0; 1931e2e58d0fSFilipe Manana } 1932e2e58d0fSFilipe Manana 1933fdf8d595SAnand Jain return btrfs_bin_search(eb, search_low_slot, key, slot); 1934e2e58d0fSFilipe Manana } 1935e2e58d0fSFilipe Manana 1936109324cfSFilipe Manana static int search_leaf(struct btrfs_trans_handle *trans, 1937109324cfSFilipe Manana struct btrfs_root *root, 1938109324cfSFilipe Manana const struct btrfs_key *key, 1939109324cfSFilipe Manana struct btrfs_path *path, 1940109324cfSFilipe Manana int ins_len, 1941109324cfSFilipe Manana int prev_cmp) 1942109324cfSFilipe Manana { 1943109324cfSFilipe Manana struct extent_buffer *leaf = path->nodes[0]; 1944109324cfSFilipe Manana int leaf_free_space = -1; 1945109324cfSFilipe Manana int search_low_slot = 0; 1946109324cfSFilipe Manana int ret; 1947109324cfSFilipe Manana bool do_bin_search = true; 1948109324cfSFilipe Manana 1949109324cfSFilipe Manana /* 1950109324cfSFilipe Manana * If we are doing an insertion, the leaf has enough free space and the 1951109324cfSFilipe Manana * destination slot for the key is not slot 0, then we can unlock our 1952109324cfSFilipe Manana * write lock on the parent, and any other upper nodes, before doing the 1953109324cfSFilipe Manana * binary search on the leaf (with search_for_key_slot()), allowing other 1954109324cfSFilipe Manana * tasks to lock the parent and any other upper nodes. 1955109324cfSFilipe Manana */ 1956109324cfSFilipe Manana if (ins_len > 0) { 1957109324cfSFilipe Manana /* 1958109324cfSFilipe Manana * Cache the leaf free space, since we will need it later and it 1959109324cfSFilipe Manana * will not change until then. 1960109324cfSFilipe Manana */ 1961109324cfSFilipe Manana leaf_free_space = btrfs_leaf_free_space(leaf); 1962109324cfSFilipe Manana 1963109324cfSFilipe Manana /* 1964109324cfSFilipe Manana * !path->locks[1] means we have a single node tree, the leaf is 1965109324cfSFilipe Manana * the root of the tree. 1966109324cfSFilipe Manana */ 1967109324cfSFilipe Manana if (path->locks[1] && leaf_free_space >= ins_len) { 1968109324cfSFilipe Manana struct btrfs_disk_key first_key; 1969109324cfSFilipe Manana 1970109324cfSFilipe Manana ASSERT(btrfs_header_nritems(leaf) > 0); 1971109324cfSFilipe Manana btrfs_item_key(leaf, &first_key, 0); 1972109324cfSFilipe Manana 1973109324cfSFilipe Manana /* 1974109324cfSFilipe Manana * Doing the extra comparison with the first key is cheap, 1975109324cfSFilipe Manana * taking into account that the first key is very likely 1976109324cfSFilipe Manana * already in a cache line because it immediately follows 1977109324cfSFilipe Manana * the extent buffer's header and we have recently accessed 1978109324cfSFilipe Manana * the header's level field. 1979109324cfSFilipe Manana */ 1980109324cfSFilipe Manana ret = comp_keys(&first_key, key); 1981109324cfSFilipe Manana if (ret < 0) { 1982109324cfSFilipe Manana /* 1983109324cfSFilipe Manana * The first key is smaller than the key we want 1984109324cfSFilipe Manana * to insert, so we are safe to unlock all upper 1985109324cfSFilipe Manana * nodes and we have to do the binary search. 1986109324cfSFilipe Manana * 1987109324cfSFilipe Manana * We do use btrfs_unlock_up_safe() and not 1988109324cfSFilipe Manana * unlock_up() because the later does not unlock 1989109324cfSFilipe Manana * nodes with a slot of 0 - we can safely unlock 1990109324cfSFilipe Manana * any node even if its slot is 0 since in this 1991109324cfSFilipe Manana * case the key does not end up at slot 0 of the 1992109324cfSFilipe Manana * leaf and there's no need to split the leaf. 1993109324cfSFilipe Manana */ 1994109324cfSFilipe Manana btrfs_unlock_up_safe(path, 1); 1995109324cfSFilipe Manana search_low_slot = 1; 1996109324cfSFilipe Manana } else { 1997109324cfSFilipe Manana /* 1998109324cfSFilipe Manana * The first key is >= then the key we want to 1999109324cfSFilipe Manana * insert, so we can skip the binary search as 2000109324cfSFilipe Manana * the target key will be at slot 0. 2001109324cfSFilipe Manana * 2002109324cfSFilipe Manana * We can not unlock upper nodes when the key is 2003109324cfSFilipe Manana * less than the first key, because we will need 2004109324cfSFilipe Manana * to update the key at slot 0 of the parent node 2005109324cfSFilipe Manana * and possibly of other upper nodes too. 2006109324cfSFilipe Manana * If the key matches the first key, then we can 2007109324cfSFilipe Manana * unlock all the upper nodes, using 2008109324cfSFilipe Manana * btrfs_unlock_up_safe() instead of unlock_up() 2009109324cfSFilipe Manana * as stated above. 2010109324cfSFilipe Manana */ 2011109324cfSFilipe Manana if (ret == 0) 2012109324cfSFilipe Manana btrfs_unlock_up_safe(path, 1); 2013109324cfSFilipe Manana /* 2014109324cfSFilipe Manana * ret is already 0 or 1, matching the result of 2015109324cfSFilipe Manana * a btrfs_bin_search() call, so there is no need 2016109324cfSFilipe Manana * to adjust it. 2017109324cfSFilipe Manana */ 2018109324cfSFilipe Manana do_bin_search = false; 2019109324cfSFilipe Manana path->slots[0] = 0; 2020109324cfSFilipe Manana } 2021109324cfSFilipe Manana } 2022109324cfSFilipe Manana } 2023109324cfSFilipe Manana 2024109324cfSFilipe Manana if (do_bin_search) { 2025109324cfSFilipe Manana ret = search_for_key_slot(leaf, search_low_slot, key, 2026109324cfSFilipe Manana prev_cmp, &path->slots[0]); 2027109324cfSFilipe Manana if (ret < 0) 2028109324cfSFilipe Manana return ret; 2029109324cfSFilipe Manana } 2030109324cfSFilipe Manana 2031109324cfSFilipe Manana if (ins_len > 0) { 2032109324cfSFilipe Manana /* 2033109324cfSFilipe Manana * Item key already exists. In this case, if we are allowed to 2034109324cfSFilipe Manana * insert the item (for example, in dir_item case, item key 2035109324cfSFilipe Manana * collision is allowed), it will be merged with the original 2036109324cfSFilipe Manana * item. Only the item size grows, no new btrfs item will be 2037109324cfSFilipe Manana * added. If search_for_extension is not set, ins_len already 2038109324cfSFilipe Manana * accounts the size btrfs_item, deduct it here so leaf space 2039109324cfSFilipe Manana * check will be correct. 2040109324cfSFilipe Manana */ 2041109324cfSFilipe Manana if (ret == 0 && !path->search_for_extension) { 2042109324cfSFilipe Manana ASSERT(ins_len >= sizeof(struct btrfs_item)); 2043109324cfSFilipe Manana ins_len -= sizeof(struct btrfs_item); 2044109324cfSFilipe Manana } 2045109324cfSFilipe Manana 2046109324cfSFilipe Manana ASSERT(leaf_free_space >= 0); 2047109324cfSFilipe Manana 2048109324cfSFilipe Manana if (leaf_free_space < ins_len) { 2049109324cfSFilipe Manana int err; 2050109324cfSFilipe Manana 2051109324cfSFilipe Manana err = split_leaf(trans, root, key, path, ins_len, 2052109324cfSFilipe Manana (ret == 0)); 2053bb8e9a60SFilipe Manana ASSERT(err <= 0); 2054bb8e9a60SFilipe Manana if (WARN_ON(err > 0)) 2055bb8e9a60SFilipe Manana err = -EUCLEAN; 2056109324cfSFilipe Manana if (err) 2057109324cfSFilipe Manana ret = err; 2058109324cfSFilipe Manana } 2059109324cfSFilipe Manana } 2060109324cfSFilipe Manana 2061109324cfSFilipe Manana return ret; 2062109324cfSFilipe Manana } 2063109324cfSFilipe Manana 2064c8c42864SChris Mason /* 20654271eceaSNikolay Borisov * btrfs_search_slot - look for a key in a tree and perform necessary 20664271eceaSNikolay Borisov * modifications to preserve tree invariants. 206774123bd7SChris Mason * 20684271eceaSNikolay Borisov * @trans: Handle of transaction, used when modifying the tree 20694271eceaSNikolay Borisov * @p: Holds all btree nodes along the search path 20704271eceaSNikolay Borisov * @root: The root node of the tree 20714271eceaSNikolay Borisov * @key: The key we are looking for 20729a664971Sethanwu * @ins_len: Indicates purpose of search: 20739a664971Sethanwu * >0 for inserts it's size of item inserted (*) 20749a664971Sethanwu * <0 for deletions 20759a664971Sethanwu * 0 for plain searches, not modifying the tree 20769a664971Sethanwu * 20779a664971Sethanwu * (*) If size of item inserted doesn't include 20789a664971Sethanwu * sizeof(struct btrfs_item), then p->search_for_extension must 20799a664971Sethanwu * be set. 20804271eceaSNikolay Borisov * @cow: boolean should CoW operations be performed. Must always be 1 20814271eceaSNikolay Borisov * when modifying the tree. 208297571fd0SChris Mason * 20834271eceaSNikolay Borisov * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. 20844271eceaSNikolay Borisov * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) 20854271eceaSNikolay Borisov * 20864271eceaSNikolay Borisov * If @key is found, 0 is returned and you can find the item in the leaf level 20874271eceaSNikolay Borisov * of the path (level 0) 20884271eceaSNikolay Borisov * 20894271eceaSNikolay Borisov * If @key isn't found, 1 is returned and the leaf level of the path (level 0) 20904271eceaSNikolay Borisov * points to the slot where it should be inserted 20914271eceaSNikolay Borisov * 20924271eceaSNikolay Borisov * If an error is encountered while searching the tree a negative error number 20934271eceaSNikolay Borisov * is returned 209474123bd7SChris Mason */ 2095310712b2SOmar Sandoval int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, 2096310712b2SOmar Sandoval const struct btrfs_key *key, struct btrfs_path *p, 2097310712b2SOmar Sandoval int ins_len, int cow) 2098be0e5c09SChris Mason { 2099d96b3424SFilipe Manana struct btrfs_fs_info *fs_info = root->fs_info; 21005f39d397SChris Mason struct extent_buffer *b; 2101be0e5c09SChris Mason int slot; 2102be0e5c09SChris Mason int ret; 210333c66f43SYan Zheng int err; 2104be0e5c09SChris Mason int level; 2105925baeddSChris Mason int lowest_unlock = 1; 2106bd681513SChris Mason /* everything at write_lock_level or lower must be write locked */ 2107bd681513SChris Mason int write_lock_level = 0; 21089f3a7427SChris Mason u8 lowest_level = 0; 2109f7c79f30SChris Mason int min_write_lock_level; 2110d7396f07SFilipe David Borba Manana int prev_cmp; 21119f3a7427SChris Mason 2112a4c853afSChenXiaoSong might_sleep(); 2113a4c853afSChenXiaoSong 21146702ed49SChris Mason lowest_level = p->lowest_level; 2115323ac95bSChris Mason WARN_ON(lowest_level && ins_len > 0); 211622b0ebdaSChris Mason WARN_ON(p->nodes[0] != NULL); 2117eb653de1SFilipe David Borba Manana BUG_ON(!cow && ins_len); 211825179201SJosef Bacik 2119857bc13fSJosef Bacik /* 2120857bc13fSJosef Bacik * For now only allow nowait for read only operations. There's no 2121857bc13fSJosef Bacik * strict reason why we can't, we just only need it for reads so it's 2122857bc13fSJosef Bacik * only implemented for reads. 2123857bc13fSJosef Bacik */ 2124857bc13fSJosef Bacik ASSERT(!p->nowait || !cow); 2125857bc13fSJosef Bacik 2126bd681513SChris Mason if (ins_len < 0) { 2127925baeddSChris Mason lowest_unlock = 2; 212865b51a00SChris Mason 2129bd681513SChris Mason /* when we are removing items, we might have to go up to level 2130bd681513SChris Mason * two as we update tree pointers Make sure we keep write 2131bd681513SChris Mason * for those levels as well 2132bd681513SChris Mason */ 2133bd681513SChris Mason write_lock_level = 2; 2134bd681513SChris Mason } else if (ins_len > 0) { 2135bd681513SChris Mason /* 2136bd681513SChris Mason * for inserting items, make sure we have a write lock on 2137bd681513SChris Mason * level 1 so we can update keys 2138bd681513SChris Mason */ 2139bd681513SChris Mason write_lock_level = 1; 2140bd681513SChris Mason } 2141bd681513SChris Mason 2142bd681513SChris Mason if (!cow) 2143bd681513SChris Mason write_lock_level = -1; 2144bd681513SChris Mason 214509a2a8f9SJosef Bacik if (cow && (p->keep_locks || p->lowest_level)) 2146bd681513SChris Mason write_lock_level = BTRFS_MAX_LEVEL; 2147bd681513SChris Mason 2148f7c79f30SChris Mason min_write_lock_level = write_lock_level; 2149f7c79f30SChris Mason 2150d96b3424SFilipe Manana if (p->need_commit_sem) { 2151d96b3424SFilipe Manana ASSERT(p->search_commit_root); 2152857bc13fSJosef Bacik if (p->nowait) { 2153857bc13fSJosef Bacik if (!down_read_trylock(&fs_info->commit_root_sem)) 2154857bc13fSJosef Bacik return -EAGAIN; 2155857bc13fSJosef Bacik } else { 2156d96b3424SFilipe Manana down_read(&fs_info->commit_root_sem); 2157d96b3424SFilipe Manana } 2158857bc13fSJosef Bacik } 2159d96b3424SFilipe Manana 2160bb803951SChris Mason again: 2161d7396f07SFilipe David Borba Manana prev_cmp = -1; 21621fc28d8eSLiu Bo b = btrfs_search_slot_get_root(root, p, write_lock_level); 2163be6821f8SFilipe Manana if (IS_ERR(b)) { 2164be6821f8SFilipe Manana ret = PTR_ERR(b); 2165be6821f8SFilipe Manana goto done; 2166be6821f8SFilipe Manana } 2167925baeddSChris Mason 2168eb60ceacSChris Mason while (b) { 2169f624d976SQu Wenruo int dec = 0; 2170f624d976SQu Wenruo 21715f39d397SChris Mason level = btrfs_header_level(b); 217265b51a00SChris Mason 217302217ed2SChris Mason if (cow) { 21749ea2c7c9SNikolay Borisov bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); 21759ea2c7c9SNikolay Borisov 2176c8c42864SChris Mason /* 2177c8c42864SChris Mason * if we don't really need to cow this block 2178c8c42864SChris Mason * then we don't want to set the path blocking, 2179c8c42864SChris Mason * so we test it here 2180c8c42864SChris Mason */ 21815963ffcaSJosef Bacik if (!should_cow_block(trans, root, b)) 218265b51a00SChris Mason goto cow_done; 21835d4f98a2SYan Zheng 2184bd681513SChris Mason /* 2185bd681513SChris Mason * must have write locks on this node and the 2186bd681513SChris Mason * parent 2187bd681513SChris Mason */ 21885124e00eSJosef Bacik if (level > write_lock_level || 21895124e00eSJosef Bacik (level + 1 > write_lock_level && 21905124e00eSJosef Bacik level + 1 < BTRFS_MAX_LEVEL && 21915124e00eSJosef Bacik p->nodes[level + 1])) { 2192bd681513SChris Mason write_lock_level = level + 1; 2193bd681513SChris Mason btrfs_release_path(p); 2194bd681513SChris Mason goto again; 2195bd681513SChris Mason } 2196bd681513SChris Mason 21979ea2c7c9SNikolay Borisov if (last_level) 21989ea2c7c9SNikolay Borisov err = btrfs_cow_block(trans, root, b, NULL, 0, 21999631e4ccSJosef Bacik &b, 22009631e4ccSJosef Bacik BTRFS_NESTING_COW); 22019ea2c7c9SNikolay Borisov else 220233c66f43SYan Zheng err = btrfs_cow_block(trans, root, b, 2203e20d96d6SChris Mason p->nodes[level + 1], 22049631e4ccSJosef Bacik p->slots[level + 1], &b, 22059631e4ccSJosef Bacik BTRFS_NESTING_COW); 220633c66f43SYan Zheng if (err) { 220733c66f43SYan Zheng ret = err; 220865b51a00SChris Mason goto done; 220954aa1f4dSChris Mason } 221002217ed2SChris Mason } 221165b51a00SChris Mason cow_done: 2212eb60ceacSChris Mason p->nodes[level] = b; 2213b4ce94deSChris Mason 2214b4ce94deSChris Mason /* 2215b4ce94deSChris Mason * we have a lock on b and as long as we aren't changing 2216b4ce94deSChris Mason * the tree, there is no way to for the items in b to change. 2217b4ce94deSChris Mason * It is safe to drop the lock on our parent before we 2218b4ce94deSChris Mason * go through the expensive btree search on b. 2219b4ce94deSChris Mason * 2220eb653de1SFilipe David Borba Manana * If we're inserting or deleting (ins_len != 0), then we might 2221eb653de1SFilipe David Borba Manana * be changing slot zero, which may require changing the parent. 2222eb653de1SFilipe David Borba Manana * So, we can't drop the lock until after we know which slot 2223eb653de1SFilipe David Borba Manana * we're operating on. 2224b4ce94deSChris Mason */ 2225eb653de1SFilipe David Borba Manana if (!ins_len && !p->keep_locks) { 2226eb653de1SFilipe David Borba Manana int u = level + 1; 2227eb653de1SFilipe David Borba Manana 2228eb653de1SFilipe David Borba Manana if (u < BTRFS_MAX_LEVEL && p->locks[u]) { 2229eb653de1SFilipe David Borba Manana btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); 2230eb653de1SFilipe David Borba Manana p->locks[u] = 0; 2231eb653de1SFilipe David Borba Manana } 2232eb653de1SFilipe David Borba Manana } 2233b4ce94deSChris Mason 2234e2e58d0fSFilipe Manana if (level == 0) { 2235109324cfSFilipe Manana if (ins_len > 0) 2236e5e1c174SFilipe Manana ASSERT(write_lock_level >= 1); 2237bd681513SChris Mason 2238109324cfSFilipe Manana ret = search_leaf(trans, root, key, p, ins_len, prev_cmp); 2239459931ecSChris Mason if (!p->search_for_split) 2240f7c79f30SChris Mason unlock_up(p, level, lowest_unlock, 22414b6f8e96SLiu Bo min_write_lock_level, NULL); 224265b51a00SChris Mason goto done; 224365b51a00SChris Mason } 2244e2e58d0fSFilipe Manana 2245e2e58d0fSFilipe Manana ret = search_for_key_slot(b, 0, key, prev_cmp, &slot); 2246e2e58d0fSFilipe Manana if (ret < 0) 2247e2e58d0fSFilipe Manana goto done; 2248e2e58d0fSFilipe Manana prev_cmp = ret; 2249e2e58d0fSFilipe Manana 2250f624d976SQu Wenruo if (ret && slot > 0) { 2251f624d976SQu Wenruo dec = 1; 2252f624d976SQu Wenruo slot--; 2253f624d976SQu Wenruo } 2254f624d976SQu Wenruo p->slots[level] = slot; 2255f624d976SQu Wenruo err = setup_nodes_for_search(trans, root, p, b, level, ins_len, 2256f624d976SQu Wenruo &write_lock_level); 2257f624d976SQu Wenruo if (err == -EAGAIN) 2258f624d976SQu Wenruo goto again; 2259f624d976SQu Wenruo if (err) { 2260f624d976SQu Wenruo ret = err; 2261f624d976SQu Wenruo goto done; 2262f624d976SQu Wenruo } 2263f624d976SQu Wenruo b = p->nodes[level]; 2264f624d976SQu Wenruo slot = p->slots[level]; 2265f624d976SQu Wenruo 2266f624d976SQu Wenruo /* 2267f624d976SQu Wenruo * Slot 0 is special, if we change the key we have to update 2268f624d976SQu Wenruo * the parent pointer which means we must have a write lock on 2269f624d976SQu Wenruo * the parent 2270f624d976SQu Wenruo */ 2271f624d976SQu Wenruo if (slot == 0 && ins_len && write_lock_level < level + 1) { 2272f624d976SQu Wenruo write_lock_level = level + 1; 2273f624d976SQu Wenruo btrfs_release_path(p); 2274f624d976SQu Wenruo goto again; 2275f624d976SQu Wenruo } 2276f624d976SQu Wenruo 2277f624d976SQu Wenruo unlock_up(p, level, lowest_unlock, min_write_lock_level, 2278f624d976SQu Wenruo &write_lock_level); 2279f624d976SQu Wenruo 2280f624d976SQu Wenruo if (level == lowest_level) { 2281f624d976SQu Wenruo if (dec) 2282f624d976SQu Wenruo p->slots[level]++; 2283f624d976SQu Wenruo goto done; 2284f624d976SQu Wenruo } 2285f624d976SQu Wenruo 2286f624d976SQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 2287f624d976SQu Wenruo if (err == -EAGAIN) 2288f624d976SQu Wenruo goto again; 2289f624d976SQu Wenruo if (err) { 2290f624d976SQu Wenruo ret = err; 2291f624d976SQu Wenruo goto done; 2292f624d976SQu Wenruo } 2293f624d976SQu Wenruo 2294f624d976SQu Wenruo if (!p->skip_locking) { 2295f624d976SQu Wenruo level = btrfs_header_level(b); 2296b40130b2SJosef Bacik 2297b40130b2SJosef Bacik btrfs_maybe_reset_lockdep_class(root, b); 2298b40130b2SJosef Bacik 2299f624d976SQu Wenruo if (level <= write_lock_level) { 2300f624d976SQu Wenruo btrfs_tree_lock(b); 2301f624d976SQu Wenruo p->locks[level] = BTRFS_WRITE_LOCK; 2302f624d976SQu Wenruo } else { 2303857bc13fSJosef Bacik if (p->nowait) { 2304857bc13fSJosef Bacik if (!btrfs_try_tree_read_lock(b)) { 2305857bc13fSJosef Bacik free_extent_buffer(b); 2306857bc13fSJosef Bacik ret = -EAGAIN; 2307857bc13fSJosef Bacik goto done; 2308857bc13fSJosef Bacik } 2309857bc13fSJosef Bacik } else { 2310fe596ca3SJosef Bacik btrfs_tree_read_lock(b); 2311857bc13fSJosef Bacik } 2312f624d976SQu Wenruo p->locks[level] = BTRFS_READ_LOCK; 2313f624d976SQu Wenruo } 2314f624d976SQu Wenruo p->nodes[level] = b; 2315f624d976SQu Wenruo } 231665b51a00SChris Mason } 231765b51a00SChris Mason ret = 1; 231865b51a00SChris Mason done: 23195f5bc6b1SFilipe Manana if (ret < 0 && !p->skip_release_on_error) 2320b3b4aa74SDavid Sterba btrfs_release_path(p); 2321d96b3424SFilipe Manana 2322d96b3424SFilipe Manana if (p->need_commit_sem) { 2323d96b3424SFilipe Manana int ret2; 2324d96b3424SFilipe Manana 2325d96b3424SFilipe Manana ret2 = finish_need_commit_sem_search(p); 2326d96b3424SFilipe Manana up_read(&fs_info->commit_root_sem); 2327d96b3424SFilipe Manana if (ret2) 2328d96b3424SFilipe Manana ret = ret2; 2329d96b3424SFilipe Manana } 2330d96b3424SFilipe Manana 2331be0e5c09SChris Mason return ret; 2332be0e5c09SChris Mason } 2333f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); 2334be0e5c09SChris Mason 233574123bd7SChris Mason /* 23365d9e75c4SJan Schmidt * Like btrfs_search_slot, this looks for a key in the given tree. It uses the 23375d9e75c4SJan Schmidt * current state of the tree together with the operations recorded in the tree 23385d9e75c4SJan Schmidt * modification log to search for the key in a previous version of this tree, as 23395d9e75c4SJan Schmidt * denoted by the time_seq parameter. 23405d9e75c4SJan Schmidt * 23415d9e75c4SJan Schmidt * Naturally, there is no support for insert, delete or cow operations. 23425d9e75c4SJan Schmidt * 23435d9e75c4SJan Schmidt * The resulting path and return value will be set up as if we called 23445d9e75c4SJan Schmidt * btrfs_search_slot at that point in time with ins_len and cow both set to 0. 23455d9e75c4SJan Schmidt */ 2346310712b2SOmar Sandoval int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, 23475d9e75c4SJan Schmidt struct btrfs_path *p, u64 time_seq) 23485d9e75c4SJan Schmidt { 23490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23505d9e75c4SJan Schmidt struct extent_buffer *b; 23515d9e75c4SJan Schmidt int slot; 23525d9e75c4SJan Schmidt int ret; 23535d9e75c4SJan Schmidt int err; 23545d9e75c4SJan Schmidt int level; 23555d9e75c4SJan Schmidt int lowest_unlock = 1; 23565d9e75c4SJan Schmidt u8 lowest_level = 0; 23575d9e75c4SJan Schmidt 23585d9e75c4SJan Schmidt lowest_level = p->lowest_level; 23595d9e75c4SJan Schmidt WARN_ON(p->nodes[0] != NULL); 2360c922b016SStefan Roesch ASSERT(!p->nowait); 23615d9e75c4SJan Schmidt 23625d9e75c4SJan Schmidt if (p->search_commit_root) { 23635d9e75c4SJan Schmidt BUG_ON(time_seq); 23645d9e75c4SJan Schmidt return btrfs_search_slot(NULL, root, key, p, 0, 0); 23655d9e75c4SJan Schmidt } 23665d9e75c4SJan Schmidt 23675d9e75c4SJan Schmidt again: 2368f3a84ccdSFilipe Manana b = btrfs_get_old_root(root, time_seq); 2369315bed43SNikolay Borisov if (!b) { 2370315bed43SNikolay Borisov ret = -EIO; 2371315bed43SNikolay Borisov goto done; 2372315bed43SNikolay Borisov } 23735d9e75c4SJan Schmidt level = btrfs_header_level(b); 23745d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 23755d9e75c4SJan Schmidt 23765d9e75c4SJan Schmidt while (b) { 2377abe9339dSQu Wenruo int dec = 0; 2378abe9339dSQu Wenruo 23795d9e75c4SJan Schmidt level = btrfs_header_level(b); 23805d9e75c4SJan Schmidt p->nodes[level] = b; 23815d9e75c4SJan Schmidt 23825d9e75c4SJan Schmidt /* 23835d9e75c4SJan Schmidt * we have a lock on b and as long as we aren't changing 23845d9e75c4SJan Schmidt * the tree, there is no way to for the items in b to change. 23855d9e75c4SJan Schmidt * It is safe to drop the lock on our parent before we 23865d9e75c4SJan Schmidt * go through the expensive btree search on b. 23875d9e75c4SJan Schmidt */ 23885d9e75c4SJan Schmidt btrfs_unlock_up_safe(p, level + 1); 23895d9e75c4SJan Schmidt 2390fdf8d595SAnand Jain ret = btrfs_bin_search(b, 0, key, &slot); 2391cbca7d59SFilipe Manana if (ret < 0) 2392cbca7d59SFilipe Manana goto done; 23935d9e75c4SJan Schmidt 2394abe9339dSQu Wenruo if (level == 0) { 2395abe9339dSQu Wenruo p->slots[level] = slot; 2396abe9339dSQu Wenruo unlock_up(p, level, lowest_unlock, 0, NULL); 2397abe9339dSQu Wenruo goto done; 2398abe9339dSQu Wenruo } 2399abe9339dSQu Wenruo 24005d9e75c4SJan Schmidt if (ret && slot > 0) { 24015d9e75c4SJan Schmidt dec = 1; 2402abe9339dSQu Wenruo slot--; 24035d9e75c4SJan Schmidt } 24045d9e75c4SJan Schmidt p->slots[level] = slot; 24055d9e75c4SJan Schmidt unlock_up(p, level, lowest_unlock, 0, NULL); 24065d9e75c4SJan Schmidt 24075d9e75c4SJan Schmidt if (level == lowest_level) { 24085d9e75c4SJan Schmidt if (dec) 24095d9e75c4SJan Schmidt p->slots[level]++; 24105d9e75c4SJan Schmidt goto done; 24115d9e75c4SJan Schmidt } 24125d9e75c4SJan Schmidt 2413abe9339dSQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 24145d9e75c4SJan Schmidt if (err == -EAGAIN) 24155d9e75c4SJan Schmidt goto again; 24165d9e75c4SJan Schmidt if (err) { 24175d9e75c4SJan Schmidt ret = err; 24185d9e75c4SJan Schmidt goto done; 24195d9e75c4SJan Schmidt } 24205d9e75c4SJan Schmidt 24215d9e75c4SJan Schmidt level = btrfs_header_level(b); 24225d9e75c4SJan Schmidt btrfs_tree_read_lock(b); 2423f3a84ccdSFilipe Manana b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq); 2424db7f3436SJosef Bacik if (!b) { 2425db7f3436SJosef Bacik ret = -ENOMEM; 2426db7f3436SJosef Bacik goto done; 2427db7f3436SJosef Bacik } 24285d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 24295d9e75c4SJan Schmidt p->nodes[level] = b; 24305d9e75c4SJan Schmidt } 24315d9e75c4SJan Schmidt ret = 1; 24325d9e75c4SJan Schmidt done: 24335d9e75c4SJan Schmidt if (ret < 0) 24345d9e75c4SJan Schmidt btrfs_release_path(p); 24355d9e75c4SJan Schmidt 24365d9e75c4SJan Schmidt return ret; 24375d9e75c4SJan Schmidt } 24385d9e75c4SJan Schmidt 24395d9e75c4SJan Schmidt /* 2440f469c8bdSFilipe Manana * Search the tree again to find a leaf with smaller keys. 2441f469c8bdSFilipe Manana * Returns 0 if it found something. 2442f469c8bdSFilipe Manana * Returns 1 if there are no smaller keys. 2443f469c8bdSFilipe Manana * Returns < 0 on error. 2444f469c8bdSFilipe Manana * 2445f469c8bdSFilipe Manana * This may release the path, and so you may lose any locks held at the 2446f469c8bdSFilipe Manana * time you call it. 2447f469c8bdSFilipe Manana */ 2448f469c8bdSFilipe Manana static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) 2449f469c8bdSFilipe Manana { 2450f469c8bdSFilipe Manana struct btrfs_key key; 2451f469c8bdSFilipe Manana struct btrfs_key orig_key; 2452f469c8bdSFilipe Manana struct btrfs_disk_key found_key; 2453f469c8bdSFilipe Manana int ret; 2454f469c8bdSFilipe Manana 2455f469c8bdSFilipe Manana btrfs_item_key_to_cpu(path->nodes[0], &key, 0); 2456f469c8bdSFilipe Manana orig_key = key; 2457f469c8bdSFilipe Manana 2458f469c8bdSFilipe Manana if (key.offset > 0) { 2459f469c8bdSFilipe Manana key.offset--; 2460f469c8bdSFilipe Manana } else if (key.type > 0) { 2461f469c8bdSFilipe Manana key.type--; 2462f469c8bdSFilipe Manana key.offset = (u64)-1; 2463f469c8bdSFilipe Manana } else if (key.objectid > 0) { 2464f469c8bdSFilipe Manana key.objectid--; 2465f469c8bdSFilipe Manana key.type = (u8)-1; 2466f469c8bdSFilipe Manana key.offset = (u64)-1; 2467f469c8bdSFilipe Manana } else { 2468f469c8bdSFilipe Manana return 1; 2469f469c8bdSFilipe Manana } 2470f469c8bdSFilipe Manana 2471f469c8bdSFilipe Manana btrfs_release_path(path); 2472f469c8bdSFilipe Manana ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2473f469c8bdSFilipe Manana if (ret <= 0) 2474f469c8bdSFilipe Manana return ret; 2475f469c8bdSFilipe Manana 2476f469c8bdSFilipe Manana /* 2477f469c8bdSFilipe Manana * Previous key not found. Even if we were at slot 0 of the leaf we had 2478f469c8bdSFilipe Manana * before releasing the path and calling btrfs_search_slot(), we now may 2479f469c8bdSFilipe Manana * be in a slot pointing to the same original key - this can happen if 2480f469c8bdSFilipe Manana * after we released the path, one of more items were moved from a 2481f469c8bdSFilipe Manana * sibling leaf into the front of the leaf we had due to an insertion 2482f469c8bdSFilipe Manana * (see push_leaf_right()). 2483f469c8bdSFilipe Manana * If we hit this case and our slot is > 0 and just decrement the slot 2484f469c8bdSFilipe Manana * so that the caller does not process the same key again, which may or 2485f469c8bdSFilipe Manana * may not break the caller, depending on its logic. 2486f469c8bdSFilipe Manana */ 2487f469c8bdSFilipe Manana if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { 2488f469c8bdSFilipe Manana btrfs_item_key(path->nodes[0], &found_key, path->slots[0]); 2489f469c8bdSFilipe Manana ret = comp_keys(&found_key, &orig_key); 2490f469c8bdSFilipe Manana if (ret == 0) { 2491f469c8bdSFilipe Manana if (path->slots[0] > 0) { 2492f469c8bdSFilipe Manana path->slots[0]--; 2493f469c8bdSFilipe Manana return 0; 2494f469c8bdSFilipe Manana } 2495f469c8bdSFilipe Manana /* 2496f469c8bdSFilipe Manana * At slot 0, same key as before, it means orig_key is 2497f469c8bdSFilipe Manana * the lowest, leftmost, key in the tree. We're done. 2498f469c8bdSFilipe Manana */ 2499f469c8bdSFilipe Manana return 1; 2500f469c8bdSFilipe Manana } 2501f469c8bdSFilipe Manana } 2502f469c8bdSFilipe Manana 2503f469c8bdSFilipe Manana btrfs_item_key(path->nodes[0], &found_key, 0); 2504f469c8bdSFilipe Manana ret = comp_keys(&found_key, &key); 2505f469c8bdSFilipe Manana /* 2506f469c8bdSFilipe Manana * We might have had an item with the previous key in the tree right 2507f469c8bdSFilipe Manana * before we released our path. And after we released our path, that 2508f469c8bdSFilipe Manana * item might have been pushed to the first slot (0) of the leaf we 2509f469c8bdSFilipe Manana * were holding due to a tree balance. Alternatively, an item with the 2510f469c8bdSFilipe Manana * previous key can exist as the only element of a leaf (big fat item). 2511f469c8bdSFilipe Manana * Therefore account for these 2 cases, so that our callers (like 2512f469c8bdSFilipe Manana * btrfs_previous_item) don't miss an existing item with a key matching 2513f469c8bdSFilipe Manana * the previous key we computed above. 2514f469c8bdSFilipe Manana */ 2515f469c8bdSFilipe Manana if (ret <= 0) 2516f469c8bdSFilipe Manana return 0; 2517f469c8bdSFilipe Manana return 1; 2518f469c8bdSFilipe Manana } 2519f469c8bdSFilipe Manana 2520f469c8bdSFilipe Manana /* 25212f38b3e1SArne Jansen * helper to use instead of search slot if no exact match is needed but 25222f38b3e1SArne Jansen * instead the next or previous item should be returned. 25232f38b3e1SArne Jansen * When find_higher is true, the next higher item is returned, the next lower 25242f38b3e1SArne Jansen * otherwise. 25252f38b3e1SArne Jansen * When return_any and find_higher are both true, and no higher item is found, 25262f38b3e1SArne Jansen * return the next lower instead. 25272f38b3e1SArne Jansen * When return_any is true and find_higher is false, and no lower item is found, 25282f38b3e1SArne Jansen * return the next higher instead. 25292f38b3e1SArne Jansen * It returns 0 if any item is found, 1 if none is found (tree empty), and 25302f38b3e1SArne Jansen * < 0 on error 25312f38b3e1SArne Jansen */ 25322f38b3e1SArne Jansen int btrfs_search_slot_for_read(struct btrfs_root *root, 2533310712b2SOmar Sandoval const struct btrfs_key *key, 2534310712b2SOmar Sandoval struct btrfs_path *p, int find_higher, 2535310712b2SOmar Sandoval int return_any) 25362f38b3e1SArne Jansen { 25372f38b3e1SArne Jansen int ret; 25382f38b3e1SArne Jansen struct extent_buffer *leaf; 25392f38b3e1SArne Jansen 25402f38b3e1SArne Jansen again: 25412f38b3e1SArne Jansen ret = btrfs_search_slot(NULL, root, key, p, 0, 0); 25422f38b3e1SArne Jansen if (ret <= 0) 25432f38b3e1SArne Jansen return ret; 25442f38b3e1SArne Jansen /* 25452f38b3e1SArne Jansen * a return value of 1 means the path is at the position where the 25462f38b3e1SArne Jansen * item should be inserted. Normally this is the next bigger item, 25472f38b3e1SArne Jansen * but in case the previous item is the last in a leaf, path points 25482f38b3e1SArne Jansen * to the first free slot in the previous leaf, i.e. at an invalid 25492f38b3e1SArne Jansen * item. 25502f38b3e1SArne Jansen */ 25512f38b3e1SArne Jansen leaf = p->nodes[0]; 25522f38b3e1SArne Jansen 25532f38b3e1SArne Jansen if (find_higher) { 25542f38b3e1SArne Jansen if (p->slots[0] >= btrfs_header_nritems(leaf)) { 25552f38b3e1SArne Jansen ret = btrfs_next_leaf(root, p); 25562f38b3e1SArne Jansen if (ret <= 0) 25572f38b3e1SArne Jansen return ret; 25582f38b3e1SArne Jansen if (!return_any) 25592f38b3e1SArne Jansen return 1; 25602f38b3e1SArne Jansen /* 25612f38b3e1SArne Jansen * no higher item found, return the next 25622f38b3e1SArne Jansen * lower instead 25632f38b3e1SArne Jansen */ 25642f38b3e1SArne Jansen return_any = 0; 25652f38b3e1SArne Jansen find_higher = 0; 25662f38b3e1SArne Jansen btrfs_release_path(p); 25672f38b3e1SArne Jansen goto again; 25682f38b3e1SArne Jansen } 25692f38b3e1SArne Jansen } else { 25702f38b3e1SArne Jansen if (p->slots[0] == 0) { 25712f38b3e1SArne Jansen ret = btrfs_prev_leaf(root, p); 2572e6793769SArne Jansen if (ret < 0) 25732f38b3e1SArne Jansen return ret; 2574e6793769SArne Jansen if (!ret) { 257523c6bf6aSFilipe David Borba Manana leaf = p->nodes[0]; 257623c6bf6aSFilipe David Borba Manana if (p->slots[0] == btrfs_header_nritems(leaf)) 257723c6bf6aSFilipe David Borba Manana p->slots[0]--; 2578e6793769SArne Jansen return 0; 2579e6793769SArne Jansen } 25802f38b3e1SArne Jansen if (!return_any) 25812f38b3e1SArne Jansen return 1; 25822f38b3e1SArne Jansen /* 25832f38b3e1SArne Jansen * no lower item found, return the next 25842f38b3e1SArne Jansen * higher instead 25852f38b3e1SArne Jansen */ 25862f38b3e1SArne Jansen return_any = 0; 25872f38b3e1SArne Jansen find_higher = 1; 25882f38b3e1SArne Jansen btrfs_release_path(p); 25892f38b3e1SArne Jansen goto again; 2590e6793769SArne Jansen } else { 25912f38b3e1SArne Jansen --p->slots[0]; 25922f38b3e1SArne Jansen } 25932f38b3e1SArne Jansen } 25942f38b3e1SArne Jansen return 0; 25952f38b3e1SArne Jansen } 25962f38b3e1SArne Jansen 25972f38b3e1SArne Jansen /* 25980ff40a91SMarcos Paulo de Souza * Execute search and call btrfs_previous_item to traverse backwards if the item 25990ff40a91SMarcos Paulo de Souza * was not found. 26000ff40a91SMarcos Paulo de Souza * 26010ff40a91SMarcos Paulo de Souza * Return 0 if found, 1 if not found and < 0 if error. 26020ff40a91SMarcos Paulo de Souza */ 26030ff40a91SMarcos Paulo de Souza int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key, 26040ff40a91SMarcos Paulo de Souza struct btrfs_path *path) 26050ff40a91SMarcos Paulo de Souza { 26060ff40a91SMarcos Paulo de Souza int ret; 26070ff40a91SMarcos Paulo de Souza 26080ff40a91SMarcos Paulo de Souza ret = btrfs_search_slot(NULL, root, key, path, 0, 0); 26090ff40a91SMarcos Paulo de Souza if (ret > 0) 26100ff40a91SMarcos Paulo de Souza ret = btrfs_previous_item(root, path, key->objectid, key->type); 26110ff40a91SMarcos Paulo de Souza 26120ff40a91SMarcos Paulo de Souza if (ret == 0) 26130ff40a91SMarcos Paulo de Souza btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); 26140ff40a91SMarcos Paulo de Souza 26150ff40a91SMarcos Paulo de Souza return ret; 26160ff40a91SMarcos Paulo de Souza } 26170ff40a91SMarcos Paulo de Souza 261843dd529aSDavid Sterba /* 261962142be3SGabriel Niebler * Search for a valid slot for the given path. 262062142be3SGabriel Niebler * 262162142be3SGabriel Niebler * @root: The root node of the tree. 262262142be3SGabriel Niebler * @key: Will contain a valid item if found. 262362142be3SGabriel Niebler * @path: The starting point to validate the slot. 262462142be3SGabriel Niebler * 262562142be3SGabriel Niebler * Return: 0 if the item is valid 262662142be3SGabriel Niebler * 1 if not found 262762142be3SGabriel Niebler * <0 if error. 262862142be3SGabriel Niebler */ 262962142be3SGabriel Niebler int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key, 263062142be3SGabriel Niebler struct btrfs_path *path) 263162142be3SGabriel Niebler { 2632524f14bbSFilipe Manana if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 263362142be3SGabriel Niebler int ret; 263462142be3SGabriel Niebler 263562142be3SGabriel Niebler ret = btrfs_next_leaf(root, path); 263662142be3SGabriel Niebler if (ret) 263762142be3SGabriel Niebler return ret; 263862142be3SGabriel Niebler } 2639524f14bbSFilipe Manana 2640524f14bbSFilipe Manana btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); 264162142be3SGabriel Niebler return 0; 264262142be3SGabriel Niebler } 264362142be3SGabriel Niebler 26440ff40a91SMarcos Paulo de Souza /* 264574123bd7SChris Mason * adjust the pointers going up the tree, starting at level 264674123bd7SChris Mason * making sure the right key of each node is points to 'key'. 264774123bd7SChris Mason * This is used after shifting pointers to the left, so it stops 264874123bd7SChris Mason * fixing up pointers when a given leaf/node is not in slot 0 of the 264974123bd7SChris Mason * higher levels 2650aa5d6bedSChris Mason * 265174123bd7SChris Mason */ 2652b167fa91SNikolay Borisov static void fixup_low_keys(struct btrfs_path *path, 26535f39d397SChris Mason struct btrfs_disk_key *key, int level) 2654be0e5c09SChris Mason { 2655be0e5c09SChris Mason int i; 26565f39d397SChris Mason struct extent_buffer *t; 26570e82bcfeSDavid Sterba int ret; 26585f39d397SChris Mason 2659234b63a0SChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 2660be0e5c09SChris Mason int tslot = path->slots[i]; 26610e82bcfeSDavid Sterba 2662eb60ceacSChris Mason if (!path->nodes[i]) 2663be0e5c09SChris Mason break; 26645f39d397SChris Mason t = path->nodes[i]; 2665f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(t, tslot, 266633cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REPLACE); 26670e82bcfeSDavid Sterba BUG_ON(ret < 0); 26685f39d397SChris Mason btrfs_set_node_key(t, key, tslot); 2669d6025579SChris Mason btrfs_mark_buffer_dirty(path->nodes[i]); 2670be0e5c09SChris Mason if (tslot != 0) 2671be0e5c09SChris Mason break; 2672be0e5c09SChris Mason } 2673be0e5c09SChris Mason } 2674be0e5c09SChris Mason 267574123bd7SChris Mason /* 267631840ae1SZheng Yan * update item key. 267731840ae1SZheng Yan * 267831840ae1SZheng Yan * This function isn't completely safe. It's the caller's responsibility 267931840ae1SZheng Yan * that the new key won't break the order 268031840ae1SZheng Yan */ 2681b7a0365eSDaniel Dressler void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, 2682b7a0365eSDaniel Dressler struct btrfs_path *path, 2683310712b2SOmar Sandoval const struct btrfs_key *new_key) 268431840ae1SZheng Yan { 268531840ae1SZheng Yan struct btrfs_disk_key disk_key; 268631840ae1SZheng Yan struct extent_buffer *eb; 268731840ae1SZheng Yan int slot; 268831840ae1SZheng Yan 268931840ae1SZheng Yan eb = path->nodes[0]; 269031840ae1SZheng Yan slot = path->slots[0]; 269131840ae1SZheng Yan if (slot > 0) { 269231840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot - 1); 26937c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { 2694eee3b811SQu Wenruo btrfs_print_leaf(eb); 26957c15d410SQu Wenruo btrfs_crit(fs_info, 26967c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 26977c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 26987c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 26997c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 27007c15d410SQu Wenruo new_key->objectid, new_key->type, 27017c15d410SQu Wenruo new_key->offset); 27027c15d410SQu Wenruo BUG(); 27037c15d410SQu Wenruo } 270431840ae1SZheng Yan } 270531840ae1SZheng Yan if (slot < btrfs_header_nritems(eb) - 1) { 270631840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot + 1); 27077c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { 2708eee3b811SQu Wenruo btrfs_print_leaf(eb); 27097c15d410SQu Wenruo btrfs_crit(fs_info, 27107c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 27117c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 27127c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 27137c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 27147c15d410SQu Wenruo new_key->objectid, new_key->type, 27157c15d410SQu Wenruo new_key->offset); 27167c15d410SQu Wenruo BUG(); 27177c15d410SQu Wenruo } 271831840ae1SZheng Yan } 271931840ae1SZheng Yan 272031840ae1SZheng Yan btrfs_cpu_key_to_disk(&disk_key, new_key); 272131840ae1SZheng Yan btrfs_set_item_key(eb, &disk_key, slot); 272231840ae1SZheng Yan btrfs_mark_buffer_dirty(eb); 272331840ae1SZheng Yan if (slot == 0) 2724b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 272531840ae1SZheng Yan } 272631840ae1SZheng Yan 272731840ae1SZheng Yan /* 2728d16c702fSQu Wenruo * Check key order of two sibling extent buffers. 2729d16c702fSQu Wenruo * 2730d16c702fSQu Wenruo * Return true if something is wrong. 2731d16c702fSQu Wenruo * Return false if everything is fine. 2732d16c702fSQu Wenruo * 2733d16c702fSQu Wenruo * Tree-checker only works inside one tree block, thus the following 2734d16c702fSQu Wenruo * corruption can not be detected by tree-checker: 2735d16c702fSQu Wenruo * 2736d16c702fSQu Wenruo * Leaf @left | Leaf @right 2737d16c702fSQu Wenruo * -------------------------------------------------------------- 2738d16c702fSQu Wenruo * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | 2739d16c702fSQu Wenruo * 2740d16c702fSQu Wenruo * Key f6 in leaf @left itself is valid, but not valid when the next 2741d16c702fSQu Wenruo * key in leaf @right is 7. 2742d16c702fSQu Wenruo * This can only be checked at tree block merge time. 2743d16c702fSQu Wenruo * And since tree checker has ensured all key order in each tree block 2744d16c702fSQu Wenruo * is correct, we only need to bother the last key of @left and the first 2745d16c702fSQu Wenruo * key of @right. 2746d16c702fSQu Wenruo */ 2747d16c702fSQu Wenruo static bool check_sibling_keys(struct extent_buffer *left, 2748d16c702fSQu Wenruo struct extent_buffer *right) 2749d16c702fSQu Wenruo { 2750d16c702fSQu Wenruo struct btrfs_key left_last; 2751d16c702fSQu Wenruo struct btrfs_key right_first; 2752d16c702fSQu Wenruo int level = btrfs_header_level(left); 2753d16c702fSQu Wenruo int nr_left = btrfs_header_nritems(left); 2754d16c702fSQu Wenruo int nr_right = btrfs_header_nritems(right); 2755d16c702fSQu Wenruo 2756d16c702fSQu Wenruo /* No key to check in one of the tree blocks */ 2757d16c702fSQu Wenruo if (!nr_left || !nr_right) 2758d16c702fSQu Wenruo return false; 2759d16c702fSQu Wenruo 2760d16c702fSQu Wenruo if (level) { 2761d16c702fSQu Wenruo btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); 2762d16c702fSQu Wenruo btrfs_node_key_to_cpu(right, &right_first, 0); 2763d16c702fSQu Wenruo } else { 2764d16c702fSQu Wenruo btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); 2765d16c702fSQu Wenruo btrfs_item_key_to_cpu(right, &right_first, 0); 2766d16c702fSQu Wenruo } 2767d16c702fSQu Wenruo 276888ad95b0SFilipe Manana if (unlikely(btrfs_comp_cpu_keys(&left_last, &right_first) >= 0)) { 2769a2cea677SFilipe Manana btrfs_crit(left->fs_info, "left extent buffer:"); 2770a2cea677SFilipe Manana btrfs_print_tree(left, false); 2771a2cea677SFilipe Manana btrfs_crit(left->fs_info, "right extent buffer:"); 2772a2cea677SFilipe Manana btrfs_print_tree(right, false); 2773d16c702fSQu Wenruo btrfs_crit(left->fs_info, 2774d16c702fSQu Wenruo "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", 2775d16c702fSQu Wenruo left_last.objectid, left_last.type, 2776d16c702fSQu Wenruo left_last.offset, right_first.objectid, 2777d16c702fSQu Wenruo right_first.type, right_first.offset); 2778d16c702fSQu Wenruo return true; 2779d16c702fSQu Wenruo } 2780d16c702fSQu Wenruo return false; 2781d16c702fSQu Wenruo } 2782d16c702fSQu Wenruo 2783d16c702fSQu Wenruo /* 278474123bd7SChris Mason * try to push data from one node into the next node left in the 278579f95c82SChris Mason * tree. 2786aa5d6bedSChris Mason * 2787aa5d6bedSChris Mason * returns 0 if some ptrs were pushed left, < 0 if there was some horrible 2788aa5d6bedSChris Mason * error, and > 0 if there was no room in the left hand block. 278974123bd7SChris Mason */ 279098ed5174SChris Mason static int push_node_left(struct btrfs_trans_handle *trans, 27912ff7e61eSJeff Mahoney struct extent_buffer *dst, 2792971a1f66SChris Mason struct extent_buffer *src, int empty) 2793be0e5c09SChris Mason { 2794d30a668fSDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 2795be0e5c09SChris Mason int push_items = 0; 2796bb803951SChris Mason int src_nritems; 2797bb803951SChris Mason int dst_nritems; 2798aa5d6bedSChris Mason int ret = 0; 2799be0e5c09SChris Mason 28005f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 28015f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 28020b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 28037bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 28047bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 280554aa1f4dSChris Mason 2806bce4eae9SChris Mason if (!empty && src_nritems <= 8) 2807971a1f66SChris Mason return 1; 2808971a1f66SChris Mason 2809d397712bSChris Mason if (push_items <= 0) 2810be0e5c09SChris Mason return 1; 2811be0e5c09SChris Mason 2812bce4eae9SChris Mason if (empty) { 2813971a1f66SChris Mason push_items = min(src_nritems, push_items); 2814bce4eae9SChris Mason if (push_items < src_nritems) { 2815bce4eae9SChris Mason /* leave at least 8 pointers in the node if 2816bce4eae9SChris Mason * we aren't going to empty it 2817bce4eae9SChris Mason */ 2818bce4eae9SChris Mason if (src_nritems - push_items < 8) { 2819bce4eae9SChris Mason if (push_items <= 8) 2820bce4eae9SChris Mason return 1; 2821bce4eae9SChris Mason push_items -= 8; 2822bce4eae9SChris Mason } 2823bce4eae9SChris Mason } 2824bce4eae9SChris Mason } else 2825bce4eae9SChris Mason push_items = min(src_nritems - 8, push_items); 282679f95c82SChris Mason 2827d16c702fSQu Wenruo /* dst is the left eb, src is the middle eb */ 2828d16c702fSQu Wenruo if (check_sibling_keys(dst, src)) { 2829d16c702fSQu Wenruo ret = -EUCLEAN; 2830d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 2831d16c702fSQu Wenruo return ret; 2832d16c702fSQu Wenruo } 2833f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); 28345de865eeSFilipe David Borba Manana if (ret) { 283566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 28365de865eeSFilipe David Borba Manana return ret; 28375de865eeSFilipe David Borba Manana } 28385f39d397SChris Mason copy_extent_buffer(dst, src, 2839e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(dst, dst_nritems), 2840e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(src, 0), 2841123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 28425f39d397SChris Mason 2843bb803951SChris Mason if (push_items < src_nritems) { 284457911b8bSJan Schmidt /* 28455cead542SBoris Burkov * btrfs_tree_mod_log_eb_copy handles logging the move, so we 28465cead542SBoris Burkov * don't need to do an explicit tree mod log operation for it. 284757911b8bSJan Schmidt */ 2848e23efd8eSJosef Bacik memmove_extent_buffer(src, btrfs_node_key_ptr_offset(src, 0), 2849e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(src, push_items), 2850e2fa7227SChris Mason (src_nritems - push_items) * 2851123abc88SChris Mason sizeof(struct btrfs_key_ptr)); 2852bb803951SChris Mason } 28535f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 28545f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 28555f39d397SChris Mason btrfs_mark_buffer_dirty(src); 28565f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 285731840ae1SZheng Yan 2858bb803951SChris Mason return ret; 2859be0e5c09SChris Mason } 2860be0e5c09SChris Mason 286197571fd0SChris Mason /* 286279f95c82SChris Mason * try to push data from one node into the next node right in the 286379f95c82SChris Mason * tree. 286479f95c82SChris Mason * 286579f95c82SChris Mason * returns 0 if some ptrs were pushed, < 0 if there was some horrible 286679f95c82SChris Mason * error, and > 0 if there was no room in the right hand block. 286779f95c82SChris Mason * 286879f95c82SChris Mason * this will only push up to 1/2 the contents of the left node over 286979f95c82SChris Mason */ 28705f39d397SChris Mason static int balance_node_right(struct btrfs_trans_handle *trans, 28715f39d397SChris Mason struct extent_buffer *dst, 28725f39d397SChris Mason struct extent_buffer *src) 287379f95c82SChris Mason { 287455d32ed8SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 287579f95c82SChris Mason int push_items = 0; 287679f95c82SChris Mason int max_push; 287779f95c82SChris Mason int src_nritems; 287879f95c82SChris Mason int dst_nritems; 287979f95c82SChris Mason int ret = 0; 288079f95c82SChris Mason 28817bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 28827bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 28837bb86316SChris Mason 28845f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 28855f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 28860b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 2887d397712bSChris Mason if (push_items <= 0) 288879f95c82SChris Mason return 1; 2889bce4eae9SChris Mason 2890d397712bSChris Mason if (src_nritems < 4) 2891bce4eae9SChris Mason return 1; 289279f95c82SChris Mason 289379f95c82SChris Mason max_push = src_nritems / 2 + 1; 289479f95c82SChris Mason /* don't try to empty the node */ 2895d397712bSChris Mason if (max_push >= src_nritems) 289679f95c82SChris Mason return 1; 2897252c38f0SYan 289879f95c82SChris Mason if (max_push < push_items) 289979f95c82SChris Mason push_items = max_push; 290079f95c82SChris Mason 2901d16c702fSQu Wenruo /* dst is the right eb, src is the middle eb */ 2902d16c702fSQu Wenruo if (check_sibling_keys(src, dst)) { 2903d16c702fSQu Wenruo ret = -EUCLEAN; 2904d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 2905d16c702fSQu Wenruo return ret; 2906d16c702fSQu Wenruo } 29075cead542SBoris Burkov 29085cead542SBoris Burkov /* 29095cead542SBoris Burkov * btrfs_tree_mod_log_eb_copy handles logging the move, so we don't 29105cead542SBoris Burkov * need to do an explicit tree mod log operation for it. 29115cead542SBoris Burkov */ 2912e23efd8eSJosef Bacik memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(dst, push_items), 2913e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(dst, 0), 29145f39d397SChris Mason (dst_nritems) * 29155f39d397SChris Mason sizeof(struct btrfs_key_ptr)); 2916d6025579SChris Mason 2917f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, 2918ed874f0dSDavid Sterba push_items); 29195de865eeSFilipe David Borba Manana if (ret) { 292066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 29215de865eeSFilipe David Borba Manana return ret; 29225de865eeSFilipe David Borba Manana } 29235f39d397SChris Mason copy_extent_buffer(dst, src, 2924e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(dst, 0), 2925e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(src, src_nritems - push_items), 2926123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 292779f95c82SChris Mason 29285f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 29295f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 293079f95c82SChris Mason 29315f39d397SChris Mason btrfs_mark_buffer_dirty(src); 29325f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 293331840ae1SZheng Yan 293479f95c82SChris Mason return ret; 293579f95c82SChris Mason } 293679f95c82SChris Mason 293779f95c82SChris Mason /* 293897571fd0SChris Mason * helper function to insert a new root level in the tree. 293997571fd0SChris Mason * A new node is allocated, and a single item is inserted to 294097571fd0SChris Mason * point to the existing root 2941aa5d6bedSChris Mason * 2942aa5d6bedSChris Mason * returns zero on success or < 0 on failure. 294397571fd0SChris Mason */ 2944d397712bSChris Mason static noinline int insert_new_root(struct btrfs_trans_handle *trans, 29455f39d397SChris Mason struct btrfs_root *root, 2946fdd99c72SLiu Bo struct btrfs_path *path, int level) 294774123bd7SChris Mason { 29480b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 29497bb86316SChris Mason u64 lower_gen; 29505f39d397SChris Mason struct extent_buffer *lower; 29515f39d397SChris Mason struct extent_buffer *c; 2952925baeddSChris Mason struct extent_buffer *old; 29535f39d397SChris Mason struct btrfs_disk_key lower_key; 2954d9d19a01SDavid Sterba int ret; 29555c680ed6SChris Mason 29565c680ed6SChris Mason BUG_ON(path->nodes[level]); 29575c680ed6SChris Mason BUG_ON(path->nodes[level-1] != root->node); 29585c680ed6SChris Mason 29597bb86316SChris Mason lower = path->nodes[level-1]; 29607bb86316SChris Mason if (level == 1) 29617bb86316SChris Mason btrfs_item_key(lower, &lower_key, 0); 29627bb86316SChris Mason else 29637bb86316SChris Mason btrfs_node_key(lower, &lower_key, 0); 29647bb86316SChris Mason 296579bd3712SFilipe Manana c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 296679bd3712SFilipe Manana &lower_key, level, root->node->start, 0, 2967cf6f34aaSJosef Bacik BTRFS_NESTING_NEW_ROOT); 29685f39d397SChris Mason if (IS_ERR(c)) 29695f39d397SChris Mason return PTR_ERR(c); 2970925baeddSChris Mason 29710b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 2972f0486c68SYan, Zheng 29735f39d397SChris Mason btrfs_set_header_nritems(c, 1); 29745f39d397SChris Mason btrfs_set_node_key(c, &lower_key, 0); 2975db94535dSChris Mason btrfs_set_node_blockptr(c, 0, lower->start); 29767bb86316SChris Mason lower_gen = btrfs_header_generation(lower); 297731840ae1SZheng Yan WARN_ON(lower_gen != trans->transid); 29787bb86316SChris Mason 29797bb86316SChris Mason btrfs_set_node_ptr_generation(c, 0, lower_gen); 29805f39d397SChris Mason 29815f39d397SChris Mason btrfs_mark_buffer_dirty(c); 2982d5719762SChris Mason 2983925baeddSChris Mason old = root->node; 2984406808abSFilipe Manana ret = btrfs_tree_mod_log_insert_root(root->node, c, false); 2985f61aa7baSFilipe Manana if (ret < 0) { 2986f61aa7baSFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), c, 0, 1); 2987f61aa7baSFilipe Manana btrfs_tree_unlock(c); 2988f61aa7baSFilipe Manana free_extent_buffer(c); 2989f61aa7baSFilipe Manana return ret; 2990f61aa7baSFilipe Manana } 2991240f62c8SChris Mason rcu_assign_pointer(root->node, c); 2992925baeddSChris Mason 2993925baeddSChris Mason /* the super has an extra ref to root->node */ 2994925baeddSChris Mason free_extent_buffer(old); 2995925baeddSChris Mason 29960b86a832SChris Mason add_root_to_dirty_list(root); 299767439dadSDavid Sterba atomic_inc(&c->refs); 29985f39d397SChris Mason path->nodes[level] = c; 2999ac5887c8SJosef Bacik path->locks[level] = BTRFS_WRITE_LOCK; 300074123bd7SChris Mason path->slots[level] = 0; 300174123bd7SChris Mason return 0; 300274123bd7SChris Mason } 30035c680ed6SChris Mason 30045c680ed6SChris Mason /* 30055c680ed6SChris Mason * worker function to insert a single pointer in a node. 30065c680ed6SChris Mason * the node should have enough room for the pointer already 300797571fd0SChris Mason * 30085c680ed6SChris Mason * slot and level indicate where you want the key to go, and 30095c680ed6SChris Mason * blocknr is the block the key points to. 30105c680ed6SChris Mason */ 301150b5d1fcSFilipe Manana static int insert_ptr(struct btrfs_trans_handle *trans, 30126ad3cf6dSDavid Sterba struct btrfs_path *path, 3013143bede5SJeff Mahoney struct btrfs_disk_key *key, u64 bytenr, 3014c3e06965SJan Schmidt int slot, int level) 30155c680ed6SChris Mason { 30165f39d397SChris Mason struct extent_buffer *lower; 30175c680ed6SChris Mason int nritems; 3018f3ea38daSJan Schmidt int ret; 30195c680ed6SChris Mason 30205c680ed6SChris Mason BUG_ON(!path->nodes[level]); 302149d0c642SFilipe Manana btrfs_assert_tree_write_locked(path->nodes[level]); 30225f39d397SChris Mason lower = path->nodes[level]; 30235f39d397SChris Mason nritems = btrfs_header_nritems(lower); 3024c293498bSStoyan Gaydarov BUG_ON(slot > nritems); 30256ad3cf6dSDavid Sterba BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); 302674123bd7SChris Mason if (slot != nritems) { 3027bf1d3425SDavid Sterba if (level) { 3028f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_move(lower, slot + 1, 3029f3a84ccdSFilipe Manana slot, nritems - slot); 303050b5d1fcSFilipe Manana if (ret < 0) { 303150b5d1fcSFilipe Manana btrfs_abort_transaction(trans, ret); 303250b5d1fcSFilipe Manana return ret; 303350b5d1fcSFilipe Manana } 3034bf1d3425SDavid Sterba } 30355f39d397SChris Mason memmove_extent_buffer(lower, 3036e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(lower, slot + 1), 3037e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(lower, slot), 3038123abc88SChris Mason (nritems - slot) * sizeof(struct btrfs_key_ptr)); 303974123bd7SChris Mason } 3040c3e06965SJan Schmidt if (level) { 3041f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(lower, slot, 304233cff222SFilipe Manana BTRFS_MOD_LOG_KEY_ADD); 304350b5d1fcSFilipe Manana if (ret < 0) { 304450b5d1fcSFilipe Manana btrfs_abort_transaction(trans, ret); 304550b5d1fcSFilipe Manana return ret; 304650b5d1fcSFilipe Manana } 3047f3ea38daSJan Schmidt } 30485f39d397SChris Mason btrfs_set_node_key(lower, key, slot); 3049db94535dSChris Mason btrfs_set_node_blockptr(lower, slot, bytenr); 305074493f7aSChris Mason WARN_ON(trans->transid == 0); 305174493f7aSChris Mason btrfs_set_node_ptr_generation(lower, slot, trans->transid); 30525f39d397SChris Mason btrfs_set_header_nritems(lower, nritems + 1); 30535f39d397SChris Mason btrfs_mark_buffer_dirty(lower); 305450b5d1fcSFilipe Manana 305550b5d1fcSFilipe Manana return 0; 305674123bd7SChris Mason } 305774123bd7SChris Mason 305897571fd0SChris Mason /* 305997571fd0SChris Mason * split the node at the specified level in path in two. 306097571fd0SChris Mason * The path is corrected to point to the appropriate node after the split 306197571fd0SChris Mason * 306297571fd0SChris Mason * Before splitting this tries to make some room in the node by pushing 306397571fd0SChris Mason * left and right, if either one works, it returns right away. 3064aa5d6bedSChris Mason * 3065aa5d6bedSChris Mason * returns 0 on success and < 0 on failure 306697571fd0SChris Mason */ 3067e02119d5SChris Mason static noinline int split_node(struct btrfs_trans_handle *trans, 3068e02119d5SChris Mason struct btrfs_root *root, 3069e02119d5SChris Mason struct btrfs_path *path, int level) 3070be0e5c09SChris Mason { 30710b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 30725f39d397SChris Mason struct extent_buffer *c; 30735f39d397SChris Mason struct extent_buffer *split; 30745f39d397SChris Mason struct btrfs_disk_key disk_key; 3075be0e5c09SChris Mason int mid; 30765c680ed6SChris Mason int ret; 30777518a238SChris Mason u32 c_nritems; 3078be0e5c09SChris Mason 30795f39d397SChris Mason c = path->nodes[level]; 30807bb86316SChris Mason WARN_ON(btrfs_header_generation(c) != trans->transid); 30815f39d397SChris Mason if (c == root->node) { 3082d9abbf1cSJan Schmidt /* 308390f8d62eSJan Schmidt * trying to split the root, lets make a new one 308490f8d62eSJan Schmidt * 3085fdd99c72SLiu Bo * tree mod log: We don't log_removal old root in 308690f8d62eSJan Schmidt * insert_new_root, because that root buffer will be kept as a 308790f8d62eSJan Schmidt * normal node. We are going to log removal of half of the 3088f3a84ccdSFilipe Manana * elements below with btrfs_tree_mod_log_eb_copy(). We're 3089f3a84ccdSFilipe Manana * holding a tree lock on the buffer, which is why we cannot 3090f3a84ccdSFilipe Manana * race with other tree_mod_log users. 3091d9abbf1cSJan Schmidt */ 3092fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, level + 1); 30935c680ed6SChris Mason if (ret) 30945c680ed6SChris Mason return ret; 3095b3612421SChris Mason } else { 3096e66f709bSChris Mason ret = push_nodes_for_insert(trans, root, path, level); 30975f39d397SChris Mason c = path->nodes[level]; 30985f39d397SChris Mason if (!ret && btrfs_header_nritems(c) < 30990b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) 3100e66f709bSChris Mason return 0; 310154aa1f4dSChris Mason if (ret < 0) 310254aa1f4dSChris Mason return ret; 31035c680ed6SChris Mason } 3104e66f709bSChris Mason 31055f39d397SChris Mason c_nritems = btrfs_header_nritems(c); 31065d4f98a2SYan Zheng mid = (c_nritems + 1) / 2; 31075d4f98a2SYan Zheng btrfs_node_key(c, &disk_key, mid); 31087bb86316SChris Mason 310979bd3712SFilipe Manana split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 311079bd3712SFilipe Manana &disk_key, level, c->start, 0, 311179bd3712SFilipe Manana BTRFS_NESTING_SPLIT); 31125f39d397SChris Mason if (IS_ERR(split)) 31135f39d397SChris Mason return PTR_ERR(split); 311454aa1f4dSChris Mason 31150b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 3116bc877d28SNikolay Borisov ASSERT(btrfs_header_level(c) == level); 31175f39d397SChris Mason 3118f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); 31195de865eeSFilipe David Borba Manana if (ret) { 3120ede600e4SFilipe Manana btrfs_tree_unlock(split); 3121ede600e4SFilipe Manana free_extent_buffer(split); 312266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 31235de865eeSFilipe David Borba Manana return ret; 31245de865eeSFilipe David Borba Manana } 31255f39d397SChris Mason copy_extent_buffer(split, c, 3126e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(split, 0), 3127e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(c, mid), 3128123abc88SChris Mason (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); 31295f39d397SChris Mason btrfs_set_header_nritems(split, c_nritems - mid); 31305f39d397SChris Mason btrfs_set_header_nritems(c, mid); 3131aa5d6bedSChris Mason 31325f39d397SChris Mason btrfs_mark_buffer_dirty(c); 31335f39d397SChris Mason btrfs_mark_buffer_dirty(split); 31345f39d397SChris Mason 313550b5d1fcSFilipe Manana ret = insert_ptr(trans, path, &disk_key, split->start, 3136c3e06965SJan Schmidt path->slots[level + 1] + 1, level + 1); 313750b5d1fcSFilipe Manana if (ret < 0) { 313850b5d1fcSFilipe Manana btrfs_tree_unlock(split); 313950b5d1fcSFilipe Manana free_extent_buffer(split); 314050b5d1fcSFilipe Manana return ret; 314150b5d1fcSFilipe Manana } 3142aa5d6bedSChris Mason 31435de08d7dSChris Mason if (path->slots[level] >= mid) { 31445c680ed6SChris Mason path->slots[level] -= mid; 3145925baeddSChris Mason btrfs_tree_unlock(c); 31465f39d397SChris Mason free_extent_buffer(c); 31475f39d397SChris Mason path->nodes[level] = split; 31485c680ed6SChris Mason path->slots[level + 1] += 1; 3149eb60ceacSChris Mason } else { 3150925baeddSChris Mason btrfs_tree_unlock(split); 31515f39d397SChris Mason free_extent_buffer(split); 3152be0e5c09SChris Mason } 3153d5286a92SNikolay Borisov return 0; 3154be0e5c09SChris Mason } 3155be0e5c09SChris Mason 315674123bd7SChris Mason /* 315774123bd7SChris Mason * how many bytes are required to store the items in a leaf. start 315874123bd7SChris Mason * and nr indicate which items in the leaf to check. This totals up the 315974123bd7SChris Mason * space used both by the item structs and the item data 316074123bd7SChris Mason */ 31616c75a589SQu Wenruo static int leaf_space_used(const struct extent_buffer *l, int start, int nr) 3162be0e5c09SChris Mason { 3163be0e5c09SChris Mason int data_len; 31645f39d397SChris Mason int nritems = btrfs_header_nritems(l); 3165d4dbff95SChris Mason int end = min(nritems, start + nr) - 1; 3166be0e5c09SChris Mason 3167be0e5c09SChris Mason if (!nr) 3168be0e5c09SChris Mason return 0; 31693212fa14SJosef Bacik data_len = btrfs_item_offset(l, start) + btrfs_item_size(l, start); 31703212fa14SJosef Bacik data_len = data_len - btrfs_item_offset(l, end); 31710783fcfcSChris Mason data_len += sizeof(struct btrfs_item) * nr; 3172d4dbff95SChris Mason WARN_ON(data_len < 0); 3173be0e5c09SChris Mason return data_len; 3174be0e5c09SChris Mason } 3175be0e5c09SChris Mason 317674123bd7SChris Mason /* 3177d4dbff95SChris Mason * The space between the end of the leaf items and 3178d4dbff95SChris Mason * the start of the leaf data. IOW, how much room 3179d4dbff95SChris Mason * the leaf has left for both items and data 3180d4dbff95SChris Mason */ 31816c75a589SQu Wenruo int btrfs_leaf_free_space(const struct extent_buffer *leaf) 3182d4dbff95SChris Mason { 3183e902baacSDavid Sterba struct btrfs_fs_info *fs_info = leaf->fs_info; 31845f39d397SChris Mason int nritems = btrfs_header_nritems(leaf); 31855f39d397SChris Mason int ret; 31860b246afaSJeff Mahoney 31870b246afaSJeff Mahoney ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); 31885f39d397SChris Mason if (ret < 0) { 31890b246afaSJeff Mahoney btrfs_crit(fs_info, 3190efe120a0SFrank Holton "leaf free space ret %d, leaf data size %lu, used %d nritems %d", 3191da17066cSJeff Mahoney ret, 31920b246afaSJeff Mahoney (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), 31935f39d397SChris Mason leaf_space_used(leaf, 0, nritems), nritems); 31945f39d397SChris Mason } 31955f39d397SChris Mason return ret; 3196d4dbff95SChris Mason } 3197d4dbff95SChris Mason 319899d8f83cSChris Mason /* 319999d8f83cSChris Mason * min slot controls the lowest index we're willing to push to the 320099d8f83cSChris Mason * right. We'll push up to and including min_slot, but no lower 320199d8f83cSChris Mason */ 3202ed25dab3SJosef Bacik static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, 3203ed25dab3SJosef Bacik struct btrfs_path *path, 320444871b1bSChris Mason int data_size, int empty, 320544871b1bSChris Mason struct extent_buffer *right, 320699d8f83cSChris Mason int free_space, u32 left_nritems, 320799d8f83cSChris Mason u32 min_slot) 320800ec4c51SChris Mason { 3209f72f0010SDavid Sterba struct btrfs_fs_info *fs_info = right->fs_info; 32105f39d397SChris Mason struct extent_buffer *left = path->nodes[0]; 321144871b1bSChris Mason struct extent_buffer *upper = path->nodes[1]; 3212cfed81a0SChris Mason struct btrfs_map_token token; 32135f39d397SChris Mason struct btrfs_disk_key disk_key; 321400ec4c51SChris Mason int slot; 321534a38218SChris Mason u32 i; 321600ec4c51SChris Mason int push_space = 0; 321700ec4c51SChris Mason int push_items = 0; 321834a38218SChris Mason u32 nr; 32197518a238SChris Mason u32 right_nritems; 32205f39d397SChris Mason u32 data_end; 3221db94535dSChris Mason u32 this_item_size; 322200ec4c51SChris Mason 322334a38218SChris Mason if (empty) 322434a38218SChris Mason nr = 0; 322534a38218SChris Mason else 322699d8f83cSChris Mason nr = max_t(u32, 1, min_slot); 322734a38218SChris Mason 322831840ae1SZheng Yan if (path->slots[0] >= left_nritems) 322987b29b20SYan Zheng push_space += data_size; 323031840ae1SZheng Yan 323144871b1bSChris Mason slot = path->slots[1]; 323234a38218SChris Mason i = left_nritems - 1; 323334a38218SChris Mason while (i >= nr) { 323431840ae1SZheng Yan if (!empty && push_items > 0) { 323531840ae1SZheng Yan if (path->slots[0] > i) 323631840ae1SZheng Yan break; 323731840ae1SZheng Yan if (path->slots[0] == i) { 3238e902baacSDavid Sterba int space = btrfs_leaf_free_space(left); 3239e902baacSDavid Sterba 324031840ae1SZheng Yan if (space + push_space * 2 > free_space) 324131840ae1SZheng Yan break; 324231840ae1SZheng Yan } 324331840ae1SZheng Yan } 324431840ae1SZheng Yan 324500ec4c51SChris Mason if (path->slots[0] == i) 324687b29b20SYan Zheng push_space += data_size; 3247db94535dSChris Mason 32483212fa14SJosef Bacik this_item_size = btrfs_item_size(left, i); 324974794207SJosef Bacik if (this_item_size + sizeof(struct btrfs_item) + 325074794207SJosef Bacik push_space > free_space) 325100ec4c51SChris Mason break; 325231840ae1SZheng Yan 325300ec4c51SChris Mason push_items++; 325474794207SJosef Bacik push_space += this_item_size + sizeof(struct btrfs_item); 325534a38218SChris Mason if (i == 0) 325634a38218SChris Mason break; 325734a38218SChris Mason i--; 3258db94535dSChris Mason } 32595f39d397SChris Mason 3260925baeddSChris Mason if (push_items == 0) 3261925baeddSChris Mason goto out_unlock; 32625f39d397SChris Mason 32636c1500f2SJulia Lawall WARN_ON(!empty && push_items == left_nritems); 32645f39d397SChris Mason 326500ec4c51SChris Mason /* push left to right */ 32665f39d397SChris Mason right_nritems = btrfs_header_nritems(right); 326734a38218SChris Mason 3268dc2e724eSJosef Bacik push_space = btrfs_item_data_end(left, left_nritems - push_items); 32698f881e8cSDavid Sterba push_space -= leaf_data_end(left); 32705f39d397SChris Mason 327100ec4c51SChris Mason /* make room in the right data area */ 32728f881e8cSDavid Sterba data_end = leaf_data_end(right); 3273637e3b48SJosef Bacik memmove_leaf_data(right, data_end - push_space, data_end, 32740b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); 32755f39d397SChris Mason 327600ec4c51SChris Mason /* copy from the left data area */ 3277637e3b48SJosef Bacik copy_leaf_data(right, left, BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 3278637e3b48SJosef Bacik leaf_data_end(left), push_space); 32795f39d397SChris Mason 3280637e3b48SJosef Bacik memmove_leaf_items(right, push_items, 0, right_nritems); 32815f39d397SChris Mason 328200ec4c51SChris Mason /* copy the items from left to right */ 3283637e3b48SJosef Bacik copy_leaf_items(right, left, 0, left_nritems - push_items, push_items); 328400ec4c51SChris Mason 328500ec4c51SChris Mason /* update the item pointers */ 3286c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 32877518a238SChris Mason right_nritems += push_items; 32885f39d397SChris Mason btrfs_set_header_nritems(right, right_nritems); 32890b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 32907518a238SChris Mason for (i = 0; i < right_nritems; i++) { 32913212fa14SJosef Bacik push_space -= btrfs_token_item_size(&token, i); 32923212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, push_space); 3293db94535dSChris Mason } 3294db94535dSChris Mason 32957518a238SChris Mason left_nritems -= push_items; 32965f39d397SChris Mason btrfs_set_header_nritems(left, left_nritems); 329700ec4c51SChris Mason 329834a38218SChris Mason if (left_nritems) 32995f39d397SChris Mason btrfs_mark_buffer_dirty(left); 3300f0486c68SYan, Zheng else 3301190a8339SJosef Bacik btrfs_clear_buffer_dirty(trans, left); 3302f0486c68SYan, Zheng 33035f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3304a429e513SChris Mason 33055f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 33065f39d397SChris Mason btrfs_set_node_key(upper, &disk_key, slot + 1); 3307d6025579SChris Mason btrfs_mark_buffer_dirty(upper); 330802217ed2SChris Mason 330900ec4c51SChris Mason /* then fixup the leaf pointer in the path */ 33107518a238SChris Mason if (path->slots[0] >= left_nritems) { 33117518a238SChris Mason path->slots[0] -= left_nritems; 3312925baeddSChris Mason if (btrfs_header_nritems(path->nodes[0]) == 0) 3313190a8339SJosef Bacik btrfs_clear_buffer_dirty(trans, path->nodes[0]); 3314925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 33155f39d397SChris Mason free_extent_buffer(path->nodes[0]); 33165f39d397SChris Mason path->nodes[0] = right; 331700ec4c51SChris Mason path->slots[1] += 1; 331800ec4c51SChris Mason } else { 3319925baeddSChris Mason btrfs_tree_unlock(right); 33205f39d397SChris Mason free_extent_buffer(right); 332100ec4c51SChris Mason } 332200ec4c51SChris Mason return 0; 3323925baeddSChris Mason 3324925baeddSChris Mason out_unlock: 3325925baeddSChris Mason btrfs_tree_unlock(right); 3326925baeddSChris Mason free_extent_buffer(right); 3327925baeddSChris Mason return 1; 332800ec4c51SChris Mason } 3329925baeddSChris Mason 333000ec4c51SChris Mason /* 333144871b1bSChris Mason * push some data in the path leaf to the right, trying to free up at 333274123bd7SChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 333344871b1bSChris Mason * 333444871b1bSChris Mason * returns 1 if the push failed because the other node didn't have enough 333544871b1bSChris Mason * room, 0 if everything worked out and < 0 if there were major errors. 333699d8f83cSChris Mason * 333799d8f83cSChris Mason * this will push starting from min_slot to the end of the leaf. It won't 333899d8f83cSChris Mason * push any slot lower than min_slot 333974123bd7SChris Mason */ 334044871b1bSChris Mason static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root 334199d8f83cSChris Mason *root, struct btrfs_path *path, 334299d8f83cSChris Mason int min_data_size, int data_size, 334399d8f83cSChris Mason int empty, u32 min_slot) 3344be0e5c09SChris Mason { 334544871b1bSChris Mason struct extent_buffer *left = path->nodes[0]; 334644871b1bSChris Mason struct extent_buffer *right; 334744871b1bSChris Mason struct extent_buffer *upper; 334844871b1bSChris Mason int slot; 334944871b1bSChris Mason int free_space; 335044871b1bSChris Mason u32 left_nritems; 335144871b1bSChris Mason int ret; 335244871b1bSChris Mason 335344871b1bSChris Mason if (!path->nodes[1]) 335444871b1bSChris Mason return 1; 335544871b1bSChris Mason 335644871b1bSChris Mason slot = path->slots[1]; 335744871b1bSChris Mason upper = path->nodes[1]; 335844871b1bSChris Mason if (slot >= btrfs_header_nritems(upper) - 1) 335944871b1bSChris Mason return 1; 336044871b1bSChris Mason 336149d0c642SFilipe Manana btrfs_assert_tree_write_locked(path->nodes[1]); 336244871b1bSChris Mason 33634b231ae4SDavid Sterba right = btrfs_read_node_slot(upper, slot + 1); 3364fb770ae4SLiu Bo if (IS_ERR(right)) 33659cf14029SJosef Bacik return PTR_ERR(right); 336691ca338dSTsutomu Itoh 3367bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 336844871b1bSChris Mason 3369e902baacSDavid Sterba free_space = btrfs_leaf_free_space(right); 337044871b1bSChris Mason if (free_space < data_size) 337144871b1bSChris Mason goto out_unlock; 337244871b1bSChris Mason 337344871b1bSChris Mason ret = btrfs_cow_block(trans, root, right, upper, 3374bf59a5a2SJosef Bacik slot + 1, &right, BTRFS_NESTING_RIGHT_COW); 337544871b1bSChris Mason if (ret) 337644871b1bSChris Mason goto out_unlock; 337744871b1bSChris Mason 337844871b1bSChris Mason left_nritems = btrfs_header_nritems(left); 337944871b1bSChris Mason if (left_nritems == 0) 338044871b1bSChris Mason goto out_unlock; 338144871b1bSChris Mason 3382d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3383d16c702fSQu Wenruo ret = -EUCLEAN; 33849ae5afd0SFilipe Manana btrfs_abort_transaction(trans, ret); 3385d16c702fSQu Wenruo btrfs_tree_unlock(right); 3386d16c702fSQu Wenruo free_extent_buffer(right); 3387d16c702fSQu Wenruo return ret; 3388d16c702fSQu Wenruo } 33892ef1fed2SFilipe David Borba Manana if (path->slots[0] == left_nritems && !empty) { 33902ef1fed2SFilipe David Borba Manana /* Key greater than all keys in the leaf, right neighbor has 33912ef1fed2SFilipe David Borba Manana * enough room for it and we're not emptying our leaf to delete 33922ef1fed2SFilipe David Borba Manana * it, therefore use right neighbor to insert the new item and 339352042d8eSAndrea Gelmini * no need to touch/dirty our left leaf. */ 33942ef1fed2SFilipe David Borba Manana btrfs_tree_unlock(left); 33952ef1fed2SFilipe David Borba Manana free_extent_buffer(left); 33962ef1fed2SFilipe David Borba Manana path->nodes[0] = right; 33972ef1fed2SFilipe David Borba Manana path->slots[0] = 0; 33982ef1fed2SFilipe David Borba Manana path->slots[1]++; 33992ef1fed2SFilipe David Borba Manana return 0; 34002ef1fed2SFilipe David Borba Manana } 34012ef1fed2SFilipe David Borba Manana 3402ed25dab3SJosef Bacik return __push_leaf_right(trans, path, min_data_size, empty, right, 3403ed25dab3SJosef Bacik free_space, left_nritems, min_slot); 340444871b1bSChris Mason out_unlock: 340544871b1bSChris Mason btrfs_tree_unlock(right); 340644871b1bSChris Mason free_extent_buffer(right); 340744871b1bSChris Mason return 1; 340844871b1bSChris Mason } 340944871b1bSChris Mason 341044871b1bSChris Mason /* 341144871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 341244871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 341399d8f83cSChris Mason * 341499d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 341599d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the 341699d8f83cSChris Mason * items 341744871b1bSChris Mason */ 3418ed25dab3SJosef Bacik static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, 3419ed25dab3SJosef Bacik struct btrfs_path *path, int data_size, 342044871b1bSChris Mason int empty, struct extent_buffer *left, 342199d8f83cSChris Mason int free_space, u32 right_nritems, 342299d8f83cSChris Mason u32 max_slot) 342344871b1bSChris Mason { 34248087c193SDavid Sterba struct btrfs_fs_info *fs_info = left->fs_info; 34255f39d397SChris Mason struct btrfs_disk_key disk_key; 34265f39d397SChris Mason struct extent_buffer *right = path->nodes[0]; 3427be0e5c09SChris Mason int i; 3428be0e5c09SChris Mason int push_space = 0; 3429be0e5c09SChris Mason int push_items = 0; 34307518a238SChris Mason u32 old_left_nritems; 343134a38218SChris Mason u32 nr; 3432aa5d6bedSChris Mason int ret = 0; 3433db94535dSChris Mason u32 this_item_size; 3434db94535dSChris Mason u32 old_left_item_size; 3435cfed81a0SChris Mason struct btrfs_map_token token; 3436cfed81a0SChris Mason 343734a38218SChris Mason if (empty) 343899d8f83cSChris Mason nr = min(right_nritems, max_slot); 343934a38218SChris Mason else 344099d8f83cSChris Mason nr = min(right_nritems - 1, max_slot); 344134a38218SChris Mason 344234a38218SChris Mason for (i = 0; i < nr; i++) { 344331840ae1SZheng Yan if (!empty && push_items > 0) { 344431840ae1SZheng Yan if (path->slots[0] < i) 344531840ae1SZheng Yan break; 344631840ae1SZheng Yan if (path->slots[0] == i) { 3447e902baacSDavid Sterba int space = btrfs_leaf_free_space(right); 3448e902baacSDavid Sterba 344931840ae1SZheng Yan if (space + push_space * 2 > free_space) 345031840ae1SZheng Yan break; 345131840ae1SZheng Yan } 345231840ae1SZheng Yan } 345331840ae1SZheng Yan 3454be0e5c09SChris Mason if (path->slots[0] == i) 345587b29b20SYan Zheng push_space += data_size; 3456db94535dSChris Mason 34573212fa14SJosef Bacik this_item_size = btrfs_item_size(right, i); 345874794207SJosef Bacik if (this_item_size + sizeof(struct btrfs_item) + push_space > 345974794207SJosef Bacik free_space) 3460be0e5c09SChris Mason break; 3461db94535dSChris Mason 3462be0e5c09SChris Mason push_items++; 346374794207SJosef Bacik push_space += this_item_size + sizeof(struct btrfs_item); 3464be0e5c09SChris Mason } 3465db94535dSChris Mason 3466be0e5c09SChris Mason if (push_items == 0) { 3467925baeddSChris Mason ret = 1; 3468925baeddSChris Mason goto out; 3469be0e5c09SChris Mason } 3470fae7f21cSDulshani Gunawardhana WARN_ON(!empty && push_items == btrfs_header_nritems(right)); 34715f39d397SChris Mason 3472be0e5c09SChris Mason /* push data from right to left */ 3473637e3b48SJosef Bacik copy_leaf_items(left, right, btrfs_header_nritems(left), 0, push_items); 34745f39d397SChris Mason 34750b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - 34763212fa14SJosef Bacik btrfs_item_offset(right, push_items - 1); 34775f39d397SChris Mason 3478637e3b48SJosef Bacik copy_leaf_data(left, right, leaf_data_end(left) - push_space, 3479637e3b48SJosef Bacik btrfs_item_offset(right, push_items - 1), push_space); 34805f39d397SChris Mason old_left_nritems = btrfs_header_nritems(left); 348187b29b20SYan Zheng BUG_ON(old_left_nritems <= 0); 3482eb60ceacSChris Mason 3483c82f823cSDavid Sterba btrfs_init_map_token(&token, left); 34843212fa14SJosef Bacik old_left_item_size = btrfs_item_offset(left, old_left_nritems - 1); 3485be0e5c09SChris Mason for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { 34865f39d397SChris Mason u32 ioff; 3487db94535dSChris Mason 34883212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 34893212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, 3490cc4c13d5SDavid Sterba ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); 3491be0e5c09SChris Mason } 34925f39d397SChris Mason btrfs_set_header_nritems(left, old_left_nritems + push_items); 3493be0e5c09SChris Mason 3494be0e5c09SChris Mason /* fixup right node */ 349531b1a2bdSJulia Lawall if (push_items > right_nritems) 349631b1a2bdSJulia Lawall WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, 3497d397712bSChris Mason right_nritems); 349834a38218SChris Mason 349934a38218SChris Mason if (push_items < right_nritems) { 35003212fa14SJosef Bacik push_space = btrfs_item_offset(right, push_items - 1) - 35018f881e8cSDavid Sterba leaf_data_end(right); 3502637e3b48SJosef Bacik memmove_leaf_data(right, 35030b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 35048f881e8cSDavid Sterba leaf_data_end(right), push_space); 35055f39d397SChris Mason 3506637e3b48SJosef Bacik memmove_leaf_items(right, 0, push_items, 3507637e3b48SJosef Bacik btrfs_header_nritems(right) - push_items); 350834a38218SChris Mason } 3509c82f823cSDavid Sterba 3510c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 3511eef1c494SYan right_nritems -= push_items; 3512eef1c494SYan btrfs_set_header_nritems(right, right_nritems); 35130b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 35145f39d397SChris Mason for (i = 0; i < right_nritems; i++) { 35153212fa14SJosef Bacik push_space = push_space - btrfs_token_item_size(&token, i); 35163212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, push_space); 3517db94535dSChris Mason } 3518eb60ceacSChris Mason 35195f39d397SChris Mason btrfs_mark_buffer_dirty(left); 352034a38218SChris Mason if (right_nritems) 35215f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3522f0486c68SYan, Zheng else 3523190a8339SJosef Bacik btrfs_clear_buffer_dirty(trans, right); 3524098f59c2SChris Mason 35255f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 3526b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 3527be0e5c09SChris Mason 3528be0e5c09SChris Mason /* then fixup the leaf pointer in the path */ 3529be0e5c09SChris Mason if (path->slots[0] < push_items) { 3530be0e5c09SChris Mason path->slots[0] += old_left_nritems; 3531925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 35325f39d397SChris Mason free_extent_buffer(path->nodes[0]); 35335f39d397SChris Mason path->nodes[0] = left; 3534be0e5c09SChris Mason path->slots[1] -= 1; 3535be0e5c09SChris Mason } else { 3536925baeddSChris Mason btrfs_tree_unlock(left); 35375f39d397SChris Mason free_extent_buffer(left); 3538be0e5c09SChris Mason path->slots[0] -= push_items; 3539be0e5c09SChris Mason } 3540eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 3541aa5d6bedSChris Mason return ret; 3542925baeddSChris Mason out: 3543925baeddSChris Mason btrfs_tree_unlock(left); 3544925baeddSChris Mason free_extent_buffer(left); 3545925baeddSChris Mason return ret; 3546be0e5c09SChris Mason } 3547be0e5c09SChris Mason 354874123bd7SChris Mason /* 354944871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 355044871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 355199d8f83cSChris Mason * 355299d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 355399d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the 355499d8f83cSChris Mason * items 355544871b1bSChris Mason */ 355644871b1bSChris Mason static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root 355799d8f83cSChris Mason *root, struct btrfs_path *path, int min_data_size, 355899d8f83cSChris Mason int data_size, int empty, u32 max_slot) 355944871b1bSChris Mason { 356044871b1bSChris Mason struct extent_buffer *right = path->nodes[0]; 356144871b1bSChris Mason struct extent_buffer *left; 356244871b1bSChris Mason int slot; 356344871b1bSChris Mason int free_space; 356444871b1bSChris Mason u32 right_nritems; 356544871b1bSChris Mason int ret = 0; 356644871b1bSChris Mason 356744871b1bSChris Mason slot = path->slots[1]; 356844871b1bSChris Mason if (slot == 0) 356944871b1bSChris Mason return 1; 357044871b1bSChris Mason if (!path->nodes[1]) 357144871b1bSChris Mason return 1; 357244871b1bSChris Mason 357344871b1bSChris Mason right_nritems = btrfs_header_nritems(right); 357444871b1bSChris Mason if (right_nritems == 0) 357544871b1bSChris Mason return 1; 357644871b1bSChris Mason 357749d0c642SFilipe Manana btrfs_assert_tree_write_locked(path->nodes[1]); 357844871b1bSChris Mason 35794b231ae4SDavid Sterba left = btrfs_read_node_slot(path->nodes[1], slot - 1); 3580fb770ae4SLiu Bo if (IS_ERR(left)) 35819cf14029SJosef Bacik return PTR_ERR(left); 358291ca338dSTsutomu Itoh 3583bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 358444871b1bSChris Mason 3585e902baacSDavid Sterba free_space = btrfs_leaf_free_space(left); 358644871b1bSChris Mason if (free_space < data_size) { 358744871b1bSChris Mason ret = 1; 358844871b1bSChris Mason goto out; 358944871b1bSChris Mason } 359044871b1bSChris Mason 359144871b1bSChris Mason ret = btrfs_cow_block(trans, root, left, 35929631e4ccSJosef Bacik path->nodes[1], slot - 1, &left, 3593bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 359444871b1bSChris Mason if (ret) { 359544871b1bSChris Mason /* we hit -ENOSPC, but it isn't fatal here */ 359679787eaaSJeff Mahoney if (ret == -ENOSPC) 359744871b1bSChris Mason ret = 1; 359844871b1bSChris Mason goto out; 359944871b1bSChris Mason } 360044871b1bSChris Mason 3601d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3602d16c702fSQu Wenruo ret = -EUCLEAN; 36039ae5afd0SFilipe Manana btrfs_abort_transaction(trans, ret); 3604d16c702fSQu Wenruo goto out; 3605d16c702fSQu Wenruo } 3606ed25dab3SJosef Bacik return __push_leaf_left(trans, path, min_data_size, empty, left, 3607ed25dab3SJosef Bacik free_space, right_nritems, max_slot); 360844871b1bSChris Mason out: 360944871b1bSChris Mason btrfs_tree_unlock(left); 361044871b1bSChris Mason free_extent_buffer(left); 361144871b1bSChris Mason return ret; 361244871b1bSChris Mason } 361344871b1bSChris Mason 361444871b1bSChris Mason /* 361574123bd7SChris Mason * split the path's leaf in two, making sure there is at least data_size 361674123bd7SChris Mason * available for the resulting leaf level of the path. 361774123bd7SChris Mason */ 361850b5d1fcSFilipe Manana static noinline int copy_for_split(struct btrfs_trans_handle *trans, 361944871b1bSChris Mason struct btrfs_path *path, 362044871b1bSChris Mason struct extent_buffer *l, 362144871b1bSChris Mason struct extent_buffer *right, 362244871b1bSChris Mason int slot, int mid, int nritems) 3623be0e5c09SChris Mason { 362494f94ad9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 3625be0e5c09SChris Mason int data_copy_size; 3626be0e5c09SChris Mason int rt_data_off; 3627be0e5c09SChris Mason int i; 362850b5d1fcSFilipe Manana int ret; 3629d4dbff95SChris Mason struct btrfs_disk_key disk_key; 3630cfed81a0SChris Mason struct btrfs_map_token token; 3631cfed81a0SChris Mason 36325f39d397SChris Mason nritems = nritems - mid; 36335f39d397SChris Mason btrfs_set_header_nritems(right, nritems); 3634dc2e724eSJosef Bacik data_copy_size = btrfs_item_data_end(l, mid) - leaf_data_end(l); 36355f39d397SChris Mason 3636637e3b48SJosef Bacik copy_leaf_items(right, l, 0, mid, nritems); 36375f39d397SChris Mason 3638637e3b48SJosef Bacik copy_leaf_data(right, l, BTRFS_LEAF_DATA_SIZE(fs_info) - data_copy_size, 36398f881e8cSDavid Sterba leaf_data_end(l), data_copy_size); 364074123bd7SChris Mason 3641dc2e724eSJosef Bacik rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_data_end(l, mid); 36425f39d397SChris Mason 3643c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 36445f39d397SChris Mason for (i = 0; i < nritems; i++) { 3645db94535dSChris Mason u32 ioff; 3646db94535dSChris Mason 36473212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 36483212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff + rt_data_off); 36490783fcfcSChris Mason } 365074123bd7SChris Mason 36515f39d397SChris Mason btrfs_set_header_nritems(l, mid); 36525f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 365350b5d1fcSFilipe Manana ret = insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); 365450b5d1fcSFilipe Manana if (ret < 0) 365550b5d1fcSFilipe Manana return ret; 36565f39d397SChris Mason 36575f39d397SChris Mason btrfs_mark_buffer_dirty(right); 36585f39d397SChris Mason btrfs_mark_buffer_dirty(l); 3659eb60ceacSChris Mason BUG_ON(path->slots[0] != slot); 36605f39d397SChris Mason 3661be0e5c09SChris Mason if (mid <= slot) { 3662925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 36635f39d397SChris Mason free_extent_buffer(path->nodes[0]); 36645f39d397SChris Mason path->nodes[0] = right; 3665be0e5c09SChris Mason path->slots[0] -= mid; 3666be0e5c09SChris Mason path->slots[1] += 1; 3667925baeddSChris Mason } else { 3668925baeddSChris Mason btrfs_tree_unlock(right); 36695f39d397SChris Mason free_extent_buffer(right); 3670925baeddSChris Mason } 36715f39d397SChris Mason 3672eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 367350b5d1fcSFilipe Manana 367450b5d1fcSFilipe Manana return 0; 367544871b1bSChris Mason } 367644871b1bSChris Mason 367744871b1bSChris Mason /* 367899d8f83cSChris Mason * double splits happen when we need to insert a big item in the middle 367999d8f83cSChris Mason * of a leaf. A double split can leave us with 3 mostly empty leaves: 368099d8f83cSChris Mason * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] 368199d8f83cSChris Mason * A B C 368299d8f83cSChris Mason * 368399d8f83cSChris Mason * We avoid this by trying to push the items on either side of our target 368499d8f83cSChris Mason * into the adjacent leaves. If all goes well we can avoid the double split 368599d8f83cSChris Mason * completely. 368699d8f83cSChris Mason */ 368799d8f83cSChris Mason static noinline int push_for_double_split(struct btrfs_trans_handle *trans, 368899d8f83cSChris Mason struct btrfs_root *root, 368999d8f83cSChris Mason struct btrfs_path *path, 369099d8f83cSChris Mason int data_size) 369199d8f83cSChris Mason { 369299d8f83cSChris Mason int ret; 369399d8f83cSChris Mason int progress = 0; 369499d8f83cSChris Mason int slot; 369599d8f83cSChris Mason u32 nritems; 36965a4267caSFilipe David Borba Manana int space_needed = data_size; 369799d8f83cSChris Mason 369899d8f83cSChris Mason slot = path->slots[0]; 36995a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(path->nodes[0])) 3700e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 370199d8f83cSChris Mason 370299d8f83cSChris Mason /* 370399d8f83cSChris Mason * try to push all the items after our slot into the 370499d8f83cSChris Mason * right leaf 370599d8f83cSChris Mason */ 37065a4267caSFilipe David Borba Manana ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); 370799d8f83cSChris Mason if (ret < 0) 370899d8f83cSChris Mason return ret; 370999d8f83cSChris Mason 371099d8f83cSChris Mason if (ret == 0) 371199d8f83cSChris Mason progress++; 371299d8f83cSChris Mason 371399d8f83cSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 371499d8f83cSChris Mason /* 371599d8f83cSChris Mason * our goal is to get our slot at the start or end of a leaf. If 371699d8f83cSChris Mason * we've done so we're done 371799d8f83cSChris Mason */ 371899d8f83cSChris Mason if (path->slots[0] == 0 || path->slots[0] == nritems) 371999d8f83cSChris Mason return 0; 372099d8f83cSChris Mason 3721e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 372299d8f83cSChris Mason return 0; 372399d8f83cSChris Mason 372499d8f83cSChris Mason /* try to push all the items before our slot into the next leaf */ 372599d8f83cSChris Mason slot = path->slots[0]; 3726263d3995SFilipe Manana space_needed = data_size; 3727263d3995SFilipe Manana if (slot > 0) 3728e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 37295a4267caSFilipe David Borba Manana ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); 373099d8f83cSChris Mason if (ret < 0) 373199d8f83cSChris Mason return ret; 373299d8f83cSChris Mason 373399d8f83cSChris Mason if (ret == 0) 373499d8f83cSChris Mason progress++; 373599d8f83cSChris Mason 373699d8f83cSChris Mason if (progress) 373799d8f83cSChris Mason return 0; 373899d8f83cSChris Mason return 1; 373999d8f83cSChris Mason } 374099d8f83cSChris Mason 374199d8f83cSChris Mason /* 374244871b1bSChris Mason * split the path's leaf in two, making sure there is at least data_size 374344871b1bSChris Mason * available for the resulting leaf level of the path. 374444871b1bSChris Mason * 374544871b1bSChris Mason * returns 0 if all went well and < 0 on failure. 374644871b1bSChris Mason */ 374744871b1bSChris Mason static noinline int split_leaf(struct btrfs_trans_handle *trans, 374844871b1bSChris Mason struct btrfs_root *root, 3749310712b2SOmar Sandoval const struct btrfs_key *ins_key, 375044871b1bSChris Mason struct btrfs_path *path, int data_size, 375144871b1bSChris Mason int extend) 375244871b1bSChris Mason { 37535d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 375444871b1bSChris Mason struct extent_buffer *l; 375544871b1bSChris Mason u32 nritems; 375644871b1bSChris Mason int mid; 375744871b1bSChris Mason int slot; 375844871b1bSChris Mason struct extent_buffer *right; 3759b7a0365eSDaniel Dressler struct btrfs_fs_info *fs_info = root->fs_info; 376044871b1bSChris Mason int ret = 0; 376144871b1bSChris Mason int wret; 37625d4f98a2SYan Zheng int split; 376344871b1bSChris Mason int num_doubles = 0; 376499d8f83cSChris Mason int tried_avoid_double = 0; 376544871b1bSChris Mason 3766a5719521SYan, Zheng l = path->nodes[0]; 3767a5719521SYan, Zheng slot = path->slots[0]; 37683212fa14SJosef Bacik if (extend && data_size + btrfs_item_size(l, slot) + 37690b246afaSJeff Mahoney sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) 3770a5719521SYan, Zheng return -EOVERFLOW; 3771a5719521SYan, Zheng 377244871b1bSChris Mason /* first try to make some room by pushing left and right */ 377333157e05SLiu Bo if (data_size && path->nodes[1]) { 37745a4267caSFilipe David Borba Manana int space_needed = data_size; 37755a4267caSFilipe David Borba Manana 37765a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(l)) 3777e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 37785a4267caSFilipe David Borba Manana 37795a4267caSFilipe David Borba Manana wret = push_leaf_right(trans, root, path, space_needed, 37805a4267caSFilipe David Borba Manana space_needed, 0, 0); 378144871b1bSChris Mason if (wret < 0) 378244871b1bSChris Mason return wret; 378344871b1bSChris Mason if (wret) { 3784263d3995SFilipe Manana space_needed = data_size; 3785263d3995SFilipe Manana if (slot > 0) 3786e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 37875a4267caSFilipe David Borba Manana wret = push_leaf_left(trans, root, path, space_needed, 37885a4267caSFilipe David Borba Manana space_needed, 0, (u32)-1); 378944871b1bSChris Mason if (wret < 0) 379044871b1bSChris Mason return wret; 379144871b1bSChris Mason } 379244871b1bSChris Mason l = path->nodes[0]; 379344871b1bSChris Mason 379444871b1bSChris Mason /* did the pushes work? */ 3795e902baacSDavid Sterba if (btrfs_leaf_free_space(l) >= data_size) 379644871b1bSChris Mason return 0; 379744871b1bSChris Mason } 379844871b1bSChris Mason 379944871b1bSChris Mason if (!path->nodes[1]) { 3800fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, 1); 380144871b1bSChris Mason if (ret) 380244871b1bSChris Mason return ret; 380344871b1bSChris Mason } 380444871b1bSChris Mason again: 38055d4f98a2SYan Zheng split = 1; 380644871b1bSChris Mason l = path->nodes[0]; 380744871b1bSChris Mason slot = path->slots[0]; 380844871b1bSChris Mason nritems = btrfs_header_nritems(l); 380944871b1bSChris Mason mid = (nritems + 1) / 2; 381044871b1bSChris Mason 38115d4f98a2SYan Zheng if (mid <= slot) { 38125d4f98a2SYan Zheng if (nritems == 1 || 38135d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + data_size > 38140b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 38155d4f98a2SYan Zheng if (slot >= nritems) { 38165d4f98a2SYan Zheng split = 0; 38175d4f98a2SYan Zheng } else { 38185d4f98a2SYan Zheng mid = slot; 38195d4f98a2SYan Zheng if (mid != nritems && 38205d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 38210b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 382299d8f83cSChris Mason if (data_size && !tried_avoid_double) 382399d8f83cSChris Mason goto push_for_double; 38245d4f98a2SYan Zheng split = 2; 38255d4f98a2SYan Zheng } 38265d4f98a2SYan Zheng } 38275d4f98a2SYan Zheng } 38285d4f98a2SYan Zheng } else { 38295d4f98a2SYan Zheng if (leaf_space_used(l, 0, mid) + data_size > 38300b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 38315d4f98a2SYan Zheng if (!extend && data_size && slot == 0) { 38325d4f98a2SYan Zheng split = 0; 38335d4f98a2SYan Zheng } else if ((extend || !data_size) && slot == 0) { 38345d4f98a2SYan Zheng mid = 1; 38355d4f98a2SYan Zheng } else { 38365d4f98a2SYan Zheng mid = slot; 38375d4f98a2SYan Zheng if (mid != nritems && 38385d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 38390b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 384099d8f83cSChris Mason if (data_size && !tried_avoid_double) 384199d8f83cSChris Mason goto push_for_double; 38425d4f98a2SYan Zheng split = 2; 38435d4f98a2SYan Zheng } 38445d4f98a2SYan Zheng } 38455d4f98a2SYan Zheng } 38465d4f98a2SYan Zheng } 38475d4f98a2SYan Zheng 38485d4f98a2SYan Zheng if (split == 0) 38495d4f98a2SYan Zheng btrfs_cpu_key_to_disk(&disk_key, ins_key); 38505d4f98a2SYan Zheng else 38515d4f98a2SYan Zheng btrfs_item_key(l, &disk_key, mid); 38525d4f98a2SYan Zheng 3853ca9d473aSJosef Bacik /* 3854ca9d473aSJosef Bacik * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double 3855ca9d473aSJosef Bacik * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES 3856ca9d473aSJosef Bacik * subclasses, which is 8 at the time of this patch, and we've maxed it 3857ca9d473aSJosef Bacik * out. In the future we could add a 3858ca9d473aSJosef Bacik * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just 3859ca9d473aSJosef Bacik * use BTRFS_NESTING_NEW_ROOT. 3860ca9d473aSJosef Bacik */ 386179bd3712SFilipe Manana right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 386279bd3712SFilipe Manana &disk_key, 0, l->start, 0, 386379bd3712SFilipe Manana num_doubles ? BTRFS_NESTING_NEW_ROOT : 3864ca9d473aSJosef Bacik BTRFS_NESTING_SPLIT); 3865f0486c68SYan, Zheng if (IS_ERR(right)) 386644871b1bSChris Mason return PTR_ERR(right); 3867f0486c68SYan, Zheng 38680b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 386944871b1bSChris Mason 38705d4f98a2SYan Zheng if (split == 0) { 387144871b1bSChris Mason if (mid <= slot) { 387244871b1bSChris Mason btrfs_set_header_nritems(right, 0); 387350b5d1fcSFilipe Manana ret = insert_ptr(trans, path, &disk_key, 38742ff7e61eSJeff Mahoney right->start, path->slots[1] + 1, 1); 387550b5d1fcSFilipe Manana if (ret < 0) { 387650b5d1fcSFilipe Manana btrfs_tree_unlock(right); 387750b5d1fcSFilipe Manana free_extent_buffer(right); 387850b5d1fcSFilipe Manana return ret; 387950b5d1fcSFilipe Manana } 388044871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 388144871b1bSChris Mason free_extent_buffer(path->nodes[0]); 388244871b1bSChris Mason path->nodes[0] = right; 388344871b1bSChris Mason path->slots[0] = 0; 388444871b1bSChris Mason path->slots[1] += 1; 388544871b1bSChris Mason } else { 388644871b1bSChris Mason btrfs_set_header_nritems(right, 0); 388750b5d1fcSFilipe Manana ret = insert_ptr(trans, path, &disk_key, 38882ff7e61eSJeff Mahoney right->start, path->slots[1], 1); 388950b5d1fcSFilipe Manana if (ret < 0) { 389050b5d1fcSFilipe Manana btrfs_tree_unlock(right); 389150b5d1fcSFilipe Manana free_extent_buffer(right); 389250b5d1fcSFilipe Manana return ret; 389350b5d1fcSFilipe Manana } 389444871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 389544871b1bSChris Mason free_extent_buffer(path->nodes[0]); 389644871b1bSChris Mason path->nodes[0] = right; 389744871b1bSChris Mason path->slots[0] = 0; 3898143bede5SJeff Mahoney if (path->slots[1] == 0) 3899b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 39005d4f98a2SYan Zheng } 3901196e0249SLiu Bo /* 3902196e0249SLiu Bo * We create a new leaf 'right' for the required ins_len and 3903196e0249SLiu Bo * we'll do btrfs_mark_buffer_dirty() on this leaf after copying 3904196e0249SLiu Bo * the content of ins_len to 'right'. 3905196e0249SLiu Bo */ 390644871b1bSChris Mason return ret; 390744871b1bSChris Mason } 390844871b1bSChris Mason 390950b5d1fcSFilipe Manana ret = copy_for_split(trans, path, l, right, slot, mid, nritems); 391050b5d1fcSFilipe Manana if (ret < 0) { 391150b5d1fcSFilipe Manana btrfs_tree_unlock(right); 391250b5d1fcSFilipe Manana free_extent_buffer(right); 391350b5d1fcSFilipe Manana return ret; 391450b5d1fcSFilipe Manana } 391544871b1bSChris Mason 39165d4f98a2SYan Zheng if (split == 2) { 3917cc0c5538SChris Mason BUG_ON(num_doubles != 0); 3918cc0c5538SChris Mason num_doubles++; 3919cc0c5538SChris Mason goto again; 39203326d1b0SChris Mason } 392144871b1bSChris Mason 3922143bede5SJeff Mahoney return 0; 392399d8f83cSChris Mason 392499d8f83cSChris Mason push_for_double: 392599d8f83cSChris Mason push_for_double_split(trans, root, path, data_size); 392699d8f83cSChris Mason tried_avoid_double = 1; 3927e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 392899d8f83cSChris Mason return 0; 392999d8f83cSChris Mason goto again; 3930be0e5c09SChris Mason } 3931be0e5c09SChris Mason 3932ad48fd75SYan, Zheng static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, 3933ad48fd75SYan, Zheng struct btrfs_root *root, 3934ad48fd75SYan, Zheng struct btrfs_path *path, int ins_len) 3935ad48fd75SYan, Zheng { 3936ad48fd75SYan, Zheng struct btrfs_key key; 3937ad48fd75SYan, Zheng struct extent_buffer *leaf; 3938ad48fd75SYan, Zheng struct btrfs_file_extent_item *fi; 3939ad48fd75SYan, Zheng u64 extent_len = 0; 3940ad48fd75SYan, Zheng u32 item_size; 3941ad48fd75SYan, Zheng int ret; 3942ad48fd75SYan, Zheng 3943ad48fd75SYan, Zheng leaf = path->nodes[0]; 3944ad48fd75SYan, Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3945ad48fd75SYan, Zheng 3946ad48fd75SYan, Zheng BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && 3947ad48fd75SYan, Zheng key.type != BTRFS_EXTENT_CSUM_KEY); 3948ad48fd75SYan, Zheng 3949e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) >= ins_len) 3950ad48fd75SYan, Zheng return 0; 3951ad48fd75SYan, Zheng 39523212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 3953ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 3954ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 3955ad48fd75SYan, Zheng struct btrfs_file_extent_item); 3956ad48fd75SYan, Zheng extent_len = btrfs_file_extent_num_bytes(leaf, fi); 3957ad48fd75SYan, Zheng } 3958b3b4aa74SDavid Sterba btrfs_release_path(path); 3959ad48fd75SYan, Zheng 3960ad48fd75SYan, Zheng path->keep_locks = 1; 3961ad48fd75SYan, Zheng path->search_for_split = 1; 3962ad48fd75SYan, Zheng ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 3963ad48fd75SYan, Zheng path->search_for_split = 0; 3964a8df6fe6SFilipe Manana if (ret > 0) 3965a8df6fe6SFilipe Manana ret = -EAGAIN; 3966ad48fd75SYan, Zheng if (ret < 0) 3967ad48fd75SYan, Zheng goto err; 3968ad48fd75SYan, Zheng 3969ad48fd75SYan, Zheng ret = -EAGAIN; 3970ad48fd75SYan, Zheng leaf = path->nodes[0]; 3971a8df6fe6SFilipe Manana /* if our item isn't there, return now */ 39723212fa14SJosef Bacik if (item_size != btrfs_item_size(leaf, path->slots[0])) 3973ad48fd75SYan, Zheng goto err; 3974ad48fd75SYan, Zheng 3975109f6aefSChris Mason /* the leaf has changed, it now has room. return now */ 3976e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) 3977109f6aefSChris Mason goto err; 3978109f6aefSChris Mason 3979ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 3980ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 3981ad48fd75SYan, Zheng struct btrfs_file_extent_item); 3982ad48fd75SYan, Zheng if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) 3983ad48fd75SYan, Zheng goto err; 3984ad48fd75SYan, Zheng } 3985ad48fd75SYan, Zheng 3986ad48fd75SYan, Zheng ret = split_leaf(trans, root, &key, path, ins_len, 1); 3987f0486c68SYan, Zheng if (ret) 3988f0486c68SYan, Zheng goto err; 3989ad48fd75SYan, Zheng 3990ad48fd75SYan, Zheng path->keep_locks = 0; 3991ad48fd75SYan, Zheng btrfs_unlock_up_safe(path, 1); 3992ad48fd75SYan, Zheng return 0; 3993ad48fd75SYan, Zheng err: 3994ad48fd75SYan, Zheng path->keep_locks = 0; 3995ad48fd75SYan, Zheng return ret; 3996ad48fd75SYan, Zheng } 3997ad48fd75SYan, Zheng 399825263cd7SDavid Sterba static noinline int split_item(struct btrfs_path *path, 3999310712b2SOmar Sandoval const struct btrfs_key *new_key, 4000459931ecSChris Mason unsigned long split_offset) 4001459931ecSChris Mason { 4002459931ecSChris Mason struct extent_buffer *leaf; 4003c91666b1SJosef Bacik int orig_slot, slot; 4004ad48fd75SYan, Zheng char *buf; 4005459931ecSChris Mason u32 nritems; 4006ad48fd75SYan, Zheng u32 item_size; 4007459931ecSChris Mason u32 orig_offset; 4008459931ecSChris Mason struct btrfs_disk_key disk_key; 4009459931ecSChris Mason 4010459931ecSChris Mason leaf = path->nodes[0]; 40117569141eSFilipe Manana /* 40127569141eSFilipe Manana * Shouldn't happen because the caller must have previously called 40137569141eSFilipe Manana * setup_leaf_for_split() to make room for the new item in the leaf. 40147569141eSFilipe Manana */ 40157569141eSFilipe Manana if (WARN_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item))) 40167569141eSFilipe Manana return -ENOSPC; 4017b9473439SChris Mason 4018c91666b1SJosef Bacik orig_slot = path->slots[0]; 40193212fa14SJosef Bacik orig_offset = btrfs_item_offset(leaf, path->slots[0]); 40203212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 4021459931ecSChris Mason 4022459931ecSChris Mason buf = kmalloc(item_size, GFP_NOFS); 4023ad48fd75SYan, Zheng if (!buf) 4024ad48fd75SYan, Zheng return -ENOMEM; 4025ad48fd75SYan, Zheng 4026459931ecSChris Mason read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, 4027459931ecSChris Mason path->slots[0]), item_size); 4028ad48fd75SYan, Zheng 4029459931ecSChris Mason slot = path->slots[0] + 1; 4030459931ecSChris Mason nritems = btrfs_header_nritems(leaf); 4031459931ecSChris Mason if (slot != nritems) { 4032459931ecSChris Mason /* shift the items */ 4033637e3b48SJosef Bacik memmove_leaf_items(leaf, slot + 1, slot, nritems - slot); 4034459931ecSChris Mason } 4035459931ecSChris Mason 4036459931ecSChris Mason btrfs_cpu_key_to_disk(&disk_key, new_key); 4037459931ecSChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 4038459931ecSChris Mason 40393212fa14SJosef Bacik btrfs_set_item_offset(leaf, slot, orig_offset); 40403212fa14SJosef Bacik btrfs_set_item_size(leaf, slot, item_size - split_offset); 4041459931ecSChris Mason 40423212fa14SJosef Bacik btrfs_set_item_offset(leaf, orig_slot, 4043459931ecSChris Mason orig_offset + item_size - split_offset); 40443212fa14SJosef Bacik btrfs_set_item_size(leaf, orig_slot, split_offset); 4045459931ecSChris Mason 4046459931ecSChris Mason btrfs_set_header_nritems(leaf, nritems + 1); 4047459931ecSChris Mason 4048459931ecSChris Mason /* write the data for the start of the original item */ 4049459931ecSChris Mason write_extent_buffer(leaf, buf, 4050459931ecSChris Mason btrfs_item_ptr_offset(leaf, path->slots[0]), 4051459931ecSChris Mason split_offset); 4052459931ecSChris Mason 4053459931ecSChris Mason /* write the data for the new item */ 4054459931ecSChris Mason write_extent_buffer(leaf, buf + split_offset, 4055459931ecSChris Mason btrfs_item_ptr_offset(leaf, slot), 4056459931ecSChris Mason item_size - split_offset); 4057459931ecSChris Mason btrfs_mark_buffer_dirty(leaf); 4058459931ecSChris Mason 4059e902baacSDavid Sterba BUG_ON(btrfs_leaf_free_space(leaf) < 0); 4060459931ecSChris Mason kfree(buf); 4061ad48fd75SYan, Zheng return 0; 4062ad48fd75SYan, Zheng } 4063ad48fd75SYan, Zheng 4064ad48fd75SYan, Zheng /* 4065ad48fd75SYan, Zheng * This function splits a single item into two items, 4066ad48fd75SYan, Zheng * giving 'new_key' to the new item and splitting the 4067ad48fd75SYan, Zheng * old one at split_offset (from the start of the item). 4068ad48fd75SYan, Zheng * 4069ad48fd75SYan, Zheng * The path may be released by this operation. After 4070ad48fd75SYan, Zheng * the split, the path is pointing to the old item. The 4071ad48fd75SYan, Zheng * new item is going to be in the same node as the old one. 4072ad48fd75SYan, Zheng * 4073ad48fd75SYan, Zheng * Note, the item being split must be smaller enough to live alone on 4074ad48fd75SYan, Zheng * a tree block with room for one extra struct btrfs_item 4075ad48fd75SYan, Zheng * 4076ad48fd75SYan, Zheng * This allows us to split the item in place, keeping a lock on the 4077ad48fd75SYan, Zheng * leaf the entire time. 4078ad48fd75SYan, Zheng */ 4079ad48fd75SYan, Zheng int btrfs_split_item(struct btrfs_trans_handle *trans, 4080ad48fd75SYan, Zheng struct btrfs_root *root, 4081ad48fd75SYan, Zheng struct btrfs_path *path, 4082310712b2SOmar Sandoval const struct btrfs_key *new_key, 4083ad48fd75SYan, Zheng unsigned long split_offset) 4084ad48fd75SYan, Zheng { 4085ad48fd75SYan, Zheng int ret; 4086ad48fd75SYan, Zheng ret = setup_leaf_for_split(trans, root, path, 4087ad48fd75SYan, Zheng sizeof(struct btrfs_item)); 4088ad48fd75SYan, Zheng if (ret) 4089459931ecSChris Mason return ret; 4090ad48fd75SYan, Zheng 409125263cd7SDavid Sterba ret = split_item(path, new_key, split_offset); 4092ad48fd75SYan, Zheng return ret; 4093ad48fd75SYan, Zheng } 4094ad48fd75SYan, Zheng 4095ad48fd75SYan, Zheng /* 4096d352ac68SChris Mason * make the item pointed to by the path smaller. new_size indicates 4097d352ac68SChris Mason * how small to make it, and from_end tells us if we just chop bytes 4098d352ac68SChris Mason * off the end of the item or if we shift the item to chop bytes off 4099d352ac68SChris Mason * the front. 4100d352ac68SChris Mason */ 410178ac4f9eSDavid Sterba void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) 4102b18c6685SChris Mason { 4103b18c6685SChris Mason int slot; 41045f39d397SChris Mason struct extent_buffer *leaf; 4105b18c6685SChris Mason u32 nritems; 4106b18c6685SChris Mason unsigned int data_end; 4107b18c6685SChris Mason unsigned int old_data_start; 4108b18c6685SChris Mason unsigned int old_size; 4109b18c6685SChris Mason unsigned int size_diff; 4110b18c6685SChris Mason int i; 4111cfed81a0SChris Mason struct btrfs_map_token token; 4112cfed81a0SChris Mason 41135f39d397SChris Mason leaf = path->nodes[0]; 4114179e29e4SChris Mason slot = path->slots[0]; 4115179e29e4SChris Mason 41163212fa14SJosef Bacik old_size = btrfs_item_size(leaf, slot); 4117179e29e4SChris Mason if (old_size == new_size) 4118143bede5SJeff Mahoney return; 4119b18c6685SChris Mason 41205f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 41218f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 4122b18c6685SChris Mason 41233212fa14SJosef Bacik old_data_start = btrfs_item_offset(leaf, slot); 4124179e29e4SChris Mason 4125b18c6685SChris Mason size_diff = old_size - new_size; 4126b18c6685SChris Mason 4127b18c6685SChris Mason BUG_ON(slot < 0); 4128b18c6685SChris Mason BUG_ON(slot >= nritems); 4129b18c6685SChris Mason 4130b18c6685SChris Mason /* 4131b18c6685SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 4132b18c6685SChris Mason */ 4133b18c6685SChris Mason /* first correct the data pointers */ 4134c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 4135b18c6685SChris Mason for (i = slot; i < nritems; i++) { 41365f39d397SChris Mason u32 ioff; 4137db94535dSChris Mason 41383212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 41393212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff + size_diff); 4140b18c6685SChris Mason } 4141db94535dSChris Mason 4142b18c6685SChris Mason /* shift the data */ 4143179e29e4SChris Mason if (from_end) { 4144637e3b48SJosef Bacik memmove_leaf_data(leaf, data_end + size_diff, data_end, 4145637e3b48SJosef Bacik old_data_start + new_size - data_end); 4146179e29e4SChris Mason } else { 4147179e29e4SChris Mason struct btrfs_disk_key disk_key; 4148179e29e4SChris Mason u64 offset; 4149179e29e4SChris Mason 4150179e29e4SChris Mason btrfs_item_key(leaf, &disk_key, slot); 4151179e29e4SChris Mason 4152179e29e4SChris Mason if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { 4153179e29e4SChris Mason unsigned long ptr; 4154179e29e4SChris Mason struct btrfs_file_extent_item *fi; 4155179e29e4SChris Mason 4156179e29e4SChris Mason fi = btrfs_item_ptr(leaf, slot, 4157179e29e4SChris Mason struct btrfs_file_extent_item); 4158179e29e4SChris Mason fi = (struct btrfs_file_extent_item *)( 4159179e29e4SChris Mason (unsigned long)fi - size_diff); 4160179e29e4SChris Mason 4161179e29e4SChris Mason if (btrfs_file_extent_type(leaf, fi) == 4162179e29e4SChris Mason BTRFS_FILE_EXTENT_INLINE) { 4163179e29e4SChris Mason ptr = btrfs_item_ptr_offset(leaf, slot); 4164179e29e4SChris Mason memmove_extent_buffer(leaf, ptr, 4165179e29e4SChris Mason (unsigned long)fi, 41667ec20afbSDavid Sterba BTRFS_FILE_EXTENT_INLINE_DATA_START); 4167179e29e4SChris Mason } 4168179e29e4SChris Mason } 4169179e29e4SChris Mason 4170637e3b48SJosef Bacik memmove_leaf_data(leaf, data_end + size_diff, data_end, 4171637e3b48SJosef Bacik old_data_start - data_end); 4172179e29e4SChris Mason 4173179e29e4SChris Mason offset = btrfs_disk_key_offset(&disk_key); 4174179e29e4SChris Mason btrfs_set_disk_key_offset(&disk_key, offset + size_diff); 4175179e29e4SChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 4176179e29e4SChris Mason if (slot == 0) 4177b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 4178179e29e4SChris Mason } 41795f39d397SChris Mason 41803212fa14SJosef Bacik btrfs_set_item_size(leaf, slot, new_size); 41815f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 4182b18c6685SChris Mason 4183e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4184a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4185b18c6685SChris Mason BUG(); 41865f39d397SChris Mason } 4187b18c6685SChris Mason } 4188b18c6685SChris Mason 4189d352ac68SChris Mason /* 41908f69dbd2SStefan Behrens * make the item pointed to by the path bigger, data_size is the added size. 4191d352ac68SChris Mason */ 4192c71dd880SDavid Sterba void btrfs_extend_item(struct btrfs_path *path, u32 data_size) 41936567e837SChris Mason { 41946567e837SChris Mason int slot; 41955f39d397SChris Mason struct extent_buffer *leaf; 41966567e837SChris Mason u32 nritems; 41976567e837SChris Mason unsigned int data_end; 41986567e837SChris Mason unsigned int old_data; 41996567e837SChris Mason unsigned int old_size; 42006567e837SChris Mason int i; 4201cfed81a0SChris Mason struct btrfs_map_token token; 4202cfed81a0SChris Mason 42035f39d397SChris Mason leaf = path->nodes[0]; 42046567e837SChris Mason 42055f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 42068f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 42076567e837SChris Mason 4208e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < data_size) { 4209a4f78750SDavid Sterba btrfs_print_leaf(leaf); 42106567e837SChris Mason BUG(); 42115f39d397SChris Mason } 42126567e837SChris Mason slot = path->slots[0]; 4213dc2e724eSJosef Bacik old_data = btrfs_item_data_end(leaf, slot); 42146567e837SChris Mason 42156567e837SChris Mason BUG_ON(slot < 0); 42163326d1b0SChris Mason if (slot >= nritems) { 4217a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4218c71dd880SDavid Sterba btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", 4219d397712bSChris Mason slot, nritems); 4220290342f6SArnd Bergmann BUG(); 42213326d1b0SChris Mason } 42226567e837SChris Mason 42236567e837SChris Mason /* 42246567e837SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 42256567e837SChris Mason */ 42266567e837SChris Mason /* first correct the data pointers */ 4227c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 42286567e837SChris Mason for (i = slot; i < nritems; i++) { 42295f39d397SChris Mason u32 ioff; 4230db94535dSChris Mason 42313212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 42323212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff - data_size); 42336567e837SChris Mason } 42345f39d397SChris Mason 42356567e837SChris Mason /* shift the data */ 4236637e3b48SJosef Bacik memmove_leaf_data(leaf, data_end - data_size, data_end, 4237637e3b48SJosef Bacik old_data - data_end); 42385f39d397SChris Mason 42396567e837SChris Mason data_end = old_data; 42403212fa14SJosef Bacik old_size = btrfs_item_size(leaf, slot); 42413212fa14SJosef Bacik btrfs_set_item_size(leaf, slot, old_size + data_size); 42425f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 42436567e837SChris Mason 4244e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4245a4f78750SDavid Sterba btrfs_print_leaf(leaf); 42466567e837SChris Mason BUG(); 42475f39d397SChris Mason } 42486567e837SChris Mason } 42496567e837SChris Mason 425043dd529aSDavid Sterba /* 425143dd529aSDavid Sterba * Make space in the node before inserting one or more items. 4252da9ffb24SNikolay Borisov * 4253da9ffb24SNikolay Borisov * @root: root we are inserting items to 4254da9ffb24SNikolay Borisov * @path: points to the leaf/slot where we are going to insert new items 4255b7ef5f3aSFilipe Manana * @batch: information about the batch of items to insert 425643dd529aSDavid Sterba * 425743dd529aSDavid Sterba * Main purpose is to save stack depth by doing the bulk of the work in a 425843dd529aSDavid Sterba * function that doesn't call btrfs_search_slot 425974123bd7SChris Mason */ 4260f0641656SFilipe Manana static void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, 4261b7ef5f3aSFilipe Manana const struct btrfs_item_batch *batch) 4262be0e5c09SChris Mason { 42630b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 42649c58309dSChris Mason int i; 42657518a238SChris Mason u32 nritems; 4266be0e5c09SChris Mason unsigned int data_end; 4267e2fa7227SChris Mason struct btrfs_disk_key disk_key; 426844871b1bSChris Mason struct extent_buffer *leaf; 426944871b1bSChris Mason int slot; 4270cfed81a0SChris Mason struct btrfs_map_token token; 4271fc0d82e1SNikolay Borisov u32 total_size; 4272fc0d82e1SNikolay Borisov 4273b7ef5f3aSFilipe Manana /* 4274b7ef5f3aSFilipe Manana * Before anything else, update keys in the parent and other ancestors 4275b7ef5f3aSFilipe Manana * if needed, then release the write locks on them, so that other tasks 4276b7ef5f3aSFilipe Manana * can use them while we modify the leaf. 4277b7ef5f3aSFilipe Manana */ 427824cdc847SFilipe Manana if (path->slots[0] == 0) { 4279b7ef5f3aSFilipe Manana btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]); 4280b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 428124cdc847SFilipe Manana } 428224cdc847SFilipe Manana btrfs_unlock_up_safe(path, 1); 428324cdc847SFilipe Manana 42845f39d397SChris Mason leaf = path->nodes[0]; 428544871b1bSChris Mason slot = path->slots[0]; 428674123bd7SChris Mason 42875f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 42888f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 4289b7ef5f3aSFilipe Manana total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); 4290eb60ceacSChris Mason 4291e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < total_size) { 4292a4f78750SDavid Sterba btrfs_print_leaf(leaf); 42930b246afaSJeff Mahoney btrfs_crit(fs_info, "not enough freespace need %u have %d", 4294e902baacSDavid Sterba total_size, btrfs_leaf_free_space(leaf)); 4295be0e5c09SChris Mason BUG(); 4296d4dbff95SChris Mason } 42975f39d397SChris Mason 4298c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 4299be0e5c09SChris Mason if (slot != nritems) { 4300dc2e724eSJosef Bacik unsigned int old_data = btrfs_item_data_end(leaf, slot); 4301be0e5c09SChris Mason 43025f39d397SChris Mason if (old_data < data_end) { 4303a4f78750SDavid Sterba btrfs_print_leaf(leaf); 43047269ddd2SNikolay Borisov btrfs_crit(fs_info, 43057269ddd2SNikolay Borisov "item at slot %d with data offset %u beyond data end of leaf %u", 43065f39d397SChris Mason slot, old_data, data_end); 4307290342f6SArnd Bergmann BUG(); 43085f39d397SChris Mason } 4309be0e5c09SChris Mason /* 4310be0e5c09SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 4311be0e5c09SChris Mason */ 4312be0e5c09SChris Mason /* first correct the data pointers */ 43130783fcfcSChris Mason for (i = slot; i < nritems; i++) { 43145f39d397SChris Mason u32 ioff; 4315db94535dSChris Mason 43163212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 43173212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, 4318b7ef5f3aSFilipe Manana ioff - batch->total_data_size); 43190783fcfcSChris Mason } 4320be0e5c09SChris Mason /* shift the items */ 4321637e3b48SJosef Bacik memmove_leaf_items(leaf, slot + batch->nr, slot, nritems - slot); 4322be0e5c09SChris Mason 4323be0e5c09SChris Mason /* shift the data */ 4324637e3b48SJosef Bacik memmove_leaf_data(leaf, data_end - batch->total_data_size, 4325637e3b48SJosef Bacik data_end, old_data - data_end); 4326be0e5c09SChris Mason data_end = old_data; 4327be0e5c09SChris Mason } 43285f39d397SChris Mason 432962e2749eSChris Mason /* setup the item for the new data */ 4330b7ef5f3aSFilipe Manana for (i = 0; i < batch->nr; i++) { 4331b7ef5f3aSFilipe Manana btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]); 43329c58309dSChris Mason btrfs_set_item_key(leaf, &disk_key, slot + i); 4333b7ef5f3aSFilipe Manana data_end -= batch->data_sizes[i]; 43343212fa14SJosef Bacik btrfs_set_token_item_offset(&token, slot + i, data_end); 43353212fa14SJosef Bacik btrfs_set_token_item_size(&token, slot + i, batch->data_sizes[i]); 43369c58309dSChris Mason } 433744871b1bSChris Mason 4338b7ef5f3aSFilipe Manana btrfs_set_header_nritems(leaf, nritems + batch->nr); 4339b9473439SChris Mason btrfs_mark_buffer_dirty(leaf); 4340aa5d6bedSChris Mason 4341e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4342a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4343be0e5c09SChris Mason BUG(); 43445f39d397SChris Mason } 434544871b1bSChris Mason } 434644871b1bSChris Mason 434744871b1bSChris Mason /* 4348f0641656SFilipe Manana * Insert a new item into a leaf. 4349f0641656SFilipe Manana * 4350f0641656SFilipe Manana * @root: The root of the btree. 4351f0641656SFilipe Manana * @path: A path pointing to the target leaf and slot. 4352f0641656SFilipe Manana * @key: The key of the new item. 4353f0641656SFilipe Manana * @data_size: The size of the data associated with the new key. 4354f0641656SFilipe Manana */ 4355f0641656SFilipe Manana void btrfs_setup_item_for_insert(struct btrfs_root *root, 4356f0641656SFilipe Manana struct btrfs_path *path, 4357f0641656SFilipe Manana const struct btrfs_key *key, 4358f0641656SFilipe Manana u32 data_size) 4359f0641656SFilipe Manana { 4360f0641656SFilipe Manana struct btrfs_item_batch batch; 4361f0641656SFilipe Manana 4362f0641656SFilipe Manana batch.keys = key; 4363f0641656SFilipe Manana batch.data_sizes = &data_size; 4364f0641656SFilipe Manana batch.total_data_size = data_size; 4365f0641656SFilipe Manana batch.nr = 1; 4366f0641656SFilipe Manana 4367f0641656SFilipe Manana setup_items_for_insert(root, path, &batch); 4368f0641656SFilipe Manana } 4369f0641656SFilipe Manana 4370f0641656SFilipe Manana /* 437144871b1bSChris Mason * Given a key and some data, insert items into the tree. 437244871b1bSChris Mason * This does all the path init required, making room in the tree if needed. 437344871b1bSChris Mason */ 437444871b1bSChris Mason int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, 437544871b1bSChris Mason struct btrfs_root *root, 437644871b1bSChris Mason struct btrfs_path *path, 4377b7ef5f3aSFilipe Manana const struct btrfs_item_batch *batch) 437844871b1bSChris Mason { 437944871b1bSChris Mason int ret = 0; 438044871b1bSChris Mason int slot; 4381b7ef5f3aSFilipe Manana u32 total_size; 438244871b1bSChris Mason 4383b7ef5f3aSFilipe Manana total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); 4384b7ef5f3aSFilipe Manana ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1); 438544871b1bSChris Mason if (ret == 0) 438644871b1bSChris Mason return -EEXIST; 438744871b1bSChris Mason if (ret < 0) 4388143bede5SJeff Mahoney return ret; 438944871b1bSChris Mason 439044871b1bSChris Mason slot = path->slots[0]; 439144871b1bSChris Mason BUG_ON(slot < 0); 439244871b1bSChris Mason 4393b7ef5f3aSFilipe Manana setup_items_for_insert(root, path, batch); 4394143bede5SJeff Mahoney return 0; 439562e2749eSChris Mason } 439662e2749eSChris Mason 439762e2749eSChris Mason /* 439862e2749eSChris Mason * Given a key and some data, insert an item into the tree. 439962e2749eSChris Mason * This does all the path init required, making room in the tree if needed. 440062e2749eSChris Mason */ 4401310712b2SOmar Sandoval int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, 4402310712b2SOmar Sandoval const struct btrfs_key *cpu_key, void *data, 4403310712b2SOmar Sandoval u32 data_size) 440462e2749eSChris Mason { 440562e2749eSChris Mason int ret = 0; 44062c90e5d6SChris Mason struct btrfs_path *path; 44075f39d397SChris Mason struct extent_buffer *leaf; 44085f39d397SChris Mason unsigned long ptr; 440962e2749eSChris Mason 44102c90e5d6SChris Mason path = btrfs_alloc_path(); 4411db5b493aSTsutomu Itoh if (!path) 4412db5b493aSTsutomu Itoh return -ENOMEM; 44132c90e5d6SChris Mason ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); 441462e2749eSChris Mason if (!ret) { 44155f39d397SChris Mason leaf = path->nodes[0]; 44165f39d397SChris Mason ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); 44175f39d397SChris Mason write_extent_buffer(leaf, data, ptr, data_size); 44185f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 441962e2749eSChris Mason } 44202c90e5d6SChris Mason btrfs_free_path(path); 4421aa5d6bedSChris Mason return ret; 4422be0e5c09SChris Mason } 4423be0e5c09SChris Mason 442474123bd7SChris Mason /* 4425f0641656SFilipe Manana * This function duplicates an item, giving 'new_key' to the new item. 4426f0641656SFilipe Manana * It guarantees both items live in the same tree leaf and the new item is 4427f0641656SFilipe Manana * contiguous with the original item. 4428f0641656SFilipe Manana * 4429f0641656SFilipe Manana * This allows us to split a file extent in place, keeping a lock on the leaf 4430f0641656SFilipe Manana * the entire time. 4431f0641656SFilipe Manana */ 4432f0641656SFilipe Manana int btrfs_duplicate_item(struct btrfs_trans_handle *trans, 4433f0641656SFilipe Manana struct btrfs_root *root, 4434f0641656SFilipe Manana struct btrfs_path *path, 4435f0641656SFilipe Manana const struct btrfs_key *new_key) 4436f0641656SFilipe Manana { 4437f0641656SFilipe Manana struct extent_buffer *leaf; 4438f0641656SFilipe Manana int ret; 4439f0641656SFilipe Manana u32 item_size; 4440f0641656SFilipe Manana 4441f0641656SFilipe Manana leaf = path->nodes[0]; 44423212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 4443f0641656SFilipe Manana ret = setup_leaf_for_split(trans, root, path, 4444f0641656SFilipe Manana item_size + sizeof(struct btrfs_item)); 4445f0641656SFilipe Manana if (ret) 4446f0641656SFilipe Manana return ret; 4447f0641656SFilipe Manana 4448f0641656SFilipe Manana path->slots[0]++; 4449f0641656SFilipe Manana btrfs_setup_item_for_insert(root, path, new_key, item_size); 4450f0641656SFilipe Manana leaf = path->nodes[0]; 4451f0641656SFilipe Manana memcpy_extent_buffer(leaf, 4452f0641656SFilipe Manana btrfs_item_ptr_offset(leaf, path->slots[0]), 4453f0641656SFilipe Manana btrfs_item_ptr_offset(leaf, path->slots[0] - 1), 4454f0641656SFilipe Manana item_size); 4455f0641656SFilipe Manana return 0; 4456f0641656SFilipe Manana } 4457f0641656SFilipe Manana 4458f0641656SFilipe Manana /* 44595de08d7dSChris Mason * delete the pointer from a given node. 446074123bd7SChris Mason * 4461d352ac68SChris Mason * the tree should have been previously balanced so the deletion does not 4462d352ac68SChris Mason * empty a node. 4463016f9d0bSJosef Bacik * 4464016f9d0bSJosef Bacik * This is exported for use inside btrfs-progs, don't un-export it. 446574123bd7SChris Mason */ 4466751a2761SFilipe Manana int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, 4467751a2761SFilipe Manana struct btrfs_path *path, int level, int slot) 4468be0e5c09SChris Mason { 44695f39d397SChris Mason struct extent_buffer *parent = path->nodes[level]; 44707518a238SChris Mason u32 nritems; 4471f3ea38daSJan Schmidt int ret; 4472be0e5c09SChris Mason 44735f39d397SChris Mason nritems = btrfs_header_nritems(parent); 4474be0e5c09SChris Mason if (slot != nritems - 1) { 4475bf1d3425SDavid Sterba if (level) { 4476f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_move(parent, slot, 4477f3a84ccdSFilipe Manana slot + 1, nritems - slot - 1); 4478751a2761SFilipe Manana if (ret < 0) { 4479751a2761SFilipe Manana btrfs_abort_transaction(trans, ret); 4480751a2761SFilipe Manana return ret; 4481751a2761SFilipe Manana } 4482bf1d3425SDavid Sterba } 44835f39d397SChris Mason memmove_extent_buffer(parent, 4484e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(parent, slot), 4485e23efd8eSJosef Bacik btrfs_node_key_ptr_offset(parent, slot + 1), 4486d6025579SChris Mason sizeof(struct btrfs_key_ptr) * 4487d6025579SChris Mason (nritems - slot - 1)); 448857ba86c0SChris Mason } else if (level) { 4489f3a84ccdSFilipe Manana ret = btrfs_tree_mod_log_insert_key(parent, slot, 449033cff222SFilipe Manana BTRFS_MOD_LOG_KEY_REMOVE); 4491751a2761SFilipe Manana if (ret < 0) { 4492751a2761SFilipe Manana btrfs_abort_transaction(trans, ret); 4493751a2761SFilipe Manana return ret; 4494751a2761SFilipe Manana } 4495be0e5c09SChris Mason } 4496f3ea38daSJan Schmidt 44977518a238SChris Mason nritems--; 44985f39d397SChris Mason btrfs_set_header_nritems(parent, nritems); 44997518a238SChris Mason if (nritems == 0 && parent == root->node) { 45005f39d397SChris Mason BUG_ON(btrfs_header_level(root->node) != 1); 4501eb60ceacSChris Mason /* just turn the root into a leaf and break */ 45025f39d397SChris Mason btrfs_set_header_level(root->node, 0); 4503bb803951SChris Mason } else if (slot == 0) { 45045f39d397SChris Mason struct btrfs_disk_key disk_key; 45055f39d397SChris Mason 45065f39d397SChris Mason btrfs_node_key(parent, &disk_key, 0); 4507b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, level + 1); 4508be0e5c09SChris Mason } 4509d6025579SChris Mason btrfs_mark_buffer_dirty(parent); 4510751a2761SFilipe Manana return 0; 4511be0e5c09SChris Mason } 4512be0e5c09SChris Mason 451374123bd7SChris Mason /* 4514323ac95bSChris Mason * a helper function to delete the leaf pointed to by path->slots[1] and 45155d4f98a2SYan Zheng * path->nodes[1]. 4516323ac95bSChris Mason * 4517323ac95bSChris Mason * This deletes the pointer in path->nodes[1] and frees the leaf 4518323ac95bSChris Mason * block extent. zero is returned if it all worked out, < 0 otherwise. 4519323ac95bSChris Mason * 4520323ac95bSChris Mason * The path must have already been setup for deleting the leaf, including 4521323ac95bSChris Mason * all the proper balancing. path->nodes[1] must be locked. 4522323ac95bSChris Mason */ 4523751a2761SFilipe Manana static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans, 4524323ac95bSChris Mason struct btrfs_root *root, 45255d4f98a2SYan Zheng struct btrfs_path *path, 45265d4f98a2SYan Zheng struct extent_buffer *leaf) 4527323ac95bSChris Mason { 4528751a2761SFilipe Manana int ret; 4529751a2761SFilipe Manana 45305d4f98a2SYan Zheng WARN_ON(btrfs_header_generation(leaf) != trans->transid); 4531751a2761SFilipe Manana ret = btrfs_del_ptr(trans, root, path, 1, path->slots[1]); 4532751a2761SFilipe Manana if (ret < 0) 4533751a2761SFilipe Manana return ret; 4534323ac95bSChris Mason 45354d081c41SChris Mason /* 45364d081c41SChris Mason * btrfs_free_extent is expensive, we want to make sure we 45374d081c41SChris Mason * aren't holding any locks when we call it 45384d081c41SChris Mason */ 45394d081c41SChris Mason btrfs_unlock_up_safe(path, 0); 45404d081c41SChris Mason 4541f0486c68SYan, Zheng root_sub_used(root, leaf->len); 4542f0486c68SYan, Zheng 454367439dadSDavid Sterba atomic_inc(&leaf->refs); 45447a163608SFilipe Manana btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1); 45453083ee2eSJosef Bacik free_extent_buffer_stale(leaf); 4546751a2761SFilipe Manana return 0; 4547323ac95bSChris Mason } 4548323ac95bSChris Mason /* 454974123bd7SChris Mason * delete the item at the leaf level in path. If that empties 455074123bd7SChris Mason * the leaf, remove it from the tree 455174123bd7SChris Mason */ 455285e21bacSChris Mason int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, 455385e21bacSChris Mason struct btrfs_path *path, int slot, int nr) 4554be0e5c09SChris Mason { 45550b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 45565f39d397SChris Mason struct extent_buffer *leaf; 4557aa5d6bedSChris Mason int ret = 0; 4558aa5d6bedSChris Mason int wret; 45597518a238SChris Mason u32 nritems; 4560be0e5c09SChris Mason 45615f39d397SChris Mason leaf = path->nodes[0]; 45625f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 4563be0e5c09SChris Mason 456485e21bacSChris Mason if (slot + nr != nritems) { 45650cae23b6SFilipe Manana const u32 last_off = btrfs_item_offset(leaf, slot + nr - 1); 45660cae23b6SFilipe Manana const int data_end = leaf_data_end(leaf); 4567c82f823cSDavid Sterba struct btrfs_map_token token; 45680cae23b6SFilipe Manana u32 dsize = 0; 45690cae23b6SFilipe Manana int i; 45700cae23b6SFilipe Manana 45710cae23b6SFilipe Manana for (i = 0; i < nr; i++) 45720cae23b6SFilipe Manana dsize += btrfs_item_size(leaf, slot + i); 45735f39d397SChris Mason 4574637e3b48SJosef Bacik memmove_leaf_data(leaf, data_end + dsize, data_end, 457585e21bacSChris Mason last_off - data_end); 45765f39d397SChris Mason 4577c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 457885e21bacSChris Mason for (i = slot + nr; i < nritems; i++) { 45795f39d397SChris Mason u32 ioff; 4580db94535dSChris Mason 45813212fa14SJosef Bacik ioff = btrfs_token_item_offset(&token, i); 45823212fa14SJosef Bacik btrfs_set_token_item_offset(&token, i, ioff + dsize); 45830783fcfcSChris Mason } 4584db94535dSChris Mason 4585637e3b48SJosef Bacik memmove_leaf_items(leaf, slot, slot + nr, nritems - slot - nr); 4586be0e5c09SChris Mason } 458785e21bacSChris Mason btrfs_set_header_nritems(leaf, nritems - nr); 458885e21bacSChris Mason nritems -= nr; 45895f39d397SChris Mason 459074123bd7SChris Mason /* delete the leaf if we've emptied it */ 45917518a238SChris Mason if (nritems == 0) { 45925f39d397SChris Mason if (leaf == root->node) { 45935f39d397SChris Mason btrfs_set_header_level(leaf, 0); 45949a8dd150SChris Mason } else { 4595190a8339SJosef Bacik btrfs_clear_buffer_dirty(trans, leaf); 4596751a2761SFilipe Manana ret = btrfs_del_leaf(trans, root, path, leaf); 4597751a2761SFilipe Manana if (ret < 0) 4598751a2761SFilipe Manana return ret; 45999a8dd150SChris Mason } 4600be0e5c09SChris Mason } else { 46017518a238SChris Mason int used = leaf_space_used(leaf, 0, nritems); 4602aa5d6bedSChris Mason if (slot == 0) { 46035f39d397SChris Mason struct btrfs_disk_key disk_key; 46045f39d397SChris Mason 46055f39d397SChris Mason btrfs_item_key(leaf, &disk_key, 0); 4606b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 4607aa5d6bedSChris Mason } 4608aa5d6bedSChris Mason 46097c4063d1SFilipe Manana /* 46107c4063d1SFilipe Manana * Try to delete the leaf if it is mostly empty. We do this by 46117c4063d1SFilipe Manana * trying to move all its items into its left and right neighbours. 46127c4063d1SFilipe Manana * If we can't move all the items, then we don't delete it - it's 46137c4063d1SFilipe Manana * not ideal, but future insertions might fill the leaf with more 46147c4063d1SFilipe Manana * items, or items from other leaves might be moved later into our 46157c4063d1SFilipe Manana * leaf due to deletions on those leaves. 46167c4063d1SFilipe Manana */ 46170b246afaSJeff Mahoney if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { 46187c4063d1SFilipe Manana u32 min_push_space; 46197c4063d1SFilipe Manana 4620be0e5c09SChris Mason /* push_leaf_left fixes the path. 4621be0e5c09SChris Mason * make sure the path still points to our leaf 4622016f9d0bSJosef Bacik * for possible call to btrfs_del_ptr below 4623be0e5c09SChris Mason */ 46244920c9acSChris Mason slot = path->slots[1]; 462567439dadSDavid Sterba atomic_inc(&leaf->refs); 46267c4063d1SFilipe Manana /* 46277c4063d1SFilipe Manana * We want to be able to at least push one item to the 46287c4063d1SFilipe Manana * left neighbour leaf, and that's the first item. 46297c4063d1SFilipe Manana */ 46307c4063d1SFilipe Manana min_push_space = sizeof(struct btrfs_item) + 46317c4063d1SFilipe Manana btrfs_item_size(leaf, 0); 46327c4063d1SFilipe Manana wret = push_leaf_left(trans, root, path, 0, 46337c4063d1SFilipe Manana min_push_space, 1, (u32)-1); 463454aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4635aa5d6bedSChris Mason ret = wret; 46365f39d397SChris Mason 46375f39d397SChris Mason if (path->nodes[0] == leaf && 46385f39d397SChris Mason btrfs_header_nritems(leaf)) { 46397c4063d1SFilipe Manana /* 46407c4063d1SFilipe Manana * If we were not able to push all items from our 46417c4063d1SFilipe Manana * leaf to its left neighbour, then attempt to 46427c4063d1SFilipe Manana * either push all the remaining items to the 46437c4063d1SFilipe Manana * right neighbour or none. There's no advantage 46447c4063d1SFilipe Manana * in pushing only some items, instead of all, as 46457c4063d1SFilipe Manana * it's pointless to end up with a leaf having 46467c4063d1SFilipe Manana * too few items while the neighbours can be full 46477c4063d1SFilipe Manana * or nearly full. 46487c4063d1SFilipe Manana */ 46497c4063d1SFilipe Manana nritems = btrfs_header_nritems(leaf); 46507c4063d1SFilipe Manana min_push_space = leaf_space_used(leaf, 0, nritems); 46517c4063d1SFilipe Manana wret = push_leaf_right(trans, root, path, 0, 46527c4063d1SFilipe Manana min_push_space, 1, 0); 465354aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4654aa5d6bedSChris Mason ret = wret; 4655aa5d6bedSChris Mason } 46565f39d397SChris Mason 46575f39d397SChris Mason if (btrfs_header_nritems(leaf) == 0) { 4658323ac95bSChris Mason path->slots[1] = slot; 4659751a2761SFilipe Manana ret = btrfs_del_leaf(trans, root, path, leaf); 4660751a2761SFilipe Manana if (ret < 0) 4661751a2761SFilipe Manana return ret; 46625f39d397SChris Mason free_extent_buffer(leaf); 4663143bede5SJeff Mahoney ret = 0; 46645de08d7dSChris Mason } else { 4665925baeddSChris Mason /* if we're still in the path, make sure 4666925baeddSChris Mason * we're dirty. Otherwise, one of the 4667925baeddSChris Mason * push_leaf functions must have already 4668925baeddSChris Mason * dirtied this buffer 4669925baeddSChris Mason */ 4670925baeddSChris Mason if (path->nodes[0] == leaf) 46715f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 46725f39d397SChris Mason free_extent_buffer(leaf); 4673be0e5c09SChris Mason } 4674d5719762SChris Mason } else { 46755f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 4676be0e5c09SChris Mason } 46779a8dd150SChris Mason } 4678aa5d6bedSChris Mason return ret; 46799a8dd150SChris Mason } 46809a8dd150SChris Mason 468197571fd0SChris Mason /* 46823f157a2fSChris Mason * A helper function to walk down the tree starting at min_key, and looking 4683de78b51aSEric Sandeen * for nodes or leaves that are have a minimum transaction id. 4684de78b51aSEric Sandeen * This is used by the btree defrag code, and tree logging 46853f157a2fSChris Mason * 46863f157a2fSChris Mason * This does not cow, but it does stuff the starting key it finds back 46873f157a2fSChris Mason * into min_key, so you can call btrfs_search_slot with cow=1 on the 46883f157a2fSChris Mason * key and get a writable path. 46893f157a2fSChris Mason * 46903f157a2fSChris Mason * This honors path->lowest_level to prevent descent past a given level 46913f157a2fSChris Mason * of the tree. 46923f157a2fSChris Mason * 4693d352ac68SChris Mason * min_trans indicates the oldest transaction that you are interested 4694d352ac68SChris Mason * in walking through. Any nodes or leaves older than min_trans are 4695d352ac68SChris Mason * skipped over (without reading them). 4696d352ac68SChris Mason * 46973f157a2fSChris Mason * returns zero if something useful was found, < 0 on error and 1 if there 46983f157a2fSChris Mason * was nothing in the tree that matched the search criteria. 46993f157a2fSChris Mason */ 47003f157a2fSChris Mason int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, 4701de78b51aSEric Sandeen struct btrfs_path *path, 47023f157a2fSChris Mason u64 min_trans) 47033f157a2fSChris Mason { 47043f157a2fSChris Mason struct extent_buffer *cur; 47053f157a2fSChris Mason struct btrfs_key found_key; 47063f157a2fSChris Mason int slot; 47079652480bSYan int sret; 47083f157a2fSChris Mason u32 nritems; 47093f157a2fSChris Mason int level; 47103f157a2fSChris Mason int ret = 1; 4711f98de9b9SFilipe Manana int keep_locks = path->keep_locks; 47123f157a2fSChris Mason 4713c922b016SStefan Roesch ASSERT(!path->nowait); 4714f98de9b9SFilipe Manana path->keep_locks = 1; 47153f157a2fSChris Mason again: 4716bd681513SChris Mason cur = btrfs_read_lock_root_node(root); 47173f157a2fSChris Mason level = btrfs_header_level(cur); 4718e02119d5SChris Mason WARN_ON(path->nodes[level]); 47193f157a2fSChris Mason path->nodes[level] = cur; 4720bd681513SChris Mason path->locks[level] = BTRFS_READ_LOCK; 47213f157a2fSChris Mason 47223f157a2fSChris Mason if (btrfs_header_generation(cur) < min_trans) { 47233f157a2fSChris Mason ret = 1; 47243f157a2fSChris Mason goto out; 47253f157a2fSChris Mason } 47263f157a2fSChris Mason while (1) { 47273f157a2fSChris Mason nritems = btrfs_header_nritems(cur); 47283f157a2fSChris Mason level = btrfs_header_level(cur); 4729fdf8d595SAnand Jain sret = btrfs_bin_search(cur, 0, min_key, &slot); 4730cbca7d59SFilipe Manana if (sret < 0) { 4731cbca7d59SFilipe Manana ret = sret; 4732cbca7d59SFilipe Manana goto out; 4733cbca7d59SFilipe Manana } 47343f157a2fSChris Mason 4735323ac95bSChris Mason /* at the lowest level, we're done, setup the path and exit */ 4736323ac95bSChris Mason if (level == path->lowest_level) { 4737e02119d5SChris Mason if (slot >= nritems) 4738e02119d5SChris Mason goto find_next_key; 47393f157a2fSChris Mason ret = 0; 47403f157a2fSChris Mason path->slots[level] = slot; 47413f157a2fSChris Mason btrfs_item_key_to_cpu(cur, &found_key, slot); 47423f157a2fSChris Mason goto out; 47433f157a2fSChris Mason } 47449652480bSYan if (sret && slot > 0) 47459652480bSYan slot--; 47463f157a2fSChris Mason /* 4747de78b51aSEric Sandeen * check this node pointer against the min_trans parameters. 4748260db43cSRandy Dunlap * If it is too old, skip to the next one. 47493f157a2fSChris Mason */ 47503f157a2fSChris Mason while (slot < nritems) { 47513f157a2fSChris Mason u64 gen; 4752e02119d5SChris Mason 47533f157a2fSChris Mason gen = btrfs_node_ptr_generation(cur, slot); 47543f157a2fSChris Mason if (gen < min_trans) { 47553f157a2fSChris Mason slot++; 47563f157a2fSChris Mason continue; 47573f157a2fSChris Mason } 47583f157a2fSChris Mason break; 47593f157a2fSChris Mason } 4760e02119d5SChris Mason find_next_key: 47613f157a2fSChris Mason /* 47623f157a2fSChris Mason * we didn't find a candidate key in this node, walk forward 47633f157a2fSChris Mason * and find another one 47643f157a2fSChris Mason */ 47653f157a2fSChris Mason if (slot >= nritems) { 4766e02119d5SChris Mason path->slots[level] = slot; 4767e02119d5SChris Mason sret = btrfs_find_next_key(root, path, min_key, level, 4768de78b51aSEric Sandeen min_trans); 4769e02119d5SChris Mason if (sret == 0) { 4770b3b4aa74SDavid Sterba btrfs_release_path(path); 47713f157a2fSChris Mason goto again; 47723f157a2fSChris Mason } else { 47733f157a2fSChris Mason goto out; 47743f157a2fSChris Mason } 47753f157a2fSChris Mason } 47763f157a2fSChris Mason /* save our key for returning back */ 47773f157a2fSChris Mason btrfs_node_key_to_cpu(cur, &found_key, slot); 47783f157a2fSChris Mason path->slots[level] = slot; 47793f157a2fSChris Mason if (level == path->lowest_level) { 47803f157a2fSChris Mason ret = 0; 47813f157a2fSChris Mason goto out; 47823f157a2fSChris Mason } 47834b231ae4SDavid Sterba cur = btrfs_read_node_slot(cur, slot); 4784fb770ae4SLiu Bo if (IS_ERR(cur)) { 4785fb770ae4SLiu Bo ret = PTR_ERR(cur); 4786fb770ae4SLiu Bo goto out; 4787fb770ae4SLiu Bo } 47883f157a2fSChris Mason 4789bd681513SChris Mason btrfs_tree_read_lock(cur); 4790b4ce94deSChris Mason 4791bd681513SChris Mason path->locks[level - 1] = BTRFS_READ_LOCK; 47923f157a2fSChris Mason path->nodes[level - 1] = cur; 4793f7c79f30SChris Mason unlock_up(path, level, 1, 0, NULL); 47943f157a2fSChris Mason } 47953f157a2fSChris Mason out: 4796f98de9b9SFilipe Manana path->keep_locks = keep_locks; 4797f98de9b9SFilipe Manana if (ret == 0) { 4798f98de9b9SFilipe Manana btrfs_unlock_up_safe(path, path->lowest_level + 1); 4799f98de9b9SFilipe Manana memcpy(min_key, &found_key, sizeof(found_key)); 4800f98de9b9SFilipe Manana } 48013f157a2fSChris Mason return ret; 48023f157a2fSChris Mason } 48033f157a2fSChris Mason 48043f157a2fSChris Mason /* 48053f157a2fSChris Mason * this is similar to btrfs_next_leaf, but does not try to preserve 48063f157a2fSChris Mason * and fixup the path. It looks for and returns the next key in the 4807de78b51aSEric Sandeen * tree based on the current path and the min_trans parameters. 48083f157a2fSChris Mason * 48093f157a2fSChris Mason * 0 is returned if another key is found, < 0 if there are any errors 48103f157a2fSChris Mason * and 1 is returned if there are no higher keys in the tree 48113f157a2fSChris Mason * 48123f157a2fSChris Mason * path->keep_locks should be set to 1 on the search made before 48133f157a2fSChris Mason * calling this function. 48143f157a2fSChris Mason */ 4815e7a84565SChris Mason int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, 4816de78b51aSEric Sandeen struct btrfs_key *key, int level, u64 min_trans) 4817e7a84565SChris Mason { 4818e7a84565SChris Mason int slot; 4819e7a84565SChris Mason struct extent_buffer *c; 4820e7a84565SChris Mason 48216a9fb468SJosef Bacik WARN_ON(!path->keep_locks && !path->skip_locking); 4822e7a84565SChris Mason while (level < BTRFS_MAX_LEVEL) { 4823e7a84565SChris Mason if (!path->nodes[level]) 4824e7a84565SChris Mason return 1; 4825e7a84565SChris Mason 4826e7a84565SChris Mason slot = path->slots[level] + 1; 4827e7a84565SChris Mason c = path->nodes[level]; 48283f157a2fSChris Mason next: 4829e7a84565SChris Mason if (slot >= btrfs_header_nritems(c)) { 483033c66f43SYan Zheng int ret; 483133c66f43SYan Zheng int orig_lowest; 483233c66f43SYan Zheng struct btrfs_key cur_key; 483333c66f43SYan Zheng if (level + 1 >= BTRFS_MAX_LEVEL || 483433c66f43SYan Zheng !path->nodes[level + 1]) 4835e7a84565SChris Mason return 1; 483633c66f43SYan Zheng 48376a9fb468SJosef Bacik if (path->locks[level + 1] || path->skip_locking) { 483833c66f43SYan Zheng level++; 4839e7a84565SChris Mason continue; 4840e7a84565SChris Mason } 484133c66f43SYan Zheng 484233c66f43SYan Zheng slot = btrfs_header_nritems(c) - 1; 484333c66f43SYan Zheng if (level == 0) 484433c66f43SYan Zheng btrfs_item_key_to_cpu(c, &cur_key, slot); 484533c66f43SYan Zheng else 484633c66f43SYan Zheng btrfs_node_key_to_cpu(c, &cur_key, slot); 484733c66f43SYan Zheng 484833c66f43SYan Zheng orig_lowest = path->lowest_level; 4849b3b4aa74SDavid Sterba btrfs_release_path(path); 485033c66f43SYan Zheng path->lowest_level = level; 485133c66f43SYan Zheng ret = btrfs_search_slot(NULL, root, &cur_key, path, 485233c66f43SYan Zheng 0, 0); 485333c66f43SYan Zheng path->lowest_level = orig_lowest; 485433c66f43SYan Zheng if (ret < 0) 485533c66f43SYan Zheng return ret; 485633c66f43SYan Zheng 485733c66f43SYan Zheng c = path->nodes[level]; 485833c66f43SYan Zheng slot = path->slots[level]; 485933c66f43SYan Zheng if (ret == 0) 486033c66f43SYan Zheng slot++; 486133c66f43SYan Zheng goto next; 486233c66f43SYan Zheng } 486333c66f43SYan Zheng 4864e7a84565SChris Mason if (level == 0) 4865e7a84565SChris Mason btrfs_item_key_to_cpu(c, key, slot); 48663f157a2fSChris Mason else { 48673f157a2fSChris Mason u64 gen = btrfs_node_ptr_generation(c, slot); 48683f157a2fSChris Mason 48693f157a2fSChris Mason if (gen < min_trans) { 48703f157a2fSChris Mason slot++; 48713f157a2fSChris Mason goto next; 48723f157a2fSChris Mason } 4873e7a84565SChris Mason btrfs_node_key_to_cpu(c, key, slot); 48743f157a2fSChris Mason } 4875e7a84565SChris Mason return 0; 4876e7a84565SChris Mason } 4877e7a84565SChris Mason return 1; 4878e7a84565SChris Mason } 4879e7a84565SChris Mason 48803d7806ecSJan Schmidt int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, 48813d7806ecSJan Schmidt u64 time_seq) 48823d7806ecSJan Schmidt { 4883d97e63b6SChris Mason int slot; 48848e73f275SChris Mason int level; 48855f39d397SChris Mason struct extent_buffer *c; 48868e73f275SChris Mason struct extent_buffer *next; 4887d96b3424SFilipe Manana struct btrfs_fs_info *fs_info = root->fs_info; 4888925baeddSChris Mason struct btrfs_key key; 4889d96b3424SFilipe Manana bool need_commit_sem = false; 4890925baeddSChris Mason u32 nritems; 4891925baeddSChris Mason int ret; 48920e46318dSJosef Bacik int i; 4893925baeddSChris Mason 4894bdcdd86cSFilipe Manana /* 4895bdcdd86cSFilipe Manana * The nowait semantics are used only for write paths, where we don't 4896bdcdd86cSFilipe Manana * use the tree mod log and sequence numbers. 4897bdcdd86cSFilipe Manana */ 4898bdcdd86cSFilipe Manana if (time_seq) 4899c922b016SStefan Roesch ASSERT(!path->nowait); 4900c922b016SStefan Roesch 4901925baeddSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 4902d397712bSChris Mason if (nritems == 0) 4903925baeddSChris Mason return 1; 4904925baeddSChris Mason 49058e73f275SChris Mason btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); 49068e73f275SChris Mason again: 49078e73f275SChris Mason level = 1; 49088e73f275SChris Mason next = NULL; 4909b3b4aa74SDavid Sterba btrfs_release_path(path); 49108e73f275SChris Mason 4911a2135011SChris Mason path->keep_locks = 1; 49128e73f275SChris Mason 4913d96b3424SFilipe Manana if (time_seq) { 49143d7806ecSJan Schmidt ret = btrfs_search_old_slot(root, &key, path, time_seq); 4915d96b3424SFilipe Manana } else { 4916d96b3424SFilipe Manana if (path->need_commit_sem) { 4917d96b3424SFilipe Manana path->need_commit_sem = 0; 4918d96b3424SFilipe Manana need_commit_sem = true; 4919bdcdd86cSFilipe Manana if (path->nowait) { 4920bdcdd86cSFilipe Manana if (!down_read_trylock(&fs_info->commit_root_sem)) { 4921bdcdd86cSFilipe Manana ret = -EAGAIN; 4922bdcdd86cSFilipe Manana goto done; 4923bdcdd86cSFilipe Manana } 4924bdcdd86cSFilipe Manana } else { 4925d96b3424SFilipe Manana down_read(&fs_info->commit_root_sem); 4926d96b3424SFilipe Manana } 4927bdcdd86cSFilipe Manana } 4928925baeddSChris Mason ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4929d96b3424SFilipe Manana } 4930925baeddSChris Mason path->keep_locks = 0; 4931925baeddSChris Mason 4932925baeddSChris Mason if (ret < 0) 4933d96b3424SFilipe Manana goto done; 4934925baeddSChris Mason 4935a2135011SChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 4936168fd7d2SChris Mason /* 4937168fd7d2SChris Mason * by releasing the path above we dropped all our locks. A balance 4938168fd7d2SChris Mason * could have added more items next to the key that used to be 4939168fd7d2SChris Mason * at the very end of the block. So, check again here and 4940168fd7d2SChris Mason * advance the path if there are now more items available. 4941168fd7d2SChris Mason */ 4942a2135011SChris Mason if (nritems > 0 && path->slots[0] < nritems - 1) { 4943e457afecSYan Zheng if (ret == 0) 4944168fd7d2SChris Mason path->slots[0]++; 49458e73f275SChris Mason ret = 0; 4946925baeddSChris Mason goto done; 4947925baeddSChris Mason } 49480b43e04fSLiu Bo /* 49490b43e04fSLiu Bo * So the above check misses one case: 49500b43e04fSLiu Bo * - after releasing the path above, someone has removed the item that 49510b43e04fSLiu Bo * used to be at the very end of the block, and balance between leafs 49520b43e04fSLiu Bo * gets another one with bigger key.offset to replace it. 49530b43e04fSLiu Bo * 49540b43e04fSLiu Bo * This one should be returned as well, or we can get leaf corruption 49550b43e04fSLiu Bo * later(esp. in __btrfs_drop_extents()). 49560b43e04fSLiu Bo * 49570b43e04fSLiu Bo * And a bit more explanation about this check, 49580b43e04fSLiu Bo * with ret > 0, the key isn't found, the path points to the slot 49590b43e04fSLiu Bo * where it should be inserted, so the path->slots[0] item must be the 49600b43e04fSLiu Bo * bigger one. 49610b43e04fSLiu Bo */ 49620b43e04fSLiu Bo if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { 49630b43e04fSLiu Bo ret = 0; 49640b43e04fSLiu Bo goto done; 49650b43e04fSLiu Bo } 4966d97e63b6SChris Mason 4967234b63a0SChris Mason while (level < BTRFS_MAX_LEVEL) { 49688e73f275SChris Mason if (!path->nodes[level]) { 49698e73f275SChris Mason ret = 1; 49708e73f275SChris Mason goto done; 49718e73f275SChris Mason } 49725f39d397SChris Mason 4973d97e63b6SChris Mason slot = path->slots[level] + 1; 4974d97e63b6SChris Mason c = path->nodes[level]; 49755f39d397SChris Mason if (slot >= btrfs_header_nritems(c)) { 4976d97e63b6SChris Mason level++; 49778e73f275SChris Mason if (level == BTRFS_MAX_LEVEL) { 49788e73f275SChris Mason ret = 1; 49798e73f275SChris Mason goto done; 49808e73f275SChris Mason } 4981d97e63b6SChris Mason continue; 4982d97e63b6SChris Mason } 49835f39d397SChris Mason 49840e46318dSJosef Bacik 49850e46318dSJosef Bacik /* 49860e46318dSJosef Bacik * Our current level is where we're going to start from, and to 49870e46318dSJosef Bacik * make sure lockdep doesn't complain we need to drop our locks 49880e46318dSJosef Bacik * and nodes from 0 to our current level. 49890e46318dSJosef Bacik */ 49900e46318dSJosef Bacik for (i = 0; i < level; i++) { 49910e46318dSJosef Bacik if (path->locks[level]) { 49920e46318dSJosef Bacik btrfs_tree_read_unlock(path->nodes[i]); 49930e46318dSJosef Bacik path->locks[i] = 0; 49940e46318dSJosef Bacik } 49950e46318dSJosef Bacik free_extent_buffer(path->nodes[i]); 49960e46318dSJosef Bacik path->nodes[i] = NULL; 4997925baeddSChris Mason } 49985f39d397SChris Mason 49998e73f275SChris Mason next = c; 5000d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 5001cda79c54SDavid Sterba slot, &key); 5002bdcdd86cSFilipe Manana if (ret == -EAGAIN && !path->nowait) 50038e73f275SChris Mason goto again; 50045f39d397SChris Mason 500576a05b35SChris Mason if (ret < 0) { 5006b3b4aa74SDavid Sterba btrfs_release_path(path); 500776a05b35SChris Mason goto done; 500876a05b35SChris Mason } 500976a05b35SChris Mason 50105cd57b2cSChris Mason if (!path->skip_locking) { 5011bd681513SChris Mason ret = btrfs_try_tree_read_lock(next); 5012bdcdd86cSFilipe Manana if (!ret && path->nowait) { 5013bdcdd86cSFilipe Manana ret = -EAGAIN; 5014bdcdd86cSFilipe Manana goto done; 5015bdcdd86cSFilipe Manana } 5016d42244a0SJan Schmidt if (!ret && time_seq) { 5017d42244a0SJan Schmidt /* 5018d42244a0SJan Schmidt * If we don't get the lock, we may be racing 5019d42244a0SJan Schmidt * with push_leaf_left, holding that lock while 5020d42244a0SJan Schmidt * itself waiting for the leaf we've currently 5021d42244a0SJan Schmidt * locked. To solve this situation, we give up 5022d42244a0SJan Schmidt * on our lock and cycle. 5023d42244a0SJan Schmidt */ 5024cf538830SJan Schmidt free_extent_buffer(next); 5025d42244a0SJan Schmidt btrfs_release_path(path); 5026d42244a0SJan Schmidt cond_resched(); 5027d42244a0SJan Schmidt goto again; 5028d42244a0SJan Schmidt } 50290e46318dSJosef Bacik if (!ret) 50300e46318dSJosef Bacik btrfs_tree_read_lock(next); 5031bd681513SChris Mason } 5032d97e63b6SChris Mason break; 5033d97e63b6SChris Mason } 5034d97e63b6SChris Mason path->slots[level] = slot; 5035d97e63b6SChris Mason while (1) { 5036d97e63b6SChris Mason level--; 5037d97e63b6SChris Mason path->nodes[level] = next; 5038d97e63b6SChris Mason path->slots[level] = 0; 5039a74a4b97SChris Mason if (!path->skip_locking) 5040ffeb03cfSJosef Bacik path->locks[level] = BTRFS_READ_LOCK; 5041d97e63b6SChris Mason if (!level) 5042d97e63b6SChris Mason break; 5043b4ce94deSChris Mason 5044d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 5045cda79c54SDavid Sterba 0, &key); 5046bdcdd86cSFilipe Manana if (ret == -EAGAIN && !path->nowait) 50478e73f275SChris Mason goto again; 50488e73f275SChris Mason 504976a05b35SChris Mason if (ret < 0) { 5050b3b4aa74SDavid Sterba btrfs_release_path(path); 505176a05b35SChris Mason goto done; 505276a05b35SChris Mason } 505376a05b35SChris Mason 5054bdcdd86cSFilipe Manana if (!path->skip_locking) { 5055bdcdd86cSFilipe Manana if (path->nowait) { 5056bdcdd86cSFilipe Manana if (!btrfs_try_tree_read_lock(next)) { 5057bdcdd86cSFilipe Manana ret = -EAGAIN; 5058bdcdd86cSFilipe Manana goto done; 5059bdcdd86cSFilipe Manana } 5060bdcdd86cSFilipe Manana } else { 50610e46318dSJosef Bacik btrfs_tree_read_lock(next); 5062d97e63b6SChris Mason } 5063bdcdd86cSFilipe Manana } 5064bdcdd86cSFilipe Manana } 50658e73f275SChris Mason ret = 0; 5066925baeddSChris Mason done: 5067f7c79f30SChris Mason unlock_up(path, 0, 1, 0, NULL); 5068d96b3424SFilipe Manana if (need_commit_sem) { 5069d96b3424SFilipe Manana int ret2; 5070d96b3424SFilipe Manana 5071d96b3424SFilipe Manana path->need_commit_sem = 1; 5072d96b3424SFilipe Manana ret2 = finish_need_commit_sem_search(path); 5073d96b3424SFilipe Manana up_read(&fs_info->commit_root_sem); 5074d96b3424SFilipe Manana if (ret2) 5075d96b3424SFilipe Manana ret = ret2; 5076d96b3424SFilipe Manana } 50778e73f275SChris Mason 50788e73f275SChris Mason return ret; 5079d97e63b6SChris Mason } 50800b86a832SChris Mason 5081890d2b1aSJosef Bacik int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq) 5082890d2b1aSJosef Bacik { 5083890d2b1aSJosef Bacik path->slots[0]++; 5084890d2b1aSJosef Bacik if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) 5085890d2b1aSJosef Bacik return btrfs_next_old_leaf(root, path, time_seq); 5086890d2b1aSJosef Bacik return 0; 5087890d2b1aSJosef Bacik } 5088890d2b1aSJosef Bacik 50893f157a2fSChris Mason /* 50903f157a2fSChris Mason * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps 50913f157a2fSChris Mason * searching until it gets past min_objectid or finds an item of 'type' 50923f157a2fSChris Mason * 50933f157a2fSChris Mason * returns 0 if something is found, 1 if nothing was found and < 0 on error 50943f157a2fSChris Mason */ 50950b86a832SChris Mason int btrfs_previous_item(struct btrfs_root *root, 50960b86a832SChris Mason struct btrfs_path *path, u64 min_objectid, 50970b86a832SChris Mason int type) 50980b86a832SChris Mason { 50990b86a832SChris Mason struct btrfs_key found_key; 51000b86a832SChris Mason struct extent_buffer *leaf; 5101e02119d5SChris Mason u32 nritems; 51020b86a832SChris Mason int ret; 51030b86a832SChris Mason 51040b86a832SChris Mason while (1) { 51050b86a832SChris Mason if (path->slots[0] == 0) { 51060b86a832SChris Mason ret = btrfs_prev_leaf(root, path); 51070b86a832SChris Mason if (ret != 0) 51080b86a832SChris Mason return ret; 51090b86a832SChris Mason } else { 51100b86a832SChris Mason path->slots[0]--; 51110b86a832SChris Mason } 51120b86a832SChris Mason leaf = path->nodes[0]; 5113e02119d5SChris Mason nritems = btrfs_header_nritems(leaf); 5114e02119d5SChris Mason if (nritems == 0) 5115e02119d5SChris Mason return 1; 5116e02119d5SChris Mason if (path->slots[0] == nritems) 5117e02119d5SChris Mason path->slots[0]--; 5118e02119d5SChris Mason 51190b86a832SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 5120e02119d5SChris Mason if (found_key.objectid < min_objectid) 5121e02119d5SChris Mason break; 51220a4eefbbSYan Zheng if (found_key.type == type) 51230a4eefbbSYan Zheng return 0; 5124e02119d5SChris Mason if (found_key.objectid == min_objectid && 5125e02119d5SChris Mason found_key.type < type) 5126e02119d5SChris Mason break; 51270b86a832SChris Mason } 51280b86a832SChris Mason return 1; 51290b86a832SChris Mason } 5130ade2e0b3SWang Shilong 5131ade2e0b3SWang Shilong /* 5132ade2e0b3SWang Shilong * search in extent tree to find a previous Metadata/Data extent item with 5133ade2e0b3SWang Shilong * min objecitd. 5134ade2e0b3SWang Shilong * 5135ade2e0b3SWang Shilong * returns 0 if something is found, 1 if nothing was found and < 0 on error 5136ade2e0b3SWang Shilong */ 5137ade2e0b3SWang Shilong int btrfs_previous_extent_item(struct btrfs_root *root, 5138ade2e0b3SWang Shilong struct btrfs_path *path, u64 min_objectid) 5139ade2e0b3SWang Shilong { 5140ade2e0b3SWang Shilong struct btrfs_key found_key; 5141ade2e0b3SWang Shilong struct extent_buffer *leaf; 5142ade2e0b3SWang Shilong u32 nritems; 5143ade2e0b3SWang Shilong int ret; 5144ade2e0b3SWang Shilong 5145ade2e0b3SWang Shilong while (1) { 5146ade2e0b3SWang Shilong if (path->slots[0] == 0) { 5147ade2e0b3SWang Shilong ret = btrfs_prev_leaf(root, path); 5148ade2e0b3SWang Shilong if (ret != 0) 5149ade2e0b3SWang Shilong return ret; 5150ade2e0b3SWang Shilong } else { 5151ade2e0b3SWang Shilong path->slots[0]--; 5152ade2e0b3SWang Shilong } 5153ade2e0b3SWang Shilong leaf = path->nodes[0]; 5154ade2e0b3SWang Shilong nritems = btrfs_header_nritems(leaf); 5155ade2e0b3SWang Shilong if (nritems == 0) 5156ade2e0b3SWang Shilong return 1; 5157ade2e0b3SWang Shilong if (path->slots[0] == nritems) 5158ade2e0b3SWang Shilong path->slots[0]--; 5159ade2e0b3SWang Shilong 5160ade2e0b3SWang Shilong btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 5161ade2e0b3SWang Shilong if (found_key.objectid < min_objectid) 5162ade2e0b3SWang Shilong break; 5163ade2e0b3SWang Shilong if (found_key.type == BTRFS_EXTENT_ITEM_KEY || 5164ade2e0b3SWang Shilong found_key.type == BTRFS_METADATA_ITEM_KEY) 5165ade2e0b3SWang Shilong return 0; 5166ade2e0b3SWang Shilong if (found_key.objectid == min_objectid && 5167ade2e0b3SWang Shilong found_key.type < BTRFS_EXTENT_ITEM_KEY) 5168ade2e0b3SWang Shilong break; 5169ade2e0b3SWang Shilong } 5170ade2e0b3SWang Shilong return 1; 5171ade2e0b3SWang Shilong } 5172226463d7SJosef Bacik 5173226463d7SJosef Bacik int __init btrfs_ctree_init(void) 5174226463d7SJosef Bacik { 5175226463d7SJosef Bacik btrfs_path_cachep = kmem_cache_create("btrfs_path", 5176226463d7SJosef Bacik sizeof(struct btrfs_path), 0, 5177226463d7SJosef Bacik SLAB_MEM_SPREAD, NULL); 5178226463d7SJosef Bacik if (!btrfs_path_cachep) 5179226463d7SJosef Bacik return -ENOMEM; 5180226463d7SJosef Bacik return 0; 5181226463d7SJosef Bacik } 5182226463d7SJosef Bacik 5183226463d7SJosef Bacik void __cold btrfs_ctree_exit(void) 5184226463d7SJosef Bacik { 5185226463d7SJosef Bacik kmem_cache_destroy(btrfs_path_cachep); 5186226463d7SJosef Bacik } 5187