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); 224*867ed321SJosef Bacik if (ret) { 225*867ed321SJosef Bacik btrfs_abort_transaction(trans, ret); 226be20aa9dSChris Mason return ret; 227*867ed321SJosef Bacik } 228be20aa9dSChris Mason 229be20aa9dSChris Mason btrfs_mark_buffer_dirty(cow); 230be20aa9dSChris Mason *cow_ret = cow; 231be20aa9dSChris Mason return 0; 232be20aa9dSChris Mason } 233be20aa9dSChris Mason 234bd989ba3SJan Schmidt enum mod_log_op { 235bd989ba3SJan Schmidt MOD_LOG_KEY_REPLACE, 236bd989ba3SJan Schmidt MOD_LOG_KEY_ADD, 237bd989ba3SJan Schmidt MOD_LOG_KEY_REMOVE, 238bd989ba3SJan Schmidt MOD_LOG_KEY_REMOVE_WHILE_FREEING, 239bd989ba3SJan Schmidt MOD_LOG_KEY_REMOVE_WHILE_MOVING, 240bd989ba3SJan Schmidt MOD_LOG_MOVE_KEYS, 241bd989ba3SJan Schmidt MOD_LOG_ROOT_REPLACE, 242bd989ba3SJan Schmidt }; 243bd989ba3SJan Schmidt 244bd989ba3SJan Schmidt struct tree_mod_root { 245bd989ba3SJan Schmidt u64 logical; 246bd989ba3SJan Schmidt u8 level; 247bd989ba3SJan Schmidt }; 248bd989ba3SJan Schmidt 249bd989ba3SJan Schmidt struct tree_mod_elem { 250bd989ba3SJan Schmidt struct rb_node node; 251298cfd36SChandan Rajendra u64 logical; 252097b8a7cSJan Schmidt u64 seq; 253bd989ba3SJan Schmidt enum mod_log_op op; 254bd989ba3SJan Schmidt 255bd989ba3SJan Schmidt /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ 256bd989ba3SJan Schmidt int slot; 257bd989ba3SJan Schmidt 258bd989ba3SJan Schmidt /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ 259bd989ba3SJan Schmidt u64 generation; 260bd989ba3SJan Schmidt 261bd989ba3SJan Schmidt /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ 262bd989ba3SJan Schmidt struct btrfs_disk_key key; 263bd989ba3SJan Schmidt u64 blockptr; 264bd989ba3SJan Schmidt 265bd989ba3SJan Schmidt /* this is used for op == MOD_LOG_MOVE_KEYS */ 266b6dfa35bSDavid Sterba struct { 267b6dfa35bSDavid Sterba int dst_slot; 268b6dfa35bSDavid Sterba int nr_items; 269b6dfa35bSDavid Sterba } move; 270bd989ba3SJan Schmidt 271bd989ba3SJan Schmidt /* this is used for op == MOD_LOG_ROOT_REPLACE */ 272bd989ba3SJan Schmidt struct tree_mod_root old_root; 273bd989ba3SJan Schmidt }; 274bd989ba3SJan Schmidt 275097b8a7cSJan Schmidt /* 276fcebe456SJosef Bacik * Pull a new tree mod seq number for our operation. 277fc36ed7eSJan Schmidt */ 278fcebe456SJosef Bacik static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) 279fc36ed7eSJan Schmidt { 280fc36ed7eSJan Schmidt return atomic64_inc_return(&fs_info->tree_mod_seq); 281fc36ed7eSJan Schmidt } 282fc36ed7eSJan Schmidt 283fc36ed7eSJan Schmidt /* 284097b8a7cSJan Schmidt * This adds a new blocker to the tree mod log's blocker list if the @elem 285097b8a7cSJan Schmidt * passed does not already have a sequence number set. So when a caller expects 286097b8a7cSJan Schmidt * to record tree modifications, it should ensure to set elem->seq to zero 287097b8a7cSJan Schmidt * before calling btrfs_get_tree_mod_seq. 288097b8a7cSJan Schmidt * Returns a fresh, unused tree log modification sequence number, even if no new 289097b8a7cSJan Schmidt * blocker was added. 290097b8a7cSJan Schmidt */ 291097b8a7cSJan Schmidt u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, 292bd989ba3SJan Schmidt struct seq_list *elem) 293bd989ba3SJan Schmidt { 294b1a09f1eSDavid Sterba write_lock(&fs_info->tree_mod_log_lock); 295097b8a7cSJan Schmidt if (!elem->seq) { 296fcebe456SJosef Bacik elem->seq = btrfs_inc_tree_mod_seq(fs_info); 297097b8a7cSJan Schmidt list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); 298097b8a7cSJan Schmidt } 299b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 300097b8a7cSJan Schmidt 301fcebe456SJosef Bacik return elem->seq; 302bd989ba3SJan Schmidt } 303bd989ba3SJan Schmidt 304bd989ba3SJan Schmidt void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, 305bd989ba3SJan Schmidt struct seq_list *elem) 306bd989ba3SJan Schmidt { 307bd989ba3SJan Schmidt struct rb_root *tm_root; 308bd989ba3SJan Schmidt struct rb_node *node; 309bd989ba3SJan Schmidt struct rb_node *next; 310bd989ba3SJan Schmidt struct tree_mod_elem *tm; 311bd989ba3SJan Schmidt u64 min_seq = (u64)-1; 312bd989ba3SJan Schmidt u64 seq_putting = elem->seq; 313bd989ba3SJan Schmidt 314bd989ba3SJan Schmidt if (!seq_putting) 315bd989ba3SJan Schmidt return; 316bd989ba3SJan Schmidt 3177227ff4dSFilipe Manana write_lock(&fs_info->tree_mod_log_lock); 318bd989ba3SJan Schmidt list_del(&elem->list); 319097b8a7cSJan Schmidt elem->seq = 0; 320bd989ba3SJan Schmidt 32142836cf4SFilipe Manana if (!list_empty(&fs_info->tree_mod_seq_list)) { 32242836cf4SFilipe Manana struct seq_list *first; 32342836cf4SFilipe Manana 32442836cf4SFilipe Manana first = list_first_entry(&fs_info->tree_mod_seq_list, 32542836cf4SFilipe Manana struct seq_list, list); 32642836cf4SFilipe Manana if (seq_putting > first->seq) { 327bd989ba3SJan Schmidt /* 32842836cf4SFilipe Manana * Blocker with lower sequence number exists, we 32942836cf4SFilipe Manana * cannot remove anything from the log. 330bd989ba3SJan Schmidt */ 3317227ff4dSFilipe Manana write_unlock(&fs_info->tree_mod_log_lock); 332097b8a7cSJan Schmidt return; 333bd989ba3SJan Schmidt } 33442836cf4SFilipe Manana min_seq = first->seq; 335bd989ba3SJan Schmidt } 336097b8a7cSJan Schmidt 337097b8a7cSJan Schmidt /* 338bd989ba3SJan Schmidt * anything that's lower than the lowest existing (read: blocked) 339bd989ba3SJan Schmidt * sequence number can be removed from the tree. 340bd989ba3SJan Schmidt */ 341bd989ba3SJan Schmidt tm_root = &fs_info->tree_mod_log; 342bd989ba3SJan Schmidt for (node = rb_first(tm_root); node; node = next) { 343bd989ba3SJan Schmidt next = rb_next(node); 3446b4df8b6SGeliang Tang tm = rb_entry(node, struct tree_mod_elem, node); 3456609fee8SFilipe Manana if (tm->seq >= min_seq) 346bd989ba3SJan Schmidt continue; 347bd989ba3SJan Schmidt rb_erase(node, tm_root); 348bd989ba3SJan Schmidt kfree(tm); 349bd989ba3SJan Schmidt } 350b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 351bd989ba3SJan Schmidt } 352bd989ba3SJan Schmidt 353bd989ba3SJan Schmidt /* 354bd989ba3SJan Schmidt * key order of the log: 355298cfd36SChandan Rajendra * node/leaf start address -> sequence 356bd989ba3SJan Schmidt * 357298cfd36SChandan Rajendra * The 'start address' is the logical address of the *new* root node 358298cfd36SChandan Rajendra * for root replace operations, or the logical address of the affected 359298cfd36SChandan Rajendra * block for all other operations. 360bd989ba3SJan Schmidt */ 361bd989ba3SJan Schmidt static noinline int 362bd989ba3SJan Schmidt __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) 363bd989ba3SJan Schmidt { 364bd989ba3SJan Schmidt struct rb_root *tm_root; 365bd989ba3SJan Schmidt struct rb_node **new; 366bd989ba3SJan Schmidt struct rb_node *parent = NULL; 367bd989ba3SJan Schmidt struct tree_mod_elem *cur; 368bd989ba3SJan Schmidt 36973e82fe4SDavid Sterba lockdep_assert_held_write(&fs_info->tree_mod_log_lock); 37073e82fe4SDavid Sterba 371fcebe456SJosef Bacik tm->seq = btrfs_inc_tree_mod_seq(fs_info); 372bd989ba3SJan Schmidt 373bd989ba3SJan Schmidt tm_root = &fs_info->tree_mod_log; 374bd989ba3SJan Schmidt new = &tm_root->rb_node; 375bd989ba3SJan Schmidt while (*new) { 3766b4df8b6SGeliang Tang cur = rb_entry(*new, struct tree_mod_elem, node); 377bd989ba3SJan Schmidt parent = *new; 378298cfd36SChandan Rajendra if (cur->logical < tm->logical) 379bd989ba3SJan Schmidt new = &((*new)->rb_left); 380298cfd36SChandan Rajendra else if (cur->logical > tm->logical) 381bd989ba3SJan Schmidt new = &((*new)->rb_right); 382097b8a7cSJan Schmidt else if (cur->seq < tm->seq) 383bd989ba3SJan Schmidt new = &((*new)->rb_left); 384097b8a7cSJan Schmidt else if (cur->seq > tm->seq) 385bd989ba3SJan Schmidt new = &((*new)->rb_right); 3865de865eeSFilipe David Borba Manana else 3875de865eeSFilipe David Borba Manana return -EEXIST; 388bd989ba3SJan Schmidt } 389bd989ba3SJan Schmidt 390bd989ba3SJan Schmidt rb_link_node(&tm->node, parent, new); 391bd989ba3SJan Schmidt rb_insert_color(&tm->node, tm_root); 3925de865eeSFilipe David Borba Manana return 0; 393bd989ba3SJan Schmidt } 394bd989ba3SJan Schmidt 395097b8a7cSJan Schmidt /* 396097b8a7cSJan Schmidt * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it 397097b8a7cSJan Schmidt * returns zero with the tree_mod_log_lock acquired. The caller must hold 398097b8a7cSJan Schmidt * this until all tree mod log insertions are recorded in the rb tree and then 399b1a09f1eSDavid Sterba * write unlock fs_info::tree_mod_log_lock. 400097b8a7cSJan Schmidt */ 401e9b7fd4dSJan Schmidt static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, 402e9b7fd4dSJan Schmidt struct extent_buffer *eb) { 403e9b7fd4dSJan Schmidt smp_mb(); 404e9b7fd4dSJan Schmidt if (list_empty(&(fs_info)->tree_mod_seq_list)) 405e9b7fd4dSJan Schmidt return 1; 406097b8a7cSJan Schmidt if (eb && btrfs_header_level(eb) == 0) 407e9b7fd4dSJan Schmidt return 1; 4085de865eeSFilipe David Borba Manana 409b1a09f1eSDavid Sterba write_lock(&fs_info->tree_mod_log_lock); 4105de865eeSFilipe David Borba Manana if (list_empty(&(fs_info)->tree_mod_seq_list)) { 411b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 4125de865eeSFilipe David Borba Manana return 1; 4135de865eeSFilipe David Borba Manana } 4145de865eeSFilipe David Borba Manana 415e9b7fd4dSJan Schmidt return 0; 416e9b7fd4dSJan Schmidt } 417e9b7fd4dSJan Schmidt 4185de865eeSFilipe David Borba Manana /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ 4195de865eeSFilipe David Borba Manana static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, 4205de865eeSFilipe David Borba Manana struct extent_buffer *eb) 4215de865eeSFilipe David Borba Manana { 4225de865eeSFilipe David Borba Manana smp_mb(); 4235de865eeSFilipe David Borba Manana if (list_empty(&(fs_info)->tree_mod_seq_list)) 4245de865eeSFilipe David Borba Manana return 0; 4255de865eeSFilipe David Borba Manana if (eb && btrfs_header_level(eb) == 0) 4265de865eeSFilipe David Borba Manana return 0; 4275de865eeSFilipe David Borba Manana 4285de865eeSFilipe David Borba Manana return 1; 4295de865eeSFilipe David Borba Manana } 4305de865eeSFilipe David Borba Manana 4315de865eeSFilipe David Borba Manana static struct tree_mod_elem * 4325de865eeSFilipe David Borba Manana alloc_tree_mod_elem(struct extent_buffer *eb, int slot, 433bd989ba3SJan Schmidt enum mod_log_op op, gfp_t flags) 434bd989ba3SJan Schmidt { 435097b8a7cSJan Schmidt struct tree_mod_elem *tm; 436bd989ba3SJan Schmidt 437c8cc6341SJosef Bacik tm = kzalloc(sizeof(*tm), flags); 438c8cc6341SJosef Bacik if (!tm) 4395de865eeSFilipe David Borba Manana return NULL; 440bd989ba3SJan Schmidt 441298cfd36SChandan Rajendra tm->logical = eb->start; 442bd989ba3SJan Schmidt if (op != MOD_LOG_KEY_ADD) { 443bd989ba3SJan Schmidt btrfs_node_key(eb, &tm->key, slot); 444bd989ba3SJan Schmidt tm->blockptr = btrfs_node_blockptr(eb, slot); 445bd989ba3SJan Schmidt } 446bd989ba3SJan Schmidt tm->op = op; 447bd989ba3SJan Schmidt tm->slot = slot; 448bd989ba3SJan Schmidt tm->generation = btrfs_node_ptr_generation(eb, slot); 4495de865eeSFilipe David Borba Manana RB_CLEAR_NODE(&tm->node); 450bd989ba3SJan Schmidt 4515de865eeSFilipe David Borba Manana return tm; 452097b8a7cSJan Schmidt } 453097b8a7cSJan Schmidt 454e09c2efeSDavid Sterba static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, 455097b8a7cSJan Schmidt enum mod_log_op op, gfp_t flags) 456097b8a7cSJan Schmidt { 4575de865eeSFilipe David Borba Manana struct tree_mod_elem *tm; 4585de865eeSFilipe David Borba Manana int ret; 4595de865eeSFilipe David Borba Manana 460e09c2efeSDavid Sterba if (!tree_mod_need_log(eb->fs_info, eb)) 461097b8a7cSJan Schmidt return 0; 462097b8a7cSJan Schmidt 4635de865eeSFilipe David Borba Manana tm = alloc_tree_mod_elem(eb, slot, op, flags); 4645de865eeSFilipe David Borba Manana if (!tm) 4655de865eeSFilipe David Borba Manana return -ENOMEM; 4665de865eeSFilipe David Borba Manana 467e09c2efeSDavid Sterba if (tree_mod_dont_log(eb->fs_info, eb)) { 4685de865eeSFilipe David Borba Manana kfree(tm); 4695de865eeSFilipe David Borba Manana return 0; 4705de865eeSFilipe David Borba Manana } 4715de865eeSFilipe David Borba Manana 472e09c2efeSDavid Sterba ret = __tree_mod_log_insert(eb->fs_info, tm); 473b1a09f1eSDavid Sterba write_unlock(&eb->fs_info->tree_mod_log_lock); 4745de865eeSFilipe David Borba Manana if (ret) 4755de865eeSFilipe David Borba Manana kfree(tm); 4765de865eeSFilipe David Borba Manana 4775de865eeSFilipe David Borba Manana return ret; 478097b8a7cSJan Schmidt } 479097b8a7cSJan Schmidt 4806074d45fSDavid Sterba static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, 4816074d45fSDavid Sterba int dst_slot, int src_slot, int nr_items) 482bd989ba3SJan Schmidt { 4835de865eeSFilipe David Borba Manana struct tree_mod_elem *tm = NULL; 4845de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list = NULL; 4855de865eeSFilipe David Borba Manana int ret = 0; 486bd989ba3SJan Schmidt int i; 4875de865eeSFilipe David Borba Manana int locked = 0; 488bd989ba3SJan Schmidt 4896074d45fSDavid Sterba if (!tree_mod_need_log(eb->fs_info, eb)) 490f395694cSJan Schmidt return 0; 491bd989ba3SJan Schmidt 492176ef8f5SDavid Sterba tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); 4935de865eeSFilipe David Borba Manana if (!tm_list) 4945de865eeSFilipe David Borba Manana return -ENOMEM; 495bd989ba3SJan Schmidt 496176ef8f5SDavid Sterba tm = kzalloc(sizeof(*tm), GFP_NOFS); 4975de865eeSFilipe David Borba Manana if (!tm) { 4985de865eeSFilipe David Borba Manana ret = -ENOMEM; 4995de865eeSFilipe David Borba Manana goto free_tms; 5005de865eeSFilipe David Borba Manana } 501f395694cSJan Schmidt 502298cfd36SChandan Rajendra tm->logical = eb->start; 503bd989ba3SJan Schmidt tm->slot = src_slot; 504bd989ba3SJan Schmidt tm->move.dst_slot = dst_slot; 505bd989ba3SJan Schmidt tm->move.nr_items = nr_items; 506bd989ba3SJan Schmidt tm->op = MOD_LOG_MOVE_KEYS; 507bd989ba3SJan Schmidt 5085de865eeSFilipe David Borba Manana for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { 5095de865eeSFilipe David Borba Manana tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, 510176ef8f5SDavid Sterba MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); 5115de865eeSFilipe David Borba Manana if (!tm_list[i]) { 5125de865eeSFilipe David Borba Manana ret = -ENOMEM; 5135de865eeSFilipe David Borba Manana goto free_tms; 5145de865eeSFilipe David Borba Manana } 515bd989ba3SJan Schmidt } 516bd989ba3SJan Schmidt 5176074d45fSDavid Sterba if (tree_mod_dont_log(eb->fs_info, eb)) 5185de865eeSFilipe David Borba Manana goto free_tms; 5195de865eeSFilipe David Borba Manana locked = 1; 5205de865eeSFilipe David Borba Manana 5215de865eeSFilipe David Borba Manana /* 5225de865eeSFilipe David Borba Manana * When we override something during the move, we log these removals. 5235de865eeSFilipe David Borba Manana * This can only happen when we move towards the beginning of the 5245de865eeSFilipe David Borba Manana * buffer, i.e. dst_slot < src_slot. 5255de865eeSFilipe David Borba Manana */ 5265de865eeSFilipe David Borba Manana for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { 5276074d45fSDavid Sterba ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); 5285de865eeSFilipe David Borba Manana if (ret) 5295de865eeSFilipe David Borba Manana goto free_tms; 5305de865eeSFilipe David Borba Manana } 5315de865eeSFilipe David Borba Manana 5326074d45fSDavid Sterba ret = __tree_mod_log_insert(eb->fs_info, tm); 5335de865eeSFilipe David Borba Manana if (ret) 5345de865eeSFilipe David Borba Manana goto free_tms; 535b1a09f1eSDavid Sterba write_unlock(&eb->fs_info->tree_mod_log_lock); 5365de865eeSFilipe David Borba Manana kfree(tm_list); 5375de865eeSFilipe David Borba Manana 5385de865eeSFilipe David Borba Manana return 0; 5395de865eeSFilipe David Borba Manana free_tms: 5405de865eeSFilipe David Borba Manana for (i = 0; i < nr_items; i++) { 5415de865eeSFilipe David Borba Manana if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) 5426074d45fSDavid Sterba rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); 5435de865eeSFilipe David Borba Manana kfree(tm_list[i]); 5445de865eeSFilipe David Borba Manana } 5455de865eeSFilipe David Borba Manana if (locked) 546b1a09f1eSDavid Sterba write_unlock(&eb->fs_info->tree_mod_log_lock); 5475de865eeSFilipe David Borba Manana kfree(tm_list); 5485de865eeSFilipe David Borba Manana kfree(tm); 5495de865eeSFilipe David Borba Manana 5505de865eeSFilipe David Borba Manana return ret; 5515de865eeSFilipe David Borba Manana } 5525de865eeSFilipe David Borba Manana 5535de865eeSFilipe David Borba Manana static inline int 5545de865eeSFilipe David Borba Manana __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, 5555de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list, 5565de865eeSFilipe David Borba Manana int nritems) 557097b8a7cSJan Schmidt { 5585de865eeSFilipe David Borba Manana int i, j; 559097b8a7cSJan Schmidt int ret; 560097b8a7cSJan Schmidt 561097b8a7cSJan Schmidt for (i = nritems - 1; i >= 0; i--) { 5625de865eeSFilipe David Borba Manana ret = __tree_mod_log_insert(fs_info, tm_list[i]); 5635de865eeSFilipe David Borba Manana if (ret) { 5645de865eeSFilipe David Borba Manana for (j = nritems - 1; j > i; j--) 5655de865eeSFilipe David Borba Manana rb_erase(&tm_list[j]->node, 5665de865eeSFilipe David Borba Manana &fs_info->tree_mod_log); 5675de865eeSFilipe David Borba Manana return ret; 568097b8a7cSJan Schmidt } 569097b8a7cSJan Schmidt } 570097b8a7cSJan Schmidt 5715de865eeSFilipe David Borba Manana return 0; 5725de865eeSFilipe David Borba Manana } 5735de865eeSFilipe David Borba Manana 57495b757c1SDavid Sterba static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, 57595b757c1SDavid Sterba struct extent_buffer *new_root, int log_removal) 576bd989ba3SJan Schmidt { 57795b757c1SDavid Sterba struct btrfs_fs_info *fs_info = old_root->fs_info; 5785de865eeSFilipe David Borba Manana struct tree_mod_elem *tm = NULL; 5795de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list = NULL; 5805de865eeSFilipe David Borba Manana int nritems = 0; 5815de865eeSFilipe David Borba Manana int ret = 0; 5825de865eeSFilipe David Borba Manana int i; 583bd989ba3SJan Schmidt 5845de865eeSFilipe David Borba Manana if (!tree_mod_need_log(fs_info, NULL)) 585097b8a7cSJan Schmidt return 0; 586097b8a7cSJan Schmidt 5875de865eeSFilipe David Borba Manana if (log_removal && btrfs_header_level(old_root) > 0) { 5885de865eeSFilipe David Borba Manana nritems = btrfs_header_nritems(old_root); 58931e818feSDavid Sterba tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), 590bcc8e07fSDavid Sterba GFP_NOFS); 5915de865eeSFilipe David Borba Manana if (!tm_list) { 5925de865eeSFilipe David Borba Manana ret = -ENOMEM; 5935de865eeSFilipe David Borba Manana goto free_tms; 5945de865eeSFilipe David Borba Manana } 5955de865eeSFilipe David Borba Manana for (i = 0; i < nritems; i++) { 5965de865eeSFilipe David Borba Manana tm_list[i] = alloc_tree_mod_elem(old_root, i, 597bcc8e07fSDavid Sterba MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); 5985de865eeSFilipe David Borba Manana if (!tm_list[i]) { 5995de865eeSFilipe David Borba Manana ret = -ENOMEM; 6005de865eeSFilipe David Borba Manana goto free_tms; 6015de865eeSFilipe David Borba Manana } 6025de865eeSFilipe David Borba Manana } 6035de865eeSFilipe David Borba Manana } 604d9abbf1cSJan Schmidt 605bcc8e07fSDavid Sterba tm = kzalloc(sizeof(*tm), GFP_NOFS); 6065de865eeSFilipe David Borba Manana if (!tm) { 6075de865eeSFilipe David Borba Manana ret = -ENOMEM; 6085de865eeSFilipe David Borba Manana goto free_tms; 6095de865eeSFilipe David Borba Manana } 610bd989ba3SJan Schmidt 611298cfd36SChandan Rajendra tm->logical = new_root->start; 612bd989ba3SJan Schmidt tm->old_root.logical = old_root->start; 613bd989ba3SJan Schmidt tm->old_root.level = btrfs_header_level(old_root); 614bd989ba3SJan Schmidt tm->generation = btrfs_header_generation(old_root); 615bd989ba3SJan Schmidt tm->op = MOD_LOG_ROOT_REPLACE; 616bd989ba3SJan Schmidt 6175de865eeSFilipe David Borba Manana if (tree_mod_dont_log(fs_info, NULL)) 6185de865eeSFilipe David Borba Manana goto free_tms; 6195de865eeSFilipe David Borba Manana 6205de865eeSFilipe David Borba Manana if (tm_list) 6215de865eeSFilipe David Borba Manana ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); 6225de865eeSFilipe David Borba Manana if (!ret) 6235de865eeSFilipe David Borba Manana ret = __tree_mod_log_insert(fs_info, tm); 6245de865eeSFilipe David Borba Manana 625b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 6265de865eeSFilipe David Borba Manana if (ret) 6275de865eeSFilipe David Borba Manana goto free_tms; 6285de865eeSFilipe David Borba Manana kfree(tm_list); 6295de865eeSFilipe David Borba Manana 6305de865eeSFilipe David Borba Manana return ret; 6315de865eeSFilipe David Borba Manana 6325de865eeSFilipe David Borba Manana free_tms: 6335de865eeSFilipe David Borba Manana if (tm_list) { 6345de865eeSFilipe David Borba Manana for (i = 0; i < nritems; i++) 6355de865eeSFilipe David Borba Manana kfree(tm_list[i]); 6365de865eeSFilipe David Borba Manana kfree(tm_list); 6375de865eeSFilipe David Borba Manana } 6385de865eeSFilipe David Borba Manana kfree(tm); 6395de865eeSFilipe David Borba Manana 6405de865eeSFilipe David Borba Manana return ret; 641bd989ba3SJan Schmidt } 642bd989ba3SJan Schmidt 643bd989ba3SJan Schmidt static struct tree_mod_elem * 644bd989ba3SJan Schmidt __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, 645bd989ba3SJan Schmidt int smallest) 646bd989ba3SJan Schmidt { 647bd989ba3SJan Schmidt struct rb_root *tm_root; 648bd989ba3SJan Schmidt struct rb_node *node; 649bd989ba3SJan Schmidt struct tree_mod_elem *cur = NULL; 650bd989ba3SJan Schmidt struct tree_mod_elem *found = NULL; 651bd989ba3SJan Schmidt 652b1a09f1eSDavid Sterba read_lock(&fs_info->tree_mod_log_lock); 653bd989ba3SJan Schmidt tm_root = &fs_info->tree_mod_log; 654bd989ba3SJan Schmidt node = tm_root->rb_node; 655bd989ba3SJan Schmidt while (node) { 6566b4df8b6SGeliang Tang cur = rb_entry(node, struct tree_mod_elem, node); 657298cfd36SChandan Rajendra if (cur->logical < start) { 658bd989ba3SJan Schmidt node = node->rb_left; 659298cfd36SChandan Rajendra } else if (cur->logical > start) { 660bd989ba3SJan Schmidt node = node->rb_right; 661097b8a7cSJan Schmidt } else if (cur->seq < min_seq) { 662bd989ba3SJan Schmidt node = node->rb_left; 663bd989ba3SJan Schmidt } else if (!smallest) { 664bd989ba3SJan Schmidt /* we want the node with the highest seq */ 665bd989ba3SJan Schmidt if (found) 666097b8a7cSJan Schmidt BUG_ON(found->seq > cur->seq); 667bd989ba3SJan Schmidt found = cur; 668bd989ba3SJan Schmidt node = node->rb_left; 669097b8a7cSJan Schmidt } else if (cur->seq > min_seq) { 670bd989ba3SJan Schmidt /* we want the node with the smallest seq */ 671bd989ba3SJan Schmidt if (found) 672097b8a7cSJan Schmidt BUG_ON(found->seq < cur->seq); 673bd989ba3SJan Schmidt found = cur; 674bd989ba3SJan Schmidt node = node->rb_right; 675bd989ba3SJan Schmidt } else { 676bd989ba3SJan Schmidt found = cur; 677bd989ba3SJan Schmidt break; 678bd989ba3SJan Schmidt } 679bd989ba3SJan Schmidt } 680b1a09f1eSDavid Sterba read_unlock(&fs_info->tree_mod_log_lock); 681bd989ba3SJan Schmidt 682bd989ba3SJan Schmidt return found; 683bd989ba3SJan Schmidt } 684bd989ba3SJan Schmidt 685bd989ba3SJan Schmidt /* 686bd989ba3SJan Schmidt * this returns the element from the log with the smallest time sequence 687bd989ba3SJan Schmidt * value that's in the log (the oldest log item). any element with a time 688bd989ba3SJan Schmidt * sequence lower than min_seq will be ignored. 689bd989ba3SJan Schmidt */ 690bd989ba3SJan Schmidt static struct tree_mod_elem * 691bd989ba3SJan Schmidt tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, 692bd989ba3SJan Schmidt u64 min_seq) 693bd989ba3SJan Schmidt { 694bd989ba3SJan Schmidt return __tree_mod_log_search(fs_info, start, min_seq, 1); 695bd989ba3SJan Schmidt } 696bd989ba3SJan Schmidt 697bd989ba3SJan Schmidt /* 698bd989ba3SJan Schmidt * this returns the element from the log with the largest time sequence 699bd989ba3SJan Schmidt * value that's in the log (the most recent log item). any element with 700bd989ba3SJan Schmidt * a time sequence lower than min_seq will be ignored. 701bd989ba3SJan Schmidt */ 702bd989ba3SJan Schmidt static struct tree_mod_elem * 703bd989ba3SJan Schmidt tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) 704bd989ba3SJan Schmidt { 705bd989ba3SJan Schmidt return __tree_mod_log_search(fs_info, start, min_seq, 0); 706bd989ba3SJan Schmidt } 707bd989ba3SJan Schmidt 708ed874f0dSDavid Sterba static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst, 709bd989ba3SJan Schmidt struct extent_buffer *src, unsigned long dst_offset, 71090f8d62eSJan Schmidt unsigned long src_offset, int nr_items) 711bd989ba3SJan Schmidt { 712ed874f0dSDavid Sterba struct btrfs_fs_info *fs_info = dst->fs_info; 7135de865eeSFilipe David Borba Manana int ret = 0; 7145de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list = NULL; 7155de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list_add, **tm_list_rem; 716bd989ba3SJan Schmidt int i; 7175de865eeSFilipe David Borba Manana int locked = 0; 718bd989ba3SJan Schmidt 7195de865eeSFilipe David Borba Manana if (!tree_mod_need_log(fs_info, NULL)) 7205de865eeSFilipe David Borba Manana return 0; 721bd989ba3SJan Schmidt 722c8cc6341SJosef Bacik if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) 7235de865eeSFilipe David Borba Manana return 0; 7245de865eeSFilipe David Borba Manana 72531e818feSDavid Sterba tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), 7265de865eeSFilipe David Borba Manana GFP_NOFS); 7275de865eeSFilipe David Borba Manana if (!tm_list) 7285de865eeSFilipe David Borba Manana return -ENOMEM; 7295de865eeSFilipe David Borba Manana 7305de865eeSFilipe David Borba Manana tm_list_add = tm_list; 7315de865eeSFilipe David Borba Manana tm_list_rem = tm_list + nr_items; 7325de865eeSFilipe David Borba Manana for (i = 0; i < nr_items; i++) { 7335de865eeSFilipe David Borba Manana tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, 7345de865eeSFilipe David Borba Manana MOD_LOG_KEY_REMOVE, GFP_NOFS); 7355de865eeSFilipe David Borba Manana if (!tm_list_rem[i]) { 7365de865eeSFilipe David Borba Manana ret = -ENOMEM; 7375de865eeSFilipe David Borba Manana goto free_tms; 7385de865eeSFilipe David Borba Manana } 7395de865eeSFilipe David Borba Manana 7405de865eeSFilipe David Borba Manana tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, 7415de865eeSFilipe David Borba Manana MOD_LOG_KEY_ADD, GFP_NOFS); 7425de865eeSFilipe David Borba Manana if (!tm_list_add[i]) { 7435de865eeSFilipe David Borba Manana ret = -ENOMEM; 7445de865eeSFilipe David Borba Manana goto free_tms; 7455de865eeSFilipe David Borba Manana } 7465de865eeSFilipe David Borba Manana } 7475de865eeSFilipe David Borba Manana 7485de865eeSFilipe David Borba Manana if (tree_mod_dont_log(fs_info, NULL)) 7495de865eeSFilipe David Borba Manana goto free_tms; 7505de865eeSFilipe David Borba Manana locked = 1; 751bd989ba3SJan Schmidt 752bd989ba3SJan Schmidt for (i = 0; i < nr_items; i++) { 7535de865eeSFilipe David Borba Manana ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); 7545de865eeSFilipe David Borba Manana if (ret) 7555de865eeSFilipe David Borba Manana goto free_tms; 7565de865eeSFilipe David Borba Manana ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); 7575de865eeSFilipe David Borba Manana if (ret) 7585de865eeSFilipe David Borba Manana goto free_tms; 759bd989ba3SJan Schmidt } 7605de865eeSFilipe David Borba Manana 761b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 7625de865eeSFilipe David Borba Manana kfree(tm_list); 7635de865eeSFilipe David Borba Manana 7645de865eeSFilipe David Borba Manana return 0; 7655de865eeSFilipe David Borba Manana 7665de865eeSFilipe David Borba Manana free_tms: 7675de865eeSFilipe David Borba Manana for (i = 0; i < nr_items * 2; i++) { 7685de865eeSFilipe David Borba Manana if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) 7695de865eeSFilipe David Borba Manana rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); 7705de865eeSFilipe David Borba Manana kfree(tm_list[i]); 7715de865eeSFilipe David Borba Manana } 7725de865eeSFilipe David Borba Manana if (locked) 773b1a09f1eSDavid Sterba write_unlock(&fs_info->tree_mod_log_lock); 7745de865eeSFilipe David Borba Manana kfree(tm_list); 7755de865eeSFilipe David Borba Manana 7765de865eeSFilipe David Borba Manana return ret; 777bd989ba3SJan Schmidt } 778bd989ba3SJan Schmidt 779db7279a2SDavid Sterba static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) 780bd989ba3SJan Schmidt { 7815de865eeSFilipe David Borba Manana struct tree_mod_elem **tm_list = NULL; 7825de865eeSFilipe David Borba Manana int nritems = 0; 7835de865eeSFilipe David Borba Manana int i; 7845de865eeSFilipe David Borba Manana int ret = 0; 7855de865eeSFilipe David Borba Manana 7865de865eeSFilipe David Borba Manana if (btrfs_header_level(eb) == 0) 7875de865eeSFilipe David Borba Manana return 0; 7885de865eeSFilipe David Borba Manana 789db7279a2SDavid Sterba if (!tree_mod_need_log(eb->fs_info, NULL)) 7905de865eeSFilipe David Borba Manana return 0; 7915de865eeSFilipe David Borba Manana 7925de865eeSFilipe David Borba Manana nritems = btrfs_header_nritems(eb); 79331e818feSDavid Sterba tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); 7945de865eeSFilipe David Borba Manana if (!tm_list) 7955de865eeSFilipe David Borba Manana return -ENOMEM; 7965de865eeSFilipe David Borba Manana 7975de865eeSFilipe David Borba Manana for (i = 0; i < nritems; i++) { 7985de865eeSFilipe David Borba Manana tm_list[i] = alloc_tree_mod_elem(eb, i, 7995de865eeSFilipe David Borba Manana MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); 8005de865eeSFilipe David Borba Manana if (!tm_list[i]) { 8015de865eeSFilipe David Borba Manana ret = -ENOMEM; 8025de865eeSFilipe David Borba Manana goto free_tms; 8035de865eeSFilipe David Borba Manana } 8045de865eeSFilipe David Borba Manana } 8055de865eeSFilipe David Borba Manana 806db7279a2SDavid Sterba if (tree_mod_dont_log(eb->fs_info, eb)) 8075de865eeSFilipe David Borba Manana goto free_tms; 8085de865eeSFilipe David Borba Manana 809db7279a2SDavid Sterba ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); 810b1a09f1eSDavid Sterba write_unlock(&eb->fs_info->tree_mod_log_lock); 8115de865eeSFilipe David Borba Manana if (ret) 8125de865eeSFilipe David Borba Manana goto free_tms; 8135de865eeSFilipe David Borba Manana kfree(tm_list); 8145de865eeSFilipe David Borba Manana 8155de865eeSFilipe David Borba Manana return 0; 8165de865eeSFilipe David Borba Manana 8175de865eeSFilipe David Borba Manana free_tms: 8185de865eeSFilipe David Borba Manana for (i = 0; i < nritems; i++) 8195de865eeSFilipe David Borba Manana kfree(tm_list[i]); 8205de865eeSFilipe David Borba Manana kfree(tm_list); 8215de865eeSFilipe David Borba Manana 8225de865eeSFilipe David Borba Manana return ret; 823bd989ba3SJan Schmidt } 824bd989ba3SJan Schmidt 825d352ac68SChris Mason /* 8265d4f98a2SYan Zheng * check if the tree block can be shared by multiple trees 8275d4f98a2SYan Zheng */ 8285d4f98a2SYan Zheng int btrfs_block_can_be_shared(struct btrfs_root *root, 8295d4f98a2SYan Zheng struct extent_buffer *buf) 8305d4f98a2SYan Zheng { 8315d4f98a2SYan Zheng /* 83292a7cc42SQu Wenruo * Tree blocks not in shareable trees and tree roots are never shared. 83392a7cc42SQu Wenruo * If a block was allocated after the last snapshot and the block was 83492a7cc42SQu Wenruo * not allocated by tree relocation, we know the block is not shared. 8355d4f98a2SYan Zheng */ 83692a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 8375d4f98a2SYan Zheng buf != root->node && buf != root->commit_root && 8385d4f98a2SYan Zheng (btrfs_header_generation(buf) <= 8395d4f98a2SYan Zheng btrfs_root_last_snapshot(&root->root_item) || 8405d4f98a2SYan Zheng btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) 8415d4f98a2SYan Zheng return 1; 842a79865c6SNikolay Borisov 8435d4f98a2SYan Zheng return 0; 8445d4f98a2SYan Zheng } 8455d4f98a2SYan Zheng 8465d4f98a2SYan Zheng static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, 8475d4f98a2SYan Zheng struct btrfs_root *root, 8485d4f98a2SYan Zheng struct extent_buffer *buf, 849f0486c68SYan, Zheng struct extent_buffer *cow, 850f0486c68SYan, Zheng int *last_ref) 8515d4f98a2SYan Zheng { 8520b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 8535d4f98a2SYan Zheng u64 refs; 8545d4f98a2SYan Zheng u64 owner; 8555d4f98a2SYan Zheng u64 flags; 8565d4f98a2SYan Zheng u64 new_flags = 0; 8575d4f98a2SYan Zheng int ret; 8585d4f98a2SYan Zheng 8595d4f98a2SYan Zheng /* 8605d4f98a2SYan Zheng * Backrefs update rules: 8615d4f98a2SYan Zheng * 8625d4f98a2SYan Zheng * Always use full backrefs for extent pointers in tree block 8635d4f98a2SYan Zheng * allocated by tree relocation. 8645d4f98a2SYan Zheng * 8655d4f98a2SYan Zheng * If a shared tree block is no longer referenced by its owner 8665d4f98a2SYan Zheng * tree (btrfs_header_owner(buf) == root->root_key.objectid), 8675d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 8685d4f98a2SYan Zheng * 8695d4f98a2SYan Zheng * If a tree block is been relocating 8705d4f98a2SYan Zheng * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), 8715d4f98a2SYan Zheng * use full backrefs for extent pointers in tree block. 8725d4f98a2SYan Zheng * The reason for this is some operations (such as drop tree) 8735d4f98a2SYan Zheng * are only allowed for blocks use full backrefs. 8745d4f98a2SYan Zheng */ 8755d4f98a2SYan Zheng 8765d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, buf)) { 8772ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, 8783173a18fSJosef Bacik btrfs_header_level(buf), 1, 8793173a18fSJosef Bacik &refs, &flags); 880be1a5564SMark Fasheh if (ret) 881be1a5564SMark Fasheh return ret; 882e5df9573SMark Fasheh if (refs == 0) { 883e5df9573SMark Fasheh ret = -EROFS; 8840b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 885e5df9573SMark Fasheh return ret; 886e5df9573SMark Fasheh } 8875d4f98a2SYan Zheng } else { 8885d4f98a2SYan Zheng refs = 1; 8895d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 8905d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 8915d4f98a2SYan Zheng flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; 8925d4f98a2SYan Zheng else 8935d4f98a2SYan Zheng flags = 0; 8945d4f98a2SYan Zheng } 8955d4f98a2SYan Zheng 8965d4f98a2SYan Zheng owner = btrfs_header_owner(buf); 8975d4f98a2SYan Zheng BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && 8985d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); 8995d4f98a2SYan Zheng 9005d4f98a2SYan Zheng if (refs > 1) { 9015d4f98a2SYan Zheng if ((owner == root->root_key.objectid || 9025d4f98a2SYan Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && 9035d4f98a2SYan Zheng !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { 904e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, buf, 1); 905692826b2SJeff Mahoney if (ret) 906692826b2SJeff Mahoney return ret; 9075d4f98a2SYan Zheng 9085d4f98a2SYan Zheng if (root->root_key.objectid == 9095d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) { 910e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 0); 911692826b2SJeff Mahoney if (ret) 912692826b2SJeff Mahoney return ret; 913e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 914692826b2SJeff Mahoney if (ret) 915692826b2SJeff Mahoney return ret; 9165d4f98a2SYan Zheng } 9175d4f98a2SYan Zheng new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; 9185d4f98a2SYan Zheng } else { 9195d4f98a2SYan Zheng 9205d4f98a2SYan Zheng if (root->root_key.objectid == 9215d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 922e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 9235d4f98a2SYan Zheng else 924e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 925692826b2SJeff Mahoney if (ret) 926692826b2SJeff Mahoney return ret; 9275d4f98a2SYan Zheng } 9285d4f98a2SYan Zheng if (new_flags != 0) { 929b1c79e09SJosef Bacik int level = btrfs_header_level(buf); 930b1c79e09SJosef Bacik 93142c9d0b5SDavid Sterba ret = btrfs_set_disk_extent_flags(trans, buf, 932b1c79e09SJosef Bacik new_flags, level, 0); 933be1a5564SMark Fasheh if (ret) 934be1a5564SMark Fasheh return ret; 9355d4f98a2SYan Zheng } 9365d4f98a2SYan Zheng } else { 9375d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { 9385d4f98a2SYan Zheng if (root->root_key.objectid == 9395d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID) 940e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 1); 9415d4f98a2SYan Zheng else 942e339a6b0SJosef Bacik ret = btrfs_inc_ref(trans, root, cow, 0); 943692826b2SJeff Mahoney if (ret) 944692826b2SJeff Mahoney return ret; 945e339a6b0SJosef Bacik ret = btrfs_dec_ref(trans, root, buf, 1); 946692826b2SJeff Mahoney if (ret) 947692826b2SJeff Mahoney return ret; 9485d4f98a2SYan Zheng } 9496a884d7dSDavid Sterba btrfs_clean_tree_block(buf); 950f0486c68SYan, Zheng *last_ref = 1; 9515d4f98a2SYan Zheng } 9525d4f98a2SYan Zheng return 0; 9535d4f98a2SYan Zheng } 9545d4f98a2SYan Zheng 955a6279470SFilipe Manana static struct extent_buffer *alloc_tree_block_no_bg_flush( 956a6279470SFilipe Manana struct btrfs_trans_handle *trans, 957a6279470SFilipe Manana struct btrfs_root *root, 958a6279470SFilipe Manana u64 parent_start, 959a6279470SFilipe Manana const struct btrfs_disk_key *disk_key, 960a6279470SFilipe Manana int level, 961a6279470SFilipe Manana u64 hint, 9629631e4ccSJosef Bacik u64 empty_size, 9639631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 964a6279470SFilipe Manana { 965a6279470SFilipe Manana struct btrfs_fs_info *fs_info = root->fs_info; 966a6279470SFilipe Manana struct extent_buffer *ret; 967a6279470SFilipe Manana 968a6279470SFilipe Manana /* 969a6279470SFilipe Manana * If we are COWing a node/leaf from the extent, chunk, device or free 970a6279470SFilipe Manana * space trees, make sure that we do not finish block group creation of 971a6279470SFilipe Manana * pending block groups. We do this to avoid a deadlock. 972a6279470SFilipe Manana * COWing can result in allocation of a new chunk, and flushing pending 973a6279470SFilipe Manana * block groups (btrfs_create_pending_block_groups()) can be triggered 974a6279470SFilipe Manana * when finishing allocation of a new chunk. Creation of a pending block 975a6279470SFilipe Manana * group modifies the extent, chunk, device and free space trees, 976a6279470SFilipe Manana * therefore we could deadlock with ourselves since we are holding a 977a6279470SFilipe Manana * lock on an extent buffer that btrfs_create_pending_block_groups() may 978a6279470SFilipe Manana * try to COW later. 979a6279470SFilipe Manana * For similar reasons, we also need to delay flushing pending block 980a6279470SFilipe Manana * groups when splitting a leaf or node, from one of those trees, since 981a6279470SFilipe Manana * we are holding a write lock on it and its parent or when inserting a 982a6279470SFilipe Manana * new root node for one of those trees. 983a6279470SFilipe Manana */ 984a6279470SFilipe Manana if (root == fs_info->extent_root || 985a6279470SFilipe Manana root == fs_info->chunk_root || 986a6279470SFilipe Manana root == fs_info->dev_root || 987a6279470SFilipe Manana root == fs_info->free_space_root) 988a6279470SFilipe Manana trans->can_flush_pending_bgs = false; 989a6279470SFilipe Manana 990a6279470SFilipe Manana ret = btrfs_alloc_tree_block(trans, root, parent_start, 991a6279470SFilipe Manana root->root_key.objectid, disk_key, level, 9929631e4ccSJosef Bacik hint, empty_size, nest); 993a6279470SFilipe Manana trans->can_flush_pending_bgs = true; 994a6279470SFilipe Manana 995a6279470SFilipe Manana return ret; 996a6279470SFilipe Manana } 997a6279470SFilipe Manana 9985d4f98a2SYan Zheng /* 999d397712bSChris Mason * does the dirty work in cow of a single block. The parent block (if 1000d397712bSChris Mason * supplied) is updated to point to the new cow copy. The new buffer is marked 1001d397712bSChris Mason * dirty and returned locked. If you modify the block it needs to be marked 1002d397712bSChris Mason * dirty again. 1003d352ac68SChris Mason * 1004d352ac68SChris Mason * search_start -- an allocation hint for the new block 1005d352ac68SChris Mason * 1006d397712bSChris Mason * empty_size -- a hint that you plan on doing more cow. This is the size in 1007d397712bSChris Mason * bytes the allocator should try to find free next to the block it returns. 1008d397712bSChris Mason * This is just a hint and may be ignored by the allocator. 1009d352ac68SChris Mason */ 1010d397712bSChris Mason static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, 10115f39d397SChris Mason struct btrfs_root *root, 10125f39d397SChris Mason struct extent_buffer *buf, 10135f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 10145f39d397SChris Mason struct extent_buffer **cow_ret, 10159631e4ccSJosef Bacik u64 search_start, u64 empty_size, 10169631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 10176702ed49SChris Mason { 10180b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 10195d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 10205f39d397SChris Mason struct extent_buffer *cow; 1021be1a5564SMark Fasheh int level, ret; 1022f0486c68SYan, Zheng int last_ref = 0; 1023925baeddSChris Mason int unlock_orig = 0; 10240f5053ebSGoldwyn Rodrigues u64 parent_start = 0; 10256702ed49SChris Mason 1026925baeddSChris Mason if (*cow_ret == buf) 1027925baeddSChris Mason unlock_orig = 1; 1028925baeddSChris Mason 1029b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1030925baeddSChris Mason 103192a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 10320b246afaSJeff Mahoney trans->transid != fs_info->running_transaction->transid); 103392a7cc42SQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 103427cdeb70SMiao Xie trans->transid != root->last_trans); 10355f39d397SChris Mason 10367bb86316SChris Mason level = btrfs_header_level(buf); 103731840ae1SZheng Yan 10385d4f98a2SYan Zheng if (level == 0) 10395d4f98a2SYan Zheng btrfs_item_key(buf, &disk_key, 0); 10405d4f98a2SYan Zheng else 10415d4f98a2SYan Zheng btrfs_node_key(buf, &disk_key, 0); 10425d4f98a2SYan Zheng 10430f5053ebSGoldwyn Rodrigues if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) 10445d4f98a2SYan Zheng parent_start = parent->start; 10455d4f98a2SYan Zheng 1046a6279470SFilipe Manana cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key, 10479631e4ccSJosef Bacik level, search_start, empty_size, nest); 10486702ed49SChris Mason if (IS_ERR(cow)) 10496702ed49SChris Mason return PTR_ERR(cow); 10506702ed49SChris Mason 1051b4ce94deSChris Mason /* cow is set to blocking by btrfs_init_new_buffer */ 1052b4ce94deSChris Mason 105358e8012cSDavid Sterba copy_extent_buffer_full(cow, buf); 1054db94535dSChris Mason btrfs_set_header_bytenr(cow, cow->start); 10555f39d397SChris Mason btrfs_set_header_generation(cow, trans->transid); 10565d4f98a2SYan Zheng btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 10575d4f98a2SYan Zheng btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 10585d4f98a2SYan Zheng BTRFS_HEADER_FLAG_RELOC); 10595d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 10605d4f98a2SYan Zheng btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 10615d4f98a2SYan Zheng else 10625f39d397SChris Mason btrfs_set_header_owner(cow, root->root_key.objectid); 10636702ed49SChris Mason 1064de37aa51SNikolay Borisov write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 10652b82032cSYan Zheng 1066be1a5564SMark Fasheh ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); 1067b68dc2a9SMark Fasheh if (ret) { 1068572c83acSJosef Bacik btrfs_tree_unlock(cow); 1069572c83acSJosef Bacik free_extent_buffer(cow); 107066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1071b68dc2a9SMark Fasheh return ret; 1072b68dc2a9SMark Fasheh } 10731a40e23bSZheng Yan 107492a7cc42SQu Wenruo if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 107583d4cfd4SJosef Bacik ret = btrfs_reloc_cow_block(trans, root, buf, cow); 107693314e3bSZhaolei if (ret) { 1077572c83acSJosef Bacik btrfs_tree_unlock(cow); 1078572c83acSJosef Bacik free_extent_buffer(cow); 107966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 108083d4cfd4SJosef Bacik return ret; 108183d4cfd4SJosef Bacik } 108293314e3bSZhaolei } 10833fd0a558SYan, Zheng 10846702ed49SChris Mason if (buf == root->node) { 1085925baeddSChris Mason WARN_ON(parent && parent != buf); 10865d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 10875d4f98a2SYan Zheng btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 10885d4f98a2SYan Zheng parent_start = buf->start; 1089925baeddSChris Mason 109067439dadSDavid Sterba atomic_inc(&cow->refs); 1091d9d19a01SDavid Sterba ret = tree_mod_log_insert_root(root->node, cow, 1); 1092d9d19a01SDavid Sterba BUG_ON(ret < 0); 1093240f62c8SChris Mason rcu_assign_pointer(root->node, cow); 1094925baeddSChris Mason 1095f0486c68SYan, Zheng btrfs_free_tree_block(trans, root, buf, parent_start, 10965581a51aSJan Schmidt last_ref); 10975f39d397SChris Mason free_extent_buffer(buf); 10980b86a832SChris Mason add_root_to_dirty_list(root); 10996702ed49SChris Mason } else { 11005d4f98a2SYan Zheng WARN_ON(trans->transid != btrfs_header_generation(parent)); 1101e09c2efeSDavid Sterba tree_mod_log_insert_key(parent, parent_slot, 1102c8cc6341SJosef Bacik MOD_LOG_KEY_REPLACE, GFP_NOFS); 11035f39d397SChris Mason btrfs_set_node_blockptr(parent, parent_slot, 1104db94535dSChris Mason cow->start); 110574493f7aSChris Mason btrfs_set_node_ptr_generation(parent, parent_slot, 110674493f7aSChris Mason trans->transid); 11076702ed49SChris Mason btrfs_mark_buffer_dirty(parent); 11085de865eeSFilipe David Borba Manana if (last_ref) { 1109db7279a2SDavid Sterba ret = tree_mod_log_free_eb(buf); 11105de865eeSFilipe David Borba Manana if (ret) { 1111572c83acSJosef Bacik btrfs_tree_unlock(cow); 1112572c83acSJosef Bacik free_extent_buffer(cow); 111366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 11145de865eeSFilipe David Borba Manana return ret; 11155de865eeSFilipe David Borba Manana } 11165de865eeSFilipe David Borba Manana } 1117f0486c68SYan, Zheng btrfs_free_tree_block(trans, root, buf, parent_start, 11185581a51aSJan Schmidt last_ref); 11196702ed49SChris Mason } 1120925baeddSChris Mason if (unlock_orig) 1121925baeddSChris Mason btrfs_tree_unlock(buf); 11223083ee2eSJosef Bacik free_extent_buffer_stale(buf); 11236702ed49SChris Mason btrfs_mark_buffer_dirty(cow); 11246702ed49SChris Mason *cow_ret = cow; 11256702ed49SChris Mason return 0; 11266702ed49SChris Mason } 11276702ed49SChris Mason 11285d9e75c4SJan Schmidt /* 11295d9e75c4SJan Schmidt * returns the logical address of the oldest predecessor of the given root. 11305d9e75c4SJan Schmidt * entries older than time_seq are ignored. 11315d9e75c4SJan Schmidt */ 1132bcd24dabSDavid Sterba static struct tree_mod_elem *__tree_mod_log_oldest_root( 113330b0463aSJan Schmidt struct extent_buffer *eb_root, u64 time_seq) 11345d9e75c4SJan Schmidt { 11355d9e75c4SJan Schmidt struct tree_mod_elem *tm; 11365d9e75c4SJan Schmidt struct tree_mod_elem *found = NULL; 113730b0463aSJan Schmidt u64 root_logical = eb_root->start; 11385d9e75c4SJan Schmidt int looped = 0; 11395d9e75c4SJan Schmidt 11405d9e75c4SJan Schmidt if (!time_seq) 114135a3621bSStefan Behrens return NULL; 11425d9e75c4SJan Schmidt 11435d9e75c4SJan Schmidt /* 1144298cfd36SChandan Rajendra * the very last operation that's logged for a root is the 1145298cfd36SChandan Rajendra * replacement operation (if it is replaced at all). this has 1146298cfd36SChandan Rajendra * the logical address of the *new* root, making it the very 1147298cfd36SChandan Rajendra * first operation that's logged for this root. 11485d9e75c4SJan Schmidt */ 11495d9e75c4SJan Schmidt while (1) { 1150bcd24dabSDavid Sterba tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, 11515d9e75c4SJan Schmidt time_seq); 11525d9e75c4SJan Schmidt if (!looped && !tm) 115335a3621bSStefan Behrens return NULL; 11545d9e75c4SJan Schmidt /* 115528da9fb4SJan Schmidt * if there are no tree operation for the oldest root, we simply 115628da9fb4SJan Schmidt * return it. this should only happen if that (old) root is at 115728da9fb4SJan Schmidt * level 0. 11585d9e75c4SJan Schmidt */ 115928da9fb4SJan Schmidt if (!tm) 116028da9fb4SJan Schmidt break; 11615d9e75c4SJan Schmidt 116228da9fb4SJan Schmidt /* 116328da9fb4SJan Schmidt * if there's an operation that's not a root replacement, we 116428da9fb4SJan Schmidt * found the oldest version of our root. normally, we'll find a 116528da9fb4SJan Schmidt * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. 116628da9fb4SJan Schmidt */ 11675d9e75c4SJan Schmidt if (tm->op != MOD_LOG_ROOT_REPLACE) 11685d9e75c4SJan Schmidt break; 11695d9e75c4SJan Schmidt 11705d9e75c4SJan Schmidt found = tm; 11715d9e75c4SJan Schmidt root_logical = tm->old_root.logical; 11725d9e75c4SJan Schmidt looped = 1; 11735d9e75c4SJan Schmidt } 11745d9e75c4SJan Schmidt 1175a95236d9SJan Schmidt /* if there's no old root to return, return what we found instead */ 1176a95236d9SJan Schmidt if (!found) 1177a95236d9SJan Schmidt found = tm; 1178a95236d9SJan Schmidt 11795d9e75c4SJan Schmidt return found; 11805d9e75c4SJan Schmidt } 11815d9e75c4SJan Schmidt 11825d9e75c4SJan Schmidt /* 11835d9e75c4SJan Schmidt * tm is a pointer to the first operation to rewind within eb. then, all 118401327610SNicholas D Steeves * previous operations will be rewound (until we reach something older than 11855d9e75c4SJan Schmidt * time_seq). 11865d9e75c4SJan Schmidt */ 11875d9e75c4SJan Schmidt static void 1188f1ca7e98SJosef Bacik __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, 1189f1ca7e98SJosef Bacik u64 time_seq, struct tree_mod_elem *first_tm) 11905d9e75c4SJan Schmidt { 11915d9e75c4SJan Schmidt u32 n; 11925d9e75c4SJan Schmidt struct rb_node *next; 11935d9e75c4SJan Schmidt struct tree_mod_elem *tm = first_tm; 11945d9e75c4SJan Schmidt unsigned long o_dst; 11955d9e75c4SJan Schmidt unsigned long o_src; 11965d9e75c4SJan Schmidt unsigned long p_size = sizeof(struct btrfs_key_ptr); 11975d9e75c4SJan Schmidt 11985d9e75c4SJan Schmidt n = btrfs_header_nritems(eb); 1199b1a09f1eSDavid Sterba read_lock(&fs_info->tree_mod_log_lock); 1200097b8a7cSJan Schmidt while (tm && tm->seq >= time_seq) { 12015d9e75c4SJan Schmidt /* 12025d9e75c4SJan Schmidt * all the operations are recorded with the operator used for 12035d9e75c4SJan Schmidt * the modification. as we're going backwards, we do the 12045d9e75c4SJan Schmidt * opposite of each operation here. 12055d9e75c4SJan Schmidt */ 12065d9e75c4SJan Schmidt switch (tm->op) { 12075d9e75c4SJan Schmidt case MOD_LOG_KEY_REMOVE_WHILE_FREEING: 12085d9e75c4SJan Schmidt BUG_ON(tm->slot < n); 1209c730ae0cSMarcos Paulo de Souza fallthrough; 121095c80bb1SLiu Bo case MOD_LOG_KEY_REMOVE_WHILE_MOVING: 12114c3e6969SChris Mason case MOD_LOG_KEY_REMOVE: 12125d9e75c4SJan Schmidt btrfs_set_node_key(eb, &tm->key, tm->slot); 12135d9e75c4SJan Schmidt btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); 12145d9e75c4SJan Schmidt btrfs_set_node_ptr_generation(eb, tm->slot, 12155d9e75c4SJan Schmidt tm->generation); 12164c3e6969SChris Mason n++; 12175d9e75c4SJan Schmidt break; 12185d9e75c4SJan Schmidt case MOD_LOG_KEY_REPLACE: 12195d9e75c4SJan Schmidt BUG_ON(tm->slot >= n); 12205d9e75c4SJan Schmidt btrfs_set_node_key(eb, &tm->key, tm->slot); 12215d9e75c4SJan Schmidt btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); 12225d9e75c4SJan Schmidt btrfs_set_node_ptr_generation(eb, tm->slot, 12235d9e75c4SJan Schmidt tm->generation); 12245d9e75c4SJan Schmidt break; 12255d9e75c4SJan Schmidt case MOD_LOG_KEY_ADD: 122619956c7eSJan Schmidt /* if a move operation is needed it's in the log */ 12275d9e75c4SJan Schmidt n--; 12285d9e75c4SJan Schmidt break; 12295d9e75c4SJan Schmidt case MOD_LOG_MOVE_KEYS: 1230c3193108SJan Schmidt o_dst = btrfs_node_key_ptr_offset(tm->slot); 1231c3193108SJan Schmidt o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); 1232c3193108SJan Schmidt memmove_extent_buffer(eb, o_dst, o_src, 12335d9e75c4SJan Schmidt tm->move.nr_items * p_size); 12345d9e75c4SJan Schmidt break; 12355d9e75c4SJan Schmidt case MOD_LOG_ROOT_REPLACE: 12365d9e75c4SJan Schmidt /* 12375d9e75c4SJan Schmidt * this operation is special. for roots, this must be 12385d9e75c4SJan Schmidt * handled explicitly before rewinding. 12395d9e75c4SJan Schmidt * for non-roots, this operation may exist if the node 12405d9e75c4SJan Schmidt * was a root: root A -> child B; then A gets empty and 12415d9e75c4SJan Schmidt * B is promoted to the new root. in the mod log, we'll 12425d9e75c4SJan Schmidt * have a root-replace operation for B, a tree block 12435d9e75c4SJan Schmidt * that is no root. we simply ignore that operation. 12445d9e75c4SJan Schmidt */ 12455d9e75c4SJan Schmidt break; 12465d9e75c4SJan Schmidt } 12475d9e75c4SJan Schmidt next = rb_next(&tm->node); 12485d9e75c4SJan Schmidt if (!next) 12495d9e75c4SJan Schmidt break; 12506b4df8b6SGeliang Tang tm = rb_entry(next, struct tree_mod_elem, node); 1251298cfd36SChandan Rajendra if (tm->logical != first_tm->logical) 12525d9e75c4SJan Schmidt break; 12535d9e75c4SJan Schmidt } 1254b1a09f1eSDavid Sterba read_unlock(&fs_info->tree_mod_log_lock); 12555d9e75c4SJan Schmidt btrfs_set_header_nritems(eb, n); 12565d9e75c4SJan Schmidt } 12575d9e75c4SJan Schmidt 125847fb091fSJan Schmidt /* 125901327610SNicholas D Steeves * Called with eb read locked. If the buffer cannot be rewound, the same buffer 126047fb091fSJan Schmidt * is returned. If rewind operations happen, a fresh buffer is returned. The 126147fb091fSJan Schmidt * returned buffer is always read-locked. If the returned buffer is not the 126247fb091fSJan Schmidt * input buffer, the lock on the input buffer is released and the input buffer 126347fb091fSJan Schmidt * is freed (its refcount is decremented). 126447fb091fSJan Schmidt */ 12655d9e75c4SJan Schmidt static struct extent_buffer * 12669ec72677SJosef Bacik tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, 12679ec72677SJosef Bacik struct extent_buffer *eb, u64 time_seq) 12685d9e75c4SJan Schmidt { 12695d9e75c4SJan Schmidt struct extent_buffer *eb_rewin; 12705d9e75c4SJan Schmidt struct tree_mod_elem *tm; 12715d9e75c4SJan Schmidt 12725d9e75c4SJan Schmidt if (!time_seq) 12735d9e75c4SJan Schmidt return eb; 12745d9e75c4SJan Schmidt 12755d9e75c4SJan Schmidt if (btrfs_header_level(eb) == 0) 12765d9e75c4SJan Schmidt return eb; 12775d9e75c4SJan Schmidt 12785d9e75c4SJan Schmidt tm = tree_mod_log_search(fs_info, eb->start, time_seq); 12795d9e75c4SJan Schmidt if (!tm) 12805d9e75c4SJan Schmidt return eb; 12815d9e75c4SJan Schmidt 12825d9e75c4SJan Schmidt if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { 12835d9e75c4SJan Schmidt BUG_ON(tm->slot != 0); 1284da17066cSJeff Mahoney eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); 1285db7f3436SJosef Bacik if (!eb_rewin) { 1286ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb); 1287db7f3436SJosef Bacik free_extent_buffer(eb); 1288db7f3436SJosef Bacik return NULL; 1289db7f3436SJosef Bacik } 12905d9e75c4SJan Schmidt btrfs_set_header_bytenr(eb_rewin, eb->start); 12915d9e75c4SJan Schmidt btrfs_set_header_backref_rev(eb_rewin, 12925d9e75c4SJan Schmidt btrfs_header_backref_rev(eb)); 12935d9e75c4SJan Schmidt btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); 1294c3193108SJan Schmidt btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); 12955d9e75c4SJan Schmidt } else { 12965d9e75c4SJan Schmidt eb_rewin = btrfs_clone_extent_buffer(eb); 1297db7f3436SJosef Bacik if (!eb_rewin) { 1298ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb); 1299db7f3436SJosef Bacik free_extent_buffer(eb); 1300db7f3436SJosef Bacik return NULL; 1301db7f3436SJosef Bacik } 13025d9e75c4SJan Schmidt } 13035d9e75c4SJan Schmidt 1304ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb); 13055d9e75c4SJan Schmidt free_extent_buffer(eb); 13065d9e75c4SJan Schmidt 1307d3beaa25SJosef Bacik btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb_rewin), 1308d3beaa25SJosef Bacik eb_rewin, btrfs_header_level(eb_rewin)); 130947fb091fSJan Schmidt btrfs_tree_read_lock(eb_rewin); 1310f1ca7e98SJosef Bacik __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); 131157911b8bSJan Schmidt WARN_ON(btrfs_header_nritems(eb_rewin) > 1312da17066cSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info)); 13135d9e75c4SJan Schmidt 13145d9e75c4SJan Schmidt return eb_rewin; 13155d9e75c4SJan Schmidt } 13165d9e75c4SJan Schmidt 13178ba97a15SJan Schmidt /* 13188ba97a15SJan Schmidt * get_old_root() rewinds the state of @root's root node to the given @time_seq 13198ba97a15SJan Schmidt * value. If there are no changes, the current root->root_node is returned. If 13208ba97a15SJan Schmidt * anything changed in between, there's a fresh buffer allocated on which the 13218ba97a15SJan Schmidt * rewind operations are done. In any case, the returned buffer is read locked. 13228ba97a15SJan Schmidt * Returns NULL on error (with no locks held). 13238ba97a15SJan Schmidt */ 13245d9e75c4SJan Schmidt static inline struct extent_buffer * 13255d9e75c4SJan Schmidt get_old_root(struct btrfs_root *root, u64 time_seq) 13265d9e75c4SJan Schmidt { 13270b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13285d9e75c4SJan Schmidt struct tree_mod_elem *tm; 132930b0463aSJan Schmidt struct extent_buffer *eb = NULL; 133030b0463aSJan Schmidt struct extent_buffer *eb_root; 1331efad8a85SFilipe Manana u64 eb_root_owner = 0; 13327bfdcf7fSLiu Bo struct extent_buffer *old; 1333a95236d9SJan Schmidt struct tree_mod_root *old_root = NULL; 13344325edd0SChris Mason u64 old_generation = 0; 1335a95236d9SJan Schmidt u64 logical; 1336581c1760SQu Wenruo int level; 13375d9e75c4SJan Schmidt 133830b0463aSJan Schmidt eb_root = btrfs_read_lock_root_node(root); 1339bcd24dabSDavid Sterba tm = __tree_mod_log_oldest_root(eb_root, time_seq); 13405d9e75c4SJan Schmidt if (!tm) 134130b0463aSJan Schmidt return eb_root; 13425d9e75c4SJan Schmidt 1343a95236d9SJan Schmidt if (tm->op == MOD_LOG_ROOT_REPLACE) { 13445d9e75c4SJan Schmidt old_root = &tm->old_root; 13455d9e75c4SJan Schmidt old_generation = tm->generation; 1346a95236d9SJan Schmidt logical = old_root->logical; 1347581c1760SQu Wenruo level = old_root->level; 1348a95236d9SJan Schmidt } else { 134930b0463aSJan Schmidt logical = eb_root->start; 1350581c1760SQu Wenruo level = btrfs_header_level(eb_root); 1351a95236d9SJan Schmidt } 13525d9e75c4SJan Schmidt 13530b246afaSJeff Mahoney tm = tree_mod_log_search(fs_info, logical, time_seq); 1354834328a8SJan Schmidt if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { 135530b0463aSJan Schmidt btrfs_tree_read_unlock(eb_root); 135630b0463aSJan Schmidt free_extent_buffer(eb_root); 13571b7ec85eSJosef Bacik old = read_tree_block(fs_info, logical, root->root_key.objectid, 13581b7ec85eSJosef Bacik 0, level, NULL); 135964c043deSLiu Bo if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { 136064c043deSLiu Bo if (!IS_ERR(old)) 1361416bc658SJosef Bacik free_extent_buffer(old); 13620b246afaSJeff Mahoney btrfs_warn(fs_info, 13630b246afaSJeff Mahoney "failed to read tree block %llu from get_old_root", 13640b246afaSJeff Mahoney logical); 1365834328a8SJan Schmidt } else { 13667bfdcf7fSLiu Bo eb = btrfs_clone_extent_buffer(old); 13677bfdcf7fSLiu Bo free_extent_buffer(old); 1368834328a8SJan Schmidt } 1369834328a8SJan Schmidt } else if (old_root) { 1370efad8a85SFilipe Manana eb_root_owner = btrfs_header_owner(eb_root); 137130b0463aSJan Schmidt btrfs_tree_read_unlock(eb_root); 137230b0463aSJan Schmidt free_extent_buffer(eb_root); 13730b246afaSJeff Mahoney eb = alloc_dummy_extent_buffer(fs_info, logical); 1374834328a8SJan Schmidt } else { 137530b0463aSJan Schmidt eb = btrfs_clone_extent_buffer(eb_root); 1376ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb_root); 137730b0463aSJan Schmidt free_extent_buffer(eb_root); 1378834328a8SJan Schmidt } 1379834328a8SJan Schmidt 13808ba97a15SJan Schmidt if (!eb) 13818ba97a15SJan Schmidt return NULL; 1382a95236d9SJan Schmidt if (old_root) { 13835d9e75c4SJan Schmidt btrfs_set_header_bytenr(eb, eb->start); 13845d9e75c4SJan Schmidt btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); 1385efad8a85SFilipe Manana btrfs_set_header_owner(eb, eb_root_owner); 13865d9e75c4SJan Schmidt btrfs_set_header_level(eb, old_root->level); 13875d9e75c4SJan Schmidt btrfs_set_header_generation(eb, old_generation); 1388a95236d9SJan Schmidt } 1389d3beaa25SJosef Bacik btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), eb, 1390d3beaa25SJosef Bacik btrfs_header_level(eb)); 1391d3beaa25SJosef Bacik btrfs_tree_read_lock(eb); 139228da9fb4SJan Schmidt if (tm) 13930b246afaSJeff Mahoney __tree_mod_log_rewind(fs_info, eb, time_seq, tm); 139428da9fb4SJan Schmidt else 139528da9fb4SJan Schmidt WARN_ON(btrfs_header_level(eb) != 0); 13960b246afaSJeff Mahoney WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); 13975d9e75c4SJan Schmidt 13985d9e75c4SJan Schmidt return eb; 13995d9e75c4SJan Schmidt } 14005d9e75c4SJan Schmidt 14015b6602e7SJan Schmidt int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) 14025b6602e7SJan Schmidt { 14035b6602e7SJan Schmidt struct tree_mod_elem *tm; 14045b6602e7SJan Schmidt int level; 140530b0463aSJan Schmidt struct extent_buffer *eb_root = btrfs_root_node(root); 14065b6602e7SJan Schmidt 1407bcd24dabSDavid Sterba tm = __tree_mod_log_oldest_root(eb_root, time_seq); 14085b6602e7SJan Schmidt if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { 14095b6602e7SJan Schmidt level = tm->old_root.level; 14105b6602e7SJan Schmidt } else { 141130b0463aSJan Schmidt level = btrfs_header_level(eb_root); 14125b6602e7SJan Schmidt } 141330b0463aSJan Schmidt free_extent_buffer(eb_root); 14145b6602e7SJan Schmidt 14155b6602e7SJan Schmidt return level; 14165b6602e7SJan Schmidt } 14175b6602e7SJan Schmidt 14185d4f98a2SYan Zheng static inline int should_cow_block(struct btrfs_trans_handle *trans, 14195d4f98a2SYan Zheng struct btrfs_root *root, 14205d4f98a2SYan Zheng struct extent_buffer *buf) 14215d4f98a2SYan Zheng { 1422f5ee5c9aSJeff Mahoney if (btrfs_is_testing(root->fs_info)) 1423faa2dbf0SJosef Bacik return 0; 1424fccb84c9SDavid Sterba 1425d1980131SDavid Sterba /* Ensure we can see the FORCE_COW bit */ 1426d1980131SDavid Sterba smp_mb__before_atomic(); 1427f1ebcc74SLiu Bo 1428f1ebcc74SLiu Bo /* 1429f1ebcc74SLiu Bo * We do not need to cow a block if 1430f1ebcc74SLiu Bo * 1) this block is not created or changed in this transaction; 1431f1ebcc74SLiu Bo * 2) this block does not belong to TREE_RELOC tree; 1432f1ebcc74SLiu Bo * 3) the root is not forced COW. 1433f1ebcc74SLiu Bo * 1434f1ebcc74SLiu Bo * What is forced COW: 143501327610SNicholas D Steeves * when we create snapshot during committing the transaction, 143652042d8eSAndrea Gelmini * after we've finished copying src root, we must COW the shared 1437f1ebcc74SLiu Bo * block to ensure the metadata consistency. 1438f1ebcc74SLiu Bo */ 14395d4f98a2SYan Zheng if (btrfs_header_generation(buf) == trans->transid && 14405d4f98a2SYan Zheng !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && 14415d4f98a2SYan Zheng !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 1442f1ebcc74SLiu Bo btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && 144327cdeb70SMiao Xie !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) 14445d4f98a2SYan Zheng return 0; 14455d4f98a2SYan Zheng return 1; 14465d4f98a2SYan Zheng } 14475d4f98a2SYan Zheng 1448d352ac68SChris Mason /* 1449d352ac68SChris Mason * cows a single block, see __btrfs_cow_block for the real work. 145001327610SNicholas D Steeves * This version of it has extra checks so that a block isn't COWed more than 1451d352ac68SChris Mason * once per transaction, as long as it hasn't been written yet 1452d352ac68SChris Mason */ 1453d397712bSChris Mason noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, 14545f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *buf, 14555f39d397SChris Mason struct extent_buffer *parent, int parent_slot, 14569631e4ccSJosef Bacik struct extent_buffer **cow_ret, 14579631e4ccSJosef Bacik enum btrfs_lock_nesting nest) 145802217ed2SChris Mason { 14590b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14606702ed49SChris Mason u64 search_start; 1461f510cfecSChris Mason int ret; 1462dc17ff8fSChris Mason 146383354f07SJosef Bacik if (test_bit(BTRFS_ROOT_DELETING, &root->state)) 146483354f07SJosef Bacik btrfs_err(fs_info, 146583354f07SJosef Bacik "COW'ing blocks on a fs root that's being dropped"); 146683354f07SJosef Bacik 14670b246afaSJeff Mahoney if (trans->transaction != fs_info->running_transaction) 146831b1a2bdSJulia Lawall WARN(1, KERN_CRIT "trans %llu running %llu\n", 1469c1c9ff7cSGeert Uytterhoeven trans->transid, 14700b246afaSJeff Mahoney fs_info->running_transaction->transid); 147131b1a2bdSJulia Lawall 14720b246afaSJeff Mahoney if (trans->transid != fs_info->generation) 147331b1a2bdSJulia Lawall WARN(1, KERN_CRIT "trans %llu running %llu\n", 14740b246afaSJeff Mahoney trans->transid, fs_info->generation); 1475dc17ff8fSChris Mason 14765d4f98a2SYan Zheng if (!should_cow_block(trans, root, buf)) { 147764c12921SJeff Mahoney trans->dirty = true; 147802217ed2SChris Mason *cow_ret = buf; 147902217ed2SChris Mason return 0; 148002217ed2SChris Mason } 1481c487685dSChris Mason 1482ee22184bSByongho Lee search_start = buf->start & ~((u64)SZ_1G - 1); 1483b4ce94deSChris Mason 1484f616f5cdSQu Wenruo /* 1485f616f5cdSQu Wenruo * Before CoWing this block for later modification, check if it's 1486f616f5cdSQu Wenruo * the subtree root and do the delayed subtree trace if needed. 1487f616f5cdSQu Wenruo * 1488f616f5cdSQu Wenruo * Also We don't care about the error, as it's handled internally. 1489f616f5cdSQu Wenruo */ 1490f616f5cdSQu Wenruo btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); 1491f510cfecSChris Mason ret = __btrfs_cow_block(trans, root, buf, parent, 14929631e4ccSJosef Bacik parent_slot, cow_ret, search_start, 0, nest); 14931abe9b8aSliubo 14941abe9b8aSliubo trace_btrfs_cow_block(root, buf, *cow_ret); 14951abe9b8aSliubo 1496f510cfecSChris Mason return ret; 14972c90e5d6SChris Mason } 1498f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); 14996702ed49SChris Mason 1500d352ac68SChris Mason /* 1501d352ac68SChris Mason * helper function for defrag to decide if two blocks pointed to by a 1502d352ac68SChris Mason * node are actually close by 1503d352ac68SChris Mason */ 15046b80053dSChris Mason static int close_blocks(u64 blocknr, u64 other, u32 blocksize) 15056702ed49SChris Mason { 15066b80053dSChris Mason if (blocknr < other && other - (blocknr + blocksize) < 32768) 15076702ed49SChris Mason return 1; 15086b80053dSChris Mason if (blocknr > other && blocknr - (other + blocksize) < 32768) 15096702ed49SChris Mason return 1; 151002217ed2SChris Mason return 0; 151102217ed2SChris Mason } 151202217ed2SChris Mason 1513ce6ef5abSDavid Sterba #ifdef __LITTLE_ENDIAN 1514ce6ef5abSDavid Sterba 1515ce6ef5abSDavid Sterba /* 1516ce6ef5abSDavid Sterba * Compare two keys, on little-endian the disk order is same as CPU order and 1517ce6ef5abSDavid Sterba * we can avoid the conversion. 1518ce6ef5abSDavid Sterba */ 1519ce6ef5abSDavid Sterba static int comp_keys(const struct btrfs_disk_key *disk_key, 1520ce6ef5abSDavid Sterba const struct btrfs_key *k2) 1521ce6ef5abSDavid Sterba { 1522ce6ef5abSDavid Sterba const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; 1523ce6ef5abSDavid Sterba 1524ce6ef5abSDavid Sterba return btrfs_comp_cpu_keys(k1, k2); 1525ce6ef5abSDavid Sterba } 1526ce6ef5abSDavid Sterba 1527ce6ef5abSDavid Sterba #else 1528ce6ef5abSDavid Sterba 1529081e9573SChris Mason /* 1530081e9573SChris Mason * compare two keys in a memcmp fashion 1531081e9573SChris Mason */ 1532310712b2SOmar Sandoval static int comp_keys(const struct btrfs_disk_key *disk, 1533310712b2SOmar Sandoval const struct btrfs_key *k2) 1534081e9573SChris Mason { 1535081e9573SChris Mason struct btrfs_key k1; 1536081e9573SChris Mason 1537081e9573SChris Mason btrfs_disk_key_to_cpu(&k1, disk); 1538081e9573SChris Mason 153920736abaSDiego Calleja return btrfs_comp_cpu_keys(&k1, k2); 1540081e9573SChris Mason } 1541ce6ef5abSDavid Sterba #endif 1542081e9573SChris Mason 1543f3465ca4SJosef Bacik /* 1544f3465ca4SJosef Bacik * same as comp_keys only with two btrfs_key's 1545f3465ca4SJosef Bacik */ 1546e1f60a65SDavid Sterba int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) 1547f3465ca4SJosef Bacik { 1548f3465ca4SJosef Bacik if (k1->objectid > k2->objectid) 1549f3465ca4SJosef Bacik return 1; 1550f3465ca4SJosef Bacik if (k1->objectid < k2->objectid) 1551f3465ca4SJosef Bacik return -1; 1552f3465ca4SJosef Bacik if (k1->type > k2->type) 1553f3465ca4SJosef Bacik return 1; 1554f3465ca4SJosef Bacik if (k1->type < k2->type) 1555f3465ca4SJosef Bacik return -1; 1556f3465ca4SJosef Bacik if (k1->offset > k2->offset) 1557f3465ca4SJosef Bacik return 1; 1558f3465ca4SJosef Bacik if (k1->offset < k2->offset) 1559f3465ca4SJosef Bacik return -1; 1560f3465ca4SJosef Bacik return 0; 1561f3465ca4SJosef Bacik } 1562081e9573SChris Mason 1563d352ac68SChris Mason /* 1564d352ac68SChris Mason * this is used by the defrag code to go through all the 1565d352ac68SChris Mason * leaves pointed to by a node and reallocate them so that 1566d352ac68SChris Mason * disk order is close to key order 1567d352ac68SChris Mason */ 15686702ed49SChris Mason int btrfs_realloc_node(struct btrfs_trans_handle *trans, 15695f39d397SChris Mason struct btrfs_root *root, struct extent_buffer *parent, 1570de78b51aSEric Sandeen int start_slot, u64 *last_ret, 1571a6b6e75eSChris Mason struct btrfs_key *progress) 15726702ed49SChris Mason { 15730b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15746b80053dSChris Mason struct extent_buffer *cur; 15756702ed49SChris Mason u64 blocknr; 1576e9d0b13bSChris Mason u64 search_start = *last_ret; 1577e9d0b13bSChris Mason u64 last_block = 0; 15786702ed49SChris Mason u64 other; 15796702ed49SChris Mason u32 parent_nritems; 15806702ed49SChris Mason int end_slot; 15816702ed49SChris Mason int i; 15826702ed49SChris Mason int err = 0; 15836b80053dSChris Mason u32 blocksize; 1584081e9573SChris Mason int progress_passed = 0; 1585081e9573SChris Mason struct btrfs_disk_key disk_key; 15866702ed49SChris Mason 15870b246afaSJeff Mahoney WARN_ON(trans->transaction != fs_info->running_transaction); 15880b246afaSJeff Mahoney WARN_ON(trans->transid != fs_info->generation); 158986479a04SChris Mason 15906b80053dSChris Mason parent_nritems = btrfs_header_nritems(parent); 15910b246afaSJeff Mahoney blocksize = fs_info->nodesize; 15925dfe2be7SFilipe Manana end_slot = parent_nritems - 1; 15936702ed49SChris Mason 15945dfe2be7SFilipe Manana if (parent_nritems <= 1) 15956702ed49SChris Mason return 0; 15966702ed49SChris Mason 15975dfe2be7SFilipe Manana for (i = start_slot; i <= end_slot; i++) { 15986702ed49SChris Mason int close = 1; 1599a6b6e75eSChris Mason 1600081e9573SChris Mason btrfs_node_key(parent, &disk_key, i); 1601081e9573SChris Mason if (!progress_passed && comp_keys(&disk_key, progress) < 0) 1602081e9573SChris Mason continue; 1603081e9573SChris Mason 1604081e9573SChris Mason progress_passed = 1; 16056b80053dSChris Mason blocknr = btrfs_node_blockptr(parent, i); 1606e9d0b13bSChris Mason if (last_block == 0) 1607e9d0b13bSChris Mason last_block = blocknr; 16085708b959SChris Mason 16096702ed49SChris Mason if (i > 0) { 16106b80053dSChris Mason other = btrfs_node_blockptr(parent, i - 1); 16116b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 16126702ed49SChris Mason } 16135dfe2be7SFilipe Manana if (!close && i < end_slot) { 16146b80053dSChris Mason other = btrfs_node_blockptr(parent, i + 1); 16156b80053dSChris Mason close = close_blocks(blocknr, other, blocksize); 16166702ed49SChris Mason } 1617e9d0b13bSChris Mason if (close) { 1618e9d0b13bSChris Mason last_block = blocknr; 16196702ed49SChris Mason continue; 1620e9d0b13bSChris Mason } 16216702ed49SChris Mason 1622206983b7SJosef Bacik cur = btrfs_read_node_slot(parent, i); 1623206983b7SJosef Bacik if (IS_ERR(cur)) 162464c043deSLiu Bo return PTR_ERR(cur); 1625e9d0b13bSChris Mason if (search_start == 0) 16266b80053dSChris Mason search_start = last_block; 1627e9d0b13bSChris Mason 1628e7a84565SChris Mason btrfs_tree_lock(cur); 16296b80053dSChris Mason err = __btrfs_cow_block(trans, root, cur, parent, i, 1630e7a84565SChris Mason &cur, search_start, 16316b80053dSChris Mason min(16 * blocksize, 16329631e4ccSJosef Bacik (end_slot - i) * blocksize), 16339631e4ccSJosef Bacik BTRFS_NESTING_COW); 1634252c38f0SYan if (err) { 1635e7a84565SChris Mason btrfs_tree_unlock(cur); 16366b80053dSChris Mason free_extent_buffer(cur); 16376702ed49SChris Mason break; 1638252c38f0SYan } 1639e7a84565SChris Mason search_start = cur->start; 1640e7a84565SChris Mason last_block = cur->start; 1641f2183bdeSChris Mason *last_ret = search_start; 1642e7a84565SChris Mason btrfs_tree_unlock(cur); 1643e7a84565SChris Mason free_extent_buffer(cur); 16446702ed49SChris Mason } 16456702ed49SChris Mason return err; 16466702ed49SChris Mason } 16476702ed49SChris Mason 164874123bd7SChris Mason /* 16495f39d397SChris Mason * search for key in the extent_buffer. The items start at offset p, 16505f39d397SChris Mason * and they are item_size apart. There are 'max' items in p. 16515f39d397SChris Mason * 165274123bd7SChris Mason * the slot in the array is returned via slot, and it points to 165374123bd7SChris Mason * the place where you would insert key if it is not found in 165474123bd7SChris Mason * the array. 165574123bd7SChris Mason * 165674123bd7SChris Mason * slot may point to max if the key is bigger than all of the keys 165774123bd7SChris Mason */ 1658e02119d5SChris Mason static noinline int generic_bin_search(struct extent_buffer *eb, 1659310712b2SOmar Sandoval unsigned long p, int item_size, 1660310712b2SOmar Sandoval const struct btrfs_key *key, 1661be0e5c09SChris Mason int max, int *slot) 1662be0e5c09SChris Mason { 1663be0e5c09SChris Mason int low = 0; 1664be0e5c09SChris Mason int high = max; 1665be0e5c09SChris Mason int ret; 16665cd17f34SDavid Sterba const int key_size = sizeof(struct btrfs_disk_key); 1667be0e5c09SChris Mason 16685e24e9afSLiu Bo if (low > high) { 16695e24e9afSLiu Bo btrfs_err(eb->fs_info, 16705e24e9afSLiu Bo "%s: low (%d) > high (%d) eb %llu owner %llu level %d", 16715e24e9afSLiu Bo __func__, low, high, eb->start, 16725e24e9afSLiu Bo btrfs_header_owner(eb), btrfs_header_level(eb)); 16735e24e9afSLiu Bo return -EINVAL; 16745e24e9afSLiu Bo } 16755e24e9afSLiu Bo 1676be0e5c09SChris Mason while (low < high) { 16775cd17f34SDavid Sterba unsigned long oip; 16785cd17f34SDavid Sterba unsigned long offset; 16795cd17f34SDavid Sterba struct btrfs_disk_key *tmp; 16805cd17f34SDavid Sterba struct btrfs_disk_key unaligned; 16815cd17f34SDavid Sterba int mid; 16825cd17f34SDavid Sterba 1683be0e5c09SChris Mason mid = (low + high) / 2; 16845f39d397SChris Mason offset = p + mid * item_size; 16855cd17f34SDavid Sterba oip = offset_in_page(offset); 16865f39d397SChris Mason 16875cd17f34SDavid Sterba if (oip + key_size <= PAGE_SIZE) { 1688884b07d0SQu Wenruo const unsigned long idx = get_eb_page_index(offset); 16895cd17f34SDavid Sterba char *kaddr = page_address(eb->pages[idx]); 1690934d375bSChris Mason 1691884b07d0SQu Wenruo oip = get_eb_offset_in_page(eb, offset); 16925cd17f34SDavid Sterba tmp = (struct btrfs_disk_key *)(kaddr + oip); 16935cd17f34SDavid Sterba } else { 16945cd17f34SDavid Sterba read_extent_buffer(eb, &unaligned, offset, key_size); 16955f39d397SChris Mason tmp = &unaligned; 1696479965d6SChris Mason } 1697479965d6SChris Mason 1698be0e5c09SChris Mason ret = comp_keys(tmp, key); 1699be0e5c09SChris Mason 1700be0e5c09SChris Mason if (ret < 0) 1701be0e5c09SChris Mason low = mid + 1; 1702be0e5c09SChris Mason else if (ret > 0) 1703be0e5c09SChris Mason high = mid; 1704be0e5c09SChris Mason else { 1705be0e5c09SChris Mason *slot = mid; 1706be0e5c09SChris Mason return 0; 1707be0e5c09SChris Mason } 1708be0e5c09SChris Mason } 1709be0e5c09SChris Mason *slot = low; 1710be0e5c09SChris Mason return 1; 1711be0e5c09SChris Mason } 1712be0e5c09SChris Mason 171397571fd0SChris Mason /* 171497571fd0SChris Mason * simple bin_search frontend that does the right thing for 171597571fd0SChris Mason * leaves vs nodes 171697571fd0SChris Mason */ 1717a74b35ecSNikolay Borisov int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, 1718e3b83361SQu Wenruo int *slot) 1719be0e5c09SChris Mason { 1720e3b83361SQu Wenruo if (btrfs_header_level(eb) == 0) 17215f39d397SChris Mason return generic_bin_search(eb, 17225f39d397SChris Mason offsetof(struct btrfs_leaf, items), 17230783fcfcSChris Mason sizeof(struct btrfs_item), 17245f39d397SChris Mason key, btrfs_header_nritems(eb), 17257518a238SChris Mason slot); 1726f775738fSWang Sheng-Hui else 17275f39d397SChris Mason return generic_bin_search(eb, 17285f39d397SChris Mason offsetof(struct btrfs_node, ptrs), 1729123abc88SChris Mason sizeof(struct btrfs_key_ptr), 17305f39d397SChris Mason key, btrfs_header_nritems(eb), 17317518a238SChris Mason slot); 1732be0e5c09SChris Mason } 1733be0e5c09SChris Mason 1734f0486c68SYan, Zheng static void root_add_used(struct btrfs_root *root, u32 size) 1735f0486c68SYan, Zheng { 1736f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 1737f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 1738f0486c68SYan, Zheng btrfs_root_used(&root->root_item) + size); 1739f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 1740f0486c68SYan, Zheng } 1741f0486c68SYan, Zheng 1742f0486c68SYan, Zheng static void root_sub_used(struct btrfs_root *root, u32 size) 1743f0486c68SYan, Zheng { 1744f0486c68SYan, Zheng spin_lock(&root->accounting_lock); 1745f0486c68SYan, Zheng btrfs_set_root_used(&root->root_item, 1746f0486c68SYan, Zheng btrfs_root_used(&root->root_item) - size); 1747f0486c68SYan, Zheng spin_unlock(&root->accounting_lock); 1748f0486c68SYan, Zheng } 1749f0486c68SYan, Zheng 1750d352ac68SChris Mason /* given a node and slot number, this reads the blocks it points to. The 1751d352ac68SChris Mason * extent buffer is returned with a reference taken (but unlocked). 1752d352ac68SChris Mason */ 17534b231ae4SDavid Sterba struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, 17544b231ae4SDavid Sterba int slot) 1755bb803951SChris Mason { 1756ca7a79adSChris Mason int level = btrfs_header_level(parent); 1757416bc658SJosef Bacik struct extent_buffer *eb; 1758581c1760SQu Wenruo struct btrfs_key first_key; 1759416bc658SJosef Bacik 1760fb770ae4SLiu Bo if (slot < 0 || slot >= btrfs_header_nritems(parent)) 1761fb770ae4SLiu Bo return ERR_PTR(-ENOENT); 1762ca7a79adSChris Mason 1763ca7a79adSChris Mason BUG_ON(level == 0); 1764ca7a79adSChris Mason 1765581c1760SQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 1766d0d20b0fSDavid Sterba eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), 17671b7ec85eSJosef Bacik btrfs_header_owner(parent), 1768581c1760SQu Wenruo btrfs_node_ptr_generation(parent, slot), 1769581c1760SQu Wenruo level - 1, &first_key); 1770fb770ae4SLiu Bo if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { 1771416bc658SJosef Bacik free_extent_buffer(eb); 1772fb770ae4SLiu Bo eb = ERR_PTR(-EIO); 1773416bc658SJosef Bacik } 1774416bc658SJosef Bacik 1775416bc658SJosef Bacik return eb; 1776bb803951SChris Mason } 1777bb803951SChris Mason 1778d352ac68SChris Mason /* 1779d352ac68SChris Mason * node level balancing, used to make sure nodes are in proper order for 1780d352ac68SChris Mason * item deletion. We balance from the top down, so we have to make sure 1781d352ac68SChris Mason * that a deletion won't leave an node completely empty later on. 1782d352ac68SChris Mason */ 1783e02119d5SChris Mason static noinline int balance_level(struct btrfs_trans_handle *trans, 178498ed5174SChris Mason struct btrfs_root *root, 178598ed5174SChris Mason struct btrfs_path *path, int level) 1786bb803951SChris Mason { 17870b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 17885f39d397SChris Mason struct extent_buffer *right = NULL; 17895f39d397SChris Mason struct extent_buffer *mid; 17905f39d397SChris Mason struct extent_buffer *left = NULL; 17915f39d397SChris Mason struct extent_buffer *parent = NULL; 1792bb803951SChris Mason int ret = 0; 1793bb803951SChris Mason int wret; 1794bb803951SChris Mason int pslot; 1795bb803951SChris Mason int orig_slot = path->slots[level]; 179679f95c82SChris Mason u64 orig_ptr; 1797bb803951SChris Mason 179898e6b1ebSLiu Bo ASSERT(level > 0); 1799bb803951SChris Mason 18005f39d397SChris Mason mid = path->nodes[level]; 1801b4ce94deSChris Mason 1802ac5887c8SJosef Bacik WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); 18037bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 18047bb86316SChris Mason 18051d4f8a0cSChris Mason orig_ptr = btrfs_node_blockptr(mid, orig_slot); 180679f95c82SChris Mason 1807a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 18085f39d397SChris Mason parent = path->nodes[level + 1]; 1809bb803951SChris Mason pslot = path->slots[level + 1]; 1810a05a9bb1SLi Zefan } 1811bb803951SChris Mason 181240689478SChris Mason /* 181340689478SChris Mason * deal with the case where there is only one pointer in the root 181440689478SChris Mason * by promoting the node below to a root 181540689478SChris Mason */ 18165f39d397SChris Mason if (!parent) { 18175f39d397SChris Mason struct extent_buffer *child; 1818bb803951SChris Mason 18195f39d397SChris Mason if (btrfs_header_nritems(mid) != 1) 1820bb803951SChris Mason return 0; 1821bb803951SChris Mason 1822bb803951SChris Mason /* promote the child to a root */ 18234b231ae4SDavid Sterba child = btrfs_read_node_slot(mid, 0); 1824fb770ae4SLiu Bo if (IS_ERR(child)) { 1825fb770ae4SLiu Bo ret = PTR_ERR(child); 18260b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 1827305a26afSMark Fasheh goto enospc; 1828305a26afSMark Fasheh } 1829305a26afSMark Fasheh 1830925baeddSChris Mason btrfs_tree_lock(child); 18319631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 18329631e4ccSJosef Bacik BTRFS_NESTING_COW); 1833f0486c68SYan, Zheng if (ret) { 1834f0486c68SYan, Zheng btrfs_tree_unlock(child); 1835f0486c68SYan, Zheng free_extent_buffer(child); 1836f0486c68SYan, Zheng goto enospc; 1837f0486c68SYan, Zheng } 18382f375ab9SYan 1839d9d19a01SDavid Sterba ret = tree_mod_log_insert_root(root->node, child, 1); 1840d9d19a01SDavid Sterba BUG_ON(ret < 0); 1841240f62c8SChris Mason rcu_assign_pointer(root->node, child); 1842925baeddSChris Mason 18430b86a832SChris Mason add_root_to_dirty_list(root); 1844925baeddSChris Mason btrfs_tree_unlock(child); 1845b4ce94deSChris Mason 1846925baeddSChris Mason path->locks[level] = 0; 1847bb803951SChris Mason path->nodes[level] = NULL; 18486a884d7dSDavid Sterba btrfs_clean_tree_block(mid); 1849925baeddSChris Mason btrfs_tree_unlock(mid); 1850bb803951SChris Mason /* once for the path */ 18515f39d397SChris Mason free_extent_buffer(mid); 1852f0486c68SYan, Zheng 1853f0486c68SYan, Zheng root_sub_used(root, mid->len); 18545581a51aSJan Schmidt btrfs_free_tree_block(trans, root, mid, 0, 1); 1855bb803951SChris Mason /* once for the root ptr */ 18563083ee2eSJosef Bacik free_extent_buffer_stale(mid); 1857f0486c68SYan, Zheng return 0; 1858bb803951SChris Mason } 18595f39d397SChris Mason if (btrfs_header_nritems(mid) > 18600b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) 1861bb803951SChris Mason return 0; 1862bb803951SChris Mason 18634b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 1864fb770ae4SLiu Bo if (IS_ERR(left)) 1865fb770ae4SLiu Bo left = NULL; 1866fb770ae4SLiu Bo 18675f39d397SChris Mason if (left) { 1868bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 18695f39d397SChris Mason wret = btrfs_cow_block(trans, root, left, 18709631e4ccSJosef Bacik parent, pslot - 1, &left, 1871bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 187254aa1f4dSChris Mason if (wret) { 187354aa1f4dSChris Mason ret = wret; 187454aa1f4dSChris Mason goto enospc; 187554aa1f4dSChris Mason } 18762cc58cf2SChris Mason } 1877fb770ae4SLiu Bo 18784b231ae4SDavid Sterba right = btrfs_read_node_slot(parent, pslot + 1); 1879fb770ae4SLiu Bo if (IS_ERR(right)) 1880fb770ae4SLiu Bo right = NULL; 1881fb770ae4SLiu Bo 18825f39d397SChris Mason if (right) { 1883bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 18845f39d397SChris Mason wret = btrfs_cow_block(trans, root, right, 18859631e4ccSJosef Bacik parent, pslot + 1, &right, 1886bf59a5a2SJosef Bacik BTRFS_NESTING_RIGHT_COW); 18872cc58cf2SChris Mason if (wret) { 18882cc58cf2SChris Mason ret = wret; 18892cc58cf2SChris Mason goto enospc; 18902cc58cf2SChris Mason } 18912cc58cf2SChris Mason } 18922cc58cf2SChris Mason 18932cc58cf2SChris Mason /* first, try to make some room in the middle buffer */ 18945f39d397SChris Mason if (left) { 18955f39d397SChris Mason orig_slot += btrfs_header_nritems(left); 1896d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 189779f95c82SChris Mason if (wret < 0) 189879f95c82SChris Mason ret = wret; 1899bb803951SChris Mason } 190079f95c82SChris Mason 190179f95c82SChris Mason /* 190279f95c82SChris Mason * then try to empty the right most buffer into the middle 190379f95c82SChris Mason */ 19045f39d397SChris Mason if (right) { 1905d30a668fSDavid Sterba wret = push_node_left(trans, mid, right, 1); 190654aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 190779f95c82SChris Mason ret = wret; 19085f39d397SChris Mason if (btrfs_header_nritems(right) == 0) { 19096a884d7dSDavid Sterba btrfs_clean_tree_block(right); 1910925baeddSChris Mason btrfs_tree_unlock(right); 1911afe5fea7STsutomu Itoh del_ptr(root, path, level + 1, pslot + 1); 1912f0486c68SYan, Zheng root_sub_used(root, right->len); 19135581a51aSJan Schmidt btrfs_free_tree_block(trans, root, right, 0, 1); 19143083ee2eSJosef Bacik free_extent_buffer_stale(right); 1915f0486c68SYan, Zheng right = NULL; 1916bb803951SChris Mason } else { 19175f39d397SChris Mason struct btrfs_disk_key right_key; 19185f39d397SChris Mason btrfs_node_key(right, &right_key, 0); 19190e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(parent, pslot + 1, 19200e82bcfeSDavid Sterba MOD_LOG_KEY_REPLACE, GFP_NOFS); 19210e82bcfeSDavid Sterba BUG_ON(ret < 0); 19225f39d397SChris Mason btrfs_set_node_key(parent, &right_key, pslot + 1); 19235f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 1924bb803951SChris Mason } 1925bb803951SChris Mason } 19265f39d397SChris Mason if (btrfs_header_nritems(mid) == 1) { 192779f95c82SChris Mason /* 192879f95c82SChris Mason * we're not allowed to leave a node with one item in the 192979f95c82SChris Mason * tree during a delete. A deletion from lower in the tree 193079f95c82SChris Mason * could try to delete the only pointer in this node. 193179f95c82SChris Mason * So, pull some keys from the left. 193279f95c82SChris Mason * There has to be a left pointer at this point because 193379f95c82SChris Mason * otherwise we would have pulled some pointers from the 193479f95c82SChris Mason * right 193579f95c82SChris Mason */ 1936305a26afSMark Fasheh if (!left) { 1937305a26afSMark Fasheh ret = -EROFS; 19380b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 1939305a26afSMark Fasheh goto enospc; 1940305a26afSMark Fasheh } 194155d32ed8SDavid Sterba wret = balance_node_right(trans, mid, left); 194254aa1f4dSChris Mason if (wret < 0) { 194379f95c82SChris Mason ret = wret; 194454aa1f4dSChris Mason goto enospc; 194554aa1f4dSChris Mason } 1946bce4eae9SChris Mason if (wret == 1) { 1947d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 1); 1948bce4eae9SChris Mason if (wret < 0) 1949bce4eae9SChris Mason ret = wret; 1950bce4eae9SChris Mason } 195179f95c82SChris Mason BUG_ON(wret == 1); 195279f95c82SChris Mason } 19535f39d397SChris Mason if (btrfs_header_nritems(mid) == 0) { 19546a884d7dSDavid Sterba btrfs_clean_tree_block(mid); 1955925baeddSChris Mason btrfs_tree_unlock(mid); 1956afe5fea7STsutomu Itoh del_ptr(root, path, level + 1, pslot); 1957f0486c68SYan, Zheng root_sub_used(root, mid->len); 19585581a51aSJan Schmidt btrfs_free_tree_block(trans, root, mid, 0, 1); 19593083ee2eSJosef Bacik free_extent_buffer_stale(mid); 1960f0486c68SYan, Zheng mid = NULL; 196179f95c82SChris Mason } else { 196279f95c82SChris Mason /* update the parent key to reflect our changes */ 19635f39d397SChris Mason struct btrfs_disk_key mid_key; 19645f39d397SChris Mason btrfs_node_key(mid, &mid_key, 0); 19650e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(parent, pslot, 19660e82bcfeSDavid Sterba MOD_LOG_KEY_REPLACE, GFP_NOFS); 19670e82bcfeSDavid Sterba BUG_ON(ret < 0); 19685f39d397SChris Mason btrfs_set_node_key(parent, &mid_key, pslot); 19695f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 197079f95c82SChris Mason } 1971bb803951SChris Mason 197279f95c82SChris Mason /* update the path */ 19735f39d397SChris Mason if (left) { 19745f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 197567439dadSDavid Sterba atomic_inc(&left->refs); 1976925baeddSChris Mason /* left was locked after cow */ 19775f39d397SChris Mason path->nodes[level] = left; 1978bb803951SChris Mason path->slots[level + 1] -= 1; 1979bb803951SChris Mason path->slots[level] = orig_slot; 1980925baeddSChris Mason if (mid) { 1981925baeddSChris Mason btrfs_tree_unlock(mid); 19825f39d397SChris Mason free_extent_buffer(mid); 1983925baeddSChris Mason } 1984bb803951SChris Mason } else { 19855f39d397SChris Mason orig_slot -= btrfs_header_nritems(left); 1986bb803951SChris Mason path->slots[level] = orig_slot; 1987bb803951SChris Mason } 1988bb803951SChris Mason } 198979f95c82SChris Mason /* double check we haven't messed things up */ 1990e20d96d6SChris Mason if (orig_ptr != 19915f39d397SChris Mason btrfs_node_blockptr(path->nodes[level], path->slots[level])) 199279f95c82SChris Mason BUG(); 199354aa1f4dSChris Mason enospc: 1994925baeddSChris Mason if (right) { 1995925baeddSChris Mason btrfs_tree_unlock(right); 19965f39d397SChris Mason free_extent_buffer(right); 1997925baeddSChris Mason } 1998925baeddSChris Mason if (left) { 1999925baeddSChris Mason if (path->nodes[level] != left) 2000925baeddSChris Mason btrfs_tree_unlock(left); 20015f39d397SChris Mason free_extent_buffer(left); 2002925baeddSChris Mason } 2003bb803951SChris Mason return ret; 2004bb803951SChris Mason } 2005bb803951SChris Mason 2006d352ac68SChris Mason /* Node balancing for insertion. Here we only split or push nodes around 2007d352ac68SChris Mason * when they are completely full. This is also done top down, so we 2008d352ac68SChris Mason * have to be pessimistic. 2009d352ac68SChris Mason */ 2010d397712bSChris Mason static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, 2011e66f709bSChris Mason struct btrfs_root *root, 2012e66f709bSChris Mason struct btrfs_path *path, int level) 2013e66f709bSChris Mason { 20140b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20155f39d397SChris Mason struct extent_buffer *right = NULL; 20165f39d397SChris Mason struct extent_buffer *mid; 20175f39d397SChris Mason struct extent_buffer *left = NULL; 20185f39d397SChris Mason struct extent_buffer *parent = NULL; 2019e66f709bSChris Mason int ret = 0; 2020e66f709bSChris Mason int wret; 2021e66f709bSChris Mason int pslot; 2022e66f709bSChris Mason int orig_slot = path->slots[level]; 2023e66f709bSChris Mason 2024e66f709bSChris Mason if (level == 0) 2025e66f709bSChris Mason return 1; 2026e66f709bSChris Mason 20275f39d397SChris Mason mid = path->nodes[level]; 20287bb86316SChris Mason WARN_ON(btrfs_header_generation(mid) != trans->transid); 2029e66f709bSChris Mason 2030a05a9bb1SLi Zefan if (level < BTRFS_MAX_LEVEL - 1) { 20315f39d397SChris Mason parent = path->nodes[level + 1]; 2032e66f709bSChris Mason pslot = path->slots[level + 1]; 2033a05a9bb1SLi Zefan } 2034e66f709bSChris Mason 20355f39d397SChris Mason if (!parent) 2036e66f709bSChris Mason return 1; 2037e66f709bSChris Mason 20384b231ae4SDavid Sterba left = btrfs_read_node_slot(parent, pslot - 1); 2039fb770ae4SLiu Bo if (IS_ERR(left)) 2040fb770ae4SLiu Bo left = NULL; 2041e66f709bSChris Mason 2042e66f709bSChris Mason /* first, try to make some room in the middle buffer */ 20435f39d397SChris Mason if (left) { 2044e66f709bSChris Mason u32 left_nr; 2045925baeddSChris Mason 2046bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 2047b4ce94deSChris Mason 20485f39d397SChris Mason left_nr = btrfs_header_nritems(left); 20490b246afaSJeff Mahoney if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 205033ade1f8SChris Mason wret = 1; 205133ade1f8SChris Mason } else { 20525f39d397SChris Mason ret = btrfs_cow_block(trans, root, left, parent, 20539631e4ccSJosef Bacik pslot - 1, &left, 2054bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 205554aa1f4dSChris Mason if (ret) 205654aa1f4dSChris Mason wret = 1; 205754aa1f4dSChris Mason else { 2058d30a668fSDavid Sterba wret = push_node_left(trans, left, mid, 0); 205954aa1f4dSChris Mason } 206033ade1f8SChris Mason } 2061e66f709bSChris Mason if (wret < 0) 2062e66f709bSChris Mason ret = wret; 2063e66f709bSChris Mason if (wret == 0) { 20645f39d397SChris Mason struct btrfs_disk_key disk_key; 2065e66f709bSChris Mason orig_slot += left_nr; 20665f39d397SChris Mason btrfs_node_key(mid, &disk_key, 0); 20670e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(parent, pslot, 20680e82bcfeSDavid Sterba MOD_LOG_KEY_REPLACE, GFP_NOFS); 20690e82bcfeSDavid Sterba BUG_ON(ret < 0); 20705f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot); 20715f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 20725f39d397SChris Mason if (btrfs_header_nritems(left) > orig_slot) { 20735f39d397SChris Mason path->nodes[level] = left; 2074e66f709bSChris Mason path->slots[level + 1] -= 1; 2075e66f709bSChris Mason path->slots[level] = orig_slot; 2076925baeddSChris Mason btrfs_tree_unlock(mid); 20775f39d397SChris Mason free_extent_buffer(mid); 2078e66f709bSChris Mason } else { 2079e66f709bSChris Mason orig_slot -= 20805f39d397SChris Mason btrfs_header_nritems(left); 2081e66f709bSChris Mason path->slots[level] = orig_slot; 2082925baeddSChris Mason btrfs_tree_unlock(left); 20835f39d397SChris Mason free_extent_buffer(left); 2084e66f709bSChris Mason } 2085e66f709bSChris Mason return 0; 2086e66f709bSChris Mason } 2087925baeddSChris Mason btrfs_tree_unlock(left); 20885f39d397SChris Mason free_extent_buffer(left); 2089e66f709bSChris Mason } 20904b231ae4SDavid Sterba right = btrfs_read_node_slot(parent, pslot + 1); 2091fb770ae4SLiu Bo if (IS_ERR(right)) 2092fb770ae4SLiu Bo right = NULL; 2093e66f709bSChris Mason 2094e66f709bSChris Mason /* 2095e66f709bSChris Mason * then try to empty the right most buffer into the middle 2096e66f709bSChris Mason */ 20975f39d397SChris Mason if (right) { 209833ade1f8SChris Mason u32 right_nr; 2099b4ce94deSChris Mason 2100bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 2101b4ce94deSChris Mason 21025f39d397SChris Mason right_nr = btrfs_header_nritems(right); 21030b246afaSJeff Mahoney if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 210433ade1f8SChris Mason wret = 1; 210533ade1f8SChris Mason } else { 21065f39d397SChris Mason ret = btrfs_cow_block(trans, root, right, 21075f39d397SChris Mason parent, pslot + 1, 2108bf59a5a2SJosef Bacik &right, BTRFS_NESTING_RIGHT_COW); 210954aa1f4dSChris Mason if (ret) 211054aa1f4dSChris Mason wret = 1; 211154aa1f4dSChris Mason else { 211255d32ed8SDavid Sterba wret = balance_node_right(trans, right, mid); 211333ade1f8SChris Mason } 211454aa1f4dSChris Mason } 2115e66f709bSChris Mason if (wret < 0) 2116e66f709bSChris Mason ret = wret; 2117e66f709bSChris Mason if (wret == 0) { 21185f39d397SChris Mason struct btrfs_disk_key disk_key; 21195f39d397SChris Mason 21205f39d397SChris Mason btrfs_node_key(right, &disk_key, 0); 21210e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(parent, pslot + 1, 21220e82bcfeSDavid Sterba MOD_LOG_KEY_REPLACE, GFP_NOFS); 21230e82bcfeSDavid Sterba BUG_ON(ret < 0); 21245f39d397SChris Mason btrfs_set_node_key(parent, &disk_key, pslot + 1); 21255f39d397SChris Mason btrfs_mark_buffer_dirty(parent); 21265f39d397SChris Mason 21275f39d397SChris Mason if (btrfs_header_nritems(mid) <= orig_slot) { 21285f39d397SChris Mason path->nodes[level] = right; 2129e66f709bSChris Mason path->slots[level + 1] += 1; 2130e66f709bSChris Mason path->slots[level] = orig_slot - 21315f39d397SChris Mason btrfs_header_nritems(mid); 2132925baeddSChris Mason btrfs_tree_unlock(mid); 21335f39d397SChris Mason free_extent_buffer(mid); 2134e66f709bSChris Mason } else { 2135925baeddSChris Mason btrfs_tree_unlock(right); 21365f39d397SChris Mason free_extent_buffer(right); 2137e66f709bSChris Mason } 2138e66f709bSChris Mason return 0; 2139e66f709bSChris Mason } 2140925baeddSChris Mason btrfs_tree_unlock(right); 21415f39d397SChris Mason free_extent_buffer(right); 2142e66f709bSChris Mason } 2143e66f709bSChris Mason return 1; 2144e66f709bSChris Mason } 2145e66f709bSChris Mason 214674123bd7SChris Mason /* 2147d352ac68SChris Mason * readahead one full node of leaves, finding things that are close 2148d352ac68SChris Mason * to the block in 'slot', and triggering ra on them. 21493c69faecSChris Mason */ 21502ff7e61eSJeff Mahoney static void reada_for_search(struct btrfs_fs_info *fs_info, 2151e02119d5SChris Mason struct btrfs_path *path, 215201f46658SChris Mason int level, int slot, u64 objectid) 21533c69faecSChris Mason { 21545f39d397SChris Mason struct extent_buffer *node; 215501f46658SChris Mason struct btrfs_disk_key disk_key; 21563c69faecSChris Mason u32 nritems; 21573c69faecSChris Mason u64 search; 2158a7175319SChris Mason u64 target; 21596b80053dSChris Mason u64 nread = 0; 21605f39d397SChris Mason struct extent_buffer *eb; 21616b80053dSChris Mason u32 nr; 21626b80053dSChris Mason u32 blocksize; 21636b80053dSChris Mason u32 nscan = 0; 2164db94535dSChris Mason 2165a6b6e75eSChris Mason if (level != 1) 21663c69faecSChris Mason return; 21673c69faecSChris Mason 21686702ed49SChris Mason if (!path->nodes[level]) 21696702ed49SChris Mason return; 21706702ed49SChris Mason 21715f39d397SChris Mason node = path->nodes[level]; 2172925baeddSChris Mason 21733c69faecSChris Mason search = btrfs_node_blockptr(node, slot); 21740b246afaSJeff Mahoney blocksize = fs_info->nodesize; 21750b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, search); 21765f39d397SChris Mason if (eb) { 21775f39d397SChris Mason free_extent_buffer(eb); 21783c69faecSChris Mason return; 21793c69faecSChris Mason } 21803c69faecSChris Mason 2181a7175319SChris Mason target = search; 21826b80053dSChris Mason 21835f39d397SChris Mason nritems = btrfs_header_nritems(node); 21846b80053dSChris Mason nr = slot; 218525b8b936SJosef Bacik 21863c69faecSChris Mason while (1) { 2187e4058b54SDavid Sterba if (path->reada == READA_BACK) { 21886b80053dSChris Mason if (nr == 0) 21893c69faecSChris Mason break; 21906b80053dSChris Mason nr--; 2191e4058b54SDavid Sterba } else if (path->reada == READA_FORWARD) { 21926b80053dSChris Mason nr++; 21936b80053dSChris Mason if (nr >= nritems) 21946b80053dSChris Mason break; 21953c69faecSChris Mason } 2196e4058b54SDavid Sterba if (path->reada == READA_BACK && objectid) { 219701f46658SChris Mason btrfs_node_key(node, &disk_key, nr); 219801f46658SChris Mason if (btrfs_disk_key_objectid(&disk_key) != objectid) 219901f46658SChris Mason break; 220001f46658SChris Mason } 22016b80053dSChris Mason search = btrfs_node_blockptr(node, nr); 2202a7175319SChris Mason if ((search <= target && target - search <= 65536) || 2203a7175319SChris Mason (search > target && search - target <= 65536)) { 2204bfb484d9SJosef Bacik btrfs_readahead_node_child(node, nr); 22056b80053dSChris Mason nread += blocksize; 22063c69faecSChris Mason } 22076b80053dSChris Mason nscan++; 2208a7175319SChris Mason if ((nread > 65536 || nscan > 32)) 22096b80053dSChris Mason break; 22103c69faecSChris Mason } 22113c69faecSChris Mason } 2212925baeddSChris Mason 2213bfb484d9SJosef Bacik static noinline void reada_for_balance(struct btrfs_path *path, int level) 2214b4ce94deSChris Mason { 2215bfb484d9SJosef Bacik struct extent_buffer *parent; 2216b4ce94deSChris Mason int slot; 2217b4ce94deSChris Mason int nritems; 2218b4ce94deSChris Mason 22198c594ea8SChris Mason parent = path->nodes[level + 1]; 2220b4ce94deSChris Mason if (!parent) 22210b08851fSJosef Bacik return; 2222b4ce94deSChris Mason 2223b4ce94deSChris Mason nritems = btrfs_header_nritems(parent); 22248c594ea8SChris Mason slot = path->slots[level + 1]; 2225b4ce94deSChris Mason 2226bfb484d9SJosef Bacik if (slot > 0) 2227bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot - 1); 2228bfb484d9SJosef Bacik if (slot + 1 < nritems) 2229bfb484d9SJosef Bacik btrfs_readahead_node_child(parent, slot + 1); 2230b4ce94deSChris Mason } 2231b4ce94deSChris Mason 2232b4ce94deSChris Mason 2233b4ce94deSChris Mason /* 2234d397712bSChris Mason * when we walk down the tree, it is usually safe to unlock the higher layers 2235d397712bSChris Mason * in the tree. The exceptions are when our path goes through slot 0, because 2236d397712bSChris Mason * operations on the tree might require changing key pointers higher up in the 2237d397712bSChris Mason * tree. 2238d352ac68SChris Mason * 2239d397712bSChris Mason * callers might also have set path->keep_locks, which tells this code to keep 2240d397712bSChris Mason * the lock if the path points to the last slot in the block. This is part of 2241d397712bSChris Mason * walking through the tree, and selecting the next slot in the higher block. 2242d352ac68SChris Mason * 2243d397712bSChris Mason * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so 2244d397712bSChris Mason * if lowest_unlock is 1, level 0 won't be unlocked 2245d352ac68SChris Mason */ 2246e02119d5SChris Mason static noinline void unlock_up(struct btrfs_path *path, int level, 2247f7c79f30SChris Mason int lowest_unlock, int min_write_lock_level, 2248f7c79f30SChris Mason int *write_lock_level) 2249925baeddSChris Mason { 2250925baeddSChris Mason int i; 2251925baeddSChris Mason int skip_level = level; 2252051e1b9fSChris Mason int no_skips = 0; 2253925baeddSChris Mason struct extent_buffer *t; 2254925baeddSChris Mason 2255925baeddSChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 2256925baeddSChris Mason if (!path->nodes[i]) 2257925baeddSChris Mason break; 2258925baeddSChris Mason if (!path->locks[i]) 2259925baeddSChris Mason break; 2260051e1b9fSChris Mason if (!no_skips && path->slots[i] == 0) { 2261925baeddSChris Mason skip_level = i + 1; 2262925baeddSChris Mason continue; 2263925baeddSChris Mason } 2264051e1b9fSChris Mason if (!no_skips && path->keep_locks) { 2265925baeddSChris Mason u32 nritems; 2266925baeddSChris Mason t = path->nodes[i]; 2267925baeddSChris Mason nritems = btrfs_header_nritems(t); 2268051e1b9fSChris Mason if (nritems < 1 || path->slots[i] >= nritems - 1) { 2269925baeddSChris Mason skip_level = i + 1; 2270925baeddSChris Mason continue; 2271925baeddSChris Mason } 2272925baeddSChris Mason } 2273051e1b9fSChris Mason if (skip_level < i && i >= lowest_unlock) 2274051e1b9fSChris Mason no_skips = 1; 2275051e1b9fSChris Mason 2276925baeddSChris Mason t = path->nodes[i]; 2277d80bb3f9SLiu Bo if (i >= lowest_unlock && i > skip_level) { 2278bd681513SChris Mason btrfs_tree_unlock_rw(t, path->locks[i]); 2279925baeddSChris Mason path->locks[i] = 0; 2280f7c79f30SChris Mason if (write_lock_level && 2281f7c79f30SChris Mason i > min_write_lock_level && 2282f7c79f30SChris Mason i <= *write_lock_level) { 2283f7c79f30SChris Mason *write_lock_level = i - 1; 2284f7c79f30SChris Mason } 2285925baeddSChris Mason } 2286925baeddSChris Mason } 2287925baeddSChris Mason } 2288925baeddSChris Mason 22893c69faecSChris Mason /* 2290c8c42864SChris Mason * helper function for btrfs_search_slot. The goal is to find a block 2291c8c42864SChris Mason * in cache without setting the path to blocking. If we find the block 2292c8c42864SChris Mason * we return zero and the path is unchanged. 2293c8c42864SChris Mason * 2294c8c42864SChris Mason * If we can't find the block, we set the path blocking and do some 2295c8c42864SChris Mason * reada. -EAGAIN is returned and the search must be repeated. 2296c8c42864SChris Mason */ 2297c8c42864SChris Mason static int 2298d07b8528SLiu Bo read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, 2299c8c42864SChris Mason struct extent_buffer **eb_ret, int level, int slot, 2300cda79c54SDavid Sterba const struct btrfs_key *key) 2301c8c42864SChris Mason { 23020b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 2303c8c42864SChris Mason u64 blocknr; 2304c8c42864SChris Mason u64 gen; 2305c8c42864SChris Mason struct extent_buffer *tmp; 2306581c1760SQu Wenruo struct btrfs_key first_key; 230776a05b35SChris Mason int ret; 2308581c1760SQu Wenruo int parent_level; 2309c8c42864SChris Mason 2310213ff4b7SNikolay Borisov blocknr = btrfs_node_blockptr(*eb_ret, slot); 2311213ff4b7SNikolay Borisov gen = btrfs_node_ptr_generation(*eb_ret, slot); 2312213ff4b7SNikolay Borisov parent_level = btrfs_header_level(*eb_ret); 2313213ff4b7SNikolay Borisov btrfs_node_key_to_cpu(*eb_ret, &first_key, slot); 2314c8c42864SChris Mason 23150b246afaSJeff Mahoney tmp = find_extent_buffer(fs_info, blocknr); 2316cb44921aSChris Mason if (tmp) { 2317b9fab919SChris Mason /* first we do an atomic uptodate check */ 2318b9fab919SChris Mason if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { 2319448de471SQu Wenruo /* 2320448de471SQu Wenruo * Do extra check for first_key, eb can be stale due to 2321448de471SQu Wenruo * being cached, read from scrub, or have multiple 2322448de471SQu Wenruo * parents (shared tree blocks). 2323448de471SQu Wenruo */ 2324e064d5e9SDavid Sterba if (btrfs_verify_level_key(tmp, 2325448de471SQu Wenruo parent_level - 1, &first_key, gen)) { 2326448de471SQu Wenruo free_extent_buffer(tmp); 2327448de471SQu Wenruo return -EUCLEAN; 2328448de471SQu Wenruo } 2329c8c42864SChris Mason *eb_ret = tmp; 2330c8c42864SChris Mason return 0; 2331c8c42864SChris Mason } 2332bdf7c00eSJosef Bacik 2333b9fab919SChris Mason /* now we're allowed to do a blocking uptodate check */ 2334581c1760SQu Wenruo ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); 2335bdf7c00eSJosef Bacik if (!ret) { 2336cb44921aSChris Mason *eb_ret = tmp; 2337cb44921aSChris Mason return 0; 2338cb44921aSChris Mason } 2339cb44921aSChris Mason free_extent_buffer(tmp); 2340b3b4aa74SDavid Sterba btrfs_release_path(p); 2341cb44921aSChris Mason return -EIO; 2342cb44921aSChris Mason } 2343c8c42864SChris Mason 2344c8c42864SChris Mason /* 2345c8c42864SChris Mason * reduce lock contention at high levels 2346c8c42864SChris Mason * of the btree by dropping locks before 234776a05b35SChris Mason * we read. Don't release the lock on the current 234876a05b35SChris Mason * level because we need to walk this node to figure 234976a05b35SChris Mason * out which blocks to read. 2350c8c42864SChris Mason */ 23518c594ea8SChris Mason btrfs_unlock_up_safe(p, level + 1); 23528c594ea8SChris Mason 2353e4058b54SDavid Sterba if (p->reada != READA_NONE) 23542ff7e61eSJeff Mahoney reada_for_search(fs_info, p, level, slot, key->objectid); 2355c8c42864SChris Mason 235676a05b35SChris Mason ret = -EAGAIN; 23571b7ec85eSJosef Bacik tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid, 23581b7ec85eSJosef Bacik gen, parent_level - 1, &first_key); 235964c043deSLiu Bo if (!IS_ERR(tmp)) { 236076a05b35SChris Mason /* 236176a05b35SChris Mason * If the read above didn't mark this buffer up to date, 236276a05b35SChris Mason * it will never end up being up to date. Set ret to EIO now 236376a05b35SChris Mason * and give up so that our caller doesn't loop forever 236476a05b35SChris Mason * on our EAGAINs. 236576a05b35SChris Mason */ 2366e6a1d6fdSLiu Bo if (!extent_buffer_uptodate(tmp)) 236776a05b35SChris Mason ret = -EIO; 2368c8c42864SChris Mason free_extent_buffer(tmp); 2369c871b0f2SLiu Bo } else { 2370c871b0f2SLiu Bo ret = PTR_ERR(tmp); 237176a05b35SChris Mason } 237202a3307aSLiu Bo 237302a3307aSLiu Bo btrfs_release_path(p); 237476a05b35SChris Mason return ret; 2375c8c42864SChris Mason } 2376c8c42864SChris Mason 2377c8c42864SChris Mason /* 2378c8c42864SChris Mason * helper function for btrfs_search_slot. This does all of the checks 2379c8c42864SChris Mason * for node-level blocks and does any balancing required based on 2380c8c42864SChris Mason * the ins_len. 2381c8c42864SChris Mason * 2382c8c42864SChris Mason * If no extra work was required, zero is returned. If we had to 2383c8c42864SChris Mason * drop the path, -EAGAIN is returned and btrfs_search_slot must 2384c8c42864SChris Mason * start over 2385c8c42864SChris Mason */ 2386c8c42864SChris Mason static int 2387c8c42864SChris Mason setup_nodes_for_search(struct btrfs_trans_handle *trans, 2388c8c42864SChris Mason struct btrfs_root *root, struct btrfs_path *p, 2389bd681513SChris Mason struct extent_buffer *b, int level, int ins_len, 2390bd681513SChris Mason int *write_lock_level) 2391c8c42864SChris Mason { 23920b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 239395b982deSNikolay Borisov int ret = 0; 23940b246afaSJeff Mahoney 2395c8c42864SChris Mason if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= 23960b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { 2397c8c42864SChris Mason 2398bd681513SChris Mason if (*write_lock_level < level + 1) { 2399bd681513SChris Mason *write_lock_level = level + 1; 2400bd681513SChris Mason btrfs_release_path(p); 240195b982deSNikolay Borisov return -EAGAIN; 2402bd681513SChris Mason } 2403bd681513SChris Mason 2404bfb484d9SJosef Bacik reada_for_balance(p, level); 240595b982deSNikolay Borisov ret = split_node(trans, root, p, level); 2406c8c42864SChris Mason 2407c8c42864SChris Mason b = p->nodes[level]; 2408c8c42864SChris Mason } else if (ins_len < 0 && btrfs_header_nritems(b) < 24090b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { 2410c8c42864SChris Mason 2411bd681513SChris Mason if (*write_lock_level < level + 1) { 2412bd681513SChris Mason *write_lock_level = level + 1; 2413bd681513SChris Mason btrfs_release_path(p); 241495b982deSNikolay Borisov return -EAGAIN; 2415bd681513SChris Mason } 2416bd681513SChris Mason 2417bfb484d9SJosef Bacik reada_for_balance(p, level); 241895b982deSNikolay Borisov ret = balance_level(trans, root, p, level); 241995b982deSNikolay Borisov if (ret) 242095b982deSNikolay Borisov return ret; 2421c8c42864SChris Mason 2422c8c42864SChris Mason b = p->nodes[level]; 2423c8c42864SChris Mason if (!b) { 2424b3b4aa74SDavid Sterba btrfs_release_path(p); 242595b982deSNikolay Borisov return -EAGAIN; 2426c8c42864SChris Mason } 2427c8c42864SChris Mason BUG_ON(btrfs_header_nritems(b) == 1); 2428c8c42864SChris Mason } 2429c8c42864SChris Mason return ret; 2430c8c42864SChris Mason } 2431c8c42864SChris Mason 2432381cf658SDavid Sterba int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, 2433e33d5c3dSKelley Nielsen u64 iobjectid, u64 ioff, u8 key_type, 2434e33d5c3dSKelley Nielsen struct btrfs_key *found_key) 2435e33d5c3dSKelley Nielsen { 2436e33d5c3dSKelley Nielsen int ret; 2437e33d5c3dSKelley Nielsen struct btrfs_key key; 2438e33d5c3dSKelley Nielsen struct extent_buffer *eb; 2439381cf658SDavid Sterba 2440381cf658SDavid Sterba ASSERT(path); 24411d4c08e0SDavid Sterba ASSERT(found_key); 2442e33d5c3dSKelley Nielsen 2443e33d5c3dSKelley Nielsen key.type = key_type; 2444e33d5c3dSKelley Nielsen key.objectid = iobjectid; 2445e33d5c3dSKelley Nielsen key.offset = ioff; 2446e33d5c3dSKelley Nielsen 2447e33d5c3dSKelley Nielsen ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); 24481d4c08e0SDavid Sterba if (ret < 0) 2449e33d5c3dSKelley Nielsen return ret; 2450e33d5c3dSKelley Nielsen 2451e33d5c3dSKelley Nielsen eb = path->nodes[0]; 2452e33d5c3dSKelley Nielsen if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { 2453e33d5c3dSKelley Nielsen ret = btrfs_next_leaf(fs_root, path); 2454e33d5c3dSKelley Nielsen if (ret) 2455e33d5c3dSKelley Nielsen return ret; 2456e33d5c3dSKelley Nielsen eb = path->nodes[0]; 2457e33d5c3dSKelley Nielsen } 2458e33d5c3dSKelley Nielsen 2459e33d5c3dSKelley Nielsen btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); 2460e33d5c3dSKelley Nielsen if (found_key->type != key.type || 2461e33d5c3dSKelley Nielsen found_key->objectid != key.objectid) 2462e33d5c3dSKelley Nielsen return 1; 2463e33d5c3dSKelley Nielsen 2464e33d5c3dSKelley Nielsen return 0; 2465e33d5c3dSKelley Nielsen } 2466e33d5c3dSKelley Nielsen 24671fc28d8eSLiu Bo static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, 24681fc28d8eSLiu Bo struct btrfs_path *p, 24691fc28d8eSLiu Bo int write_lock_level) 24701fc28d8eSLiu Bo { 24711fc28d8eSLiu Bo struct btrfs_fs_info *fs_info = root->fs_info; 24721fc28d8eSLiu Bo struct extent_buffer *b; 24731fc28d8eSLiu Bo int root_lock; 24741fc28d8eSLiu Bo int level = 0; 24751fc28d8eSLiu Bo 24761fc28d8eSLiu Bo /* We try very hard to do read locks on the root */ 24771fc28d8eSLiu Bo root_lock = BTRFS_READ_LOCK; 24781fc28d8eSLiu Bo 24791fc28d8eSLiu Bo if (p->search_commit_root) { 2480be6821f8SFilipe Manana /* 2481be6821f8SFilipe Manana * The commit roots are read only so we always do read locks, 2482be6821f8SFilipe Manana * and we always must hold the commit_root_sem when doing 2483be6821f8SFilipe Manana * searches on them, the only exception is send where we don't 2484be6821f8SFilipe Manana * want to block transaction commits for a long time, so 2485be6821f8SFilipe Manana * we need to clone the commit root in order to avoid races 2486be6821f8SFilipe Manana * with transaction commits that create a snapshot of one of 2487be6821f8SFilipe Manana * the roots used by a send operation. 2488be6821f8SFilipe Manana */ 2489be6821f8SFilipe Manana if (p->need_commit_sem) { 24901fc28d8eSLiu Bo down_read(&fs_info->commit_root_sem); 2491be6821f8SFilipe Manana b = btrfs_clone_extent_buffer(root->commit_root); 2492be6821f8SFilipe Manana up_read(&fs_info->commit_root_sem); 2493be6821f8SFilipe Manana if (!b) 2494be6821f8SFilipe Manana return ERR_PTR(-ENOMEM); 2495be6821f8SFilipe Manana 2496be6821f8SFilipe Manana } else { 24971fc28d8eSLiu Bo b = root->commit_root; 249867439dadSDavid Sterba atomic_inc(&b->refs); 2499be6821f8SFilipe Manana } 25001fc28d8eSLiu Bo level = btrfs_header_level(b); 2501f9ddfd05SLiu Bo /* 2502f9ddfd05SLiu Bo * Ensure that all callers have set skip_locking when 2503f9ddfd05SLiu Bo * p->search_commit_root = 1. 2504f9ddfd05SLiu Bo */ 2505f9ddfd05SLiu Bo ASSERT(p->skip_locking == 1); 25061fc28d8eSLiu Bo 25071fc28d8eSLiu Bo goto out; 25081fc28d8eSLiu Bo } 25091fc28d8eSLiu Bo 25101fc28d8eSLiu Bo if (p->skip_locking) { 25111fc28d8eSLiu Bo b = btrfs_root_node(root); 25121fc28d8eSLiu Bo level = btrfs_header_level(b); 25131fc28d8eSLiu Bo goto out; 25141fc28d8eSLiu Bo } 25151fc28d8eSLiu Bo 25161fc28d8eSLiu Bo /* 2517662c653bSLiu Bo * If the level is set to maximum, we can skip trying to get the read 2518662c653bSLiu Bo * lock. 2519662c653bSLiu Bo */ 2520662c653bSLiu Bo if (write_lock_level < BTRFS_MAX_LEVEL) { 2521662c653bSLiu Bo /* 2522662c653bSLiu Bo * We don't know the level of the root node until we actually 2523662c653bSLiu Bo * have it read locked 25241fc28d8eSLiu Bo */ 25251bb96598SJosef Bacik b = btrfs_read_lock_root_node(root); 25261fc28d8eSLiu Bo level = btrfs_header_level(b); 25271fc28d8eSLiu Bo if (level > write_lock_level) 25281fc28d8eSLiu Bo goto out; 25291fc28d8eSLiu Bo 2530662c653bSLiu Bo /* Whoops, must trade for write lock */ 25311fc28d8eSLiu Bo btrfs_tree_read_unlock(b); 25321fc28d8eSLiu Bo free_extent_buffer(b); 2533662c653bSLiu Bo } 2534662c653bSLiu Bo 25351fc28d8eSLiu Bo b = btrfs_lock_root_node(root); 25361fc28d8eSLiu Bo root_lock = BTRFS_WRITE_LOCK; 25371fc28d8eSLiu Bo 25381fc28d8eSLiu Bo /* The level might have changed, check again */ 25391fc28d8eSLiu Bo level = btrfs_header_level(b); 25401fc28d8eSLiu Bo 25411fc28d8eSLiu Bo out: 25421fc28d8eSLiu Bo p->nodes[level] = b; 25431fc28d8eSLiu Bo if (!p->skip_locking) 25441fc28d8eSLiu Bo p->locks[level] = root_lock; 25451fc28d8eSLiu Bo /* 25461fc28d8eSLiu Bo * Callers are responsible for dropping b's references. 25471fc28d8eSLiu Bo */ 25481fc28d8eSLiu Bo return b; 25491fc28d8eSLiu Bo } 25501fc28d8eSLiu Bo 25511fc28d8eSLiu Bo 2552c8c42864SChris Mason /* 25534271eceaSNikolay Borisov * btrfs_search_slot - look for a key in a tree and perform necessary 25544271eceaSNikolay Borisov * modifications to preserve tree invariants. 255574123bd7SChris Mason * 25564271eceaSNikolay Borisov * @trans: Handle of transaction, used when modifying the tree 25574271eceaSNikolay Borisov * @p: Holds all btree nodes along the search path 25584271eceaSNikolay Borisov * @root: The root node of the tree 25594271eceaSNikolay Borisov * @key: The key we are looking for 25609a664971Sethanwu * @ins_len: Indicates purpose of search: 25619a664971Sethanwu * >0 for inserts it's size of item inserted (*) 25629a664971Sethanwu * <0 for deletions 25639a664971Sethanwu * 0 for plain searches, not modifying the tree 25649a664971Sethanwu * 25659a664971Sethanwu * (*) If size of item inserted doesn't include 25669a664971Sethanwu * sizeof(struct btrfs_item), then p->search_for_extension must 25679a664971Sethanwu * be set. 25684271eceaSNikolay Borisov * @cow: boolean should CoW operations be performed. Must always be 1 25694271eceaSNikolay Borisov * when modifying the tree. 257097571fd0SChris Mason * 25714271eceaSNikolay Borisov * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. 25724271eceaSNikolay Borisov * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) 25734271eceaSNikolay Borisov * 25744271eceaSNikolay Borisov * If @key is found, 0 is returned and you can find the item in the leaf level 25754271eceaSNikolay Borisov * of the path (level 0) 25764271eceaSNikolay Borisov * 25774271eceaSNikolay Borisov * If @key isn't found, 1 is returned and the leaf level of the path (level 0) 25784271eceaSNikolay Borisov * points to the slot where it should be inserted 25794271eceaSNikolay Borisov * 25804271eceaSNikolay Borisov * If an error is encountered while searching the tree a negative error number 25814271eceaSNikolay Borisov * is returned 258274123bd7SChris Mason */ 2583310712b2SOmar Sandoval int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, 2584310712b2SOmar Sandoval const struct btrfs_key *key, struct btrfs_path *p, 2585310712b2SOmar Sandoval int ins_len, int cow) 2586be0e5c09SChris Mason { 25875f39d397SChris Mason struct extent_buffer *b; 2588be0e5c09SChris Mason int slot; 2589be0e5c09SChris Mason int ret; 259033c66f43SYan Zheng int err; 2591be0e5c09SChris Mason int level; 2592925baeddSChris Mason int lowest_unlock = 1; 2593bd681513SChris Mason /* everything at write_lock_level or lower must be write locked */ 2594bd681513SChris Mason int write_lock_level = 0; 25959f3a7427SChris Mason u8 lowest_level = 0; 2596f7c79f30SChris Mason int min_write_lock_level; 2597d7396f07SFilipe David Borba Manana int prev_cmp; 25989f3a7427SChris Mason 25996702ed49SChris Mason lowest_level = p->lowest_level; 2600323ac95bSChris Mason WARN_ON(lowest_level && ins_len > 0); 260122b0ebdaSChris Mason WARN_ON(p->nodes[0] != NULL); 2602eb653de1SFilipe David Borba Manana BUG_ON(!cow && ins_len); 260325179201SJosef Bacik 2604bd681513SChris Mason if (ins_len < 0) { 2605925baeddSChris Mason lowest_unlock = 2; 260665b51a00SChris Mason 2607bd681513SChris Mason /* when we are removing items, we might have to go up to level 2608bd681513SChris Mason * two as we update tree pointers Make sure we keep write 2609bd681513SChris Mason * for those levels as well 2610bd681513SChris Mason */ 2611bd681513SChris Mason write_lock_level = 2; 2612bd681513SChris Mason } else if (ins_len > 0) { 2613bd681513SChris Mason /* 2614bd681513SChris Mason * for inserting items, make sure we have a write lock on 2615bd681513SChris Mason * level 1 so we can update keys 2616bd681513SChris Mason */ 2617bd681513SChris Mason write_lock_level = 1; 2618bd681513SChris Mason } 2619bd681513SChris Mason 2620bd681513SChris Mason if (!cow) 2621bd681513SChris Mason write_lock_level = -1; 2622bd681513SChris Mason 262309a2a8f9SJosef Bacik if (cow && (p->keep_locks || p->lowest_level)) 2624bd681513SChris Mason write_lock_level = BTRFS_MAX_LEVEL; 2625bd681513SChris Mason 2626f7c79f30SChris Mason min_write_lock_level = write_lock_level; 2627f7c79f30SChris Mason 2628bb803951SChris Mason again: 2629d7396f07SFilipe David Borba Manana prev_cmp = -1; 26301fc28d8eSLiu Bo b = btrfs_search_slot_get_root(root, p, write_lock_level); 2631be6821f8SFilipe Manana if (IS_ERR(b)) { 2632be6821f8SFilipe Manana ret = PTR_ERR(b); 2633be6821f8SFilipe Manana goto done; 2634be6821f8SFilipe Manana } 2635925baeddSChris Mason 2636eb60ceacSChris Mason while (b) { 2637f624d976SQu Wenruo int dec = 0; 2638f624d976SQu Wenruo 26395f39d397SChris Mason level = btrfs_header_level(b); 264065b51a00SChris Mason 264102217ed2SChris Mason if (cow) { 26429ea2c7c9SNikolay Borisov bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); 26439ea2c7c9SNikolay Borisov 2644c8c42864SChris Mason /* 2645c8c42864SChris Mason * if we don't really need to cow this block 2646c8c42864SChris Mason * then we don't want to set the path blocking, 2647c8c42864SChris Mason * so we test it here 2648c8c42864SChris Mason */ 264964c12921SJeff Mahoney if (!should_cow_block(trans, root, b)) { 265064c12921SJeff Mahoney trans->dirty = true; 265165b51a00SChris Mason goto cow_done; 265264c12921SJeff Mahoney } 26535d4f98a2SYan Zheng 2654bd681513SChris Mason /* 2655bd681513SChris Mason * must have write locks on this node and the 2656bd681513SChris Mason * parent 2657bd681513SChris Mason */ 26585124e00eSJosef Bacik if (level > write_lock_level || 26595124e00eSJosef Bacik (level + 1 > write_lock_level && 26605124e00eSJosef Bacik level + 1 < BTRFS_MAX_LEVEL && 26615124e00eSJosef Bacik p->nodes[level + 1])) { 2662bd681513SChris Mason write_lock_level = level + 1; 2663bd681513SChris Mason btrfs_release_path(p); 2664bd681513SChris Mason goto again; 2665bd681513SChris Mason } 2666bd681513SChris Mason 26679ea2c7c9SNikolay Borisov if (last_level) 26689ea2c7c9SNikolay Borisov err = btrfs_cow_block(trans, root, b, NULL, 0, 26699631e4ccSJosef Bacik &b, 26709631e4ccSJosef Bacik BTRFS_NESTING_COW); 26719ea2c7c9SNikolay Borisov else 267233c66f43SYan Zheng err = btrfs_cow_block(trans, root, b, 2673e20d96d6SChris Mason p->nodes[level + 1], 26749631e4ccSJosef Bacik p->slots[level + 1], &b, 26759631e4ccSJosef Bacik BTRFS_NESTING_COW); 267633c66f43SYan Zheng if (err) { 267733c66f43SYan Zheng ret = err; 267865b51a00SChris Mason goto done; 267954aa1f4dSChris Mason } 268002217ed2SChris Mason } 268165b51a00SChris Mason cow_done: 2682eb60ceacSChris Mason p->nodes[level] = b; 268352398340SLiu Bo /* 268452398340SLiu Bo * Leave path with blocking locks to avoid massive 268552398340SLiu Bo * lock context switch, this is made on purpose. 268652398340SLiu Bo */ 2687b4ce94deSChris Mason 2688b4ce94deSChris Mason /* 2689b4ce94deSChris Mason * we have a lock on b and as long as we aren't changing 2690b4ce94deSChris Mason * the tree, there is no way to for the items in b to change. 2691b4ce94deSChris Mason * It is safe to drop the lock on our parent before we 2692b4ce94deSChris Mason * go through the expensive btree search on b. 2693b4ce94deSChris Mason * 2694eb653de1SFilipe David Borba Manana * If we're inserting or deleting (ins_len != 0), then we might 2695eb653de1SFilipe David Borba Manana * be changing slot zero, which may require changing the parent. 2696eb653de1SFilipe David Borba Manana * So, we can't drop the lock until after we know which slot 2697eb653de1SFilipe David Borba Manana * we're operating on. 2698b4ce94deSChris Mason */ 2699eb653de1SFilipe David Borba Manana if (!ins_len && !p->keep_locks) { 2700eb653de1SFilipe David Borba Manana int u = level + 1; 2701eb653de1SFilipe David Borba Manana 2702eb653de1SFilipe David Borba Manana if (u < BTRFS_MAX_LEVEL && p->locks[u]) { 2703eb653de1SFilipe David Borba Manana btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); 2704eb653de1SFilipe David Borba Manana p->locks[u] = 0; 2705eb653de1SFilipe David Borba Manana } 2706eb653de1SFilipe David Borba Manana } 2707b4ce94deSChris Mason 2708995e9a16SNikolay Borisov /* 2709995e9a16SNikolay Borisov * If btrfs_bin_search returns an exact match (prev_cmp == 0) 2710995e9a16SNikolay Borisov * we can safely assume the target key will always be in slot 0 2711995e9a16SNikolay Borisov * on lower levels due to the invariants BTRFS' btree provides, 2712995e9a16SNikolay Borisov * namely that a btrfs_key_ptr entry always points to the 2713995e9a16SNikolay Borisov * lowest key in the child node, thus we can skip searching 2714995e9a16SNikolay Borisov * lower levels 2715995e9a16SNikolay Borisov */ 2716995e9a16SNikolay Borisov if (prev_cmp == 0) { 2717995e9a16SNikolay Borisov slot = 0; 2718995e9a16SNikolay Borisov ret = 0; 2719995e9a16SNikolay Borisov } else { 2720995e9a16SNikolay Borisov ret = btrfs_bin_search(b, key, &slot); 2721995e9a16SNikolay Borisov prev_cmp = ret; 2722415b35a5SLiu Bo if (ret < 0) 2723415b35a5SLiu Bo goto done; 2724995e9a16SNikolay Borisov } 2725b4ce94deSChris Mason 2726f624d976SQu Wenruo if (level == 0) { 2727be0e5c09SChris Mason p->slots[level] = slot; 27289a664971Sethanwu /* 27299a664971Sethanwu * Item key already exists. In this case, if we are 27309a664971Sethanwu * allowed to insert the item (for example, in dir_item 27319a664971Sethanwu * case, item key collision is allowed), it will be 27329a664971Sethanwu * merged with the original item. Only the item size 27339a664971Sethanwu * grows, no new btrfs item will be added. If 27349a664971Sethanwu * search_for_extension is not set, ins_len already 27359a664971Sethanwu * accounts the size btrfs_item, deduct it here so leaf 27369a664971Sethanwu * space check will be correct. 27379a664971Sethanwu */ 27389a664971Sethanwu if (ret == 0 && ins_len > 0 && !p->search_for_extension) { 27399a664971Sethanwu ASSERT(ins_len >= sizeof(struct btrfs_item)); 27409a664971Sethanwu ins_len -= sizeof(struct btrfs_item); 27419a664971Sethanwu } 274287b29b20SYan Zheng if (ins_len > 0 && 2743e902baacSDavid Sterba btrfs_leaf_free_space(b) < ins_len) { 2744bd681513SChris Mason if (write_lock_level < 1) { 2745bd681513SChris Mason write_lock_level = 1; 2746bd681513SChris Mason btrfs_release_path(p); 2747bd681513SChris Mason goto again; 2748bd681513SChris Mason } 2749bd681513SChris Mason 275033c66f43SYan Zheng err = split_leaf(trans, root, key, 2751cc0c5538SChris Mason p, ins_len, ret == 0); 2752b4ce94deSChris Mason 275333c66f43SYan Zheng BUG_ON(err > 0); 275433c66f43SYan Zheng if (err) { 275533c66f43SYan Zheng ret = err; 275665b51a00SChris Mason goto done; 275765b51a00SChris Mason } 27585c680ed6SChris Mason } 2759459931ecSChris Mason if (!p->search_for_split) 2760f7c79f30SChris Mason unlock_up(p, level, lowest_unlock, 27614b6f8e96SLiu Bo min_write_lock_level, NULL); 276265b51a00SChris Mason goto done; 276365b51a00SChris Mason } 2764f624d976SQu Wenruo if (ret && slot > 0) { 2765f624d976SQu Wenruo dec = 1; 2766f624d976SQu Wenruo slot--; 2767f624d976SQu Wenruo } 2768f624d976SQu Wenruo p->slots[level] = slot; 2769f624d976SQu Wenruo err = setup_nodes_for_search(trans, root, p, b, level, ins_len, 2770f624d976SQu Wenruo &write_lock_level); 2771f624d976SQu Wenruo if (err == -EAGAIN) 2772f624d976SQu Wenruo goto again; 2773f624d976SQu Wenruo if (err) { 2774f624d976SQu Wenruo ret = err; 2775f624d976SQu Wenruo goto done; 2776f624d976SQu Wenruo } 2777f624d976SQu Wenruo b = p->nodes[level]; 2778f624d976SQu Wenruo slot = p->slots[level]; 2779f624d976SQu Wenruo 2780f624d976SQu Wenruo /* 2781f624d976SQu Wenruo * Slot 0 is special, if we change the key we have to update 2782f624d976SQu Wenruo * the parent pointer which means we must have a write lock on 2783f624d976SQu Wenruo * the parent 2784f624d976SQu Wenruo */ 2785f624d976SQu Wenruo if (slot == 0 && ins_len && write_lock_level < level + 1) { 2786f624d976SQu Wenruo write_lock_level = level + 1; 2787f624d976SQu Wenruo btrfs_release_path(p); 2788f624d976SQu Wenruo goto again; 2789f624d976SQu Wenruo } 2790f624d976SQu Wenruo 2791f624d976SQu Wenruo unlock_up(p, level, lowest_unlock, min_write_lock_level, 2792f624d976SQu Wenruo &write_lock_level); 2793f624d976SQu Wenruo 2794f624d976SQu Wenruo if (level == lowest_level) { 2795f624d976SQu Wenruo if (dec) 2796f624d976SQu Wenruo p->slots[level]++; 2797f624d976SQu Wenruo goto done; 2798f624d976SQu Wenruo } 2799f624d976SQu Wenruo 2800f624d976SQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 2801f624d976SQu Wenruo if (err == -EAGAIN) 2802f624d976SQu Wenruo goto again; 2803f624d976SQu Wenruo if (err) { 2804f624d976SQu Wenruo ret = err; 2805f624d976SQu Wenruo goto done; 2806f624d976SQu Wenruo } 2807f624d976SQu Wenruo 2808f624d976SQu Wenruo if (!p->skip_locking) { 2809f624d976SQu Wenruo level = btrfs_header_level(b); 2810f624d976SQu Wenruo if (level <= write_lock_level) { 2811f624d976SQu Wenruo btrfs_tree_lock(b); 2812f624d976SQu Wenruo p->locks[level] = BTRFS_WRITE_LOCK; 2813f624d976SQu Wenruo } else { 2814fe596ca3SJosef Bacik btrfs_tree_read_lock(b); 2815f624d976SQu Wenruo p->locks[level] = BTRFS_READ_LOCK; 2816f624d976SQu Wenruo } 2817f624d976SQu Wenruo p->nodes[level] = b; 2818f624d976SQu Wenruo } 281965b51a00SChris Mason } 282065b51a00SChris Mason ret = 1; 282165b51a00SChris Mason done: 28225f5bc6b1SFilipe Manana if (ret < 0 && !p->skip_release_on_error) 2823b3b4aa74SDavid Sterba btrfs_release_path(p); 2824be0e5c09SChris Mason return ret; 2825be0e5c09SChris Mason } 2826f75e2b79SJosef Bacik ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); 2827be0e5c09SChris Mason 282874123bd7SChris Mason /* 28295d9e75c4SJan Schmidt * Like btrfs_search_slot, this looks for a key in the given tree. It uses the 28305d9e75c4SJan Schmidt * current state of the tree together with the operations recorded in the tree 28315d9e75c4SJan Schmidt * modification log to search for the key in a previous version of this tree, as 28325d9e75c4SJan Schmidt * denoted by the time_seq parameter. 28335d9e75c4SJan Schmidt * 28345d9e75c4SJan Schmidt * Naturally, there is no support for insert, delete or cow operations. 28355d9e75c4SJan Schmidt * 28365d9e75c4SJan Schmidt * The resulting path and return value will be set up as if we called 28375d9e75c4SJan Schmidt * btrfs_search_slot at that point in time with ins_len and cow both set to 0. 28385d9e75c4SJan Schmidt */ 2839310712b2SOmar Sandoval int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, 28405d9e75c4SJan Schmidt struct btrfs_path *p, u64 time_seq) 28415d9e75c4SJan Schmidt { 28420b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 28435d9e75c4SJan Schmidt struct extent_buffer *b; 28445d9e75c4SJan Schmidt int slot; 28455d9e75c4SJan Schmidt int ret; 28465d9e75c4SJan Schmidt int err; 28475d9e75c4SJan Schmidt int level; 28485d9e75c4SJan Schmidt int lowest_unlock = 1; 28495d9e75c4SJan Schmidt u8 lowest_level = 0; 28505d9e75c4SJan Schmidt 28515d9e75c4SJan Schmidt lowest_level = p->lowest_level; 28525d9e75c4SJan Schmidt WARN_ON(p->nodes[0] != NULL); 28535d9e75c4SJan Schmidt 28545d9e75c4SJan Schmidt if (p->search_commit_root) { 28555d9e75c4SJan Schmidt BUG_ON(time_seq); 28565d9e75c4SJan Schmidt return btrfs_search_slot(NULL, root, key, p, 0, 0); 28575d9e75c4SJan Schmidt } 28585d9e75c4SJan Schmidt 28595d9e75c4SJan Schmidt again: 28605d9e75c4SJan Schmidt b = get_old_root(root, time_seq); 2861315bed43SNikolay Borisov if (!b) { 2862315bed43SNikolay Borisov ret = -EIO; 2863315bed43SNikolay Borisov goto done; 2864315bed43SNikolay Borisov } 28655d9e75c4SJan Schmidt level = btrfs_header_level(b); 28665d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 28675d9e75c4SJan Schmidt 28685d9e75c4SJan Schmidt while (b) { 2869abe9339dSQu Wenruo int dec = 0; 2870abe9339dSQu Wenruo 28715d9e75c4SJan Schmidt level = btrfs_header_level(b); 28725d9e75c4SJan Schmidt p->nodes[level] = b; 28735d9e75c4SJan Schmidt 28745d9e75c4SJan Schmidt /* 28755d9e75c4SJan Schmidt * we have a lock on b and as long as we aren't changing 28765d9e75c4SJan Schmidt * the tree, there is no way to for the items in b to change. 28775d9e75c4SJan Schmidt * It is safe to drop the lock on our parent before we 28785d9e75c4SJan Schmidt * go through the expensive btree search on b. 28795d9e75c4SJan Schmidt */ 28805d9e75c4SJan Schmidt btrfs_unlock_up_safe(p, level + 1); 28815d9e75c4SJan Schmidt 2882995e9a16SNikolay Borisov ret = btrfs_bin_search(b, key, &slot); 2883cbca7d59SFilipe Manana if (ret < 0) 2884cbca7d59SFilipe Manana goto done; 28855d9e75c4SJan Schmidt 2886abe9339dSQu Wenruo if (level == 0) { 2887abe9339dSQu Wenruo p->slots[level] = slot; 2888abe9339dSQu Wenruo unlock_up(p, level, lowest_unlock, 0, NULL); 2889abe9339dSQu Wenruo goto done; 2890abe9339dSQu Wenruo } 2891abe9339dSQu Wenruo 28925d9e75c4SJan Schmidt if (ret && slot > 0) { 28935d9e75c4SJan Schmidt dec = 1; 2894abe9339dSQu Wenruo slot--; 28955d9e75c4SJan Schmidt } 28965d9e75c4SJan Schmidt p->slots[level] = slot; 28975d9e75c4SJan Schmidt unlock_up(p, level, lowest_unlock, 0, NULL); 28985d9e75c4SJan Schmidt 28995d9e75c4SJan Schmidt if (level == lowest_level) { 29005d9e75c4SJan Schmidt if (dec) 29015d9e75c4SJan Schmidt p->slots[level]++; 29025d9e75c4SJan Schmidt goto done; 29035d9e75c4SJan Schmidt } 29045d9e75c4SJan Schmidt 2905abe9339dSQu Wenruo err = read_block_for_search(root, p, &b, level, slot, key); 29065d9e75c4SJan Schmidt if (err == -EAGAIN) 29075d9e75c4SJan Schmidt goto again; 29085d9e75c4SJan Schmidt if (err) { 29095d9e75c4SJan Schmidt ret = err; 29105d9e75c4SJan Schmidt goto done; 29115d9e75c4SJan Schmidt } 29125d9e75c4SJan Schmidt 29135d9e75c4SJan Schmidt level = btrfs_header_level(b); 29145d9e75c4SJan Schmidt btrfs_tree_read_lock(b); 29150b246afaSJeff Mahoney b = tree_mod_log_rewind(fs_info, p, b, time_seq); 2916db7f3436SJosef Bacik if (!b) { 2917db7f3436SJosef Bacik ret = -ENOMEM; 2918db7f3436SJosef Bacik goto done; 2919db7f3436SJosef Bacik } 29205d9e75c4SJan Schmidt p->locks[level] = BTRFS_READ_LOCK; 29215d9e75c4SJan Schmidt p->nodes[level] = b; 29225d9e75c4SJan Schmidt } 29235d9e75c4SJan Schmidt ret = 1; 29245d9e75c4SJan Schmidt done: 29255d9e75c4SJan Schmidt if (ret < 0) 29265d9e75c4SJan Schmidt btrfs_release_path(p); 29275d9e75c4SJan Schmidt 29285d9e75c4SJan Schmidt return ret; 29295d9e75c4SJan Schmidt } 29305d9e75c4SJan Schmidt 29315d9e75c4SJan Schmidt /* 29322f38b3e1SArne Jansen * helper to use instead of search slot if no exact match is needed but 29332f38b3e1SArne Jansen * instead the next or previous item should be returned. 29342f38b3e1SArne Jansen * When find_higher is true, the next higher item is returned, the next lower 29352f38b3e1SArne Jansen * otherwise. 29362f38b3e1SArne Jansen * When return_any and find_higher are both true, and no higher item is found, 29372f38b3e1SArne Jansen * return the next lower instead. 29382f38b3e1SArne Jansen * When return_any is true and find_higher is false, and no lower item is found, 29392f38b3e1SArne Jansen * return the next higher instead. 29402f38b3e1SArne Jansen * It returns 0 if any item is found, 1 if none is found (tree empty), and 29412f38b3e1SArne Jansen * < 0 on error 29422f38b3e1SArne Jansen */ 29432f38b3e1SArne Jansen int btrfs_search_slot_for_read(struct btrfs_root *root, 2944310712b2SOmar Sandoval const struct btrfs_key *key, 2945310712b2SOmar Sandoval struct btrfs_path *p, int find_higher, 2946310712b2SOmar Sandoval int return_any) 29472f38b3e1SArne Jansen { 29482f38b3e1SArne Jansen int ret; 29492f38b3e1SArne Jansen struct extent_buffer *leaf; 29502f38b3e1SArne Jansen 29512f38b3e1SArne Jansen again: 29522f38b3e1SArne Jansen ret = btrfs_search_slot(NULL, root, key, p, 0, 0); 29532f38b3e1SArne Jansen if (ret <= 0) 29542f38b3e1SArne Jansen return ret; 29552f38b3e1SArne Jansen /* 29562f38b3e1SArne Jansen * a return value of 1 means the path is at the position where the 29572f38b3e1SArne Jansen * item should be inserted. Normally this is the next bigger item, 29582f38b3e1SArne Jansen * but in case the previous item is the last in a leaf, path points 29592f38b3e1SArne Jansen * to the first free slot in the previous leaf, i.e. at an invalid 29602f38b3e1SArne Jansen * item. 29612f38b3e1SArne Jansen */ 29622f38b3e1SArne Jansen leaf = p->nodes[0]; 29632f38b3e1SArne Jansen 29642f38b3e1SArne Jansen if (find_higher) { 29652f38b3e1SArne Jansen if (p->slots[0] >= btrfs_header_nritems(leaf)) { 29662f38b3e1SArne Jansen ret = btrfs_next_leaf(root, p); 29672f38b3e1SArne Jansen if (ret <= 0) 29682f38b3e1SArne Jansen return ret; 29692f38b3e1SArne Jansen if (!return_any) 29702f38b3e1SArne Jansen return 1; 29712f38b3e1SArne Jansen /* 29722f38b3e1SArne Jansen * no higher item found, return the next 29732f38b3e1SArne Jansen * lower instead 29742f38b3e1SArne Jansen */ 29752f38b3e1SArne Jansen return_any = 0; 29762f38b3e1SArne Jansen find_higher = 0; 29772f38b3e1SArne Jansen btrfs_release_path(p); 29782f38b3e1SArne Jansen goto again; 29792f38b3e1SArne Jansen } 29802f38b3e1SArne Jansen } else { 29812f38b3e1SArne Jansen if (p->slots[0] == 0) { 29822f38b3e1SArne Jansen ret = btrfs_prev_leaf(root, p); 2983e6793769SArne Jansen if (ret < 0) 29842f38b3e1SArne Jansen return ret; 2985e6793769SArne Jansen if (!ret) { 298623c6bf6aSFilipe David Borba Manana leaf = p->nodes[0]; 298723c6bf6aSFilipe David Borba Manana if (p->slots[0] == btrfs_header_nritems(leaf)) 298823c6bf6aSFilipe David Borba Manana p->slots[0]--; 2989e6793769SArne Jansen return 0; 2990e6793769SArne Jansen } 29912f38b3e1SArne Jansen if (!return_any) 29922f38b3e1SArne Jansen return 1; 29932f38b3e1SArne Jansen /* 29942f38b3e1SArne Jansen * no lower item found, return the next 29952f38b3e1SArne Jansen * higher instead 29962f38b3e1SArne Jansen */ 29972f38b3e1SArne Jansen return_any = 0; 29982f38b3e1SArne Jansen find_higher = 1; 29992f38b3e1SArne Jansen btrfs_release_path(p); 30002f38b3e1SArne Jansen goto again; 3001e6793769SArne Jansen } else { 30022f38b3e1SArne Jansen --p->slots[0]; 30032f38b3e1SArne Jansen } 30042f38b3e1SArne Jansen } 30052f38b3e1SArne Jansen return 0; 30062f38b3e1SArne Jansen } 30072f38b3e1SArne Jansen 30082f38b3e1SArne Jansen /* 300974123bd7SChris Mason * adjust the pointers going up the tree, starting at level 301074123bd7SChris Mason * making sure the right key of each node is points to 'key'. 301174123bd7SChris Mason * This is used after shifting pointers to the left, so it stops 301274123bd7SChris Mason * fixing up pointers when a given leaf/node is not in slot 0 of the 301374123bd7SChris Mason * higher levels 3014aa5d6bedSChris Mason * 301574123bd7SChris Mason */ 3016b167fa91SNikolay Borisov static void fixup_low_keys(struct btrfs_path *path, 30175f39d397SChris Mason struct btrfs_disk_key *key, int level) 3018be0e5c09SChris Mason { 3019be0e5c09SChris Mason int i; 30205f39d397SChris Mason struct extent_buffer *t; 30210e82bcfeSDavid Sterba int ret; 30225f39d397SChris Mason 3023234b63a0SChris Mason for (i = level; i < BTRFS_MAX_LEVEL; i++) { 3024be0e5c09SChris Mason int tslot = path->slots[i]; 30250e82bcfeSDavid Sterba 3026eb60ceacSChris Mason if (!path->nodes[i]) 3027be0e5c09SChris Mason break; 30285f39d397SChris Mason t = path->nodes[i]; 30290e82bcfeSDavid Sterba ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, 30300e82bcfeSDavid Sterba GFP_ATOMIC); 30310e82bcfeSDavid Sterba BUG_ON(ret < 0); 30325f39d397SChris Mason btrfs_set_node_key(t, key, tslot); 3033d6025579SChris Mason btrfs_mark_buffer_dirty(path->nodes[i]); 3034be0e5c09SChris Mason if (tslot != 0) 3035be0e5c09SChris Mason break; 3036be0e5c09SChris Mason } 3037be0e5c09SChris Mason } 3038be0e5c09SChris Mason 303974123bd7SChris Mason /* 304031840ae1SZheng Yan * update item key. 304131840ae1SZheng Yan * 304231840ae1SZheng Yan * This function isn't completely safe. It's the caller's responsibility 304331840ae1SZheng Yan * that the new key won't break the order 304431840ae1SZheng Yan */ 3045b7a0365eSDaniel Dressler void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, 3046b7a0365eSDaniel Dressler struct btrfs_path *path, 3047310712b2SOmar Sandoval const struct btrfs_key *new_key) 304831840ae1SZheng Yan { 304931840ae1SZheng Yan struct btrfs_disk_key disk_key; 305031840ae1SZheng Yan struct extent_buffer *eb; 305131840ae1SZheng Yan int slot; 305231840ae1SZheng Yan 305331840ae1SZheng Yan eb = path->nodes[0]; 305431840ae1SZheng Yan slot = path->slots[0]; 305531840ae1SZheng Yan if (slot > 0) { 305631840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot - 1); 30577c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { 30587c15d410SQu Wenruo btrfs_crit(fs_info, 30597c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 30607c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 30617c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 30627c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 30637c15d410SQu Wenruo new_key->objectid, new_key->type, 30647c15d410SQu Wenruo new_key->offset); 30657c15d410SQu Wenruo btrfs_print_leaf(eb); 30667c15d410SQu Wenruo BUG(); 30677c15d410SQu Wenruo } 306831840ae1SZheng Yan } 306931840ae1SZheng Yan if (slot < btrfs_header_nritems(eb) - 1) { 307031840ae1SZheng Yan btrfs_item_key(eb, &disk_key, slot + 1); 30717c15d410SQu Wenruo if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { 30727c15d410SQu Wenruo btrfs_crit(fs_info, 30737c15d410SQu Wenruo "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 30747c15d410SQu Wenruo slot, btrfs_disk_key_objectid(&disk_key), 30757c15d410SQu Wenruo btrfs_disk_key_type(&disk_key), 30767c15d410SQu Wenruo btrfs_disk_key_offset(&disk_key), 30777c15d410SQu Wenruo new_key->objectid, new_key->type, 30787c15d410SQu Wenruo new_key->offset); 30797c15d410SQu Wenruo btrfs_print_leaf(eb); 30807c15d410SQu Wenruo BUG(); 30817c15d410SQu Wenruo } 308231840ae1SZheng Yan } 308331840ae1SZheng Yan 308431840ae1SZheng Yan btrfs_cpu_key_to_disk(&disk_key, new_key); 308531840ae1SZheng Yan btrfs_set_item_key(eb, &disk_key, slot); 308631840ae1SZheng Yan btrfs_mark_buffer_dirty(eb); 308731840ae1SZheng Yan if (slot == 0) 3088b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 308931840ae1SZheng Yan } 309031840ae1SZheng Yan 309131840ae1SZheng Yan /* 3092d16c702fSQu Wenruo * Check key order of two sibling extent buffers. 3093d16c702fSQu Wenruo * 3094d16c702fSQu Wenruo * Return true if something is wrong. 3095d16c702fSQu Wenruo * Return false if everything is fine. 3096d16c702fSQu Wenruo * 3097d16c702fSQu Wenruo * Tree-checker only works inside one tree block, thus the following 3098d16c702fSQu Wenruo * corruption can not be detected by tree-checker: 3099d16c702fSQu Wenruo * 3100d16c702fSQu Wenruo * Leaf @left | Leaf @right 3101d16c702fSQu Wenruo * -------------------------------------------------------------- 3102d16c702fSQu Wenruo * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | 3103d16c702fSQu Wenruo * 3104d16c702fSQu Wenruo * Key f6 in leaf @left itself is valid, but not valid when the next 3105d16c702fSQu Wenruo * key in leaf @right is 7. 3106d16c702fSQu Wenruo * This can only be checked at tree block merge time. 3107d16c702fSQu Wenruo * And since tree checker has ensured all key order in each tree block 3108d16c702fSQu Wenruo * is correct, we only need to bother the last key of @left and the first 3109d16c702fSQu Wenruo * key of @right. 3110d16c702fSQu Wenruo */ 3111d16c702fSQu Wenruo static bool check_sibling_keys(struct extent_buffer *left, 3112d16c702fSQu Wenruo struct extent_buffer *right) 3113d16c702fSQu Wenruo { 3114d16c702fSQu Wenruo struct btrfs_key left_last; 3115d16c702fSQu Wenruo struct btrfs_key right_first; 3116d16c702fSQu Wenruo int level = btrfs_header_level(left); 3117d16c702fSQu Wenruo int nr_left = btrfs_header_nritems(left); 3118d16c702fSQu Wenruo int nr_right = btrfs_header_nritems(right); 3119d16c702fSQu Wenruo 3120d16c702fSQu Wenruo /* No key to check in one of the tree blocks */ 3121d16c702fSQu Wenruo if (!nr_left || !nr_right) 3122d16c702fSQu Wenruo return false; 3123d16c702fSQu Wenruo 3124d16c702fSQu Wenruo if (level) { 3125d16c702fSQu Wenruo btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); 3126d16c702fSQu Wenruo btrfs_node_key_to_cpu(right, &right_first, 0); 3127d16c702fSQu Wenruo } else { 3128d16c702fSQu Wenruo btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); 3129d16c702fSQu Wenruo btrfs_item_key_to_cpu(right, &right_first, 0); 3130d16c702fSQu Wenruo } 3131d16c702fSQu Wenruo 3132d16c702fSQu Wenruo if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) { 3133d16c702fSQu Wenruo btrfs_crit(left->fs_info, 3134d16c702fSQu Wenruo "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", 3135d16c702fSQu Wenruo left_last.objectid, left_last.type, 3136d16c702fSQu Wenruo left_last.offset, right_first.objectid, 3137d16c702fSQu Wenruo right_first.type, right_first.offset); 3138d16c702fSQu Wenruo return true; 3139d16c702fSQu Wenruo } 3140d16c702fSQu Wenruo return false; 3141d16c702fSQu Wenruo } 3142d16c702fSQu Wenruo 3143d16c702fSQu Wenruo /* 314474123bd7SChris Mason * try to push data from one node into the next node left in the 314579f95c82SChris Mason * tree. 3146aa5d6bedSChris Mason * 3147aa5d6bedSChris Mason * returns 0 if some ptrs were pushed left, < 0 if there was some horrible 3148aa5d6bedSChris Mason * error, and > 0 if there was no room in the left hand block. 314974123bd7SChris Mason */ 315098ed5174SChris Mason static int push_node_left(struct btrfs_trans_handle *trans, 31512ff7e61eSJeff Mahoney struct extent_buffer *dst, 3152971a1f66SChris Mason struct extent_buffer *src, int empty) 3153be0e5c09SChris Mason { 3154d30a668fSDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 3155be0e5c09SChris Mason int push_items = 0; 3156bb803951SChris Mason int src_nritems; 3157bb803951SChris Mason int dst_nritems; 3158aa5d6bedSChris Mason int ret = 0; 3159be0e5c09SChris Mason 31605f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 31615f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 31620b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 31637bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 31647bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 316554aa1f4dSChris Mason 3166bce4eae9SChris Mason if (!empty && src_nritems <= 8) 3167971a1f66SChris Mason return 1; 3168971a1f66SChris Mason 3169d397712bSChris Mason if (push_items <= 0) 3170be0e5c09SChris Mason return 1; 3171be0e5c09SChris Mason 3172bce4eae9SChris Mason if (empty) { 3173971a1f66SChris Mason push_items = min(src_nritems, push_items); 3174bce4eae9SChris Mason if (push_items < src_nritems) { 3175bce4eae9SChris Mason /* leave at least 8 pointers in the node if 3176bce4eae9SChris Mason * we aren't going to empty it 3177bce4eae9SChris Mason */ 3178bce4eae9SChris Mason if (src_nritems - push_items < 8) { 3179bce4eae9SChris Mason if (push_items <= 8) 3180bce4eae9SChris Mason return 1; 3181bce4eae9SChris Mason push_items -= 8; 3182bce4eae9SChris Mason } 3183bce4eae9SChris Mason } 3184bce4eae9SChris Mason } else 3185bce4eae9SChris Mason push_items = min(src_nritems - 8, push_items); 318679f95c82SChris Mason 3187d16c702fSQu Wenruo /* dst is the left eb, src is the middle eb */ 3188d16c702fSQu Wenruo if (check_sibling_keys(dst, src)) { 3189d16c702fSQu Wenruo ret = -EUCLEAN; 3190d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 3191d16c702fSQu Wenruo return ret; 3192d16c702fSQu Wenruo } 3193ed874f0dSDavid Sterba ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); 31945de865eeSFilipe David Borba Manana if (ret) { 319566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 31965de865eeSFilipe David Borba Manana return ret; 31975de865eeSFilipe David Borba Manana } 31985f39d397SChris Mason copy_extent_buffer(dst, src, 31995f39d397SChris Mason btrfs_node_key_ptr_offset(dst_nritems), 32005f39d397SChris Mason btrfs_node_key_ptr_offset(0), 3201123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 32025f39d397SChris Mason 3203bb803951SChris Mason if (push_items < src_nritems) { 320457911b8bSJan Schmidt /* 3205bf1d3425SDavid Sterba * Don't call tree_mod_log_insert_move here, key removal was 3206bf1d3425SDavid Sterba * already fully logged by tree_mod_log_eb_copy above. 320757911b8bSJan Schmidt */ 32085f39d397SChris Mason memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), 32095f39d397SChris Mason btrfs_node_key_ptr_offset(push_items), 3210e2fa7227SChris Mason (src_nritems - push_items) * 3211123abc88SChris Mason sizeof(struct btrfs_key_ptr)); 3212bb803951SChris Mason } 32135f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 32145f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 32155f39d397SChris Mason btrfs_mark_buffer_dirty(src); 32165f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 321731840ae1SZheng Yan 3218bb803951SChris Mason return ret; 3219be0e5c09SChris Mason } 3220be0e5c09SChris Mason 322197571fd0SChris Mason /* 322279f95c82SChris Mason * try to push data from one node into the next node right in the 322379f95c82SChris Mason * tree. 322479f95c82SChris Mason * 322579f95c82SChris Mason * returns 0 if some ptrs were pushed, < 0 if there was some horrible 322679f95c82SChris Mason * error, and > 0 if there was no room in the right hand block. 322779f95c82SChris Mason * 322879f95c82SChris Mason * this will only push up to 1/2 the contents of the left node over 322979f95c82SChris Mason */ 32305f39d397SChris Mason static int balance_node_right(struct btrfs_trans_handle *trans, 32315f39d397SChris Mason struct extent_buffer *dst, 32325f39d397SChris Mason struct extent_buffer *src) 323379f95c82SChris Mason { 323455d32ed8SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 323579f95c82SChris Mason int push_items = 0; 323679f95c82SChris Mason int max_push; 323779f95c82SChris Mason int src_nritems; 323879f95c82SChris Mason int dst_nritems; 323979f95c82SChris Mason int ret = 0; 324079f95c82SChris Mason 32417bb86316SChris Mason WARN_ON(btrfs_header_generation(src) != trans->transid); 32427bb86316SChris Mason WARN_ON(btrfs_header_generation(dst) != trans->transid); 32437bb86316SChris Mason 32445f39d397SChris Mason src_nritems = btrfs_header_nritems(src); 32455f39d397SChris Mason dst_nritems = btrfs_header_nritems(dst); 32460b246afaSJeff Mahoney push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 3247d397712bSChris Mason if (push_items <= 0) 324879f95c82SChris Mason return 1; 3249bce4eae9SChris Mason 3250d397712bSChris Mason if (src_nritems < 4) 3251bce4eae9SChris Mason return 1; 325279f95c82SChris Mason 325379f95c82SChris Mason max_push = src_nritems / 2 + 1; 325479f95c82SChris Mason /* don't try to empty the node */ 3255d397712bSChris Mason if (max_push >= src_nritems) 325679f95c82SChris Mason return 1; 3257252c38f0SYan 325879f95c82SChris Mason if (max_push < push_items) 325979f95c82SChris Mason push_items = max_push; 326079f95c82SChris Mason 3261d16c702fSQu Wenruo /* dst is the right eb, src is the middle eb */ 3262d16c702fSQu Wenruo if (check_sibling_keys(src, dst)) { 3263d16c702fSQu Wenruo ret = -EUCLEAN; 3264d16c702fSQu Wenruo btrfs_abort_transaction(trans, ret); 3265d16c702fSQu Wenruo return ret; 3266d16c702fSQu Wenruo } 3267bf1d3425SDavid Sterba ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); 3268bf1d3425SDavid Sterba BUG_ON(ret < 0); 32695f39d397SChris Mason memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), 32705f39d397SChris Mason btrfs_node_key_ptr_offset(0), 32715f39d397SChris Mason (dst_nritems) * 32725f39d397SChris Mason sizeof(struct btrfs_key_ptr)); 3273d6025579SChris Mason 3274ed874f0dSDavid Sterba ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, 3275ed874f0dSDavid Sterba push_items); 32765de865eeSFilipe David Borba Manana if (ret) { 327766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 32785de865eeSFilipe David Borba Manana return ret; 32795de865eeSFilipe David Borba Manana } 32805f39d397SChris Mason copy_extent_buffer(dst, src, 32815f39d397SChris Mason btrfs_node_key_ptr_offset(0), 32825f39d397SChris Mason btrfs_node_key_ptr_offset(src_nritems - push_items), 3283123abc88SChris Mason push_items * sizeof(struct btrfs_key_ptr)); 328479f95c82SChris Mason 32855f39d397SChris Mason btrfs_set_header_nritems(src, src_nritems - push_items); 32865f39d397SChris Mason btrfs_set_header_nritems(dst, dst_nritems + push_items); 328779f95c82SChris Mason 32885f39d397SChris Mason btrfs_mark_buffer_dirty(src); 32895f39d397SChris Mason btrfs_mark_buffer_dirty(dst); 329031840ae1SZheng Yan 329179f95c82SChris Mason return ret; 329279f95c82SChris Mason } 329379f95c82SChris Mason 329479f95c82SChris Mason /* 329597571fd0SChris Mason * helper function to insert a new root level in the tree. 329697571fd0SChris Mason * A new node is allocated, and a single item is inserted to 329797571fd0SChris Mason * point to the existing root 3298aa5d6bedSChris Mason * 3299aa5d6bedSChris Mason * returns zero on success or < 0 on failure. 330097571fd0SChris Mason */ 3301d397712bSChris Mason static noinline int insert_new_root(struct btrfs_trans_handle *trans, 33025f39d397SChris Mason struct btrfs_root *root, 3303fdd99c72SLiu Bo struct btrfs_path *path, int level) 330474123bd7SChris Mason { 33050b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 33067bb86316SChris Mason u64 lower_gen; 33075f39d397SChris Mason struct extent_buffer *lower; 33085f39d397SChris Mason struct extent_buffer *c; 3309925baeddSChris Mason struct extent_buffer *old; 33105f39d397SChris Mason struct btrfs_disk_key lower_key; 3311d9d19a01SDavid Sterba int ret; 33125c680ed6SChris Mason 33135c680ed6SChris Mason BUG_ON(path->nodes[level]); 33145c680ed6SChris Mason BUG_ON(path->nodes[level-1] != root->node); 33155c680ed6SChris Mason 33167bb86316SChris Mason lower = path->nodes[level-1]; 33177bb86316SChris Mason if (level == 1) 33187bb86316SChris Mason btrfs_item_key(lower, &lower_key, 0); 33197bb86316SChris Mason else 33207bb86316SChris Mason btrfs_node_key(lower, &lower_key, 0); 33217bb86316SChris Mason 3322a6279470SFilipe Manana c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level, 33239631e4ccSJosef Bacik root->node->start, 0, 3324cf6f34aaSJosef Bacik BTRFS_NESTING_NEW_ROOT); 33255f39d397SChris Mason if (IS_ERR(c)) 33265f39d397SChris Mason return PTR_ERR(c); 3327925baeddSChris Mason 33280b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 3329f0486c68SYan, Zheng 33305f39d397SChris Mason btrfs_set_header_nritems(c, 1); 33315f39d397SChris Mason btrfs_set_node_key(c, &lower_key, 0); 3332db94535dSChris Mason btrfs_set_node_blockptr(c, 0, lower->start); 33337bb86316SChris Mason lower_gen = btrfs_header_generation(lower); 333431840ae1SZheng Yan WARN_ON(lower_gen != trans->transid); 33357bb86316SChris Mason 33367bb86316SChris Mason btrfs_set_node_ptr_generation(c, 0, lower_gen); 33375f39d397SChris Mason 33385f39d397SChris Mason btrfs_mark_buffer_dirty(c); 3339d5719762SChris Mason 3340925baeddSChris Mason old = root->node; 3341d9d19a01SDavid Sterba ret = tree_mod_log_insert_root(root->node, c, 0); 3342d9d19a01SDavid Sterba BUG_ON(ret < 0); 3343240f62c8SChris Mason rcu_assign_pointer(root->node, c); 3344925baeddSChris Mason 3345925baeddSChris Mason /* the super has an extra ref to root->node */ 3346925baeddSChris Mason free_extent_buffer(old); 3347925baeddSChris Mason 33480b86a832SChris Mason add_root_to_dirty_list(root); 334967439dadSDavid Sterba atomic_inc(&c->refs); 33505f39d397SChris Mason path->nodes[level] = c; 3351ac5887c8SJosef Bacik path->locks[level] = BTRFS_WRITE_LOCK; 335274123bd7SChris Mason path->slots[level] = 0; 335374123bd7SChris Mason return 0; 335474123bd7SChris Mason } 33555c680ed6SChris Mason 33565c680ed6SChris Mason /* 33575c680ed6SChris Mason * worker function to insert a single pointer in a node. 33585c680ed6SChris Mason * the node should have enough room for the pointer already 335997571fd0SChris Mason * 33605c680ed6SChris Mason * slot and level indicate where you want the key to go, and 33615c680ed6SChris Mason * blocknr is the block the key points to. 33625c680ed6SChris Mason */ 3363143bede5SJeff Mahoney static void insert_ptr(struct btrfs_trans_handle *trans, 33646ad3cf6dSDavid Sterba struct btrfs_path *path, 3365143bede5SJeff Mahoney struct btrfs_disk_key *key, u64 bytenr, 3366c3e06965SJan Schmidt int slot, int level) 33675c680ed6SChris Mason { 33685f39d397SChris Mason struct extent_buffer *lower; 33695c680ed6SChris Mason int nritems; 3370f3ea38daSJan Schmidt int ret; 33715c680ed6SChris Mason 33725c680ed6SChris Mason BUG_ON(!path->nodes[level]); 3373f0486c68SYan, Zheng btrfs_assert_tree_locked(path->nodes[level]); 33745f39d397SChris Mason lower = path->nodes[level]; 33755f39d397SChris Mason nritems = btrfs_header_nritems(lower); 3376c293498bSStoyan Gaydarov BUG_ON(slot > nritems); 33776ad3cf6dSDavid Sterba BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); 337874123bd7SChris Mason if (slot != nritems) { 3379bf1d3425SDavid Sterba if (level) { 3380bf1d3425SDavid Sterba ret = tree_mod_log_insert_move(lower, slot + 1, slot, 3381a446a979SDavid Sterba nritems - slot); 3382bf1d3425SDavid Sterba BUG_ON(ret < 0); 3383bf1d3425SDavid Sterba } 33845f39d397SChris Mason memmove_extent_buffer(lower, 33855f39d397SChris Mason btrfs_node_key_ptr_offset(slot + 1), 33865f39d397SChris Mason btrfs_node_key_ptr_offset(slot), 3387123abc88SChris Mason (nritems - slot) * sizeof(struct btrfs_key_ptr)); 338874123bd7SChris Mason } 3389c3e06965SJan Schmidt if (level) { 3390e09c2efeSDavid Sterba ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, 3391e09c2efeSDavid Sterba GFP_NOFS); 3392f3ea38daSJan Schmidt BUG_ON(ret < 0); 3393f3ea38daSJan Schmidt } 33945f39d397SChris Mason btrfs_set_node_key(lower, key, slot); 3395db94535dSChris Mason btrfs_set_node_blockptr(lower, slot, bytenr); 339674493f7aSChris Mason WARN_ON(trans->transid == 0); 339774493f7aSChris Mason btrfs_set_node_ptr_generation(lower, slot, trans->transid); 33985f39d397SChris Mason btrfs_set_header_nritems(lower, nritems + 1); 33995f39d397SChris Mason btrfs_mark_buffer_dirty(lower); 340074123bd7SChris Mason } 340174123bd7SChris Mason 340297571fd0SChris Mason /* 340397571fd0SChris Mason * split the node at the specified level in path in two. 340497571fd0SChris Mason * The path is corrected to point to the appropriate node after the split 340597571fd0SChris Mason * 340697571fd0SChris Mason * Before splitting this tries to make some room in the node by pushing 340797571fd0SChris Mason * left and right, if either one works, it returns right away. 3408aa5d6bedSChris Mason * 3409aa5d6bedSChris Mason * returns 0 on success and < 0 on failure 341097571fd0SChris Mason */ 3411e02119d5SChris Mason static noinline int split_node(struct btrfs_trans_handle *trans, 3412e02119d5SChris Mason struct btrfs_root *root, 3413e02119d5SChris Mason struct btrfs_path *path, int level) 3414be0e5c09SChris Mason { 34150b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 34165f39d397SChris Mason struct extent_buffer *c; 34175f39d397SChris Mason struct extent_buffer *split; 34185f39d397SChris Mason struct btrfs_disk_key disk_key; 3419be0e5c09SChris Mason int mid; 34205c680ed6SChris Mason int ret; 34217518a238SChris Mason u32 c_nritems; 3422be0e5c09SChris Mason 34235f39d397SChris Mason c = path->nodes[level]; 34247bb86316SChris Mason WARN_ON(btrfs_header_generation(c) != trans->transid); 34255f39d397SChris Mason if (c == root->node) { 3426d9abbf1cSJan Schmidt /* 342790f8d62eSJan Schmidt * trying to split the root, lets make a new one 342890f8d62eSJan Schmidt * 3429fdd99c72SLiu Bo * tree mod log: We don't log_removal old root in 343090f8d62eSJan Schmidt * insert_new_root, because that root buffer will be kept as a 343190f8d62eSJan Schmidt * normal node. We are going to log removal of half of the 343290f8d62eSJan Schmidt * elements below with tree_mod_log_eb_copy. We're holding a 343390f8d62eSJan Schmidt * tree lock on the buffer, which is why we cannot race with 343490f8d62eSJan Schmidt * other tree_mod_log users. 3435d9abbf1cSJan Schmidt */ 3436fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, level + 1); 34375c680ed6SChris Mason if (ret) 34385c680ed6SChris Mason return ret; 3439b3612421SChris Mason } else { 3440e66f709bSChris Mason ret = push_nodes_for_insert(trans, root, path, level); 34415f39d397SChris Mason c = path->nodes[level]; 34425f39d397SChris Mason if (!ret && btrfs_header_nritems(c) < 34430b246afaSJeff Mahoney BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) 3444e66f709bSChris Mason return 0; 344554aa1f4dSChris Mason if (ret < 0) 344654aa1f4dSChris Mason return ret; 34475c680ed6SChris Mason } 3448e66f709bSChris Mason 34495f39d397SChris Mason c_nritems = btrfs_header_nritems(c); 34505d4f98a2SYan Zheng mid = (c_nritems + 1) / 2; 34515d4f98a2SYan Zheng btrfs_node_key(c, &disk_key, mid); 34527bb86316SChris Mason 3453a6279470SFilipe Manana split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level, 34544dff97e6SJosef Bacik c->start, 0, BTRFS_NESTING_SPLIT); 34555f39d397SChris Mason if (IS_ERR(split)) 34565f39d397SChris Mason return PTR_ERR(split); 345754aa1f4dSChris Mason 34580b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 3459bc877d28SNikolay Borisov ASSERT(btrfs_header_level(c) == level); 34605f39d397SChris Mason 3461ed874f0dSDavid Sterba ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); 34625de865eeSFilipe David Borba Manana if (ret) { 346366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 34645de865eeSFilipe David Borba Manana return ret; 34655de865eeSFilipe David Borba Manana } 34665f39d397SChris Mason copy_extent_buffer(split, c, 34675f39d397SChris Mason btrfs_node_key_ptr_offset(0), 34685f39d397SChris Mason btrfs_node_key_ptr_offset(mid), 3469123abc88SChris Mason (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); 34705f39d397SChris Mason btrfs_set_header_nritems(split, c_nritems - mid); 34715f39d397SChris Mason btrfs_set_header_nritems(c, mid); 3472aa5d6bedSChris Mason 34735f39d397SChris Mason btrfs_mark_buffer_dirty(c); 34745f39d397SChris Mason btrfs_mark_buffer_dirty(split); 34755f39d397SChris Mason 34766ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, split->start, 3477c3e06965SJan Schmidt path->slots[level + 1] + 1, level + 1); 3478aa5d6bedSChris Mason 34795de08d7dSChris Mason if (path->slots[level] >= mid) { 34805c680ed6SChris Mason path->slots[level] -= mid; 3481925baeddSChris Mason btrfs_tree_unlock(c); 34825f39d397SChris Mason free_extent_buffer(c); 34835f39d397SChris Mason path->nodes[level] = split; 34845c680ed6SChris Mason path->slots[level + 1] += 1; 3485eb60ceacSChris Mason } else { 3486925baeddSChris Mason btrfs_tree_unlock(split); 34875f39d397SChris Mason free_extent_buffer(split); 3488be0e5c09SChris Mason } 3489d5286a92SNikolay Borisov return 0; 3490be0e5c09SChris Mason } 3491be0e5c09SChris Mason 349274123bd7SChris Mason /* 349374123bd7SChris Mason * how many bytes are required to store the items in a leaf. start 349474123bd7SChris Mason * and nr indicate which items in the leaf to check. This totals up the 349574123bd7SChris Mason * space used both by the item structs and the item data 349674123bd7SChris Mason */ 34975f39d397SChris Mason static int leaf_space_used(struct extent_buffer *l, int start, int nr) 3498be0e5c09SChris Mason { 349941be1f3bSJosef Bacik struct btrfs_item *start_item; 350041be1f3bSJosef Bacik struct btrfs_item *end_item; 3501be0e5c09SChris Mason int data_len; 35025f39d397SChris Mason int nritems = btrfs_header_nritems(l); 3503d4dbff95SChris Mason int end = min(nritems, start + nr) - 1; 3504be0e5c09SChris Mason 3505be0e5c09SChris Mason if (!nr) 3506be0e5c09SChris Mason return 0; 3507dd3cc16bSRoss Kirk start_item = btrfs_item_nr(start); 3508dd3cc16bSRoss Kirk end_item = btrfs_item_nr(end); 3509a31356b9SDavid Sterba data_len = btrfs_item_offset(l, start_item) + 3510a31356b9SDavid Sterba btrfs_item_size(l, start_item); 3511a31356b9SDavid Sterba data_len = data_len - btrfs_item_offset(l, end_item); 35120783fcfcSChris Mason data_len += sizeof(struct btrfs_item) * nr; 3513d4dbff95SChris Mason WARN_ON(data_len < 0); 3514be0e5c09SChris Mason return data_len; 3515be0e5c09SChris Mason } 3516be0e5c09SChris Mason 351774123bd7SChris Mason /* 3518d4dbff95SChris Mason * The space between the end of the leaf items and 3519d4dbff95SChris Mason * the start of the leaf data. IOW, how much room 3520d4dbff95SChris Mason * the leaf has left for both items and data 3521d4dbff95SChris Mason */ 3522e902baacSDavid Sterba noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) 3523d4dbff95SChris Mason { 3524e902baacSDavid Sterba struct btrfs_fs_info *fs_info = leaf->fs_info; 35255f39d397SChris Mason int nritems = btrfs_header_nritems(leaf); 35265f39d397SChris Mason int ret; 35270b246afaSJeff Mahoney 35280b246afaSJeff Mahoney ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); 35295f39d397SChris Mason if (ret < 0) { 35300b246afaSJeff Mahoney btrfs_crit(fs_info, 3531efe120a0SFrank Holton "leaf free space ret %d, leaf data size %lu, used %d nritems %d", 3532da17066cSJeff Mahoney ret, 35330b246afaSJeff Mahoney (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), 35345f39d397SChris Mason leaf_space_used(leaf, 0, nritems), nritems); 35355f39d397SChris Mason } 35365f39d397SChris Mason return ret; 3537d4dbff95SChris Mason } 3538d4dbff95SChris Mason 353999d8f83cSChris Mason /* 354099d8f83cSChris Mason * min slot controls the lowest index we're willing to push to the 354199d8f83cSChris Mason * right. We'll push up to and including min_slot, but no lower 354299d8f83cSChris Mason */ 3543f72f0010SDavid Sterba static noinline int __push_leaf_right(struct btrfs_path *path, 354444871b1bSChris Mason int data_size, int empty, 354544871b1bSChris Mason struct extent_buffer *right, 354699d8f83cSChris Mason int free_space, u32 left_nritems, 354799d8f83cSChris Mason u32 min_slot) 354800ec4c51SChris Mason { 3549f72f0010SDavid Sterba struct btrfs_fs_info *fs_info = right->fs_info; 35505f39d397SChris Mason struct extent_buffer *left = path->nodes[0]; 355144871b1bSChris Mason struct extent_buffer *upper = path->nodes[1]; 3552cfed81a0SChris Mason struct btrfs_map_token token; 35535f39d397SChris Mason struct btrfs_disk_key disk_key; 355400ec4c51SChris Mason int slot; 355534a38218SChris Mason u32 i; 355600ec4c51SChris Mason int push_space = 0; 355700ec4c51SChris Mason int push_items = 0; 35580783fcfcSChris Mason struct btrfs_item *item; 355934a38218SChris Mason u32 nr; 35607518a238SChris Mason u32 right_nritems; 35615f39d397SChris Mason u32 data_end; 3562db94535dSChris Mason u32 this_item_size; 356300ec4c51SChris Mason 356434a38218SChris Mason if (empty) 356534a38218SChris Mason nr = 0; 356634a38218SChris Mason else 356799d8f83cSChris Mason nr = max_t(u32, 1, min_slot); 356834a38218SChris Mason 356931840ae1SZheng Yan if (path->slots[0] >= left_nritems) 357087b29b20SYan Zheng push_space += data_size; 357131840ae1SZheng Yan 357244871b1bSChris Mason slot = path->slots[1]; 357334a38218SChris Mason i = left_nritems - 1; 357434a38218SChris Mason while (i >= nr) { 3575dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3576db94535dSChris Mason 357731840ae1SZheng Yan if (!empty && push_items > 0) { 357831840ae1SZheng Yan if (path->slots[0] > i) 357931840ae1SZheng Yan break; 358031840ae1SZheng Yan if (path->slots[0] == i) { 3581e902baacSDavid Sterba int space = btrfs_leaf_free_space(left); 3582e902baacSDavid Sterba 358331840ae1SZheng Yan if (space + push_space * 2 > free_space) 358431840ae1SZheng Yan break; 358531840ae1SZheng Yan } 358631840ae1SZheng Yan } 358731840ae1SZheng Yan 358800ec4c51SChris Mason if (path->slots[0] == i) 358987b29b20SYan Zheng push_space += data_size; 3590db94535dSChris Mason 3591db94535dSChris Mason this_item_size = btrfs_item_size(left, item); 3592db94535dSChris Mason if (this_item_size + sizeof(*item) + push_space > free_space) 359300ec4c51SChris Mason break; 359431840ae1SZheng Yan 359500ec4c51SChris Mason push_items++; 3596db94535dSChris Mason push_space += this_item_size + sizeof(*item); 359734a38218SChris Mason if (i == 0) 359834a38218SChris Mason break; 359934a38218SChris Mason i--; 3600db94535dSChris Mason } 36015f39d397SChris Mason 3602925baeddSChris Mason if (push_items == 0) 3603925baeddSChris Mason goto out_unlock; 36045f39d397SChris Mason 36056c1500f2SJulia Lawall WARN_ON(!empty && push_items == left_nritems); 36065f39d397SChris Mason 360700ec4c51SChris Mason /* push left to right */ 36085f39d397SChris Mason right_nritems = btrfs_header_nritems(right); 360934a38218SChris Mason 36105f39d397SChris Mason push_space = btrfs_item_end_nr(left, left_nritems - push_items); 36118f881e8cSDavid Sterba push_space -= leaf_data_end(left); 36125f39d397SChris Mason 361300ec4c51SChris Mason /* make room in the right data area */ 36148f881e8cSDavid Sterba data_end = leaf_data_end(right); 36155f39d397SChris Mason memmove_extent_buffer(right, 36163d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end - push_space, 36173d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end, 36180b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); 36195f39d397SChris Mason 362000ec4c51SChris Mason /* copy from the left data area */ 36213d9ec8c4SNikolay Borisov copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + 36220b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 36238f881e8cSDavid Sterba BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), 3624d6025579SChris Mason push_space); 36255f39d397SChris Mason 36265f39d397SChris Mason memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), 36275f39d397SChris Mason btrfs_item_nr_offset(0), 36280783fcfcSChris Mason right_nritems * sizeof(struct btrfs_item)); 36295f39d397SChris Mason 363000ec4c51SChris Mason /* copy the items from left to right */ 36315f39d397SChris Mason copy_extent_buffer(right, left, btrfs_item_nr_offset(0), 36325f39d397SChris Mason btrfs_item_nr_offset(left_nritems - push_items), 36330783fcfcSChris Mason push_items * sizeof(struct btrfs_item)); 363400ec4c51SChris Mason 363500ec4c51SChris Mason /* update the item pointers */ 3636c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 36377518a238SChris Mason right_nritems += push_items; 36385f39d397SChris Mason btrfs_set_header_nritems(right, right_nritems); 36390b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 36407518a238SChris Mason for (i = 0; i < right_nritems; i++) { 3641dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3642cc4c13d5SDavid Sterba push_space -= btrfs_token_item_size(&token, item); 3643cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, push_space); 3644db94535dSChris Mason } 3645db94535dSChris Mason 36467518a238SChris Mason left_nritems -= push_items; 36475f39d397SChris Mason btrfs_set_header_nritems(left, left_nritems); 364800ec4c51SChris Mason 364934a38218SChris Mason if (left_nritems) 36505f39d397SChris Mason btrfs_mark_buffer_dirty(left); 3651f0486c68SYan, Zheng else 36526a884d7dSDavid Sterba btrfs_clean_tree_block(left); 3653f0486c68SYan, Zheng 36545f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3655a429e513SChris Mason 36565f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 36575f39d397SChris Mason btrfs_set_node_key(upper, &disk_key, slot + 1); 3658d6025579SChris Mason btrfs_mark_buffer_dirty(upper); 365902217ed2SChris Mason 366000ec4c51SChris Mason /* then fixup the leaf pointer in the path */ 36617518a238SChris Mason if (path->slots[0] >= left_nritems) { 36627518a238SChris Mason path->slots[0] -= left_nritems; 3663925baeddSChris Mason if (btrfs_header_nritems(path->nodes[0]) == 0) 36646a884d7dSDavid Sterba btrfs_clean_tree_block(path->nodes[0]); 3665925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 36665f39d397SChris Mason free_extent_buffer(path->nodes[0]); 36675f39d397SChris Mason path->nodes[0] = right; 366800ec4c51SChris Mason path->slots[1] += 1; 366900ec4c51SChris Mason } else { 3670925baeddSChris Mason btrfs_tree_unlock(right); 36715f39d397SChris Mason free_extent_buffer(right); 367200ec4c51SChris Mason } 367300ec4c51SChris Mason return 0; 3674925baeddSChris Mason 3675925baeddSChris Mason out_unlock: 3676925baeddSChris Mason btrfs_tree_unlock(right); 3677925baeddSChris Mason free_extent_buffer(right); 3678925baeddSChris Mason return 1; 367900ec4c51SChris Mason } 3680925baeddSChris Mason 368100ec4c51SChris Mason /* 368244871b1bSChris Mason * push some data in the path leaf to the right, trying to free up at 368374123bd7SChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 368444871b1bSChris Mason * 368544871b1bSChris Mason * returns 1 if the push failed because the other node didn't have enough 368644871b1bSChris Mason * room, 0 if everything worked out and < 0 if there were major errors. 368799d8f83cSChris Mason * 368899d8f83cSChris Mason * this will push starting from min_slot to the end of the leaf. It won't 368999d8f83cSChris Mason * push any slot lower than min_slot 369074123bd7SChris Mason */ 369144871b1bSChris Mason static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root 369299d8f83cSChris Mason *root, struct btrfs_path *path, 369399d8f83cSChris Mason int min_data_size, int data_size, 369499d8f83cSChris Mason int empty, u32 min_slot) 3695be0e5c09SChris Mason { 369644871b1bSChris Mason struct extent_buffer *left = path->nodes[0]; 369744871b1bSChris Mason struct extent_buffer *right; 369844871b1bSChris Mason struct extent_buffer *upper; 369944871b1bSChris Mason int slot; 370044871b1bSChris Mason int free_space; 370144871b1bSChris Mason u32 left_nritems; 370244871b1bSChris Mason int ret; 370344871b1bSChris Mason 370444871b1bSChris Mason if (!path->nodes[1]) 370544871b1bSChris Mason return 1; 370644871b1bSChris Mason 370744871b1bSChris Mason slot = path->slots[1]; 370844871b1bSChris Mason upper = path->nodes[1]; 370944871b1bSChris Mason if (slot >= btrfs_header_nritems(upper) - 1) 371044871b1bSChris Mason return 1; 371144871b1bSChris Mason 371244871b1bSChris Mason btrfs_assert_tree_locked(path->nodes[1]); 371344871b1bSChris Mason 37144b231ae4SDavid Sterba right = btrfs_read_node_slot(upper, slot + 1); 3715fb770ae4SLiu Bo /* 3716fb770ae4SLiu Bo * slot + 1 is not valid or we fail to read the right node, 3717fb770ae4SLiu Bo * no big deal, just return. 3718fb770ae4SLiu Bo */ 3719fb770ae4SLiu Bo if (IS_ERR(right)) 372091ca338dSTsutomu Itoh return 1; 372191ca338dSTsutomu Itoh 3722bf77467aSJosef Bacik __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 372344871b1bSChris Mason 3724e902baacSDavid Sterba free_space = btrfs_leaf_free_space(right); 372544871b1bSChris Mason if (free_space < data_size) 372644871b1bSChris Mason goto out_unlock; 372744871b1bSChris Mason 372844871b1bSChris Mason /* cow and double check */ 372944871b1bSChris Mason ret = btrfs_cow_block(trans, root, right, upper, 3730bf59a5a2SJosef Bacik slot + 1, &right, BTRFS_NESTING_RIGHT_COW); 373144871b1bSChris Mason if (ret) 373244871b1bSChris Mason goto out_unlock; 373344871b1bSChris Mason 3734e902baacSDavid Sterba free_space = btrfs_leaf_free_space(right); 373544871b1bSChris Mason if (free_space < data_size) 373644871b1bSChris Mason goto out_unlock; 373744871b1bSChris Mason 373844871b1bSChris Mason left_nritems = btrfs_header_nritems(left); 373944871b1bSChris Mason if (left_nritems == 0) 374044871b1bSChris Mason goto out_unlock; 374144871b1bSChris Mason 3742d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3743d16c702fSQu Wenruo ret = -EUCLEAN; 3744d16c702fSQu Wenruo btrfs_tree_unlock(right); 3745d16c702fSQu Wenruo free_extent_buffer(right); 3746d16c702fSQu Wenruo return ret; 3747d16c702fSQu Wenruo } 37482ef1fed2SFilipe David Borba Manana if (path->slots[0] == left_nritems && !empty) { 37492ef1fed2SFilipe David Borba Manana /* Key greater than all keys in the leaf, right neighbor has 37502ef1fed2SFilipe David Borba Manana * enough room for it and we're not emptying our leaf to delete 37512ef1fed2SFilipe David Borba Manana * it, therefore use right neighbor to insert the new item and 375252042d8eSAndrea Gelmini * no need to touch/dirty our left leaf. */ 37532ef1fed2SFilipe David Borba Manana btrfs_tree_unlock(left); 37542ef1fed2SFilipe David Borba Manana free_extent_buffer(left); 37552ef1fed2SFilipe David Borba Manana path->nodes[0] = right; 37562ef1fed2SFilipe David Borba Manana path->slots[0] = 0; 37572ef1fed2SFilipe David Borba Manana path->slots[1]++; 37582ef1fed2SFilipe David Borba Manana return 0; 37592ef1fed2SFilipe David Borba Manana } 37602ef1fed2SFilipe David Borba Manana 3761f72f0010SDavid Sterba return __push_leaf_right(path, min_data_size, empty, 376299d8f83cSChris Mason right, free_space, left_nritems, min_slot); 376344871b1bSChris Mason out_unlock: 376444871b1bSChris Mason btrfs_tree_unlock(right); 376544871b1bSChris Mason free_extent_buffer(right); 376644871b1bSChris Mason return 1; 376744871b1bSChris Mason } 376844871b1bSChris Mason 376944871b1bSChris Mason /* 377044871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 377144871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 377299d8f83cSChris Mason * 377399d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 377499d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the 377599d8f83cSChris Mason * items 377644871b1bSChris Mason */ 37778087c193SDavid Sterba static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, 377844871b1bSChris Mason int empty, struct extent_buffer *left, 377999d8f83cSChris Mason int free_space, u32 right_nritems, 378099d8f83cSChris Mason u32 max_slot) 378144871b1bSChris Mason { 37828087c193SDavid Sterba struct btrfs_fs_info *fs_info = left->fs_info; 37835f39d397SChris Mason struct btrfs_disk_key disk_key; 37845f39d397SChris Mason struct extent_buffer *right = path->nodes[0]; 3785be0e5c09SChris Mason int i; 3786be0e5c09SChris Mason int push_space = 0; 3787be0e5c09SChris Mason int push_items = 0; 37880783fcfcSChris Mason struct btrfs_item *item; 37897518a238SChris Mason u32 old_left_nritems; 379034a38218SChris Mason u32 nr; 3791aa5d6bedSChris Mason int ret = 0; 3792db94535dSChris Mason u32 this_item_size; 3793db94535dSChris Mason u32 old_left_item_size; 3794cfed81a0SChris Mason struct btrfs_map_token token; 3795cfed81a0SChris Mason 379634a38218SChris Mason if (empty) 379799d8f83cSChris Mason nr = min(right_nritems, max_slot); 379834a38218SChris Mason else 379999d8f83cSChris Mason nr = min(right_nritems - 1, max_slot); 380034a38218SChris Mason 380134a38218SChris Mason for (i = 0; i < nr; i++) { 3802dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3803db94535dSChris Mason 380431840ae1SZheng Yan if (!empty && push_items > 0) { 380531840ae1SZheng Yan if (path->slots[0] < i) 380631840ae1SZheng Yan break; 380731840ae1SZheng Yan if (path->slots[0] == i) { 3808e902baacSDavid Sterba int space = btrfs_leaf_free_space(right); 3809e902baacSDavid Sterba 381031840ae1SZheng Yan if (space + push_space * 2 > free_space) 381131840ae1SZheng Yan break; 381231840ae1SZheng Yan } 381331840ae1SZheng Yan } 381431840ae1SZheng Yan 3815be0e5c09SChris Mason if (path->slots[0] == i) 381687b29b20SYan Zheng push_space += data_size; 3817db94535dSChris Mason 3818db94535dSChris Mason this_item_size = btrfs_item_size(right, item); 3819db94535dSChris Mason if (this_item_size + sizeof(*item) + push_space > free_space) 3820be0e5c09SChris Mason break; 3821db94535dSChris Mason 3822be0e5c09SChris Mason push_items++; 3823db94535dSChris Mason push_space += this_item_size + sizeof(*item); 3824be0e5c09SChris Mason } 3825db94535dSChris Mason 3826be0e5c09SChris Mason if (push_items == 0) { 3827925baeddSChris Mason ret = 1; 3828925baeddSChris Mason goto out; 3829be0e5c09SChris Mason } 3830fae7f21cSDulshani Gunawardhana WARN_ON(!empty && push_items == btrfs_header_nritems(right)); 38315f39d397SChris Mason 3832be0e5c09SChris Mason /* push data from right to left */ 38335f39d397SChris Mason copy_extent_buffer(left, right, 38345f39d397SChris Mason btrfs_item_nr_offset(btrfs_header_nritems(left)), 38355f39d397SChris Mason btrfs_item_nr_offset(0), 38365f39d397SChris Mason push_items * sizeof(struct btrfs_item)); 38375f39d397SChris Mason 38380b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - 38395f39d397SChris Mason btrfs_item_offset_nr(right, push_items - 1); 38405f39d397SChris Mason 38413d9ec8c4SNikolay Borisov copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + 38428f881e8cSDavid Sterba leaf_data_end(left) - push_space, 38433d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + 38445f39d397SChris Mason btrfs_item_offset_nr(right, push_items - 1), 3845be0e5c09SChris Mason push_space); 38465f39d397SChris Mason old_left_nritems = btrfs_header_nritems(left); 384787b29b20SYan Zheng BUG_ON(old_left_nritems <= 0); 3848eb60ceacSChris Mason 3849c82f823cSDavid Sterba btrfs_init_map_token(&token, left); 3850db94535dSChris Mason old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); 3851be0e5c09SChris Mason for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { 38525f39d397SChris Mason u32 ioff; 3853db94535dSChris Mason 3854dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3855db94535dSChris Mason 3856cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 3857cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, 3858cc4c13d5SDavid Sterba ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); 3859be0e5c09SChris Mason } 38605f39d397SChris Mason btrfs_set_header_nritems(left, old_left_nritems + push_items); 3861be0e5c09SChris Mason 3862be0e5c09SChris Mason /* fixup right node */ 386331b1a2bdSJulia Lawall if (push_items > right_nritems) 386431b1a2bdSJulia Lawall WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, 3865d397712bSChris Mason right_nritems); 386634a38218SChris Mason 386734a38218SChris Mason if (push_items < right_nritems) { 38685f39d397SChris Mason push_space = btrfs_item_offset_nr(right, push_items - 1) - 38698f881e8cSDavid Sterba leaf_data_end(right); 38703d9ec8c4SNikolay Borisov memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + 38710b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 38723d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + 38738f881e8cSDavid Sterba leaf_data_end(right), push_space); 38745f39d397SChris Mason 38755f39d397SChris Mason memmove_extent_buffer(right, btrfs_item_nr_offset(0), 38765f39d397SChris Mason btrfs_item_nr_offset(push_items), 38775f39d397SChris Mason (btrfs_header_nritems(right) - push_items) * 38780783fcfcSChris Mason sizeof(struct btrfs_item)); 387934a38218SChris Mason } 3880c82f823cSDavid Sterba 3881c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 3882eef1c494SYan right_nritems -= push_items; 3883eef1c494SYan btrfs_set_header_nritems(right, right_nritems); 38840b246afaSJeff Mahoney push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 38855f39d397SChris Mason for (i = 0; i < right_nritems; i++) { 3886dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 3887db94535dSChris Mason 3888cc4c13d5SDavid Sterba push_space = push_space - btrfs_token_item_size(&token, item); 3889cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, push_space); 3890db94535dSChris Mason } 3891eb60ceacSChris Mason 38925f39d397SChris Mason btrfs_mark_buffer_dirty(left); 389334a38218SChris Mason if (right_nritems) 38945f39d397SChris Mason btrfs_mark_buffer_dirty(right); 3895f0486c68SYan, Zheng else 38966a884d7dSDavid Sterba btrfs_clean_tree_block(right); 3897098f59c2SChris Mason 38985f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 3899b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 3900be0e5c09SChris Mason 3901be0e5c09SChris Mason /* then fixup the leaf pointer in the path */ 3902be0e5c09SChris Mason if (path->slots[0] < push_items) { 3903be0e5c09SChris Mason path->slots[0] += old_left_nritems; 3904925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 39055f39d397SChris Mason free_extent_buffer(path->nodes[0]); 39065f39d397SChris Mason path->nodes[0] = left; 3907be0e5c09SChris Mason path->slots[1] -= 1; 3908be0e5c09SChris Mason } else { 3909925baeddSChris Mason btrfs_tree_unlock(left); 39105f39d397SChris Mason free_extent_buffer(left); 3911be0e5c09SChris Mason path->slots[0] -= push_items; 3912be0e5c09SChris Mason } 3913eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 3914aa5d6bedSChris Mason return ret; 3915925baeddSChris Mason out: 3916925baeddSChris Mason btrfs_tree_unlock(left); 3917925baeddSChris Mason free_extent_buffer(left); 3918925baeddSChris Mason return ret; 3919be0e5c09SChris Mason } 3920be0e5c09SChris Mason 392174123bd7SChris Mason /* 392244871b1bSChris Mason * push some data in the path leaf to the left, trying to free up at 392344871b1bSChris Mason * least data_size bytes. returns zero if the push worked, nonzero otherwise 392499d8f83cSChris Mason * 392599d8f83cSChris Mason * max_slot can put a limit on how far into the leaf we'll push items. The 392699d8f83cSChris Mason * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the 392799d8f83cSChris Mason * items 392844871b1bSChris Mason */ 392944871b1bSChris Mason static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root 393099d8f83cSChris Mason *root, struct btrfs_path *path, int min_data_size, 393199d8f83cSChris Mason int data_size, int empty, u32 max_slot) 393244871b1bSChris Mason { 393344871b1bSChris Mason struct extent_buffer *right = path->nodes[0]; 393444871b1bSChris Mason struct extent_buffer *left; 393544871b1bSChris Mason int slot; 393644871b1bSChris Mason int free_space; 393744871b1bSChris Mason u32 right_nritems; 393844871b1bSChris Mason int ret = 0; 393944871b1bSChris Mason 394044871b1bSChris Mason slot = path->slots[1]; 394144871b1bSChris Mason if (slot == 0) 394244871b1bSChris Mason return 1; 394344871b1bSChris Mason if (!path->nodes[1]) 394444871b1bSChris Mason return 1; 394544871b1bSChris Mason 394644871b1bSChris Mason right_nritems = btrfs_header_nritems(right); 394744871b1bSChris Mason if (right_nritems == 0) 394844871b1bSChris Mason return 1; 394944871b1bSChris Mason 395044871b1bSChris Mason btrfs_assert_tree_locked(path->nodes[1]); 395144871b1bSChris Mason 39524b231ae4SDavid Sterba left = btrfs_read_node_slot(path->nodes[1], slot - 1); 3953fb770ae4SLiu Bo /* 3954fb770ae4SLiu Bo * slot - 1 is not valid or we fail to read the left node, 3955fb770ae4SLiu Bo * no big deal, just return. 3956fb770ae4SLiu Bo */ 3957fb770ae4SLiu Bo if (IS_ERR(left)) 395891ca338dSTsutomu Itoh return 1; 395991ca338dSTsutomu Itoh 3960bf77467aSJosef Bacik __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 396144871b1bSChris Mason 3962e902baacSDavid Sterba free_space = btrfs_leaf_free_space(left); 396344871b1bSChris Mason if (free_space < data_size) { 396444871b1bSChris Mason ret = 1; 396544871b1bSChris Mason goto out; 396644871b1bSChris Mason } 396744871b1bSChris Mason 396844871b1bSChris Mason /* cow and double check */ 396944871b1bSChris Mason ret = btrfs_cow_block(trans, root, left, 39709631e4ccSJosef Bacik path->nodes[1], slot - 1, &left, 3971bf59a5a2SJosef Bacik BTRFS_NESTING_LEFT_COW); 397244871b1bSChris Mason if (ret) { 397344871b1bSChris Mason /* we hit -ENOSPC, but it isn't fatal here */ 397479787eaaSJeff Mahoney if (ret == -ENOSPC) 397544871b1bSChris Mason ret = 1; 397644871b1bSChris Mason goto out; 397744871b1bSChris Mason } 397844871b1bSChris Mason 3979e902baacSDavid Sterba free_space = btrfs_leaf_free_space(left); 398044871b1bSChris Mason if (free_space < data_size) { 398144871b1bSChris Mason ret = 1; 398244871b1bSChris Mason goto out; 398344871b1bSChris Mason } 398444871b1bSChris Mason 3985d16c702fSQu Wenruo if (check_sibling_keys(left, right)) { 3986d16c702fSQu Wenruo ret = -EUCLEAN; 3987d16c702fSQu Wenruo goto out; 3988d16c702fSQu Wenruo } 39898087c193SDavid Sterba return __push_leaf_left(path, min_data_size, 399099d8f83cSChris Mason empty, left, free_space, right_nritems, 399199d8f83cSChris Mason max_slot); 399244871b1bSChris Mason out: 399344871b1bSChris Mason btrfs_tree_unlock(left); 399444871b1bSChris Mason free_extent_buffer(left); 399544871b1bSChris Mason return ret; 399644871b1bSChris Mason } 399744871b1bSChris Mason 399844871b1bSChris Mason /* 399974123bd7SChris Mason * split the path's leaf in two, making sure there is at least data_size 400074123bd7SChris Mason * available for the resulting leaf level of the path. 400174123bd7SChris Mason */ 4002143bede5SJeff Mahoney static noinline void copy_for_split(struct btrfs_trans_handle *trans, 400344871b1bSChris Mason struct btrfs_path *path, 400444871b1bSChris Mason struct extent_buffer *l, 400544871b1bSChris Mason struct extent_buffer *right, 400644871b1bSChris Mason int slot, int mid, int nritems) 4007be0e5c09SChris Mason { 400894f94ad9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 4009be0e5c09SChris Mason int data_copy_size; 4010be0e5c09SChris Mason int rt_data_off; 4011be0e5c09SChris Mason int i; 4012d4dbff95SChris Mason struct btrfs_disk_key disk_key; 4013cfed81a0SChris Mason struct btrfs_map_token token; 4014cfed81a0SChris Mason 40155f39d397SChris Mason nritems = nritems - mid; 40165f39d397SChris Mason btrfs_set_header_nritems(right, nritems); 40178f881e8cSDavid Sterba data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l); 40185f39d397SChris Mason 40195f39d397SChris Mason copy_extent_buffer(right, l, btrfs_item_nr_offset(0), 40205f39d397SChris Mason btrfs_item_nr_offset(mid), 40215f39d397SChris Mason nritems * sizeof(struct btrfs_item)); 40225f39d397SChris Mason 40235f39d397SChris Mason copy_extent_buffer(right, l, 40243d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - 40253d9ec8c4SNikolay Borisov data_copy_size, BTRFS_LEAF_DATA_OFFSET + 40268f881e8cSDavid Sterba leaf_data_end(l), data_copy_size); 402774123bd7SChris Mason 40280b246afaSJeff Mahoney rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); 40295f39d397SChris Mason 4030c82f823cSDavid Sterba btrfs_init_map_token(&token, right); 40315f39d397SChris Mason for (i = 0; i < nritems; i++) { 4032dd3cc16bSRoss Kirk struct btrfs_item *item = btrfs_item_nr(i); 4033db94535dSChris Mason u32 ioff; 4034db94535dSChris Mason 4035cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4036cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, ioff + rt_data_off); 40370783fcfcSChris Mason } 403874123bd7SChris Mason 40395f39d397SChris Mason btrfs_set_header_nritems(l, mid); 40405f39d397SChris Mason btrfs_item_key(right, &disk_key, 0); 40416ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); 40425f39d397SChris Mason 40435f39d397SChris Mason btrfs_mark_buffer_dirty(right); 40445f39d397SChris Mason btrfs_mark_buffer_dirty(l); 4045eb60ceacSChris Mason BUG_ON(path->slots[0] != slot); 40465f39d397SChris Mason 4047be0e5c09SChris Mason if (mid <= slot) { 4048925baeddSChris Mason btrfs_tree_unlock(path->nodes[0]); 40495f39d397SChris Mason free_extent_buffer(path->nodes[0]); 40505f39d397SChris Mason path->nodes[0] = right; 4051be0e5c09SChris Mason path->slots[0] -= mid; 4052be0e5c09SChris Mason path->slots[1] += 1; 4053925baeddSChris Mason } else { 4054925baeddSChris Mason btrfs_tree_unlock(right); 40555f39d397SChris Mason free_extent_buffer(right); 4056925baeddSChris Mason } 40575f39d397SChris Mason 4058eb60ceacSChris Mason BUG_ON(path->slots[0] < 0); 405944871b1bSChris Mason } 406044871b1bSChris Mason 406144871b1bSChris Mason /* 406299d8f83cSChris Mason * double splits happen when we need to insert a big item in the middle 406399d8f83cSChris Mason * of a leaf. A double split can leave us with 3 mostly empty leaves: 406499d8f83cSChris Mason * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] 406599d8f83cSChris Mason * A B C 406699d8f83cSChris Mason * 406799d8f83cSChris Mason * We avoid this by trying to push the items on either side of our target 406899d8f83cSChris Mason * into the adjacent leaves. If all goes well we can avoid the double split 406999d8f83cSChris Mason * completely. 407099d8f83cSChris Mason */ 407199d8f83cSChris Mason static noinline int push_for_double_split(struct btrfs_trans_handle *trans, 407299d8f83cSChris Mason struct btrfs_root *root, 407399d8f83cSChris Mason struct btrfs_path *path, 407499d8f83cSChris Mason int data_size) 407599d8f83cSChris Mason { 407699d8f83cSChris Mason int ret; 407799d8f83cSChris Mason int progress = 0; 407899d8f83cSChris Mason int slot; 407999d8f83cSChris Mason u32 nritems; 40805a4267caSFilipe David Borba Manana int space_needed = data_size; 408199d8f83cSChris Mason 408299d8f83cSChris Mason slot = path->slots[0]; 40835a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(path->nodes[0])) 4084e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 408599d8f83cSChris Mason 408699d8f83cSChris Mason /* 408799d8f83cSChris Mason * try to push all the items after our slot into the 408899d8f83cSChris Mason * right leaf 408999d8f83cSChris Mason */ 40905a4267caSFilipe David Borba Manana ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); 409199d8f83cSChris Mason if (ret < 0) 409299d8f83cSChris Mason return ret; 409399d8f83cSChris Mason 409499d8f83cSChris Mason if (ret == 0) 409599d8f83cSChris Mason progress++; 409699d8f83cSChris Mason 409799d8f83cSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 409899d8f83cSChris Mason /* 409999d8f83cSChris Mason * our goal is to get our slot at the start or end of a leaf. If 410099d8f83cSChris Mason * we've done so we're done 410199d8f83cSChris Mason */ 410299d8f83cSChris Mason if (path->slots[0] == 0 || path->slots[0] == nritems) 410399d8f83cSChris Mason return 0; 410499d8f83cSChris Mason 4105e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 410699d8f83cSChris Mason return 0; 410799d8f83cSChris Mason 410899d8f83cSChris Mason /* try to push all the items before our slot into the next leaf */ 410999d8f83cSChris Mason slot = path->slots[0]; 4110263d3995SFilipe Manana space_needed = data_size; 4111263d3995SFilipe Manana if (slot > 0) 4112e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(path->nodes[0]); 41135a4267caSFilipe David Borba Manana ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); 411499d8f83cSChris Mason if (ret < 0) 411599d8f83cSChris Mason return ret; 411699d8f83cSChris Mason 411799d8f83cSChris Mason if (ret == 0) 411899d8f83cSChris Mason progress++; 411999d8f83cSChris Mason 412099d8f83cSChris Mason if (progress) 412199d8f83cSChris Mason return 0; 412299d8f83cSChris Mason return 1; 412399d8f83cSChris Mason } 412499d8f83cSChris Mason 412599d8f83cSChris Mason /* 412644871b1bSChris Mason * split the path's leaf in two, making sure there is at least data_size 412744871b1bSChris Mason * available for the resulting leaf level of the path. 412844871b1bSChris Mason * 412944871b1bSChris Mason * returns 0 if all went well and < 0 on failure. 413044871b1bSChris Mason */ 413144871b1bSChris Mason static noinline int split_leaf(struct btrfs_trans_handle *trans, 413244871b1bSChris Mason struct btrfs_root *root, 4133310712b2SOmar Sandoval const struct btrfs_key *ins_key, 413444871b1bSChris Mason struct btrfs_path *path, int data_size, 413544871b1bSChris Mason int extend) 413644871b1bSChris Mason { 41375d4f98a2SYan Zheng struct btrfs_disk_key disk_key; 413844871b1bSChris Mason struct extent_buffer *l; 413944871b1bSChris Mason u32 nritems; 414044871b1bSChris Mason int mid; 414144871b1bSChris Mason int slot; 414244871b1bSChris Mason struct extent_buffer *right; 4143b7a0365eSDaniel Dressler struct btrfs_fs_info *fs_info = root->fs_info; 414444871b1bSChris Mason int ret = 0; 414544871b1bSChris Mason int wret; 41465d4f98a2SYan Zheng int split; 414744871b1bSChris Mason int num_doubles = 0; 414899d8f83cSChris Mason int tried_avoid_double = 0; 414944871b1bSChris Mason 4150a5719521SYan, Zheng l = path->nodes[0]; 4151a5719521SYan, Zheng slot = path->slots[0]; 4152a5719521SYan, Zheng if (extend && data_size + btrfs_item_size_nr(l, slot) + 41530b246afaSJeff Mahoney sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) 4154a5719521SYan, Zheng return -EOVERFLOW; 4155a5719521SYan, Zheng 415644871b1bSChris Mason /* first try to make some room by pushing left and right */ 415733157e05SLiu Bo if (data_size && path->nodes[1]) { 41585a4267caSFilipe David Borba Manana int space_needed = data_size; 41595a4267caSFilipe David Borba Manana 41605a4267caSFilipe David Borba Manana if (slot < btrfs_header_nritems(l)) 4161e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 41625a4267caSFilipe David Borba Manana 41635a4267caSFilipe David Borba Manana wret = push_leaf_right(trans, root, path, space_needed, 41645a4267caSFilipe David Borba Manana space_needed, 0, 0); 416544871b1bSChris Mason if (wret < 0) 416644871b1bSChris Mason return wret; 416744871b1bSChris Mason if (wret) { 4168263d3995SFilipe Manana space_needed = data_size; 4169263d3995SFilipe Manana if (slot > 0) 4170e902baacSDavid Sterba space_needed -= btrfs_leaf_free_space(l); 41715a4267caSFilipe David Borba Manana wret = push_leaf_left(trans, root, path, space_needed, 41725a4267caSFilipe David Borba Manana space_needed, 0, (u32)-1); 417344871b1bSChris Mason if (wret < 0) 417444871b1bSChris Mason return wret; 417544871b1bSChris Mason } 417644871b1bSChris Mason l = path->nodes[0]; 417744871b1bSChris Mason 417844871b1bSChris Mason /* did the pushes work? */ 4179e902baacSDavid Sterba if (btrfs_leaf_free_space(l) >= data_size) 418044871b1bSChris Mason return 0; 418144871b1bSChris Mason } 418244871b1bSChris Mason 418344871b1bSChris Mason if (!path->nodes[1]) { 4184fdd99c72SLiu Bo ret = insert_new_root(trans, root, path, 1); 418544871b1bSChris Mason if (ret) 418644871b1bSChris Mason return ret; 418744871b1bSChris Mason } 418844871b1bSChris Mason again: 41895d4f98a2SYan Zheng split = 1; 419044871b1bSChris Mason l = path->nodes[0]; 419144871b1bSChris Mason slot = path->slots[0]; 419244871b1bSChris Mason nritems = btrfs_header_nritems(l); 419344871b1bSChris Mason mid = (nritems + 1) / 2; 419444871b1bSChris Mason 41955d4f98a2SYan Zheng if (mid <= slot) { 41965d4f98a2SYan Zheng if (nritems == 1 || 41975d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + data_size > 41980b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 41995d4f98a2SYan Zheng if (slot >= nritems) { 42005d4f98a2SYan Zheng split = 0; 42015d4f98a2SYan Zheng } else { 42025d4f98a2SYan Zheng mid = slot; 42035d4f98a2SYan Zheng if (mid != nritems && 42045d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 42050b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 420699d8f83cSChris Mason if (data_size && !tried_avoid_double) 420799d8f83cSChris Mason goto push_for_double; 42085d4f98a2SYan Zheng split = 2; 42095d4f98a2SYan Zheng } 42105d4f98a2SYan Zheng } 42115d4f98a2SYan Zheng } 42125d4f98a2SYan Zheng } else { 42135d4f98a2SYan Zheng if (leaf_space_used(l, 0, mid) + data_size > 42140b246afaSJeff Mahoney BTRFS_LEAF_DATA_SIZE(fs_info)) { 42155d4f98a2SYan Zheng if (!extend && data_size && slot == 0) { 42165d4f98a2SYan Zheng split = 0; 42175d4f98a2SYan Zheng } else if ((extend || !data_size) && slot == 0) { 42185d4f98a2SYan Zheng mid = 1; 42195d4f98a2SYan Zheng } else { 42205d4f98a2SYan Zheng mid = slot; 42215d4f98a2SYan Zheng if (mid != nritems && 42225d4f98a2SYan Zheng leaf_space_used(l, mid, nritems - mid) + 42230b246afaSJeff Mahoney data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 422499d8f83cSChris Mason if (data_size && !tried_avoid_double) 422599d8f83cSChris Mason goto push_for_double; 42265d4f98a2SYan Zheng split = 2; 42275d4f98a2SYan Zheng } 42285d4f98a2SYan Zheng } 42295d4f98a2SYan Zheng } 42305d4f98a2SYan Zheng } 42315d4f98a2SYan Zheng 42325d4f98a2SYan Zheng if (split == 0) 42335d4f98a2SYan Zheng btrfs_cpu_key_to_disk(&disk_key, ins_key); 42345d4f98a2SYan Zheng else 42355d4f98a2SYan Zheng btrfs_item_key(l, &disk_key, mid); 42365d4f98a2SYan Zheng 4237ca9d473aSJosef Bacik /* 4238ca9d473aSJosef Bacik * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double 4239ca9d473aSJosef Bacik * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES 4240ca9d473aSJosef Bacik * subclasses, which is 8 at the time of this patch, and we've maxed it 4241ca9d473aSJosef Bacik * out. In the future we could add a 4242ca9d473aSJosef Bacik * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just 4243ca9d473aSJosef Bacik * use BTRFS_NESTING_NEW_ROOT. 4244ca9d473aSJosef Bacik */ 4245a6279470SFilipe Manana right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0, 4246ca9d473aSJosef Bacik l->start, 0, num_doubles ? 4247ca9d473aSJosef Bacik BTRFS_NESTING_NEW_ROOT : 4248ca9d473aSJosef Bacik BTRFS_NESTING_SPLIT); 4249f0486c68SYan, Zheng if (IS_ERR(right)) 425044871b1bSChris Mason return PTR_ERR(right); 4251f0486c68SYan, Zheng 42520b246afaSJeff Mahoney root_add_used(root, fs_info->nodesize); 425344871b1bSChris Mason 42545d4f98a2SYan Zheng if (split == 0) { 425544871b1bSChris Mason if (mid <= slot) { 425644871b1bSChris Mason btrfs_set_header_nritems(right, 0); 42576ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, 42582ff7e61eSJeff Mahoney right->start, path->slots[1] + 1, 1); 425944871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 426044871b1bSChris Mason free_extent_buffer(path->nodes[0]); 426144871b1bSChris Mason path->nodes[0] = right; 426244871b1bSChris Mason path->slots[0] = 0; 426344871b1bSChris Mason path->slots[1] += 1; 426444871b1bSChris Mason } else { 426544871b1bSChris Mason btrfs_set_header_nritems(right, 0); 42666ad3cf6dSDavid Sterba insert_ptr(trans, path, &disk_key, 42672ff7e61eSJeff Mahoney right->start, path->slots[1], 1); 426844871b1bSChris Mason btrfs_tree_unlock(path->nodes[0]); 426944871b1bSChris Mason free_extent_buffer(path->nodes[0]); 427044871b1bSChris Mason path->nodes[0] = right; 427144871b1bSChris Mason path->slots[0] = 0; 4272143bede5SJeff Mahoney if (path->slots[1] == 0) 4273b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 42745d4f98a2SYan Zheng } 4275196e0249SLiu Bo /* 4276196e0249SLiu Bo * We create a new leaf 'right' for the required ins_len and 4277196e0249SLiu Bo * we'll do btrfs_mark_buffer_dirty() on this leaf after copying 4278196e0249SLiu Bo * the content of ins_len to 'right'. 4279196e0249SLiu Bo */ 428044871b1bSChris Mason return ret; 428144871b1bSChris Mason } 428244871b1bSChris Mason 428394f94ad9SDavid Sterba copy_for_split(trans, path, l, right, slot, mid, nritems); 428444871b1bSChris Mason 42855d4f98a2SYan Zheng if (split == 2) { 4286cc0c5538SChris Mason BUG_ON(num_doubles != 0); 4287cc0c5538SChris Mason num_doubles++; 4288cc0c5538SChris Mason goto again; 42893326d1b0SChris Mason } 429044871b1bSChris Mason 4291143bede5SJeff Mahoney return 0; 429299d8f83cSChris Mason 429399d8f83cSChris Mason push_for_double: 429499d8f83cSChris Mason push_for_double_split(trans, root, path, data_size); 429599d8f83cSChris Mason tried_avoid_double = 1; 4296e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 429799d8f83cSChris Mason return 0; 429899d8f83cSChris Mason goto again; 4299be0e5c09SChris Mason } 4300be0e5c09SChris Mason 4301ad48fd75SYan, Zheng static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, 4302ad48fd75SYan, Zheng struct btrfs_root *root, 4303ad48fd75SYan, Zheng struct btrfs_path *path, int ins_len) 4304ad48fd75SYan, Zheng { 4305ad48fd75SYan, Zheng struct btrfs_key key; 4306ad48fd75SYan, Zheng struct extent_buffer *leaf; 4307ad48fd75SYan, Zheng struct btrfs_file_extent_item *fi; 4308ad48fd75SYan, Zheng u64 extent_len = 0; 4309ad48fd75SYan, Zheng u32 item_size; 4310ad48fd75SYan, Zheng int ret; 4311ad48fd75SYan, Zheng 4312ad48fd75SYan, Zheng leaf = path->nodes[0]; 4313ad48fd75SYan, Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4314ad48fd75SYan, Zheng 4315ad48fd75SYan, Zheng BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && 4316ad48fd75SYan, Zheng key.type != BTRFS_EXTENT_CSUM_KEY); 4317ad48fd75SYan, Zheng 4318e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) >= ins_len) 4319ad48fd75SYan, Zheng return 0; 4320ad48fd75SYan, Zheng 4321ad48fd75SYan, Zheng item_size = btrfs_item_size_nr(leaf, path->slots[0]); 4322ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 4323ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 4324ad48fd75SYan, Zheng struct btrfs_file_extent_item); 4325ad48fd75SYan, Zheng extent_len = btrfs_file_extent_num_bytes(leaf, fi); 4326ad48fd75SYan, Zheng } 4327b3b4aa74SDavid Sterba btrfs_release_path(path); 4328ad48fd75SYan, Zheng 4329ad48fd75SYan, Zheng path->keep_locks = 1; 4330ad48fd75SYan, Zheng path->search_for_split = 1; 4331ad48fd75SYan, Zheng ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 4332ad48fd75SYan, Zheng path->search_for_split = 0; 4333a8df6fe6SFilipe Manana if (ret > 0) 4334a8df6fe6SFilipe Manana ret = -EAGAIN; 4335ad48fd75SYan, Zheng if (ret < 0) 4336ad48fd75SYan, Zheng goto err; 4337ad48fd75SYan, Zheng 4338ad48fd75SYan, Zheng ret = -EAGAIN; 4339ad48fd75SYan, Zheng leaf = path->nodes[0]; 4340a8df6fe6SFilipe Manana /* if our item isn't there, return now */ 4341a8df6fe6SFilipe Manana if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) 4342ad48fd75SYan, Zheng goto err; 4343ad48fd75SYan, Zheng 4344109f6aefSChris Mason /* the leaf has changed, it now has room. return now */ 4345e902baacSDavid Sterba if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) 4346109f6aefSChris Mason goto err; 4347109f6aefSChris Mason 4348ad48fd75SYan, Zheng if (key.type == BTRFS_EXTENT_DATA_KEY) { 4349ad48fd75SYan, Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 4350ad48fd75SYan, Zheng struct btrfs_file_extent_item); 4351ad48fd75SYan, Zheng if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) 4352ad48fd75SYan, Zheng goto err; 4353ad48fd75SYan, Zheng } 4354ad48fd75SYan, Zheng 4355ad48fd75SYan, Zheng ret = split_leaf(trans, root, &key, path, ins_len, 1); 4356f0486c68SYan, Zheng if (ret) 4357f0486c68SYan, Zheng goto err; 4358ad48fd75SYan, Zheng 4359ad48fd75SYan, Zheng path->keep_locks = 0; 4360ad48fd75SYan, Zheng btrfs_unlock_up_safe(path, 1); 4361ad48fd75SYan, Zheng return 0; 4362ad48fd75SYan, Zheng err: 4363ad48fd75SYan, Zheng path->keep_locks = 0; 4364ad48fd75SYan, Zheng return ret; 4365ad48fd75SYan, Zheng } 4366ad48fd75SYan, Zheng 436725263cd7SDavid Sterba static noinline int split_item(struct btrfs_path *path, 4368310712b2SOmar Sandoval const struct btrfs_key *new_key, 4369459931ecSChris Mason unsigned long split_offset) 4370459931ecSChris Mason { 4371459931ecSChris Mason struct extent_buffer *leaf; 4372459931ecSChris Mason struct btrfs_item *item; 4373459931ecSChris Mason struct btrfs_item *new_item; 4374459931ecSChris Mason int slot; 4375ad48fd75SYan, Zheng char *buf; 4376459931ecSChris Mason u32 nritems; 4377ad48fd75SYan, Zheng u32 item_size; 4378459931ecSChris Mason u32 orig_offset; 4379459931ecSChris Mason struct btrfs_disk_key disk_key; 4380459931ecSChris Mason 4381459931ecSChris Mason leaf = path->nodes[0]; 4382e902baacSDavid Sterba BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); 4383b9473439SChris Mason 4384dd3cc16bSRoss Kirk item = btrfs_item_nr(path->slots[0]); 4385459931ecSChris Mason orig_offset = btrfs_item_offset(leaf, item); 4386459931ecSChris Mason item_size = btrfs_item_size(leaf, item); 4387459931ecSChris Mason 4388459931ecSChris Mason buf = kmalloc(item_size, GFP_NOFS); 4389ad48fd75SYan, Zheng if (!buf) 4390ad48fd75SYan, Zheng return -ENOMEM; 4391ad48fd75SYan, Zheng 4392459931ecSChris Mason read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, 4393459931ecSChris Mason path->slots[0]), item_size); 4394ad48fd75SYan, Zheng 4395459931ecSChris Mason slot = path->slots[0] + 1; 4396459931ecSChris Mason nritems = btrfs_header_nritems(leaf); 4397459931ecSChris Mason if (slot != nritems) { 4398459931ecSChris Mason /* shift the items */ 4399459931ecSChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), 4400459931ecSChris Mason btrfs_item_nr_offset(slot), 4401459931ecSChris Mason (nritems - slot) * sizeof(struct btrfs_item)); 4402459931ecSChris Mason } 4403459931ecSChris Mason 4404459931ecSChris Mason btrfs_cpu_key_to_disk(&disk_key, new_key); 4405459931ecSChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 4406459931ecSChris Mason 4407dd3cc16bSRoss Kirk new_item = btrfs_item_nr(slot); 4408459931ecSChris Mason 4409459931ecSChris Mason btrfs_set_item_offset(leaf, new_item, orig_offset); 4410459931ecSChris Mason btrfs_set_item_size(leaf, new_item, item_size - split_offset); 4411459931ecSChris Mason 4412459931ecSChris Mason btrfs_set_item_offset(leaf, item, 4413459931ecSChris Mason orig_offset + item_size - split_offset); 4414459931ecSChris Mason btrfs_set_item_size(leaf, item, split_offset); 4415459931ecSChris Mason 4416459931ecSChris Mason btrfs_set_header_nritems(leaf, nritems + 1); 4417459931ecSChris Mason 4418459931ecSChris Mason /* write the data for the start of the original item */ 4419459931ecSChris Mason write_extent_buffer(leaf, buf, 4420459931ecSChris Mason btrfs_item_ptr_offset(leaf, path->slots[0]), 4421459931ecSChris Mason split_offset); 4422459931ecSChris Mason 4423459931ecSChris Mason /* write the data for the new item */ 4424459931ecSChris Mason write_extent_buffer(leaf, buf + split_offset, 4425459931ecSChris Mason btrfs_item_ptr_offset(leaf, slot), 4426459931ecSChris Mason item_size - split_offset); 4427459931ecSChris Mason btrfs_mark_buffer_dirty(leaf); 4428459931ecSChris Mason 4429e902baacSDavid Sterba BUG_ON(btrfs_leaf_free_space(leaf) < 0); 4430459931ecSChris Mason kfree(buf); 4431ad48fd75SYan, Zheng return 0; 4432ad48fd75SYan, Zheng } 4433ad48fd75SYan, Zheng 4434ad48fd75SYan, Zheng /* 4435ad48fd75SYan, Zheng * This function splits a single item into two items, 4436ad48fd75SYan, Zheng * giving 'new_key' to the new item and splitting the 4437ad48fd75SYan, Zheng * old one at split_offset (from the start of the item). 4438ad48fd75SYan, Zheng * 4439ad48fd75SYan, Zheng * The path may be released by this operation. After 4440ad48fd75SYan, Zheng * the split, the path is pointing to the old item. The 4441ad48fd75SYan, Zheng * new item is going to be in the same node as the old one. 4442ad48fd75SYan, Zheng * 4443ad48fd75SYan, Zheng * Note, the item being split must be smaller enough to live alone on 4444ad48fd75SYan, Zheng * a tree block with room for one extra struct btrfs_item 4445ad48fd75SYan, Zheng * 4446ad48fd75SYan, Zheng * This allows us to split the item in place, keeping a lock on the 4447ad48fd75SYan, Zheng * leaf the entire time. 4448ad48fd75SYan, Zheng */ 4449ad48fd75SYan, Zheng int btrfs_split_item(struct btrfs_trans_handle *trans, 4450ad48fd75SYan, Zheng struct btrfs_root *root, 4451ad48fd75SYan, Zheng struct btrfs_path *path, 4452310712b2SOmar Sandoval const struct btrfs_key *new_key, 4453ad48fd75SYan, Zheng unsigned long split_offset) 4454ad48fd75SYan, Zheng { 4455ad48fd75SYan, Zheng int ret; 4456ad48fd75SYan, Zheng ret = setup_leaf_for_split(trans, root, path, 4457ad48fd75SYan, Zheng sizeof(struct btrfs_item)); 4458ad48fd75SYan, Zheng if (ret) 4459459931ecSChris Mason return ret; 4460ad48fd75SYan, Zheng 446125263cd7SDavid Sterba ret = split_item(path, new_key, split_offset); 4462ad48fd75SYan, Zheng return ret; 4463ad48fd75SYan, Zheng } 4464ad48fd75SYan, Zheng 4465ad48fd75SYan, Zheng /* 4466ad48fd75SYan, Zheng * This function duplicate a item, giving 'new_key' to the new item. 4467ad48fd75SYan, Zheng * It guarantees both items live in the same tree leaf and the new item 4468ad48fd75SYan, Zheng * is contiguous with the original item. 4469ad48fd75SYan, Zheng * 4470ad48fd75SYan, Zheng * This allows us to split file extent in place, keeping a lock on the 4471ad48fd75SYan, Zheng * leaf the entire time. 4472ad48fd75SYan, Zheng */ 4473ad48fd75SYan, Zheng int btrfs_duplicate_item(struct btrfs_trans_handle *trans, 4474ad48fd75SYan, Zheng struct btrfs_root *root, 4475ad48fd75SYan, Zheng struct btrfs_path *path, 4476310712b2SOmar Sandoval const struct btrfs_key *new_key) 4477ad48fd75SYan, Zheng { 4478ad48fd75SYan, Zheng struct extent_buffer *leaf; 4479ad48fd75SYan, Zheng int ret; 4480ad48fd75SYan, Zheng u32 item_size; 4481ad48fd75SYan, Zheng 4482ad48fd75SYan, Zheng leaf = path->nodes[0]; 4483ad48fd75SYan, Zheng item_size = btrfs_item_size_nr(leaf, path->slots[0]); 4484ad48fd75SYan, Zheng ret = setup_leaf_for_split(trans, root, path, 4485ad48fd75SYan, Zheng item_size + sizeof(struct btrfs_item)); 4486ad48fd75SYan, Zheng if (ret) 4487ad48fd75SYan, Zheng return ret; 4488ad48fd75SYan, Zheng 4489ad48fd75SYan, Zheng path->slots[0]++; 4490fc0d82e1SNikolay Borisov setup_items_for_insert(root, path, new_key, &item_size, 1); 4491ad48fd75SYan, Zheng leaf = path->nodes[0]; 4492ad48fd75SYan, Zheng memcpy_extent_buffer(leaf, 4493ad48fd75SYan, Zheng btrfs_item_ptr_offset(leaf, path->slots[0]), 4494ad48fd75SYan, Zheng btrfs_item_ptr_offset(leaf, path->slots[0] - 1), 4495ad48fd75SYan, Zheng item_size); 4496ad48fd75SYan, Zheng return 0; 4497459931ecSChris Mason } 4498459931ecSChris Mason 4499459931ecSChris Mason /* 4500d352ac68SChris Mason * make the item pointed to by the path smaller. new_size indicates 4501d352ac68SChris Mason * how small to make it, and from_end tells us if we just chop bytes 4502d352ac68SChris Mason * off the end of the item or if we shift the item to chop bytes off 4503d352ac68SChris Mason * the front. 4504d352ac68SChris Mason */ 450578ac4f9eSDavid Sterba void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) 4506b18c6685SChris Mason { 4507b18c6685SChris Mason int slot; 45085f39d397SChris Mason struct extent_buffer *leaf; 45095f39d397SChris Mason struct btrfs_item *item; 4510b18c6685SChris Mason u32 nritems; 4511b18c6685SChris Mason unsigned int data_end; 4512b18c6685SChris Mason unsigned int old_data_start; 4513b18c6685SChris Mason unsigned int old_size; 4514b18c6685SChris Mason unsigned int size_diff; 4515b18c6685SChris Mason int i; 4516cfed81a0SChris Mason struct btrfs_map_token token; 4517cfed81a0SChris Mason 45185f39d397SChris Mason leaf = path->nodes[0]; 4519179e29e4SChris Mason slot = path->slots[0]; 4520179e29e4SChris Mason 4521179e29e4SChris Mason old_size = btrfs_item_size_nr(leaf, slot); 4522179e29e4SChris Mason if (old_size == new_size) 4523143bede5SJeff Mahoney return; 4524b18c6685SChris Mason 45255f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 45268f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 4527b18c6685SChris Mason 45285f39d397SChris Mason old_data_start = btrfs_item_offset_nr(leaf, slot); 4529179e29e4SChris Mason 4530b18c6685SChris Mason size_diff = old_size - new_size; 4531b18c6685SChris Mason 4532b18c6685SChris Mason BUG_ON(slot < 0); 4533b18c6685SChris Mason BUG_ON(slot >= nritems); 4534b18c6685SChris Mason 4535b18c6685SChris Mason /* 4536b18c6685SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 4537b18c6685SChris Mason */ 4538b18c6685SChris Mason /* first correct the data pointers */ 4539c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 4540b18c6685SChris Mason for (i = slot; i < nritems; i++) { 45415f39d397SChris Mason u32 ioff; 4542dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 4543db94535dSChris Mason 4544cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4545cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, ioff + size_diff); 4546b18c6685SChris Mason } 4547db94535dSChris Mason 4548b18c6685SChris Mason /* shift the data */ 4549179e29e4SChris Mason if (from_end) { 45503d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 45513d9ec8c4SNikolay Borisov data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + 4552b18c6685SChris Mason data_end, old_data_start + new_size - data_end); 4553179e29e4SChris Mason } else { 4554179e29e4SChris Mason struct btrfs_disk_key disk_key; 4555179e29e4SChris Mason u64 offset; 4556179e29e4SChris Mason 4557179e29e4SChris Mason btrfs_item_key(leaf, &disk_key, slot); 4558179e29e4SChris Mason 4559179e29e4SChris Mason if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { 4560179e29e4SChris Mason unsigned long ptr; 4561179e29e4SChris Mason struct btrfs_file_extent_item *fi; 4562179e29e4SChris Mason 4563179e29e4SChris Mason fi = btrfs_item_ptr(leaf, slot, 4564179e29e4SChris Mason struct btrfs_file_extent_item); 4565179e29e4SChris Mason fi = (struct btrfs_file_extent_item *)( 4566179e29e4SChris Mason (unsigned long)fi - size_diff); 4567179e29e4SChris Mason 4568179e29e4SChris Mason if (btrfs_file_extent_type(leaf, fi) == 4569179e29e4SChris Mason BTRFS_FILE_EXTENT_INLINE) { 4570179e29e4SChris Mason ptr = btrfs_item_ptr_offset(leaf, slot); 4571179e29e4SChris Mason memmove_extent_buffer(leaf, ptr, 4572179e29e4SChris Mason (unsigned long)fi, 45737ec20afbSDavid Sterba BTRFS_FILE_EXTENT_INLINE_DATA_START); 4574179e29e4SChris Mason } 4575179e29e4SChris Mason } 4576179e29e4SChris Mason 45773d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 45783d9ec8c4SNikolay Borisov data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + 4579179e29e4SChris Mason data_end, old_data_start - data_end); 4580179e29e4SChris Mason 4581179e29e4SChris Mason offset = btrfs_disk_key_offset(&disk_key); 4582179e29e4SChris Mason btrfs_set_disk_key_offset(&disk_key, offset + size_diff); 4583179e29e4SChris Mason btrfs_set_item_key(leaf, &disk_key, slot); 4584179e29e4SChris Mason if (slot == 0) 4585b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 4586179e29e4SChris Mason } 45875f39d397SChris Mason 4588dd3cc16bSRoss Kirk item = btrfs_item_nr(slot); 45895f39d397SChris Mason btrfs_set_item_size(leaf, item, new_size); 45905f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 4591b18c6685SChris Mason 4592e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4593a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4594b18c6685SChris Mason BUG(); 45955f39d397SChris Mason } 4596b18c6685SChris Mason } 4597b18c6685SChris Mason 4598d352ac68SChris Mason /* 45998f69dbd2SStefan Behrens * make the item pointed to by the path bigger, data_size is the added size. 4600d352ac68SChris Mason */ 4601c71dd880SDavid Sterba void btrfs_extend_item(struct btrfs_path *path, u32 data_size) 46026567e837SChris Mason { 46036567e837SChris Mason int slot; 46045f39d397SChris Mason struct extent_buffer *leaf; 46055f39d397SChris Mason struct btrfs_item *item; 46066567e837SChris Mason u32 nritems; 46076567e837SChris Mason unsigned int data_end; 46086567e837SChris Mason unsigned int old_data; 46096567e837SChris Mason unsigned int old_size; 46106567e837SChris Mason int i; 4611cfed81a0SChris Mason struct btrfs_map_token token; 4612cfed81a0SChris Mason 46135f39d397SChris Mason leaf = path->nodes[0]; 46146567e837SChris Mason 46155f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 46168f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 46176567e837SChris Mason 4618e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < data_size) { 4619a4f78750SDavid Sterba btrfs_print_leaf(leaf); 46206567e837SChris Mason BUG(); 46215f39d397SChris Mason } 46226567e837SChris Mason slot = path->slots[0]; 46235f39d397SChris Mason old_data = btrfs_item_end_nr(leaf, slot); 46246567e837SChris Mason 46256567e837SChris Mason BUG_ON(slot < 0); 46263326d1b0SChris Mason if (slot >= nritems) { 4627a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4628c71dd880SDavid Sterba btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", 4629d397712bSChris Mason slot, nritems); 4630290342f6SArnd Bergmann BUG(); 46313326d1b0SChris Mason } 46326567e837SChris Mason 46336567e837SChris Mason /* 46346567e837SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 46356567e837SChris Mason */ 46366567e837SChris Mason /* first correct the data pointers */ 4637c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 46386567e837SChris Mason for (i = slot; i < nritems; i++) { 46395f39d397SChris Mason u32 ioff; 4640dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 4641db94535dSChris Mason 4642cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4643cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, ioff - data_size); 46446567e837SChris Mason } 46455f39d397SChris Mason 46466567e837SChris Mason /* shift the data */ 46473d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 46483d9ec8c4SNikolay Borisov data_end - data_size, BTRFS_LEAF_DATA_OFFSET + 46496567e837SChris Mason data_end, old_data - data_end); 46505f39d397SChris Mason 46516567e837SChris Mason data_end = old_data; 46525f39d397SChris Mason old_size = btrfs_item_size_nr(leaf, slot); 4653dd3cc16bSRoss Kirk item = btrfs_item_nr(slot); 46545f39d397SChris Mason btrfs_set_item_size(leaf, item, old_size + data_size); 46555f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 46566567e837SChris Mason 4657e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4658a4f78750SDavid Sterba btrfs_print_leaf(leaf); 46596567e837SChris Mason BUG(); 46605f39d397SChris Mason } 46616567e837SChris Mason } 46626567e837SChris Mason 4663da9ffb24SNikolay Borisov /** 4664da9ffb24SNikolay Borisov * setup_items_for_insert - Helper called before inserting one or more items 4665da9ffb24SNikolay Borisov * to a leaf. Main purpose is to save stack depth by doing the bulk of the work 4666da9ffb24SNikolay Borisov * in a function that doesn't call btrfs_search_slot 4667da9ffb24SNikolay Borisov * 4668da9ffb24SNikolay Borisov * @root: root we are inserting items to 4669da9ffb24SNikolay Borisov * @path: points to the leaf/slot where we are going to insert new items 4670da9ffb24SNikolay Borisov * @cpu_key: array of keys for items to be inserted 4671da9ffb24SNikolay Borisov * @data_size: size of the body of each item we are going to insert 4672da9ffb24SNikolay Borisov * @nr: size of @cpu_key/@data_size arrays 467374123bd7SChris Mason */ 4674afe5fea7STsutomu Itoh void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, 4675310712b2SOmar Sandoval const struct btrfs_key *cpu_key, u32 *data_size, 4676fc0d82e1SNikolay Borisov int nr) 4677be0e5c09SChris Mason { 46780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46795f39d397SChris Mason struct btrfs_item *item; 46809c58309dSChris Mason int i; 46817518a238SChris Mason u32 nritems; 4682be0e5c09SChris Mason unsigned int data_end; 4683e2fa7227SChris Mason struct btrfs_disk_key disk_key; 468444871b1bSChris Mason struct extent_buffer *leaf; 468544871b1bSChris Mason int slot; 4686cfed81a0SChris Mason struct btrfs_map_token token; 4687fc0d82e1SNikolay Borisov u32 total_size; 4688fc0d82e1SNikolay Borisov u32 total_data = 0; 4689fc0d82e1SNikolay Borisov 4690fc0d82e1SNikolay Borisov for (i = 0; i < nr; i++) 4691fc0d82e1SNikolay Borisov total_data += data_size[i]; 4692fc0d82e1SNikolay Borisov total_size = total_data + (nr * sizeof(struct btrfs_item)); 4693cfed81a0SChris Mason 469424cdc847SFilipe Manana if (path->slots[0] == 0) { 469524cdc847SFilipe Manana btrfs_cpu_key_to_disk(&disk_key, cpu_key); 4696b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 469724cdc847SFilipe Manana } 469824cdc847SFilipe Manana btrfs_unlock_up_safe(path, 1); 469924cdc847SFilipe Manana 47005f39d397SChris Mason leaf = path->nodes[0]; 470144871b1bSChris Mason slot = path->slots[0]; 470274123bd7SChris Mason 47035f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 47048f881e8cSDavid Sterba data_end = leaf_data_end(leaf); 4705eb60ceacSChris Mason 4706e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < total_size) { 4707a4f78750SDavid Sterba btrfs_print_leaf(leaf); 47080b246afaSJeff Mahoney btrfs_crit(fs_info, "not enough freespace need %u have %d", 4709e902baacSDavid Sterba total_size, btrfs_leaf_free_space(leaf)); 4710be0e5c09SChris Mason BUG(); 4711d4dbff95SChris Mason } 47125f39d397SChris Mason 4713c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 4714be0e5c09SChris Mason if (slot != nritems) { 47155f39d397SChris Mason unsigned int old_data = btrfs_item_end_nr(leaf, slot); 4716be0e5c09SChris Mason 47175f39d397SChris Mason if (old_data < data_end) { 4718a4f78750SDavid Sterba btrfs_print_leaf(leaf); 47197269ddd2SNikolay Borisov btrfs_crit(fs_info, 47207269ddd2SNikolay Borisov "item at slot %d with data offset %u beyond data end of leaf %u", 47215f39d397SChris Mason slot, old_data, data_end); 4722290342f6SArnd Bergmann BUG(); 47235f39d397SChris Mason } 4724be0e5c09SChris Mason /* 4725be0e5c09SChris Mason * item0..itemN ... dataN.offset..dataN.size .. data0.size 4726be0e5c09SChris Mason */ 4727be0e5c09SChris Mason /* first correct the data pointers */ 47280783fcfcSChris Mason for (i = slot; i < nritems; i++) { 47295f39d397SChris Mason u32 ioff; 4730db94535dSChris Mason 4731dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 4732cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4733cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, 4734cc4c13d5SDavid Sterba ioff - total_data); 47350783fcfcSChris Mason } 4736be0e5c09SChris Mason /* shift the items */ 47379c58309dSChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), 47385f39d397SChris Mason btrfs_item_nr_offset(slot), 47390783fcfcSChris Mason (nritems - slot) * sizeof(struct btrfs_item)); 4740be0e5c09SChris Mason 4741be0e5c09SChris Mason /* shift the data */ 47423d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 47433d9ec8c4SNikolay Borisov data_end - total_data, BTRFS_LEAF_DATA_OFFSET + 4744be0e5c09SChris Mason data_end, old_data - data_end); 4745be0e5c09SChris Mason data_end = old_data; 4746be0e5c09SChris Mason } 47475f39d397SChris Mason 474862e2749eSChris Mason /* setup the item for the new data */ 47499c58309dSChris Mason for (i = 0; i < nr; i++) { 47509c58309dSChris Mason btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); 47519c58309dSChris Mason btrfs_set_item_key(leaf, &disk_key, slot + i); 4752dd3cc16bSRoss Kirk item = btrfs_item_nr(slot + i); 47539c58309dSChris Mason data_end -= data_size[i]; 4754fc0716c2SNikolay Borisov btrfs_set_token_item_offset(&token, item, data_end); 4755cc4c13d5SDavid Sterba btrfs_set_token_item_size(&token, item, data_size[i]); 47569c58309dSChris Mason } 475744871b1bSChris Mason 47589c58309dSChris Mason btrfs_set_header_nritems(leaf, nritems + nr); 4759b9473439SChris Mason btrfs_mark_buffer_dirty(leaf); 4760aa5d6bedSChris Mason 4761e902baacSDavid Sterba if (btrfs_leaf_free_space(leaf) < 0) { 4762a4f78750SDavid Sterba btrfs_print_leaf(leaf); 4763be0e5c09SChris Mason BUG(); 47645f39d397SChris Mason } 476544871b1bSChris Mason } 476644871b1bSChris Mason 476744871b1bSChris Mason /* 476844871b1bSChris Mason * Given a key and some data, insert items into the tree. 476944871b1bSChris Mason * This does all the path init required, making room in the tree if needed. 477044871b1bSChris Mason */ 477144871b1bSChris Mason int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, 477244871b1bSChris Mason struct btrfs_root *root, 477344871b1bSChris Mason struct btrfs_path *path, 4774310712b2SOmar Sandoval const struct btrfs_key *cpu_key, u32 *data_size, 477544871b1bSChris Mason int nr) 477644871b1bSChris Mason { 477744871b1bSChris Mason int ret = 0; 477844871b1bSChris Mason int slot; 477944871b1bSChris Mason int i; 478044871b1bSChris Mason u32 total_size = 0; 478144871b1bSChris Mason u32 total_data = 0; 478244871b1bSChris Mason 478344871b1bSChris Mason for (i = 0; i < nr; i++) 478444871b1bSChris Mason total_data += data_size[i]; 478544871b1bSChris Mason 478644871b1bSChris Mason total_size = total_data + (nr * sizeof(struct btrfs_item)); 478744871b1bSChris Mason ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); 478844871b1bSChris Mason if (ret == 0) 478944871b1bSChris Mason return -EEXIST; 479044871b1bSChris Mason if (ret < 0) 4791143bede5SJeff Mahoney return ret; 479244871b1bSChris Mason 479344871b1bSChris Mason slot = path->slots[0]; 479444871b1bSChris Mason BUG_ON(slot < 0); 479544871b1bSChris Mason 4796fc0d82e1SNikolay Borisov setup_items_for_insert(root, path, cpu_key, data_size, nr); 4797143bede5SJeff Mahoney return 0; 479862e2749eSChris Mason } 479962e2749eSChris Mason 480062e2749eSChris Mason /* 480162e2749eSChris Mason * Given a key and some data, insert an item into the tree. 480262e2749eSChris Mason * This does all the path init required, making room in the tree if needed. 480362e2749eSChris Mason */ 4804310712b2SOmar Sandoval int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, 4805310712b2SOmar Sandoval const struct btrfs_key *cpu_key, void *data, 4806310712b2SOmar Sandoval u32 data_size) 480762e2749eSChris Mason { 480862e2749eSChris Mason int ret = 0; 48092c90e5d6SChris Mason struct btrfs_path *path; 48105f39d397SChris Mason struct extent_buffer *leaf; 48115f39d397SChris Mason unsigned long ptr; 481262e2749eSChris Mason 48132c90e5d6SChris Mason path = btrfs_alloc_path(); 4814db5b493aSTsutomu Itoh if (!path) 4815db5b493aSTsutomu Itoh return -ENOMEM; 48162c90e5d6SChris Mason ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); 481762e2749eSChris Mason if (!ret) { 48185f39d397SChris Mason leaf = path->nodes[0]; 48195f39d397SChris Mason ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); 48205f39d397SChris Mason write_extent_buffer(leaf, data, ptr, data_size); 48215f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 482262e2749eSChris Mason } 48232c90e5d6SChris Mason btrfs_free_path(path); 4824aa5d6bedSChris Mason return ret; 4825be0e5c09SChris Mason } 4826be0e5c09SChris Mason 482774123bd7SChris Mason /* 48285de08d7dSChris Mason * delete the pointer from a given node. 482974123bd7SChris Mason * 4830d352ac68SChris Mason * the tree should have been previously balanced so the deletion does not 4831d352ac68SChris Mason * empty a node. 483274123bd7SChris Mason */ 4833afe5fea7STsutomu Itoh static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, 4834afe5fea7STsutomu Itoh int level, int slot) 4835be0e5c09SChris Mason { 48365f39d397SChris Mason struct extent_buffer *parent = path->nodes[level]; 48377518a238SChris Mason u32 nritems; 4838f3ea38daSJan Schmidt int ret; 4839be0e5c09SChris Mason 48405f39d397SChris Mason nritems = btrfs_header_nritems(parent); 4841be0e5c09SChris Mason if (slot != nritems - 1) { 4842bf1d3425SDavid Sterba if (level) { 4843bf1d3425SDavid Sterba ret = tree_mod_log_insert_move(parent, slot, slot + 1, 4844a446a979SDavid Sterba nritems - slot - 1); 4845bf1d3425SDavid Sterba BUG_ON(ret < 0); 4846bf1d3425SDavid Sterba } 48475f39d397SChris Mason memmove_extent_buffer(parent, 48485f39d397SChris Mason btrfs_node_key_ptr_offset(slot), 48495f39d397SChris Mason btrfs_node_key_ptr_offset(slot + 1), 4850d6025579SChris Mason sizeof(struct btrfs_key_ptr) * 4851d6025579SChris Mason (nritems - slot - 1)); 485257ba86c0SChris Mason } else if (level) { 4853e09c2efeSDavid Sterba ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, 4854e09c2efeSDavid Sterba GFP_NOFS); 485557ba86c0SChris Mason BUG_ON(ret < 0); 4856be0e5c09SChris Mason } 4857f3ea38daSJan Schmidt 48587518a238SChris Mason nritems--; 48595f39d397SChris Mason btrfs_set_header_nritems(parent, nritems); 48607518a238SChris Mason if (nritems == 0 && parent == root->node) { 48615f39d397SChris Mason BUG_ON(btrfs_header_level(root->node) != 1); 4862eb60ceacSChris Mason /* just turn the root into a leaf and break */ 48635f39d397SChris Mason btrfs_set_header_level(root->node, 0); 4864bb803951SChris Mason } else if (slot == 0) { 48655f39d397SChris Mason struct btrfs_disk_key disk_key; 48665f39d397SChris Mason 48675f39d397SChris Mason btrfs_node_key(parent, &disk_key, 0); 4868b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, level + 1); 4869be0e5c09SChris Mason } 4870d6025579SChris Mason btrfs_mark_buffer_dirty(parent); 4871be0e5c09SChris Mason } 4872be0e5c09SChris Mason 487374123bd7SChris Mason /* 4874323ac95bSChris Mason * a helper function to delete the leaf pointed to by path->slots[1] and 48755d4f98a2SYan Zheng * path->nodes[1]. 4876323ac95bSChris Mason * 4877323ac95bSChris Mason * This deletes the pointer in path->nodes[1] and frees the leaf 4878323ac95bSChris Mason * block extent. zero is returned if it all worked out, < 0 otherwise. 4879323ac95bSChris Mason * 4880323ac95bSChris Mason * The path must have already been setup for deleting the leaf, including 4881323ac95bSChris Mason * all the proper balancing. path->nodes[1] must be locked. 4882323ac95bSChris Mason */ 4883143bede5SJeff Mahoney static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, 4884323ac95bSChris Mason struct btrfs_root *root, 48855d4f98a2SYan Zheng struct btrfs_path *path, 48865d4f98a2SYan Zheng struct extent_buffer *leaf) 4887323ac95bSChris Mason { 48885d4f98a2SYan Zheng WARN_ON(btrfs_header_generation(leaf) != trans->transid); 4889afe5fea7STsutomu Itoh del_ptr(root, path, 1, path->slots[1]); 4890323ac95bSChris Mason 48914d081c41SChris Mason /* 48924d081c41SChris Mason * btrfs_free_extent is expensive, we want to make sure we 48934d081c41SChris Mason * aren't holding any locks when we call it 48944d081c41SChris Mason */ 48954d081c41SChris Mason btrfs_unlock_up_safe(path, 0); 48964d081c41SChris Mason 4897f0486c68SYan, Zheng root_sub_used(root, leaf->len); 4898f0486c68SYan, Zheng 489967439dadSDavid Sterba atomic_inc(&leaf->refs); 49005581a51aSJan Schmidt btrfs_free_tree_block(trans, root, leaf, 0, 1); 49013083ee2eSJosef Bacik free_extent_buffer_stale(leaf); 4902323ac95bSChris Mason } 4903323ac95bSChris Mason /* 490474123bd7SChris Mason * delete the item at the leaf level in path. If that empties 490574123bd7SChris Mason * the leaf, remove it from the tree 490674123bd7SChris Mason */ 490785e21bacSChris Mason int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, 490885e21bacSChris Mason struct btrfs_path *path, int slot, int nr) 4909be0e5c09SChris Mason { 49100b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 49115f39d397SChris Mason struct extent_buffer *leaf; 49125f39d397SChris Mason struct btrfs_item *item; 4913ce0eac2aSAlexandru Moise u32 last_off; 4914ce0eac2aSAlexandru Moise u32 dsize = 0; 4915aa5d6bedSChris Mason int ret = 0; 4916aa5d6bedSChris Mason int wret; 491785e21bacSChris Mason int i; 49187518a238SChris Mason u32 nritems; 4919be0e5c09SChris Mason 49205f39d397SChris Mason leaf = path->nodes[0]; 492185e21bacSChris Mason last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); 492285e21bacSChris Mason 492385e21bacSChris Mason for (i = 0; i < nr; i++) 492485e21bacSChris Mason dsize += btrfs_item_size_nr(leaf, slot + i); 492585e21bacSChris Mason 49265f39d397SChris Mason nritems = btrfs_header_nritems(leaf); 4927be0e5c09SChris Mason 492885e21bacSChris Mason if (slot + nr != nritems) { 49298f881e8cSDavid Sterba int data_end = leaf_data_end(leaf); 4930c82f823cSDavid Sterba struct btrfs_map_token token; 49315f39d397SChris Mason 49323d9ec8c4SNikolay Borisov memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + 4933d6025579SChris Mason data_end + dsize, 49343d9ec8c4SNikolay Borisov BTRFS_LEAF_DATA_OFFSET + data_end, 493585e21bacSChris Mason last_off - data_end); 49365f39d397SChris Mason 4937c82f823cSDavid Sterba btrfs_init_map_token(&token, leaf); 493885e21bacSChris Mason for (i = slot + nr; i < nritems; i++) { 49395f39d397SChris Mason u32 ioff; 4940db94535dSChris Mason 4941dd3cc16bSRoss Kirk item = btrfs_item_nr(i); 4942cc4c13d5SDavid Sterba ioff = btrfs_token_item_offset(&token, item); 4943cc4c13d5SDavid Sterba btrfs_set_token_item_offset(&token, item, ioff + dsize); 49440783fcfcSChris Mason } 4945db94535dSChris Mason 49465f39d397SChris Mason memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), 494785e21bacSChris Mason btrfs_item_nr_offset(slot + nr), 49480783fcfcSChris Mason sizeof(struct btrfs_item) * 494985e21bacSChris Mason (nritems - slot - nr)); 4950be0e5c09SChris Mason } 495185e21bacSChris Mason btrfs_set_header_nritems(leaf, nritems - nr); 495285e21bacSChris Mason nritems -= nr; 49535f39d397SChris Mason 495474123bd7SChris Mason /* delete the leaf if we've emptied it */ 49557518a238SChris Mason if (nritems == 0) { 49565f39d397SChris Mason if (leaf == root->node) { 49575f39d397SChris Mason btrfs_set_header_level(leaf, 0); 49589a8dd150SChris Mason } else { 49596a884d7dSDavid Sterba btrfs_clean_tree_block(leaf); 4960143bede5SJeff Mahoney btrfs_del_leaf(trans, root, path, leaf); 49619a8dd150SChris Mason } 4962be0e5c09SChris Mason } else { 49637518a238SChris Mason int used = leaf_space_used(leaf, 0, nritems); 4964aa5d6bedSChris Mason if (slot == 0) { 49655f39d397SChris Mason struct btrfs_disk_key disk_key; 49665f39d397SChris Mason 49675f39d397SChris Mason btrfs_item_key(leaf, &disk_key, 0); 4968b167fa91SNikolay Borisov fixup_low_keys(path, &disk_key, 1); 4969aa5d6bedSChris Mason } 4970aa5d6bedSChris Mason 497174123bd7SChris Mason /* delete the leaf if it is mostly empty */ 49720b246afaSJeff Mahoney if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { 4973be0e5c09SChris Mason /* push_leaf_left fixes the path. 4974be0e5c09SChris Mason * make sure the path still points to our leaf 4975be0e5c09SChris Mason * for possible call to del_ptr below 4976be0e5c09SChris Mason */ 49774920c9acSChris Mason slot = path->slots[1]; 497867439dadSDavid Sterba atomic_inc(&leaf->refs); 49795f39d397SChris Mason 498099d8f83cSChris Mason wret = push_leaf_left(trans, root, path, 1, 1, 498199d8f83cSChris Mason 1, (u32)-1); 498254aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4983aa5d6bedSChris Mason ret = wret; 49845f39d397SChris Mason 49855f39d397SChris Mason if (path->nodes[0] == leaf && 49865f39d397SChris Mason btrfs_header_nritems(leaf)) { 498799d8f83cSChris Mason wret = push_leaf_right(trans, root, path, 1, 498899d8f83cSChris Mason 1, 1, 0); 498954aa1f4dSChris Mason if (wret < 0 && wret != -ENOSPC) 4990aa5d6bedSChris Mason ret = wret; 4991aa5d6bedSChris Mason } 49925f39d397SChris Mason 49935f39d397SChris Mason if (btrfs_header_nritems(leaf) == 0) { 4994323ac95bSChris Mason path->slots[1] = slot; 4995143bede5SJeff Mahoney btrfs_del_leaf(trans, root, path, leaf); 49965f39d397SChris Mason free_extent_buffer(leaf); 4997143bede5SJeff Mahoney ret = 0; 49985de08d7dSChris Mason } else { 4999925baeddSChris Mason /* if we're still in the path, make sure 5000925baeddSChris Mason * we're dirty. Otherwise, one of the 5001925baeddSChris Mason * push_leaf functions must have already 5002925baeddSChris Mason * dirtied this buffer 5003925baeddSChris Mason */ 5004925baeddSChris Mason if (path->nodes[0] == leaf) 50055f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 50065f39d397SChris Mason free_extent_buffer(leaf); 5007be0e5c09SChris Mason } 5008d5719762SChris Mason } else { 50095f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 5010be0e5c09SChris Mason } 50119a8dd150SChris Mason } 5012aa5d6bedSChris Mason return ret; 50139a8dd150SChris Mason } 50149a8dd150SChris Mason 501597571fd0SChris Mason /* 5016925baeddSChris Mason * search the tree again to find a leaf with lesser keys 50177bb86316SChris Mason * returns 0 if it found something or 1 if there are no lesser leaves. 50187bb86316SChris Mason * returns < 0 on io errors. 5019d352ac68SChris Mason * 5020d352ac68SChris Mason * This may release the path, and so you may lose any locks held at the 5021d352ac68SChris Mason * time you call it. 50227bb86316SChris Mason */ 502316e7549fSJosef Bacik int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) 50247bb86316SChris Mason { 5025925baeddSChris Mason struct btrfs_key key; 5026925baeddSChris Mason struct btrfs_disk_key found_key; 5027925baeddSChris Mason int ret; 50287bb86316SChris Mason 5029925baeddSChris Mason btrfs_item_key_to_cpu(path->nodes[0], &key, 0); 5030925baeddSChris Mason 5031e8b0d724SFilipe David Borba Manana if (key.offset > 0) { 5032925baeddSChris Mason key.offset--; 5033e8b0d724SFilipe David Borba Manana } else if (key.type > 0) { 5034925baeddSChris Mason key.type--; 5035e8b0d724SFilipe David Borba Manana key.offset = (u64)-1; 5036e8b0d724SFilipe David Borba Manana } else if (key.objectid > 0) { 5037925baeddSChris Mason key.objectid--; 5038e8b0d724SFilipe David Borba Manana key.type = (u8)-1; 5039e8b0d724SFilipe David Borba Manana key.offset = (u64)-1; 5040e8b0d724SFilipe David Borba Manana } else { 50417bb86316SChris Mason return 1; 5042e8b0d724SFilipe David Borba Manana } 50437bb86316SChris Mason 5044b3b4aa74SDavid Sterba btrfs_release_path(path); 5045925baeddSChris Mason ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 5046925baeddSChris Mason if (ret < 0) 5047925baeddSChris Mason return ret; 5048925baeddSChris Mason btrfs_item_key(path->nodes[0], &found_key, 0); 5049925baeddSChris Mason ret = comp_keys(&found_key, &key); 5050337c6f68SFilipe Manana /* 5051337c6f68SFilipe Manana * We might have had an item with the previous key in the tree right 5052337c6f68SFilipe Manana * before we released our path. And after we released our path, that 5053337c6f68SFilipe Manana * item might have been pushed to the first slot (0) of the leaf we 5054337c6f68SFilipe Manana * were holding due to a tree balance. Alternatively, an item with the 5055337c6f68SFilipe Manana * previous key can exist as the only element of a leaf (big fat item). 5056337c6f68SFilipe Manana * Therefore account for these 2 cases, so that our callers (like 5057337c6f68SFilipe Manana * btrfs_previous_item) don't miss an existing item with a key matching 5058337c6f68SFilipe Manana * the previous key we computed above. 5059337c6f68SFilipe Manana */ 5060337c6f68SFilipe Manana if (ret <= 0) 50617bb86316SChris Mason return 0; 5062925baeddSChris Mason return 1; 50637bb86316SChris Mason } 50647bb86316SChris Mason 50653f157a2fSChris Mason /* 50663f157a2fSChris Mason * A helper function to walk down the tree starting at min_key, and looking 5067de78b51aSEric Sandeen * for nodes or leaves that are have a minimum transaction id. 5068de78b51aSEric Sandeen * This is used by the btree defrag code, and tree logging 50693f157a2fSChris Mason * 50703f157a2fSChris Mason * This does not cow, but it does stuff the starting key it finds back 50713f157a2fSChris Mason * into min_key, so you can call btrfs_search_slot with cow=1 on the 50723f157a2fSChris Mason * key and get a writable path. 50733f157a2fSChris Mason * 50743f157a2fSChris Mason * This honors path->lowest_level to prevent descent past a given level 50753f157a2fSChris Mason * of the tree. 50763f157a2fSChris Mason * 5077d352ac68SChris Mason * min_trans indicates the oldest transaction that you are interested 5078d352ac68SChris Mason * in walking through. Any nodes or leaves older than min_trans are 5079d352ac68SChris Mason * skipped over (without reading them). 5080d352ac68SChris Mason * 50813f157a2fSChris Mason * returns zero if something useful was found, < 0 on error and 1 if there 50823f157a2fSChris Mason * was nothing in the tree that matched the search criteria. 50833f157a2fSChris Mason */ 50843f157a2fSChris Mason int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, 5085de78b51aSEric Sandeen struct btrfs_path *path, 50863f157a2fSChris Mason u64 min_trans) 50873f157a2fSChris Mason { 50883f157a2fSChris Mason struct extent_buffer *cur; 50893f157a2fSChris Mason struct btrfs_key found_key; 50903f157a2fSChris Mason int slot; 50919652480bSYan int sret; 50923f157a2fSChris Mason u32 nritems; 50933f157a2fSChris Mason int level; 50943f157a2fSChris Mason int ret = 1; 5095f98de9b9SFilipe Manana int keep_locks = path->keep_locks; 50963f157a2fSChris Mason 5097f98de9b9SFilipe Manana path->keep_locks = 1; 50983f157a2fSChris Mason again: 5099bd681513SChris Mason cur = btrfs_read_lock_root_node(root); 51003f157a2fSChris Mason level = btrfs_header_level(cur); 5101e02119d5SChris Mason WARN_ON(path->nodes[level]); 51023f157a2fSChris Mason path->nodes[level] = cur; 5103bd681513SChris Mason path->locks[level] = BTRFS_READ_LOCK; 51043f157a2fSChris Mason 51053f157a2fSChris Mason if (btrfs_header_generation(cur) < min_trans) { 51063f157a2fSChris Mason ret = 1; 51073f157a2fSChris Mason goto out; 51083f157a2fSChris Mason } 51093f157a2fSChris Mason while (1) { 51103f157a2fSChris Mason nritems = btrfs_header_nritems(cur); 51113f157a2fSChris Mason level = btrfs_header_level(cur); 5112e3b83361SQu Wenruo sret = btrfs_bin_search(cur, min_key, &slot); 5113cbca7d59SFilipe Manana if (sret < 0) { 5114cbca7d59SFilipe Manana ret = sret; 5115cbca7d59SFilipe Manana goto out; 5116cbca7d59SFilipe Manana } 51173f157a2fSChris Mason 5118323ac95bSChris Mason /* at the lowest level, we're done, setup the path and exit */ 5119323ac95bSChris Mason if (level == path->lowest_level) { 5120e02119d5SChris Mason if (slot >= nritems) 5121e02119d5SChris Mason goto find_next_key; 51223f157a2fSChris Mason ret = 0; 51233f157a2fSChris Mason path->slots[level] = slot; 51243f157a2fSChris Mason btrfs_item_key_to_cpu(cur, &found_key, slot); 51253f157a2fSChris Mason goto out; 51263f157a2fSChris Mason } 51279652480bSYan if (sret && slot > 0) 51289652480bSYan slot--; 51293f157a2fSChris Mason /* 5130de78b51aSEric Sandeen * check this node pointer against the min_trans parameters. 5131260db43cSRandy Dunlap * If it is too old, skip to the next one. 51323f157a2fSChris Mason */ 51333f157a2fSChris Mason while (slot < nritems) { 51343f157a2fSChris Mason u64 gen; 5135e02119d5SChris Mason 51363f157a2fSChris Mason gen = btrfs_node_ptr_generation(cur, slot); 51373f157a2fSChris Mason if (gen < min_trans) { 51383f157a2fSChris Mason slot++; 51393f157a2fSChris Mason continue; 51403f157a2fSChris Mason } 51413f157a2fSChris Mason break; 51423f157a2fSChris Mason } 5143e02119d5SChris Mason find_next_key: 51443f157a2fSChris Mason /* 51453f157a2fSChris Mason * we didn't find a candidate key in this node, walk forward 51463f157a2fSChris Mason * and find another one 51473f157a2fSChris Mason */ 51483f157a2fSChris Mason if (slot >= nritems) { 5149e02119d5SChris Mason path->slots[level] = slot; 5150e02119d5SChris Mason sret = btrfs_find_next_key(root, path, min_key, level, 5151de78b51aSEric Sandeen min_trans); 5152e02119d5SChris Mason if (sret == 0) { 5153b3b4aa74SDavid Sterba btrfs_release_path(path); 51543f157a2fSChris Mason goto again; 51553f157a2fSChris Mason } else { 51563f157a2fSChris Mason goto out; 51573f157a2fSChris Mason } 51583f157a2fSChris Mason } 51593f157a2fSChris Mason /* save our key for returning back */ 51603f157a2fSChris Mason btrfs_node_key_to_cpu(cur, &found_key, slot); 51613f157a2fSChris Mason path->slots[level] = slot; 51623f157a2fSChris Mason if (level == path->lowest_level) { 51633f157a2fSChris Mason ret = 0; 51643f157a2fSChris Mason goto out; 51653f157a2fSChris Mason } 51664b231ae4SDavid Sterba cur = btrfs_read_node_slot(cur, slot); 5167fb770ae4SLiu Bo if (IS_ERR(cur)) { 5168fb770ae4SLiu Bo ret = PTR_ERR(cur); 5169fb770ae4SLiu Bo goto out; 5170fb770ae4SLiu Bo } 51713f157a2fSChris Mason 5172bd681513SChris Mason btrfs_tree_read_lock(cur); 5173b4ce94deSChris Mason 5174bd681513SChris Mason path->locks[level - 1] = BTRFS_READ_LOCK; 51753f157a2fSChris Mason path->nodes[level - 1] = cur; 5176f7c79f30SChris Mason unlock_up(path, level, 1, 0, NULL); 51773f157a2fSChris Mason } 51783f157a2fSChris Mason out: 5179f98de9b9SFilipe Manana path->keep_locks = keep_locks; 5180f98de9b9SFilipe Manana if (ret == 0) { 5181f98de9b9SFilipe Manana btrfs_unlock_up_safe(path, path->lowest_level + 1); 5182f98de9b9SFilipe Manana memcpy(min_key, &found_key, sizeof(found_key)); 5183f98de9b9SFilipe Manana } 51843f157a2fSChris Mason return ret; 51853f157a2fSChris Mason } 51863f157a2fSChris Mason 51873f157a2fSChris Mason /* 51883f157a2fSChris Mason * this is similar to btrfs_next_leaf, but does not try to preserve 51893f157a2fSChris Mason * and fixup the path. It looks for and returns the next key in the 5190de78b51aSEric Sandeen * tree based on the current path and the min_trans parameters. 51913f157a2fSChris Mason * 51923f157a2fSChris Mason * 0 is returned if another key is found, < 0 if there are any errors 51933f157a2fSChris Mason * and 1 is returned if there are no higher keys in the tree 51943f157a2fSChris Mason * 51953f157a2fSChris Mason * path->keep_locks should be set to 1 on the search made before 51963f157a2fSChris Mason * calling this function. 51973f157a2fSChris Mason */ 5198e7a84565SChris Mason int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, 5199de78b51aSEric Sandeen struct btrfs_key *key, int level, u64 min_trans) 5200e7a84565SChris Mason { 5201e7a84565SChris Mason int slot; 5202e7a84565SChris Mason struct extent_buffer *c; 5203e7a84565SChris Mason 52046a9fb468SJosef Bacik WARN_ON(!path->keep_locks && !path->skip_locking); 5205e7a84565SChris Mason while (level < BTRFS_MAX_LEVEL) { 5206e7a84565SChris Mason if (!path->nodes[level]) 5207e7a84565SChris Mason return 1; 5208e7a84565SChris Mason 5209e7a84565SChris Mason slot = path->slots[level] + 1; 5210e7a84565SChris Mason c = path->nodes[level]; 52113f157a2fSChris Mason next: 5212e7a84565SChris Mason if (slot >= btrfs_header_nritems(c)) { 521333c66f43SYan Zheng int ret; 521433c66f43SYan Zheng int orig_lowest; 521533c66f43SYan Zheng struct btrfs_key cur_key; 521633c66f43SYan Zheng if (level + 1 >= BTRFS_MAX_LEVEL || 521733c66f43SYan Zheng !path->nodes[level + 1]) 5218e7a84565SChris Mason return 1; 521933c66f43SYan Zheng 52206a9fb468SJosef Bacik if (path->locks[level + 1] || path->skip_locking) { 522133c66f43SYan Zheng level++; 5222e7a84565SChris Mason continue; 5223e7a84565SChris Mason } 522433c66f43SYan Zheng 522533c66f43SYan Zheng slot = btrfs_header_nritems(c) - 1; 522633c66f43SYan Zheng if (level == 0) 522733c66f43SYan Zheng btrfs_item_key_to_cpu(c, &cur_key, slot); 522833c66f43SYan Zheng else 522933c66f43SYan Zheng btrfs_node_key_to_cpu(c, &cur_key, slot); 523033c66f43SYan Zheng 523133c66f43SYan Zheng orig_lowest = path->lowest_level; 5232b3b4aa74SDavid Sterba btrfs_release_path(path); 523333c66f43SYan Zheng path->lowest_level = level; 523433c66f43SYan Zheng ret = btrfs_search_slot(NULL, root, &cur_key, path, 523533c66f43SYan Zheng 0, 0); 523633c66f43SYan Zheng path->lowest_level = orig_lowest; 523733c66f43SYan Zheng if (ret < 0) 523833c66f43SYan Zheng return ret; 523933c66f43SYan Zheng 524033c66f43SYan Zheng c = path->nodes[level]; 524133c66f43SYan Zheng slot = path->slots[level]; 524233c66f43SYan Zheng if (ret == 0) 524333c66f43SYan Zheng slot++; 524433c66f43SYan Zheng goto next; 524533c66f43SYan Zheng } 524633c66f43SYan Zheng 5247e7a84565SChris Mason if (level == 0) 5248e7a84565SChris Mason btrfs_item_key_to_cpu(c, key, slot); 52493f157a2fSChris Mason else { 52503f157a2fSChris Mason u64 gen = btrfs_node_ptr_generation(c, slot); 52513f157a2fSChris Mason 52523f157a2fSChris Mason if (gen < min_trans) { 52533f157a2fSChris Mason slot++; 52543f157a2fSChris Mason goto next; 52553f157a2fSChris Mason } 5256e7a84565SChris Mason btrfs_node_key_to_cpu(c, key, slot); 52573f157a2fSChris Mason } 5258e7a84565SChris Mason return 0; 5259e7a84565SChris Mason } 5260e7a84565SChris Mason return 1; 5261e7a84565SChris Mason } 5262e7a84565SChris Mason 52637bb86316SChris Mason /* 5264925baeddSChris Mason * search the tree again to find a leaf with greater keys 52650f70abe2SChris Mason * returns 0 if it found something or 1 if there are no greater leaves. 52660f70abe2SChris Mason * returns < 0 on io errors. 526797571fd0SChris Mason */ 5268234b63a0SChris Mason int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) 5269d97e63b6SChris Mason { 52703d7806ecSJan Schmidt return btrfs_next_old_leaf(root, path, 0); 52713d7806ecSJan Schmidt } 52723d7806ecSJan Schmidt 52733d7806ecSJan Schmidt int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, 52743d7806ecSJan Schmidt u64 time_seq) 52753d7806ecSJan Schmidt { 5276d97e63b6SChris Mason int slot; 52778e73f275SChris Mason int level; 52785f39d397SChris Mason struct extent_buffer *c; 52798e73f275SChris Mason struct extent_buffer *next; 5280925baeddSChris Mason struct btrfs_key key; 5281925baeddSChris Mason u32 nritems; 5282925baeddSChris Mason int ret; 52830e46318dSJosef Bacik int i; 5284925baeddSChris Mason 5285925baeddSChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 5286d397712bSChris Mason if (nritems == 0) 5287925baeddSChris Mason return 1; 5288925baeddSChris Mason 52898e73f275SChris Mason btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); 52908e73f275SChris Mason again: 52918e73f275SChris Mason level = 1; 52928e73f275SChris Mason next = NULL; 5293b3b4aa74SDavid Sterba btrfs_release_path(path); 52948e73f275SChris Mason 5295a2135011SChris Mason path->keep_locks = 1; 52968e73f275SChris Mason 52973d7806ecSJan Schmidt if (time_seq) 52983d7806ecSJan Schmidt ret = btrfs_search_old_slot(root, &key, path, time_seq); 52993d7806ecSJan Schmidt else 5300925baeddSChris Mason ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 5301925baeddSChris Mason path->keep_locks = 0; 5302925baeddSChris Mason 5303925baeddSChris Mason if (ret < 0) 5304925baeddSChris Mason return ret; 5305925baeddSChris Mason 5306a2135011SChris Mason nritems = btrfs_header_nritems(path->nodes[0]); 5307168fd7d2SChris Mason /* 5308168fd7d2SChris Mason * by releasing the path above we dropped all our locks. A balance 5309168fd7d2SChris Mason * could have added more items next to the key that used to be 5310168fd7d2SChris Mason * at the very end of the block. So, check again here and 5311168fd7d2SChris Mason * advance the path if there are now more items available. 5312168fd7d2SChris Mason */ 5313a2135011SChris Mason if (nritems > 0 && path->slots[0] < nritems - 1) { 5314e457afecSYan Zheng if (ret == 0) 5315168fd7d2SChris Mason path->slots[0]++; 53168e73f275SChris Mason ret = 0; 5317925baeddSChris Mason goto done; 5318925baeddSChris Mason } 53190b43e04fSLiu Bo /* 53200b43e04fSLiu Bo * So the above check misses one case: 53210b43e04fSLiu Bo * - after releasing the path above, someone has removed the item that 53220b43e04fSLiu Bo * used to be at the very end of the block, and balance between leafs 53230b43e04fSLiu Bo * gets another one with bigger key.offset to replace it. 53240b43e04fSLiu Bo * 53250b43e04fSLiu Bo * This one should be returned as well, or we can get leaf corruption 53260b43e04fSLiu Bo * later(esp. in __btrfs_drop_extents()). 53270b43e04fSLiu Bo * 53280b43e04fSLiu Bo * And a bit more explanation about this check, 53290b43e04fSLiu Bo * with ret > 0, the key isn't found, the path points to the slot 53300b43e04fSLiu Bo * where it should be inserted, so the path->slots[0] item must be the 53310b43e04fSLiu Bo * bigger one. 53320b43e04fSLiu Bo */ 53330b43e04fSLiu Bo if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { 53340b43e04fSLiu Bo ret = 0; 53350b43e04fSLiu Bo goto done; 53360b43e04fSLiu Bo } 5337d97e63b6SChris Mason 5338234b63a0SChris Mason while (level < BTRFS_MAX_LEVEL) { 53398e73f275SChris Mason if (!path->nodes[level]) { 53408e73f275SChris Mason ret = 1; 53418e73f275SChris Mason goto done; 53428e73f275SChris Mason } 53435f39d397SChris Mason 5344d97e63b6SChris Mason slot = path->slots[level] + 1; 5345d97e63b6SChris Mason c = path->nodes[level]; 53465f39d397SChris Mason if (slot >= btrfs_header_nritems(c)) { 5347d97e63b6SChris Mason level++; 53488e73f275SChris Mason if (level == BTRFS_MAX_LEVEL) { 53498e73f275SChris Mason ret = 1; 53508e73f275SChris Mason goto done; 53518e73f275SChris Mason } 5352d97e63b6SChris Mason continue; 5353d97e63b6SChris Mason } 53545f39d397SChris Mason 53550e46318dSJosef Bacik 53560e46318dSJosef Bacik /* 53570e46318dSJosef Bacik * Our current level is where we're going to start from, and to 53580e46318dSJosef Bacik * make sure lockdep doesn't complain we need to drop our locks 53590e46318dSJosef Bacik * and nodes from 0 to our current level. 53600e46318dSJosef Bacik */ 53610e46318dSJosef Bacik for (i = 0; i < level; i++) { 53620e46318dSJosef Bacik if (path->locks[level]) { 53630e46318dSJosef Bacik btrfs_tree_read_unlock(path->nodes[i]); 53640e46318dSJosef Bacik path->locks[i] = 0; 53650e46318dSJosef Bacik } 53660e46318dSJosef Bacik free_extent_buffer(path->nodes[i]); 53670e46318dSJosef Bacik path->nodes[i] = NULL; 5368925baeddSChris Mason } 53695f39d397SChris Mason 53708e73f275SChris Mason next = c; 5371d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 5372cda79c54SDavid Sterba slot, &key); 53738e73f275SChris Mason if (ret == -EAGAIN) 53748e73f275SChris Mason goto again; 53755f39d397SChris Mason 537676a05b35SChris Mason if (ret < 0) { 5377b3b4aa74SDavid Sterba btrfs_release_path(path); 537876a05b35SChris Mason goto done; 537976a05b35SChris Mason } 538076a05b35SChris Mason 53815cd57b2cSChris Mason if (!path->skip_locking) { 5382bd681513SChris Mason ret = btrfs_try_tree_read_lock(next); 5383d42244a0SJan Schmidt if (!ret && time_seq) { 5384d42244a0SJan Schmidt /* 5385d42244a0SJan Schmidt * If we don't get the lock, we may be racing 5386d42244a0SJan Schmidt * with push_leaf_left, holding that lock while 5387d42244a0SJan Schmidt * itself waiting for the leaf we've currently 5388d42244a0SJan Schmidt * locked. To solve this situation, we give up 5389d42244a0SJan Schmidt * on our lock and cycle. 5390d42244a0SJan Schmidt */ 5391cf538830SJan Schmidt free_extent_buffer(next); 5392d42244a0SJan Schmidt btrfs_release_path(path); 5393d42244a0SJan Schmidt cond_resched(); 5394d42244a0SJan Schmidt goto again; 5395d42244a0SJan Schmidt } 53960e46318dSJosef Bacik if (!ret) 53970e46318dSJosef Bacik btrfs_tree_read_lock(next); 5398bd681513SChris Mason } 5399d97e63b6SChris Mason break; 5400d97e63b6SChris Mason } 5401d97e63b6SChris Mason path->slots[level] = slot; 5402d97e63b6SChris Mason while (1) { 5403d97e63b6SChris Mason level--; 5404d97e63b6SChris Mason path->nodes[level] = next; 5405d97e63b6SChris Mason path->slots[level] = 0; 5406a74a4b97SChris Mason if (!path->skip_locking) 5407ffeb03cfSJosef Bacik path->locks[level] = BTRFS_READ_LOCK; 5408d97e63b6SChris Mason if (!level) 5409d97e63b6SChris Mason break; 5410b4ce94deSChris Mason 5411d07b8528SLiu Bo ret = read_block_for_search(root, path, &next, level, 5412cda79c54SDavid Sterba 0, &key); 54138e73f275SChris Mason if (ret == -EAGAIN) 54148e73f275SChris Mason goto again; 54158e73f275SChris Mason 541676a05b35SChris Mason if (ret < 0) { 5417b3b4aa74SDavid Sterba btrfs_release_path(path); 541876a05b35SChris Mason goto done; 541976a05b35SChris Mason } 542076a05b35SChris Mason 5421ffeb03cfSJosef Bacik if (!path->skip_locking) 54220e46318dSJosef Bacik btrfs_tree_read_lock(next); 5423d97e63b6SChris Mason } 54248e73f275SChris Mason ret = 0; 5425925baeddSChris Mason done: 5426f7c79f30SChris Mason unlock_up(path, 0, 1, 0, NULL); 54278e73f275SChris Mason 54288e73f275SChris Mason return ret; 5429d97e63b6SChris Mason } 54300b86a832SChris Mason 54313f157a2fSChris Mason /* 54323f157a2fSChris Mason * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps 54333f157a2fSChris Mason * searching until it gets past min_objectid or finds an item of 'type' 54343f157a2fSChris Mason * 54353f157a2fSChris Mason * returns 0 if something is found, 1 if nothing was found and < 0 on error 54363f157a2fSChris Mason */ 54370b86a832SChris Mason int btrfs_previous_item(struct btrfs_root *root, 54380b86a832SChris Mason struct btrfs_path *path, u64 min_objectid, 54390b86a832SChris Mason int type) 54400b86a832SChris Mason { 54410b86a832SChris Mason struct btrfs_key found_key; 54420b86a832SChris Mason struct extent_buffer *leaf; 5443e02119d5SChris Mason u32 nritems; 54440b86a832SChris Mason int ret; 54450b86a832SChris Mason 54460b86a832SChris Mason while (1) { 54470b86a832SChris Mason if (path->slots[0] == 0) { 54480b86a832SChris Mason ret = btrfs_prev_leaf(root, path); 54490b86a832SChris Mason if (ret != 0) 54500b86a832SChris Mason return ret; 54510b86a832SChris Mason } else { 54520b86a832SChris Mason path->slots[0]--; 54530b86a832SChris Mason } 54540b86a832SChris Mason leaf = path->nodes[0]; 5455e02119d5SChris Mason nritems = btrfs_header_nritems(leaf); 5456e02119d5SChris Mason if (nritems == 0) 5457e02119d5SChris Mason return 1; 5458e02119d5SChris Mason if (path->slots[0] == nritems) 5459e02119d5SChris Mason path->slots[0]--; 5460e02119d5SChris Mason 54610b86a832SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 5462e02119d5SChris Mason if (found_key.objectid < min_objectid) 5463e02119d5SChris Mason break; 54640a4eefbbSYan Zheng if (found_key.type == type) 54650a4eefbbSYan Zheng return 0; 5466e02119d5SChris Mason if (found_key.objectid == min_objectid && 5467e02119d5SChris Mason found_key.type < type) 5468e02119d5SChris Mason break; 54690b86a832SChris Mason } 54700b86a832SChris Mason return 1; 54710b86a832SChris Mason } 5472ade2e0b3SWang Shilong 5473ade2e0b3SWang Shilong /* 5474ade2e0b3SWang Shilong * search in extent tree to find a previous Metadata/Data extent item with 5475ade2e0b3SWang Shilong * min objecitd. 5476ade2e0b3SWang Shilong * 5477ade2e0b3SWang Shilong * returns 0 if something is found, 1 if nothing was found and < 0 on error 5478ade2e0b3SWang Shilong */ 5479ade2e0b3SWang Shilong int btrfs_previous_extent_item(struct btrfs_root *root, 5480ade2e0b3SWang Shilong struct btrfs_path *path, u64 min_objectid) 5481ade2e0b3SWang Shilong { 5482ade2e0b3SWang Shilong struct btrfs_key found_key; 5483ade2e0b3SWang Shilong struct extent_buffer *leaf; 5484ade2e0b3SWang Shilong u32 nritems; 5485ade2e0b3SWang Shilong int ret; 5486ade2e0b3SWang Shilong 5487ade2e0b3SWang Shilong while (1) { 5488ade2e0b3SWang Shilong if (path->slots[0] == 0) { 5489ade2e0b3SWang Shilong ret = btrfs_prev_leaf(root, path); 5490ade2e0b3SWang Shilong if (ret != 0) 5491ade2e0b3SWang Shilong return ret; 5492ade2e0b3SWang Shilong } else { 5493ade2e0b3SWang Shilong path->slots[0]--; 5494ade2e0b3SWang Shilong } 5495ade2e0b3SWang Shilong leaf = path->nodes[0]; 5496ade2e0b3SWang Shilong nritems = btrfs_header_nritems(leaf); 5497ade2e0b3SWang Shilong if (nritems == 0) 5498ade2e0b3SWang Shilong return 1; 5499ade2e0b3SWang Shilong if (path->slots[0] == nritems) 5500ade2e0b3SWang Shilong path->slots[0]--; 5501ade2e0b3SWang Shilong 5502ade2e0b3SWang Shilong btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 5503ade2e0b3SWang Shilong if (found_key.objectid < min_objectid) 5504ade2e0b3SWang Shilong break; 5505ade2e0b3SWang Shilong if (found_key.type == BTRFS_EXTENT_ITEM_KEY || 5506ade2e0b3SWang Shilong found_key.type == BTRFS_METADATA_ITEM_KEY) 5507ade2e0b3SWang Shilong return 0; 5508ade2e0b3SWang Shilong if (found_key.objectid == min_objectid && 5509ade2e0b3SWang Shilong found_key.type < BTRFS_EXTENT_ITEM_KEY) 5510ade2e0b3SWang Shilong break; 5511ade2e0b3SWang Shilong } 5512ade2e0b3SWang Shilong return 1; 5513ade2e0b3SWang Shilong } 5514