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> 10eb60ceacSChris Mason #include "ctree.h" 11eb60ceacSChris Mason #include "disk-io.h" 127f5c1516SChris Mason #include "transaction.h" 135f39d397SChris Mason #include "print-tree.h" 14925baeddSChris Mason #include "locking.h" 15de37aa51SNikolay Borisov #include "volumes.h" 16f616f5cdSQu Wenruo #include "qgroup.h" 179a8dd150SChris Mason 18e089f05cSChris Mason static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root 19e089f05cSChris Mason *root, struct btrfs_path *path, int level); 20310712b2SOmar Sandoval static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, 21310712b2SOmar Sandoval const struct btrfs_key *ins_key, struct btrfs_path *path, 22310712b2SOmar Sandoval int data_size, int extend); 235f39d397SChris Mason static int push_node_left(struct btrfs_trans_handle *trans, 242ff7e61eSJeff Mahoney struct extent_buffer *dst, 25971a1f66SChris Mason struct extent_buffer *src, int empty); 265f39d397SChris Mason static int balance_node_right(struct btrfs_trans_handle *trans, 275f39d397SChris Mason struct extent_buffer *dst_buf, 285f39d397SChris Mason struct extent_buffer *src_buf); 29afe5fea7STsutomu Itoh static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, 30afe5fea7STsutomu Itoh int level, int slot); 31d97e63b6SChris Mason 32af024ed2SJohannes Thumshirn static const struct btrfs_csums { 33af024ed2SJohannes Thumshirn u16 size; 3459a0fcdbSDavid Sterba const char name[10]; 3559a0fcdbSDavid Sterba const char driver[12]; 36af024ed2SJohannes Thumshirn } btrfs_csums[] = { 37af024ed2SJohannes Thumshirn [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, 383951e7f0SJohannes Thumshirn [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" }, 393831bf00SJohannes Thumshirn [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" }, 40352ae07bSDavid Sterba [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b", 41352ae07bSDavid Sterba .driver = "blake2b-256" }, 42af024ed2SJohannes Thumshirn }; 43af024ed2SJohannes Thumshirn 44af024ed2SJohannes Thumshirn int btrfs_super_csum_size(const struct btrfs_super_block *s) 45af024ed2SJohannes Thumshirn { 46af024ed2SJohannes Thumshirn u16 t = btrfs_super_csum_type(s); 47af024ed2SJohannes Thumshirn /* 48af024ed2SJohannes Thumshirn * csum type is validated at mount time 49af024ed2SJohannes Thumshirn */ 50af024ed2SJohannes Thumshirn return btrfs_csums[t].size; 51af024ed2SJohannes Thumshirn } 52af024ed2SJohannes Thumshirn 53af024ed2SJohannes Thumshirn const char *btrfs_super_csum_name(u16 csum_type) 54af024ed2SJohannes Thumshirn { 55af024ed2SJohannes Thumshirn /* csum type is validated at mount time */ 56af024ed2SJohannes Thumshirn return btrfs_csums[csum_type].name; 57af024ed2SJohannes Thumshirn } 58af024ed2SJohannes Thumshirn 59b4e967beSDavid Sterba /* 60b4e967beSDavid Sterba * Return driver name if defined, otherwise the name that's also a valid driver 61b4e967beSDavid Sterba * name 62b4e967beSDavid Sterba */ 63b4e967beSDavid Sterba const char *btrfs_super_csum_driver(u16 csum_type) 64b4e967beSDavid Sterba { 65b4e967beSDavid Sterba /* csum type is validated at mount time */ 6659a0fcdbSDavid Sterba return btrfs_csums[csum_type].driver[0] ? 6759a0fcdbSDavid Sterba btrfs_csums[csum_type].driver : 68b4e967beSDavid Sterba btrfs_csums[csum_type].name; 69b4e967beSDavid Sterba } 70b4e967beSDavid Sterba 71604997b4SDavid Sterba size_t __attribute_const__ btrfs_get_num_csums(void) 72f7cea56cSDavid Sterba { 73f7cea56cSDavid Sterba return ARRAY_SIZE(btrfs_csums); 74f7cea56cSDavid Sterba } 75f7cea56cSDavid Sterba 762c90e5d6SChris Mason struct btrfs_path *btrfs_alloc_path(void) 772c90e5d6SChris Mason { 78e2c89907SMasahiro Yamada return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); 792c90e5d6SChris Mason } 802c90e5d6SChris Mason 81d352ac68SChris Mason /* this also releases the path */ 822c90e5d6SChris Mason void btrfs_free_path(struct btrfs_path *p) 832c90e5d6SChris Mason { 84ff175d57SJesper Juhl if (!p) 85ff175d57SJesper Juhl return; 86b3b4aa74SDavid Sterba btrfs_release_path(p); 872c90e5d6SChris Mason kmem_cache_free(btrfs_path_cachep, p); 882c90e5d6SChris Mason } 892c90e5d6SChris Mason 90d352ac68SChris Mason /* 91d352ac68SChris Mason * path release drops references on the extent buffers in the path 92d352ac68SChris Mason * and it drops any locks held by this path 93d352ac68SChris Mason * 94d352ac68SChris Mason * It is safe to call this on paths that no locks or extent buffers held. 95d352ac68SChris Mason */ 96b3b4aa74SDavid Sterba noinline void btrfs_release_path(struct btrfs_path *p) 97eb60ceacSChris Mason { 98eb60ceacSChris Mason int i; 99a2135011SChris Mason 100234b63a0SChris Mason for (i = 0; i < BTRFS_MAX_LEVEL; i++) { 1013f157a2fSChris Mason p->slots[i] = 0; 102eb60ceacSChris Mason if (!p->nodes[i]) 103925baeddSChris Mason continue; 104925baeddSChris Mason if (p->locks[i]) { 105bd681513SChris Mason btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); 106925baeddSChris Mason p->locks[i] = 0; 107925baeddSChris Mason } 1085f39d397SChris Mason free_extent_buffer(p->nodes[i]); 1093f157a2fSChris Mason p->nodes[i] = NULL; 110eb60ceacSChris Mason } 111eb60ceacSChris Mason } 112eb60ceacSChris Mason 113d352ac68SChris Mason /* 114d352ac68SChris Mason * safely gets a reference on the root node of a tree. A lock 115d352ac68SChris Mason * is not taken, so a concurrent writer may put a different node 116d352ac68SChris Mason * at the root of the tree. See btrfs_lock_root_node for the 117d352ac68SChris Mason * looping required. 118d352ac68SChris Mason * 119d352ac68SChris Mason * The extent buffer returned by this has a reference taken, so 120d352ac68SChris Mason * it won't disappear. It may stop being the root of the tree 121d352ac68SChris Mason * at any time because there are no locks held. 122d352ac68SChris Mason */ 123925baeddSChris Mason struct extent_buffer *btrfs_root_node(struct btrfs_root *root) 124925baeddSChris Mason { 125925baeddSChris Mason struct extent_buffer *eb; 126240f62c8SChris Mason 1273083ee2eSJosef Bacik while (1) { 128240f62c8SChris Mason rcu_read_lock(); 129240f62c8SChris Mason eb = rcu_dereference(root->node); 1303083ee2eSJosef Bacik 1313083ee2eSJosef Bacik /* 1323083ee2eSJosef Bacik * RCU really hurts here, we could free up the root node because 13301327610SNicholas D Steeves * it was COWed but we may not get the new root node yet so do 1343083ee2eSJosef Bacik * the inc_not_zero dance and if it doesn't work then 1353083ee2eSJosef Bacik * synchronize_rcu and try again. 1363083ee2eSJosef Bacik */ 1373083ee2eSJosef Bacik if (atomic_inc_not_zero(&eb->refs)) { 138240f62c8SChris Mason rcu_read_unlock(); 1393083ee2eSJosef Bacik break; 1403083ee2eSJosef Bacik } 1413083ee2eSJosef Bacik rcu_read_unlock(); 1423083ee2eSJosef Bacik synchronize_rcu(); 1433083ee2eSJosef Bacik } 144925baeddSChris Mason return eb; 145925baeddSChris Mason } 146925baeddSChris Mason 14792a7cc42SQu Wenruo /* 14892a7cc42SQu Wenruo * Cowonly root (not-shareable trees, everything not subvolume or reloc roots), 14992a7cc42SQu Wenruo * just get put onto a simple dirty list. Transaction walks this list to make 15092a7cc42SQu Wenruo * sure they get properly updated on disk. 151d352ac68SChris Mason */ 1520b86a832SChris Mason static void add_root_to_dirty_list(struct btrfs_root *root) 1530b86a832SChris Mason { 1540b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1550b246afaSJeff Mahoney 156e7070be1SJosef Bacik if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || 157e7070be1SJosef Bacik !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) 158e7070be1SJosef Bacik return; 159e7070be1SJosef Bacik 1600b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 161e7070be1SJosef Bacik if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { 162e7070be1SJosef Bacik /* Want the extent tree to be the last on the list */ 1634fd786e6SMisono Tomohiro if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) 164e7070be1SJosef Bacik list_move_tail(&root->dirty_list, 1650b246afaSJeff Mahoney &fs_info->dirty_cowonly_roots); 166e7070be1SJosef Bacik else 167e7070be1SJosef Bacik list_move(&root->dirty_list, 1680b246afaSJeff Mahoney &fs_info->dirty_cowonly_roots); 1690b86a832SChris Mason } 1700b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1710b86a832SChris Mason } 1720b86a832SChris Mason 173d352ac68SChris Mason /* 174d352ac68SChris Mason * used by snapshot creation to make a copy of a root for a tree with 175d352ac68SChris Mason * a given objectid. The buffer with the new root node is returned in 176d352ac68SChris Mason * cow_ret, and this func returns zero on success or a negative error code. 177d352ac68SChris Mason */ 178be20aa9dSChris Mason int btrfs_copy_root(struct btrfs_trans_handle *trans, 179be20aa9dSChris Mason struct btrfs_root *root, 180be20aa9dSChris Mason struct extent_buffer *buf, 181be20aa9dSChris Mason struct extent_buffer **cow_ret, u64 new_root_objectid) 182be20aa9dSChris Mason { 1830b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 184be20aa9dSChris Mason struct extent_buffer *cow; 185be20aa9dSChris Mason int ret = 0; 186be20aa9dSChris Mason int level; 1875d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 188be20aa9dSChris Mason 18992a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 1900b246afaSJeff Mahoney trans->transid != fs_info->running_transaction->transid); 19192a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 19227cdeb70SMiao Xie trans->transid != root->last_trans); 193be20aa9dSChris Mason 194be20aa9dSChris Mason level = btrfs_header_level(buf); 1955d4f98a2SYan Zheng if (level == 0) 1965d4f98a2SYan Zheng btrfs_item_key(buf, &disk_key, 0); 1975d4f98a2SYan Zheng else 1985d4f98a2SYan Zheng btrfs_node_key(buf, &disk_key, 0); 19931840ae1SZheng Yan 2004d75f8a9SDavid Sterba cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, 201cf6f34aaSJosef Bacik &disk_key, level, buf->start, 0, 202cf6f34aaSJosef Bacik BTRFS_NESTING_NEW_ROOT); 2035d4f98a2SYan Zheng if (IS_ERR(cow)) 204be20aa9dSChris Mason return PTR_ERR(cow); 205be20aa9dSChris Mason 20658e8012cSDavid Sterba copy_extent_buffer_full(cow, buf); 207be20aa9dSChris Mason btrfs_set_header_bytenr(cow, cow->start); 208be20aa9dSChris Mason btrfs_set_header_generation(cow, trans->transid); 2095d4f98a2SYan Zheng btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 2105d4f98a2SYan Zheng btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 2115d4f98a2SYan Zheng BTRFS_HEADER_FLAG_RELOC); 2125d4f98a2SYan Zheng if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 2135d4f98a2SYan Zheng btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 2145d4f98a2SYan Zheng else 215be20aa9dSChris Mason btrfs_set_header_owner(cow, new_root_objectid); 216be20aa9dSChris Mason 217de37aa51SNikolay Borisov write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 2182b82032cSYan Zheng 219be20aa9dSChris Mason WARN_ON(btrfs_header_generation(buf) > trans->transid); 2205d4f98a2SYan Zheng if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 221e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 2225d4f98a2SYan Zheng else 223e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 224867ed321SJosef Bacik if (ret) { 225*72c9925fSFilipe Manana btrfs_tree_unlock(cow); 226*72c9925fSFilipe Manana free_extent_buffer(cow); 227867ed321SJosef Bacik btrfs_abort_transaction(trans, ret); 228be20aa9dSChris Mason return ret; 229867ed321SJosef Bacik } 230be20aa9dSChris Mason 231be20aa9dSChris Mason btrfs_mark_buffer_dirty(cow); 232be20aa9dSChris Mason *cow_ret = cow; 233be20aa9dSChris Mason return 0; 234be20aa9dSChris Mason } 235be20aa9dSChris Mason 236bd989ba3SJan Schmidt enum mod_log_op { 237bd989ba3SJan Schmidt MOD_LOG_KEY_REPLACE, 238bd989ba3SJan Schmidt MOD_LOG_KEY_ADD, 239bd989ba3SJan Schmidt MOD_LOG_KEY_REMOVE, 240bd989ba3SJan Schmidt MOD_LOG_KEY_REMOVE_WHILE_FREEING, 241bd989ba3SJan Schmidt MOD_LOG_KEY_REMOVE_WHILE_MOVING, 242bd989ba3SJan Schmidt MOD_LOG_MOVE_KEYS, 243bd989ba3SJan Schmidt MOD_LOG_ROOT_REPLACE, 244bd989ba3SJan Schmidt }; 245bd989ba3SJan Schmidt 246bd989ba3SJan Schmidt struct tree_mod_root { 247bd989ba3SJan Schmidt u64 logical; 248bd989ba3SJan Schmidt u8 level; 249bd989ba3SJan Schmidt }; 250bd989ba3SJan Schmidt 251bd989ba3SJan Schmidt struct tree_mod_elem { 252bd989ba3SJan Schmidt struct rb_node node; 253298cfd36SChandan Rajendra u64 logical; 254097b8a7cSJan Schmidt u64 seq; 255bd989ba3SJan Schmidt enum mod_log_op op; 256bd989ba3SJan Schmidt 257bd989ba3SJan Schmidt /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ 258bd989ba3SJan Schmidt int slot; 259bd989ba3SJan Schmidt 260bd989ba3SJan Schmidt /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ 261bd989ba3SJan Schmidt u64 generation; 262bd989ba3SJan Schmidt 263bd989ba3SJan Schmidt /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ 264bd989ba3SJan Schmidt struct btrfs_disk_key key; 265bd989ba3SJan Schmidt u64 blockptr; 266bd989ba3SJan Schmidt 267bd989ba3SJan Schmidt /* this is used for op == MOD_LOG_MOVE_KEYS */ 268b6dfa35bSDavid Sterba struct { 269b6dfa35bSDavid Sterba int dst_slot; 270b6dfa35bSDavid Sterba int nr_items; 271b6dfa35bSDavid Sterba } move; 272bd989ba3SJan Schmidt 273bd989ba3SJan Schmidt /* this is used for op == MOD_LOG_ROOT_REPLACE */ 274bd989ba3SJan Schmidt struct tree_mod_root old_root; 275bd989ba3SJan Schmidt }; 276bd989ba3SJan Schmidt 277097b8a7cSJan Schmidt /* 278fcebe456SJosef Bacik * Pull a new tree mod seq number for our operation. 279fc36ed7eSJan Schmidt */ 280fcebe456SJosef Bacik static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) 281fc36ed7eSJan Schmidt { 282fc36ed7eSJan Schmidt return atomic64_inc_return(&fs_info->tree_mod_seq); 283fc36ed7eSJan Schmidt } 284fc36ed7eSJan Schmidt 285fc36ed7eSJan Schmidt /* 286097b8a7cSJan Schmidt * This adds a new blocker to the tree mod log's blocker list if the @elem 287097b8a7cSJan Schmidt * passed does not already have a sequence number set. So when a caller expects 288097b8a7cSJan Schmidt * to record tree modifications, it should ensure to set elem->seq to zero 289097b8a7cSJan Schmidt * before calling btrfs_get_tree_mod_seq. 290097b8a7cSJan Schmidt * Returns a fresh, unused tree log modification sequence number, even if no new 291097b8a7cSJan Schmidt * blocker was added. 292097b8a7cSJan Schmidt */ 293097b8a7cSJan Schmidt u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, 294bd989ba3SJan Schmidt struct seq_list *elem) 295bd989ba3SJan Schmidt { 296b1a09f1eSDavid Sterba write_lock(&fs_info->tree_mod_log_lock); 297097b8a7cSJan Schmidt if (!elem->seq) { 298fcebe456SJosef Bacik elem->seq = btrfs_inc_tree_mod_seq(fs_info); 299097b8a7cSJan Schmidt list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); 300097b8a7cSJan Schmidt } 301b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 302097b8a7cSJan Schmidt 303fcebe456SJosef Bacik return elem->seq; 304bd989ba3SJan Schmidt } 305bd989ba3SJan Schmidt 306bd989ba3SJan Schmidt void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, 307bd989ba3SJan Schmidt struct seq_list *elem) 308bd989ba3SJan Schmidt { 309bd989ba3SJan Schmidt struct rb_root *tm_root; 310bd989ba3SJan Schmidt struct rb_node *node; 311bd989ba3SJan Schmidt struct rb_node *next; 312bd989ba3SJan Schmidt struct tree_mod_elem *tm; 313bd989ba3SJan Schmidt u64 min_seq = (u64)-1; 314bd989ba3SJan Schmidt u64 seq_putting = elem->seq; 315bd989ba3SJan Schmidt 316bd989ba3SJan Schmidt if (!seq_putting) 317bd989ba3SJan Schmidt return; 318bd989ba3SJan Schmidt 3197227ff4dSFilipe Manana write_lock(&fs_info->tree_mod_log_lock); 320bd989ba3SJan Schmidt list_del(&elem->list); 321097b8a7cSJan Schmidt elem->seq = 0; 322bd989ba3SJan Schmidt 32342836cf4SFilipe Manana if (!list_empty(&fs_info->tree_mod_seq_list)) { 32442836cf4SFilipe Manana struct seq_list *first; 32542836cf4SFilipe Manana 32642836cf4SFilipe Manana first = list_first_entry(&fs_info->tree_mod_seq_list, 32742836cf4SFilipe Manana struct seq_list, list); 32842836cf4SFilipe Manana if (seq_putting > first->seq) { 329bd989ba3SJan Schmidt /* 33042836cf4SFilipe Manana * Blocker with lower sequence number exists, we 33142836cf4SFilipe Manana * cannot remove anything from the log. 332bd989ba3SJan Schmidt */ 3337227ff4dSFilipe Manana write_unlock(&fs_info->tree_mod_log_lock); 334097b8a7cSJan Schmidt return; 335bd989ba3SJan Schmidt } 33642836cf4SFilipe Manana min_seq = first->seq; 337bd989ba3SJan Schmidt } 338097b8a7cSJan Schmidt 339097b8a7cSJan Schmidt /* 340bd989ba3SJan Schmidt * anything that's lower than the lowest existing (read: blocked) 341bd989ba3SJan Schmidt * sequence number can be removed from the tree. 342bd989ba3SJan Schmidt */ 343bd989ba3SJan Schmidt tm_root = &fs_info->tree_mod_log; 344bd989ba3SJan Schmidt for (node = rb_first(tm_root); node; node = next) { 345bd989ba3SJan Schmidt next = rb_next(node); 3466b4df8b6SGeliang Tang tm = rb_entry(node, struct tree_mod_elem, node); 3476609fee8SFilipe Manana if (tm->seq >= min_seq) 348bd989ba3SJan Schmidt continue; 349bd989ba3SJan Schmidt rb_erase(node, tm_root); 350bd989ba3SJan Schmidt kfree(tm); 351bd989ba3SJan Schmidt } 352b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 353bd989ba3SJan Schmidt } 354bd989ba3SJan Schmidt 355bd989ba3SJan Schmidt /* 356bd989ba3SJan Schmidt * key order of the log: 357298cfd36SChandan Rajendra * node/leaf start address -> sequence 358bd989ba3SJan Schmidt * 359298cfd36SChandan Rajendra * The 'start address' is the logical address of the *new* root node 360298cfd36SChandan Rajendra * for root replace operations, or the logical address of the affected 361298cfd36SChandan Rajendra * block for all other operations. 362bd989ba3SJan Schmidt */ 363bd989ba3SJan Schmidt static noinline int 364bd989ba3SJan Schmidt __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) 365bd989ba3SJan Schmidt { 366bd989ba3SJan Schmidt struct rb_root *tm_root; 367bd989ba3SJan Schmidt struct rb_node **new; 368bd989ba3SJan Schmidt struct rb_node *parent = NULL; 369bd989ba3SJan Schmidt struct tree_mod_elem *cur; 370bd989ba3SJan Schmidt 37173e82fe4SDavid Sterba lockdep_assert_held_write(&fs_info->tree_mod_log_lock); 37273e82fe4SDavid Sterba 373fcebe456SJosef Bacik tm->seq = btrfs_inc_tree_mod_seq(fs_info); 374bd989ba3SJan Schmidt 375bd989ba3SJan Schmidt tm_root = &fs_info->tree_mod_log; 376bd989ba3SJan Schmidt new = &tm_root->rb_node; 377bd989ba3SJan Schmidt while (*new) { 3786b4df8b6SGeliang Tang cur = rb_entry(*new, struct tree_mod_elem, node); 379bd989ba3SJan Schmidt parent = *new; 380298cfd36SChandan Rajendra if (cur->logical < tm->logical) 381bd989ba3SJan Schmidt new = &((*new)->rb_left); 382298cfd36SChandan Rajendra else if (cur->logical > tm->logical) 383bd989ba3SJan Schmidt new = &((*new)->rb_right); 384097b8a7cSJan Schmidt else if (cur->seq < tm->seq) 385bd989ba3SJan Schmidt new = &((*new)->rb_left); 386097b8a7cSJan Schmidt else if (cur->seq > tm->seq) 387bd989ba3SJan Schmidt new = &((*new)->rb_right); 3885de865eeSFilipe David Borba Manana else 3895de865eeSFilipe David Borba Manana return -EEXIST; 390bd989ba3SJan Schmidt } 391bd989ba3SJan Schmidt 392bd989ba3SJan Schmidt rb_link_node(&tm->node, parent, new); 393bd989ba3SJan Schmidt rb_insert_color(&tm->node, tm_root); 3945de865eeSFilipe David Borba Manana return 0; 395bd989ba3SJan Schmidt } 396bd989ba3SJan Schmidt 397097b8a7cSJan Schmidt /* 398097b8a7cSJan Schmidt * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it 399097b8a7cSJan Schmidt * returns zero with the tree_mod_log_lock acquired. The caller must hold 400097b8a7cSJan Schmidt * this until all tree mod log insertions are recorded in the rb tree and then 401b1a09f1eSDavid Sterba * write unlock fs_info::tree_mod_log_lock. 402097b8a7cSJan Schmidt */ 403e9b7fd4dSJan Schmidt static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, 404e9b7fd4dSJan Schmidt struct extent_buffer *eb) { 405e9b7fd4dSJan Schmidt smp_mb(); 406e9b7fd4dSJan Schmidt if (list_empty(&(fs_info)->tree_mod_seq_list)) 407e9b7fd4dSJan Schmidt return 1; 408097b8a7cSJan Schmidt if (eb && btrfs_header_level(eb) == 0) 409e9b7fd4dSJan Schmidt return 1; 4105de865eeSFilipe David Borba Manana 411b1a09f1eSDavid Sterba write_lock(&fs_info->tree_mod_log_lock); 4125de865eeSFilipe David Borba Manana if (list_empty(&(fs_info)->tree_mod_seq_list)) { 413b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 4145de865eeSFilipe David Borba Manana return 1; 4155de865eeSFilipe David Borba Manana } 4165de865eeSFilipe David Borba Manana 417e9b7fd4dSJan Schmidt return 0; 418e9b7fd4dSJan Schmidt } 419e9b7fd4dSJan Schmidt 4205de865eeSFilipe David Borba Manana /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ 4215de865eeSFilipe David Borba Manana static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, 4225de865eeSFilipe David Borba Manana struct extent_buffer *eb) 4235de865eeSFilipe David Borba Manana { 4245de865eeSFilipe David Borba Manana smp_mb(); 4255de865eeSFilipe David Borba Manana if (list_empty(&(fs_info)->tree_mod_seq_list)) 4265de865eeSFilipe David Borba Manana return 0; 4275de865eeSFilipe David Borba Manana if (eb && btrfs_header_level(eb) == 0) 4285de865eeSFilipe David Borba Manana return 0; 4295de865eeSFilipe David Borba Manana 4305de865eeSFilipe David Borba Manana return 1; 4315de865eeSFilipe David Borba Manana } 4325de865eeSFilipe David Borba Manana 4335de865eeSFilipe David Borba Manana static struct tree_mod_elem * 4345de865eeSFilipe David Borba Manana alloc_tree_mod_elem(struct extent_buffer *eb, int slot, 435bd989ba3SJan Schmidt enum mod_log_op op, gfp_t flags) 436bd989ba3SJan Schmidt { 437097b8a7cSJan Schmidt struct tree_mod_elem *tm; 438bd989ba3SJan Schmidt 439c8cc6341SJosef Bacik tm = kzalloc(sizeof(*tm), flags); 440c8cc6341SJosef Bacik if (!tm) 4415de865eeSFilipe David Borba Manana return NULL; 442bd989ba3SJan Schmidt 443298cfd36SChandan Rajendra tm->logical = eb->start; 444bd989ba3SJan Schmidt if (op != MOD_LOG_KEY_ADD) { 445bd989ba3SJan Schmidt btrfs_node_key(eb, &tm->key, slot); 446bd989ba3SJan Schmidt tm->blockptr = btrfs_node_blockptr(eb, slot); 447bd989ba3SJan Schmidt } 448bd989ba3SJan Schmidt tm->op = op; 449bd989ba3SJan Schmidt tm->slot = slot; 450bd989ba3SJan Schmidt tm->generation = btrfs_node_ptr_generation(eb, slot); 4515de865eeSFilipe David Borba Manana RB_CLEAR_NODE(&tm->node); 452bd989ba3SJan Schmidt 4535de865eeSFilipe David Borba Manana return tm; 454097b8a7cSJan Schmidt } 455097b8a7cSJan Schmidt 456e09c2efeSDavid Sterba static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, 457097b8a7cSJan Schmidt enum mod_log_op op, gfp_t flags) 458097b8a7cSJan Schmidt { 4595de865eeSFilipe David Borba Manana struct tree_mod_elem *tm; 4605de865eeSFilipe David Borba Manana int ret; 4615de865eeSFilipe David Borba Manana 462e09c2efeSDavid Sterba if (!tree_mod_need_log(eb->fs_info, eb)) 463097b8a7cSJan Schmidt return 0; 464097b8a7cSJan Schmidt 4655de865eeSFilipe David Borba Manana tm = alloc_tree_mod_elem(eb, slot, op, flags); 4665de865eeSFilipe David Borba Manana if (!tm) 4675de865eeSFilipe David Borba Manana return -ENOMEM; 4685de865eeSFilipe David Borba Manana 469e09c2efeSDavid Sterba if (tree_mod_dont_log(eb->fs_info, eb)) { 4705de865eeSFilipe David Borba Manana kfree(tm); 4715de865eeSFilipe David Borba Manana return 0; 4725de865eeSFilipe David Borba Manana } 4735de865eeSFilipe David Borba Manana 474e09c2efeSDavid Sterba ret = __tree_mod_log_insert(eb->fs_info, tm); 475b1a09f1eSDavid Sterba write_unlock(&eb->fs_info->tree_mod_log_lock); 4765de865eeSFilipe David Borba Manana if (ret) 4775de865eeSFilipe David Borba Manana kfree(tm); 4785de865eeSFilipe David Borba Manana 4795de865eeSFilipe David Borba Manana return ret; 480097b8a7cSJan Schmidt } 481097b8a7cSJan Schmidt 4826074d45fSDavid Sterba static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, 4836074d45fSDavid Sterba int dst_slot, int src_slot, int nr_items) 484bd989ba3SJan Schmidt { 4855de865eeSFilipe David Borba Manana struct tree_mod_elem *tm = NULL; 4865de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list = NULL; 4875de865eeSFilipe David Borba Manana int ret = 0; 488bd989ba3SJan Schmidt int i; 4895de865eeSFilipe David Borba Manana int locked = 0; 490bd989ba3SJan Schmidt 4916074d45fSDavid Sterba if (!tree_mod_need_log(eb->fs_info, eb)) 492f395694cSJan Schmidt return 0; 493bd989ba3SJan Schmidt 494176ef8f5SDavid Sterba tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); 4955de865eeSFilipe David Borba Manana if (!tm_list) 4965de865eeSFilipe David Borba Manana return -ENOMEM; 497bd989ba3SJan Schmidt 498176ef8f5SDavid Sterba tm = kzalloc(sizeof(*tm), GFP_NOFS); 4995de865eeSFilipe David Borba Manana if (!tm) { 5005de865eeSFilipe David Borba Manana ret = -ENOMEM; 5015de865eeSFilipe David Borba Manana goto free_tms; 5025de865eeSFilipe David Borba Manana } 503f395694cSJan Schmidt 504298cfd36SChandan Rajendra tm->logical = eb->start; 505bd989ba3SJan Schmidt tm->slot = src_slot; 506bd989ba3SJan Schmidt tm->move.dst_slot = dst_slot; 507bd989ba3SJan Schmidt tm->move.nr_items = nr_items; 508bd989ba3SJan Schmidt tm->op = MOD_LOG_MOVE_KEYS; 509bd989ba3SJan Schmidt 5105de865eeSFilipe David Borba Manana for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { 5115de865eeSFilipe David Borba Manana tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, 512176ef8f5SDavid Sterba MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); 5135de865eeSFilipe David Borba Manana if (!tm_list[i]) { 5145de865eeSFilipe David Borba Manana ret = -ENOMEM; 5155de865eeSFilipe David Borba Manana goto free_tms; 5165de865eeSFilipe David Borba Manana } 517bd989ba3SJan Schmidt } 518bd989ba3SJan Schmidt 5196074d45fSDavid Sterba if (tree_mod_dont_log(eb->fs_info, eb)) 5205de865eeSFilipe David Borba Manana goto free_tms; 5215de865eeSFilipe David Borba Manana locked = 1; 5225de865eeSFilipe David Borba Manana 5235de865eeSFilipe David Borba Manana /* 5245de865eeSFilipe David Borba Manana * When we override something during the move, we log these removals. 5255de865eeSFilipe David Borba Manana * This can only happen when we move towards the beginning of the 5265de865eeSFilipe David Borba Manana * buffer, i.e. dst_slot < src_slot. 5275de865eeSFilipe David Borba Manana */ 5285de865eeSFilipe David Borba Manana for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { 5296074d45fSDavid Sterba ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); 5305de865eeSFilipe David Borba Manana if (ret) 5315de865eeSFilipe David Borba Manana goto free_tms; 5325de865eeSFilipe David Borba Manana } 5335de865eeSFilipe David Borba Manana 5346074d45fSDavid Sterba ret = __tree_mod_log_insert(eb->fs_info, tm); 5355de865eeSFilipe David Borba Manana if (ret) 5365de865eeSFilipe David Borba Manana goto free_tms; 537b1a09f1eSDavid Sterba write_unlock(&eb->fs_info->tree_mod_log_lock); 5385de865eeSFilipe David Borba Manana kfree(tm_list); 5395de865eeSFilipe David Borba Manana 5405de865eeSFilipe David Borba Manana return 0; 5415de865eeSFilipe David Borba Manana free_tms: 5425de865eeSFilipe David Borba Manana for (i = 0; i < nr_items; i++) { 5435de865eeSFilipe David Borba Manana if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) 5446074d45fSDavid Sterba rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); 5455de865eeSFilipe David Borba Manana kfree(tm_list[i]); 5465de865eeSFilipe David Borba Manana } 5475de865eeSFilipe David Borba Manana if (locked) 548b1a09f1eSDavid Sterba write_unlock(&eb->fs_info->tree_mod_log_lock); 5495de865eeSFilipe David Borba Manana kfree(tm_list); 5505de865eeSFilipe David Borba Manana kfree(tm); 5515de865eeSFilipe David Borba Manana 5525de865eeSFilipe David Borba Manana return ret; 5535de865eeSFilipe David Borba Manana } 5545de865eeSFilipe David Borba Manana 5555de865eeSFilipe David Borba Manana static inline int 5565de865eeSFilipe David Borba Manana __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, 5575de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list, 5585de865eeSFilipe David Borba Manana int nritems) 559097b8a7cSJan Schmidt { 5605de865eeSFilipe David Borba Manana int i, j; 561097b8a7cSJan Schmidt int ret; 562097b8a7cSJan Schmidt 563097b8a7cSJan Schmidt for (i = nritems - 1; i >= 0; i--) { 5645de865eeSFilipe David Borba Manana ret = __tree_mod_log_insert(fs_info, tm_list[i]); 5655de865eeSFilipe David Borba Manana if (ret) { 5665de865eeSFilipe David Borba Manana for (j = nritems - 1; j > i; j--) 5675de865eeSFilipe David Borba Manana rb_erase(&tm_list[j]->node, 5685de865eeSFilipe David Borba Manana &fs_info->tree_mod_log); 5695de865eeSFilipe David Borba Manana return ret; 570097b8a7cSJan Schmidt } 571097b8a7cSJan Schmidt } 572097b8a7cSJan Schmidt 5735de865eeSFilipe David Borba Manana return 0; 5745de865eeSFilipe David Borba Manana } 5755de865eeSFilipe David Borba Manana 57695b757c1SDavid Sterba static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, 57795b757c1SDavid Sterba struct extent_buffer *new_root, int log_removal) 578bd989ba3SJan Schmidt { 57995b757c1SDavid Sterba struct btrfs_fs_info *fs_info = old_root->fs_info; 5805de865eeSFilipe David Borba Manana struct tree_mod_elem *tm = NULL; 5815de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list = NULL; 5825de865eeSFilipe David Borba Manana int nritems = 0; 5835de865eeSFilipe David Borba Manana int ret = 0; 5845de865eeSFilipe David Borba Manana int i; 585bd989ba3SJan Schmidt 5865de865eeSFilipe David Borba Manana if (!tree_mod_need_log(fs_info, NULL)) 587097b8a7cSJan Schmidt return 0; 588097b8a7cSJan Schmidt 5895de865eeSFilipe David Borba Manana if (log_removal && btrfs_header_level(old_root) > 0) { 5905de865eeSFilipe David Borba Manana nritems = btrfs_header_nritems(old_root); 59131e818feSDavid Sterba tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), 592bcc8e07fSDavid Sterba GFP_NOFS); 5935de865eeSFilipe David Borba Manana if (!tm_list) { 5945de865eeSFilipe David Borba Manana ret = -ENOMEM; 5955de865eeSFilipe David Borba Manana goto free_tms; 5965de865eeSFilipe David Borba Manana } 5975de865eeSFilipe David Borba Manana for (i = 0; i < nritems; i++) { 5985de865eeSFilipe David Borba Manana tm_list[i] = alloc_tree_mod_elem(old_root, i, 599bcc8e07fSDavid Sterba MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); 6005de865eeSFilipe David Borba Manana if (!tm_list[i]) { 6015de865eeSFilipe David Borba Manana ret = -ENOMEM; 6025de865eeSFilipe David Borba Manana goto free_tms; 6035de865eeSFilipe David Borba Manana } 6045de865eeSFilipe David Borba Manana } 6055de865eeSFilipe David Borba Manana } 606d9abbf1cSJan Schmidt 607bcc8e07fSDavid Sterba tm = kzalloc(sizeof(*tm), GFP_NOFS); 6085de865eeSFilipe David Borba Manana if (!tm) { 6095de865eeSFilipe David Borba Manana ret = -ENOMEM; 6105de865eeSFilipe David Borba Manana goto free_tms; 6115de865eeSFilipe David Borba Manana } 612bd989ba3SJan Schmidt 613298cfd36SChandan Rajendra tm->logical = new_root->start; 614bd989ba3SJan Schmidt tm->old_root.logical = old_root->start; 615bd989ba3SJan Schmidt tm->old_root.level = btrfs_header_level(old_root); 616bd989ba3SJan Schmidt tm->generation = btrfs_header_generation(old_root); 617bd989ba3SJan Schmidt tm->op = MOD_LOG_ROOT_REPLACE; 618bd989ba3SJan Schmidt 6195de865eeSFilipe David Borba Manana if (tree_mod_dont_log(fs_info, NULL)) 6205de865eeSFilipe David Borba Manana goto free_tms; 6215de865eeSFilipe David Borba Manana 6225de865eeSFilipe David Borba Manana if (tm_list) 6235de865eeSFilipe David Borba Manana ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); 6245de865eeSFilipe David Borba Manana if (!ret) 6255de865eeSFilipe David Borba Manana ret = __tree_mod_log_insert(fs_info, tm); 6265de865eeSFilipe David Borba Manana 627b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 6285de865eeSFilipe David Borba Manana if (ret) 6295de865eeSFilipe David Borba Manana goto free_tms; 6305de865eeSFilipe David Borba Manana kfree(tm_list); 6315de865eeSFilipe David Borba Manana 6325de865eeSFilipe David Borba Manana return ret; 6335de865eeSFilipe David Borba Manana 6345de865eeSFilipe David Borba Manana free_tms: 6355de865eeSFilipe David Borba Manana if (tm_list) { 6365de865eeSFilipe David Borba Manana for (i = 0; i < nritems; i++) 6375de865eeSFilipe David Borba Manana kfree(tm_list[i]); 6385de865eeSFilipe David Borba Manana kfree(tm_list); 6395de865eeSFilipe David Borba Manana } 6405de865eeSFilipe David Borba Manana kfree(tm); 6415de865eeSFilipe David Borba Manana 6425de865eeSFilipe David Borba Manana return ret; 643bd989ba3SJan Schmidt } 644bd989ba3SJan Schmidt 645bd989ba3SJan Schmidt static struct tree_mod_elem * 646bd989ba3SJan Schmidt __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, 647bd989ba3SJan Schmidt int smallest) 648bd989ba3SJan Schmidt { 649bd989ba3SJan Schmidt struct rb_root *tm_root; 650bd989ba3SJan Schmidt struct rb_node *node; 651bd989ba3SJan Schmidt struct tree_mod_elem *cur = NULL; 652bd989ba3SJan Schmidt struct tree_mod_elem *found = NULL; 653bd989ba3SJan Schmidt 654b1a09f1eSDavid Sterba read_lock(&fs_info->tree_mod_log_lock); 655bd989ba3SJan Schmidt tm_root = &fs_info->tree_mod_log; 656bd989ba3SJan Schmidt node = tm_root->rb_node; 657bd989ba3SJan Schmidt while (node) { 6586b4df8b6SGeliang Tang cur = rb_entry(node, struct tree_mod_elem, node); 659298cfd36SChandan Rajendra if (cur->logical < start) { 660bd989ba3SJan Schmidt node = node->rb_left; 661298cfd36SChandan Rajendra } else if (cur->logical > start) { 662bd989ba3SJan Schmidt node = node->rb_right; 663097b8a7cSJan Schmidt } else if (cur->seq < min_seq) { 664bd989ba3SJan Schmidt node = node->rb_left; 665bd989ba3SJan Schmidt } else if (!smallest) { 666bd989ba3SJan Schmidt /* we want the node with the highest seq */ 667bd989ba3SJan Schmidt if (found) 668097b8a7cSJan Schmidt BUG_ON(found->seq > cur->seq); 669bd989ba3SJan Schmidt found = cur; 670bd989ba3SJan Schmidt node = node->rb_left; 671097b8a7cSJan Schmidt } else if (cur->seq > min_seq) { 672bd989ba3SJan Schmidt /* we want the node with the smallest seq */ 673bd989ba3SJan Schmidt if (found) 674097b8a7cSJan Schmidt BUG_ON(found->seq < cur->seq); 675bd989ba3SJan Schmidt found = cur; 676bd989ba3SJan Schmidt node = node->rb_right; 677bd989ba3SJan Schmidt } else { 678bd989ba3SJan Schmidt found = cur; 679bd989ba3SJan Schmidt break; 680bd989ba3SJan Schmidt } 681bd989ba3SJan Schmidt } 682b1a09f1eSDavid Sterba read_unlock(&fs_info->tree_mod_log_lock); 683bd989ba3SJan Schmidt 684bd989ba3SJan Schmidt return found; 685bd989ba3SJan Schmidt } 686bd989ba3SJan Schmidt 687bd989ba3SJan Schmidt /* 688bd989ba3SJan Schmidt * this returns the element from the log with the smallest time sequence 689bd989ba3SJan Schmidt * value that's in the log (the oldest log item). any element with a time 690bd989ba3SJan Schmidt * sequence lower than min_seq will be ignored. 691bd989ba3SJan Schmidt */ 692bd989ba3SJan Schmidt static struct tree_mod_elem * 693bd989ba3SJan Schmidt tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, 694bd989ba3SJan Schmidt u64 min_seq) 695bd989ba3SJan Schmidt { 696bd989ba3SJan Schmidt return __tree_mod_log_search(fs_info, start, min_seq, 1); 697bd989ba3SJan Schmidt } 698bd989ba3SJan Schmidt 699bd989ba3SJan Schmidt /* 700bd989ba3SJan Schmidt * this returns the element from the log with the largest time sequence 701bd989ba3SJan Schmidt * value that's in the log (the most recent log item). any element with 702bd989ba3SJan Schmidt * a time sequence lower than min_seq will be ignored. 703bd989ba3SJan Schmidt */ 704bd989ba3SJan Schmidt static struct tree_mod_elem * 705bd989ba3SJan Schmidt tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) 706bd989ba3SJan Schmidt { 707bd989ba3SJan Schmidt return __tree_mod_log_search(fs_info, start, min_seq, 0); 708bd989ba3SJan Schmidt } 709bd989ba3SJan Schmidt 710ed874f0dSDavid Sterba static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst, 711bd989ba3SJan Schmidt struct extent_buffer *src, unsigned long dst_offset, 71290f8d62eSJan Schmidt unsigned long src_offset, int nr_items) 713bd989ba3SJan Schmidt { 714ed874f0dSDavid Sterba struct btrfs_fs_info *fs_info = dst->fs_info; 7155de865eeSFilipe David Borba Manana int ret = 0; 7165de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list = NULL; 7175de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list_add, **tm_list_rem; 718bd989ba3SJan Schmidt int i; 7195de865eeSFilipe David Borba Manana int locked = 0; 720bd989ba3SJan Schmidt 7215de865eeSFilipe David Borba Manana if (!tree_mod_need_log(fs_info, NULL)) 7225de865eeSFilipe David Borba Manana return 0; 723bd989ba3SJan Schmidt 724c8cc6341SJosef Bacik if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) 7255de865eeSFilipe David Borba Manana return 0; 7265de865eeSFilipe David Borba Manana 72731e818feSDavid Sterba tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), 7285de865eeSFilipe David Borba Manana GFP_NOFS); 7295de865eeSFilipe David Borba Manana if (!tm_list) 7305de865eeSFilipe David Borba Manana return -ENOMEM; 7315de865eeSFilipe David Borba Manana 7325de865eeSFilipe David Borba Manana tm_list_add = tm_list; 7335de865eeSFilipe David Borba Manana tm_list_rem = tm_list + nr_items; 7345de865eeSFilipe David Borba Manana for (i = 0; i < nr_items; i++) { 7355de865eeSFilipe David Borba Manana tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, 7365de865eeSFilipe David Borba Manana MOD_LOG_KEY_REMOVE, GFP_NOFS); 7375de865eeSFilipe David Borba Manana if (!tm_list_rem[i]) { 7385de865eeSFilipe David Borba Manana ret = -ENOMEM; 7395de865eeSFilipe David Borba Manana goto free_tms; 7405de865eeSFilipe David Borba Manana } 7415de865eeSFilipe David Borba Manana 7425de865eeSFilipe David Borba Manana tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, 7435de865eeSFilipe David Borba Manana MOD_LOG_KEY_ADD, GFP_NOFS); 7445de865eeSFilipe David Borba Manana if (!tm_list_add[i]) { 7455de865eeSFilipe David Borba Manana ret = -ENOMEM; 7465de865eeSFilipe David Borba Manana goto free_tms; 7475de865eeSFilipe David Borba Manana } 7485de865eeSFilipe David Borba Manana } 7495de865eeSFilipe David Borba Manana 7505de865eeSFilipe David Borba Manana if (tree_mod_dont_log(fs_info, NULL)) 7515de865eeSFilipe David Borba Manana goto free_tms; 7525de865eeSFilipe David Borba Manana locked = 1; 753bd989ba3SJan Schmidt 754bd989ba3SJan Schmidt for (i = 0; i < nr_items; i++) { 7555de865eeSFilipe David Borba Manana ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); 7565de865eeSFilipe David Borba Manana if (ret) 7575de865eeSFilipe David Borba Manana goto free_tms; 7585de865eeSFilipe David Borba Manana ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); 7595de865eeSFilipe David Borba Manana if (ret) 7605de865eeSFilipe David Borba Manana goto free_tms; 761bd989ba3SJan Schmidt } 7625de865eeSFilipe David Borba Manana 763b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 7645de865eeSFilipe David Borba Manana kfree(tm_list); 7655de865eeSFilipe David Borba Manana 7665de865eeSFilipe David Borba Manana return 0; 7675de865eeSFilipe David Borba Manana 7685de865eeSFilipe David Borba Manana free_tms: 7695de865eeSFilipe David Borba Manana for (i = 0; i < nr_items * 2; i++) { 7705de865eeSFilipe David Borba Manana if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) 7715de865eeSFilipe David Borba Manana rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); 7725de865eeSFilipe David Borba Manana kfree(tm_list[i]); 7735de865eeSFilipe David Borba Manana } 7745de865eeSFilipe David Borba Manana if (locked) 775b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 7765de865eeSFilipe David Borba Manana kfree(tm_list); 7775de865eeSFilipe David Borba Manana 7785de865eeSFilipe David Borba Manana return ret; 779bd989ba3SJan Schmidt } 780bd989ba3SJan Schmidt 781db7279a2SDavid Sterba static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) 782bd989ba3SJan Schmidt { 7835de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list = NULL; 7845de865eeSFilipe David Borba Manana int nritems = 0; 7855de865eeSFilipe David Borba Manana int i; 7865de865eeSFilipe David Borba Manana int ret = 0; 7875de865eeSFilipe David Borba Manana 7885de865eeSFilipe David Borba Manana if (btrfs_header_level(eb) == 0) 7895de865eeSFilipe David Borba Manana return 0; 7905de865eeSFilipe David Borba Manana 791db7279a2SDavid Sterba if (!tree_mod_need_log(eb->fs_info, NULL)) 7925de865eeSFilipe David Borba Manana return 0; 7935de865eeSFilipe David Borba Manana 7945de865eeSFilipe David Borba Manana nritems = btrfs_header_nritems(eb); 79531e818feSDavid Sterba tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); 7965de865eeSFilipe David Borba Manana if (!tm_list) 7975de865eeSFilipe David Borba Manana return -ENOMEM; 7985de865eeSFilipe David Borba Manana 7995de865eeSFilipe David Borba Manana for (i = 0; i < nritems; i++) { 8005de865eeSFilipe David Borba Manana tm_list[i] = alloc_tree_mod_elem(eb, i, 8015de865eeSFilipe David Borba Manana MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); 8025de865eeSFilipe David Borba Manana if (!tm_list[i]) { 8035de865eeSFilipe David Borba Manana ret = -ENOMEM; 8045de865eeSFilipe David Borba Manana goto free_tms; 8055de865eeSFilipe David Borba Manana } 8065de865eeSFilipe David Borba Manana } 8075de865eeSFilipe David Borba Manana 808db7279a2SDavid Sterba if (tree_mod_dont_log(eb->fs_info, eb)) 8095de865eeSFilipe David Borba Manana goto free_tms; 8105de865eeSFilipe David Borba Manana 811db7279a2SDavid Sterba ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); 812b1a09f1eSDavid Sterba write_unlock(&eb->fs_info->tree_mod_log_lock); 8135de865eeSFilipe David Borba Manana if (ret) 8145de865eeSFilipe David Borba Manana goto free_tms; 8155de865eeSFilipe David Borba Manana kfree(tm_list); 8165de865eeSFilipe David Borba Manana 8175de865eeSFilipe David Borba Manana return 0; 8185de865eeSFilipe David Borba Manana 8195de865eeSFilipe David Borba Manana free_tms: 8205de865eeSFilipe David Borba Manana for (i = 0; i < nritems; i++) 8215de865eeSFilipe David Borba Manana kfree(tm_list[i]); 8225de865eeSFilipe David Borba Manana kfree(tm_list); 8235de865eeSFilipe David Borba Manana 8245de865eeSFilipe David Borba Manana return ret; 825bd989ba3SJan Schmidt } 826bd989ba3SJan Schmidt 827d352ac68SChris Mason /* 8285d4f98a2SYan Zheng * check if the tree block can be shared by multiple trees 8295d4f98a2SYan Zheng */ 8305d4f98a2SYan Zheng int btrfs_block_can_be_shared(struct btrfs_root *root, 8315d4f98a2SYan Zheng struct extent_buffer *buf) 8325d4f98a2SYan Zheng { 8335d4f98a2SYan Zheng /* 83492a7cc42SQu Wenruo * Tree blocks not in shareable trees and tree roots are never shared. 83592a7cc42SQu Wenruo * If a block was allocated after the last snapshot and the block was 83692a7cc42SQu Wenruo * not allocated by tree relocation, we know the block is not shared. 8375d4f98a2SYan Zheng */ 83892a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 8395d4f98a2SYan Zheng buf != root->node && buf != root->commit_root && 8405d4f98a2SYan Zheng (btrfs_header_generation(buf) <= 8415d4f98a2SYan Zheng btrfs_root_last_snapshot(&root->root_item) || 8425d4f98a2SYan Zheng btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) 8435d4f98a2SYan Zheng return 1; 844a79865c6SNikolay Borisov 8455d4f98a2SYan Zheng return 0; 8465d4f98a2SYan Zheng } 8475d4f98a2SYan Zheng 8485d4f98a2SYan Zheng static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, 8495d4f98a2SYan Zheng struct btrfs_root *root, 8505d4f98a2SYan Zheng struct extent_buffer *buf, 851f0486c68SYan, Zheng struct extent_buffer *cow, 852f0486c68SYan, Zheng int *last_ref) 8535d4f98a2SYan Zheng { 8540b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8555d4f98a2SYan Zheng u64 refs; 8565d4f98a2SYan Zheng u64 owner; 8575d4f98a2SYan Zheng u64 flags; 8585d4f98a2SYan Zheng u64 new_flags = 0; 8595d4f98a2SYan Zheng int ret; 8605d4f98a2SYan Zheng 8615d4f98a2SYan Zheng /* 8625d4f98a2SYan Zheng * Backrefs update rules: 8635d4f98a2SYan Zheng * 8645d4f98a2SYan Zheng * Always use full backrefs for extent pointers in tree block 8655d4f98a2SYan Zheng * allocated by tree relocation. 8665d4f98a2SYan Zheng * 8675d4f98a2SYan Zheng * If a shared tree block is no longer referenced by its owner 8685d4f98a2SYan Zheng * tree (btrfs_header_owner(buf) == root->root_key.objectid), 8695d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 8705d4f98a2SYan Zheng * 8715d4f98a2SYan Zheng * If a tree block is been relocating 8725d4f98a2SYan Zheng * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), 8735d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 8745d4f98a2SYan Zheng * The reason for this is some operations (such as drop tree) 8755d4f98a2SYan Zheng * are only allowed for blocks use full backrefs. 8765d4f98a2SYan Zheng */ 8775d4f98a2SYan Zheng 8785d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, buf)) { 8792ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, 8803173a18fSJosef Bacik btrfs_header_level(buf), 1, 8813173a18fSJosef Bacik &refs, &flags); 882be1a5564SMark Fasheh if (ret) 883be1a5564SMark Fasheh return ret; 884e5df9573SMark Fasheh if (refs == 0) { 885e5df9573SMark Fasheh ret = -EROFS; 8860b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 887e5df9573SMark Fasheh return ret; 888e5df9573SMark Fasheh } 8895d4f98a2SYan Zheng } else { 8905d4f98a2SYan Zheng refs = 1; 8915d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 8925d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 8935d4f98a2SYan Zheng flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; 8945d4f98a2SYan Zheng else 8955d4f98a2SYan Zheng flags = 0; 8965d4f98a2SYan Zheng } 8975d4f98a2SYan Zheng 8985d4f98a2SYan Zheng owner = btrfs_header_owner(buf); 8995d4f98a2SYan Zheng BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && 9005d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); 9015d4f98a2SYan Zheng 9025d4f98a2SYan Zheng if (refs > 1) { 9035d4f98a2SYan Zheng if ((owner == root->root_key.objectid || 9045d4f98a2SYan Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && 9055d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { 906e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, buf, 1); 907692826b2SJeff Mahoney if (ret) 908692826b2SJeff Mahoney return ret; 9095d4f98a2SYan Zheng 9105d4f98a2SYan Zheng if (root->root_key.objectid == 9115d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) { 912e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 0); 913692826b2SJeff Mahoney if (ret) 914692826b2SJeff Mahoney return ret; 915e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 916692826b2SJeff Mahoney if (ret) 917692826b2SJeff Mahoney return ret; 9185d4f98a2SYan Zheng } 9195d4f98a2SYan Zheng new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; 9205d4f98a2SYan Zheng } else { 9215d4f98a2SYan Zheng 9225d4f98a2SYan Zheng if (root->root_key.objectid == 9235d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 924e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 9255d4f98a2SYan Zheng else 926e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 927692826b2SJeff Mahoney if (ret) 928692826b2SJeff Mahoney return ret; 9295d4f98a2SYan Zheng } 9305d4f98a2SYan Zheng if (new_flags != 0) { 931b1c79e09SJosef Bacik int level = btrfs_header_level(buf); 932b1c79e09SJosef Bacik 93342c9d0b5SDavid Sterba ret = btrfs_set_disk_extent_flags(trans, buf, 934b1c79e09SJosef Bacik new_flags, level, 0); 935be1a5564SMark Fasheh if (ret) 936be1a5564SMark Fasheh return ret; 9375d4f98a2SYan Zheng } 9385d4f98a2SYan Zheng } else { 9395d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { 9405d4f98a2SYan Zheng if (root->root_key.objectid == 9415d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 942e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 9435d4f98a2SYan Zheng else 944e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 945692826b2SJeff Mahoney if (ret) 946692826b2SJeff Mahoney return ret; 947e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 1); 948692826b2SJeff Mahoney if (ret) 949692826b2SJeff Mahoney return ret; 9505d4f98a2SYan Zheng } 9516a884d7dSDavid Sterba btrfs_clean_tree_block(buf); 952f0486c68SYan, Zheng *last_ref = 1; 9535d4f98a2SYan Zheng } 9545d4f98a2SYan Zheng return 0; 9555d4f98a2SYan Zheng } 9565d4f98a2SYan Zheng 957a6279470SFilipe Manana static struct extent_buffer *alloc_tree_block_no_bg_flush( 958a6279470SFilipe Manana struct btrfs_trans_handle *trans, 959a6279470SFilipe Manana struct btrfs_root *root, 960a6279470SFilipe Manana u64 parent_start, 961a6279470SFilipe Manana const struct btrfs_disk_key *disk_key, 962a6279470SFilipe Manana int level, 963a6279470SFilipe Manana u64 hint, 9649631e4ccSJosef Bacik u64 empty_size, 9659631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 966a6279470SFilipe Manana { 967a6279470SFilipe Manana struct btrfs_fs_info *fs_info = root->fs_info; 968a6279470SFilipe Manana struct extent_buffer *ret; 969a6279470SFilipe Manana 970a6279470SFilipe Manana /* 971a6279470SFilipe Manana * If we are COWing a node/leaf from the extent, chunk, device or free 972a6279470SFilipe Manana * space trees, make sure that we do not finish block group creation of 973a6279470SFilipe Manana * pending block groups. We do this to avoid a deadlock. 974a6279470SFilipe Manana * COWing can result in allocation of a new chunk, and flushing pending 975a6279470SFilipe Manana * block groups (btrfs_create_pending_block_groups()) can be triggered 976a6279470SFilipe Manana * when finishing allocation of a new chunk. Creation of a pending block 977a6279470SFilipe Manana * group modifies the extent, chunk, device and free space trees, 978a6279470SFilipe Manana * therefore we could deadlock with ourselves since we are holding a 979a6279470SFilipe Manana * lock on an extent buffer that btrfs_create_pending_block_groups() may 980a6279470SFilipe Manana * try to COW later. 981a6279470SFilipe Manana * For similar reasons, we also need to delay flushing pending block 982a6279470SFilipe Manana * groups when splitting a leaf or node, from one of those trees, since 983a6279470SFilipe Manana * we are holding a write lock on it and its parent or when inserting a 984a6279470SFilipe Manana * new root node for one of those trees. 985a6279470SFilipe Manana */ 986a6279470SFilipe Manana if (root == fs_info->extent_root || 987a6279470SFilipe Manana root == fs_info->chunk_root || 988a6279470SFilipe Manana root == fs_info->dev_root || 989a6279470SFilipe Manana root == fs_info->free_space_root) 990a6279470SFilipe Manana trans->can_flush_pending_bgs = false; 991a6279470SFilipe Manana 992a6279470SFilipe Manana ret = btrfs_alloc_tree_block(trans, root, parent_start, 993a6279470SFilipe Manana root->root_key.objectid, disk_key, level, 9949631e4ccSJosef Bacik hint, empty_size, nest); 995a6279470SFilipe Manana trans->can_flush_pending_bgs = true; 996a6279470SFilipe Manana 997a6279470SFilipe Manana return ret; 998a6279470SFilipe Manana } 999a6279470SFilipe Manana 10005d4f98a2SYan Zheng /* 1001d397712bSChris Mason * does the dirty work in cow of a single block. The parent block (if 1002d397712bSChris Mason * supplied) is updated to point to the new cow copy. The new buffer is marked 1003d397712bSChris Mason * dirty and returned locked. If you modify the block it needs to be marked 1004d397712bSChris Mason * dirty again. 1005d352ac68SChris Mason * 1006d352ac68SChris Mason * search_start -- an allocation hint for the new block 1007d352ac68SChris Mason * 1008d397712bSChris Mason * empty_size -- a hint that you plan on doing more cow. This is the size in 1009d397712bSChris Mason * bytes the allocator should try to find free next to the block it returns. 1010d397712bSChris Mason * This is just a hint and may be ignored by the allocator. 1011d352ac68SChris Mason */ 1012d397712bSChris Mason static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, 10135f39d397SChris Mason struct btrfs_root *root, 10145f39d397SChris Mason struct extent_buffer *buf, 10155f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 10165f39d397SChris Mason struct extent_buffer **cow_ret, 10179631e4ccSJosef Bacik u64 search_start, u64 empty_size, 10189631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 10196702ed49SChris Mason { 10200b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 10215d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 10225f39d397SChris Mason struct extent_buffer *cow; 1023be1a5564SMark Fasheh int level, ret; 1024f0486c68SYan, Zheng int last_ref = 0; 1025925baeddSChris Mason int unlock_orig = 0; 10260f5053ebSGoldwyn Rodrigues u64 parent_start = 0; 10276702ed49SChris Mason 1028925baeddSChris Mason if (*cow_ret == buf) 1029925baeddSChris Mason unlock_orig = 1; 1030925baeddSChris Mason 1031b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1032925baeddSChris Mason 103392a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 10340b246afaSJeff Mahoney trans->transid != fs_info->running_transaction->transid); 103592a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 103627cdeb70SMiao Xie trans->transid != root->last_trans); 10375f39d397SChris Mason 10387bb86316SChris Mason level = btrfs_header_level(buf); 103931840ae1SZheng Yan 10405d4f98a2SYan Zheng if (level == 0) 10415d4f98a2SYan Zheng btrfs_item_key(buf, &disk_key, 0); 10425d4f98a2SYan Zheng else 10435d4f98a2SYan Zheng btrfs_node_key(buf, &disk_key, 0); 10445d4f98a2SYan Zheng 10450f5053ebSGoldwyn Rodrigues if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) 10465d4f98a2SYan Zheng parent_start = parent->start; 10475d4f98a2SYan Zheng 1048a6279470SFilipe Manana cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key, 10499631e4ccSJosef Bacik level, search_start, empty_size, nest); 10506702ed49SChris Mason if (IS_ERR(cow)) 10516702ed49SChris Mason return PTR_ERR(cow); 10526702ed49SChris Mason 1053b4ce94deSChris Mason /* cow is set to blocking by btrfs_init_new_buffer */ 1054b4ce94deSChris Mason 105558e8012cSDavid Sterba copy_extent_buffer_full(cow, buf); 1056db94535dSChris Mason btrfs_set_header_bytenr(cow, cow->start); 10575f39d397SChris Mason btrfs_set_header_generation(cow, trans->transid); 10585d4f98a2SYan Zheng btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 10595d4f98a2SYan Zheng btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 10605d4f98a2SYan Zheng BTRFS_HEADER_FLAG_RELOC); 10615d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 10625d4f98a2SYan Zheng btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 10635d4f98a2SYan Zheng else 10645f39d397SChris Mason btrfs_set_header_owner(cow, root->root_key.objectid); 10656702ed49SChris Mason 1066de37aa51SNikolay Borisov write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 10672b82032cSYan Zheng 1068be1a5564SMark Fasheh ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); 1069b68dc2a9SMark Fasheh if (ret) { 1070572c83acSJosef Bacik btrfs_tree_unlock(cow); 1071572c83acSJosef Bacik free_extent_buffer(cow); 107266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1073b68dc2a9SMark Fasheh return ret; 1074b68dc2a9SMark Fasheh } 10751a40e23bSZheng Yan 107692a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 107783d4cfd4SJosef Bacik ret = btrfs_reloc_cow_block(trans, root, buf, cow); 107893314e3bSZhaolei if (ret) { 1079572c83acSJosef Bacik btrfs_tree_unlock(cow); 1080572c83acSJosef Bacik free_extent_buffer(cow); 108166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 108283d4cfd4SJosef Bacik return ret; 108383d4cfd4SJosef Bacik } 108493314e3bSZhaolei } 10853fd0a558SYan, Zheng 10866702ed49SChris Mason if (buf == root->node) { 1087925baeddSChris Mason WARN_ON(parent && parent != buf); 10885d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 10895d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 10905d4f98a2SYan Zheng parent_start = buf->start; 1091925baeddSChris Mason 109267439dadSDavid Sterba atomic_inc(&cow->refs); 1093d9d19a01SDavid Sterba ret = tree_mod_log_insert_root(root->node, cow, 1); 1094d9d19a01SDavid Sterba BUG_ON(ret < 0); 1095240f62c8SChris Mason rcu_assign_pointer(root->node, cow); 1096925baeddSChris Mason 1097f0486c68SYan, Zheng btrfs_free_tree_block(trans, root, buf, parent_start, 10985581a51aSJan Schmidt last_ref); 10995f39d397SChris Mason free_extent_buffer(buf); 11000b86a832SChris Mason add_root_to_dirty_list(root); 11016702ed49SChris Mason } else { 11025d4f98a2SYan Zheng WARN_ON(trans->transid != btrfs_header_generation(parent)); 1103e09c2efeSDavid Sterba tree_mod_log_insert_key(parent, parent_slot, 1104c8cc6341SJosef Bacik MOD_LOG_KEY_REPLACE, GFP_NOFS); 11055f39d397SChris Mason btrfs_set_node_blockptr(parent, parent_slot, 1106db94535dSChris Mason cow->start); 110774493f7aSChris Mason btrfs_set_node_ptr_generation(parent, parent_slot, 110874493f7aSChris Mason trans->transid); 11096702ed49SChris Mason btrfs_mark_buffer_dirty(parent); 11105de865eeSFilipe David Borba Manana if (last_ref) { 1111db7279a2SDavid Sterba ret = tree_mod_log_free_eb(buf); 11125de865eeSFilipe David Borba Manana if (ret) { 1113572c83acSJosef Bacik btrfs_tree_unlock(cow); 1114572c83acSJosef Bacik free_extent_buffer(cow); 111566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 11165de865eeSFilipe David Borba Manana return ret; 11175de865eeSFilipe David Borba Manana } 11185de865eeSFilipe David Borba Manana } 1119f0486c68SYan, Zheng btrfs_free_tree_block(trans, root, buf, parent_start, 11205581a51aSJan Schmidt last_ref); 11216702ed49SChris Mason } 1122925baeddSChris Mason if (unlock_orig) 1123925baeddSChris Mason btrfs_tree_unlock(buf); 11243083ee2eSJosef Bacik free_extent_buffer_stale(buf); 11256702ed49SChris Mason btrfs_mark_buffer_dirty(cow); 11266702ed49SChris Mason *cow_ret = cow; 11276702ed49SChris Mason return 0; 11286702ed49SChris Mason } 11296702ed49SChris Mason 11305d9e75c4SJan Schmidt /* 11315d9e75c4SJan Schmidt * returns the logical address of the oldest predecessor of the given root. 11325d9e75c4SJan Schmidt * entries older than time_seq are ignored. 11335d9e75c4SJan Schmidt */ 1134bcd24dabSDavid Sterba static struct tree_mod_elem *__tree_mod_log_oldest_root( 113530b0463aSJan Schmidt struct extent_buffer *eb_root, u64 time_seq) 11365d9e75c4SJan Schmidt { 11375d9e75c4SJan Schmidt struct tree_mod_elem *tm; 11385d9e75c4SJan Schmidt struct tree_mod_elem *found = NULL; 113930b0463aSJan Schmidt u64 root_logical = eb_root->start; 11405d9e75c4SJan Schmidt int looped = 0; 11415d9e75c4SJan Schmidt 11425d9e75c4SJan Schmidt if (!time_seq) 114335a3621bSStefan Behrens return NULL; 11445d9e75c4SJan Schmidt 11455d9e75c4SJan Schmidt /* 1146298cfd36SChandan Rajendra * the very last operation that's logged for a root is the 1147298cfd36SChandan Rajendra * replacement operation (if it is replaced at all). this has 1148298cfd36SChandan Rajendra * the logical address of the *new* root, making it the very 1149298cfd36SChandan Rajendra * first operation that's logged for this root. 11505d9e75c4SJan Schmidt */ 11515d9e75c4SJan Schmidt while (1) { 1152bcd24dabSDavid Sterba tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, 11535d9e75c4SJan Schmidt time_seq); 11545d9e75c4SJan Schmidt if (!looped && !tm) 115535a3621bSStefan Behrens return NULL; 11565d9e75c4SJan Schmidt /* 115728da9fb4SJan Schmidt * if there are no tree operation for the oldest root, we simply 115828da9fb4SJan Schmidt * return it. this should only happen if that (old) root is at 115928da9fb4SJan Schmidt * level 0. 11605d9e75c4SJan Schmidt */ 116128da9fb4SJan Schmidt if (!tm) 116228da9fb4SJan Schmidt break; 11635d9e75c4SJan Schmidt 116428da9fb4SJan Schmidt /* 116528da9fb4SJan Schmidt * if there's an operation that's not a root replacement, we 116628da9fb4SJan Schmidt * found the oldest version of our root. normally, we'll find a 116728da9fb4SJan Schmidt * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. 116828da9fb4SJan Schmidt */ 11695d9e75c4SJan Schmidt if (tm->op != MOD_LOG_ROOT_REPLACE) 11705d9e75c4SJan Schmidt break; 11715d9e75c4SJan Schmidt 11725d9e75c4SJan Schmidt found = tm; 11735d9e75c4SJan Schmidt root_logical = tm->old_root.logical; 11745d9e75c4SJan Schmidt looped = 1; 11755d9e75c4SJan Schmidt } 11765d9e75c4SJan Schmidt 1177a95236d9SJan Schmidt /* if there's no old root to return, return what we found instead */ 1178a95236d9SJan Schmidt if (!found) 1179a95236d9SJan Schmidt found = tm; 1180a95236d9SJan Schmidt 11815d9e75c4SJan Schmidt return found; 11825d9e75c4SJan Schmidt } 11835d9e75c4SJan Schmidt 11845d9e75c4SJan Schmidt /* 11855d9e75c4SJan Schmidt * tm is a pointer to the first operation to rewind within eb. then, all 118601327610SNicholas D Steeves * previous operations will be rewound (until we reach something older than 11875d9e75c4SJan Schmidt * time_seq). 11885d9e75c4SJan Schmidt */ 11895d9e75c4SJan Schmidt static void 1190f1ca7e98SJosef Bacik __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, 1191f1ca7e98SJosef Bacik u64 time_seq, struct tree_mod_elem *first_tm) 11925d9e75c4SJan Schmidt { 11935d9e75c4SJan Schmidt u32 n; 11945d9e75c4SJan Schmidt struct rb_node *next; 11955d9e75c4SJan Schmidt struct tree_mod_elem *tm = first_tm; 11965d9e75c4SJan Schmidt unsigned long o_dst; 11975d9e75c4SJan Schmidt unsigned long o_src; 11985d9e75c4SJan Schmidt unsigned long p_size = sizeof(struct btrfs_key_ptr); 11995d9e75c4SJan Schmidt 12005d9e75c4SJan Schmidt n = btrfs_header_nritems(eb); 1201b1a09f1eSDavid Sterba read_lock(&fs_info->tree_mod_log_lock); 1202097b8a7cSJan Schmidt while (tm && tm->seq >= time_seq) { 12035d9e75c4SJan Schmidt /* 12045d9e75c4SJan Schmidt * all the operations are recorded with the operator used for 12055d9e75c4SJan Schmidt * the modification. as we're going backwards, we do the 12065d9e75c4SJan Schmidt * opposite of each operation here. 12075d9e75c4SJan Schmidt */ 12085d9e75c4SJan Schmidt switch (tm->op) { 12095d9e75c4SJan Schmidt case MOD_LOG_KEY_REMOVE_WHILE_FREEING: 12105d9e75c4SJan Schmidt BUG_ON(tm->slot < n); 1211c730ae0cSMarcos Paulo de Souza fallthrough; 121295c80bb1SLiu Bo case MOD_LOG_KEY_REMOVE_WHILE_MOVING: 12134c3e6969SChris Mason case MOD_LOG_KEY_REMOVE: 12145d9e75c4SJan Schmidt btrfs_set_node_key(eb, &tm->key, tm->slot); 12155d9e75c4SJan Schmidt btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); 12165d9e75c4SJan Schmidt btrfs_set_node_ptr_generation(eb, tm->slot, 12175d9e75c4SJan Schmidt tm->generation); 12184c3e6969SChris Mason n++; 12195d9e75c4SJan Schmidt break; 12205d9e75c4SJan Schmidt case MOD_LOG_KEY_REPLACE: 12215d9e75c4SJan Schmidt BUG_ON(tm->slot >= n); 12225d9e75c4SJan Schmidt btrfs_set_node_key(eb, &tm->key, tm->slot); 12235d9e75c4SJan Schmidt btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); 12245d9e75c4SJan Schmidt btrfs_set_node_ptr_generation(eb, tm->slot, 12255d9e75c4SJan Schmidt tm->generation); 12265d9e75c4SJan Schmidt break; 12275d9e75c4SJan Schmidt case MOD_LOG_KEY_ADD: 122819956c7eSJan Schmidt /* if a move operation is needed it's in the log */ 12295d9e75c4SJan Schmidt n--; 12305d9e75c4SJan Schmidt break; 12315d9e75c4SJan Schmidt case MOD_LOG_MOVE_KEYS: 1232c3193108SJan Schmidt o_dst = btrfs_node_key_ptr_offset(tm->slot); 1233c3193108SJan Schmidt o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); 1234c3193108SJan Schmidt memmove_extent_buffer(eb, o_dst, o_src, 12355d9e75c4SJan Schmidt tm->move.nr_items * p_size); 12365d9e75c4SJan Schmidt break; 12375d9e75c4SJan Schmidt case MOD_LOG_ROOT_REPLACE: 12385d9e75c4SJan Schmidt /* 12395d9e75c4SJan Schmidt * this operation is special. for roots, this must be 12405d9e75c4SJan Schmidt * handled explicitly before rewinding. 12415d9e75c4SJan Schmidt * for non-roots, this operation may exist if the node 12425d9e75c4SJan Schmidt * was a root: root A -> child B; then A gets empty and 12435d9e75c4SJan Schmidt * B is promoted to the new root. in the mod log, we'll 12445d9e75c4SJan Schmidt * have a root-replace operation for B, a tree block 12455d9e75c4SJan Schmidt * that is no root. we simply ignore that operation. 12465d9e75c4SJan Schmidt */ 12475d9e75c4SJan Schmidt break; 12485d9e75c4SJan Schmidt } 12495d9e75c4SJan Schmidt next = rb_next(&tm->node); 12505d9e75c4SJan Schmidt if (!next) 12515d9e75c4SJan Schmidt break; 12526b4df8b6SGeliang Tang tm = rb_entry(next, struct tree_mod_elem, node); 1253298cfd36SChandan Rajendra if (tm->logical != first_tm->logical) 12545d9e75c4SJan Schmidt break; 12555d9e75c4SJan Schmidt } 1256b1a09f1eSDavid Sterba read_unlock(&fs_info->tree_mod_log_lock); 12575d9e75c4SJan Schmidt btrfs_set_header_nritems(eb, n); 12585d9e75c4SJan Schmidt } 12595d9e75c4SJan Schmidt 126047fb091fSJan Schmidt /* 126101327610SNicholas D Steeves * Called with eb read locked. If the buffer cannot be rewound, the same buffer 126247fb091fSJan Schmidt * is returned. If rewind operations happen, a fresh buffer is returned. The 126347fb091fSJan Schmidt * returned buffer is always read-locked. If the returned buffer is not the 126447fb091fSJan Schmidt * input buffer, the lock on the input buffer is released and the input buffer 126547fb091fSJan Schmidt * is freed (its refcount is decremented). 126647fb091fSJan Schmidt */ 12675d9e75c4SJan Schmidt static struct extent_buffer * 12689ec72677SJosef Bacik tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, 12699ec72677SJosef Bacik struct extent_buffer *eb, u64 time_seq) 12705d9e75c4SJan Schmidt { 12715d9e75c4SJan Schmidt struct extent_buffer *eb_rewin; 12725d9e75c4SJan Schmidt struct tree_mod_elem *tm; 12735d9e75c4SJan Schmidt 12745d9e75c4SJan Schmidt if (!time_seq) 12755d9e75c4SJan Schmidt return eb; 12765d9e75c4SJan Schmidt 12775d9e75c4SJan Schmidt if (btrfs_header_level(eb) == 0) 12785d9e75c4SJan Schmidt return eb; 12795d9e75c4SJan Schmidt 12805d9e75c4SJan Schmidt tm = tree_mod_log_search(fs_info, eb->start, time_seq); 12815d9e75c4SJan Schmidt if (!tm) 12825d9e75c4SJan Schmidt return eb; 12835d9e75c4SJan Schmidt 12845d9e75c4SJan Schmidt if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { 12855d9e75c4SJan Schmidt BUG_ON(tm->slot != 0); 1286da17066cSJeff Mahoney eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); 1287db7f3436SJosef Bacik if (!eb_rewin) { 1288ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb); 1289db7f3436SJosef Bacik free_extent_buffer(eb); 1290db7f3436SJosef Bacik return NULL; 1291db7f3436SJosef Bacik } 12925d9e75c4SJan Schmidt btrfs_set_header_bytenr(eb_rewin, eb->start); 12935d9e75c4SJan Schmidt btrfs_set_header_backref_rev(eb_rewin, 12945d9e75c4SJan Schmidt btrfs_header_backref_rev(eb)); 12955d9e75c4SJan Schmidt btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); 1296c3193108SJan Schmidt btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); 12975d9e75c4SJan Schmidt } else { 12985d9e75c4SJan Schmidt eb_rewin = btrfs_clone_extent_buffer(eb); 1299db7f3436SJosef Bacik if (!eb_rewin) { 1300ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb); 1301db7f3436SJosef Bacik free_extent_buffer(eb); 1302db7f3436SJosef Bacik return NULL; 1303db7f3436SJosef Bacik } 13045d9e75c4SJan Schmidt } 13055d9e75c4SJan Schmidt 1306ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb); 13075d9e75c4SJan Schmidt free_extent_buffer(eb); 13085d9e75c4SJan Schmidt 1309d3beaa25SJosef Bacik btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb_rewin), 1310d3beaa25SJosef Bacik eb_rewin, btrfs_header_level(eb_rewin)); 131147fb091fSJan Schmidt btrfs_tree_read_lock(eb_rewin); 1312f1ca7e98SJosef Bacik __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); 131357911b8bSJan Schmidt WARN_ON(btrfs_header_nritems(eb_rewin) > 1314da17066cSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info)); 13155d9e75c4SJan Schmidt 13165d9e75c4SJan Schmidt return eb_rewin; 13175d9e75c4SJan Schmidt } 13185d9e75c4SJan Schmidt 13198ba97a15SJan Schmidt /* 13208ba97a15SJan Schmidt * get_old_root() rewinds the state of @root's root node to the given @time_seq 13218ba97a15SJan Schmidt * value. If there are no changes, the current root->root_node is returned. If 13228ba97a15SJan Schmidt * anything changed in between, there's a fresh buffer allocated on which the 13238ba97a15SJan Schmidt * rewind operations are done. In any case, the returned buffer is read locked. 13248ba97a15SJan Schmidt * Returns NULL on error (with no locks held). 13258ba97a15SJan Schmidt */ 13265d9e75c4SJan Schmidt static inline struct extent_buffer * 13275d9e75c4SJan Schmidt get_old_root(struct btrfs_root *root, u64 time_seq) 13285d9e75c4SJan Schmidt { 13290b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13305d9e75c4SJan Schmidt struct tree_mod_elem *tm; 133130b0463aSJan Schmidt struct extent_buffer *eb = NULL; 133230b0463aSJan Schmidt struct extent_buffer *eb_root; 1333efad8a85SFilipe Manana u64 eb_root_owner = 0; 13347bfdcf7fSLiu Bo struct extent_buffer *old; 1335a95236d9SJan Schmidt struct tree_mod_root *old_root = NULL; 13364325edd0SChris Mason u64 old_generation = 0; 1337a95236d9SJan Schmidt u64 logical; 1338581c1760SQu Wenruo int level; 13395d9e75c4SJan Schmidt 134030b0463aSJan Schmidt eb_root = btrfs_read_lock_root_node(root); 1341bcd24dabSDavid Sterba tm = __tree_mod_log_oldest_root(eb_root, time_seq); 13425d9e75c4SJan Schmidt if (!tm) 134330b0463aSJan Schmidt return eb_root; 13445d9e75c4SJan Schmidt 1345a95236d9SJan Schmidt if (tm->op == MOD_LOG_ROOT_REPLACE) { 13465d9e75c4SJan Schmidt old_root = &tm->old_root; 13475d9e75c4SJan Schmidt old_generation = tm->generation; 1348a95236d9SJan Schmidt logical = old_root->logical; 1349581c1760SQu Wenruo level = old_root->level; 1350a95236d9SJan Schmidt } else { 135130b0463aSJan Schmidt logical = eb_root->start; 1352581c1760SQu Wenruo level = btrfs_header_level(eb_root); 1353a95236d9SJan Schmidt } 13545d9e75c4SJan Schmidt 13550b246afaSJeff Mahoney tm = tree_mod_log_search(fs_info, logical, time_seq); 1356834328a8SJan Schmidt if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { 135730b0463aSJan Schmidt btrfs_tree_read_unlock(eb_root); 135830b0463aSJan Schmidt free_extent_buffer(eb_root); 13591b7ec85eSJosef Bacik old = read_tree_block(fs_info, logical, root->root_key.objectid, 13601b7ec85eSJosef Bacik 0, level, NULL); 136164c043deSLiu Bo if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { 136264c043deSLiu Bo if (!IS_ERR(old)) 1363416bc658SJosef Bacik free_extent_buffer(old); 13640b246afaSJeff Mahoney btrfs_warn(fs_info, 13650b246afaSJeff Mahoney "failed to read tree block %llu from get_old_root", 13660b246afaSJeff Mahoney logical); 1367834328a8SJan Schmidt } else { 13687bfdcf7fSLiu Bo eb = btrfs_clone_extent_buffer(old); 13697bfdcf7fSLiu Bo free_extent_buffer(old); 1370834328a8SJan Schmidt } 1371834328a8SJan Schmidt } else if (old_root) { 1372efad8a85SFilipe Manana eb_root_owner = btrfs_header_owner(eb_root); 137330b0463aSJan Schmidt btrfs_tree_read_unlock(eb_root); 137430b0463aSJan Schmidt free_extent_buffer(eb_root); 13750b246afaSJeff Mahoney eb = alloc_dummy_extent_buffer(fs_info, logical); 1376834328a8SJan Schmidt } else { 137730b0463aSJan Schmidt eb = btrfs_clone_extent_buffer(eb_root); 1378ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb_root); 137930b0463aSJan Schmidt free_extent_buffer(eb_root); 1380834328a8SJan Schmidt } 1381834328a8SJan Schmidt 13828ba97a15SJan Schmidt if (!eb) 13838ba97a15SJan Schmidt return NULL; 1384a95236d9SJan Schmidt if (old_root) { 13855d9e75c4SJan Schmidt btrfs_set_header_bytenr(eb, eb->start); 13865d9e75c4SJan Schmidt btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); 1387efad8a85SFilipe Manana btrfs_set_header_owner(eb, eb_root_owner); 13885d9e75c4SJan Schmidt btrfs_set_header_level(eb, old_root->level); 13895d9e75c4SJan Schmidt btrfs_set_header_generation(eb, old_generation); 1390a95236d9SJan Schmidt } 1391d3beaa25SJosef Bacik btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), eb, 1392d3beaa25SJosef Bacik btrfs_header_level(eb)); 1393d3beaa25SJosef Bacik btrfs_tree_read_lock(eb); 139428da9fb4SJan Schmidt if (tm) 13950b246afaSJeff Mahoney __tree_mod_log_rewind(fs_info, eb, time_seq, tm); 139628da9fb4SJan Schmidt else 139728da9fb4SJan Schmidt WARN_ON(btrfs_header_level(eb) != 0); 13980b246afaSJeff Mahoney WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); 13995d9e75c4SJan Schmidt 14005d9e75c4SJan Schmidt return eb; 14015d9e75c4SJan Schmidt } 14025d9e75c4SJan Schmidt 14035b6602e7SJan Schmidt int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) 14045b6602e7SJan Schmidt { 14055b6602e7SJan Schmidt struct tree_mod_elem *tm; 14065b6602e7SJan Schmidt int level; 140730b0463aSJan Schmidt struct extent_buffer *eb_root = btrfs_root_node(root); 14085b6602e7SJan Schmidt 1409bcd24dabSDavid Sterba tm = __tree_mod_log_oldest_root(eb_root, time_seq); 14105b6602e7SJan Schmidt if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { 14115b6602e7SJan Schmidt level = tm->old_root.level; 14125b6602e7SJan Schmidt } else { 141330b0463aSJan Schmidt level = btrfs_header_level(eb_root); 14145b6602e7SJan Schmidt } 141530b0463aSJan Schmidt free_extent_buffer(eb_root); 14165b6602e7SJan Schmidt 14175b6602e7SJan Schmidt return level; 14185b6602e7SJan Schmidt } 14195b6602e7SJan Schmidt 14205d4f98a2SYan Zheng static inline int should_cow_block(struct btrfs_trans_handle *trans, 14215d4f98a2SYan Zheng struct btrfs_root *root, 14225d4f98a2SYan Zheng struct extent_buffer *buf) 14235d4f98a2SYan Zheng { 1424f5ee5c9aSJeff Mahoney if (btrfs_is_testing(root->fs_info)) 1425faa2dbf0SJosef Bacik return 0; 1426fccb84c9SDavid Sterba 1427d1980131SDavid Sterba /* Ensure we can see the FORCE_COW bit */ 1428d1980131SDavid Sterba smp_mb__before_atomic(); 1429f1ebcc74SLiu Bo 1430f1ebcc74SLiu Bo /* 1431f1ebcc74SLiu Bo * We do not need to cow a block if 1432f1ebcc74SLiu Bo * 1) this block is not created or changed in this transaction; 1433f1ebcc74SLiu Bo * 2) this block does not belong to TREE_RELOC tree; 1434f1ebcc74SLiu Bo * 3) the root is not forced COW. 1435f1ebcc74SLiu Bo * 1436f1ebcc74SLiu Bo * What is forced COW: 143701327610SNicholas D Steeves * when we create snapshot during committing the transaction, 143852042d8eSAndrea Gelmini * after we've finished copying src root, we must COW the shared 1439f1ebcc74SLiu Bo * block to ensure the metadata consistency. 1440f1ebcc74SLiu Bo */ 14415d4f98a2SYan Zheng if (btrfs_header_generation(buf) == trans->transid && 14425d4f98a2SYan Zheng !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && 14435d4f98a2SYan Zheng !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 1444f1ebcc74SLiu Bo btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && 144527cdeb70SMiao Xie !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) 14465d4f98a2SYan Zheng return 0; 14475d4f98a2SYan Zheng return 1; 14485d4f98a2SYan Zheng } 14495d4f98a2SYan Zheng 1450d352ac68SChris Mason /* 1451d352ac68SChris Mason * cows a single block, see __btrfs_cow_block for the real work. 145201327610SNicholas D Steeves * This version of it has extra checks so that a block isn't COWed more than 1453d352ac68SChris Mason * once per transaction, as long as it hasn't been written yet 1454d352ac68SChris Mason */ 1455d397712bSChris Mason noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, 14565f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *buf, 14575f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 14589631e4ccSJosef Bacik struct extent_buffer **cow_ret, 14599631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 146002217ed2SChris Mason { 14610b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14626702ed49SChris Mason u64 search_start; 1463f510cfecSChris Mason int ret; 1464dc17ff8fSChris Mason 146583354f07SJosef Bacik if (test_bit(BTRFS_ROOT_DELETING, &root->state)) 146683354f07SJosef Bacik btrfs_err(fs_info, 146783354f07SJosef Bacik "COW'ing blocks on a fs root that's being dropped"); 146883354f07SJosef Bacik 14690b246afaSJeff Mahoney if (trans->transaction != fs_info->running_transaction) 147031b1a2bdSJulia Lawall WARN(1, KERN_CRIT "trans %llu running %llu\n", 1471c1c9ff7cSGeert Uytterhoeven trans->transid, 14720b246afaSJeff Mahoney fs_info->running_transaction->transid); 147331b1a2bdSJulia Lawall 14740b246afaSJeff Mahoney if (trans->transid != fs_info->generation) 147531b1a2bdSJulia Lawall WARN(1, KERN_CRIT "trans %llu running %llu\n", 14760b246afaSJeff Mahoney trans->transid, fs_info->generation); 1477dc17ff8fSChris Mason 14785d4f98a2SYan Zheng if (!should_cow_block(trans, root, buf)) { 147964c12921SJeff Mahoney trans->dirty = true; 148002217ed2SChris Mason *cow_ret = buf; 148102217ed2SChris Mason return 0; 148202217ed2SChris Mason } 1483c487685dSChris Mason 1484ee22184bSByongho Lee search_start = buf->start & ~((u64)SZ_1G - 1); 1485b4ce94deSChris Mason 1486f616f5cdSQu Wenruo /* 1487f616f5cdSQu Wenruo * Before CoWing this block for later modification, check if it's 1488f616f5cdSQu Wenruo * the subtree root and do the delayed subtree trace if needed. 1489f616f5cdSQu Wenruo * 1490f616f5cdSQu Wenruo * Also We don't care about the error, as it's handled internally. 1491f616f5cdSQu Wenruo */ 1492f616f5cdSQu Wenruo btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); 1493f510cfecSChris Mason ret = __btrfs_cow_block(trans, root, buf, parent, 14949631e4ccSJosef Bacik parent_slot, cow_ret, search_start, 0, nest); 14951abe9b8aSliubo 14961abe9b8aSliubo trace_btrfs_cow_block(root, buf, *cow_ret); 14971abe9b8aSliubo 1498f510cfecSChris Mason return ret; 14992c90e5d6SChris Mason } 1500f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); 15016702ed49SChris Mason 1502d352ac68SChris Mason /* 1503d352ac68SChris Mason * helper function for defrag to decide if two blocks pointed to by a 1504d352ac68SChris Mason * node are actually close by 1505d352ac68SChris Mason */ 15066b80053dSChris Mason static int close_blocks(u64 blocknr, u64 other, u32 blocksize) 15076702ed49SChris Mason { 15086b80053dSChris Mason if (blocknr < other && other - (blocknr + blocksize) < 32768) 15096702ed49SChris Mason return 1; 15106b80053dSChris Mason if (blocknr > other && blocknr - (other + blocksize) < 32768) 15116702ed49SChris Mason return 1; 151202217ed2SChris Mason return 0; 151302217ed2SChris Mason } 151402217ed2SChris Mason 1515ce6ef5abSDavid Sterba #ifdef __LITTLE_ENDIAN 1516ce6ef5abSDavid Sterba 1517ce6ef5abSDavid Sterba /* 1518ce6ef5abSDavid Sterba * Compare two keys, on little-endian the disk order is same as CPU order and 1519ce6ef5abSDavid Sterba * we can avoid the conversion. 1520ce6ef5abSDavid Sterba */ 1521ce6ef5abSDavid Sterba static int comp_keys(const struct btrfs_disk_key *disk_key, 1522ce6ef5abSDavid Sterba const struct btrfs_key *k2) 1523ce6ef5abSDavid Sterba { 1524ce6ef5abSDavid Sterba const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; 1525ce6ef5abSDavid Sterba 1526ce6ef5abSDavid Sterba return btrfs_comp_cpu_keys(k1, k2); 1527ce6ef5abSDavid Sterba } 1528ce6ef5abSDavid Sterba 1529ce6ef5abSDavid Sterba #else 1530ce6ef5abSDavid Sterba 1531081e9573SChris Mason /* 1532081e9573SChris Mason * compare two keys in a memcmp fashion 1533081e9573SChris Mason */ 1534310712b2SOmar Sandoval static int comp_keys(const struct btrfs_disk_key *disk, 1535310712b2SOmar Sandoval const struct btrfs_key *k2) 1536081e9573SChris Mason { 1537081e9573SChris Mason struct btrfs_key k1; 1538081e9573SChris Mason 1539081e9573SChris Mason btrfs_disk_key_to_cpu(&k1, disk); 1540081e9573SChris Mason 154120736abaSDiego Calleja return btrfs_comp_cpu_keys(&k1, k2); 1542081e9573SChris Mason } 1543ce6ef5abSDavid Sterba #endif 1544081e9573SChris Mason 1545f3465ca4SJosef Bacik /* 1546f3465ca4SJosef Bacik * same as comp_keys only with two btrfs_key's 1547f3465ca4SJosef Bacik */ 1548e1f60a65SDavid Sterba int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) 1549f3465ca4SJosef Bacik { 1550f3465ca4SJosef Bacik if (k1->objectid > k2->objectid) 1551f3465ca4SJosef Bacik return 1; 1552f3465ca4SJosef Bacik if (k1->objectid < k2->objectid) 1553f3465ca4SJosef Bacik return -1; 1554f3465ca4SJosef Bacik if (k1->type > k2->type) 1555f3465ca4SJosef Bacik return 1; 1556f3465ca4SJosef Bacik if (k1->type < k2->type) 1557f3465ca4SJosef Bacik return -1; 1558f3465ca4SJosef Bacik if (k1->offset > k2->offset) 1559f3465ca4SJosef Bacik return 1; 1560f3465ca4SJosef Bacik if (k1->offset < k2->offset) 1561f3465ca4SJosef Bacik return -1; 1562f3465ca4SJosef Bacik return 0; 1563f3465ca4SJosef Bacik } 1564081e9573SChris Mason 1565d352ac68SChris Mason /* 1566d352ac68SChris Mason * this is used by the defrag code to go through all the 1567d352ac68SChris Mason * leaves pointed to by a node and reallocate them so that 1568d352ac68SChris Mason * disk order is close to key order 1569d352ac68SChris Mason */ 15706702ed49SChris Mason int btrfs_realloc_node(struct btrfs_trans_handle *trans, 15715f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *parent, 1572de78b51aSEric Sandeen int start_slot, u64 *last_ret, 1573a6b6e75eSChris Mason struct btrfs_key *progress) 15746702ed49SChris Mason { 15750b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15766b80053dSChris Mason struct extent_buffer *cur; 15776702ed49SChris Mason u64 blocknr; 1578e9d0b13bSChris Mason u64 search_start = *last_ret; 1579e9d0b13bSChris Mason u64 last_block = 0; 15806702ed49SChris Mason u64 other; 15816702ed49SChris Mason u32 parent_nritems; 15826702ed49SChris Mason int end_slot; 15836702ed49SChris Mason int i; 15846702ed49SChris Mason int err = 0; 15856b80053dSChris Mason u32 blocksize; 1586081e9573SChris Mason int progress_passed = 0; 1587081e9573SChris Mason struct btrfs_disk_key disk_key; 15886702ed49SChris Mason 15890b246afaSJeff Mahoney WARN_ON(trans->transaction != fs_info->running_transaction); 15900b246afaSJeff Mahoney WARN_ON(trans->transid != fs_info->generation); 159186479a04SChris Mason 15926b80053dSChris Mason parent_nritems = btrfs_header_nritems(parent); 15930b246afaSJeff Mahoney blocksize = fs_info->nodesize; 15945dfe2be7SFilipe Manana end_slot = parent_nritems - 1; 15956702ed49SChris Mason 15965dfe2be7SFilipe Manana if (parent_nritems <= 1) 15976702ed49SChris Mason return 0; 15986702ed49SChris Mason 15995dfe2be7SFilipe Manana for (i = start_slot; i <= end_slot; i++) { 16006702ed49SChris Mason int close = 1; 1601a6b6e75eSChris Mason 1602081e9573SChris Mason btrfs_node_key(parent, &disk_key, i); 1603081e9573SChris Mason if (!progress_passed && comp_keys(&disk_key, progress) < 0) 1604081e9573SChris Mason continue; 1605081e9573SChris Mason 1606081e9573SChris Mason progress_passed = 1; 16076b80053dSChris Mason blocknr = btrfs_node_blockptr(parent, i); 1608e9d0b13bSChris Mason if (last_block == 0) 1609e9d0b13bSChris Mason last_block = blocknr; 16105708b959SChris Mason 16116702ed49SChris Mason if (i > 0) { 16126b80053dSChris Mason other = btrfs_node_blockptr(parent, i - 1); 16136b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 16146702ed49SChris Mason } 16155dfe2be7SFilipe Manana if (!close && i < end_slot) { 16166b80053dSChris Mason other = btrfs_node_blockptr(parent, i + 1); 16176b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 16186702ed49SChris Mason } 1619e9d0b13bSChris Mason if (close) { 1620e9d0b13bSChris Mason last_block = blocknr; 16216702ed49SChris Mason continue; 1622e9d0b13bSChris Mason } 16236702ed49SChris Mason 1624206983b7SJosef Bacik cur = btrfs_read_node_slot(parent, i); 1625206983b7SJosef Bacik if (IS_ERR(cur)) 162664c043deSLiu Bo return PTR_ERR(cur); 1627e9d0b13bSChris Mason if (search_start == 0) 16286b80053dSChris Mason search_start = last_block; 1629e9d0b13bSChris Mason 1630e7a84565SChris Mason btrfs_tree_lock(cur); 16316b80053dSChris Mason err = __btrfs_cow_block(trans, root, cur, parent, i, 1632e7a84565SChris Mason &cur, search_start, 16336b80053dSChris Mason min(16 * blocksize, 16349631e4ccSJosef Bacik (end_slot - i) * blocksize), 16359631e4ccSJosef Bacik BTRFS_NESTING_COW); 1636252c38f0SYan if (err) { 1637e7a84565SChris Mason btrfs_tree_unlock(cur); 16386b80053dSChris Mason free_extent_buffer(cur); 16396702ed49SChris Mason break; 1640252c38f0SYan } 1641e7a84565SChris Mason search_start = cur->start; 1642e7a84565SChris Mason last_block = cur->start; 1643f2183bdeSChris Mason *last_ret = search_start; 1644e7a84565SChris Mason btrfs_tree_unlock(cur); 1645e7a84565SChris Mason free_extent_buffer(cur); 16466702ed49SChris Mason } 16476702ed49SChris Mason return err; 16486702ed49SChris Mason } 16496702ed49SChris Mason 165074123bd7SChris Mason /* 16515f39d397SChris Mason * search for key in the extent_buffer. The items start at offset p, 16525f39d397SChris Mason * and they are item_size apart. There are 'max' items in p. 16535f39d397SChris Mason * 165474123bd7SChris Mason * the slot in the array is returned via slot, and it points to 165574123bd7SChris Mason * the place where you would insert key if it is not found in 165674123bd7SChris Mason * the array. 165774123bd7SChris Mason * 165874123bd7SChris Mason * slot may point to max if the key is bigger than all of the keys 165974123bd7SChris Mason */ 1660e02119d5SChris Mason static noinline int generic_bin_search(struct extent_buffer *eb, 1661310712b2SOmar Sandoval unsigned long p, int item_size, 1662310712b2SOmar Sandoval const struct btrfs_key *key, 1663be0e5c09SChris Mason int max, int *slot) 1664be0e5c09SChris Mason { 1665be0e5c09SChris Mason int low = 0; 1666be0e5c09SChris Mason int high = max; 1667be0e5c09SChris Mason int ret; 16685cd17f34SDavid Sterba const int key_size = sizeof(struct btrfs_disk_key); 1669be0e5c09SChris Mason 16705e24e9afSLiu Bo if (low > high) { 16715e24e9afSLiu Bo btrfs_err(eb->fs_info, 16725e24e9afSLiu Bo "%s: low (%d) > high (%d) eb %llu owner %llu level %d", 16735e24e9afSLiu Bo __func__, low, high, eb->start, 16745e24e9afSLiu Bo btrfs_header_owner(eb), btrfs_header_level(eb)); 16755e24e9afSLiu Bo return -EINVAL; 16765e24e9afSLiu Bo } 16775e24e9afSLiu Bo 1678be0e5c09SChris Mason while (low < high) { 16795cd17f34SDavid Sterba unsigned long oip; 16805cd17f34SDavid Sterba unsigned long offset; 16815cd17f34SDavid Sterba struct btrfs_disk_key *tmp; 16825cd17f34SDavid Sterba struct btrfs_disk_key unaligned; 16835cd17f34SDavid Sterba int mid; 16845cd17f34SDavid Sterba 1685be0e5c09SChris Mason mid = (low + high) / 2; 16865f39d397SChris Mason offset = p + mid * item_size; 16875cd17f34SDavid Sterba oip = offset_in_page(offset); 16885f39d397SChris Mason 16895cd17f34SDavid Sterba if (oip + key_size <= PAGE_SIZE) { 1690884b07d0SQu Wenruo const unsigned long idx = get_eb_page_index(offset); 16915cd17f34SDavid Sterba char *kaddr = page_address(eb->pages[idx]); 1692934d375bSChris Mason 1693884b07d0SQu Wenruo oip = get_eb_offset_in_page(eb, offset); 16945cd17f34SDavid Sterba tmp = (struct btrfs_disk_key *)(kaddr + oip); 16955cd17f34SDavid Sterba } else { 16965cd17f34SDavid Sterba read_extent_buffer(eb, &unaligned, offset, key_size); 16975f39d397SChris Mason tmp = &unaligned; 1698479965d6SChris Mason } 1699479965d6SChris Mason 1700be0e5c09SChris Mason ret = comp_keys(tmp, key); 1701be0e5c09SChris Mason 1702be0e5c09SChris Mason if (ret < 0) 1703be0e5c09SChris Mason low = mid + 1; 1704be0e5c09SChris Mason else if (ret > 0) 1705be0e5c09SChris Mason high = mid; 1706be0e5c09SChris Mason else { 1707be0e5c09SChris Mason *slot = mid; 1708be0e5c09SChris Mason return 0; 1709be0e5c09SChris Mason } 1710be0e5c09SChris Mason } 1711be0e5c09SChris Mason *slot = low; 1712be0e5c09SChris Mason return 1; 1713be0e5c09SChris Mason } 1714be0e5c09SChris Mason 171597571fd0SChris Mason /* 171697571fd0SChris Mason * simple bin_search frontend that does the right thing for 171797571fd0SChris Mason * leaves vs nodes 171897571fd0SChris Mason */ 1719a74b35ecSNikolay Borisov int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, 1720e3b83361SQu Wenruo int *slot) 1721be0e5c09SChris Mason { 1722e3b83361SQu Wenruo if (btrfs_header_level(eb) == 0) 17235f39d397SChris Mason return generic_bin_search(eb, 17245f39d397SChris Mason offsetof(struct btrfs_leaf, items), 17250783fcfcSChris Mason sizeof(struct btrfs_item), 17265f39d397SChris Mason key, btrfs_header_nritems(eb), 17277518a238SChris Mason slot); 1728f775738fSWang Sheng-Hui else 17295f39d397SChris Mason return generic_bin_search(eb, 17305f39d397SChris Mason offsetof(struct btrfs_node, ptrs), 1731123abc88SChris Mason sizeof(struct btrfs_key_ptr), 17325f39d397SChris Mason key, btrfs_header_nritems(eb), 17337518a238SChris Mason slot); 1734be0e5c09SChris Mason } 1735be0e5c09SChris Mason 1736f0486c68SYan, Zheng static void root_add_used(struct btrfs_root *root, u32 size) 1737f0486c68SYan, Zheng { 1738f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 1739f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 1740f0486c68SYan, Zheng btrfs_root_used(&root->root_item) + size); 1741f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 1742f0486c68SYan, Zheng } 1743f0486c68SYan, Zheng 1744f0486c68SYan, Zheng static void root_sub_used(struct btrfs_root *root, u32 size) 1745f0486c68SYan, Zheng { 1746f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 1747f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 1748f0486c68SYan, Zheng btrfs_root_used(&root->root_item) - size); 1749f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 1750f0486c68SYan, Zheng } 1751f0486c68SYan, Zheng 1752d352ac68SChris Mason /* given a node and slot number, this reads the blocks it points to. The 1753d352ac68SChris Mason * extent buffer is returned with a reference taken (but unlocked). 1754d352ac68SChris Mason */ 17554b231ae4SDavid Sterba struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, 17564b231ae4SDavid Sterba int slot) 1757bb803951SChris Mason { 1758ca7a79adSChris Mason int level = btrfs_header_level(parent); 1759416bc658SJosef Bacik struct extent_buffer *eb; 1760581c1760SQu Wenruo struct btrfs_key first_key; 1761416bc658SJosef Bacik 1762fb770ae4SLiu Bo if (slot < 0 || slot >= btrfs_header_nritems(parent)) 1763fb770ae4SLiu Bo return ERR_PTR(-ENOENT); 1764ca7a79adSChris Mason 1765ca7a79adSChris Mason BUG_ON(level == 0); 1766ca7a79adSChris Mason 1767581c1760SQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 1768d0d20b0fSDavid Sterba eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), 17691b7ec85eSJosef Bacik btrfs_header_owner(parent), 1770581c1760SQu Wenruo btrfs_node_ptr_generation(parent, slot), 1771581c1760SQu Wenruo level - 1, &first_key); 1772fb770ae4SLiu Bo if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { 1773416bc658SJosef Bacik free_extent_buffer(eb); 1774fb770ae4SLiu Bo eb = ERR_PTR(-EIO); 1775416bc658SJosef Bacik } 1776416bc658SJosef Bacik 1777416bc658SJosef Bacik return eb; 1778bb803951SChris Mason } 1779bb803951SChris Mason 1780d352ac68SChris Mason /* 1781d352ac68SChris Mason * node level balancing, used to make sure nodes are in proper order for 1782d352ac68SChris Mason * item deletion. We balance from the top down, so we have to make sure 1783d352ac68SChris Mason * that a deletion won't leave an node completely empty later on. 1784d352ac68SChris Mason */ 1785e02119d5SChris Mason static noinline int balance_level(struct btrfs_trans_handle *trans, 178698ed5174SChris Mason struct btrfs_root *root, 178798ed5174SChris Mason struct btrfs_path *path, int level) 1788bb803951SChris Mason { 17890b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 17905f39d397SChris Mason struct extent_buffer *right = NULL; 17915f39d397SChris Mason struct extent_buffer *mid; 17925f39d397SChris Mason struct extent_buffer *left = NULL; 17935f39d397SChris Mason struct extent_buffer *parent = NULL; 1794bb803951SChris Mason int ret = 0; 1795bb803951SChris Mason int wret; 1796bb803951SChris Mason int pslot; 1797bb803951SChris Mason int orig_slot = path->slots[level]; 179879f95c82SChris Mason u64 orig_ptr; 1799bb803951SChris Mason 180098e6b1ebSLiu Bo ASSERT(level > 0); 1801bb803951SChris Mason 18025f39d397SChris Mason mid = path->nodes[level]; 1803b4ce94deSChris Mason 1804ac5887c8SJosef Bacik WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); 18057bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 18067bb86316SChris Mason 18071d4f8a0cSChris Mason orig_ptr = btrfs_node_blockptr(mid, orig_slot); 180879f95c82SChris Mason 1809a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 18105f39d397SChris Mason parent = path->nodes[level + 1]; 1811bb803951SChris Mason pslot = path->slots[level + 1]; 1812a05a9bb1SLi Zefan } 1813bb803951SChris Mason 181440689478SChris Mason /* 181540689478SChris Mason * deal with the case where there is only one pointer in the root 181640689478SChris Mason * by promoting the node below to a root 181740689478SChris Mason */ 18185f39d397SChris Mason if (!parent) { 18195f39d397SChris Mason struct extent_buffer *child; 1820bb803951SChris Mason 18215f39d397SChris Mason if (btrfs_header_nritems(mid) != 1) 1822bb803951SChris Mason return 0; 1823bb803951SChris Mason 1824bb803951SChris Mason /* promote the child to a root */ 18254b231ae4SDavid Sterba child = btrfs_read_node_slot(mid, 0); 1826fb770ae4SLiu Bo if (IS_ERR(child)) { 1827fb770ae4SLiu Bo ret = PTR_ERR(child); 18280b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 1829305a26afSMark Fasheh goto enospc; 1830305a26afSMark Fasheh } 1831305a26afSMark Fasheh 1832925baeddSChris Mason btrfs_tree_lock(child); 18339631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 18349631e4ccSJosef Bacik BTRFS_NESTING_COW); 1835f0486c68SYan, Zheng if (ret) { 1836f0486c68SYan, Zheng btrfs_tree_unlock(child); 1837f0486c68SYan, Zheng free_extent_buffer(child); 1838f0486c68SYan, Zheng goto enospc; 1839f0486c68SYan, Zheng } 18402f375ab9SYan 1841d9d19a01SDavid Sterba ret = tree_mod_log_insert_root(root->node, child, 1); 1842d9d19a01SDavid Sterba BUG_ON(ret < 0); 1843240f62c8SChris Mason rcu_assign_pointer(root->node, child); 1844925baeddSChris Mason 18450b86a832SChris Mason add_root_to_dirty_list(root); 1846925baeddSChris Mason btrfs_tree_unlock(child); 1847b4ce94deSChris Mason 1848925baeddSChris Mason path->locks[level] = 0; 1849bb803951SChris Mason path->nodes[level] = NULL; 18506a884d7dSDavid Sterba btrfs_clean_tree_block(mid); 1851925baeddSChris Mason btrfs_tree_unlock(mid); 1852bb803951SChris Mason /* once for the path */ 18535f39d397SChris Mason free_extent_buffer(mid); 1854f0486c68SYan, Zheng 1855f0486c68SYan, Zheng root_sub_used(root, mid->len); 18565581a51aSJan Schmidt btrfs_free_tree_block(trans, root, mid, 0, 1); 1857bb803951SChris Mason /* once for the root ptr */ 18583083ee2eSJosef Bacik free_extent_buffer_stale(mid); 1859f0486c68SYan, Zheng return 0; 1860bb803951SChris Mason } 18615f39d397SChris Mason if (btrfs_header_nritems(mid) > 18620b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) 1863bb803951SChris Mason return 0; 1864bb803951SChris Mason 18654b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 1866fb770ae4SLiu Bo if (IS_ERR(left)) 1867fb770ae4SLiu Bo left = NULL; 1868fb770ae4SLiu Bo 18695f39d397SChris Mason if (left) { 1870bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 18715f39d397SChris Mason wret = btrfs_cow_block(trans, root, left, 18729631e4ccSJosef Bacik parent, pslot - 1, &left, 1873bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 187454aa1f4dSChris Mason if (wret) { 187554aa1f4dSChris Mason ret = wret; 187654aa1f4dSChris Mason goto enospc; 187754aa1f4dSChris Mason } 18782cc58cf2SChris Mason } 1879fb770ae4SLiu Bo 18804b231ae4SDavid Sterba right = btrfs_read_node_slot(parent, pslot + 1); 1881fb770ae4SLiu Bo if (IS_ERR(right)) 1882fb770ae4SLiu Bo right = NULL; 1883fb770ae4SLiu Bo 18845f39d397SChris Mason if (right) { 1885bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 18865f39d397SChris Mason wret = btrfs_cow_block(trans, root, right, 18879631e4ccSJosef Bacik parent, pslot + 1, &right, 1888bf59a5a2SJosef Bacik BTRFS_NESTING_RIGHT_COW); 18892cc58cf2SChris Mason if (wret) { 18902cc58cf2SChris Mason ret = wret; 18912cc58cf2SChris Mason goto enospc; 18922cc58cf2SChris Mason } 18932cc58cf2SChris Mason } 18942cc58cf2SChris Mason 18952cc58cf2SChris Mason /* first, try to make some room in the middle buffer */ 18965f39d397SChris Mason if (left) { 18975f39d397SChris Mason orig_slot += btrfs_header_nritems(left); 1898d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 189979f95c82SChris Mason if (wret < 0) 190079f95c82SChris Mason ret = wret; 1901bb803951SChris Mason } 190279f95c82SChris Mason 190379f95c82SChris Mason /* 190479f95c82SChris Mason * then try to empty the right most buffer into the middle 190579f95c82SChris Mason */ 19065f39d397SChris Mason if (right) { 1907d30a668fSDavid Sterba wret = push_node_left(trans, mid, right, 1); 190854aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 190979f95c82SChris Mason ret = wret; 19105f39d397SChris Mason if (btrfs_header_nritems(right) == 0) { 19116a884d7dSDavid Sterba btrfs_clean_tree_block(right); 1912925baeddSChris Mason btrfs_tree_unlock(right); 1913afe5fea7STsutomu Itoh del_ptr(root, path, level + 1, pslot + 1); 1914f0486c68SYan, Zheng root_sub_used(root, right->len); 19155581a51aSJan Schmidt btrfs_free_tree_block(trans, root, right, 0, 1); 19163083ee2eSJosef Bacik free_extent_buffer_stale(right); 1917f0486c68SYan, Zheng right = NULL; 1918bb803951SChris Mason } else { 19195f39d397SChris Mason struct btrfs_disk_key right_key; 19205f39d397SChris Mason btrfs_node_key(right, &right_key, 0); 19210e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(parent, pslot + 1, 19220e82bcfeSDavid Sterba MOD_LOG_KEY_REPLACE, GFP_NOFS); 19230e82bcfeSDavid Sterba BUG_ON(ret < 0); 19245f39d397SChris Mason btrfs_set_node_key(parent, &right_key, pslot + 1); 19255f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 1926bb803951SChris Mason } 1927bb803951SChris Mason } 19285f39d397SChris Mason if (btrfs_header_nritems(mid) == 1) { 192979f95c82SChris Mason /* 193079f95c82SChris Mason * we're not allowed to leave a node with one item in the 193179f95c82SChris Mason * tree during a delete. A deletion from lower in the tree 193279f95c82SChris Mason * could try to delete the only pointer in this node. 193379f95c82SChris Mason * So, pull some keys from the left. 193479f95c82SChris Mason * There has to be a left pointer at this point because 193579f95c82SChris Mason * otherwise we would have pulled some pointers from the 193679f95c82SChris Mason * right 193779f95c82SChris Mason */ 1938305a26afSMark Fasheh if (!left) { 1939305a26afSMark Fasheh ret = -EROFS; 19400b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 1941305a26afSMark Fasheh goto enospc; 1942305a26afSMark Fasheh } 194355d32ed8SDavid Sterba wret = balance_node_right(trans, mid, left); 194454aa1f4dSChris Mason if (wret < 0) { 194579f95c82SChris Mason ret = wret; 194654aa1f4dSChris Mason goto enospc; 194754aa1f4dSChris Mason } 1948bce4eae9SChris Mason if (wret == 1) { 1949d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 1950bce4eae9SChris Mason if (wret < 0) 1951bce4eae9SChris Mason ret = wret; 1952bce4eae9SChris Mason } 195379f95c82SChris Mason BUG_ON(wret == 1); 195479f95c82SChris Mason } 19555f39d397SChris Mason if (btrfs_header_nritems(mid) == 0) { 19566a884d7dSDavid Sterba btrfs_clean_tree_block(mid); 1957925baeddSChris Mason btrfs_tree_unlock(mid); 1958afe5fea7STsutomu Itoh del_ptr(root, path, level + 1, pslot); 1959f0486c68SYan, Zheng root_sub_used(root, mid->len); 19605581a51aSJan Schmidt btrfs_free_tree_block(trans, root, mid, 0, 1); 19613083ee2eSJosef Bacik free_extent_buffer_stale(mid); 1962f0486c68SYan, Zheng mid = NULL; 196379f95c82SChris Mason } else { 196479f95c82SChris Mason /* update the parent key to reflect our changes */ 19655f39d397SChris Mason struct btrfs_disk_key mid_key; 19665f39d397SChris Mason btrfs_node_key(mid, &mid_key, 0); 19670e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(parent, pslot, 19680e82bcfeSDavid Sterba MOD_LOG_KEY_REPLACE, GFP_NOFS); 19690e82bcfeSDavid Sterba BUG_ON(ret < 0); 19705f39d397SChris Mason btrfs_set_node_key(parent, &mid_key, pslot); 19715f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 197279f95c82SChris Mason } 1973bb803951SChris Mason 197479f95c82SChris Mason /* update the path */ 19755f39d397SChris Mason if (left) { 19765f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 197767439dadSDavid Sterba atomic_inc(&left->refs); 1978925baeddSChris Mason /* left was locked after cow */ 19795f39d397SChris Mason path->nodes[level] = left; 1980bb803951SChris Mason path->slots[level + 1] -= 1; 1981bb803951SChris Mason path->slots[level] = orig_slot; 1982925baeddSChris Mason if (mid) { 1983925baeddSChris Mason btrfs_tree_unlock(mid); 19845f39d397SChris Mason free_extent_buffer(mid); 1985925baeddSChris Mason } 1986bb803951SChris Mason } else { 19875f39d397SChris Mason orig_slot -= btrfs_header_nritems(left); 1988bb803951SChris Mason path->slots[level] = orig_slot; 1989bb803951SChris Mason } 1990bb803951SChris Mason } 199179f95c82SChris Mason /* double check we haven't messed things up */ 1992e20d96d6SChris Mason if (orig_ptr != 19935f39d397SChris Mason btrfs_node_blockptr(path->nodes[level], path->slots[level])) 199479f95c82SChris Mason BUG(); 199554aa1f4dSChris Mason enospc: 1996925baeddSChris Mason if (right) { 1997925baeddSChris Mason btrfs_tree_unlock(right); 19985f39d397SChris Mason free_extent_buffer(right); 1999925baeddSChris Mason } 2000925baeddSChris Mason if (left) { 2001925baeddSChris Mason if (path->nodes[level] != left) 2002925baeddSChris Mason btrfs_tree_unlock(left); 20035f39d397SChris Mason free_extent_buffer(left); 2004925baeddSChris Mason } 2005bb803951SChris Mason return ret; 2006bb803951SChris Mason } 2007bb803951SChris Mason 2008d352ac68SChris Mason /* Node balancing for insertion. Here we only split or push nodes around 2009d352ac68SChris Mason * when they are completely full. This is also done top down, so we 2010d352ac68SChris Mason * have to be pessimistic. 2011d352ac68SChris Mason */ 2012d397712bSChris Mason static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, 2013e66f709bSChris Mason struct btrfs_root *root, 2014e66f709bSChris Mason struct btrfs_path *path, int level) 2015e66f709bSChris Mason { 20160b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20175f39d397SChris Mason struct extent_buffer *right = NULL; 20185f39d397SChris Mason struct extent_buffer *mid; 20195f39d397SChris Mason struct extent_buffer *left = NULL; 20205f39d397SChris Mason struct extent_buffer *parent = NULL; 2021e66f709bSChris Mason int ret = 0; 2022e66f709bSChris Mason int wret; 2023e66f709bSChris Mason int pslot; 2024e66f709bSChris Mason int orig_slot = path->slots[level]; 2025e66f709bSChris Mason 2026e66f709bSChris Mason if (level == 0) 2027e66f709bSChris Mason return 1; 2028e66f709bSChris Mason 20295f39d397SChris Mason mid = path->nodes[level]; 20307bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 2031e66f709bSChris Mason 2032a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 20335f39d397SChris Mason parent = path->nodes[level + 1]; 2034e66f709bSChris Mason pslot = path->slots[level + 1]; 2035a05a9bb1SLi Zefan } 2036e66f709bSChris Mason 20375f39d397SChris Mason if (!parent) 2038e66f709bSChris Mason return 1; 2039e66f709bSChris Mason 20404b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 2041fb770ae4SLiu Bo if (IS_ERR(left)) 2042fb770ae4SLiu Bo left = NULL; 2043e66f709bSChris Mason 2044e66f709bSChris Mason /* first, try to make some room in the middle buffer */ 20455f39d397SChris Mason if (left) { 2046e66f709bSChris Mason u32 left_nr; 2047925baeddSChris Mason 2048bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 2049b4ce94deSChris Mason 20505f39d397SChris Mason left_nr = btrfs_header_nritems(left); 20510b246afaSJeff Mahoney if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 205233ade1f8SChris Mason wret = 1; 205333ade1f8SChris Mason } else { 20545f39d397SChris Mason ret = btrfs_cow_block(trans, root, left, parent, 20559631e4ccSJosef Bacik pslot - 1, &left, 2056bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 205754aa1f4dSChris Mason if (ret) 205854aa1f4dSChris Mason wret = 1; 205954aa1f4dSChris Mason else { 2060d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 0); 206154aa1f4dSChris Mason } 206233ade1f8SChris Mason } 2063e66f709bSChris Mason if (wret < 0) 2064e66f709bSChris Mason ret = wret; 2065e66f709bSChris Mason if (wret == 0) { 20665f39d397SChris Mason struct btrfs_disk_key disk_key; 2067e66f709bSChris Mason orig_slot += left_nr; 20685f39d397SChris Mason btrfs_node_key(mid, &disk_key, 0); 20690e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(parent, pslot, 20700e82bcfeSDavid Sterba MOD_LOG_KEY_REPLACE, GFP_NOFS); 20710e82bcfeSDavid Sterba BUG_ON(ret < 0); 20725f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot); 20735f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 20745f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 20755f39d397SChris Mason path->nodes[level] = left; 2076e66f709bSChris Mason path->slots[level + 1] -= 1; 2077e66f709bSChris Mason path->slots[level] = orig_slot; 2078925baeddSChris Mason btrfs_tree_unlock(mid); 20795f39d397SChris Mason free_extent_buffer(mid); 2080e66f709bSChris Mason } else { 2081e66f709bSChris Mason orig_slot -= 20825f39d397SChris Mason btrfs_header_nritems(left); 2083e66f709bSChris Mason path->slots[level] = orig_slot; 2084925baeddSChris Mason btrfs_tree_unlock(left); 20855f39d397SChris Mason free_extent_buffer(left); 2086e66f709bSChris Mason } 2087e66f709bSChris Mason return 0; 2088e66f709bSChris Mason } 2089925baeddSChris Mason btrfs_tree_unlock(left); 20905f39d397SChris Mason free_extent_buffer(left); 2091e66f709bSChris Mason } 20924b231ae4SDavid Sterba right = btrfs_read_node_slot(parent, pslot + 1); 2093fb770ae4SLiu Bo if (IS_ERR(right)) 2094fb770ae4SLiu Bo right = NULL; 2095e66f709bSChris Mason 2096e66f709bSChris Mason /* 2097e66f709bSChris Mason * then try to empty the right most buffer into the middle 2098e66f709bSChris Mason */ 20995f39d397SChris Mason if (right) { 210033ade1f8SChris Mason u32 right_nr; 2101b4ce94deSChris Mason 2102bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 2103b4ce94deSChris Mason 21045f39d397SChris Mason right_nr = btrfs_header_nritems(right); 21050b246afaSJeff Mahoney if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 210633ade1f8SChris Mason wret = 1; 210733ade1f8SChris Mason } else { 21085f39d397SChris Mason ret = btrfs_cow_block(trans, root, right, 21095f39d397SChris Mason parent, pslot + 1, 2110bf59a5a2SJosef Bacik &right, BTRFS_NESTING_RIGHT_COW); 211154aa1f4dSChris Mason if (ret) 211254aa1f4dSChris Mason wret = 1; 211354aa1f4dSChris Mason else { 211455d32ed8SDavid Sterba wret = balance_node_right(trans, right, mid); 211533ade1f8SChris Mason } 211654aa1f4dSChris Mason } 2117e66f709bSChris Mason if (wret < 0) 2118e66f709bSChris Mason ret = wret; 2119e66f709bSChris Mason if (wret == 0) { 21205f39d397SChris Mason struct btrfs_disk_key disk_key; 21215f39d397SChris Mason 21225f39d397SChris Mason btrfs_node_key(right, &disk_key, 0); 21230e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(parent, pslot + 1, 21240e82bcfeSDavid Sterba MOD_LOG_KEY_REPLACE, GFP_NOFS); 21250e82bcfeSDavid Sterba BUG_ON(ret < 0); 21265f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot + 1); 21275f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 21285f39d397SChris Mason 21295f39d397SChris Mason if (btrfs_header_nritems(mid) <= orig_slot) { 21305f39d397SChris Mason path->nodes[level] = right; 2131e66f709bSChris Mason path->slots[level + 1] += 1; 2132e66f709bSChris Mason path->slots[level] = orig_slot - 21335f39d397SChris Mason btrfs_header_nritems(mid); 2134925baeddSChris Mason btrfs_tree_unlock(mid); 21355f39d397SChris Mason free_extent_buffer(mid); 2136e66f709bSChris Mason } else { 2137925baeddSChris Mason btrfs_tree_unlock(right); 21385f39d397SChris Mason free_extent_buffer(right); 2139e66f709bSChris Mason } 2140e66f709bSChris Mason return 0; 2141e66f709bSChris Mason } 2142925baeddSChris Mason btrfs_tree_unlock(right); 21435f39d397SChris Mason free_extent_buffer(right); 2144e66f709bSChris Mason } 2145e66f709bSChris Mason return 1; 2146e66f709bSChris Mason } 2147e66f709bSChris Mason 214874123bd7SChris Mason /* 2149d352ac68SChris Mason * readahead one full node of leaves, finding things that are close 2150d352ac68SChris Mason * to the block in 'slot', and triggering ra on them. 21513c69faecSChris Mason */ 21522ff7e61eSJeff Mahoney static void reada_for_search(struct btrfs_fs_info *fs_info, 2153e02119d5SChris Mason struct btrfs_path *path, 215401f46658SChris Mason int level, int slot, u64 objectid) 21553c69faecSChris Mason { 21565f39d397SChris Mason struct extent_buffer *node; 215701f46658SChris Mason struct btrfs_disk_key disk_key; 21583c69faecSChris Mason u32 nritems; 21593c69faecSChris Mason u64 search; 2160a7175319SChris Mason u64 target; 21616b80053dSChris Mason u64 nread = 0; 21625f39d397SChris Mason struct extent_buffer *eb; 21636b80053dSChris Mason u32 nr; 21646b80053dSChris Mason u32 blocksize; 21656b80053dSChris Mason u32 nscan = 0; 2166db94535dSChris Mason 2167a6b6e75eSChris Mason if (level != 1) 21683c69faecSChris Mason return; 21693c69faecSChris Mason 21706702ed49SChris Mason if (!path->nodes[level]) 21716702ed49SChris Mason return; 21726702ed49SChris Mason 21735f39d397SChris Mason node = path->nodes[level]; 2174925baeddSChris Mason 21753c69faecSChris Mason search = btrfs_node_blockptr(node, slot); 21760b246afaSJeff Mahoney blocksize = fs_info->nodesize; 21770b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, search); 21785f39d397SChris Mason if (eb) { 21795f39d397SChris Mason free_extent_buffer(eb); 21803c69faecSChris Mason return; 21813c69faecSChris Mason } 21823c69faecSChris Mason 2183a7175319SChris Mason target = search; 21846b80053dSChris Mason 21855f39d397SChris Mason nritems = btrfs_header_nritems(node); 21866b80053dSChris Mason nr = slot; 218725b8b936SJosef Bacik 21883c69faecSChris Mason while (1) { 2189e4058b54SDavid Sterba if (path->reada == READA_BACK) { 21906b80053dSChris Mason if (nr == 0) 21913c69faecSChris Mason break; 21926b80053dSChris Mason nr--; 2193e4058b54SDavid Sterba } else if (path->reada == READA_FORWARD) { 21946b80053dSChris Mason nr++; 21956b80053dSChris Mason if (nr >= nritems) 21966b80053dSChris Mason break; 21973c69faecSChris Mason } 2198e4058b54SDavid Sterba if (path->reada == READA_BACK && objectid) { 219901f46658SChris Mason btrfs_node_key(node, &disk_key, nr); 220001f46658SChris Mason if (btrfs_disk_key_objectid(&disk_key) != objectid) 220101f46658SChris Mason break; 220201f46658SChris Mason } 22036b80053dSChris Mason search = btrfs_node_blockptr(node, nr); 2204a7175319SChris Mason if ((search <= target && target - search <= 65536) || 2205a7175319SChris Mason (search > target && search - target <= 65536)) { 2206bfb484d9SJosef Bacik btrfs_readahead_node_child(node, nr); 22076b80053dSChris Mason nread += blocksize; 22083c69faecSChris Mason } 22096b80053dSChris Mason nscan++; 2210a7175319SChris Mason if ((nread > 65536 || nscan > 32)) 22116b80053dSChris Mason break; 22123c69faecSChris Mason } 22133c69faecSChris Mason } 2214925baeddSChris Mason 2215bfb484d9SJosef Bacik static noinline void reada_for_balance(struct btrfs_path *path, int level) 2216b4ce94deSChris Mason { 2217bfb484d9SJosef Bacik struct extent_buffer *parent; 2218b4ce94deSChris Mason int slot; 2219b4ce94deSChris Mason int nritems; 2220b4ce94deSChris Mason 22218c594ea8SChris Mason parent = path->nodes[level + 1]; 2222b4ce94deSChris Mason if (!parent) 22230b08851fSJosef Bacik return; 2224b4ce94deSChris Mason 2225b4ce94deSChris Mason nritems = btrfs_header_nritems(parent); 22268c594ea8SChris Mason slot = path->slots[level + 1]; 2227b4ce94deSChris Mason 2228bfb484d9SJosef Bacik if (slot > 0) 2229bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot - 1); 2230bfb484d9SJosef Bacik if (slot + 1 < nritems) 2231bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot + 1); 2232b4ce94deSChris Mason } 2233b4ce94deSChris Mason 2234b4ce94deSChris Mason 2235b4ce94deSChris Mason /* 2236d397712bSChris Mason * when we walk down the tree, it is usually safe to unlock the higher layers 2237d397712bSChris Mason * in the tree. The exceptions are when our path goes through slot 0, because 2238d397712bSChris Mason * operations on the tree might require changing key pointers higher up in the 2239d397712bSChris Mason * tree. 2240d352ac68SChris Mason * 2241d397712bSChris Mason * callers might also have set path->keep_locks, which tells this code to keep 2242d397712bSChris Mason * the lock if the path points to the last slot in the block. This is part of 2243d397712bSChris Mason * walking through the tree, and selecting the next slot in the higher block. 2244d352ac68SChris Mason * 2245d397712bSChris Mason * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so 2246d397712bSChris Mason * if lowest_unlock is 1, level 0 won't be unlocked 2247d352ac68SChris Mason */ 2248e02119d5SChris Mason static noinline void unlock_up(struct btrfs_path *path, int level, 2249f7c79f30SChris Mason int lowest_unlock, int min_write_lock_level, 2250f7c79f30SChris Mason int *write_lock_level) 2251925baeddSChris Mason { 2252925baeddSChris Mason int i; 2253925baeddSChris Mason int skip_level = level; 2254051e1b9fSChris Mason int no_skips = 0; 2255925baeddSChris Mason struct extent_buffer *t; 2256925baeddSChris Mason 2257925baeddSChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 2258925baeddSChris Mason if (!path->nodes[i]) 2259925baeddSChris Mason break; 2260925baeddSChris Mason if (!path->locks[i]) 2261925baeddSChris Mason break; 2262051e1b9fSChris Mason if (!no_skips && path->slots[i] == 0) { 2263925baeddSChris Mason skip_level = i + 1; 2264925baeddSChris Mason continue; 2265925baeddSChris Mason } 2266051e1b9fSChris Mason if (!no_skips && path->keep_locks) { 2267925baeddSChris Mason u32 nritems; 2268925baeddSChris Mason t = path->nodes[i]; 2269925baeddSChris Mason nritems = btrfs_header_nritems(t); 2270051e1b9fSChris Mason if (nritems < 1 || path->slots[i] >= nritems - 1) { 2271925baeddSChris Mason skip_level = i + 1; 2272925baeddSChris Mason continue; 2273925baeddSChris Mason } 2274925baeddSChris Mason } 2275051e1b9fSChris Mason if (skip_level < i && i >= lowest_unlock) 2276051e1b9fSChris Mason no_skips = 1; 2277051e1b9fSChris Mason 2278925baeddSChris Mason t = path->nodes[i]; 2279d80bb3f9SLiu Bo if (i >= lowest_unlock && i > skip_level) { 2280bd681513SChris Mason btrfs_tree_unlock_rw(t, path->locks[i]); 2281925baeddSChris Mason path->locks[i] = 0; 2282f7c79f30SChris Mason if (write_lock_level && 2283f7c79f30SChris Mason i > min_write_lock_level && 2284f7c79f30SChris Mason i <= *write_lock_level) { 2285f7c79f30SChris Mason *write_lock_level = i - 1; 2286f7c79f30SChris Mason } 2287925baeddSChris Mason } 2288925baeddSChris Mason } 2289925baeddSChris Mason } 2290925baeddSChris Mason 22913c69faecSChris Mason /* 2292c8c42864SChris Mason * helper function for btrfs_search_slot. The goal is to find a block 2293c8c42864SChris Mason * in cache without setting the path to blocking. If we find the block 2294c8c42864SChris Mason * we return zero and the path is unchanged. 2295c8c42864SChris Mason * 2296c8c42864SChris Mason * If we can't find the block, we set the path blocking and do some 2297c8c42864SChris Mason * reada. -EAGAIN is returned and the search must be repeated. 2298c8c42864SChris Mason */ 2299c8c42864SChris Mason static int 2300d07b8528SLiu Bo read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, 2301c8c42864SChris Mason struct extent_buffer **eb_ret, int level, int slot, 2302cda79c54SDavid Sterba const struct btrfs_key *key) 2303c8c42864SChris Mason { 23040b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 2305c8c42864SChris Mason u64 blocknr; 2306c8c42864SChris Mason u64 gen; 2307c8c42864SChris Mason struct extent_buffer *tmp; 2308581c1760SQu Wenruo struct btrfs_key first_key; 230976a05b35SChris Mason int ret; 2310581c1760SQu Wenruo int parent_level; 2311c8c42864SChris Mason 2312213ff4b7SNikolay Borisov blocknr = btrfs_node_blockptr(*eb_ret, slot); 2313213ff4b7SNikolay Borisov gen = btrfs_node_ptr_generation(*eb_ret, slot); 2314213ff4b7SNikolay Borisov parent_level = btrfs_header_level(*eb_ret); 2315213ff4b7SNikolay Borisov btrfs_node_key_to_cpu(*eb_ret, &first_key, slot); 2316c8c42864SChris Mason 23170b246afaSJeff Mahoney tmp = find_extent_buffer(fs_info, blocknr); 2318cb44921aSChris Mason if (tmp) { 2319b9fab919SChris Mason /* first we do an atomic uptodate check */ 2320b9fab919SChris Mason if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { 2321448de471SQu Wenruo /* 2322448de471SQu Wenruo * Do extra check for first_key, eb can be stale due to 2323448de471SQu Wenruo * being cached, read from scrub, or have multiple 2324448de471SQu Wenruo * parents (shared tree blocks). 2325448de471SQu Wenruo */ 2326e064d5e9SDavid Sterba if (btrfs_verify_level_key(tmp, 2327448de471SQu Wenruo parent_level - 1, &first_key, gen)) { 2328448de471SQu Wenruo free_extent_buffer(tmp); 2329448de471SQu Wenruo return -EUCLEAN; 2330448de471SQu Wenruo } 2331c8c42864SChris Mason *eb_ret = tmp; 2332c8c42864SChris Mason return 0; 2333c8c42864SChris Mason } 2334bdf7c00eSJosef Bacik 2335b9fab919SChris Mason /* now we're allowed to do a blocking uptodate check */ 2336581c1760SQu Wenruo ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); 2337bdf7c00eSJosef Bacik if (!ret) { 2338cb44921aSChris Mason *eb_ret = tmp; 2339cb44921aSChris Mason return 0; 2340cb44921aSChris Mason } 2341cb44921aSChris Mason free_extent_buffer(tmp); 2342b3b4aa74SDavid Sterba btrfs_release_path(p); 2343cb44921aSChris Mason return -EIO; 2344cb44921aSChris Mason } 2345c8c42864SChris Mason 2346c8c42864SChris Mason /* 2347c8c42864SChris Mason * reduce lock contention at high levels 2348c8c42864SChris Mason * of the btree by dropping locks before 234976a05b35SChris Mason * we read. Don't release the lock on the current 235076a05b35SChris Mason * level because we need to walk this node to figure 235176a05b35SChris Mason * out which blocks to read. 2352c8c42864SChris Mason */ 23538c594ea8SChris Mason btrfs_unlock_up_safe(p, level + 1); 23548c594ea8SChris Mason 2355e4058b54SDavid Sterba if (p->reada != READA_NONE) 23562ff7e61eSJeff Mahoney reada_for_search(fs_info, p, level, slot, key->objectid); 2357c8c42864SChris Mason 235876a05b35SChris Mason ret = -EAGAIN; 23591b7ec85eSJosef Bacik tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid, 23601b7ec85eSJosef Bacik gen, parent_level - 1, &first_key); 236164c043deSLiu Bo if (!IS_ERR(tmp)) { 236276a05b35SChris Mason /* 236376a05b35SChris Mason * If the read above didn't mark this buffer up to date, 236476a05b35SChris Mason * it will never end up being up to date. Set ret to EIO now 236576a05b35SChris Mason * and give up so that our caller doesn't loop forever 236676a05b35SChris Mason * on our EAGAINs. 236776a05b35SChris Mason */ 2368e6a1d6fdSLiu Bo if (!extent_buffer_uptodate(tmp)) 236976a05b35SChris Mason ret = -EIO; 2370c8c42864SChris Mason free_extent_buffer(tmp); 2371c871b0f2SLiu Bo } else { 2372c871b0f2SLiu Bo ret = PTR_ERR(tmp); 237376a05b35SChris Mason } 237402a3307aSLiu Bo 237502a3307aSLiu Bo btrfs_release_path(p); 237676a05b35SChris Mason return ret; 2377c8c42864SChris Mason } 2378c8c42864SChris Mason 2379c8c42864SChris Mason /* 2380c8c42864SChris Mason * helper function for btrfs_search_slot. This does all of the checks 2381c8c42864SChris Mason * for node-level blocks and does any balancing required based on 2382c8c42864SChris Mason * the ins_len. 2383c8c42864SChris Mason * 2384c8c42864SChris Mason * If no extra work was required, zero is returned. If we had to 2385c8c42864SChris Mason * drop the path, -EAGAIN is returned and btrfs_search_slot must 2386c8c42864SChris Mason * start over 2387c8c42864SChris Mason */ 2388c8c42864SChris Mason static int 2389c8c42864SChris Mason setup_nodes_for_search(struct btrfs_trans_handle *trans, 2390c8c42864SChris Mason struct btrfs_root *root, struct btrfs_path *p, 2391bd681513SChris Mason struct extent_buffer *b, int level, int ins_len, 2392bd681513SChris Mason int *write_lock_level) 2393c8c42864SChris Mason { 23940b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 239595b982deSNikolay Borisov int ret = 0; 23960b246afaSJeff Mahoney 2397c8c42864SChris Mason if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= 23980b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { 2399c8c42864SChris Mason 2400bd681513SChris Mason if (*write_lock_level < level + 1) { 2401bd681513SChris Mason *write_lock_level = level + 1; 2402bd681513SChris Mason btrfs_release_path(p); 240395b982deSNikolay Borisov return -EAGAIN; 2404bd681513SChris Mason } 2405bd681513SChris Mason 2406bfb484d9SJosef Bacik reada_for_balance(p, level); 240795b982deSNikolay Borisov ret = split_node(trans, root, p, level); 2408c8c42864SChris Mason 2409c8c42864SChris Mason b = p->nodes[level]; 2410c8c42864SChris Mason } else if (ins_len < 0 && btrfs_header_nritems(b) < 24110b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { 2412c8c42864SChris Mason 2413bd681513SChris Mason if (*write_lock_level < level + 1) { 2414bd681513SChris Mason *write_lock_level = level + 1; 2415bd681513SChris Mason btrfs_release_path(p); 241695b982deSNikolay Borisov return -EAGAIN; 2417bd681513SChris Mason } 2418bd681513SChris Mason 2419bfb484d9SJosef Bacik reada_for_balance(p, level); 242095b982deSNikolay Borisov ret = balance_level(trans, root, p, level); 242195b982deSNikolay Borisov if (ret) 242295b982deSNikolay Borisov return ret; 2423c8c42864SChris Mason 2424c8c42864SChris Mason b = p->nodes[level]; 2425c8c42864SChris Mason if (!b) { 2426b3b4aa74SDavid Sterba btrfs_release_path(p); 242795b982deSNikolay Borisov return -EAGAIN; 2428c8c42864SChris Mason } 2429c8c42864SChris Mason BUG_ON(btrfs_header_nritems(b) == 1); 2430c8c42864SChris Mason } 2431c8c42864SChris Mason return ret; 2432c8c42864SChris Mason } 2433c8c42864SChris Mason 2434381cf658SDavid Sterba int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, 2435e33d5c3dSKelley Nielsen u64 iobjectid, u64 ioff, u8 key_type, 2436e33d5c3dSKelley Nielsen struct btrfs_key *found_key) 2437e33d5c3dSKelley Nielsen { 2438e33d5c3dSKelley Nielsen int ret; 2439e33d5c3dSKelley Nielsen struct btrfs_key key; 2440e33d5c3dSKelley Nielsen struct extent_buffer *eb; 2441381cf658SDavid Sterba 2442381cf658SDavid Sterba ASSERT(path); 24431d4c08e0SDavid Sterba ASSERT(found_key); 2444e33d5c3dSKelley Nielsen 2445e33d5c3dSKelley Nielsen key.type = key_type; 2446e33d5c3dSKelley Nielsen key.objectid = iobjectid; 2447e33d5c3dSKelley Nielsen key.offset = ioff; 2448e33d5c3dSKelley Nielsen 2449e33d5c3dSKelley Nielsen ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); 24501d4c08e0SDavid Sterba if (ret < 0) 2451e33d5c3dSKelley Nielsen return ret; 2452e33d5c3dSKelley Nielsen 2453e33d5c3dSKelley Nielsen eb = path->nodes[0]; 2454e33d5c3dSKelley Nielsen if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { 2455e33d5c3dSKelley Nielsen ret = btrfs_next_leaf(fs_root, path); 2456e33d5c3dSKelley Nielsen if (ret) 2457e33d5c3dSKelley Nielsen return ret; 2458e33d5c3dSKelley Nielsen eb = path->nodes[0]; 2459e33d5c3dSKelley Nielsen } 2460e33d5c3dSKelley Nielsen 2461e33d5c3dSKelley Nielsen btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); 2462e33d5c3dSKelley Nielsen if (found_key->type != key.type || 2463e33d5c3dSKelley Nielsen found_key->objectid != key.objectid) 2464e33d5c3dSKelley Nielsen return 1; 2465e33d5c3dSKelley Nielsen 2466e33d5c3dSKelley Nielsen return 0; 2467e33d5c3dSKelley Nielsen } 2468e33d5c3dSKelley Nielsen 24691fc28d8eSLiu Bo static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, 24701fc28d8eSLiu Bo struct btrfs_path *p, 24711fc28d8eSLiu Bo int write_lock_level) 24721fc28d8eSLiu Bo { 24731fc28d8eSLiu Bo struct btrfs_fs_info *fs_info = root->fs_info; 24741fc28d8eSLiu Bo struct extent_buffer *b; 24751fc28d8eSLiu Bo int root_lock; 24761fc28d8eSLiu Bo int level = 0; 24771fc28d8eSLiu Bo 24781fc28d8eSLiu Bo /* We try very hard to do read locks on the root */ 24791fc28d8eSLiu Bo root_lock = BTRFS_READ_LOCK; 24801fc28d8eSLiu Bo 24811fc28d8eSLiu Bo if (p->search_commit_root) { 2482be6821f8SFilipe Manana /* 2483be6821f8SFilipe Manana * The commit roots are read only so we always do read locks, 2484be6821f8SFilipe Manana * and we always must hold the commit_root_sem when doing 2485be6821f8SFilipe Manana * searches on them, the only exception is send where we don't 2486be6821f8SFilipe Manana * want to block transaction commits for a long time, so 2487be6821f8SFilipe Manana * we need to clone the commit root in order to avoid races 2488be6821f8SFilipe Manana * with transaction commits that create a snapshot of one of 2489be6821f8SFilipe Manana * the roots used by a send operation. 2490be6821f8SFilipe Manana */ 2491be6821f8SFilipe Manana if (p->need_commit_sem) { 24921fc28d8eSLiu Bo down_read(&fs_info->commit_root_sem); 2493be6821f8SFilipe Manana b = btrfs_clone_extent_buffer(root->commit_root); 2494be6821f8SFilipe Manana up_read(&fs_info->commit_root_sem); 2495be6821f8SFilipe Manana if (!b) 2496be6821f8SFilipe Manana return ERR_PTR(-ENOMEM); 2497be6821f8SFilipe Manana 2498be6821f8SFilipe Manana } else { 24991fc28d8eSLiu Bo b = root->commit_root; 250067439dadSDavid Sterba atomic_inc(&b->refs); 2501be6821f8SFilipe Manana } 25021fc28d8eSLiu Bo level = btrfs_header_level(b); 2503f9ddfd05SLiu Bo /* 2504f9ddfd05SLiu Bo * Ensure that all callers have set skip_locking when 2505f9ddfd05SLiu Bo * p->search_commit_root = 1. 2506f9ddfd05SLiu Bo */ 2507f9ddfd05SLiu Bo ASSERT(p->skip_locking == 1); 25081fc28d8eSLiu Bo 25091fc28d8eSLiu Bo goto out; 25101fc28d8eSLiu Bo } 25111fc28d8eSLiu Bo 25121fc28d8eSLiu Bo if (p->skip_locking) { 25131fc28d8eSLiu Bo b = btrfs_root_node(root); 25141fc28d8eSLiu Bo level = btrfs_header_level(b); 25151fc28d8eSLiu Bo goto out; 25161fc28d8eSLiu Bo } 25171fc28d8eSLiu Bo 25181fc28d8eSLiu Bo /* 2519662c653bSLiu Bo * If the level is set to maximum, we can skip trying to get the read 2520662c653bSLiu Bo * lock. 2521662c653bSLiu Bo */ 2522662c653bSLiu Bo if (write_lock_level < BTRFS_MAX_LEVEL) { 2523662c653bSLiu Bo /* 2524662c653bSLiu Bo * We don't know the level of the root node until we actually 2525662c653bSLiu Bo * have it read locked 25261fc28d8eSLiu Bo */ 25271bb96598SJosef Bacik b = btrfs_read_lock_root_node(root); 25281fc28d8eSLiu Bo level = btrfs_header_level(b); 25291fc28d8eSLiu Bo if (level > write_lock_level) 25301fc28d8eSLiu Bo goto out; 25311fc28d8eSLiu Bo 2532662c653bSLiu Bo /* Whoops, must trade for write lock */ 25331fc28d8eSLiu Bo btrfs_tree_read_unlock(b); 25341fc28d8eSLiu Bo free_extent_buffer(b); 2535662c653bSLiu Bo } 2536662c653bSLiu Bo 25371fc28d8eSLiu Bo b = btrfs_lock_root_node(root); 25381fc28d8eSLiu Bo root_lock = BTRFS_WRITE_LOCK; 25391fc28d8eSLiu Bo 25401fc28d8eSLiu Bo /* The level might have changed, check again */ 25411fc28d8eSLiu Bo level = btrfs_header_level(b); 25421fc28d8eSLiu Bo 25431fc28d8eSLiu Bo out: 25441fc28d8eSLiu Bo p->nodes[level] = b; 25451fc28d8eSLiu Bo if (!p->skip_locking) 25461fc28d8eSLiu Bo p->locks[level] = root_lock; 25471fc28d8eSLiu Bo /* 25481fc28d8eSLiu Bo * Callers are responsible for dropping b's references. 25491fc28d8eSLiu Bo */ 25501fc28d8eSLiu Bo return b; 25511fc28d8eSLiu Bo } 25521fc28d8eSLiu Bo 25531fc28d8eSLiu Bo 2554c8c42864SChris Mason /* 25554271eceaSNikolay Borisov * btrfs_search_slot - look for a key in a tree and perform necessary 25564271eceaSNikolay Borisov * modifications to preserve tree invariants. 255774123bd7SChris Mason * 25584271eceaSNikolay Borisov * @trans: Handle of transaction, used when modifying the tree 25594271eceaSNikolay Borisov * @p: Holds all btree nodes along the search path 25604271eceaSNikolay Borisov * @root: The root node of the tree 25614271eceaSNikolay Borisov * @key: The key we are looking for 25629a664971Sethanwu * @ins_len: Indicates purpose of search: 25639a664971Sethanwu * >0 for inserts it's size of item inserted (*) 25649a664971Sethanwu * <0 for deletions 25659a664971Sethanwu * 0 for plain searches, not modifying the tree 25669a664971Sethanwu * 25679a664971Sethanwu * (*) If size of item inserted doesn't include 25689a664971Sethanwu * sizeof(struct btrfs_item), then p->search_for_extension must 25699a664971Sethanwu * be set. 25704271eceaSNikolay Borisov * @cow: boolean should CoW operations be performed. Must always be 1 25714271eceaSNikolay Borisov * when modifying the tree. 257297571fd0SChris Mason * 25734271eceaSNikolay Borisov * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. 25744271eceaSNikolay Borisov * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) 25754271eceaSNikolay Borisov * 25764271eceaSNikolay Borisov * If @key is found, 0 is returned and you can find the item in the leaf level 25774271eceaSNikolay Borisov * of the path (level 0) 25784271eceaSNikolay Borisov * 25794271eceaSNikolay Borisov * If @key isn't found, 1 is returned and the leaf level of the path (level 0) 25804271eceaSNikolay Borisov * points to the slot where it should be inserted 25814271eceaSNikolay Borisov * 25824271eceaSNikolay Borisov * If an error is encountered while searching the tree a negative error number 25834271eceaSNikolay Borisov * is returned 258474123bd7SChris Mason */ 2585310712b2SOmar Sandoval int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, 2586310712b2SOmar Sandoval const struct btrfs_key *key, struct btrfs_path *p, 2587310712b2SOmar Sandoval int ins_len, int cow) 2588be0e5c09SChris Mason { 25895f39d397SChris Mason struct extent_buffer *b; 2590be0e5c09SChris Mason int slot; 2591be0e5c09SChris Mason int ret; 259233c66f43SYan Zheng int err; 2593be0e5c09SChris Mason int level; 2594925baeddSChris Mason int lowest_unlock = 1; 2595bd681513SChris Mason /* everything at write_lock_level or lower must be write locked */ 2596bd681513SChris Mason int write_lock_level = 0; 25979f3a7427SChris Mason u8 lowest_level = 0; 2598f7c79f30SChris Mason int min_write_lock_level; 2599d7396f07SFilipe David Borba Manana int prev_cmp; 26009f3a7427SChris Mason 26016702ed49SChris Mason lowest_level = p->lowest_level; 2602323ac95bSChris Mason WARN_ON(lowest_level && ins_len > 0); 260322b0ebdaSChris Mason WARN_ON(p->nodes[0] != NULL); 2604eb653de1SFilipe David Borba Manana BUG_ON(!cow && ins_len); 260525179201SJosef Bacik 2606bd681513SChris Mason if (ins_len < 0) { 2607925baeddSChris Mason lowest_unlock = 2; 260865b51a00SChris Mason 2609bd681513SChris Mason /* when we are removing items, we might have to go up to level 2610bd681513SChris Mason * two as we update tree pointers Make sure we keep write 2611bd681513SChris Mason * for those levels as well 2612bd681513SChris Mason */ 2613bd681513SChris Mason write_lock_level = 2; 2614bd681513SChris Mason } else if (ins_len > 0) { 2615bd681513SChris Mason /* 2616bd681513SChris Mason * for inserting items, make sure we have a write lock on 2617bd681513SChris Mason * level 1 so we can update keys 2618bd681513SChris Mason */ 2619bd681513SChris Mason write_lock_level = 1; 2620bd681513SChris Mason } 2621bd681513SChris Mason 2622bd681513SChris Mason if (!cow) 2623bd681513SChris Mason write_lock_level = -1; 2624bd681513SChris Mason 262509a2a8f9SJosef Bacik if (cow && (p->keep_locks || p->lowest_level)) 2626bd681513SChris Mason write_lock_level = BTRFS_MAX_LEVEL; 2627bd681513SChris Mason 2628f7c79f30SChris Mason min_write_lock_level = write_lock_level; 2629f7c79f30SChris Mason 2630bb803951SChris Mason again: 2631d7396f07SFilipe David Borba Manana prev_cmp = -1; 26321fc28d8eSLiu Bo b = btrfs_search_slot_get_root(root, p, write_lock_level); 2633be6821f8SFilipe Manana if (IS_ERR(b)) { 2634be6821f8SFilipe Manana ret = PTR_ERR(b); 2635be6821f8SFilipe Manana goto done; 2636be6821f8SFilipe Manana } 2637925baeddSChris Mason 2638eb60ceacSChris Mason while (b) { 2639f624d976SQu Wenruo int dec = 0; 2640f624d976SQu Wenruo 26415f39d397SChris Mason level = btrfs_header_level(b); 264265b51a00SChris Mason 264302217ed2SChris Mason if (cow) { 26449ea2c7c9SNikolay Borisov bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); 26459ea2c7c9SNikolay Borisov 2646c8c42864SChris Mason /* 2647c8c42864SChris Mason * if we don't really need to cow this block 2648c8c42864SChris Mason * then we don't want to set the path blocking, 2649c8c42864SChris Mason * so we test it here 2650c8c42864SChris Mason */ 265164c12921SJeff Mahoney if (!should_cow_block(trans, root, b)) { 265264c12921SJeff Mahoney trans->dirty = true; 265365b51a00SChris Mason goto cow_done; 265464c12921SJeff Mahoney } 26555d4f98a2SYan Zheng 2656bd681513SChris Mason /* 2657bd681513SChris Mason * must have write locks on this node and the 2658bd681513SChris Mason * parent 2659bd681513SChris Mason */ 26605124e00eSJosef Bacik if (level > write_lock_level || 26615124e00eSJosef Bacik (level + 1 > write_lock_level && 26625124e00eSJosef Bacik level + 1 < BTRFS_MAX_LEVEL && 26635124e00eSJosef Bacik p->nodes[level + 1])) { 2664bd681513SChris Mason write_lock_level = level + 1; 2665bd681513SChris Mason btrfs_release_path(p); 2666bd681513SChris Mason goto again; 2667bd681513SChris Mason } 2668bd681513SChris Mason 26699ea2c7c9SNikolay Borisov if (last_level) 26709ea2c7c9SNikolay Borisov err = btrfs_cow_block(trans, root, b, NULL, 0, 26719631e4ccSJosef Bacik &b, 26729631e4ccSJosef Bacik BTRFS_NESTING_COW); 26739ea2c7c9SNikolay Borisov else 267433c66f43SYan Zheng err = btrfs_cow_block(trans, root, b, 2675e20d96d6SChris Mason p->nodes[level + 1], 26769631e4ccSJosef Bacik p->slots[level + 1], &b, 26779631e4ccSJosef Bacik BTRFS_NESTING_COW); 267833c66f43SYan Zheng if (err) { 267933c66f43SYan Zheng ret = err; 268065b51a00SChris Mason goto done; 268154aa1f4dSChris Mason } 268202217ed2SChris Mason } 268365b51a00SChris Mason cow_done: 2684eb60ceacSChris Mason p->nodes[level] = b; 268552398340SLiu Bo /* 268652398340SLiu Bo * Leave path with blocking locks to avoid massive 268752398340SLiu Bo * lock context switch, this is made on purpose. 268852398340SLiu Bo */ 2689b4ce94deSChris Mason 2690b4ce94deSChris Mason /* 2691b4ce94deSChris Mason * we have a lock on b and as long as we aren't changing 2692b4ce94deSChris Mason * the tree, there is no way to for the items in b to change. 2693b4ce94deSChris Mason * It is safe to drop the lock on our parent before we 2694b4ce94deSChris Mason * go through the expensive btree search on b. 2695b4ce94deSChris Mason * 2696eb653de1SFilipe David Borba Manana * If we're inserting or deleting (ins_len != 0), then we might 2697eb653de1SFilipe David Borba Manana * be changing slot zero, which may require changing the parent. 2698eb653de1SFilipe David Borba Manana * So, we can't drop the lock until after we know which slot 2699eb653de1SFilipe David Borba Manana * we're operating on. 2700b4ce94deSChris Mason */ 2701eb653de1SFilipe David Borba Manana if (!ins_len && !p->keep_locks) { 2702eb653de1SFilipe David Borba Manana int u = level + 1; 2703eb653de1SFilipe David Borba Manana 2704eb653de1SFilipe David Borba Manana if (u < BTRFS_MAX_LEVEL && p->locks[u]) { 2705eb653de1SFilipe David Borba Manana btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); 2706eb653de1SFilipe David Borba Manana p->locks[u] = 0; 2707eb653de1SFilipe David Borba Manana } 2708eb653de1SFilipe David Borba Manana } 2709b4ce94deSChris Mason 2710995e9a16SNikolay Borisov /* 2711995e9a16SNikolay Borisov * If btrfs_bin_search returns an exact match (prev_cmp == 0) 2712995e9a16SNikolay Borisov * we can safely assume the target key will always be in slot 0 2713995e9a16SNikolay Borisov * on lower levels due to the invariants BTRFS' btree provides, 2714995e9a16SNikolay Borisov * namely that a btrfs_key_ptr entry always points to the 2715995e9a16SNikolay Borisov * lowest key in the child node, thus we can skip searching 2716995e9a16SNikolay Borisov * lower levels 2717995e9a16SNikolay Borisov */ 2718995e9a16SNikolay Borisov if (prev_cmp == 0) { 2719995e9a16SNikolay Borisov slot = 0; 2720995e9a16SNikolay Borisov ret = 0; 2721995e9a16SNikolay Borisov } else { 2722995e9a16SNikolay Borisov ret = btrfs_bin_search(b, key, &slot); 2723995e9a16SNikolay Borisov prev_cmp = ret; 2724415b35a5SLiu Bo if (ret < 0) 2725415b35a5SLiu Bo goto done; 2726995e9a16SNikolay Borisov } 2727b4ce94deSChris Mason 2728f624d976SQu Wenruo if (level == 0) { 2729be0e5c09SChris Mason p->slots[level] = slot; 27309a664971Sethanwu /* 27319a664971Sethanwu * Item key already exists. In this case, if we are 27329a664971Sethanwu * allowed to insert the item (for example, in dir_item 27339a664971Sethanwu * case, item key collision is allowed), it will be 27349a664971Sethanwu * merged with the original item. Only the item size 27359a664971Sethanwu * grows, no new btrfs item will be added. If 27369a664971Sethanwu * search_for_extension is not set, ins_len already 27379a664971Sethanwu * accounts the size btrfs_item, deduct it here so leaf 27389a664971Sethanwu * space check will be correct. 27399a664971Sethanwu */ 27409a664971Sethanwu if (ret == 0 && ins_len > 0 && !p->search_for_extension) { 27419a664971Sethanwu ASSERT(ins_len >= sizeof(struct btrfs_item)); 27429a664971Sethanwu ins_len -= sizeof(struct btrfs_item); 27439a664971Sethanwu } 274487b29b20SYan Zheng if (ins_len > 0 && 2745e902baacSDavid Sterba btrfs_leaf_free_space(b) < ins_len) { 2746bd681513SChris Mason if (write_lock_level < 1) { 2747bd681513SChris Mason write_lock_level = 1; 2748bd681513SChris Mason btrfs_release_path(p); 2749bd681513SChris Mason goto again; 2750bd681513SChris Mason } 2751bd681513SChris Mason 275233c66f43SYan Zheng err = split_leaf(trans, root, key, 2753cc0c5538SChris Mason p, ins_len, ret == 0); 2754b4ce94deSChris Mason 275533c66f43SYan Zheng BUG_ON(err > 0); 275633c66f43SYan Zheng if (err) { 275733c66f43SYan Zheng ret = err; 275865b51a00SChris Mason goto done; 275965b51a00SChris Mason } 27605c680ed6SChris Mason } 2761459931ecSChris Mason if (!p->search_for_split) 2762f7c79f30SChris Mason unlock_up(p, level, lowest_unlock, 27634b6f8e96SLiu Bo min_write_lock_level, NULL); 276465b51a00SChris Mason goto done; 276565b51a00SChris Mason } 2766f624d976SQu Wenruo if (ret && slot > 0) { 2767f624d976SQu Wenruo dec = 1; 2768f624d976SQu Wenruo slot--; 2769f624d976SQu Wenruo } 2770f624d976SQu Wenruo p->slots[level] = slot; 2771f624d976SQu Wenruo err = setup_nodes_for_search(trans, root, p, b, level, ins_len, 2772f624d976SQu Wenruo &write_lock_level); 2773f624d976SQu Wenruo if (err == -EAGAIN) 2774f624d976SQu Wenruo goto again; 2775f624d976SQu Wenruo if (err) { 2776f624d976SQu Wenruo ret = err; 2777f624d976SQu Wenruo goto done; 2778f624d976SQu Wenruo } 2779f624d976SQu Wenruo b = p->nodes[level]; 2780f624d976SQu Wenruo slot = p->slots[level]; 2781f624d976SQu Wenruo 2782f624d976SQu Wenruo /* 2783f624d976SQu Wenruo * Slot 0 is special, if we change the key we have to update 2784f624d976SQu Wenruo * the parent pointer which means we must have a write lock on 2785f624d976SQu Wenruo * the parent 2786f624d976SQu Wenruo */ 2787f624d976SQu Wenruo if (slot == 0 && ins_len && write_lock_level < level + 1) { 2788f624d976SQu Wenruo write_lock_level = level + 1; 2789f624d976SQu Wenruo btrfs_release_path(p); 2790f624d976SQu Wenruo goto again; 2791f624d976SQu Wenruo } 2792f624d976SQu Wenruo 2793f624d976SQu Wenruo unlock_up(p, level, lowest_unlock, min_write_lock_level, 2794f624d976SQu Wenruo &write_lock_level); 2795f624d976SQu Wenruo 2796f624d976SQu Wenruo if (level == lowest_level) { 2797f624d976SQu Wenruo if (dec) 2798f624d976SQu Wenruo p->slots[level]++; 2799f624d976SQu Wenruo goto done; 2800f624d976SQu Wenruo } 2801f624d976SQu Wenruo 2802f624d976SQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 2803f624d976SQu Wenruo if (err == -EAGAIN) 2804f624d976SQu Wenruo goto again; 2805f624d976SQu Wenruo if (err) { 2806f624d976SQu Wenruo ret = err; 2807f624d976SQu Wenruo goto done; 2808f624d976SQu Wenruo } 2809f624d976SQu Wenruo 2810f624d976SQu Wenruo if (!p->skip_locking) { 2811f624d976SQu Wenruo level = btrfs_header_level(b); 2812f624d976SQu Wenruo if (level <= write_lock_level) { 2813f624d976SQu Wenruo btrfs_tree_lock(b); 2814f624d976SQu Wenruo p->locks[level] = BTRFS_WRITE_LOCK; 2815f624d976SQu Wenruo } else { 2816fe596ca3SJosef Bacik btrfs_tree_read_lock(b); 2817f624d976SQu Wenruo p->locks[level] = BTRFS_READ_LOCK; 2818f624d976SQu Wenruo } 2819f624d976SQu Wenruo p->nodes[level] = b; 2820f624d976SQu Wenruo } 282165b51a00SChris Mason } 282265b51a00SChris Mason ret = 1; 282365b51a00SChris Mason done: 28245f5bc6b1SFilipe Manana if (ret < 0 && !p->skip_release_on_error) 2825b3b4aa74SDavid Sterba btrfs_release_path(p); 2826be0e5c09SChris Mason return ret; 2827be0e5c09SChris Mason } 2828f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); 2829be0e5c09SChris Mason 283074123bd7SChris Mason /* 28315d9e75c4SJan Schmidt * Like btrfs_search_slot, this looks for a key in the given tree. It uses the 28325d9e75c4SJan Schmidt * current state of the tree together with the operations recorded in the tree 28335d9e75c4SJan Schmidt * modification log to search for the key in a previous version of this tree, as 28345d9e75c4SJan Schmidt * denoted by the time_seq parameter. 28355d9e75c4SJan Schmidt * 28365d9e75c4SJan Schmidt * Naturally, there is no support for insert, delete or cow operations. 28375d9e75c4SJan Schmidt * 28385d9e75c4SJan Schmidt * The resulting path and return value will be set up as if we called 28395d9e75c4SJan Schmidt * btrfs_search_slot at that point in time with ins_len and cow both set to 0. 28405d9e75c4SJan Schmidt */ 2841310712b2SOmar Sandoval int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, 28425d9e75c4SJan Schmidt struct btrfs_path *p, u64 time_seq) 28435d9e75c4SJan Schmidt { 28440b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 28455d9e75c4SJan Schmidt struct extent_buffer *b; 28465d9e75c4SJan Schmidt int slot; 28475d9e75c4SJan Schmidt int ret; 28485d9e75c4SJan Schmidt int err; 28495d9e75c4SJan Schmidt int level; 28505d9e75c4SJan Schmidt int lowest_unlock = 1; 28515d9e75c4SJan Schmidt u8 lowest_level = 0; 28525d9e75c4SJan Schmidt 28535d9e75c4SJan Schmidt lowest_level = p->lowest_level; 28545d9e75c4SJan Schmidt WARN_ON(p->nodes[0] != NULL); 28555d9e75c4SJan Schmidt 28565d9e75c4SJan Schmidt if (p->search_commit_root) { 28575d9e75c4SJan Schmidt BUG_ON(time_seq); 28585d9e75c4SJan Schmidt return btrfs_search_slot(NULL, root, key, p, 0, 0); 28595d9e75c4SJan Schmidt } 28605d9e75c4SJan Schmidt 28615d9e75c4SJan Schmidt again: 28625d9e75c4SJan Schmidt b = get_old_root(root, time_seq); 2863315bed43SNikolay Borisov if (!b) { 2864315bed43SNikolay Borisov ret = -EIO; 2865315bed43SNikolay Borisov goto done; 2866315bed43SNikolay Borisov } 28675d9e75c4SJan Schmidt level = btrfs_header_level(b); 28685d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 28695d9e75c4SJan Schmidt 28705d9e75c4SJan Schmidt while (b) { 2871abe9339dSQu Wenruo int dec = 0; 2872abe9339dSQu Wenruo 28735d9e75c4SJan Schmidt level = btrfs_header_level(b); 28745d9e75c4SJan Schmidt p->nodes[level] = b; 28755d9e75c4SJan Schmidt 28765d9e75c4SJan Schmidt /* 28775d9e75c4SJan Schmidt * we have a lock on b and as long as we aren't changing 28785d9e75c4SJan Schmidt * the tree, there is no way to for the items in b to change. 28795d9e75c4SJan Schmidt * It is safe to drop the lock on our parent before we 28805d9e75c4SJan Schmidt * go through the expensive btree search on b. 28815d9e75c4SJan Schmidt */ 28825d9e75c4SJan Schmidt btrfs_unlock_up_safe(p, level + 1); 28835d9e75c4SJan Schmidt 2884995e9a16SNikolay Borisov ret = btrfs_bin_search(b, key, &slot); 2885cbca7d59SFilipe Manana if (ret < 0) 2886cbca7d59SFilipe Manana goto done; 28875d9e75c4SJan Schmidt 2888abe9339dSQu Wenruo if (level == 0) { 2889abe9339dSQu Wenruo p->slots[level] = slot; 2890abe9339dSQu Wenruo unlock_up(p, level, lowest_unlock, 0, NULL); 2891abe9339dSQu Wenruo goto done; 2892abe9339dSQu Wenruo } 2893abe9339dSQu Wenruo 28945d9e75c4SJan Schmidt if (ret && slot > 0) { 28955d9e75c4SJan Schmidt dec = 1; 2896abe9339dSQu Wenruo slot--; 28975d9e75c4SJan Schmidt } 28985d9e75c4SJan Schmidt p->slots[level] = slot; 28995d9e75c4SJan Schmidt unlock_up(p, level, lowest_unlock, 0, NULL); 29005d9e75c4SJan Schmidt 29015d9e75c4SJan Schmidt if (level == lowest_level) { 29025d9e75c4SJan Schmidt if (dec) 29035d9e75c4SJan Schmidt p->slots[level]++; 29045d9e75c4SJan Schmidt goto done; 29055d9e75c4SJan Schmidt } 29065d9e75c4SJan Schmidt 2907abe9339dSQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 29085d9e75c4SJan Schmidt if (err == -EAGAIN) 29095d9e75c4SJan Schmidt goto again; 29105d9e75c4SJan Schmidt if (err) { 29115d9e75c4SJan Schmidt ret = err; 29125d9e75c4SJan Schmidt goto done; 29135d9e75c4SJan Schmidt } 29145d9e75c4SJan Schmidt 29155d9e75c4SJan Schmidt level = btrfs_header_level(b); 29165d9e75c4SJan Schmidt btrfs_tree_read_lock(b); 29170b246afaSJeff Mahoney b = tree_mod_log_rewind(fs_info, p, b, time_seq); 2918db7f3436SJosef Bacik if (!b) { 2919db7f3436SJosef Bacik ret = -ENOMEM; 2920db7f3436SJosef Bacik goto done; 2921db7f3436SJosef Bacik } 29225d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 29235d9e75c4SJan Schmidt p->nodes[level] = b; 29245d9e75c4SJan Schmidt } 29255d9e75c4SJan Schmidt ret = 1; 29265d9e75c4SJan Schmidt done: 29275d9e75c4SJan Schmidt if (ret < 0) 29285d9e75c4SJan Schmidt btrfs_release_path(p); 29295d9e75c4SJan Schmidt 29305d9e75c4SJan Schmidt return ret; 29315d9e75c4SJan Schmidt } 29325d9e75c4SJan Schmidt 29335d9e75c4SJan Schmidt /* 29342f38b3e1SArne Jansen * helper to use instead of search slot if no exact match is needed but 29352f38b3e1SArne Jansen * instead the next or previous item should be returned. 29362f38b3e1SArne Jansen * When find_higher is true, the next higher item is returned, the next lower 29372f38b3e1SArne Jansen * otherwise. 29382f38b3e1SArne Jansen * When return_any and find_higher are both true, and no higher item is found, 29392f38b3e1SArne Jansen * return the next lower instead. 29402f38b3e1SArne Jansen * When return_any is true and find_higher is false, and no lower item is found, 29412f38b3e1SArne Jansen * return the next higher instead. 29422f38b3e1SArne Jansen * It returns 0 if any item is found, 1 if none is found (tree empty), and 29432f38b3e1SArne Jansen * < 0 on error 29442f38b3e1SArne Jansen */ 29452f38b3e1SArne Jansen int btrfs_search_slot_for_read(struct btrfs_root *root, 2946310712b2SOmar Sandoval const struct btrfs_key *key, 2947310712b2SOmar Sandoval struct btrfs_path *p, int find_higher, 2948310712b2SOmar Sandoval int return_any) 29492f38b3e1SArne Jansen { 29502f38b3e1SArne Jansen int ret; 29512f38b3e1SArne Jansen struct extent_buffer *leaf; 29522f38b3e1SArne Jansen 29532f38b3e1SArne Jansen again: 29542f38b3e1SArne Jansen ret = btrfs_search_slot(NULL, root, key, p, 0, 0); 29552f38b3e1SArne Jansen if (ret <= 0) 29562f38b3e1SArne Jansen return ret; 29572f38b3e1SArne Jansen /* 29582f38b3e1SArne Jansen * a return value of 1 means the path is at the position where the 29592f38b3e1SArne Jansen * item should be inserted. Normally this is the next bigger item, 29602f38b3e1SArne Jansen * but in case the previous item is the last in a leaf, path points 29612f38b3e1SArne Jansen * to the first free slot in the previous leaf, i.e. at an invalid 29622f38b3e1SArne Jansen * item. 29632f38b3e1SArne Jansen */ 29642f38b3e1SArne Jansen leaf = p->nodes[0]; 29652f38b3e1SArne Jansen 29662f38b3e1SArne Jansen if (find_higher) { 29672f38b3e1SArne Jansen if (p->slots[0] >= btrfs_header_nritems(leaf)) { 29682f38b3e1SArne Jansen ret = btrfs_next_leaf(root, p); 29692f38b3e1SArne Jansen if (ret <= 0) 29702f38b3e1SArne Jansen return ret; 29712f38b3e1SArne Jansen if (!return_any) 29722f38b3e1SArne Jansen return 1; 29732f38b3e1SArne Jansen /* 29742f38b3e1SArne Jansen * no higher item found, return the next 29752f38b3e1SArne Jansen * lower instead 29762f38b3e1SArne Jansen */ 29772f38b3e1SArne Jansen return_any = 0; 29782f38b3e1SArne Jansen find_higher = 0; 29792f38b3e1SArne Jansen btrfs_release_path(p); 29802f38b3e1SArne Jansen goto again; 29812f38b3e1SArne Jansen } 29822f38b3e1SArne Jansen } else { 29832f38b3e1SArne Jansen if (p->slots[0] == 0) { 29842f38b3e1SArne Jansen ret = btrfs_prev_leaf(root, p); 2985e6793769SArne Jansen if (ret < 0) 29862f38b3e1SArne Jansen return ret; 2987e6793769SArne Jansen if (!ret) { 298823c6bf6aSFilipe David Borba Manana leaf = p->nodes[0]; 298923c6bf6aSFilipe David Borba Manana if (p->slots[0] == btrfs_header_nritems(leaf)) 299023c6bf6aSFilipe David Borba Manana p->slots[0]--; 2991e6793769SArne Jansen return 0; 2992e6793769SArne Jansen } 29932f38b3e1SArne Jansen if (!return_any) 29942f38b3e1SArne Jansen return 1; 29952f38b3e1SArne Jansen /* 29962f38b3e1SArne Jansen * no lower item found, return the next 29972f38b3e1SArne Jansen * higher instead 29982f38b3e1SArne Jansen */ 29992f38b3e1SArne Jansen return_any = 0; 30002f38b3e1SArne Jansen find_higher = 1; 30012f38b3e1SArne Jansen btrfs_release_path(p); 30022f38b3e1SArne Jansen goto again; 3003e6793769SArne Jansen } else { 30042f38b3e1SArne Jansen --p->slots[0]; 30052f38b3e1SArne Jansen } 30062f38b3e1SArne Jansen } 30072f38b3e1SArne Jansen return 0; 30082f38b3e1SArne Jansen } 30092f38b3e1SArne Jansen 30102f38b3e1SArne Jansen /* 301174123bd7SChris Mason * adjust the pointers going up the tree, starting at level 301274123bd7SChris Mason * making sure the right key of each node is points to 'key'. 301374123bd7SChris Mason * This is used after shifting pointers to the left, so it stops 301474123bd7SChris Mason * fixing up pointers when a given leaf/node is not in slot 0 of the 301574123bd7SChris Mason * higher levels 3016aa5d6bedSChris Mason * 301774123bd7SChris Mason */ 3018b167fa91SNikolay Borisov static void fixup_low_keys(struct btrfs_path *path, 30195f39d397SChris Mason struct btrfs_disk_key *key, int level) 3020be0e5c09SChris Mason { 3021be0e5c09SChris Mason int i; 30225f39d397SChris Mason struct extent_buffer *t; 30230e82bcfeSDavid Sterba int ret; 30245f39d397SChris Mason 3025234b63a0SChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 3026be0e5c09SChris Mason int tslot = path->slots[i]; 30270e82bcfeSDavid Sterba 3028eb60ceacSChris Mason if (!path->nodes[i]) 3029be0e5c09SChris Mason break; 30305f39d397SChris Mason t = path->nodes[i]; 30310e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, 30320e82bcfeSDavid Sterba GFP_ATOMIC); 30330e82bcfeSDavid Sterba BUG_ON(ret < 0); 30345f39d397SChris Mason btrfs_set_node_key(t, key, tslot); 3035d6025579SChris Mason btrfs_mark_buffer_dirty(path->nodes[i]); 3036be0e5c09SChris Mason if (tslot != 0) 3037be0e5c09SChris Mason break; 3038be0e5c09SChris Mason } 3039be0e5c09SChris Mason } 3040be0e5c09SChris Mason 304174123bd7SChris Mason /* 304231840ae1SZheng Yan * update item key. 304331840ae1SZheng Yan * 304431840ae1SZheng Yan * This function isn't completely safe. It's the caller's responsibility 304531840ae1SZheng Yan * that the new key won't break the order 304631840ae1SZheng Yan */ 3047b7a0365eSDaniel Dressler void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, 3048b7a0365eSDaniel Dressler struct btrfs_path *path, 3049310712b2SOmar Sandoval const struct btrfs_key *new_key) 305031840ae1SZheng Yan { 305131840ae1SZheng Yan struct btrfs_disk_key disk_key; 305231840ae1SZheng Yan struct extent_buffer *eb; 305331840ae1SZheng Yan int slot; 305431840ae1SZheng Yan 305531840ae1SZheng Yan eb = path->nodes[0]; 305631840ae1SZheng Yan slot = path->slots[0]; 305731840ae1SZheng Yan if (slot > 0) { 305831840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot - 1); 30597c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { 30607c15d410SQu Wenruo btrfs_crit(fs_info, 30617c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 30627c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 30637c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 30647c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 30657c15d410SQu Wenruo new_key->objectid, new_key->type, 30667c15d410SQu Wenruo new_key->offset); 30677c15d410SQu Wenruo btrfs_print_leaf(eb); 30687c15d410SQu Wenruo BUG(); 30697c15d410SQu Wenruo } 307031840ae1SZheng Yan } 307131840ae1SZheng Yan if (slot < btrfs_header_nritems(eb) - 1) { 307231840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot + 1); 30737c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { 30747c15d410SQu Wenruo btrfs_crit(fs_info, 30757c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 30767c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 30777c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 30787c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 30797c15d410SQu Wenruo new_key->objectid, new_key->type, 30807c15d410SQu Wenruo new_key->offset); 30817c15d410SQu Wenruo btrfs_print_leaf(eb); 30827c15d410SQu Wenruo BUG(); 30837c15d410SQu Wenruo } 308431840ae1SZheng Yan } 308531840ae1SZheng Yan 308631840ae1SZheng Yan btrfs_cpu_key_to_disk(&disk_key, new_key); 308731840ae1SZheng Yan btrfs_set_item_key(eb, &disk_key, slot); 308831840ae1SZheng Yan btrfs_mark_buffer_dirty(eb); 308931840ae1SZheng Yan if (slot == 0) 3090b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 309131840ae1SZheng Yan } 309231840ae1SZheng Yan 309331840ae1SZheng Yan /* 3094d16c702fSQu Wenruo * Check key order of two sibling extent buffers. 3095d16c702fSQu Wenruo * 3096d16c702fSQu Wenruo * Return true if something is wrong. 3097d16c702fSQu Wenruo * Return false if everything is fine. 3098d16c702fSQu Wenruo * 3099d16c702fSQu Wenruo * Tree-checker only works inside one tree block, thus the following 3100d16c702fSQu Wenruo * corruption can not be detected by tree-checker: 3101d16c702fSQu Wenruo * 3102d16c702fSQu Wenruo * Leaf @left | Leaf @right 3103d16c702fSQu Wenruo * -------------------------------------------------------------- 3104d16c702fSQu Wenruo * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | 3105d16c702fSQu Wenruo * 3106d16c702fSQu Wenruo * Key f6 in leaf @left itself is valid, but not valid when the next 3107d16c702fSQu Wenruo * key in leaf @right is 7. 3108d16c702fSQu Wenruo * This can only be checked at tree block merge time. 3109d16c702fSQu Wenruo * And since tree checker has ensured all key order in each tree block 3110d16c702fSQu Wenruo * is correct, we only need to bother the last key of @left and the first 3111d16c702fSQu Wenruo * key of @right. 3112d16c702fSQu Wenruo */ 3113d16c702fSQu Wenruo static bool check_sibling_keys(struct extent_buffer *left, 3114d16c702fSQu Wenruo struct extent_buffer *right) 3115d16c702fSQu Wenruo { 3116d16c702fSQu Wenruo struct btrfs_key left_last; 3117d16c702fSQu Wenruo struct btrfs_key right_first; 3118d16c702fSQu Wenruo int level = btrfs_header_level(left); 3119d16c702fSQu Wenruo int nr_left = btrfs_header_nritems(left); 3120d16c702fSQu Wenruo int nr_right = btrfs_header_nritems(right); 3121d16c702fSQu Wenruo 3122d16c702fSQu Wenruo /* No key to check in one of the tree blocks */ 3123d16c702fSQu Wenruo if (!nr_left || !nr_right) 3124d16c702fSQu Wenruo return false; 3125d16c702fSQu Wenruo 3126d16c702fSQu Wenruo if (level) { 3127d16c702fSQu Wenruo btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); 3128d16c702fSQu Wenruo btrfs_node_key_to_cpu(right, &right_first, 0); 3129d16c702fSQu Wenruo } else { 3130d16c702fSQu Wenruo btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); 3131d16c702fSQu Wenruo btrfs_item_key_to_cpu(right, &right_first, 0); 3132d16c702fSQu Wenruo } 3133d16c702fSQu Wenruo 3134d16c702fSQu Wenruo if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) { 3135d16c702fSQu Wenruo btrfs_crit(left->fs_info, 3136d16c702fSQu Wenruo "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", 3137d16c702fSQu Wenruo left_last.objectid, left_last.type, 3138d16c702fSQu Wenruo left_last.offset, right_first.objectid, 3139d16c702fSQu Wenruo right_first.type, right_first.offset); 3140d16c702fSQu Wenruo return true; 3141d16c702fSQu Wenruo } 3142d16c702fSQu Wenruo return false; 3143d16c702fSQu Wenruo } 3144d16c702fSQu Wenruo 3145d16c702fSQu Wenruo /* 314674123bd7SChris Mason * try to push data from one node into the next node left in the 314779f95c82SChris Mason * tree. 3148aa5d6bedSChris Mason * 3149aa5d6bedSChris Mason * returns 0 if some ptrs were pushed left, < 0 if there was some horrible 3150aa5d6bedSChris Mason * error, and > 0 if there was no room in the left hand block. 315174123bd7SChris Mason */ 315298ed5174SChris Mason static int push_node_left(struct btrfs_trans_handle *trans, 31532ff7e61eSJeff Mahoney struct extent_buffer *dst, 3154971a1f66SChris Mason struct extent_buffer *src, int empty) 3155be0e5c09SChris Mason { 3156d30a668fSDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 3157be0e5c09SChris Mason int push_items = 0; 3158bb803951SChris Mason int src_nritems; 3159bb803951SChris Mason int dst_nritems; 3160aa5d6bedSChris Mason int ret = 0; 3161be0e5c09SChris Mason 31625f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 31635f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 31640b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 31657bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 31667bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 316754aa1f4dSChris Mason 3168bce4eae9SChris Mason if (!empty && src_nritems <= 8) 3169971a1f66SChris Mason return 1; 3170971a1f66SChris Mason 3171d397712bSChris Mason if (push_items <= 0) 3172be0e5c09SChris Mason return 1; 3173be0e5c09SChris Mason 3174bce4eae9SChris Mason if (empty) { 3175971a1f66SChris Mason push_items = min(src_nritems, push_items); 3176bce4eae9SChris Mason if (push_items < src_nritems) { 3177bce4eae9SChris Mason /* leave at least 8 pointers in the node if 3178bce4eae9SChris Mason * we aren't going to empty it 3179bce4eae9SChris Mason */ 3180bce4eae9SChris Mason if (src_nritems - push_items < 8) { 3181bce4eae9SChris Mason if (push_items <= 8) 3182bce4eae9SChris Mason return 1; 3183bce4eae9SChris Mason push_items -= 8; 3184bce4eae9SChris Mason } 3185bce4eae9SChris Mason } 3186bce4eae9SChris Mason } else 3187bce4eae9SChris Mason push_items = min(src_nritems - 8, push_items); 318879f95c82SChris Mason 3189d16c702fSQu Wenruo /* dst is the left eb, src is the middle eb */ 3190d16c702fSQu Wenruo if (check_sibling_keys(dst, src)) { 3191d16c702fSQu Wenruo ret = -EUCLEAN; 3192d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 3193d16c702fSQu Wenruo return ret; 3194d16c702fSQu Wenruo } 3195ed874f0dSDavid Sterba ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); 31965de865eeSFilipe David Borba Manana if (ret) { 319766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 31985de865eeSFilipe David Borba Manana return ret; 31995de865eeSFilipe David Borba Manana } 32005f39d397SChris Mason copy_extent_buffer(dst, src, 32015f39d397SChris Mason btrfs_node_key_ptr_offset(dst_nritems), 32025f39d397SChris Mason btrfs_node_key_ptr_offset(0), 3203123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 32045f39d397SChris Mason 3205bb803951SChris Mason if (push_items < src_nritems) { 320657911b8bSJan Schmidt /* 3207bf1d3425SDavid Sterba * Don't call tree_mod_log_insert_move here, key removal was 3208bf1d3425SDavid Sterba * already fully logged by tree_mod_log_eb_copy above. 320957911b8bSJan Schmidt */ 32105f39d397SChris Mason memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), 32115f39d397SChris Mason btrfs_node_key_ptr_offset(push_items), 3212e2fa7227SChris Mason (src_nritems - push_items) * 3213123abc88SChris Mason sizeof(struct btrfs_key_ptr)); 3214bb803951SChris Mason } 32155f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 32165f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 32175f39d397SChris Mason btrfs_mark_buffer_dirty(src); 32185f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 321931840ae1SZheng Yan 3220bb803951SChris Mason return ret; 3221be0e5c09SChris Mason } 3222be0e5c09SChris Mason 322397571fd0SChris Mason /* 322479f95c82SChris Mason * try to push data from one node into the next node right in the 322579f95c82SChris Mason * tree. 322679f95c82SChris Mason * 322779f95c82SChris Mason * returns 0 if some ptrs were pushed, < 0 if there was some horrible 322879f95c82SChris Mason * error, and > 0 if there was no room in the right hand block. 322979f95c82SChris Mason * 323079f95c82SChris Mason * this will only push up to 1/2 the contents of the left node over 323179f95c82SChris Mason */ 32325f39d397SChris Mason static int balance_node_right(struct btrfs_trans_handle *trans, 32335f39d397SChris Mason struct extent_buffer *dst, 32345f39d397SChris Mason struct extent_buffer *src) 323579f95c82SChris Mason { 323655d32ed8SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 323779f95c82SChris Mason int push_items = 0; 323879f95c82SChris Mason int max_push; 323979f95c82SChris Mason int src_nritems; 324079f95c82SChris Mason int dst_nritems; 324179f95c82SChris Mason int ret = 0; 324279f95c82SChris Mason 32437bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 32447bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 32457bb86316SChris Mason 32465f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 32475f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 32480b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 3249d397712bSChris Mason if (push_items <= 0) 325079f95c82SChris Mason return 1; 3251bce4eae9SChris Mason 3252d397712bSChris Mason if (src_nritems < 4) 3253bce4eae9SChris Mason return 1; 325479f95c82SChris Mason 325579f95c82SChris Mason max_push = src_nritems / 2 + 1; 325679f95c82SChris Mason /* don't try to empty the node */ 3257d397712bSChris Mason if (max_push >= src_nritems) 325879f95c82SChris Mason return 1; 3259252c38f0SYan 326079f95c82SChris Mason if (max_push < push_items) 326179f95c82SChris Mason push_items = max_push; 326279f95c82SChris Mason 3263d16c702fSQu Wenruo /* dst is the right eb, src is the middle eb */ 3264d16c702fSQu Wenruo if (check_sibling_keys(src, dst)) { 3265d16c702fSQu Wenruo ret = -EUCLEAN; 3266d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 3267d16c702fSQu Wenruo return ret; 3268d16c702fSQu Wenruo } 3269bf1d3425SDavid Sterba ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); 3270bf1d3425SDavid Sterba BUG_ON(ret < 0); 32715f39d397SChris Mason memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), 32725f39d397SChris Mason btrfs_node_key_ptr_offset(0), 32735f39d397SChris Mason (dst_nritems) * 32745f39d397SChris Mason sizeof(struct btrfs_key_ptr)); 3275d6025579SChris Mason 3276ed874f0dSDavid Sterba ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, 3277ed874f0dSDavid Sterba push_items); 32785de865eeSFilipe David Borba Manana if (ret) { 327966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 32805de865eeSFilipe David Borba Manana return ret; 32815de865eeSFilipe David Borba Manana } 32825f39d397SChris Mason copy_extent_buffer(dst, src, 32835f39d397SChris Mason btrfs_node_key_ptr_offset(0), 32845f39d397SChris Mason btrfs_node_key_ptr_offset(src_nritems - push_items), 3285123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 328679f95c82SChris Mason 32875f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 32885f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 328979f95c82SChris Mason 32905f39d397SChris Mason btrfs_mark_buffer_dirty(src); 32915f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 329231840ae1SZheng Yan 329379f95c82SChris Mason return ret; 329479f95c82SChris Mason } 329579f95c82SChris Mason 329679f95c82SChris Mason /* 329797571fd0SChris Mason * helper function to insert a new root level in the tree. 329897571fd0SChris Mason * A new node is allocated, and a single item is inserted to 329997571fd0SChris Mason * point to the existing root 3300aa5d6bedSChris Mason * 3301aa5d6bedSChris Mason * returns zero on success or < 0 on failure. 330297571fd0SChris Mason */ 3303d397712bSChris Mason static noinline int insert_new_root(struct btrfs_trans_handle *trans, 33045f39d397SChris Mason struct btrfs_root *root, 3305fdd99c72SLiu Bo struct btrfs_path *path, int level) 330674123bd7SChris Mason { 33070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 33087bb86316SChris Mason u64 lower_gen; 33095f39d397SChris Mason struct extent_buffer *lower; 33105f39d397SChris Mason struct extent_buffer *c; 3311925baeddSChris Mason struct extent_buffer *old; 33125f39d397SChris Mason struct btrfs_disk_key lower_key; 3313d9d19a01SDavid Sterba int ret; 33145c680ed6SChris Mason 33155c680ed6SChris Mason BUG_ON(path->nodes[level]); 33165c680ed6SChris Mason BUG_ON(path->nodes[level-1] != root->node); 33175c680ed6SChris Mason 33187bb86316SChris Mason lower = path->nodes[level-1]; 33197bb86316SChris Mason if (level == 1) 33207bb86316SChris Mason btrfs_item_key(lower, &lower_key, 0); 33217bb86316SChris Mason else 33227bb86316SChris Mason btrfs_node_key(lower, &lower_key, 0); 33237bb86316SChris Mason 3324a6279470SFilipe Manana c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level, 33259631e4ccSJosef Bacik root->node->start, 0, 3326cf6f34aaSJosef Bacik BTRFS_NESTING_NEW_ROOT); 33275f39d397SChris Mason if (IS_ERR(c)) 33285f39d397SChris Mason return PTR_ERR(c); 3329925baeddSChris Mason 33300b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 3331f0486c68SYan, Zheng 33325f39d397SChris Mason btrfs_set_header_nritems(c, 1); 33335f39d397SChris Mason btrfs_set_node_key(c, &lower_key, 0); 3334db94535dSChris Mason btrfs_set_node_blockptr(c, 0, lower->start); 33357bb86316SChris Mason lower_gen = btrfs_header_generation(lower); 333631840ae1SZheng Yan WARN_ON(lower_gen != trans->transid); 33377bb86316SChris Mason 33387bb86316SChris Mason btrfs_set_node_ptr_generation(c, 0, lower_gen); 33395f39d397SChris Mason 33405f39d397SChris Mason btrfs_mark_buffer_dirty(c); 3341d5719762SChris Mason 3342925baeddSChris Mason old = root->node; 3343d9d19a01SDavid Sterba ret = tree_mod_log_insert_root(root->node, c, 0); 3344d9d19a01SDavid Sterba BUG_ON(ret < 0); 3345240f62c8SChris Mason rcu_assign_pointer(root->node, c); 3346925baeddSChris Mason 3347925baeddSChris Mason /* the super has an extra ref to root->node */ 3348925baeddSChris Mason free_extent_buffer(old); 3349925baeddSChris Mason 33500b86a832SChris Mason add_root_to_dirty_list(root); 335167439dadSDavid Sterba atomic_inc(&c->refs); 33525f39d397SChris Mason path->nodes[level] = c; 3353ac5887c8SJosef Bacik path->locks[level] = BTRFS_WRITE_LOCK; 335474123bd7SChris Mason path->slots[level] = 0; 335574123bd7SChris Mason return 0; 335674123bd7SChris Mason } 33575c680ed6SChris Mason 33585c680ed6SChris Mason /* 33595c680ed6SChris Mason * worker function to insert a single pointer in a node. 33605c680ed6SChris Mason * the node should have enough room for the pointer already 336197571fd0SChris Mason * 33625c680ed6SChris Mason * slot and level indicate where you want the key to go, and 33635c680ed6SChris Mason * blocknr is the block the key points to. 33645c680ed6SChris Mason */ 3365143bede5SJeff Mahoney static void insert_ptr(struct btrfs_trans_handle *trans, 33666ad3cf6dSDavid Sterba struct btrfs_path *path, 3367143bede5SJeff Mahoney struct btrfs_disk_key *key, u64 bytenr, 3368c3e06965SJan Schmidt int slot, int level) 33695c680ed6SChris Mason { 33705f39d397SChris Mason struct extent_buffer *lower; 33715c680ed6SChris Mason int nritems; 3372f3ea38daSJan Schmidt int ret; 33735c680ed6SChris Mason 33745c680ed6SChris Mason BUG_ON(!path->nodes[level]); 3375f0486c68SYan, Zheng btrfs_assert_tree_locked(path->nodes[level]); 33765f39d397SChris Mason lower = path->nodes[level]; 33775f39d397SChris Mason nritems = btrfs_header_nritems(lower); 3378c293498bSStoyan Gaydarov BUG_ON(slot > nritems); 33796ad3cf6dSDavid Sterba BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); 338074123bd7SChris Mason if (slot != nritems) { 3381bf1d3425SDavid Sterba if (level) { 3382bf1d3425SDavid Sterba ret = tree_mod_log_insert_move(lower, slot + 1, slot, 3383a446a979SDavid Sterba nritems - slot); 3384bf1d3425SDavid Sterba BUG_ON(ret < 0); 3385bf1d3425SDavid Sterba } 33865f39d397SChris Mason memmove_extent_buffer(lower, 33875f39d397SChris Mason btrfs_node_key_ptr_offset(slot + 1), 33885f39d397SChris Mason btrfs_node_key_ptr_offset(slot), 3389123abc88SChris Mason (nritems - slot) * sizeof(struct btrfs_key_ptr)); 339074123bd7SChris Mason } 3391c3e06965SJan Schmidt if (level) { 3392e09c2efeSDavid Sterba ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, 3393e09c2efeSDavid Sterba GFP_NOFS); 3394f3ea38daSJan Schmidt BUG_ON(ret < 0); 3395f3ea38daSJan Schmidt } 33965f39d397SChris Mason btrfs_set_node_key(lower, key, slot); 3397db94535dSChris Mason btrfs_set_node_blockptr(lower, slot, bytenr); 339874493f7aSChris Mason WARN_ON(trans->transid == 0); 339974493f7aSChris Mason btrfs_set_node_ptr_generation(lower, slot, trans->transid); 34005f39d397SChris Mason btrfs_set_header_nritems(lower, nritems + 1); 34015f39d397SChris Mason btrfs_mark_buffer_dirty(lower); 340274123bd7SChris Mason } 340374123bd7SChris Mason 340497571fd0SChris Mason /* 340597571fd0SChris Mason * split the node at the specified level in path in two. 340697571fd0SChris Mason * The path is corrected to point to the appropriate node after the split 340797571fd0SChris Mason * 340897571fd0SChris Mason * Before splitting this tries to make some room in the node by pushing 340997571fd0SChris Mason * left and right, if either one works, it returns right away. 3410aa5d6bedSChris Mason * 3411aa5d6bedSChris Mason * returns 0 on success and < 0 on failure 341297571fd0SChris Mason */ 3413e02119d5SChris Mason static noinline int split_node(struct btrfs_trans_handle *trans, 3414e02119d5SChris Mason struct btrfs_root *root, 3415e02119d5SChris Mason struct btrfs_path *path, int level) 3416be0e5c09SChris Mason { 34170b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 34185f39d397SChris Mason struct extent_buffer *c; 34195f39d397SChris Mason struct extent_buffer *split; 34205f39d397SChris Mason struct btrfs_disk_key disk_key; 3421be0e5c09SChris Mason int mid; 34225c680ed6SChris Mason int ret; 34237518a238SChris Mason u32 c_nritems; 3424be0e5c09SChris Mason 34255f39d397SChris Mason c = path->nodes[level]; 34267bb86316SChris Mason WARN_ON(btrfs_header_generation(c) != trans->transid); 34275f39d397SChris Mason if (c == root->node) { 3428d9abbf1cSJan Schmidt /* 342990f8d62eSJan Schmidt * trying to split the root, lets make a new one 343090f8d62eSJan Schmidt * 3431fdd99c72SLiu Bo * tree mod log: We don't log_removal old root in 343290f8d62eSJan Schmidt * insert_new_root, because that root buffer will be kept as a 343390f8d62eSJan Schmidt * normal node. We are going to log removal of half of the 343490f8d62eSJan Schmidt * elements below with tree_mod_log_eb_copy. We're holding a 343590f8d62eSJan Schmidt * tree lock on the buffer, which is why we cannot race with 343690f8d62eSJan Schmidt * other tree_mod_log users. 3437d9abbf1cSJan Schmidt */ 3438fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, level + 1); 34395c680ed6SChris Mason if (ret) 34405c680ed6SChris Mason return ret; 3441b3612421SChris Mason } else { 3442e66f709bSChris Mason ret = push_nodes_for_insert(trans, root, path, level); 34435f39d397SChris Mason c = path->nodes[level]; 34445f39d397SChris Mason if (!ret && btrfs_header_nritems(c) < 34450b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) 3446e66f709bSChris Mason return 0; 344754aa1f4dSChris Mason if (ret < 0) 344854aa1f4dSChris Mason return ret; 34495c680ed6SChris Mason } 3450e66f709bSChris Mason 34515f39d397SChris Mason c_nritems = btrfs_header_nritems(c); 34525d4f98a2SYan Zheng mid = (c_nritems + 1) / 2; 34535d4f98a2SYan Zheng btrfs_node_key(c, &disk_key, mid); 34547bb86316SChris Mason 3455a6279470SFilipe Manana split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level, 34564dff97e6SJosef Bacik c->start, 0, BTRFS_NESTING_SPLIT); 34575f39d397SChris Mason if (IS_ERR(split)) 34585f39d397SChris Mason return PTR_ERR(split); 345954aa1f4dSChris Mason 34600b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 3461bc877d28SNikolay Borisov ASSERT(btrfs_header_level(c) == level); 34625f39d397SChris Mason 3463ed874f0dSDavid Sterba ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); 34645de865eeSFilipe David Borba Manana if (ret) { 346566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 34665de865eeSFilipe David Borba Manana return ret; 34675de865eeSFilipe David Borba Manana } 34685f39d397SChris Mason copy_extent_buffer(split, c, 34695f39d397SChris Mason btrfs_node_key_ptr_offset(0), 34705f39d397SChris Mason btrfs_node_key_ptr_offset(mid), 3471123abc88SChris Mason (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); 34725f39d397SChris Mason btrfs_set_header_nritems(split, c_nritems - mid); 34735f39d397SChris Mason btrfs_set_header_nritems(c, mid); 3474aa5d6bedSChris Mason 34755f39d397SChris Mason btrfs_mark_buffer_dirty(c); 34765f39d397SChris Mason btrfs_mark_buffer_dirty(split); 34775f39d397SChris Mason 34786ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, split->start, 3479c3e06965SJan Schmidt path->slots[level + 1] + 1, level + 1); 3480aa5d6bedSChris Mason 34815de08d7dSChris Mason if (path->slots[level] >= mid) { 34825c680ed6SChris Mason path->slots[level] -= mid; 3483925baeddSChris Mason btrfs_tree_unlock(c); 34845f39d397SChris Mason free_extent_buffer(c); 34855f39d397SChris Mason path->nodes[level] = split; 34865c680ed6SChris Mason path->slots[level + 1] += 1; 3487eb60ceacSChris Mason } else { 3488925baeddSChris Mason btrfs_tree_unlock(split); 34895f39d397SChris Mason free_extent_buffer(split); 3490be0e5c09SChris Mason } 3491d5286a92SNikolay Borisov return 0; 3492be0e5c09SChris Mason } 3493be0e5c09SChris Mason 349474123bd7SChris Mason /* 349574123bd7SChris Mason * how many bytes are required to store the items in a leaf. start 349674123bd7SChris Mason * and nr indicate which items in the leaf to check. This totals up the 349774123bd7SChris Mason * space used both by the item structs and the item data 349874123bd7SChris Mason */ 34995f39d397SChris Mason static int leaf_space_used(struct extent_buffer *l, int start, int nr) 3500be0e5c09SChris Mason { 350141be1f3bSJosef Bacik struct btrfs_item *start_item; 350241be1f3bSJosef Bacik struct btrfs_item *end_item; 3503be0e5c09SChris Mason int data_len; 35045f39d397SChris Mason int nritems = btrfs_header_nritems(l); 3505d4dbff95SChris Mason int end = min(nritems, start + nr) - 1; 3506be0e5c09SChris Mason 3507be0e5c09SChris Mason if (!nr) 3508be0e5c09SChris Mason return 0; 3509dd3cc16bSRoss Kirk start_item = btrfs_item_nr(start); 3510dd3cc16bSRoss Kirk end_item = btrfs_item_nr(end); 3511a31356b9SDavid Sterba data_len = btrfs_item_offset(l, start_item) + 3512a31356b9SDavid Sterba btrfs_item_size(l, start_item); 3513a31356b9SDavid Sterba data_len = data_len - btrfs_item_offset(l, end_item); 35140783fcfcSChris Mason data_len += sizeof(struct btrfs_item) * nr; 3515d4dbff95SChris Mason WARN_ON(data_len < 0); 3516be0e5c09SChris Mason return data_len; 3517be0e5c09SChris Mason } 3518be0e5c09SChris Mason 351974123bd7SChris Mason /* 3520d4dbff95SChris Mason * The space between the end of the leaf items and 3521d4dbff95SChris Mason * the start of the leaf data. IOW, how much room 3522d4dbff95SChris Mason * the leaf has left for both items and data 3523d4dbff95SChris Mason */ 3524e902baacSDavid Sterba noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) 3525d4dbff95SChris Mason { 3526e902baacSDavid Sterba struct btrfs_fs_info *fs_info = leaf->fs_info; 35275f39d397SChris Mason int nritems = btrfs_header_nritems(leaf); 35285f39d397SChris Mason int ret; 35290b246afaSJeff Mahoney 35300b246afaSJeff Mahoney ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); 35315f39d397SChris Mason if (ret < 0) { 35320b246afaSJeff Mahoney btrfs_crit(fs_info, 3533efe120a0SFrank Holton "leaf free space ret %d, leaf data size %lu, used %d nritems %d", 3534da17066cSJeff Mahoney ret, 35350b246afaSJeff Mahoney (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), 35365f39d397SChris Mason leaf_space_used(leaf, 0, nritems), nritems); 35375f39d397SChris Mason } 35385f39d397SChris Mason return ret; 3539d4dbff95SChris Mason } 3540d4dbff95SChris Mason 354199d8f83cSChris Mason /* 354299d8f83cSChris Mason * min slot controls the lowest index we're willing to push to the 354399d8f83cSChris Mason * right. We'll push up to and including min_slot, but no lower 354499d8f83cSChris Mason */ 3545f72f0010SDavid Sterba static noinline int __push_leaf_right(struct btrfs_path *path, 354644871b1bSChris Mason int data_size, int empty, 354744871b1bSChris Mason struct extent_buffer *right, 354899d8f83cSChris Mason int free_space, u32 left_nritems, 354999d8f83cSChris Mason u32 min_slot) 355000ec4c51SChris Mason { 3551f72f0010SDavid Sterba struct btrfs_fs_info *fs_info = right->fs_info; 35525f39d397SChris Mason struct extent_buffer *left = path->nodes[0]; 355344871b1bSChris Mason struct extent_buffer *upper = path->nodes[1]; 3554cfed81a0SChris Mason struct btrfs_map_token token; 35555f39d397SChris Mason struct btrfs_disk_key disk_key; 355600ec4c51SChris Mason int slot; 355734a38218SChris Mason u32 i; 355800ec4c51SChris Mason int push_space = 0; 355900ec4c51SChris Mason int push_items = 0; 35600783fcfcSChris Mason struct btrfs_item *item; 356134a38218SChris Mason u32 nr; 35627518a238SChris Mason u32 right_nritems; 35635f39d397SChris Mason u32 data_end; 3564db94535dSChris Mason u32 this_item_size; 356500ec4c51SChris Mason 356634a38218SChris Mason if (empty) 356734a38218SChris Mason nr = 0; 356834a38218SChris Mason else 356999d8f83cSChris Mason nr = max_t(u32, 1, min_slot); 357034a38218SChris Mason 357131840ae1SZheng Yan if (path->slots[0] >= left_nritems) 357287b29b20SYan Zheng push_space += data_size; 357331840ae1SZheng Yan 357444871b1bSChris Mason slot = path->slots[1]; 357534a38218SChris Mason i = left_nritems - 1; 357634a38218SChris Mason while (i >= nr) { 3577dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3578db94535dSChris Mason 357931840ae1SZheng Yan if (!empty && push_items > 0) { 358031840ae1SZheng Yan if (path->slots[0] > i) 358131840ae1SZheng Yan break; 358231840ae1SZheng Yan if (path->slots[0] == i) { 3583e902baacSDavid Sterba int space = btrfs_leaf_free_space(left); 3584e902baacSDavid Sterba 358531840ae1SZheng Yan if (space + push_space * 2 > free_space) 358631840ae1SZheng Yan break; 358731840ae1SZheng Yan } 358831840ae1SZheng Yan } 358931840ae1SZheng Yan 359000ec4c51SChris Mason if (path->slots[0] == i) 359187b29b20SYan Zheng push_space += data_size; 3592db94535dSChris Mason 3593db94535dSChris Mason this_item_size = btrfs_item_size(left, item); 3594db94535dSChris Mason if (this_item_size + sizeof(*item) + push_space > free_space) 359500ec4c51SChris Mason break; 359631840ae1SZheng Yan 359700ec4c51SChris Mason push_items++; 3598db94535dSChris Mason push_space += this_item_size + sizeof(*item); 359934a38218SChris Mason if (i == 0) 360034a38218SChris Mason break; 360134a38218SChris Mason i--; 3602db94535dSChris Mason } 36035f39d397SChris Mason 3604925baeddSChris Mason if (push_items == 0) 3605925baeddSChris Mason goto out_unlock; 36065f39d397SChris Mason 36076c1500f2SJulia Lawall WARN_ON(!empty && push_items == left_nritems); 36085f39d397SChris Mason 360900ec4c51SChris Mason /* push left to right */ 36105f39d397SChris Mason right_nritems = btrfs_header_nritems(right); 361134a38218SChris Mason 36125f39d397SChris Mason push_space = btrfs_item_end_nr(left, left_nritems - push_items); 36138f881e8cSDavid Sterba push_space -= leaf_data_end(left); 36145f39d397SChris Mason 361500ec4c51SChris Mason /* make room in the right data area */ 36168f881e8cSDavid Sterba data_end = leaf_data_end(right); 36175f39d397SChris Mason memmove_extent_buffer(right, 36183d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end - push_space, 36193d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end, 36200b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); 36215f39d397SChris Mason 362200ec4c51SChris Mason /* copy from the left data area */ 36233d9ec8c4SNikolay Borisov copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + 36240b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 36258f881e8cSDavid Sterba BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), 3626d6025579SChris Mason push_space); 36275f39d397SChris Mason 36285f39d397SChris Mason memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), 36295f39d397SChris Mason btrfs_item_nr_offset(0), 36300783fcfcSChris Mason right_nritems * sizeof(struct btrfs_item)); 36315f39d397SChris Mason 363200ec4c51SChris Mason /* copy the items from left to right */ 36335f39d397SChris Mason copy_extent_buffer(right, left, btrfs_item_nr_offset(0), 36345f39d397SChris Mason btrfs_item_nr_offset(left_nritems - push_items), 36350783fcfcSChris Mason push_items * sizeof(struct btrfs_item)); 363600ec4c51SChris Mason 363700ec4c51SChris Mason /* update the item pointers */ 3638c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 36397518a238SChris Mason right_nritems += push_items; 36405f39d397SChris Mason btrfs_set_header_nritems(right, right_nritems); 36410b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 36427518a238SChris Mason for (i = 0; i < right_nritems; i++) { 3643dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3644cc4c13d5SDavid Sterba push_space -= btrfs_token_item_size(&token, item); 3645cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, push_space); 3646db94535dSChris Mason } 3647db94535dSChris Mason 36487518a238SChris Mason left_nritems -= push_items; 36495f39d397SChris Mason btrfs_set_header_nritems(left, left_nritems); 365000ec4c51SChris Mason 365134a38218SChris Mason if (left_nritems) 36525f39d397SChris Mason btrfs_mark_buffer_dirty(left); 3653f0486c68SYan, Zheng else 36546a884d7dSDavid Sterba btrfs_clean_tree_block(left); 3655f0486c68SYan, Zheng 36565f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3657a429e513SChris Mason 36585f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 36595f39d397SChris Mason btrfs_set_node_key(upper, &disk_key, slot + 1); 3660d6025579SChris Mason btrfs_mark_buffer_dirty(upper); 366102217ed2SChris Mason 366200ec4c51SChris Mason /* then fixup the leaf pointer in the path */ 36637518a238SChris Mason if (path->slots[0] >= left_nritems) { 36647518a238SChris Mason path->slots[0] -= left_nritems; 3665925baeddSChris Mason if (btrfs_header_nritems(path->nodes[0]) == 0) 36666a884d7dSDavid Sterba btrfs_clean_tree_block(path->nodes[0]); 3667925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 36685f39d397SChris Mason free_extent_buffer(path->nodes[0]); 36695f39d397SChris Mason path->nodes[0] = right; 367000ec4c51SChris Mason path->slots[1] += 1; 367100ec4c51SChris Mason } else { 3672925baeddSChris Mason btrfs_tree_unlock(right); 36735f39d397SChris Mason free_extent_buffer(right); 367400ec4c51SChris Mason } 367500ec4c51SChris Mason return 0; 3676925baeddSChris Mason 3677925baeddSChris Mason out_unlock: 3678925baeddSChris Mason btrfs_tree_unlock(right); 3679925baeddSChris Mason free_extent_buffer(right); 3680925baeddSChris Mason return 1; 368100ec4c51SChris Mason } 3682925baeddSChris Mason 368300ec4c51SChris Mason /* 368444871b1bSChris Mason * push some data in the path leaf to the right, trying to free up at 368574123bd7SChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 368644871b1bSChris Mason * 368744871b1bSChris Mason * returns 1 if the push failed because the other node didn't have enough 368844871b1bSChris Mason * room, 0 if everything worked out and < 0 if there were major errors. 368999d8f83cSChris Mason * 369099d8f83cSChris Mason * this will push starting from min_slot to the end of the leaf. It won't 369199d8f83cSChris Mason * push any slot lower than min_slot 369274123bd7SChris Mason */ 369344871b1bSChris Mason static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root 369499d8f83cSChris Mason *root, struct btrfs_path *path, 369599d8f83cSChris Mason int min_data_size, int data_size, 369699d8f83cSChris Mason int empty, u32 min_slot) 3697be0e5c09SChris Mason { 369844871b1bSChris Mason struct extent_buffer *left = path->nodes[0]; 369944871b1bSChris Mason struct extent_buffer *right; 370044871b1bSChris Mason struct extent_buffer *upper; 370144871b1bSChris Mason int slot; 370244871b1bSChris Mason int free_space; 370344871b1bSChris Mason u32 left_nritems; 370444871b1bSChris Mason int ret; 370544871b1bSChris Mason 370644871b1bSChris Mason if (!path->nodes[1]) 370744871b1bSChris Mason return 1; 370844871b1bSChris Mason 370944871b1bSChris Mason slot = path->slots[1]; 371044871b1bSChris Mason upper = path->nodes[1]; 371144871b1bSChris Mason if (slot >= btrfs_header_nritems(upper) - 1) 371244871b1bSChris Mason return 1; 371344871b1bSChris Mason 371444871b1bSChris Mason btrfs_assert_tree_locked(path->nodes[1]); 371544871b1bSChris Mason 37164b231ae4SDavid Sterba right = btrfs_read_node_slot(upper, slot + 1); 3717fb770ae4SLiu Bo /* 3718fb770ae4SLiu Bo * slot + 1 is not valid or we fail to read the right node, 3719fb770ae4SLiu Bo * no big deal, just return. 3720fb770ae4SLiu Bo */ 3721fb770ae4SLiu Bo if (IS_ERR(right)) 372291ca338dSTsutomu Itoh return 1; 372391ca338dSTsutomu Itoh 3724bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 372544871b1bSChris Mason 3726e902baacSDavid Sterba free_space = btrfs_leaf_free_space(right); 372744871b1bSChris Mason if (free_space < data_size) 372844871b1bSChris Mason goto out_unlock; 372944871b1bSChris Mason 373044871b1bSChris Mason /* cow and double check */ 373144871b1bSChris Mason ret = btrfs_cow_block(trans, root, right, upper, 3732bf59a5a2SJosef Bacik slot + 1, &right, BTRFS_NESTING_RIGHT_COW); 373344871b1bSChris Mason if (ret) 373444871b1bSChris Mason goto out_unlock; 373544871b1bSChris Mason 3736e902baacSDavid Sterba free_space = btrfs_leaf_free_space(right); 373744871b1bSChris Mason if (free_space < data_size) 373844871b1bSChris Mason goto out_unlock; 373944871b1bSChris Mason 374044871b1bSChris Mason left_nritems = btrfs_header_nritems(left); 374144871b1bSChris Mason if (left_nritems == 0) 374244871b1bSChris Mason goto out_unlock; 374344871b1bSChris Mason 3744d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3745d16c702fSQu Wenruo ret = -EUCLEAN; 3746d16c702fSQu Wenruo btrfs_tree_unlock(right); 3747d16c702fSQu Wenruo free_extent_buffer(right); 3748d16c702fSQu Wenruo return ret; 3749d16c702fSQu Wenruo } 37502ef1fed2SFilipe David Borba Manana if (path->slots[0] == left_nritems && !empty) { 37512ef1fed2SFilipe David Borba Manana /* Key greater than all keys in the leaf, right neighbor has 37522ef1fed2SFilipe David Borba Manana * enough room for it and we're not emptying our leaf to delete 37532ef1fed2SFilipe David Borba Manana * it, therefore use right neighbor to insert the new item and 375452042d8eSAndrea Gelmini * no need to touch/dirty our left leaf. */ 37552ef1fed2SFilipe David Borba Manana btrfs_tree_unlock(left); 37562ef1fed2SFilipe David Borba Manana free_extent_buffer(left); 37572ef1fed2SFilipe David Borba Manana path->nodes[0] = right; 37582ef1fed2SFilipe David Borba Manana path->slots[0] = 0; 37592ef1fed2SFilipe David Borba Manana path->slots[1]++; 37602ef1fed2SFilipe David Borba Manana return 0; 37612ef1fed2SFilipe David Borba Manana } 37622ef1fed2SFilipe David Borba Manana 3763f72f0010SDavid Sterba return __push_leaf_right(path, min_data_size, empty, 376499d8f83cSChris Mason right, free_space, left_nritems, min_slot); 376544871b1bSChris Mason out_unlock: 376644871b1bSChris Mason btrfs_tree_unlock(right); 376744871b1bSChris Mason free_extent_buffer(right); 376844871b1bSChris Mason return 1; 376944871b1bSChris Mason } 377044871b1bSChris Mason 377144871b1bSChris Mason /* 377244871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 377344871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 377499d8f83cSChris Mason * 377599d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 377699d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the 377799d8f83cSChris Mason * items 377844871b1bSChris Mason */ 37798087c193SDavid Sterba static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, 378044871b1bSChris Mason int empty, struct extent_buffer *left, 378199d8f83cSChris Mason int free_space, u32 right_nritems, 378299d8f83cSChris Mason u32 max_slot) 378344871b1bSChris Mason { 37848087c193SDavid Sterba struct btrfs_fs_info *fs_info = left->fs_info; 37855f39d397SChris Mason struct btrfs_disk_key disk_key; 37865f39d397SChris Mason struct extent_buffer *right = path->nodes[0]; 3787be0e5c09SChris Mason int i; 3788be0e5c09SChris Mason int push_space = 0; 3789be0e5c09SChris Mason int push_items = 0; 37900783fcfcSChris Mason struct btrfs_item *item; 37917518a238SChris Mason u32 old_left_nritems; 379234a38218SChris Mason u32 nr; 3793aa5d6bedSChris Mason int ret = 0; 3794db94535dSChris Mason u32 this_item_size; 3795db94535dSChris Mason u32 old_left_item_size; 3796cfed81a0SChris Mason struct btrfs_map_token token; 3797cfed81a0SChris Mason 379834a38218SChris Mason if (empty) 379999d8f83cSChris Mason nr = min(right_nritems, max_slot); 380034a38218SChris Mason else 380199d8f83cSChris Mason nr = min(right_nritems - 1, max_slot); 380234a38218SChris Mason 380334a38218SChris Mason for (i = 0; i < nr; i++) { 3804dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3805db94535dSChris Mason 380631840ae1SZheng Yan if (!empty && push_items > 0) { 380731840ae1SZheng Yan if (path->slots[0] < i) 380831840ae1SZheng Yan break; 380931840ae1SZheng Yan if (path->slots[0] == i) { 3810e902baacSDavid Sterba int space = btrfs_leaf_free_space(right); 3811e902baacSDavid Sterba 381231840ae1SZheng Yan if (space + push_space * 2 > free_space) 381331840ae1SZheng Yan break; 381431840ae1SZheng Yan } 381531840ae1SZheng Yan } 381631840ae1SZheng Yan 3817be0e5c09SChris Mason if (path->slots[0] == i) 381887b29b20SYan Zheng push_space += data_size; 3819db94535dSChris Mason 3820db94535dSChris Mason this_item_size = btrfs_item_size(right, item); 3821db94535dSChris Mason if (this_item_size + sizeof(*item) + push_space > free_space) 3822be0e5c09SChris Mason break; 3823db94535dSChris Mason 3824be0e5c09SChris Mason push_items++; 3825db94535dSChris Mason push_space += this_item_size + sizeof(*item); 3826be0e5c09SChris Mason } 3827db94535dSChris Mason 3828be0e5c09SChris Mason if (push_items == 0) { 3829925baeddSChris Mason ret = 1; 3830925baeddSChris Mason goto out; 3831be0e5c09SChris Mason } 3832fae7f21cSDulshani Gunawardhana WARN_ON(!empty && push_items == btrfs_header_nritems(right)); 38335f39d397SChris Mason 3834be0e5c09SChris Mason /* push data from right to left */ 38355f39d397SChris Mason copy_extent_buffer(left, right, 38365f39d397SChris Mason btrfs_item_nr_offset(btrfs_header_nritems(left)), 38375f39d397SChris Mason btrfs_item_nr_offset(0), 38385f39d397SChris Mason push_items * sizeof(struct btrfs_item)); 38395f39d397SChris Mason 38400b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - 38415f39d397SChris Mason btrfs_item_offset_nr(right, push_items - 1); 38425f39d397SChris Mason 38433d9ec8c4SNikolay Borisov copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + 38448f881e8cSDavid Sterba leaf_data_end(left) - push_space, 38453d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + 38465f39d397SChris Mason btrfs_item_offset_nr(right, push_items - 1), 3847be0e5c09SChris Mason push_space); 38485f39d397SChris Mason old_left_nritems = btrfs_header_nritems(left); 384987b29b20SYan Zheng BUG_ON(old_left_nritems <= 0); 3850eb60ceacSChris Mason 3851c82f823cSDavid Sterba btrfs_init_map_token(&token, left); 3852db94535dSChris Mason old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); 3853be0e5c09SChris Mason for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { 38545f39d397SChris Mason u32 ioff; 3855db94535dSChris Mason 3856dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3857db94535dSChris Mason 3858cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 3859cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, 3860cc4c13d5SDavid Sterba ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); 3861be0e5c09SChris Mason } 38625f39d397SChris Mason btrfs_set_header_nritems(left, old_left_nritems + push_items); 3863be0e5c09SChris Mason 3864be0e5c09SChris Mason /* fixup right node */ 386531b1a2bdSJulia Lawall if (push_items > right_nritems) 386631b1a2bdSJulia Lawall WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, 3867d397712bSChris Mason right_nritems); 386834a38218SChris Mason 386934a38218SChris Mason if (push_items < right_nritems) { 38705f39d397SChris Mason push_space = btrfs_item_offset_nr(right, push_items - 1) - 38718f881e8cSDavid Sterba leaf_data_end(right); 38723d9ec8c4SNikolay Borisov memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + 38730b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 38743d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + 38758f881e8cSDavid Sterba leaf_data_end(right), push_space); 38765f39d397SChris Mason 38775f39d397SChris Mason memmove_extent_buffer(right, btrfs_item_nr_offset(0), 38785f39d397SChris Mason btrfs_item_nr_offset(push_items), 38795f39d397SChris Mason (btrfs_header_nritems(right) - push_items) * 38800783fcfcSChris Mason sizeof(struct btrfs_item)); 388134a38218SChris Mason } 3882c82f823cSDavid Sterba 3883c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 3884eef1c494SYan right_nritems -= push_items; 3885eef1c494SYan btrfs_set_header_nritems(right, right_nritems); 38860b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 38875f39d397SChris Mason for (i = 0; i < right_nritems; i++) { 3888dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3889db94535dSChris Mason 3890cc4c13d5SDavid Sterba push_space = push_space - btrfs_token_item_size(&token, item); 3891cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, push_space); 3892db94535dSChris Mason } 3893eb60ceacSChris Mason 38945f39d397SChris Mason btrfs_mark_buffer_dirty(left); 389534a38218SChris Mason if (right_nritems) 38965f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3897f0486c68SYan, Zheng else 38986a884d7dSDavid Sterba btrfs_clean_tree_block(right); 3899098f59c2SChris Mason 39005f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 3901b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 3902be0e5c09SChris Mason 3903be0e5c09SChris Mason /* then fixup the leaf pointer in the path */ 3904be0e5c09SChris Mason if (path->slots[0] < push_items) { 3905be0e5c09SChris Mason path->slots[0] += old_left_nritems; 3906925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 39075f39d397SChris Mason free_extent_buffer(path->nodes[0]); 39085f39d397SChris Mason path->nodes[0] = left; 3909be0e5c09SChris Mason path->slots[1] -= 1; 3910be0e5c09SChris Mason } else { 3911925baeddSChris Mason btrfs_tree_unlock(left); 39125f39d397SChris Mason free_extent_buffer(left); 3913be0e5c09SChris Mason path->slots[0] -= push_items; 3914be0e5c09SChris Mason } 3915eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 3916aa5d6bedSChris Mason return ret; 3917925baeddSChris Mason out: 3918925baeddSChris Mason btrfs_tree_unlock(left); 3919925baeddSChris Mason free_extent_buffer(left); 3920925baeddSChris Mason return ret; 3921be0e5c09SChris Mason } 3922be0e5c09SChris Mason 392374123bd7SChris Mason /* 392444871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 392544871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 392699d8f83cSChris Mason * 392799d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 392899d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the 392999d8f83cSChris Mason * items 393044871b1bSChris Mason */ 393144871b1bSChris Mason static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root 393299d8f83cSChris Mason *root, struct btrfs_path *path, int min_data_size, 393399d8f83cSChris Mason int data_size, int empty, u32 max_slot) 393444871b1bSChris Mason { 393544871b1bSChris Mason struct extent_buffer *right = path->nodes[0]; 393644871b1bSChris Mason struct extent_buffer *left; 393744871b1bSChris Mason int slot; 393844871b1bSChris Mason int free_space; 393944871b1bSChris Mason u32 right_nritems; 394044871b1bSChris Mason int ret = 0; 394144871b1bSChris Mason 394244871b1bSChris Mason slot = path->slots[1]; 394344871b1bSChris Mason if (slot == 0) 394444871b1bSChris Mason return 1; 394544871b1bSChris Mason if (!path->nodes[1]) 394644871b1bSChris Mason return 1; 394744871b1bSChris Mason 394844871b1bSChris Mason right_nritems = btrfs_header_nritems(right); 394944871b1bSChris Mason if (right_nritems == 0) 395044871b1bSChris Mason return 1; 395144871b1bSChris Mason 395244871b1bSChris Mason btrfs_assert_tree_locked(path->nodes[1]); 395344871b1bSChris Mason 39544b231ae4SDavid Sterba left = btrfs_read_node_slot(path->nodes[1], slot - 1); 3955fb770ae4SLiu Bo /* 3956fb770ae4SLiu Bo * slot - 1 is not valid or we fail to read the left node, 3957fb770ae4SLiu Bo * no big deal, just return. 3958fb770ae4SLiu Bo */ 3959fb770ae4SLiu Bo if (IS_ERR(left)) 396091ca338dSTsutomu Itoh return 1; 396191ca338dSTsutomu Itoh 3962bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 396344871b1bSChris Mason 3964e902baacSDavid Sterba free_space = btrfs_leaf_free_space(left); 396544871b1bSChris Mason if (free_space < data_size) { 396644871b1bSChris Mason ret = 1; 396744871b1bSChris Mason goto out; 396844871b1bSChris Mason } 396944871b1bSChris Mason 397044871b1bSChris Mason /* cow and double check */ 397144871b1bSChris Mason ret = btrfs_cow_block(trans, root, left, 39729631e4ccSJosef Bacik path->nodes[1], slot - 1, &left, 3973bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 397444871b1bSChris Mason if (ret) { 397544871b1bSChris Mason /* we hit -ENOSPC, but it isn't fatal here */ 397679787eaaSJeff Mahoney if (ret == -ENOSPC) 397744871b1bSChris Mason ret = 1; 397844871b1bSChris Mason goto out; 397944871b1bSChris Mason } 398044871b1bSChris Mason 3981e902baacSDavid Sterba free_space = btrfs_leaf_free_space(left); 398244871b1bSChris Mason if (free_space < data_size) { 398344871b1bSChris Mason ret = 1; 398444871b1bSChris Mason goto out; 398544871b1bSChris Mason } 398644871b1bSChris Mason 3987d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3988d16c702fSQu Wenruo ret = -EUCLEAN; 3989d16c702fSQu Wenruo goto out; 3990d16c702fSQu Wenruo } 39918087c193SDavid Sterba return __push_leaf_left(path, min_data_size, 399299d8f83cSChris Mason empty, left, free_space, right_nritems, 399399d8f83cSChris Mason max_slot); 399444871b1bSChris Mason out: 399544871b1bSChris Mason btrfs_tree_unlock(left); 399644871b1bSChris Mason free_extent_buffer(left); 399744871b1bSChris Mason return ret; 399844871b1bSChris Mason } 399944871b1bSChris Mason 400044871b1bSChris Mason /* 400174123bd7SChris Mason * split the path's leaf in two, making sure there is at least data_size 400274123bd7SChris Mason * available for the resulting leaf level of the path. 400374123bd7SChris Mason */ 4004143bede5SJeff Mahoney static noinline void copy_for_split(struct btrfs_trans_handle *trans, 400544871b1bSChris Mason struct btrfs_path *path, 400644871b1bSChris Mason struct extent_buffer *l, 400744871b1bSChris Mason struct extent_buffer *right, 400844871b1bSChris Mason int slot, int mid, int nritems) 4009be0e5c09SChris Mason { 401094f94ad9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 4011be0e5c09SChris Mason int data_copy_size; 4012be0e5c09SChris Mason int rt_data_off; 4013be0e5c09SChris Mason int i; 4014d4dbff95SChris Mason struct btrfs_disk_key disk_key; 4015cfed81a0SChris Mason struct btrfs_map_token token; 4016cfed81a0SChris Mason 40175f39d397SChris Mason nritems = nritems - mid; 40185f39d397SChris Mason btrfs_set_header_nritems(right, nritems); 40198f881e8cSDavid Sterba data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l); 40205f39d397SChris Mason 40215f39d397SChris Mason copy_extent_buffer(right, l, btrfs_item_nr_offset(0), 40225f39d397SChris Mason btrfs_item_nr_offset(mid), 40235f39d397SChris Mason nritems * sizeof(struct btrfs_item)); 40245f39d397SChris Mason 40255f39d397SChris Mason copy_extent_buffer(right, l, 40263d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - 40273d9ec8c4SNikolay Borisov data_copy_size, BTRFS_LEAF_DATA_OFFSET + 40288f881e8cSDavid Sterba leaf_data_end(l), data_copy_size); 402974123bd7SChris Mason 40300b246afaSJeff Mahoney rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); 40315f39d397SChris Mason 4032c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 40335f39d397SChris Mason for (i = 0; i < nritems; i++) { 4034dd3cc16bSRoss Kirk struct btrfs_item *item = btrfs_item_nr(i); 4035db94535dSChris Mason u32 ioff; 4036db94535dSChris Mason 4037cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4038cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, ioff + rt_data_off); 40390783fcfcSChris Mason } 404074123bd7SChris Mason 40415f39d397SChris Mason btrfs_set_header_nritems(l, mid); 40425f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 40436ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); 40445f39d397SChris Mason 40455f39d397SChris Mason btrfs_mark_buffer_dirty(right); 40465f39d397SChris Mason btrfs_mark_buffer_dirty(l); 4047eb60ceacSChris Mason BUG_ON(path->slots[0] != slot); 40485f39d397SChris Mason 4049be0e5c09SChris Mason if (mid <= slot) { 4050925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 40515f39d397SChris Mason free_extent_buffer(path->nodes[0]); 40525f39d397SChris Mason path->nodes[0] = right; 4053be0e5c09SChris Mason path->slots[0] -= mid; 4054be0e5c09SChris Mason path->slots[1] += 1; 4055925baeddSChris Mason } else { 4056925baeddSChris Mason btrfs_tree_unlock(right); 40575f39d397SChris Mason free_extent_buffer(right); 4058925baeddSChris Mason } 40595f39d397SChris Mason 4060eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 406144871b1bSChris Mason } 406244871b1bSChris Mason 406344871b1bSChris Mason /* 406499d8f83cSChris Mason * double splits happen when we need to insert a big item in the middle 406599d8f83cSChris Mason * of a leaf. A double split can leave us with 3 mostly empty leaves: 406699d8f83cSChris Mason * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] 406799d8f83cSChris Mason * A B C 406899d8f83cSChris Mason * 406999d8f83cSChris Mason * We avoid this by trying to push the items on either side of our target 407099d8f83cSChris Mason * into the adjacent leaves. If all goes well we can avoid the double split 407199d8f83cSChris Mason * completely. 407299d8f83cSChris Mason */ 407399d8f83cSChris Mason static noinline int push_for_double_split(struct btrfs_trans_handle *trans, 407499d8f83cSChris Mason struct btrfs_root *root, 407599d8f83cSChris Mason struct btrfs_path *path, 407699d8f83cSChris Mason int data_size) 407799d8f83cSChris Mason { 407899d8f83cSChris Mason int ret; 407999d8f83cSChris Mason int progress = 0; 408099d8f83cSChris Mason int slot; 408199d8f83cSChris Mason u32 nritems; 40825a4267caSFilipe David Borba Manana int space_needed = data_size; 408399d8f83cSChris Mason 408499d8f83cSChris Mason slot = path->slots[0]; 40855a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(path->nodes[0])) 4086e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 408799d8f83cSChris Mason 408899d8f83cSChris Mason /* 408999d8f83cSChris Mason * try to push all the items after our slot into the 409099d8f83cSChris Mason * right leaf 409199d8f83cSChris Mason */ 40925a4267caSFilipe David Borba Manana ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); 409399d8f83cSChris Mason if (ret < 0) 409499d8f83cSChris Mason return ret; 409599d8f83cSChris Mason 409699d8f83cSChris Mason if (ret == 0) 409799d8f83cSChris Mason progress++; 409899d8f83cSChris Mason 409999d8f83cSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 410099d8f83cSChris Mason /* 410199d8f83cSChris Mason * our goal is to get our slot at the start or end of a leaf. If 410299d8f83cSChris Mason * we've done so we're done 410399d8f83cSChris Mason */ 410499d8f83cSChris Mason if (path->slots[0] == 0 || path->slots[0] == nritems) 410599d8f83cSChris Mason return 0; 410699d8f83cSChris Mason 4107e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 410899d8f83cSChris Mason return 0; 410999d8f83cSChris Mason 411099d8f83cSChris Mason /* try to push all the items before our slot into the next leaf */ 411199d8f83cSChris Mason slot = path->slots[0]; 4112263d3995SFilipe Manana space_needed = data_size; 4113263d3995SFilipe Manana if (slot > 0) 4114e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 41155a4267caSFilipe David Borba Manana ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); 411699d8f83cSChris Mason if (ret < 0) 411799d8f83cSChris Mason return ret; 411899d8f83cSChris Mason 411999d8f83cSChris Mason if (ret == 0) 412099d8f83cSChris Mason progress++; 412199d8f83cSChris Mason 412299d8f83cSChris Mason if (progress) 412399d8f83cSChris Mason return 0; 412499d8f83cSChris Mason return 1; 412599d8f83cSChris Mason } 412699d8f83cSChris Mason 412799d8f83cSChris Mason /* 412844871b1bSChris Mason * split the path's leaf in two, making sure there is at least data_size 412944871b1bSChris Mason * available for the resulting leaf level of the path. 413044871b1bSChris Mason * 413144871b1bSChris Mason * returns 0 if all went well and < 0 on failure. 413244871b1bSChris Mason */ 413344871b1bSChris Mason static noinline int split_leaf(struct btrfs_trans_handle *trans, 413444871b1bSChris Mason struct btrfs_root *root, 4135310712b2SOmar Sandoval const struct btrfs_key *ins_key, 413644871b1bSChris Mason struct btrfs_path *path, int data_size, 413744871b1bSChris Mason int extend) 413844871b1bSChris Mason { 41395d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 414044871b1bSChris Mason struct extent_buffer *l; 414144871b1bSChris Mason u32 nritems; 414244871b1bSChris Mason int mid; 414344871b1bSChris Mason int slot; 414444871b1bSChris Mason struct extent_buffer *right; 4145b7a0365eSDaniel Dressler struct btrfs_fs_info *fs_info = root->fs_info; 414644871b1bSChris Mason int ret = 0; 414744871b1bSChris Mason int wret; 41485d4f98a2SYan Zheng int split; 414944871b1bSChris Mason int num_doubles = 0; 415099d8f83cSChris Mason int tried_avoid_double = 0; 415144871b1bSChris Mason 4152a5719521SYan, Zheng l = path->nodes[0]; 4153a5719521SYan, Zheng slot = path->slots[0]; 4154a5719521SYan, Zheng if (extend && data_size + btrfs_item_size_nr(l, slot) + 41550b246afaSJeff Mahoney sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) 4156a5719521SYan, Zheng return -EOVERFLOW; 4157a5719521SYan, Zheng 415844871b1bSChris Mason /* first try to make some room by pushing left and right */ 415933157e05SLiu Bo if (data_size && path->nodes[1]) { 41605a4267caSFilipe David Borba Manana int space_needed = data_size; 41615a4267caSFilipe David Borba Manana 41625a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(l)) 4163e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 41645a4267caSFilipe David Borba Manana 41655a4267caSFilipe David Borba Manana wret = push_leaf_right(trans, root, path, space_needed, 41665a4267caSFilipe David Borba Manana space_needed, 0, 0); 416744871b1bSChris Mason if (wret < 0) 416844871b1bSChris Mason return wret; 416944871b1bSChris Mason if (wret) { 4170263d3995SFilipe Manana space_needed = data_size; 4171263d3995SFilipe Manana if (slot > 0) 4172e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 41735a4267caSFilipe David Borba Manana wret = push_leaf_left(trans, root, path, space_needed, 41745a4267caSFilipe David Borba Manana space_needed, 0, (u32)-1); 417544871b1bSChris Mason if (wret < 0) 417644871b1bSChris Mason return wret; 417744871b1bSChris Mason } 417844871b1bSChris Mason l = path->nodes[0]; 417944871b1bSChris Mason 418044871b1bSChris Mason /* did the pushes work? */ 4181e902baacSDavid Sterba if (btrfs_leaf_free_space(l) >= data_size) 418244871b1bSChris Mason return 0; 418344871b1bSChris Mason } 418444871b1bSChris Mason 418544871b1bSChris Mason if (!path->nodes[1]) { 4186fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, 1); 418744871b1bSChris Mason if (ret) 418844871b1bSChris Mason return ret; 418944871b1bSChris Mason } 419044871b1bSChris Mason again: 41915d4f98a2SYan Zheng split = 1; 419244871b1bSChris Mason l = path->nodes[0]; 419344871b1bSChris Mason slot = path->slots[0]; 419444871b1bSChris Mason nritems = btrfs_header_nritems(l); 419544871b1bSChris Mason mid = (nritems + 1) / 2; 419644871b1bSChris Mason 41975d4f98a2SYan Zheng if (mid <= slot) { 41985d4f98a2SYan Zheng if (nritems == 1 || 41995d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + data_size > 42000b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 42015d4f98a2SYan Zheng if (slot >= nritems) { 42025d4f98a2SYan Zheng split = 0; 42035d4f98a2SYan Zheng } else { 42045d4f98a2SYan Zheng mid = slot; 42055d4f98a2SYan Zheng if (mid != nritems && 42065d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 42070b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 420899d8f83cSChris Mason if (data_size && !tried_avoid_double) 420999d8f83cSChris Mason goto push_for_double; 42105d4f98a2SYan Zheng split = 2; 42115d4f98a2SYan Zheng } 42125d4f98a2SYan Zheng } 42135d4f98a2SYan Zheng } 42145d4f98a2SYan Zheng } else { 42155d4f98a2SYan Zheng if (leaf_space_used(l, 0, mid) + data_size > 42160b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 42175d4f98a2SYan Zheng if (!extend && data_size && slot == 0) { 42185d4f98a2SYan Zheng split = 0; 42195d4f98a2SYan Zheng } else if ((extend || !data_size) && slot == 0) { 42205d4f98a2SYan Zheng mid = 1; 42215d4f98a2SYan Zheng } else { 42225d4f98a2SYan Zheng mid = slot; 42235d4f98a2SYan Zheng if (mid != nritems && 42245d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 42250b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 422699d8f83cSChris Mason if (data_size && !tried_avoid_double) 422799d8f83cSChris Mason goto push_for_double; 42285d4f98a2SYan Zheng split = 2; 42295d4f98a2SYan Zheng } 42305d4f98a2SYan Zheng } 42315d4f98a2SYan Zheng } 42325d4f98a2SYan Zheng } 42335d4f98a2SYan Zheng 42345d4f98a2SYan Zheng if (split == 0) 42355d4f98a2SYan Zheng btrfs_cpu_key_to_disk(&disk_key, ins_key); 42365d4f98a2SYan Zheng else 42375d4f98a2SYan Zheng btrfs_item_key(l, &disk_key, mid); 42385d4f98a2SYan Zheng 4239ca9d473aSJosef Bacik /* 4240ca9d473aSJosef Bacik * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double 4241ca9d473aSJosef Bacik * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES 4242ca9d473aSJosef Bacik * subclasses, which is 8 at the time of this patch, and we've maxed it 4243ca9d473aSJosef Bacik * out. In the future we could add a 4244ca9d473aSJosef Bacik * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just 4245ca9d473aSJosef Bacik * use BTRFS_NESTING_NEW_ROOT. 4246ca9d473aSJosef Bacik */ 4247a6279470SFilipe Manana right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0, 4248ca9d473aSJosef Bacik l->start, 0, num_doubles ? 4249ca9d473aSJosef Bacik BTRFS_NESTING_NEW_ROOT : 4250ca9d473aSJosef Bacik BTRFS_NESTING_SPLIT); 4251f0486c68SYan, Zheng if (IS_ERR(right)) 425244871b1bSChris Mason return PTR_ERR(right); 4253f0486c68SYan, Zheng 42540b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 425544871b1bSChris Mason 42565d4f98a2SYan Zheng if (split == 0) { 425744871b1bSChris Mason if (mid <= slot) { 425844871b1bSChris Mason btrfs_set_header_nritems(right, 0); 42596ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, 42602ff7e61eSJeff Mahoney right->start, path->slots[1] + 1, 1); 426144871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 426244871b1bSChris Mason free_extent_buffer(path->nodes[0]); 426344871b1bSChris Mason path->nodes[0] = right; 426444871b1bSChris Mason path->slots[0] = 0; 426544871b1bSChris Mason path->slots[1] += 1; 426644871b1bSChris Mason } else { 426744871b1bSChris Mason btrfs_set_header_nritems(right, 0); 42686ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, 42692ff7e61eSJeff Mahoney right->start, path->slots[1], 1); 427044871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 427144871b1bSChris Mason free_extent_buffer(path->nodes[0]); 427244871b1bSChris Mason path->nodes[0] = right; 427344871b1bSChris Mason path->slots[0] = 0; 4274143bede5SJeff Mahoney if (path->slots[1] == 0) 4275b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 42765d4f98a2SYan Zheng } 4277196e0249SLiu Bo /* 4278196e0249SLiu Bo * We create a new leaf 'right' for the required ins_len and 4279196e0249SLiu Bo * we'll do btrfs_mark_buffer_dirty() on this leaf after copying 4280196e0249SLiu Bo * the content of ins_len to 'right'. 4281196e0249SLiu Bo */ 428244871b1bSChris Mason return ret; 428344871b1bSChris Mason } 428444871b1bSChris Mason 428594f94ad9SDavid Sterba copy_for_split(trans, path, l, right, slot, mid, nritems); 428644871b1bSChris Mason 42875d4f98a2SYan Zheng if (split == 2) { 4288cc0c5538SChris Mason BUG_ON(num_doubles != 0); 4289cc0c5538SChris Mason num_doubles++; 4290cc0c5538SChris Mason goto again; 42913326d1b0SChris Mason } 429244871b1bSChris Mason 4293143bede5SJeff Mahoney return 0; 429499d8f83cSChris Mason 429599d8f83cSChris Mason push_for_double: 429699d8f83cSChris Mason push_for_double_split(trans, root, path, data_size); 429799d8f83cSChris Mason tried_avoid_double = 1; 4298e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 429999d8f83cSChris Mason return 0; 430099d8f83cSChris Mason goto again; 4301be0e5c09SChris Mason } 4302be0e5c09SChris Mason 4303ad48fd75SYan, Zheng static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, 4304ad48fd75SYan, Zheng struct btrfs_root *root, 4305ad48fd75SYan, Zheng struct btrfs_path *path, int ins_len) 4306ad48fd75SYan, Zheng { 4307ad48fd75SYan, Zheng struct btrfs_key key; 4308ad48fd75SYan, Zheng struct extent_buffer *leaf; 4309ad48fd75SYan, Zheng struct btrfs_file_extent_item *fi; 4310ad48fd75SYan, Zheng u64 extent_len = 0; 4311ad48fd75SYan, Zheng u32 item_size; 4312ad48fd75SYan, Zheng int ret; 4313ad48fd75SYan, Zheng 4314ad48fd75SYan, Zheng leaf = path->nodes[0]; 4315ad48fd75SYan, Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4316ad48fd75SYan, Zheng 4317ad48fd75SYan, Zheng BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && 4318ad48fd75SYan, Zheng key.type != BTRFS_EXTENT_CSUM_KEY); 4319ad48fd75SYan, Zheng 4320e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) >= ins_len) 4321ad48fd75SYan, Zheng return 0; 4322ad48fd75SYan, Zheng 4323ad48fd75SYan, Zheng item_size = btrfs_item_size_nr(leaf, path->slots[0]); 4324ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 4325ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 4326ad48fd75SYan, Zheng struct btrfs_file_extent_item); 4327ad48fd75SYan, Zheng extent_len = btrfs_file_extent_num_bytes(leaf, fi); 4328ad48fd75SYan, Zheng } 4329b3b4aa74SDavid Sterba btrfs_release_path(path); 4330ad48fd75SYan, Zheng 4331ad48fd75SYan, Zheng path->keep_locks = 1; 4332ad48fd75SYan, Zheng path->search_for_split = 1; 4333ad48fd75SYan, Zheng ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 4334ad48fd75SYan, Zheng path->search_for_split = 0; 4335a8df6fe6SFilipe Manana if (ret > 0) 4336a8df6fe6SFilipe Manana ret = -EAGAIN; 4337ad48fd75SYan, Zheng if (ret < 0) 4338ad48fd75SYan, Zheng goto err; 4339ad48fd75SYan, Zheng 4340ad48fd75SYan, Zheng ret = -EAGAIN; 4341ad48fd75SYan, Zheng leaf = path->nodes[0]; 4342a8df6fe6SFilipe Manana /* if our item isn't there, return now */ 4343a8df6fe6SFilipe Manana if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) 4344ad48fd75SYan, Zheng goto err; 4345ad48fd75SYan, Zheng 4346109f6aefSChris Mason /* the leaf has changed, it now has room. return now */ 4347e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) 4348109f6aefSChris Mason goto err; 4349109f6aefSChris Mason 4350ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 4351ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 4352ad48fd75SYan, Zheng struct btrfs_file_extent_item); 4353ad48fd75SYan, Zheng if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) 4354ad48fd75SYan, Zheng goto err; 4355ad48fd75SYan, Zheng } 4356ad48fd75SYan, Zheng 4357ad48fd75SYan, Zheng ret = split_leaf(trans, root, &key, path, ins_len, 1); 4358f0486c68SYan, Zheng if (ret) 4359f0486c68SYan, Zheng goto err; 4360ad48fd75SYan, Zheng 4361ad48fd75SYan, Zheng path->keep_locks = 0; 4362ad48fd75SYan, Zheng btrfs_unlock_up_safe(path, 1); 4363ad48fd75SYan, Zheng return 0; 4364ad48fd75SYan, Zheng err: 4365ad48fd75SYan, Zheng path->keep_locks = 0; 4366ad48fd75SYan, Zheng return ret; 4367ad48fd75SYan, Zheng } 4368ad48fd75SYan, Zheng 436925263cd7SDavid Sterba static noinline int split_item(struct btrfs_path *path, 4370310712b2SOmar Sandoval const struct btrfs_key *new_key, 4371459931ecSChris Mason unsigned long split_offset) 4372459931ecSChris Mason { 4373459931ecSChris Mason struct extent_buffer *leaf; 4374459931ecSChris Mason struct btrfs_item *item; 4375459931ecSChris Mason struct btrfs_item *new_item; 4376459931ecSChris Mason int slot; 4377ad48fd75SYan, Zheng char *buf; 4378459931ecSChris Mason u32 nritems; 4379ad48fd75SYan, Zheng u32 item_size; 4380459931ecSChris Mason u32 orig_offset; 4381459931ecSChris Mason struct btrfs_disk_key disk_key; 4382459931ecSChris Mason 4383459931ecSChris Mason leaf = path->nodes[0]; 4384e902baacSDavid Sterba BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); 4385b9473439SChris Mason 4386dd3cc16bSRoss Kirk item = btrfs_item_nr(path->slots[0]); 4387459931ecSChris Mason orig_offset = btrfs_item_offset(leaf, item); 4388459931ecSChris Mason item_size = btrfs_item_size(leaf, item); 4389459931ecSChris Mason 4390459931ecSChris Mason buf = kmalloc(item_size, GFP_NOFS); 4391ad48fd75SYan, Zheng if (!buf) 4392ad48fd75SYan, Zheng return -ENOMEM; 4393ad48fd75SYan, Zheng 4394459931ecSChris Mason read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, 4395459931ecSChris Mason path->slots[0]), item_size); 4396ad48fd75SYan, Zheng 4397459931ecSChris Mason slot = path->slots[0] + 1; 4398459931ecSChris Mason nritems = btrfs_header_nritems(leaf); 4399459931ecSChris Mason if (slot != nritems) { 4400459931ecSChris Mason /* shift the items */ 4401459931ecSChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), 4402459931ecSChris Mason btrfs_item_nr_offset(slot), 4403459931ecSChris Mason (nritems - slot) * sizeof(struct btrfs_item)); 4404459931ecSChris Mason } 4405459931ecSChris Mason 4406459931ecSChris Mason btrfs_cpu_key_to_disk(&disk_key, new_key); 4407459931ecSChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 4408459931ecSChris Mason 4409dd3cc16bSRoss Kirk new_item = btrfs_item_nr(slot); 4410459931ecSChris Mason 4411459931ecSChris Mason btrfs_set_item_offset(leaf, new_item, orig_offset); 4412459931ecSChris Mason btrfs_set_item_size(leaf, new_item, item_size - split_offset); 4413459931ecSChris Mason 4414459931ecSChris Mason btrfs_set_item_offset(leaf, item, 4415459931ecSChris Mason orig_offset + item_size - split_offset); 4416459931ecSChris Mason btrfs_set_item_size(leaf, item, split_offset); 4417459931ecSChris Mason 4418459931ecSChris Mason btrfs_set_header_nritems(leaf, nritems + 1); 4419459931ecSChris Mason 4420459931ecSChris Mason /* write the data for the start of the original item */ 4421459931ecSChris Mason write_extent_buffer(leaf, buf, 4422459931ecSChris Mason btrfs_item_ptr_offset(leaf, path->slots[0]), 4423459931ecSChris Mason split_offset); 4424459931ecSChris Mason 4425459931ecSChris Mason /* write the data for the new item */ 4426459931ecSChris Mason write_extent_buffer(leaf, buf + split_offset, 4427459931ecSChris Mason btrfs_item_ptr_offset(leaf, slot), 4428459931ecSChris Mason item_size - split_offset); 4429459931ecSChris Mason btrfs_mark_buffer_dirty(leaf); 4430459931ecSChris Mason 4431e902baacSDavid Sterba BUG_ON(btrfs_leaf_free_space(leaf) < 0); 4432459931ecSChris Mason kfree(buf); 4433ad48fd75SYan, Zheng return 0; 4434ad48fd75SYan, Zheng } 4435ad48fd75SYan, Zheng 4436ad48fd75SYan, Zheng /* 4437ad48fd75SYan, Zheng * This function splits a single item into two items, 4438ad48fd75SYan, Zheng * giving 'new_key' to the new item and splitting the 4439ad48fd75SYan, Zheng * old one at split_offset (from the start of the item). 4440ad48fd75SYan, Zheng * 4441ad48fd75SYan, Zheng * The path may be released by this operation. After 4442ad48fd75SYan, Zheng * the split, the path is pointing to the old item. The 4443ad48fd75SYan, Zheng * new item is going to be in the same node as the old one. 4444ad48fd75SYan, Zheng * 4445ad48fd75SYan, Zheng * Note, the item being split must be smaller enough to live alone on 4446ad48fd75SYan, Zheng * a tree block with room for one extra struct btrfs_item 4447ad48fd75SYan, Zheng * 4448ad48fd75SYan, Zheng * This allows us to split the item in place, keeping a lock on the 4449ad48fd75SYan, Zheng * leaf the entire time. 4450ad48fd75SYan, Zheng */ 4451ad48fd75SYan, Zheng int btrfs_split_item(struct btrfs_trans_handle *trans, 4452ad48fd75SYan, Zheng struct btrfs_root *root, 4453ad48fd75SYan, Zheng struct btrfs_path *path, 4454310712b2SOmar Sandoval const struct btrfs_key *new_key, 4455ad48fd75SYan, Zheng unsigned long split_offset) 4456ad48fd75SYan, Zheng { 4457ad48fd75SYan, Zheng int ret; 4458ad48fd75SYan, Zheng ret = setup_leaf_for_split(trans, root, path, 4459ad48fd75SYan, Zheng sizeof(struct btrfs_item)); 4460ad48fd75SYan, Zheng if (ret) 4461459931ecSChris Mason return ret; 4462ad48fd75SYan, Zheng 446325263cd7SDavid Sterba ret = split_item(path, new_key, split_offset); 4464ad48fd75SYan, Zheng return ret; 4465ad48fd75SYan, Zheng } 4466ad48fd75SYan, Zheng 4467ad48fd75SYan, Zheng /* 4468ad48fd75SYan, Zheng * This function duplicate a item, giving 'new_key' to the new item. 4469ad48fd75SYan, Zheng * It guarantees both items live in the same tree leaf and the new item 4470ad48fd75SYan, Zheng * is contiguous with the original item. 4471ad48fd75SYan, Zheng * 4472ad48fd75SYan, Zheng * This allows us to split file extent in place, keeping a lock on the 4473ad48fd75SYan, Zheng * leaf the entire time. 4474ad48fd75SYan, Zheng */ 4475ad48fd75SYan, Zheng int btrfs_duplicate_item(struct btrfs_trans_handle *trans, 4476ad48fd75SYan, Zheng struct btrfs_root *root, 4477ad48fd75SYan, Zheng struct btrfs_path *path, 4478310712b2SOmar Sandoval const struct btrfs_key *new_key) 4479ad48fd75SYan, Zheng { 4480ad48fd75SYan, Zheng struct extent_buffer *leaf; 4481ad48fd75SYan, Zheng int ret; 4482ad48fd75SYan, Zheng u32 item_size; 4483ad48fd75SYan, Zheng 4484ad48fd75SYan, Zheng leaf = path->nodes[0]; 4485ad48fd75SYan, Zheng item_size = btrfs_item_size_nr(leaf, path->slots[0]); 4486ad48fd75SYan, Zheng ret = setup_leaf_for_split(trans, root, path, 4487ad48fd75SYan, Zheng item_size + sizeof(struct btrfs_item)); 4488ad48fd75SYan, Zheng if (ret) 4489ad48fd75SYan, Zheng return ret; 4490ad48fd75SYan, Zheng 4491ad48fd75SYan, Zheng path->slots[0]++; 4492fc0d82e1SNikolay Borisov setup_items_for_insert(root, path, new_key, &item_size, 1); 4493ad48fd75SYan, Zheng leaf = path->nodes[0]; 4494ad48fd75SYan, Zheng memcpy_extent_buffer(leaf, 4495ad48fd75SYan, Zheng btrfs_item_ptr_offset(leaf, path->slots[0]), 4496ad48fd75SYan, Zheng btrfs_item_ptr_offset(leaf, path->slots[0] - 1), 4497ad48fd75SYan, Zheng item_size); 4498ad48fd75SYan, Zheng return 0; 4499459931ecSChris Mason } 4500459931ecSChris Mason 4501459931ecSChris Mason /* 4502d352ac68SChris Mason * make the item pointed to by the path smaller. new_size indicates 4503d352ac68SChris Mason * how small to make it, and from_end tells us if we just chop bytes 4504d352ac68SChris Mason * off the end of the item or if we shift the item to chop bytes off 4505d352ac68SChris Mason * the front. 4506d352ac68SChris Mason */ 450778ac4f9eSDavid Sterba void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) 4508b18c6685SChris Mason { 4509b18c6685SChris Mason int slot; 45105f39d397SChris Mason struct extent_buffer *leaf; 45115f39d397SChris Mason struct btrfs_item *item; 4512b18c6685SChris Mason u32 nritems; 4513b18c6685SChris Mason unsigned int data_end; 4514b18c6685SChris Mason unsigned int old_data_start; 4515b18c6685SChris Mason unsigned int old_size; 4516b18c6685SChris Mason unsigned int size_diff; 4517b18c6685SChris Mason int i; 4518cfed81a0SChris Mason struct btrfs_map_token token; 4519cfed81a0SChris Mason 45205f39d397SChris Mason leaf = path->nodes[0]; 4521179e29e4SChris Mason slot = path->slots[0]; 4522179e29e4SChris Mason 4523179e29e4SChris Mason old_size = btrfs_item_size_nr(leaf, slot); 4524179e29e4SChris Mason if (old_size == new_size) 4525143bede5SJeff Mahoney return; 4526b18c6685SChris Mason 45275f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 45288f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 4529b18c6685SChris Mason 45305f39d397SChris Mason old_data_start = btrfs_item_offset_nr(leaf, slot); 4531179e29e4SChris Mason 4532b18c6685SChris Mason size_diff = old_size - new_size; 4533b18c6685SChris Mason 4534b18c6685SChris Mason BUG_ON(slot < 0); 4535b18c6685SChris Mason BUG_ON(slot >= nritems); 4536b18c6685SChris Mason 4537b18c6685SChris Mason /* 4538b18c6685SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 4539b18c6685SChris Mason */ 4540b18c6685SChris Mason /* first correct the data pointers */ 4541c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 4542b18c6685SChris Mason for (i = slot; i < nritems; i++) { 45435f39d397SChris Mason u32 ioff; 4544dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 4545db94535dSChris Mason 4546cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4547cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, ioff + size_diff); 4548b18c6685SChris Mason } 4549db94535dSChris Mason 4550b18c6685SChris Mason /* shift the data */ 4551179e29e4SChris Mason if (from_end) { 45523d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 45533d9ec8c4SNikolay Borisov data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + 4554b18c6685SChris Mason data_end, old_data_start + new_size - data_end); 4555179e29e4SChris Mason } else { 4556179e29e4SChris Mason struct btrfs_disk_key disk_key; 4557179e29e4SChris Mason u64 offset; 4558179e29e4SChris Mason 4559179e29e4SChris Mason btrfs_item_key(leaf, &disk_key, slot); 4560179e29e4SChris Mason 4561179e29e4SChris Mason if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { 4562179e29e4SChris Mason unsigned long ptr; 4563179e29e4SChris Mason struct btrfs_file_extent_item *fi; 4564179e29e4SChris Mason 4565179e29e4SChris Mason fi = btrfs_item_ptr(leaf, slot, 4566179e29e4SChris Mason struct btrfs_file_extent_item); 4567179e29e4SChris Mason fi = (struct btrfs_file_extent_item *)( 4568179e29e4SChris Mason (unsigned long)fi - size_diff); 4569179e29e4SChris Mason 4570179e29e4SChris Mason if (btrfs_file_extent_type(leaf, fi) == 4571179e29e4SChris Mason BTRFS_FILE_EXTENT_INLINE) { 4572179e29e4SChris Mason ptr = btrfs_item_ptr_offset(leaf, slot); 4573179e29e4SChris Mason memmove_extent_buffer(leaf, ptr, 4574179e29e4SChris Mason (unsigned long)fi, 45757ec20afbSDavid Sterba BTRFS_FILE_EXTENT_INLINE_DATA_START); 4576179e29e4SChris Mason } 4577179e29e4SChris Mason } 4578179e29e4SChris Mason 45793d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 45803d9ec8c4SNikolay Borisov data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + 4581179e29e4SChris Mason data_end, old_data_start - data_end); 4582179e29e4SChris Mason 4583179e29e4SChris Mason offset = btrfs_disk_key_offset(&disk_key); 4584179e29e4SChris Mason btrfs_set_disk_key_offset(&disk_key, offset + size_diff); 4585179e29e4SChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 4586179e29e4SChris Mason if (slot == 0) 4587b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 4588179e29e4SChris Mason } 45895f39d397SChris Mason 4590dd3cc16bSRoss Kirk item = btrfs_item_nr(slot); 45915f39d397SChris Mason btrfs_set_item_size(leaf, item, new_size); 45925f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 4593b18c6685SChris Mason 4594e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4595a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4596b18c6685SChris Mason BUG(); 45975f39d397SChris Mason } 4598b18c6685SChris Mason } 4599b18c6685SChris Mason 4600d352ac68SChris Mason /* 46018f69dbd2SStefan Behrens * make the item pointed to by the path bigger, data_size is the added size. 4602d352ac68SChris Mason */ 4603c71dd880SDavid Sterba void btrfs_extend_item(struct btrfs_path *path, u32 data_size) 46046567e837SChris Mason { 46056567e837SChris Mason int slot; 46065f39d397SChris Mason struct extent_buffer *leaf; 46075f39d397SChris Mason struct btrfs_item *item; 46086567e837SChris Mason u32 nritems; 46096567e837SChris Mason unsigned int data_end; 46106567e837SChris Mason unsigned int old_data; 46116567e837SChris Mason unsigned int old_size; 46126567e837SChris Mason int i; 4613cfed81a0SChris Mason struct btrfs_map_token token; 4614cfed81a0SChris Mason 46155f39d397SChris Mason leaf = path->nodes[0]; 46166567e837SChris Mason 46175f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 46188f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 46196567e837SChris Mason 4620e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < data_size) { 4621a4f78750SDavid Sterba btrfs_print_leaf(leaf); 46226567e837SChris Mason BUG(); 46235f39d397SChris Mason } 46246567e837SChris Mason slot = path->slots[0]; 46255f39d397SChris Mason old_data = btrfs_item_end_nr(leaf, slot); 46266567e837SChris Mason 46276567e837SChris Mason BUG_ON(slot < 0); 46283326d1b0SChris Mason if (slot >= nritems) { 4629a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4630c71dd880SDavid Sterba btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", 4631d397712bSChris Mason slot, nritems); 4632290342f6SArnd Bergmann BUG(); 46333326d1b0SChris Mason } 46346567e837SChris Mason 46356567e837SChris Mason /* 46366567e837SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 46376567e837SChris Mason */ 46386567e837SChris Mason /* first correct the data pointers */ 4639c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 46406567e837SChris Mason for (i = slot; i < nritems; i++) { 46415f39d397SChris Mason u32 ioff; 4642dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 4643db94535dSChris Mason 4644cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4645cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, ioff - data_size); 46466567e837SChris Mason } 46475f39d397SChris Mason 46486567e837SChris Mason /* shift the data */ 46493d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 46503d9ec8c4SNikolay Borisov data_end - data_size, BTRFS_LEAF_DATA_OFFSET + 46516567e837SChris Mason data_end, old_data - data_end); 46525f39d397SChris Mason 46536567e837SChris Mason data_end = old_data; 46545f39d397SChris Mason old_size = btrfs_item_size_nr(leaf, slot); 4655dd3cc16bSRoss Kirk item = btrfs_item_nr(slot); 46565f39d397SChris Mason btrfs_set_item_size(leaf, item, old_size + data_size); 46575f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 46586567e837SChris Mason 4659e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4660a4f78750SDavid Sterba btrfs_print_leaf(leaf); 46616567e837SChris Mason BUG(); 46625f39d397SChris Mason } 46636567e837SChris Mason } 46646567e837SChris Mason 4665da9ffb24SNikolay Borisov /** 4666da9ffb24SNikolay Borisov * setup_items_for_insert - Helper called before inserting one or more items 4667da9ffb24SNikolay Borisov * to a leaf. Main purpose is to save stack depth by doing the bulk of the work 4668da9ffb24SNikolay Borisov * in a function that doesn't call btrfs_search_slot 4669da9ffb24SNikolay Borisov * 4670da9ffb24SNikolay Borisov * @root: root we are inserting items to 4671da9ffb24SNikolay Borisov * @path: points to the leaf/slot where we are going to insert new items 4672da9ffb24SNikolay Borisov * @cpu_key: array of keys for items to be inserted 4673da9ffb24SNikolay Borisov * @data_size: size of the body of each item we are going to insert 4674da9ffb24SNikolay Borisov * @nr: size of @cpu_key/@data_size arrays 467574123bd7SChris Mason */ 4676afe5fea7STsutomu Itoh void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, 4677310712b2SOmar Sandoval const struct btrfs_key *cpu_key, u32 *data_size, 4678fc0d82e1SNikolay Borisov int nr) 4679be0e5c09SChris Mason { 46800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46815f39d397SChris Mason struct btrfs_item *item; 46829c58309dSChris Mason int i; 46837518a238SChris Mason u32 nritems; 4684be0e5c09SChris Mason unsigned int data_end; 4685e2fa7227SChris Mason struct btrfs_disk_key disk_key; 468644871b1bSChris Mason struct extent_buffer *leaf; 468744871b1bSChris Mason int slot; 4688cfed81a0SChris Mason struct btrfs_map_token token; 4689fc0d82e1SNikolay Borisov u32 total_size; 4690fc0d82e1SNikolay Borisov u32 total_data = 0; 4691fc0d82e1SNikolay Borisov 4692fc0d82e1SNikolay Borisov for (i = 0; i < nr; i++) 4693fc0d82e1SNikolay Borisov total_data += data_size[i]; 4694fc0d82e1SNikolay Borisov total_size = total_data + (nr * sizeof(struct btrfs_item)); 4695cfed81a0SChris Mason 469624cdc847SFilipe Manana if (path->slots[0] == 0) { 469724cdc847SFilipe Manana btrfs_cpu_key_to_disk(&disk_key, cpu_key); 4698b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 469924cdc847SFilipe Manana } 470024cdc847SFilipe Manana btrfs_unlock_up_safe(path, 1); 470124cdc847SFilipe Manana 47025f39d397SChris Mason leaf = path->nodes[0]; 470344871b1bSChris Mason slot = path->slots[0]; 470474123bd7SChris Mason 47055f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 47068f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 4707eb60ceacSChris Mason 4708e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < total_size) { 4709a4f78750SDavid Sterba btrfs_print_leaf(leaf); 47100b246afaSJeff Mahoney btrfs_crit(fs_info, "not enough freespace need %u have %d", 4711e902baacSDavid Sterba total_size, btrfs_leaf_free_space(leaf)); 4712be0e5c09SChris Mason BUG(); 4713d4dbff95SChris Mason } 47145f39d397SChris Mason 4715c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 4716be0e5c09SChris Mason if (slot != nritems) { 47175f39d397SChris Mason unsigned int old_data = btrfs_item_end_nr(leaf, slot); 4718be0e5c09SChris Mason 47195f39d397SChris Mason if (old_data < data_end) { 4720a4f78750SDavid Sterba btrfs_print_leaf(leaf); 47217269ddd2SNikolay Borisov btrfs_crit(fs_info, 47227269ddd2SNikolay Borisov "item at slot %d with data offset %u beyond data end of leaf %u", 47235f39d397SChris Mason slot, old_data, data_end); 4724290342f6SArnd Bergmann BUG(); 47255f39d397SChris Mason } 4726be0e5c09SChris Mason /* 4727be0e5c09SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 4728be0e5c09SChris Mason */ 4729be0e5c09SChris Mason /* first correct the data pointers */ 47300783fcfcSChris Mason for (i = slot; i < nritems; i++) { 47315f39d397SChris Mason u32 ioff; 4732db94535dSChris Mason 4733dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 4734cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4735cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, 4736cc4c13d5SDavid Sterba ioff - total_data); 47370783fcfcSChris Mason } 4738be0e5c09SChris Mason /* shift the items */ 47399c58309dSChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), 47405f39d397SChris Mason btrfs_item_nr_offset(slot), 47410783fcfcSChris Mason (nritems - slot) * sizeof(struct btrfs_item)); 4742be0e5c09SChris Mason 4743be0e5c09SChris Mason /* shift the data */ 47443d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 47453d9ec8c4SNikolay Borisov data_end - total_data, BTRFS_LEAF_DATA_OFFSET + 4746be0e5c09SChris Mason data_end, old_data - data_end); 4747be0e5c09SChris Mason data_end = old_data; 4748be0e5c09SChris Mason } 47495f39d397SChris Mason 475062e2749eSChris Mason /* setup the item for the new data */ 47519c58309dSChris Mason for (i = 0; i < nr; i++) { 47529c58309dSChris Mason btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); 47539c58309dSChris Mason btrfs_set_item_key(leaf, &disk_key, slot + i); 4754dd3cc16bSRoss Kirk item = btrfs_item_nr(slot + i); 47559c58309dSChris Mason data_end -= data_size[i]; 4756fc0716c2SNikolay Borisov btrfs_set_token_item_offset(&token, item, data_end); 4757cc4c13d5SDavid Sterba btrfs_set_token_item_size(&token, item, data_size[i]); 47589c58309dSChris Mason } 475944871b1bSChris Mason 47609c58309dSChris Mason btrfs_set_header_nritems(leaf, nritems + nr); 4761b9473439SChris Mason btrfs_mark_buffer_dirty(leaf); 4762aa5d6bedSChris Mason 4763e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4764a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4765be0e5c09SChris Mason BUG(); 47665f39d397SChris Mason } 476744871b1bSChris Mason } 476844871b1bSChris Mason 476944871b1bSChris Mason /* 477044871b1bSChris Mason * Given a key and some data, insert items into the tree. 477144871b1bSChris Mason * This does all the path init required, making room in the tree if needed. 477244871b1bSChris Mason */ 477344871b1bSChris Mason int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, 477444871b1bSChris Mason struct btrfs_root *root, 477544871b1bSChris Mason struct btrfs_path *path, 4776310712b2SOmar Sandoval const struct btrfs_key *cpu_key, u32 *data_size, 477744871b1bSChris Mason int nr) 477844871b1bSChris Mason { 477944871b1bSChris Mason int ret = 0; 478044871b1bSChris Mason int slot; 478144871b1bSChris Mason int i; 478244871b1bSChris Mason u32 total_size = 0; 478344871b1bSChris Mason u32 total_data = 0; 478444871b1bSChris Mason 478544871b1bSChris Mason for (i = 0; i < nr; i++) 478644871b1bSChris Mason total_data += data_size[i]; 478744871b1bSChris Mason 478844871b1bSChris Mason total_size = total_data + (nr * sizeof(struct btrfs_item)); 478944871b1bSChris Mason ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); 479044871b1bSChris Mason if (ret == 0) 479144871b1bSChris Mason return -EEXIST; 479244871b1bSChris Mason if (ret < 0) 4793143bede5SJeff Mahoney return ret; 479444871b1bSChris Mason 479544871b1bSChris Mason slot = path->slots[0]; 479644871b1bSChris Mason BUG_ON(slot < 0); 479744871b1bSChris Mason 4798fc0d82e1SNikolay Borisov setup_items_for_insert(root, path, cpu_key, data_size, nr); 4799143bede5SJeff Mahoney return 0; 480062e2749eSChris Mason } 480162e2749eSChris Mason 480262e2749eSChris Mason /* 480362e2749eSChris Mason * Given a key and some data, insert an item into the tree. 480462e2749eSChris Mason * This does all the path init required, making room in the tree if needed. 480562e2749eSChris Mason */ 4806310712b2SOmar Sandoval int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, 4807310712b2SOmar Sandoval const struct btrfs_key *cpu_key, void *data, 4808310712b2SOmar Sandoval u32 data_size) 480962e2749eSChris Mason { 481062e2749eSChris Mason int ret = 0; 48112c90e5d6SChris Mason struct btrfs_path *path; 48125f39d397SChris Mason struct extent_buffer *leaf; 48135f39d397SChris Mason unsigned long ptr; 481462e2749eSChris Mason 48152c90e5d6SChris Mason path = btrfs_alloc_path(); 4816db5b493aSTsutomu Itoh if (!path) 4817db5b493aSTsutomu Itoh return -ENOMEM; 48182c90e5d6SChris Mason ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); 481962e2749eSChris Mason if (!ret) { 48205f39d397SChris Mason leaf = path->nodes[0]; 48215f39d397SChris Mason ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); 48225f39d397SChris Mason write_extent_buffer(leaf, data, ptr, data_size); 48235f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 482462e2749eSChris Mason } 48252c90e5d6SChris Mason btrfs_free_path(path); 4826aa5d6bedSChris Mason return ret; 4827be0e5c09SChris Mason } 4828be0e5c09SChris Mason 482974123bd7SChris Mason /* 48305de08d7dSChris Mason * delete the pointer from a given node. 483174123bd7SChris Mason * 4832d352ac68SChris Mason * the tree should have been previously balanced so the deletion does not 4833d352ac68SChris Mason * empty a node. 483474123bd7SChris Mason */ 4835afe5fea7STsutomu Itoh static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, 4836afe5fea7STsutomu Itoh int level, int slot) 4837be0e5c09SChris Mason { 48385f39d397SChris Mason struct extent_buffer *parent = path->nodes[level]; 48397518a238SChris Mason u32 nritems; 4840f3ea38daSJan Schmidt int ret; 4841be0e5c09SChris Mason 48425f39d397SChris Mason nritems = btrfs_header_nritems(parent); 4843be0e5c09SChris Mason if (slot != nritems - 1) { 4844bf1d3425SDavid Sterba if (level) { 4845bf1d3425SDavid Sterba ret = tree_mod_log_insert_move(parent, slot, slot + 1, 4846a446a979SDavid Sterba nritems - slot - 1); 4847bf1d3425SDavid Sterba BUG_ON(ret < 0); 4848bf1d3425SDavid Sterba } 48495f39d397SChris Mason memmove_extent_buffer(parent, 48505f39d397SChris Mason btrfs_node_key_ptr_offset(slot), 48515f39d397SChris Mason btrfs_node_key_ptr_offset(slot + 1), 4852d6025579SChris Mason sizeof(struct btrfs_key_ptr) * 4853d6025579SChris Mason (nritems - slot - 1)); 485457ba86c0SChris Mason } else if (level) { 4855e09c2efeSDavid Sterba ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, 4856e09c2efeSDavid Sterba GFP_NOFS); 485757ba86c0SChris Mason BUG_ON(ret < 0); 4858be0e5c09SChris Mason } 4859f3ea38daSJan Schmidt 48607518a238SChris Mason nritems--; 48615f39d397SChris Mason btrfs_set_header_nritems(parent, nritems); 48627518a238SChris Mason if (nritems == 0 && parent == root->node) { 48635f39d397SChris Mason BUG_ON(btrfs_header_level(root->node) != 1); 4864eb60ceacSChris Mason /* just turn the root into a leaf and break */ 48655f39d397SChris Mason btrfs_set_header_level(root->node, 0); 4866bb803951SChris Mason } else if (slot == 0) { 48675f39d397SChris Mason struct btrfs_disk_key disk_key; 48685f39d397SChris Mason 48695f39d397SChris Mason btrfs_node_key(parent, &disk_key, 0); 4870b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, level + 1); 4871be0e5c09SChris Mason } 4872d6025579SChris Mason btrfs_mark_buffer_dirty(parent); 4873be0e5c09SChris Mason } 4874be0e5c09SChris Mason 487574123bd7SChris Mason /* 4876323ac95bSChris Mason * a helper function to delete the leaf pointed to by path->slots[1] and 48775d4f98a2SYan Zheng * path->nodes[1]. 4878323ac95bSChris Mason * 4879323ac95bSChris Mason * This deletes the pointer in path->nodes[1] and frees the leaf 4880323ac95bSChris Mason * block extent. zero is returned if it all worked out, < 0 otherwise. 4881323ac95bSChris Mason * 4882323ac95bSChris Mason * The path must have already been setup for deleting the leaf, including 4883323ac95bSChris Mason * all the proper balancing. path->nodes[1] must be locked. 4884323ac95bSChris Mason */ 4885143bede5SJeff Mahoney static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, 4886323ac95bSChris Mason struct btrfs_root *root, 48875d4f98a2SYan Zheng struct btrfs_path *path, 48885d4f98a2SYan Zheng struct extent_buffer *leaf) 4889323ac95bSChris Mason { 48905d4f98a2SYan Zheng WARN_ON(btrfs_header_generation(leaf) != trans->transid); 4891afe5fea7STsutomu Itoh del_ptr(root, path, 1, path->slots[1]); 4892323ac95bSChris Mason 48934d081c41SChris Mason /* 48944d081c41SChris Mason * btrfs_free_extent is expensive, we want to make sure we 48954d081c41SChris Mason * aren't holding any locks when we call it 48964d081c41SChris Mason */ 48974d081c41SChris Mason btrfs_unlock_up_safe(path, 0); 48984d081c41SChris Mason 4899f0486c68SYan, Zheng root_sub_used(root, leaf->len); 4900f0486c68SYan, Zheng 490167439dadSDavid Sterba atomic_inc(&leaf->refs); 49025581a51aSJan Schmidt btrfs_free_tree_block(trans, root, leaf, 0, 1); 49033083ee2eSJosef Bacik free_extent_buffer_stale(leaf); 4904323ac95bSChris Mason } 4905323ac95bSChris Mason /* 490674123bd7SChris Mason * delete the item at the leaf level in path. If that empties 490774123bd7SChris Mason * the leaf, remove it from the tree 490874123bd7SChris Mason */ 490985e21bacSChris Mason int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, 491085e21bacSChris Mason struct btrfs_path *path, int slot, int nr) 4911be0e5c09SChris Mason { 49120b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 49135f39d397SChris Mason struct extent_buffer *leaf; 49145f39d397SChris Mason struct btrfs_item *item; 4915ce0eac2aSAlexandru Moise u32 last_off; 4916ce0eac2aSAlexandru Moise u32 dsize = 0; 4917aa5d6bedSChris Mason int ret = 0; 4918aa5d6bedSChris Mason int wret; 491985e21bacSChris Mason int i; 49207518a238SChris Mason u32 nritems; 4921be0e5c09SChris Mason 49225f39d397SChris Mason leaf = path->nodes[0]; 492385e21bacSChris Mason last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); 492485e21bacSChris Mason 492585e21bacSChris Mason for (i = 0; i < nr; i++) 492685e21bacSChris Mason dsize += btrfs_item_size_nr(leaf, slot + i); 492785e21bacSChris Mason 49285f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 4929be0e5c09SChris Mason 493085e21bacSChris Mason if (slot + nr != nritems) { 49318f881e8cSDavid Sterba int data_end = leaf_data_end(leaf); 4932c82f823cSDavid Sterba struct btrfs_map_token token; 49335f39d397SChris Mason 49343d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 4935d6025579SChris Mason data_end + dsize, 49363d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end, 493785e21bacSChris Mason last_off - data_end); 49385f39d397SChris Mason 4939c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 494085e21bacSChris Mason for (i = slot + nr; i < nritems; i++) { 49415f39d397SChris Mason u32 ioff; 4942db94535dSChris Mason 4943dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 4944cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4945cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, ioff + dsize); 49460783fcfcSChris Mason } 4947db94535dSChris Mason 49485f39d397SChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), 494985e21bacSChris Mason btrfs_item_nr_offset(slot + nr), 49500783fcfcSChris Mason sizeof(struct btrfs_item) * 495185e21bacSChris Mason (nritems - slot - nr)); 4952be0e5c09SChris Mason } 495385e21bacSChris Mason btrfs_set_header_nritems(leaf, nritems - nr); 495485e21bacSChris Mason nritems -= nr; 49555f39d397SChris Mason 495674123bd7SChris Mason /* delete the leaf if we've emptied it */ 49577518a238SChris Mason if (nritems == 0) { 49585f39d397SChris Mason if (leaf == root->node) { 49595f39d397SChris Mason btrfs_set_header_level(leaf, 0); 49609a8dd150SChris Mason } else { 49616a884d7dSDavid Sterba btrfs_clean_tree_block(leaf); 4962143bede5SJeff Mahoney btrfs_del_leaf(trans, root, path, leaf); 49639a8dd150SChris Mason } 4964be0e5c09SChris Mason } else { 49657518a238SChris Mason int used = leaf_space_used(leaf, 0, nritems); 4966aa5d6bedSChris Mason if (slot == 0) { 49675f39d397SChris Mason struct btrfs_disk_key disk_key; 49685f39d397SChris Mason 49695f39d397SChris Mason btrfs_item_key(leaf, &disk_key, 0); 4970b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 4971aa5d6bedSChris Mason } 4972aa5d6bedSChris Mason 497374123bd7SChris Mason /* delete the leaf if it is mostly empty */ 49740b246afaSJeff Mahoney if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { 4975be0e5c09SChris Mason /* push_leaf_left fixes the path. 4976be0e5c09SChris Mason * make sure the path still points to our leaf 4977be0e5c09SChris Mason * for possible call to del_ptr below 4978be0e5c09SChris Mason */ 49794920c9acSChris Mason slot = path->slots[1]; 498067439dadSDavid Sterba atomic_inc(&leaf->refs); 49815f39d397SChris Mason 498299d8f83cSChris Mason wret = push_leaf_left(trans, root, path, 1, 1, 498399d8f83cSChris Mason 1, (u32)-1); 498454aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4985aa5d6bedSChris Mason ret = wret; 49865f39d397SChris Mason 49875f39d397SChris Mason if (path->nodes[0] == leaf && 49885f39d397SChris Mason btrfs_header_nritems(leaf)) { 498999d8f83cSChris Mason wret = push_leaf_right(trans, root, path, 1, 499099d8f83cSChris Mason 1, 1, 0); 499154aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4992aa5d6bedSChris Mason ret = wret; 4993aa5d6bedSChris Mason } 49945f39d397SChris Mason 49955f39d397SChris Mason if (btrfs_header_nritems(leaf) == 0) { 4996323ac95bSChris Mason path->slots[1] = slot; 4997143bede5SJeff Mahoney btrfs_del_leaf(trans, root, path, leaf); 49985f39d397SChris Mason free_extent_buffer(leaf); 4999143bede5SJeff Mahoney ret = 0; 50005de08d7dSChris Mason } else { 5001925baeddSChris Mason /* if we're still in the path, make sure 5002925baeddSChris Mason * we're dirty. Otherwise, one of the 5003925baeddSChris Mason * push_leaf functions must have already 5004925baeddSChris Mason * dirtied this buffer 5005925baeddSChris Mason */ 5006925baeddSChris Mason if (path->nodes[0] == leaf) 50075f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 50085f39d397SChris Mason free_extent_buffer(leaf); 5009be0e5c09SChris Mason } 5010d5719762SChris Mason } else { 50115f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 5012be0e5c09SChris Mason } 50139a8dd150SChris Mason } 5014aa5d6bedSChris Mason return ret; 50159a8dd150SChris Mason } 50169a8dd150SChris Mason 501797571fd0SChris Mason /* 5018925baeddSChris Mason * search the tree again to find a leaf with lesser keys 50197bb86316SChris Mason * returns 0 if it found something or 1 if there are no lesser leaves. 50207bb86316SChris Mason * returns < 0 on io errors. 5021d352ac68SChris Mason * 5022d352ac68SChris Mason * This may release the path, and so you may lose any locks held at the 5023d352ac68SChris Mason * time you call it. 50247bb86316SChris Mason */ 502516e7549fSJosef Bacik int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) 50267bb86316SChris Mason { 5027925baeddSChris Mason struct btrfs_key key; 5028925baeddSChris Mason struct btrfs_disk_key found_key; 5029925baeddSChris Mason int ret; 50307bb86316SChris Mason 5031925baeddSChris Mason btrfs_item_key_to_cpu(path->nodes[0], &key, 0); 5032925baeddSChris Mason 5033e8b0d724SFilipe David Borba Manana if (key.offset > 0) { 5034925baeddSChris Mason key.offset--; 5035e8b0d724SFilipe David Borba Manana } else if (key.type > 0) { 5036925baeddSChris Mason key.type--; 5037e8b0d724SFilipe David Borba Manana key.offset = (u64)-1; 5038e8b0d724SFilipe David Borba Manana } else if (key.objectid > 0) { 5039925baeddSChris Mason key.objectid--; 5040e8b0d724SFilipe David Borba Manana key.type = (u8)-1; 5041e8b0d724SFilipe David Borba Manana key.offset = (u64)-1; 5042e8b0d724SFilipe David Borba Manana } else { 50437bb86316SChris Mason return 1; 5044e8b0d724SFilipe David Borba Manana } 50457bb86316SChris Mason 5046b3b4aa74SDavid Sterba btrfs_release_path(path); 5047925baeddSChris Mason ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 5048925baeddSChris Mason if (ret < 0) 5049925baeddSChris Mason return ret; 5050925baeddSChris Mason btrfs_item_key(path->nodes[0], &found_key, 0); 5051925baeddSChris Mason ret = comp_keys(&found_key, &key); 5052337c6f68SFilipe Manana /* 5053337c6f68SFilipe Manana * We might have had an item with the previous key in the tree right 5054337c6f68SFilipe Manana * before we released our path. And after we released our path, that 5055337c6f68SFilipe Manana * item might have been pushed to the first slot (0) of the leaf we 5056337c6f68SFilipe Manana * were holding due to a tree balance. Alternatively, an item with the 5057337c6f68SFilipe Manana * previous key can exist as the only element of a leaf (big fat item). 5058337c6f68SFilipe Manana * Therefore account for these 2 cases, so that our callers (like 5059337c6f68SFilipe Manana * btrfs_previous_item) don't miss an existing item with a key matching 5060337c6f68SFilipe Manana * the previous key we computed above. 5061337c6f68SFilipe Manana */ 5062337c6f68SFilipe Manana if (ret <= 0) 50637bb86316SChris Mason return 0; 5064925baeddSChris Mason return 1; 50657bb86316SChris Mason } 50667bb86316SChris Mason 50673f157a2fSChris Mason /* 50683f157a2fSChris Mason * A helper function to walk down the tree starting at min_key, and looking 5069de78b51aSEric Sandeen * for nodes or leaves that are have a minimum transaction id. 5070de78b51aSEric Sandeen * This is used by the btree defrag code, and tree logging 50713f157a2fSChris Mason * 50723f157a2fSChris Mason * This does not cow, but it does stuff the starting key it finds back 50733f157a2fSChris Mason * into min_key, so you can call btrfs_search_slot with cow=1 on the 50743f157a2fSChris Mason * key and get a writable path. 50753f157a2fSChris Mason * 50763f157a2fSChris Mason * This honors path->lowest_level to prevent descent past a given level 50773f157a2fSChris Mason * of the tree. 50783f157a2fSChris Mason * 5079d352ac68SChris Mason * min_trans indicates the oldest transaction that you are interested 5080d352ac68SChris Mason * in walking through. Any nodes or leaves older than min_trans are 5081d352ac68SChris Mason * skipped over (without reading them). 5082d352ac68SChris Mason * 50833f157a2fSChris Mason * returns zero if something useful was found, < 0 on error and 1 if there 50843f157a2fSChris Mason * was nothing in the tree that matched the search criteria. 50853f157a2fSChris Mason */ 50863f157a2fSChris Mason int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, 5087de78b51aSEric Sandeen struct btrfs_path *path, 50883f157a2fSChris Mason u64 min_trans) 50893f157a2fSChris Mason { 50903f157a2fSChris Mason struct extent_buffer *cur; 50913f157a2fSChris Mason struct btrfs_key found_key; 50923f157a2fSChris Mason int slot; 50939652480bSYan int sret; 50943f157a2fSChris Mason u32 nritems; 50953f157a2fSChris Mason int level; 50963f157a2fSChris Mason int ret = 1; 5097f98de9b9SFilipe Manana int keep_locks = path->keep_locks; 50983f157a2fSChris Mason 5099f98de9b9SFilipe Manana path->keep_locks = 1; 51003f157a2fSChris Mason again: 5101bd681513SChris Mason cur = btrfs_read_lock_root_node(root); 51023f157a2fSChris Mason level = btrfs_header_level(cur); 5103e02119d5SChris Mason WARN_ON(path->nodes[level]); 51043f157a2fSChris Mason path->nodes[level] = cur; 5105bd681513SChris Mason path->locks[level] = BTRFS_READ_LOCK; 51063f157a2fSChris Mason 51073f157a2fSChris Mason if (btrfs_header_generation(cur) < min_trans) { 51083f157a2fSChris Mason ret = 1; 51093f157a2fSChris Mason goto out; 51103f157a2fSChris Mason } 51113f157a2fSChris Mason while (1) { 51123f157a2fSChris Mason nritems = btrfs_header_nritems(cur); 51133f157a2fSChris Mason level = btrfs_header_level(cur); 5114e3b83361SQu Wenruo sret = btrfs_bin_search(cur, min_key, &slot); 5115cbca7d59SFilipe Manana if (sret < 0) { 5116cbca7d59SFilipe Manana ret = sret; 5117cbca7d59SFilipe Manana goto out; 5118cbca7d59SFilipe Manana } 51193f157a2fSChris Mason 5120323ac95bSChris Mason /* at the lowest level, we're done, setup the path and exit */ 5121323ac95bSChris Mason if (level == path->lowest_level) { 5122e02119d5SChris Mason if (slot >= nritems) 5123e02119d5SChris Mason goto find_next_key; 51243f157a2fSChris Mason ret = 0; 51253f157a2fSChris Mason path->slots[level] = slot; 51263f157a2fSChris Mason btrfs_item_key_to_cpu(cur, &found_key, slot); 51273f157a2fSChris Mason goto out; 51283f157a2fSChris Mason } 51299652480bSYan if (sret && slot > 0) 51309652480bSYan slot--; 51313f157a2fSChris Mason /* 5132de78b51aSEric Sandeen * check this node pointer against the min_trans parameters. 5133260db43cSRandy Dunlap * If it is too old, skip to the next one. 51343f157a2fSChris Mason */ 51353f157a2fSChris Mason while (slot < nritems) { 51363f157a2fSChris Mason u64 gen; 5137e02119d5SChris Mason 51383f157a2fSChris Mason gen = btrfs_node_ptr_generation(cur, slot); 51393f157a2fSChris Mason if (gen < min_trans) { 51403f157a2fSChris Mason slot++; 51413f157a2fSChris Mason continue; 51423f157a2fSChris Mason } 51433f157a2fSChris Mason break; 51443f157a2fSChris Mason } 5145e02119d5SChris Mason find_next_key: 51463f157a2fSChris Mason /* 51473f157a2fSChris Mason * we didn't find a candidate key in this node, walk forward 51483f157a2fSChris Mason * and find another one 51493f157a2fSChris Mason */ 51503f157a2fSChris Mason if (slot >= nritems) { 5151e02119d5SChris Mason path->slots[level] = slot; 5152e02119d5SChris Mason sret = btrfs_find_next_key(root, path, min_key, level, 5153de78b51aSEric Sandeen min_trans); 5154e02119d5SChris Mason if (sret == 0) { 5155b3b4aa74SDavid Sterba btrfs_release_path(path); 51563f157a2fSChris Mason goto again; 51573f157a2fSChris Mason } else { 51583f157a2fSChris Mason goto out; 51593f157a2fSChris Mason } 51603f157a2fSChris Mason } 51613f157a2fSChris Mason /* save our key for returning back */ 51623f157a2fSChris Mason btrfs_node_key_to_cpu(cur, &found_key, slot); 51633f157a2fSChris Mason path->slots[level] = slot; 51643f157a2fSChris Mason if (level == path->lowest_level) { 51653f157a2fSChris Mason ret = 0; 51663f157a2fSChris Mason goto out; 51673f157a2fSChris Mason } 51684b231ae4SDavid Sterba cur = btrfs_read_node_slot(cur, slot); 5169fb770ae4SLiu Bo if (IS_ERR(cur)) { 5170fb770ae4SLiu Bo ret = PTR_ERR(cur); 5171fb770ae4SLiu Bo goto out; 5172fb770ae4SLiu Bo } 51733f157a2fSChris Mason 5174bd681513SChris Mason btrfs_tree_read_lock(cur); 5175b4ce94deSChris Mason 5176bd681513SChris Mason path->locks[level - 1] = BTRFS_READ_LOCK; 51773f157a2fSChris Mason path->nodes[level - 1] = cur; 5178f7c79f30SChris Mason unlock_up(path, level, 1, 0, NULL); 51793f157a2fSChris Mason } 51803f157a2fSChris Mason out: 5181f98de9b9SFilipe Manana path->keep_locks = keep_locks; 5182f98de9b9SFilipe Manana if (ret == 0) { 5183f98de9b9SFilipe Manana btrfs_unlock_up_safe(path, path->lowest_level + 1); 5184f98de9b9SFilipe Manana memcpy(min_key, &found_key, sizeof(found_key)); 5185f98de9b9SFilipe Manana } 51863f157a2fSChris Mason return ret; 51873f157a2fSChris Mason } 51883f157a2fSChris Mason 51893f157a2fSChris Mason /* 51903f157a2fSChris Mason * this is similar to btrfs_next_leaf, but does not try to preserve 51913f157a2fSChris Mason * and fixup the path. It looks for and returns the next key in the 5192de78b51aSEric Sandeen * tree based on the current path and the min_trans parameters. 51933f157a2fSChris Mason * 51943f157a2fSChris Mason * 0 is returned if another key is found, < 0 if there are any errors 51953f157a2fSChris Mason * and 1 is returned if there are no higher keys in the tree 51963f157a2fSChris Mason * 51973f157a2fSChris Mason * path->keep_locks should be set to 1 on the search made before 51983f157a2fSChris Mason * calling this function. 51993f157a2fSChris Mason */ 5200e7a84565SChris Mason int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, 5201de78b51aSEric Sandeen struct btrfs_key *key, int level, u64 min_trans) 5202e7a84565SChris Mason { 5203e7a84565SChris Mason int slot; 5204e7a84565SChris Mason struct extent_buffer *c; 5205e7a84565SChris Mason 52066a9fb468SJosef Bacik WARN_ON(!path->keep_locks && !path->skip_locking); 5207e7a84565SChris Mason while (level < BTRFS_MAX_LEVEL) { 5208e7a84565SChris Mason if (!path->nodes[level]) 5209e7a84565SChris Mason return 1; 5210e7a84565SChris Mason 5211e7a84565SChris Mason slot = path->slots[level] + 1; 5212e7a84565SChris Mason c = path->nodes[level]; 52133f157a2fSChris Mason next: 5214e7a84565SChris Mason if (slot >= btrfs_header_nritems(c)) { 521533c66f43SYan Zheng int ret; 521633c66f43SYan Zheng int orig_lowest; 521733c66f43SYan Zheng struct btrfs_key cur_key; 521833c66f43SYan Zheng if (level + 1 >= BTRFS_MAX_LEVEL || 521933c66f43SYan Zheng !path->nodes[level + 1]) 5220e7a84565SChris Mason return 1; 522133c66f43SYan Zheng 52226a9fb468SJosef Bacik if (path->locks[level + 1] || path->skip_locking) { 522333c66f43SYan Zheng level++; 5224e7a84565SChris Mason continue; 5225e7a84565SChris Mason } 522633c66f43SYan Zheng 522733c66f43SYan Zheng slot = btrfs_header_nritems(c) - 1; 522833c66f43SYan Zheng if (level == 0) 522933c66f43SYan Zheng btrfs_item_key_to_cpu(c, &cur_key, slot); 523033c66f43SYan Zheng else 523133c66f43SYan Zheng btrfs_node_key_to_cpu(c, &cur_key, slot); 523233c66f43SYan Zheng 523333c66f43SYan Zheng orig_lowest = path->lowest_level; 5234b3b4aa74SDavid Sterba btrfs_release_path(path); 523533c66f43SYan Zheng path->lowest_level = level; 523633c66f43SYan Zheng ret = btrfs_search_slot(NULL, root, &cur_key, path, 523733c66f43SYan Zheng 0, 0); 523833c66f43SYan Zheng path->lowest_level = orig_lowest; 523933c66f43SYan Zheng if (ret < 0) 524033c66f43SYan Zheng return ret; 524133c66f43SYan Zheng 524233c66f43SYan Zheng c = path->nodes[level]; 524333c66f43SYan Zheng slot = path->slots[level]; 524433c66f43SYan Zheng if (ret == 0) 524533c66f43SYan Zheng slot++; 524633c66f43SYan Zheng goto next; 524733c66f43SYan Zheng } 524833c66f43SYan Zheng 5249e7a84565SChris Mason if (level == 0) 5250e7a84565SChris Mason btrfs_item_key_to_cpu(c, key, slot); 52513f157a2fSChris Mason else { 52523f157a2fSChris Mason u64 gen = btrfs_node_ptr_generation(c, slot); 52533f157a2fSChris Mason 52543f157a2fSChris Mason if (gen < min_trans) { 52553f157a2fSChris Mason slot++; 52563f157a2fSChris Mason goto next; 52573f157a2fSChris Mason } 5258e7a84565SChris Mason btrfs_node_key_to_cpu(c, key, slot); 52593f157a2fSChris Mason } 5260e7a84565SChris Mason return 0; 5261e7a84565SChris Mason } 5262e7a84565SChris Mason return 1; 5263e7a84565SChris Mason } 5264e7a84565SChris Mason 52657bb86316SChris Mason /* 5266925baeddSChris Mason * search the tree again to find a leaf with greater keys 52670f70abe2SChris Mason * returns 0 if it found something or 1 if there are no greater leaves. 52680f70abe2SChris Mason * returns < 0 on io errors. 526997571fd0SChris Mason */ 5270234b63a0SChris Mason int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) 5271d97e63b6SChris Mason { 52723d7806ecSJan Schmidt return btrfs_next_old_leaf(root, path, 0); 52733d7806ecSJan Schmidt } 52743d7806ecSJan Schmidt 52753d7806ecSJan Schmidt int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, 52763d7806ecSJan Schmidt u64 time_seq) 52773d7806ecSJan Schmidt { 5278d97e63b6SChris Mason int slot; 52798e73f275SChris Mason int level; 52805f39d397SChris Mason struct extent_buffer *c; 52818e73f275SChris Mason struct extent_buffer *next; 5282925baeddSChris Mason struct btrfs_key key; 5283925baeddSChris Mason u32 nritems; 5284925baeddSChris Mason int ret; 52850e46318dSJosef Bacik int i; 5286925baeddSChris Mason 5287925baeddSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 5288d397712bSChris Mason if (nritems == 0) 5289925baeddSChris Mason return 1; 5290925baeddSChris Mason 52918e73f275SChris Mason btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); 52928e73f275SChris Mason again: 52938e73f275SChris Mason level = 1; 52948e73f275SChris Mason next = NULL; 5295b3b4aa74SDavid Sterba btrfs_release_path(path); 52968e73f275SChris Mason 5297a2135011SChris Mason path->keep_locks = 1; 52988e73f275SChris Mason 52993d7806ecSJan Schmidt if (time_seq) 53003d7806ecSJan Schmidt ret = btrfs_search_old_slot(root, &key, path, time_seq); 53013d7806ecSJan Schmidt else 5302925baeddSChris Mason ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 5303925baeddSChris Mason path->keep_locks = 0; 5304925baeddSChris Mason 5305925baeddSChris Mason if (ret < 0) 5306925baeddSChris Mason return ret; 5307925baeddSChris Mason 5308a2135011SChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 5309168fd7d2SChris Mason /* 5310168fd7d2SChris Mason * by releasing the path above we dropped all our locks. A balance 5311168fd7d2SChris Mason * could have added more items next to the key that used to be 5312168fd7d2SChris Mason * at the very end of the block. So, check again here and 5313168fd7d2SChris Mason * advance the path if there are now more items available. 5314168fd7d2SChris Mason */ 5315a2135011SChris Mason if (nritems > 0 && path->slots[0] < nritems - 1) { 5316e457afecSYan Zheng if (ret == 0) 5317168fd7d2SChris Mason path->slots[0]++; 53188e73f275SChris Mason ret = 0; 5319925baeddSChris Mason goto done; 5320925baeddSChris Mason } 53210b43e04fSLiu Bo /* 53220b43e04fSLiu Bo * So the above check misses one case: 53230b43e04fSLiu Bo * - after releasing the path above, someone has removed the item that 53240b43e04fSLiu Bo * used to be at the very end of the block, and balance between leafs 53250b43e04fSLiu Bo * gets another one with bigger key.offset to replace it. 53260b43e04fSLiu Bo * 53270b43e04fSLiu Bo * This one should be returned as well, or we can get leaf corruption 53280b43e04fSLiu Bo * later(esp. in __btrfs_drop_extents()). 53290b43e04fSLiu Bo * 53300b43e04fSLiu Bo * And a bit more explanation about this check, 53310b43e04fSLiu Bo * with ret > 0, the key isn't found, the path points to the slot 53320b43e04fSLiu Bo * where it should be inserted, so the path->slots[0] item must be the 53330b43e04fSLiu Bo * bigger one. 53340b43e04fSLiu Bo */ 53350b43e04fSLiu Bo if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { 53360b43e04fSLiu Bo ret = 0; 53370b43e04fSLiu Bo goto done; 53380b43e04fSLiu Bo } 5339d97e63b6SChris Mason 5340234b63a0SChris Mason while (level < BTRFS_MAX_LEVEL) { 53418e73f275SChris Mason if (!path->nodes[level]) { 53428e73f275SChris Mason ret = 1; 53438e73f275SChris Mason goto done; 53448e73f275SChris Mason } 53455f39d397SChris Mason 5346d97e63b6SChris Mason slot = path->slots[level] + 1; 5347d97e63b6SChris Mason c = path->nodes[level]; 53485f39d397SChris Mason if (slot >= btrfs_header_nritems(c)) { 5349d97e63b6SChris Mason level++; 53508e73f275SChris Mason if (level == BTRFS_MAX_LEVEL) { 53518e73f275SChris Mason ret = 1; 53528e73f275SChris Mason goto done; 53538e73f275SChris Mason } 5354d97e63b6SChris Mason continue; 5355d97e63b6SChris Mason } 53565f39d397SChris Mason 53570e46318dSJosef Bacik 53580e46318dSJosef Bacik /* 53590e46318dSJosef Bacik * Our current level is where we're going to start from, and to 53600e46318dSJosef Bacik * make sure lockdep doesn't complain we need to drop our locks 53610e46318dSJosef Bacik * and nodes from 0 to our current level. 53620e46318dSJosef Bacik */ 53630e46318dSJosef Bacik for (i = 0; i < level; i++) { 53640e46318dSJosef Bacik if (path->locks[level]) { 53650e46318dSJosef Bacik btrfs_tree_read_unlock(path->nodes[i]); 53660e46318dSJosef Bacik path->locks[i] = 0; 53670e46318dSJosef Bacik } 53680e46318dSJosef Bacik free_extent_buffer(path->nodes[i]); 53690e46318dSJosef Bacik path->nodes[i] = NULL; 5370925baeddSChris Mason } 53715f39d397SChris Mason 53728e73f275SChris Mason next = c; 5373d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 5374cda79c54SDavid Sterba slot, &key); 53758e73f275SChris Mason if (ret == -EAGAIN) 53768e73f275SChris Mason goto again; 53775f39d397SChris Mason 537876a05b35SChris Mason if (ret < 0) { 5379b3b4aa74SDavid Sterba btrfs_release_path(path); 538076a05b35SChris Mason goto done; 538176a05b35SChris Mason } 538276a05b35SChris Mason 53835cd57b2cSChris Mason if (!path->skip_locking) { 5384bd681513SChris Mason ret = btrfs_try_tree_read_lock(next); 5385d42244a0SJan Schmidt if (!ret && time_seq) { 5386d42244a0SJan Schmidt /* 5387d42244a0SJan Schmidt * If we don't get the lock, we may be racing 5388d42244a0SJan Schmidt * with push_leaf_left, holding that lock while 5389d42244a0SJan Schmidt * itself waiting for the leaf we've currently 5390d42244a0SJan Schmidt * locked. To solve this situation, we give up 5391d42244a0SJan Schmidt * on our lock and cycle. 5392d42244a0SJan Schmidt */ 5393cf538830SJan Schmidt free_extent_buffer(next); 5394d42244a0SJan Schmidt btrfs_release_path(path); 5395d42244a0SJan Schmidt cond_resched(); 5396d42244a0SJan Schmidt goto again; 5397d42244a0SJan Schmidt } 53980e46318dSJosef Bacik if (!ret) 53990e46318dSJosef Bacik btrfs_tree_read_lock(next); 5400bd681513SChris Mason } 5401d97e63b6SChris Mason break; 5402d97e63b6SChris Mason } 5403d97e63b6SChris Mason path->slots[level] = slot; 5404d97e63b6SChris Mason while (1) { 5405d97e63b6SChris Mason level--; 5406d97e63b6SChris Mason path->nodes[level] = next; 5407d97e63b6SChris Mason path->slots[level] = 0; 5408a74a4b97SChris Mason if (!path->skip_locking) 5409ffeb03cfSJosef Bacik path->locks[level] = BTRFS_READ_LOCK; 5410d97e63b6SChris Mason if (!level) 5411d97e63b6SChris Mason break; 5412b4ce94deSChris Mason 5413d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 5414cda79c54SDavid Sterba 0, &key); 54158e73f275SChris Mason if (ret == -EAGAIN) 54168e73f275SChris Mason goto again; 54178e73f275SChris Mason 541876a05b35SChris Mason if (ret < 0) { 5419b3b4aa74SDavid Sterba btrfs_release_path(path); 542076a05b35SChris Mason goto done; 542176a05b35SChris Mason } 542276a05b35SChris Mason 5423ffeb03cfSJosef Bacik if (!path->skip_locking) 54240e46318dSJosef Bacik btrfs_tree_read_lock(next); 5425d97e63b6SChris Mason } 54268e73f275SChris Mason ret = 0; 5427925baeddSChris Mason done: 5428f7c79f30SChris Mason unlock_up(path, 0, 1, 0, NULL); 54298e73f275SChris Mason 54308e73f275SChris Mason return ret; 5431d97e63b6SChris Mason } 54320b86a832SChris Mason 54333f157a2fSChris Mason /* 54343f157a2fSChris Mason * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps 54353f157a2fSChris Mason * searching until it gets past min_objectid or finds an item of 'type' 54363f157a2fSChris Mason * 54373f157a2fSChris Mason * returns 0 if something is found, 1 if nothing was found and < 0 on error 54383f157a2fSChris Mason */ 54390b86a832SChris Mason int btrfs_previous_item(struct btrfs_root *root, 54400b86a832SChris Mason struct btrfs_path *path, u64 min_objectid, 54410b86a832SChris Mason int type) 54420b86a832SChris Mason { 54430b86a832SChris Mason struct btrfs_key found_key; 54440b86a832SChris Mason struct extent_buffer *leaf; 5445e02119d5SChris Mason u32 nritems; 54460b86a832SChris Mason int ret; 54470b86a832SChris Mason 54480b86a832SChris Mason while (1) { 54490b86a832SChris Mason if (path->slots[0] == 0) { 54500b86a832SChris Mason ret = btrfs_prev_leaf(root, path); 54510b86a832SChris Mason if (ret != 0) 54520b86a832SChris Mason return ret; 54530b86a832SChris Mason } else { 54540b86a832SChris Mason path->slots[0]--; 54550b86a832SChris Mason } 54560b86a832SChris Mason leaf = path->nodes[0]; 5457e02119d5SChris Mason nritems = btrfs_header_nritems(leaf); 5458e02119d5SChris Mason if (nritems == 0) 5459e02119d5SChris Mason return 1; 5460e02119d5SChris Mason if (path->slots[0] == nritems) 5461e02119d5SChris Mason path->slots[0]--; 5462e02119d5SChris Mason 54630b86a832SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 5464e02119d5SChris Mason if (found_key.objectid < min_objectid) 5465e02119d5SChris Mason break; 54660a4eefbbSYan Zheng if (found_key.type == type) 54670a4eefbbSYan Zheng return 0; 5468e02119d5SChris Mason if (found_key.objectid == min_objectid && 5469e02119d5SChris Mason found_key.type < type) 5470e02119d5SChris Mason break; 54710b86a832SChris Mason } 54720b86a832SChris Mason return 1; 54730b86a832SChris Mason } 5474ade2e0b3SWang Shilong 5475ade2e0b3SWang Shilong /* 5476ade2e0b3SWang Shilong * search in extent tree to find a previous Metadata/Data extent item with 5477ade2e0b3SWang Shilong * min objecitd. 5478ade2e0b3SWang Shilong * 5479ade2e0b3SWang Shilong * returns 0 if something is found, 1 if nothing was found and < 0 on error 5480ade2e0b3SWang Shilong */ 5481ade2e0b3SWang Shilong int btrfs_previous_extent_item(struct btrfs_root *root, 5482ade2e0b3SWang Shilong struct btrfs_path *path, u64 min_objectid) 5483ade2e0b3SWang Shilong { 5484ade2e0b3SWang Shilong struct btrfs_key found_key; 5485ade2e0b3SWang Shilong struct extent_buffer *leaf; 5486ade2e0b3SWang Shilong u32 nritems; 5487ade2e0b3SWang Shilong int ret; 5488ade2e0b3SWang Shilong 5489ade2e0b3SWang Shilong while (1) { 5490ade2e0b3SWang Shilong if (path->slots[0] == 0) { 5491ade2e0b3SWang Shilong ret = btrfs_prev_leaf(root, path); 5492ade2e0b3SWang Shilong if (ret != 0) 5493ade2e0b3SWang Shilong return ret; 5494ade2e0b3SWang Shilong } else { 5495ade2e0b3SWang Shilong path->slots[0]--; 5496ade2e0b3SWang Shilong } 5497ade2e0b3SWang Shilong leaf = path->nodes[0]; 5498ade2e0b3SWang Shilong nritems = btrfs_header_nritems(leaf); 5499ade2e0b3SWang Shilong if (nritems == 0) 5500ade2e0b3SWang Shilong return 1; 5501ade2e0b3SWang Shilong if (path->slots[0] == nritems) 5502ade2e0b3SWang Shilong path->slots[0]--; 5503ade2e0b3SWang Shilong 5504ade2e0b3SWang Shilong btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 5505ade2e0b3SWang Shilong if (found_key.objectid < min_objectid) 5506ade2e0b3SWang Shilong break; 5507ade2e0b3SWang Shilong if (found_key.type == BTRFS_EXTENT_ITEM_KEY || 5508ade2e0b3SWang Shilong found_key.type == BTRFS_METADATA_ITEM_KEY) 5509ade2e0b3SWang Shilong return 0; 5510ade2e0b3SWang Shilong if (found_key.objectid == min_objectid && 5511ade2e0b3SWang Shilong found_key.type < BTRFS_EXTENT_ITEM_KEY) 5512ade2e0b3SWang Shilong break; 5513ade2e0b3SWang Shilong } 5514ade2e0b3SWang Shilong return 1; 5515ade2e0b3SWang Shilong } 5516