root/fs/btrfs/ctree.c

DEFINITIONS

1. btrfs_super_csum_size
2. btrfs_super_csum_name
3. btrfs_alloc_path
4. btrfs_set_path_blocking
5. btrfs_free_path
6. btrfs_release_path
7. btrfs_root_node
8. btrfs_lock_root_node
9. btrfs_read_lock_root_node
10. add_root_to_dirty_list
11. btrfs_copy_root
12. btrfs_inc_tree_mod_seq
13. btrfs_get_tree_mod_seq
14. btrfs_put_tree_mod_seq
15. __tree_mod_log_insert
16. tree_mod_dont_log
17. tree_mod_need_log
18. alloc_tree_mod_elem
19. tree_mod_log_insert_key
20. tree_mod_log_insert_move
21. __tree_mod_log_free_eb
22. tree_mod_log_insert_root
23. __tree_mod_log_search
24. tree_mod_log_search_oldest
25. tree_mod_log_search
26. tree_mod_log_eb_copy
27. tree_mod_log_free_eb
28. btrfs_block_can_be_shared
29. update_ref_for_cow
30. alloc_tree_block_no_bg_flush
31. __btrfs_cow_block
32. __tree_mod_log_oldest_root
33. __tree_mod_log_rewind
34. tree_mod_log_rewind
35. get_old_root
36. btrfs_old_root_level
37. should_cow_block
38. btrfs_cow_block
39. close_blocks
40. comp_keys
41. btrfs_comp_cpu_keys
42. btrfs_realloc_node
43. generic_bin_search
44. btrfs_bin_search
45. root_add_used
46. root_sub_used
47. btrfs_read_node_slot
48. balance_level
49. push_nodes_for_insert
50. reada_for_search
51. reada_for_balance
52. unlock_up
53. btrfs_unlock_up_safe
54. read_block_for_search
55. setup_nodes_for_search
56. key_search
57. btrfs_find_item
58. btrfs_search_slot_get_root
59. btrfs_search_slot
60. btrfs_search_old_slot
61. btrfs_search_slot_for_read
62. fixup_low_keys
63. btrfs_set_item_key_safe
64. push_node_left
65. balance_node_right
66. insert_new_root
67. insert_ptr
68. split_node
69. leaf_space_used
70. btrfs_leaf_free_space
71. __push_leaf_right
72. push_leaf_right
73. __push_leaf_left
74. push_leaf_left
75. copy_for_split
76. push_for_double_split
77. split_leaf
78. setup_leaf_for_split
79. split_item
80. btrfs_split_item
81. btrfs_duplicate_item
82. btrfs_truncate_item
83. btrfs_extend_item
84. setup_items_for_insert
85. btrfs_insert_empty_items
86. btrfs_insert_item
87. del_ptr
88. btrfs_del_leaf
89. btrfs_del_items
90. btrfs_prev_leaf
91. btrfs_search_forward
92. btrfs_find_next_key
93. btrfs_next_leaf
94. btrfs_next_old_leaf
95. btrfs_previous_item
96. btrfs_previous_extent_item

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007,2008 Oracle.  All rights reserved.
   4  */
   5 
   6 #include <linux/sched.h>
   7 #include <linux/slab.h>
   8 #include <linux/rbtree.h>
   9 #include <linux/mm.h>
  10 #include "ctree.h"
  11 #include "disk-io.h"
  12 #include "transaction.h"
  13 #include "print-tree.h"
  14 #include "locking.h"
  15 #include "volumes.h"
  16 #include "qgroup.h"
  17 
  18 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
  19                       *root, struct btrfs_path *path, int level);
  20 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  21                       const struct btrfs_key *ins_key, struct btrfs_path *path,
  22                       int data_size, int extend);
  23 static int push_node_left(struct btrfs_trans_handle *trans,
  24                           struct extent_buffer *dst,
  25                           struct extent_buffer *src, int empty);
  26 static int balance_node_right(struct btrfs_trans_handle *trans,
  27                               struct extent_buffer *dst_buf,
  28                               struct extent_buffer *src_buf);
  29 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
  30                     int level, int slot);
  31 
  32 static const struct btrfs_csums {
  33         u16             size;
  34         const char      *name;
  35 } btrfs_csums[] = {
  36         [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" },
  37 };
  38 
  39 int btrfs_super_csum_size(const struct btrfs_super_block *s)
  40 {
  41         u16 t = btrfs_super_csum_type(s);
  42         /*
  43          * csum type is validated at mount time
  44          */
  45         return btrfs_csums[t].size;
  46 }
  47 
  48 const char *btrfs_super_csum_name(u16 csum_type)
  49 {
  50         /* csum type is validated at mount time */
  51         return btrfs_csums[csum_type].name;
  52 }
  53 
  54 struct btrfs_path *btrfs_alloc_path(void)
  55 {
  56         return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
  57 }
  58 
  59 /*
  60  * set all locked nodes in the path to blocking locks.  This should
  61  * be done before scheduling
  62  */
  63 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
  64 {
  65         int i;
  66         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
  67                 if (!p->nodes[i] || !p->locks[i])
  68                         continue;
  69                 /*
  70                  * If we currently have a spinning reader or writer lock this
  71                  * will bump the count of blocking holders and drop the
  72                  * spinlock.
  73                  */
  74                 if (p->locks[i] == BTRFS_READ_LOCK) {
  75                         btrfs_set_lock_blocking_read(p->nodes[i]);
  76                         p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
  77                 } else if (p->locks[i] == BTRFS_WRITE_LOCK) {
  78                         btrfs_set_lock_blocking_write(p->nodes[i]);
  79                         p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
  80                 }
  81         }
  82 }
  83 
  84 /* this also releases the path */
  85 void btrfs_free_path(struct btrfs_path *p)
  86 {
  87         if (!p)
  88                 return;
  89         btrfs_release_path(p);
  90         kmem_cache_free(btrfs_path_cachep, p);
  91 }
  92 
  93 /*
  94  * path release drops references on the extent buffers in the path
  95  * and it drops any locks held by this path
  96  *
  97  * It is safe to call this on paths that no locks or extent buffers held.
  98  */
  99 noinline void btrfs_release_path(struct btrfs_path *p)
 100 {
 101         int i;
 102 
 103         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
 104                 p->slots[i] = 0;
 105                 if (!p->nodes[i])
 106                         continue;
 107                 if (p->locks[i]) {
 108                         btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
 109                         p->locks[i] = 0;
 110                 }
 111                 free_extent_buffer(p->nodes[i]);
 112                 p->nodes[i] = NULL;
 113         }
 114 }
 115 
 116 /*
 117  * safely gets a reference on the root node of a tree.  A lock
 118  * is not taken, so a concurrent writer may put a different node
 119  * at the root of the tree.  See btrfs_lock_root_node for the
 120  * looping required.
 121  *
 122  * The extent buffer returned by this has a reference taken, so
 123  * it won't disappear.  It may stop being the root of the tree
 124  * at any time because there are no locks held.
 125  */
 126 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
 127 {
 128         struct extent_buffer *eb;
 129 
 130         while (1) {
 131                 rcu_read_lock();
 132                 eb = rcu_dereference(root->node);
 133 
 134                 /*
 135                  * RCU really hurts here, we could free up the root node because
 136                  * it was COWed but we may not get the new root node yet so do
 137                  * the inc_not_zero dance and if it doesn't work then
 138                  * synchronize_rcu and try again.
 139                  */
 140                 if (atomic_inc_not_zero(&eb->refs)) {
 141                         rcu_read_unlock();
 142                         break;
 143                 }
 144                 rcu_read_unlock();
 145                 synchronize_rcu();
 146         }
 147         return eb;
 148 }
 149 
 150 /* loop around taking references on and locking the root node of the
 151  * tree until you end up with a lock on the root.  A locked buffer
 152  * is returned, with a reference held.
 153  */
 154 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
 155 {
 156         struct extent_buffer *eb;
 157 
 158         while (1) {
 159                 eb = btrfs_root_node(root);
 160                 btrfs_tree_lock(eb);
 161                 if (eb == root->node)
 162                         break;
 163                 btrfs_tree_unlock(eb);
 164                 free_extent_buffer(eb);
 165         }
 166         return eb;
 167 }
 168 
 169 /* loop around taking references on and locking the root node of the
 170  * tree until you end up with a lock on the root.  A locked buffer
 171  * is returned, with a reference held.
 172  */
 173 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
 174 {
 175         struct extent_buffer *eb;
 176 
 177         while (1) {
 178                 eb = btrfs_root_node(root);
 179                 btrfs_tree_read_lock(eb);
 180                 if (eb == root->node)
 181                         break;
 182                 btrfs_tree_read_unlock(eb);
 183                 free_extent_buffer(eb);
 184         }
 185         return eb;
 186 }
 187 
 188 /* cowonly root (everything not a reference counted cow subvolume), just get
 189  * put onto a simple dirty list.  transaction.c walks this to make sure they
 190  * get properly updated on disk.
 191  */
 192 static void add_root_to_dirty_list(struct btrfs_root *root)
 193 {
 194         struct btrfs_fs_info *fs_info = root->fs_info;
 195 
 196         if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
 197             !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
 198                 return;
 199 
 200         spin_lock(&fs_info->trans_lock);
 201         if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
 202                 /* Want the extent tree to be the last on the list */
 203                 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
 204                         list_move_tail(&root->dirty_list,
 205                                        &fs_info->dirty_cowonly_roots);
 206                 else
 207                         list_move(&root->dirty_list,
 208                                   &fs_info->dirty_cowonly_roots);
 209         }
 210         spin_unlock(&fs_info->trans_lock);
 211 }
 212 
 213 /*
 214  * used by snapshot creation to make a copy of a root for a tree with
 215  * a given objectid.  The buffer with the new root node is returned in
 216  * cow_ret, and this func returns zero on success or a negative error code.
 217  */
 218 int btrfs_copy_root(struct btrfs_trans_handle *trans,
 219                       struct btrfs_root *root,
 220                       struct extent_buffer *buf,
 221                       struct extent_buffer **cow_ret, u64 new_root_objectid)
 222 {
 223         struct btrfs_fs_info *fs_info = root->fs_info;
 224         struct extent_buffer *cow;
 225         int ret = 0;
 226         int level;
 227         struct btrfs_disk_key disk_key;
 228 
 229         WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
 230                 trans->transid != fs_info->running_transaction->transid);
 231         WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
 232                 trans->transid != root->last_trans);
 233 
 234         level = btrfs_header_level(buf);
 235         if (level == 0)
 236                 btrfs_item_key(buf, &disk_key, 0);
 237         else
 238                 btrfs_node_key(buf, &disk_key, 0);
 239 
 240         cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
 241                         &disk_key, level, buf->start, 0);
 242         if (IS_ERR(cow))
 243                 return PTR_ERR(cow);
 244 
 245         copy_extent_buffer_full(cow, buf);
 246         btrfs_set_header_bytenr(cow, cow->start);
 247         btrfs_set_header_generation(cow, trans->transid);
 248         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
 249         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
 250                                      BTRFS_HEADER_FLAG_RELOC);
 251         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
 252                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
 253         else
 254                 btrfs_set_header_owner(cow, new_root_objectid);
 255 
 256         write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
 257 
 258         WARN_ON(btrfs_header_generation(buf) > trans->transid);
 259         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
 260                 ret = btrfs_inc_ref(trans, root, cow, 1);
 261         else
 262                 ret = btrfs_inc_ref(trans, root, cow, 0);
 263 
 264         if (ret)
 265                 return ret;
 266 
 267         btrfs_mark_buffer_dirty(cow);
 268         *cow_ret = cow;
 269         return 0;
 270 }
 271 
 272 enum mod_log_op {
 273         MOD_LOG_KEY_REPLACE,
 274         MOD_LOG_KEY_ADD,
 275         MOD_LOG_KEY_REMOVE,
 276         MOD_LOG_KEY_REMOVE_WHILE_FREEING,
 277         MOD_LOG_KEY_REMOVE_WHILE_MOVING,
 278         MOD_LOG_MOVE_KEYS,
 279         MOD_LOG_ROOT_REPLACE,
 280 };
 281 
 282 struct tree_mod_root {
 283         u64 logical;
 284         u8 level;
 285 };
 286 
 287 struct tree_mod_elem {
 288         struct rb_node node;
 289         u64 logical;
 290         u64 seq;
 291         enum mod_log_op op;
 292 
 293         /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
 294         int slot;
 295 
 296         /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
 297         u64 generation;
 298 
 299         /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
 300         struct btrfs_disk_key key;
 301         u64 blockptr;
 302 
 303         /* this is used for op == MOD_LOG_MOVE_KEYS */
 304         struct {
 305                 int dst_slot;
 306                 int nr_items;
 307         } move;
 308 
 309         /* this is used for op == MOD_LOG_ROOT_REPLACE */
 310         struct tree_mod_root old_root;
 311 };
 312 
 313 /*
 314  * Pull a new tree mod seq number for our operation.
 315  */
 316 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
 317 {
 318         return atomic64_inc_return(&fs_info->tree_mod_seq);
 319 }
 320 
 321 /*
 322  * This adds a new blocker to the tree mod log's blocker list if the @elem
 323  * passed does not already have a sequence number set. So when a caller expects
 324  * to record tree modifications, it should ensure to set elem->seq to zero
 325  * before calling btrfs_get_tree_mod_seq.
 326  * Returns a fresh, unused tree log modification sequence number, even if no new
 327  * blocker was added.
 328  */
 329 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
 330                            struct seq_list *elem)
 331 {
 332         write_lock(&fs_info->tree_mod_log_lock);
 333         if (!elem->seq) {
 334                 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
 335                 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
 336         }
 337         write_unlock(&fs_info->tree_mod_log_lock);
 338 
 339         return elem->seq;
 340 }
 341 
 342 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
 343                             struct seq_list *elem)
 344 {
 345         struct rb_root *tm_root;
 346         struct rb_node *node;
 347         struct rb_node *next;
 348         struct seq_list *cur_elem;
 349         struct tree_mod_elem *tm;
 350         u64 min_seq = (u64)-1;
 351         u64 seq_putting = elem->seq;
 352 
 353         if (!seq_putting)
 354                 return;
 355 
 356         write_lock(&fs_info->tree_mod_log_lock);
 357         list_del(&elem->list);
 358         elem->seq = 0;
 359 
 360         list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
 361                 if (cur_elem->seq < min_seq) {
 362                         if (seq_putting > cur_elem->seq) {
 363                                 /*
 364                                  * blocker with lower sequence number exists, we
 365                                  * cannot remove anything from the log
 366                                  */
 367                                 write_unlock(&fs_info->tree_mod_log_lock);
 368                                 return;
 369                         }
 370                         min_seq = cur_elem->seq;
 371                 }
 372         }
 373 
 374         /*
 375          * anything that's lower than the lowest existing (read: blocked)
 376          * sequence number can be removed from the tree.
 377          */
 378         tm_root = &fs_info->tree_mod_log;
 379         for (node = rb_first(tm_root); node; node = next) {
 380                 next = rb_next(node);
 381                 tm = rb_entry(node, struct tree_mod_elem, node);
 382                 if (tm->seq >= min_seq)
 383                         continue;
 384                 rb_erase(node, tm_root);
 385                 kfree(tm);
 386         }
 387         write_unlock(&fs_info->tree_mod_log_lock);
 388 }
 389 
 390 /*
 391  * key order of the log:
 392  *       node/leaf start address -> sequence
 393  *
 394  * The 'start address' is the logical address of the *new* root node
 395  * for root replace operations, or the logical address of the affected
 396  * block for all other operations.
 397  */
 398 static noinline int
 399 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
 400 {
 401         struct rb_root *tm_root;
 402         struct rb_node **new;
 403         struct rb_node *parent = NULL;
 404         struct tree_mod_elem *cur;
 405 
 406         lockdep_assert_held_write(&fs_info->tree_mod_log_lock);
 407 
 408         tm->seq = btrfs_inc_tree_mod_seq(fs_info);
 409 
 410         tm_root = &fs_info->tree_mod_log;
 411         new = &tm_root->rb_node;
 412         while (*new) {
 413                 cur = rb_entry(*new, struct tree_mod_elem, node);
 414                 parent = *new;
 415                 if (cur->logical < tm->logical)
 416                         new = &((*new)->rb_left);
 417                 else if (cur->logical > tm->logical)
 418                         new = &((*new)->rb_right);
 419                 else if (cur->seq < tm->seq)
 420                         new = &((*new)->rb_left);
 421                 else if (cur->seq > tm->seq)
 422                         new = &((*new)->rb_right);
 423                 else
 424                         return -EEXIST;
 425         }
 426 
 427         rb_link_node(&tm->node, parent, new);
 428         rb_insert_color(&tm->node, tm_root);
 429         return 0;
 430 }
 431 
 432 /*
 433  * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
 434  * returns zero with the tree_mod_log_lock acquired. The caller must hold
 435  * this until all tree mod log insertions are recorded in the rb tree and then
 436  * write unlock fs_info::tree_mod_log_lock.
 437  */
 438 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
 439                                     struct extent_buffer *eb) {
 440         smp_mb();
 441         if (list_empty(&(fs_info)->tree_mod_seq_list))
 442                 return 1;
 443         if (eb && btrfs_header_level(eb) == 0)
 444                 return 1;
 445 
 446         write_lock(&fs_info->tree_mod_log_lock);
 447         if (list_empty(&(fs_info)->tree_mod_seq_list)) {
 448                 write_unlock(&fs_info->tree_mod_log_lock);
 449                 return 1;
 450         }
 451 
 452         return 0;
 453 }
 454 
 455 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
 456 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
 457                                     struct extent_buffer *eb)
 458 {
 459         smp_mb();
 460         if (list_empty(&(fs_info)->tree_mod_seq_list))
 461                 return 0;
 462         if (eb && btrfs_header_level(eb) == 0)
 463                 return 0;
 464 
 465         return 1;
 466 }
 467 
 468 static struct tree_mod_elem *
 469 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
 470                     enum mod_log_op op, gfp_t flags)
 471 {
 472         struct tree_mod_elem *tm;
 473 
 474         tm = kzalloc(sizeof(*tm), flags);
 475         if (!tm)
 476                 return NULL;
 477 
 478         tm->logical = eb->start;
 479         if (op != MOD_LOG_KEY_ADD) {
 480                 btrfs_node_key(eb, &tm->key, slot);
 481                 tm->blockptr = btrfs_node_blockptr(eb, slot);
 482         }
 483         tm->op = op;
 484         tm->slot = slot;
 485         tm->generation = btrfs_node_ptr_generation(eb, slot);
 486         RB_CLEAR_NODE(&tm->node);
 487 
 488         return tm;
 489 }
 490 
 491 static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot,
 492                 enum mod_log_op op, gfp_t flags)
 493 {
 494         struct tree_mod_elem *tm;
 495         int ret;
 496 
 497         if (!tree_mod_need_log(eb->fs_info, eb))
 498                 return 0;
 499 
 500         tm = alloc_tree_mod_elem(eb, slot, op, flags);
 501         if (!tm)
 502                 return -ENOMEM;
 503 
 504         if (tree_mod_dont_log(eb->fs_info, eb)) {
 505                 kfree(tm);
 506                 return 0;
 507         }
 508 
 509         ret = __tree_mod_log_insert(eb->fs_info, tm);
 510         write_unlock(&eb->fs_info->tree_mod_log_lock);
 511         if (ret)
 512                 kfree(tm);
 513 
 514         return ret;
 515 }
 516 
 517 static noinline int tree_mod_log_insert_move(struct extent_buffer *eb,
 518                 int dst_slot, int src_slot, int nr_items)
 519 {
 520         struct tree_mod_elem *tm = NULL;
 521         struct tree_mod_elem **tm_list = NULL;
 522         int ret = 0;
 523         int i;
 524         int locked = 0;
 525 
 526         if (!tree_mod_need_log(eb->fs_info, eb))
 527                 return 0;
 528 
 529         tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
 530         if (!tm_list)
 531                 return -ENOMEM;
 532 
 533         tm = kzalloc(sizeof(*tm), GFP_NOFS);
 534         if (!tm) {
 535                 ret = -ENOMEM;
 536                 goto free_tms;
 537         }
 538 
 539         tm->logical = eb->start;
 540         tm->slot = src_slot;
 541         tm->move.dst_slot = dst_slot;
 542         tm->move.nr_items = nr_items;
 543         tm->op = MOD_LOG_MOVE_KEYS;
 544 
 545         for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
 546                 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
 547                     MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
 548                 if (!tm_list[i]) {
 549                         ret = -ENOMEM;
 550                         goto free_tms;
 551                 }
 552         }
 553 
 554         if (tree_mod_dont_log(eb->fs_info, eb))
 555                 goto free_tms;
 556         locked = 1;
 557 
 558         /*
 559          * When we override something during the move, we log these removals.
 560          * This can only happen when we move towards the beginning of the
 561          * buffer, i.e. dst_slot < src_slot.
 562          */
 563         for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
 564                 ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]);
 565                 if (ret)
 566                         goto free_tms;
 567         }
 568 
 569         ret = __tree_mod_log_insert(eb->fs_info, tm);
 570         if (ret)
 571                 goto free_tms;
 572         write_unlock(&eb->fs_info->tree_mod_log_lock);
 573         kfree(tm_list);
 574 
 575         return 0;
 576 free_tms:
 577         for (i = 0; i < nr_items; i++) {
 578                 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
 579                         rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
 580                 kfree(tm_list[i]);
 581         }
 582         if (locked)
 583                 write_unlock(&eb->fs_info->tree_mod_log_lock);
 584         kfree(tm_list);
 585         kfree(tm);
 586 
 587         return ret;
 588 }
 589 
 590 static inline int
 591 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
 592                        struct tree_mod_elem **tm_list,
 593                        int nritems)
 594 {
 595         int i, j;
 596         int ret;
 597 
 598         for (i = nritems - 1; i >= 0; i--) {
 599                 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
 600                 if (ret) {
 601                         for (j = nritems - 1; j > i; j--)
 602                                 rb_erase(&tm_list[j]->node,
 603                                          &fs_info->tree_mod_log);
 604                         return ret;
 605                 }
 606         }
 607 
 608         return 0;
 609 }
 610 
 611 static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root,
 612                          struct extent_buffer *new_root, int log_removal)
 613 {
 614         struct btrfs_fs_info *fs_info = old_root->fs_info;
 615         struct tree_mod_elem *tm = NULL;
 616         struct tree_mod_elem **tm_list = NULL;
 617         int nritems = 0;
 618         int ret = 0;
 619         int i;
 620 
 621         if (!tree_mod_need_log(fs_info, NULL))
 622                 return 0;
 623 
 624         if (log_removal && btrfs_header_level(old_root) > 0) {
 625                 nritems = btrfs_header_nritems(old_root);
 626                 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
 627                                   GFP_NOFS);
 628                 if (!tm_list) {
 629                         ret = -ENOMEM;
 630                         goto free_tms;
 631                 }
 632                 for (i = 0; i < nritems; i++) {
 633                         tm_list[i] = alloc_tree_mod_elem(old_root, i,
 634                             MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
 635                         if (!tm_list[i]) {
 636                                 ret = -ENOMEM;
 637                                 goto free_tms;
 638                         }
 639                 }
 640         }
 641 
 642         tm = kzalloc(sizeof(*tm), GFP_NOFS);
 643         if (!tm) {
 644                 ret = -ENOMEM;
 645                 goto free_tms;
 646         }
 647 
 648         tm->logical = new_root->start;
 649         tm->old_root.logical = old_root->start;
 650         tm->old_root.level = btrfs_header_level(old_root);
 651         tm->generation = btrfs_header_generation(old_root);
 652         tm->op = MOD_LOG_ROOT_REPLACE;
 653 
 654         if (tree_mod_dont_log(fs_info, NULL))
 655                 goto free_tms;
 656 
 657         if (tm_list)
 658                 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
 659         if (!ret)
 660                 ret = __tree_mod_log_insert(fs_info, tm);
 661 
 662         write_unlock(&fs_info->tree_mod_log_lock);
 663         if (ret)
 664                 goto free_tms;
 665         kfree(tm_list);
 666 
 667         return ret;
 668 
 669 free_tms:
 670         if (tm_list) {
 671                 for (i = 0; i < nritems; i++)
 672                         kfree(tm_list[i]);
 673                 kfree(tm_list);
 674         }
 675         kfree(tm);
 676 
 677         return ret;
 678 }
 679 
 680 static struct tree_mod_elem *
 681 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
 682                       int smallest)
 683 {
 684         struct rb_root *tm_root;
 685         struct rb_node *node;
 686         struct tree_mod_elem *cur = NULL;
 687         struct tree_mod_elem *found = NULL;
 688 
 689         read_lock(&fs_info->tree_mod_log_lock);
 690         tm_root = &fs_info->tree_mod_log;
 691         node = tm_root->rb_node;
 692         while (node) {
 693                 cur = rb_entry(node, struct tree_mod_elem, node);
 694                 if (cur->logical < start) {
 695                         node = node->rb_left;
 696                 } else if (cur->logical > start) {
 697                         node = node->rb_right;
 698                 } else if (cur->seq < min_seq) {
 699                         node = node->rb_left;
 700                 } else if (!smallest) {
 701                         /* we want the node with the highest seq */
 702                         if (found)
 703                                 BUG_ON(found->seq > cur->seq);
 704                         found = cur;
 705                         node = node->rb_left;
 706                 } else if (cur->seq > min_seq) {
 707                         /* we want the node with the smallest seq */
 708                         if (found)
 709                                 BUG_ON(found->seq < cur->seq);
 710                         found = cur;
 711                         node = node->rb_right;
 712                 } else {
 713                         found = cur;
 714                         break;
 715                 }
 716         }
 717         read_unlock(&fs_info->tree_mod_log_lock);
 718 
 719         return found;
 720 }
 721 
 722 /*
 723  * this returns the element from the log with the smallest time sequence
 724  * value that's in the log (the oldest log item). any element with a time
 725  * sequence lower than min_seq will be ignored.
 726  */
 727 static struct tree_mod_elem *
 728 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
 729                            u64 min_seq)
 730 {
 731         return __tree_mod_log_search(fs_info, start, min_seq, 1);
 732 }
 733 
 734 /*
 735  * this returns the element from the log with the largest time sequence
 736  * value that's in the log (the most recent log item). any element with
 737  * a time sequence lower than min_seq will be ignored.
 738  */
 739 static struct tree_mod_elem *
 740 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
 741 {
 742         return __tree_mod_log_search(fs_info, start, min_seq, 0);
 743 }
 744 
 745 static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst,
 746                      struct extent_buffer *src, unsigned long dst_offset,
 747                      unsigned long src_offset, int nr_items)
 748 {
 749         struct btrfs_fs_info *fs_info = dst->fs_info;
 750         int ret = 0;
 751         struct tree_mod_elem **tm_list = NULL;
 752         struct tree_mod_elem **tm_list_add, **tm_list_rem;
 753         int i;
 754         int locked = 0;
 755 
 756         if (!tree_mod_need_log(fs_info, NULL))
 757                 return 0;
 758 
 759         if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
 760                 return 0;
 761 
 762         tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
 763                           GFP_NOFS);
 764         if (!tm_list)
 765                 return -ENOMEM;
 766 
 767         tm_list_add = tm_list;
 768         tm_list_rem = tm_list + nr_items;
 769         for (i = 0; i < nr_items; i++) {
 770                 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
 771                     MOD_LOG_KEY_REMOVE, GFP_NOFS);
 772                 if (!tm_list_rem[i]) {
 773                         ret = -ENOMEM;
 774                         goto free_tms;
 775                 }
 776 
 777                 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
 778                     MOD_LOG_KEY_ADD, GFP_NOFS);
 779                 if (!tm_list_add[i]) {
 780                         ret = -ENOMEM;
 781                         goto free_tms;
 782                 }
 783         }
 784 
 785         if (tree_mod_dont_log(fs_info, NULL))
 786                 goto free_tms;
 787         locked = 1;
 788 
 789         for (i = 0; i < nr_items; i++) {
 790                 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
 791                 if (ret)
 792                         goto free_tms;
 793                 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
 794                 if (ret)
 795                         goto free_tms;
 796         }
 797 
 798         write_unlock(&fs_info->tree_mod_log_lock);
 799         kfree(tm_list);
 800 
 801         return 0;
 802 
 803 free_tms:
 804         for (i = 0; i < nr_items * 2; i++) {
 805                 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
 806                         rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
 807                 kfree(tm_list[i]);
 808         }
 809         if (locked)
 810                 write_unlock(&fs_info->tree_mod_log_lock);
 811         kfree(tm_list);
 812 
 813         return ret;
 814 }
 815 
 816 static noinline int tree_mod_log_free_eb(struct extent_buffer *eb)
 817 {
 818         struct tree_mod_elem **tm_list = NULL;
 819         int nritems = 0;
 820         int i;
 821         int ret = 0;
 822 
 823         if (btrfs_header_level(eb) == 0)
 824                 return 0;
 825 
 826         if (!tree_mod_need_log(eb->fs_info, NULL))
 827                 return 0;
 828 
 829         nritems = btrfs_header_nritems(eb);
 830         tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
 831         if (!tm_list)
 832                 return -ENOMEM;
 833 
 834         for (i = 0; i < nritems; i++) {
 835                 tm_list[i] = alloc_tree_mod_elem(eb, i,
 836                     MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
 837                 if (!tm_list[i]) {
 838                         ret = -ENOMEM;
 839                         goto free_tms;
 840                 }
 841         }
 842 
 843         if (tree_mod_dont_log(eb->fs_info, eb))
 844                 goto free_tms;
 845 
 846         ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
 847         write_unlock(&eb->fs_info->tree_mod_log_lock);
 848         if (ret)
 849                 goto free_tms;
 850         kfree(tm_list);
 851 
 852         return 0;
 853 
 854 free_tms:
 855         for (i = 0; i < nritems; i++)
 856                 kfree(tm_list[i]);
 857         kfree(tm_list);
 858 
 859         return ret;
 860 }
 861 
 862 /*
 863  * check if the tree block can be shared by multiple trees
 864  */
 865 int btrfs_block_can_be_shared(struct btrfs_root *root,
 866                               struct extent_buffer *buf)
 867 {
 868         /*
 869          * Tree blocks not in reference counted trees and tree roots
 870          * are never shared. If a block was allocated after the last
 871          * snapshot and the block was not allocated by tree relocation,
 872          * we know the block is not shared.
 873          */
 874         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
 875             buf != root->node && buf != root->commit_root &&
 876             (btrfs_header_generation(buf) <=
 877              btrfs_root_last_snapshot(&root->root_item) ||
 878              btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
 879                 return 1;
 880 
 881         return 0;
 882 }
 883 
 884 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 885                                        struct btrfs_root *root,
 886                                        struct extent_buffer *buf,
 887                                        struct extent_buffer *cow,
 888                                        int *last_ref)
 889 {
 890         struct btrfs_fs_info *fs_info = root->fs_info;
 891         u64 refs;
 892         u64 owner;
 893         u64 flags;
 894         u64 new_flags = 0;
 895         int ret;
 896 
 897         /*
 898          * Backrefs update rules:
 899          *
 900          * Always use full backrefs for extent pointers in tree block
 901          * allocated by tree relocation.
 902          *
 903          * If a shared tree block is no longer referenced by its owner
 904          * tree (btrfs_header_owner(buf) == root->root_key.objectid),
 905          * use full backrefs for extent pointers in tree block.
 906          *
 907          * If a tree block is been relocating
 908          * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
 909          * use full backrefs for extent pointers in tree block.
 910          * The reason for this is some operations (such as drop tree)
 911          * are only allowed for blocks use full backrefs.
 912          */
 913 
 914         if (btrfs_block_can_be_shared(root, buf)) {
 915                 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
 916                                                btrfs_header_level(buf), 1,
 917                                                &refs, &flags);
 918                 if (ret)
 919                         return ret;
 920                 if (refs == 0) {
 921                         ret = -EROFS;
 922                         btrfs_handle_fs_error(fs_info, ret, NULL);
 923                         return ret;
 924                 }
 925         } else {
 926                 refs = 1;
 927                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
 928                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
 929                         flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 930                 else
 931                         flags = 0;
 932         }
 933 
 934         owner = btrfs_header_owner(buf);
 935         BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
 936                !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
 937 
 938         if (refs > 1) {
 939                 if ((owner == root->root_key.objectid ||
 940                      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
 941                     !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
 942                         ret = btrfs_inc_ref(trans, root, buf, 1);
 943                         if (ret)
 944                                 return ret;
 945 
 946                         if (root->root_key.objectid ==
 947                             BTRFS_TREE_RELOC_OBJECTID) {
 948                                 ret = btrfs_dec_ref(trans, root, buf, 0);
 949                                 if (ret)
 950                                         return ret;
 951                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 952                                 if (ret)
 953                                         return ret;
 954                         }
 955                         new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
 956                 } else {
 957 
 958                         if (root->root_key.objectid ==
 959                             BTRFS_TREE_RELOC_OBJECTID)
 960                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 961                         else
 962                                 ret = btrfs_inc_ref(trans, root, cow, 0);
 963                         if (ret)
 964                                 return ret;
 965                 }
 966                 if (new_flags != 0) {
 967                         int level = btrfs_header_level(buf);
 968 
 969                         ret = btrfs_set_disk_extent_flags(trans,
 970                                                           buf->start,
 971                                                           buf->len,
 972                                                           new_flags, level, 0);
 973                         if (ret)
 974                                 return ret;
 975                 }
 976         } else {
 977                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
 978                         if (root->root_key.objectid ==
 979                             BTRFS_TREE_RELOC_OBJECTID)
 980                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 981                         else
 982                                 ret = btrfs_inc_ref(trans, root, cow, 0);
 983                         if (ret)
 984                                 return ret;
 985                         ret = btrfs_dec_ref(trans, root, buf, 1);
 986                         if (ret)
 987                                 return ret;
 988                 }
 989                 btrfs_clean_tree_block(buf);
 990                 *last_ref = 1;
 991         }
 992         return 0;
 993 }
 994 
 995 static struct extent_buffer *alloc_tree_block_no_bg_flush(
 996                                           struct btrfs_trans_handle *trans,
 997                                           struct btrfs_root *root,
 998                                           u64 parent_start,
 999                                           const struct btrfs_disk_key *disk_key,
1000                                           int level,
1001                                           u64 hint,
1002                                           u64 empty_size)
1003 {
1004         struct btrfs_fs_info *fs_info = root->fs_info;
1005         struct extent_buffer *ret;
1006 
1007         /*
1008          * If we are COWing a node/leaf from the extent, chunk, device or free
1009          * space trees, make sure that we do not finish block group creation of
1010          * pending block groups. We do this to avoid a deadlock.
1011          * COWing can result in allocation of a new chunk, and flushing pending
1012          * block groups (btrfs_create_pending_block_groups()) can be triggered
1013          * when finishing allocation of a new chunk. Creation of a pending block
1014          * group modifies the extent, chunk, device and free space trees,
1015          * therefore we could deadlock with ourselves since we are holding a
1016          * lock on an extent buffer that btrfs_create_pending_block_groups() may
1017          * try to COW later.
1018          * For similar reasons, we also need to delay flushing pending block
1019          * groups when splitting a leaf or node, from one of those trees, since
1020          * we are holding a write lock on it and its parent or when inserting a
1021          * new root node for one of those trees.
1022          */
1023         if (root == fs_info->extent_root ||
1024             root == fs_info->chunk_root ||
1025             root == fs_info->dev_root ||
1026             root == fs_info->free_space_root)
1027                 trans->can_flush_pending_bgs = false;
1028 
1029         ret = btrfs_alloc_tree_block(trans, root, parent_start,
1030                                      root->root_key.objectid, disk_key, level,
1031                                      hint, empty_size);
1032         trans->can_flush_pending_bgs = true;
1033 
1034         return ret;
1035 }
1036 
1037 /*
1038  * does the dirty work in cow of a single block.  The parent block (if
1039  * supplied) is updated to point to the new cow copy.  The new buffer is marked
1040  * dirty and returned locked.  If you modify the block it needs to be marked
1041  * dirty again.
1042  *
1043  * search_start -- an allocation hint for the new block
1044  *
1045  * empty_size -- a hint that you plan on doing more cow.  This is the size in
1046  * bytes the allocator should try to find free next to the block it returns.
1047  * This is just a hint and may be ignored by the allocator.
1048  */
1049 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1050                              struct btrfs_root *root,
1051                              struct extent_buffer *buf,
1052                              struct extent_buffer *parent, int parent_slot,
1053                              struct extent_buffer **cow_ret,
1054                              u64 search_start, u64 empty_size)
1055 {
1056         struct btrfs_fs_info *fs_info = root->fs_info;
1057         struct btrfs_disk_key disk_key;
1058         struct extent_buffer *cow;
1059         int level, ret;
1060         int last_ref = 0;
1061         int unlock_orig = 0;
1062         u64 parent_start = 0;
1063 
1064         if (*cow_ret == buf)
1065                 unlock_orig = 1;
1066 
1067         btrfs_assert_tree_locked(buf);
1068 
1069         WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1070                 trans->transid != fs_info->running_transaction->transid);
1071         WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1072                 trans->transid != root->last_trans);
1073 
1074         level = btrfs_header_level(buf);
1075 
1076         if (level == 0)
1077                 btrfs_item_key(buf, &disk_key, 0);
1078         else
1079                 btrfs_node_key(buf, &disk_key, 0);
1080 
1081         if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1082                 parent_start = parent->start;
1083 
1084         cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key,
1085                                            level, search_start, empty_size);
1086         if (IS_ERR(cow))
1087                 return PTR_ERR(cow);
1088 
1089         /* cow is set to blocking by btrfs_init_new_buffer */
1090 
1091         copy_extent_buffer_full(cow, buf);
1092         btrfs_set_header_bytenr(cow, cow->start);
1093         btrfs_set_header_generation(cow, trans->transid);
1094         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1095         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1096                                      BTRFS_HEADER_FLAG_RELOC);
1097         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1098                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1099         else
1100                 btrfs_set_header_owner(cow, root->root_key.objectid);
1101 
1102         write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
1103 
1104         ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1105         if (ret) {
1106                 btrfs_abort_transaction(trans, ret);
1107                 return ret;
1108         }
1109 
1110         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1111                 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1112                 if (ret) {
1113                         btrfs_abort_transaction(trans, ret);
1114                         return ret;
1115                 }
1116         }
1117 
1118         if (buf == root->node) {
1119                 WARN_ON(parent && parent != buf);
1120                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1121                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1122                         parent_start = buf->start;
1123 
1124                 extent_buffer_get(cow);
1125                 ret = tree_mod_log_insert_root(root->node, cow, 1);
1126                 BUG_ON(ret < 0);
1127                 rcu_assign_pointer(root->node, cow);
1128 
1129                 btrfs_free_tree_block(trans, root, buf, parent_start,
1130                                       last_ref);
1131                 free_extent_buffer(buf);
1132                 add_root_to_dirty_list(root);
1133         } else {
1134                 WARN_ON(trans->transid != btrfs_header_generation(parent));
1135                 tree_mod_log_insert_key(parent, parent_slot,
1136                                         MOD_LOG_KEY_REPLACE, GFP_NOFS);
1137                 btrfs_set_node_blockptr(parent, parent_slot,
1138                                         cow->start);
1139                 btrfs_set_node_ptr_generation(parent, parent_slot,
1140                                               trans->transid);
1141                 btrfs_mark_buffer_dirty(parent);
1142                 if (last_ref) {
1143                         ret = tree_mod_log_free_eb(buf);
1144                         if (ret) {
1145                                 btrfs_abort_transaction(trans, ret);
1146                                 return ret;
1147                         }
1148                 }
1149                 btrfs_free_tree_block(trans, root, buf, parent_start,
1150                                       last_ref);
1151         }
1152         if (unlock_orig)
1153                 btrfs_tree_unlock(buf);
1154         free_extent_buffer_stale(buf);
1155         btrfs_mark_buffer_dirty(cow);
1156         *cow_ret = cow;
1157         return 0;
1158 }
1159 
1160 /*
1161  * returns the logical address of the oldest predecessor of the given root.
1162  * entries older than time_seq are ignored.
1163  */
1164 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1165                 struct extent_buffer *eb_root, u64 time_seq)
1166 {
1167         struct tree_mod_elem *tm;
1168         struct tree_mod_elem *found = NULL;
1169         u64 root_logical = eb_root->start;
1170         int looped = 0;
1171 
1172         if (!time_seq)
1173                 return NULL;
1174 
1175         /*
1176          * the very last operation that's logged for a root is the
1177          * replacement operation (if it is replaced at all). this has
1178          * the logical address of the *new* root, making it the very
1179          * first operation that's logged for this root.
1180          */
1181         while (1) {
1182                 tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1183                                                 time_seq);
1184                 if (!looped && !tm)
1185                         return NULL;
1186                 /*
1187                  * if there are no tree operation for the oldest root, we simply
1188                  * return it. this should only happen if that (old) root is at
1189                  * level 0.
1190                  */
1191                 if (!tm)
1192                         break;
1193 
1194                 /*
1195                  * if there's an operation that's not a root replacement, we
1196                  * found the oldest version of our root. normally, we'll find a
1197                  * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1198                  */
1199                 if (tm->op != MOD_LOG_ROOT_REPLACE)
1200                         break;
1201 
1202                 found = tm;
1203                 root_logical = tm->old_root.logical;
1204                 looped = 1;
1205         }
1206 
1207         /* if there's no old root to return, return what we found instead */
1208         if (!found)
1209                 found = tm;
1210 
1211         return found;
1212 }
1213 
1214 /*
1215  * tm is a pointer to the first operation to rewind within eb. then, all
1216  * previous operations will be rewound (until we reach something older than
1217  * time_seq).
1218  */
1219 static void
1220 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1221                       u64 time_seq, struct tree_mod_elem *first_tm)
1222 {
1223         u32 n;
1224         struct rb_node *next;
1225         struct tree_mod_elem *tm = first_tm;
1226         unsigned long o_dst;
1227         unsigned long o_src;
1228         unsigned long p_size = sizeof(struct btrfs_key_ptr);
1229 
1230         n = btrfs_header_nritems(eb);
1231         read_lock(&fs_info->tree_mod_log_lock);
1232         while (tm && tm->seq >= time_seq) {
1233                 /*
1234                  * all the operations are recorded with the operator used for
1235                  * the modification. as we're going backwards, we do the
1236                  * opposite of each operation here.
1237                  */
1238                 switch (tm->op) {
1239                 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1240                         BUG_ON(tm->slot < n);
1241                         /* Fallthrough */
1242                 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1243                 case MOD_LOG_KEY_REMOVE:
1244                         btrfs_set_node_key(eb, &tm->key, tm->slot);
1245                         btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1246                         btrfs_set_node_ptr_generation(eb, tm->slot,
1247                                                       tm->generation);
1248                         n++;
1249                         break;
1250                 case MOD_LOG_KEY_REPLACE:
1251                         BUG_ON(tm->slot >= n);
1252                         btrfs_set_node_key(eb, &tm->key, tm->slot);
1253                         btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1254                         btrfs_set_node_ptr_generation(eb, tm->slot,
1255                                                       tm->generation);
1256                         break;
1257                 case MOD_LOG_KEY_ADD:
1258                         /* if a move operation is needed it's in the log */
1259                         n--;
1260                         break;
1261                 case MOD_LOG_MOVE_KEYS:
1262                         o_dst = btrfs_node_key_ptr_offset(tm->slot);
1263                         o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1264                         memmove_extent_buffer(eb, o_dst, o_src,
1265                                               tm->move.nr_items * p_size);
1266                         break;
1267                 case MOD_LOG_ROOT_REPLACE:
1268                         /*
1269                          * this operation is special. for roots, this must be
1270                          * handled explicitly before rewinding.
1271                          * for non-roots, this operation may exist if the node
1272                          * was a root: root A -> child B; then A gets empty and
1273                          * B is promoted to the new root. in the mod log, we'll
1274                          * have a root-replace operation for B, a tree block
1275                          * that is no root. we simply ignore that operation.
1276                          */
1277                         break;
1278                 }
1279                 next = rb_next(&tm->node);
1280                 if (!next)
1281                         break;
1282                 tm = rb_entry(next, struct tree_mod_elem, node);
1283                 if (tm->logical != first_tm->logical)
1284                         break;
1285         }
1286         read_unlock(&fs_info->tree_mod_log_lock);
1287         btrfs_set_header_nritems(eb, n);
1288 }
1289 
1290 /*
1291  * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1292  * is returned. If rewind operations happen, a fresh buffer is returned. The
1293  * returned buffer is always read-locked. If the returned buffer is not the
1294  * input buffer, the lock on the input buffer is released and the input buffer
1295  * is freed (its refcount is decremented).
1296  */
1297 static struct extent_buffer *
1298 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1299                     struct extent_buffer *eb, u64 time_seq)
1300 {
1301         struct extent_buffer *eb_rewin;
1302         struct tree_mod_elem *tm;
1303 
1304         if (!time_seq)
1305                 return eb;
1306 
1307         if (btrfs_header_level(eb) == 0)
1308                 return eb;
1309 
1310         tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1311         if (!tm)
1312                 return eb;
1313 
1314         btrfs_set_path_blocking(path);
1315         btrfs_set_lock_blocking_read(eb);
1316 
1317         if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1318                 BUG_ON(tm->slot != 0);
1319                 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1320                 if (!eb_rewin) {
1321                         btrfs_tree_read_unlock_blocking(eb);
1322                         free_extent_buffer(eb);
1323                         return NULL;
1324                 }
1325                 btrfs_set_header_bytenr(eb_rewin, eb->start);
1326                 btrfs_set_header_backref_rev(eb_rewin,
1327                                              btrfs_header_backref_rev(eb));
1328                 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1329                 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1330         } else {
1331                 eb_rewin = btrfs_clone_extent_buffer(eb);
1332                 if (!eb_rewin) {
1333                         btrfs_tree_read_unlock_blocking(eb);
1334                         free_extent_buffer(eb);
1335                         return NULL;
1336                 }
1337         }
1338 
1339         btrfs_tree_read_unlock_blocking(eb);
1340         free_extent_buffer(eb);
1341 
1342         btrfs_tree_read_lock(eb_rewin);
1343         __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1344         WARN_ON(btrfs_header_nritems(eb_rewin) >
1345                 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1346 
1347         return eb_rewin;
1348 }
1349 
1350 /*
1351  * get_old_root() rewinds the state of @root's root node to the given @time_seq
1352  * value. If there are no changes, the current root->root_node is returned. If
1353  * anything changed in between, there's a fresh buffer allocated on which the
1354  * rewind operations are done. In any case, the returned buffer is read locked.
1355  * Returns NULL on error (with no locks held).
1356  */
1357 static inline struct extent_buffer *
1358 get_old_root(struct btrfs_root *root, u64 time_seq)
1359 {
1360         struct btrfs_fs_info *fs_info = root->fs_info;
1361         struct tree_mod_elem *tm;
1362         struct extent_buffer *eb = NULL;
1363         struct extent_buffer *eb_root;
1364         u64 eb_root_owner = 0;
1365         struct extent_buffer *old;
1366         struct tree_mod_root *old_root = NULL;
1367         u64 old_generation = 0;
1368         u64 logical;
1369         int level;
1370 
1371         eb_root = btrfs_read_lock_root_node(root);
1372         tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1373         if (!tm)
1374                 return eb_root;
1375 
1376         if (tm->op == MOD_LOG_ROOT_REPLACE) {
1377                 old_root = &tm->old_root;
1378                 old_generation = tm->generation;
1379                 logical = old_root->logical;
1380                 level = old_root->level;
1381         } else {
1382                 logical = eb_root->start;
1383                 level = btrfs_header_level(eb_root);
1384         }
1385 
1386         tm = tree_mod_log_search(fs_info, logical, time_seq);
1387         if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1388                 btrfs_tree_read_unlock(eb_root);
1389                 free_extent_buffer(eb_root);
1390                 old = read_tree_block(fs_info, logical, 0, level, NULL);
1391                 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1392                         if (!IS_ERR(old))
1393                                 free_extent_buffer(old);
1394                         btrfs_warn(fs_info,
1395                                    "failed to read tree block %llu from get_old_root",
1396                                    logical);
1397                 } else {
1398                         eb = btrfs_clone_extent_buffer(old);
1399                         free_extent_buffer(old);
1400                 }
1401         } else if (old_root) {
1402                 eb_root_owner = btrfs_header_owner(eb_root);
1403                 btrfs_tree_read_unlock(eb_root);
1404                 free_extent_buffer(eb_root);
1405                 eb = alloc_dummy_extent_buffer(fs_info, logical);
1406         } else {
1407                 btrfs_set_lock_blocking_read(eb_root);
1408                 eb = btrfs_clone_extent_buffer(eb_root);
1409                 btrfs_tree_read_unlock_blocking(eb_root);
1410                 free_extent_buffer(eb_root);
1411         }
1412 
1413         if (!eb)
1414                 return NULL;
1415         btrfs_tree_read_lock(eb);
1416         if (old_root) {
1417                 btrfs_set_header_bytenr(eb, eb->start);
1418                 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1419                 btrfs_set_header_owner(eb, eb_root_owner);
1420                 btrfs_set_header_level(eb, old_root->level);
1421                 btrfs_set_header_generation(eb, old_generation);
1422         }
1423         if (tm)
1424                 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1425         else
1426                 WARN_ON(btrfs_header_level(eb) != 0);
1427         WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1428 
1429         return eb;
1430 }
1431 
1432 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1433 {
1434         struct tree_mod_elem *tm;
1435         int level;
1436         struct extent_buffer *eb_root = btrfs_root_node(root);
1437 
1438         tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1439         if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1440                 level = tm->old_root.level;
1441         } else {
1442                 level = btrfs_header_level(eb_root);
1443         }
1444         free_extent_buffer(eb_root);
1445 
1446         return level;
1447 }
1448 
1449 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1450                                    struct btrfs_root *root,
1451                                    struct extent_buffer *buf)
1452 {
1453         if (btrfs_is_testing(root->fs_info))
1454                 return 0;
1455 
1456         /* Ensure we can see the FORCE_COW bit */
1457         smp_mb__before_atomic();
1458 
1459         /*
1460          * We do not need to cow a block if
1461          * 1) this block is not created or changed in this transaction;
1462          * 2) this block does not belong to TREE_RELOC tree;
1463          * 3) the root is not forced COW.
1464          *
1465          * What is forced COW:
1466          *    when we create snapshot during committing the transaction,
1467          *    after we've finished copying src root, we must COW the shared
1468          *    block to ensure the metadata consistency.
1469          */
1470         if (btrfs_header_generation(buf) == trans->transid &&
1471             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1472             !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1473               btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1474             !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1475                 return 0;
1476         return 1;
1477 }
1478 
1479 /*
1480  * cows a single block, see __btrfs_cow_block for the real work.
1481  * This version of it has extra checks so that a block isn't COWed more than
1482  * once per transaction, as long as it hasn't been written yet
1483  */
1484 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1485                     struct btrfs_root *root, struct extent_buffer *buf,
1486                     struct extent_buffer *parent, int parent_slot,
1487                     struct extent_buffer **cow_ret)
1488 {
1489         struct btrfs_fs_info *fs_info = root->fs_info;
1490         u64 search_start;
1491         int ret;
1492 
1493         if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1494                 btrfs_err(fs_info,
1495                         "COW'ing blocks on a fs root that's being dropped");
1496 
1497         if (trans->transaction != fs_info->running_transaction)
1498                 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1499                        trans->transid,
1500                        fs_info->running_transaction->transid);
1501 
1502         if (trans->transid != fs_info->generation)
1503                 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1504                        trans->transid, fs_info->generation);
1505 
1506         if (!should_cow_block(trans, root, buf)) {
1507                 trans->dirty = true;
1508                 *cow_ret = buf;
1509                 return 0;
1510         }
1511 
1512         search_start = buf->start & ~((u64)SZ_1G - 1);
1513 
1514         if (parent)
1515                 btrfs_set_lock_blocking_write(parent);
1516         btrfs_set_lock_blocking_write(buf);
1517 
1518         /*
1519          * Before CoWing this block for later modification, check if it's
1520          * the subtree root and do the delayed subtree trace if needed.
1521          *
1522          * Also We don't care about the error, as it's handled internally.
1523          */
1524         btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
1525         ret = __btrfs_cow_block(trans, root, buf, parent,
1526                                  parent_slot, cow_ret, search_start, 0);
1527 
1528         trace_btrfs_cow_block(root, buf, *cow_ret);
1529 
1530         return ret;
1531 }
1532 
1533 /*
1534  * helper function for defrag to decide if two blocks pointed to by a
1535  * node are actually close by
1536  */
1537 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1538 {
1539         if (blocknr < other && other - (blocknr + blocksize) < 32768)
1540                 return 1;
1541         if (blocknr > other && blocknr - (other + blocksize) < 32768)
1542                 return 1;
1543         return 0;
1544 }
1545 
1546 /*
1547  * compare two keys in a memcmp fashion
1548  */
1549 static int comp_keys(const struct btrfs_disk_key *disk,
1550                      const struct btrfs_key *k2)
1551 {
1552         struct btrfs_key k1;
1553 
1554         btrfs_disk_key_to_cpu(&k1, disk);
1555 
1556         return btrfs_comp_cpu_keys(&k1, k2);
1557 }
1558 
1559 /*
1560  * same as comp_keys only with two btrfs_key's
1561  */
1562 int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1563 {
1564         if (k1->objectid > k2->objectid)
1565                 return 1;
1566         if (k1->objectid < k2->objectid)
1567                 return -1;
1568         if (k1->type > k2->type)
1569                 return 1;
1570         if (k1->type < k2->type)
1571                 return -1;
1572         if (k1->offset > k2->offset)
1573                 return 1;
1574         if (k1->offset < k2->offset)
1575                 return -1;
1576         return 0;
1577 }
1578 
1579 /*
1580  * this is used by the defrag code to go through all the
1581  * leaves pointed to by a node and reallocate them so that
1582  * disk order is close to key order
1583  */
1584 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1585                        struct btrfs_root *root, struct extent_buffer *parent,
1586                        int start_slot, u64 *last_ret,
1587                        struct btrfs_key *progress)
1588 {
1589         struct btrfs_fs_info *fs_info = root->fs_info;
1590         struct extent_buffer *cur;
1591         u64 blocknr;
1592         u64 gen;
1593         u64 search_start = *last_ret;
1594         u64 last_block = 0;
1595         u64 other;
1596         u32 parent_nritems;
1597         int end_slot;
1598         int i;
1599         int err = 0;
1600         int parent_level;
1601         int uptodate;
1602         u32 blocksize;
1603         int progress_passed = 0;
1604         struct btrfs_disk_key disk_key;
1605 
1606         parent_level = btrfs_header_level(parent);
1607 
1608         WARN_ON(trans->transaction != fs_info->running_transaction);
1609         WARN_ON(trans->transid != fs_info->generation);
1610 
1611         parent_nritems = btrfs_header_nritems(parent);
1612         blocksize = fs_info->nodesize;
1613         end_slot = parent_nritems - 1;
1614 
1615         if (parent_nritems <= 1)
1616                 return 0;
1617 
1618         btrfs_set_lock_blocking_write(parent);
1619 
1620         for (i = start_slot; i <= end_slot; i++) {
1621                 struct btrfs_key first_key;
1622                 int close = 1;
1623 
1624                 btrfs_node_key(parent, &disk_key, i);
1625                 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1626                         continue;
1627 
1628                 progress_passed = 1;
1629                 blocknr = btrfs_node_blockptr(parent, i);
1630                 gen = btrfs_node_ptr_generation(parent, i);
1631                 btrfs_node_key_to_cpu(parent, &first_key, i);
1632                 if (last_block == 0)
1633                         last_block = blocknr;
1634 
1635                 if (i > 0) {
1636                         other = btrfs_node_blockptr(parent, i - 1);
1637                         close = close_blocks(blocknr, other, blocksize);
1638                 }
1639                 if (!close && i < end_slot) {
1640                         other = btrfs_node_blockptr(parent, i + 1);
1641                         close = close_blocks(blocknr, other, blocksize);
1642                 }
1643                 if (close) {
1644                         last_block = blocknr;
1645                         continue;
1646                 }
1647 
1648                 cur = find_extent_buffer(fs_info, blocknr);
1649                 if (cur)
1650                         uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1651                 else
1652                         uptodate = 0;
1653                 if (!cur || !uptodate) {
1654                         if (!cur) {
1655                                 cur = read_tree_block(fs_info, blocknr, gen,
1656                                                       parent_level - 1,
1657                                                       &first_key);
1658                                 if (IS_ERR(cur)) {
1659                                         return PTR_ERR(cur);
1660                                 } else if (!extent_buffer_uptodate(cur)) {
1661                                         free_extent_buffer(cur);
1662                                         return -EIO;
1663                                 }
1664                         } else if (!uptodate) {
1665                                 err = btrfs_read_buffer(cur, gen,
1666                                                 parent_level - 1,&first_key);
1667                                 if (err) {
1668                                         free_extent_buffer(cur);
1669                                         return err;
1670                                 }
1671                         }
1672                 }
1673                 if (search_start == 0)
1674                         search_start = last_block;
1675 
1676                 btrfs_tree_lock(cur);
1677                 btrfs_set_lock_blocking_write(cur);
1678                 err = __btrfs_cow_block(trans, root, cur, parent, i,
1679                                         &cur, search_start,
1680                                         min(16 * blocksize,
1681                                             (end_slot - i) * blocksize));
1682                 if (err) {
1683                         btrfs_tree_unlock(cur);
1684                         free_extent_buffer(cur);
1685                         break;
1686                 }
1687                 search_start = cur->start;
1688                 last_block = cur->start;
1689                 *last_ret = search_start;
1690                 btrfs_tree_unlock(cur);
1691                 free_extent_buffer(cur);
1692         }
1693         return err;
1694 }
1695 
1696 /*
1697  * search for key in the extent_buffer.  The items start at offset p,
1698  * and they are item_size apart.  There are 'max' items in p.
1699  *
1700  * the slot in the array is returned via slot, and it points to
1701  * the place where you would insert key if it is not found in
1702  * the array.
1703  *
1704  * slot may point to max if the key is bigger than all of the keys
1705  */
1706 static noinline int generic_bin_search(struct extent_buffer *eb,
1707                                        unsigned long p, int item_size,
1708                                        const struct btrfs_key *key,
1709                                        int max, int *slot)
1710 {
1711         int low = 0;
1712         int high = max;
1713         int mid;
1714         int ret;
1715         struct btrfs_disk_key *tmp = NULL;
1716         struct btrfs_disk_key unaligned;
1717         unsigned long offset;
1718         char *kaddr = NULL;
1719         unsigned long map_start = 0;
1720         unsigned long map_len = 0;
1721         int err;
1722 
1723         if (low > high) {
1724                 btrfs_err(eb->fs_info,
1725                  "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1726                           __func__, low, high, eb->start,
1727                           btrfs_header_owner(eb), btrfs_header_level(eb));
1728                 return -EINVAL;
1729         }
1730 
1731         while (low < high) {
1732                 mid = (low + high) / 2;
1733                 offset = p + mid * item_size;
1734 
1735                 if (!kaddr || offset < map_start ||
1736                     (offset + sizeof(struct btrfs_disk_key)) >
1737                     map_start + map_len) {
1738 
1739                         err = map_private_extent_buffer(eb, offset,
1740                                                 sizeof(struct btrfs_disk_key),
1741                                                 &kaddr, &map_start, &map_len);
1742 
1743                         if (!err) {
1744                                 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1745                                                         map_start);
1746                         } else if (err == 1) {
1747                                 read_extent_buffer(eb, &unaligned,
1748                                                    offset, sizeof(unaligned));
1749                                 tmp = &unaligned;
1750                         } else {
1751                                 return err;
1752                         }
1753 
1754                 } else {
1755                         tmp = (struct btrfs_disk_key *)(kaddr + offset -
1756                                                         map_start);
1757                 }
1758                 ret = comp_keys(tmp, key);
1759 
1760                 if (ret < 0)
1761                         low = mid + 1;
1762                 else if (ret > 0)
1763                         high = mid;
1764                 else {
1765                         *slot = mid;
1766                         return 0;
1767                 }
1768         }
1769         *slot = low;
1770         return 1;
1771 }
1772 
1773 /*
1774  * simple bin_search frontend that does the right thing for
1775  * leaves vs nodes
1776  */
1777 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1778                      int level, int *slot)
1779 {
1780         if (level == 0)
1781                 return generic_bin_search(eb,
1782                                           offsetof(struct btrfs_leaf, items),
1783                                           sizeof(struct btrfs_item),
1784                                           key, btrfs_header_nritems(eb),
1785                                           slot);
1786         else
1787                 return generic_bin_search(eb,
1788                                           offsetof(struct btrfs_node, ptrs),
1789                                           sizeof(struct btrfs_key_ptr),
1790                                           key, btrfs_header_nritems(eb),
1791                                           slot);
1792 }
1793 
1794 static void root_add_used(struct btrfs_root *root, u32 size)
1795 {
1796         spin_lock(&root->accounting_lock);
1797         btrfs_set_root_used(&root->root_item,
1798                             btrfs_root_used(&root->root_item) + size);
1799         spin_unlock(&root->accounting_lock);
1800 }
1801 
1802 static void root_sub_used(struct btrfs_root *root, u32 size)
1803 {
1804         spin_lock(&root->accounting_lock);
1805         btrfs_set_root_used(&root->root_item,
1806                             btrfs_root_used(&root->root_item) - size);
1807         spin_unlock(&root->accounting_lock);
1808 }
1809 
1810 /* given a node and slot number, this reads the blocks it points to.  The
1811  * extent buffer is returned with a reference taken (but unlocked).
1812  */
1813 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
1814                                            int slot)
1815 {
1816         int level = btrfs_header_level(parent);
1817         struct extent_buffer *eb;
1818         struct btrfs_key first_key;
1819 
1820         if (slot < 0 || slot >= btrfs_header_nritems(parent))
1821                 return ERR_PTR(-ENOENT);
1822 
1823         BUG_ON(level == 0);
1824 
1825         btrfs_node_key_to_cpu(parent, &first_key, slot);
1826         eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
1827                              btrfs_node_ptr_generation(parent, slot),
1828                              level - 1, &first_key);
1829         if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1830                 free_extent_buffer(eb);
1831                 eb = ERR_PTR(-EIO);
1832         }
1833 
1834         return eb;
1835 }
1836 
1837 /*
1838  * node level balancing, used to make sure nodes are in proper order for
1839  * item deletion.  We balance from the top down, so we have to make sure
1840  * that a deletion won't leave an node completely empty later on.
1841  */
1842 static noinline int balance_level(struct btrfs_trans_handle *trans,
1843                          struct btrfs_root *root,
1844                          struct btrfs_path *path, int level)
1845 {
1846         struct btrfs_fs_info *fs_info = root->fs_info;
1847         struct extent_buffer *right = NULL;
1848         struct extent_buffer *mid;
1849         struct extent_buffer *left = NULL;
1850         struct extent_buffer *parent = NULL;
1851         int ret = 0;
1852         int wret;
1853         int pslot;
1854         int orig_slot = path->slots[level];
1855         u64 orig_ptr;
1856 
1857         ASSERT(level > 0);
1858 
1859         mid = path->nodes[level];
1860 
1861         WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1862                 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1863         WARN_ON(btrfs_header_generation(mid) != trans->transid);
1864 
1865         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1866 
1867         if (level < BTRFS_MAX_LEVEL - 1) {
1868                 parent = path->nodes[level + 1];
1869                 pslot = path->slots[level + 1];
1870         }
1871 
1872         /*
1873          * deal with the case where there is only one pointer in the root
1874          * by promoting the node below to a root
1875          */
1876         if (!parent) {
1877                 struct extent_buffer *child;
1878 
1879                 if (btrfs_header_nritems(mid) != 1)
1880                         return 0;
1881 
1882                 /* promote the child to a root */
1883                 child = btrfs_read_node_slot(mid, 0);
1884                 if (IS_ERR(child)) {
1885                         ret = PTR_ERR(child);
1886                         btrfs_handle_fs_error(fs_info, ret, NULL);
1887                         goto enospc;
1888                 }
1889 
1890                 btrfs_tree_lock(child);
1891                 btrfs_set_lock_blocking_write(child);
1892                 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1893                 if (ret) {
1894                         btrfs_tree_unlock(child);
1895                         free_extent_buffer(child);
1896                         goto enospc;
1897                 }
1898 
1899                 ret = tree_mod_log_insert_root(root->node, child, 1);
1900                 BUG_ON(ret < 0);
1901                 rcu_assign_pointer(root->node, child);
1902 
1903                 add_root_to_dirty_list(root);
1904                 btrfs_tree_unlock(child);
1905 
1906                 path->locks[level] = 0;
1907                 path->nodes[level] = NULL;
1908                 btrfs_clean_tree_block(mid);
1909                 btrfs_tree_unlock(mid);
1910                 /* once for the path */
1911                 free_extent_buffer(mid);
1912 
1913                 root_sub_used(root, mid->len);
1914                 btrfs_free_tree_block(trans, root, mid, 0, 1);
1915                 /* once for the root ptr */
1916                 free_extent_buffer_stale(mid);
1917                 return 0;
1918         }
1919         if (btrfs_header_nritems(mid) >
1920             BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1921                 return 0;
1922 
1923         left = btrfs_read_node_slot(parent, pslot - 1);
1924         if (IS_ERR(left))
1925                 left = NULL;
1926 
1927         if (left) {
1928                 btrfs_tree_lock(left);
1929                 btrfs_set_lock_blocking_write(left);
1930                 wret = btrfs_cow_block(trans, root, left,
1931                                        parent, pslot - 1, &left);
1932                 if (wret) {
1933                         ret = wret;
1934                         goto enospc;
1935                 }
1936         }
1937 
1938         right = btrfs_read_node_slot(parent, pslot + 1);
1939         if (IS_ERR(right))
1940                 right = NULL;
1941 
1942         if (right) {
1943                 btrfs_tree_lock(right);
1944                 btrfs_set_lock_blocking_write(right);
1945                 wret = btrfs_cow_block(trans, root, right,
1946                                        parent, pslot + 1, &right);
1947                 if (wret) {
1948                         ret = wret;
1949                         goto enospc;
1950                 }
1951         }
1952 
1953         /* first, try to make some room in the middle buffer */
1954         if (left) {
1955                 orig_slot += btrfs_header_nritems(left);
1956                 wret = push_node_left(trans, left, mid, 1);
1957                 if (wret < 0)
1958                         ret = wret;
1959         }
1960 
1961         /*
1962          * then try to empty the right most buffer into the middle
1963          */
1964         if (right) {
1965                 wret = push_node_left(trans, mid, right, 1);
1966                 if (wret < 0 && wret != -ENOSPC)
1967                         ret = wret;
1968                 if (btrfs_header_nritems(right) == 0) {
1969                         btrfs_clean_tree_block(right);
1970                         btrfs_tree_unlock(right);
1971                         del_ptr(root, path, level + 1, pslot + 1);
1972                         root_sub_used(root, right->len);
1973                         btrfs_free_tree_block(trans, root, right, 0, 1);
1974                         free_extent_buffer_stale(right);
1975                         right = NULL;
1976                 } else {
1977                         struct btrfs_disk_key right_key;
1978                         btrfs_node_key(right, &right_key, 0);
1979                         ret = tree_mod_log_insert_key(parent, pslot + 1,
1980                                         MOD_LOG_KEY_REPLACE, GFP_NOFS);
1981                         BUG_ON(ret < 0);
1982                         btrfs_set_node_key(parent, &right_key, pslot + 1);
1983                         btrfs_mark_buffer_dirty(parent);
1984                 }
1985         }
1986         if (btrfs_header_nritems(mid) == 1) {
1987                 /*
1988                  * we're not allowed to leave a node with one item in the
1989                  * tree during a delete.  A deletion from lower in the tree
1990                  * could try to delete the only pointer in this node.
1991                  * So, pull some keys from the left.
1992                  * There has to be a left pointer at this point because
1993                  * otherwise we would have pulled some pointers from the
1994                  * right
1995                  */
1996                 if (!left) {
1997                         ret = -EROFS;
1998                         btrfs_handle_fs_error(fs_info, ret, NULL);
1999                         goto enospc;
2000                 }
2001                 wret = balance_node_right(trans, mid, left);
2002                 if (wret < 0) {
2003                         ret = wret;
2004                         goto enospc;
2005                 }
2006                 if (wret == 1) {
2007                         wret = push_node_left(trans, left, mid, 1);
2008                         if (wret < 0)
2009                                 ret = wret;
2010                 }
2011                 BUG_ON(wret == 1);
2012         }
2013         if (btrfs_header_nritems(mid) == 0) {
2014                 btrfs_clean_tree_block(mid);
2015                 btrfs_tree_unlock(mid);
2016                 del_ptr(root, path, level + 1, pslot);
2017                 root_sub_used(root, mid->len);
2018                 btrfs_free_tree_block(trans, root, mid, 0, 1);
2019                 free_extent_buffer_stale(mid);
2020                 mid = NULL;
2021         } else {
2022                 /* update the parent key to reflect our changes */
2023                 struct btrfs_disk_key mid_key;
2024                 btrfs_node_key(mid, &mid_key, 0);
2025                 ret = tree_mod_log_insert_key(parent, pslot,
2026                                 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2027                 BUG_ON(ret < 0);
2028                 btrfs_set_node_key(parent, &mid_key, pslot);
2029                 btrfs_mark_buffer_dirty(parent);
2030         }
2031 
2032         /* update the path */
2033         if (left) {
2034                 if (btrfs_header_nritems(left) > orig_slot) {
2035                         extent_buffer_get(left);
2036                         /* left was locked after cow */
2037                         path->nodes[level] = left;
2038                         path->slots[level + 1] -= 1;
2039                         path->slots[level] = orig_slot;
2040                         if (mid) {
2041                                 btrfs_tree_unlock(mid);
2042                                 free_extent_buffer(mid);
2043                         }
2044                 } else {
2045                         orig_slot -= btrfs_header_nritems(left);
2046                         path->slots[level] = orig_slot;
2047                 }
2048         }
2049         /* double check we haven't messed things up */
2050         if (orig_ptr !=
2051             btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2052                 BUG();
2053 enospc:
2054         if (right) {
2055                 btrfs_tree_unlock(right);
2056                 free_extent_buffer(right);
2057         }
2058         if (left) {
2059                 if (path->nodes[level] != left)
2060                         btrfs_tree_unlock(left);
2061                 free_extent_buffer(left);
2062         }
2063         return ret;
2064 }
2065 
2066 /* Node balancing for insertion.  Here we only split or push nodes around
2067  * when they are completely full.  This is also done top down, so we
2068  * have to be pessimistic.
2069  */
2070 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2071                                           struct btrfs_root *root,
2072                                           struct btrfs_path *path, int level)
2073 {
2074         struct btrfs_fs_info *fs_info = root->fs_info;
2075         struct extent_buffer *right = NULL;
2076         struct extent_buffer *mid;
2077         struct extent_buffer *left = NULL;
2078         struct extent_buffer *parent = NULL;
2079         int ret = 0;
2080         int wret;
2081         int pslot;
2082         int orig_slot = path->slots[level];
2083 
2084         if (level == 0)
2085                 return 1;
2086 
2087         mid = path->nodes[level];
2088         WARN_ON(btrfs_header_generation(mid) != trans->transid);
2089 
2090         if (level < BTRFS_MAX_LEVEL - 1) {
2091                 parent = path->nodes[level + 1];
2092                 pslot = path->slots[level + 1];
2093         }
2094 
2095         if (!parent)
2096                 return 1;
2097 
2098         left = btrfs_read_node_slot(parent, pslot - 1);
2099         if (IS_ERR(left))
2100                 left = NULL;
2101 
2102         /* first, try to make some room in the middle buffer */
2103         if (left) {
2104                 u32 left_nr;
2105 
2106                 btrfs_tree_lock(left);
2107                 btrfs_set_lock_blocking_write(left);
2108 
2109                 left_nr = btrfs_header_nritems(left);
2110                 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2111                         wret = 1;
2112                 } else {
2113                         ret = btrfs_cow_block(trans, root, left, parent,
2114                                               pslot - 1, &left);
2115                         if (ret)
2116                                 wret = 1;
2117                         else {
2118                                 wret = push_node_left(trans, left, mid, 0);
2119                         }
2120                 }
2121                 if (wret < 0)
2122                         ret = wret;
2123                 if (wret == 0) {
2124                         struct btrfs_disk_key disk_key;
2125                         orig_slot += left_nr;
2126                         btrfs_node_key(mid, &disk_key, 0);
2127                         ret = tree_mod_log_insert_key(parent, pslot,
2128                                         MOD_LOG_KEY_REPLACE, GFP_NOFS);
2129                         BUG_ON(ret < 0);
2130                         btrfs_set_node_key(parent, &disk_key, pslot);
2131                         btrfs_mark_buffer_dirty(parent);
2132                         if (btrfs_header_nritems(left) > orig_slot) {
2133                                 path->nodes[level] = left;
2134                                 path->slots[level + 1] -= 1;
2135                                 path->slots[level] = orig_slot;
2136                                 btrfs_tree_unlock(mid);
2137                                 free_extent_buffer(mid);
2138                         } else {
2139                                 orig_slot -=
2140                                         btrfs_header_nritems(left);
2141                                 path->slots[level] = orig_slot;
2142                                 btrfs_tree_unlock(left);
2143                                 free_extent_buffer(left);
2144                         }
2145                         return 0;
2146                 }
2147                 btrfs_tree_unlock(left);
2148                 free_extent_buffer(left);
2149         }
2150         right = btrfs_read_node_slot(parent, pslot + 1);
2151         if (IS_ERR(right))
2152                 right = NULL;
2153 
2154         /*
2155          * then try to empty the right most buffer into the middle
2156          */
2157         if (right) {
2158                 u32 right_nr;
2159 
2160                 btrfs_tree_lock(right);
2161                 btrfs_set_lock_blocking_write(right);
2162 
2163                 right_nr = btrfs_header_nritems(right);
2164                 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2165                         wret = 1;
2166                 } else {
2167                         ret = btrfs_cow_block(trans, root, right,
2168                                               parent, pslot + 1,
2169                                               &right);
2170                         if (ret)
2171                                 wret = 1;
2172                         else {
2173                                 wret = balance_node_right(trans, right, mid);
2174                         }
2175                 }
2176                 if (wret < 0)
2177                         ret = wret;
2178                 if (wret == 0) {
2179                         struct btrfs_disk_key disk_key;
2180 
2181                         btrfs_node_key(right, &disk_key, 0);
2182                         ret = tree_mod_log_insert_key(parent, pslot + 1,
2183                                         MOD_LOG_KEY_REPLACE, GFP_NOFS);
2184                         BUG_ON(ret < 0);
2185                         btrfs_set_node_key(parent, &disk_key, pslot + 1);
2186                         btrfs_mark_buffer_dirty(parent);
2187 
2188                         if (btrfs_header_nritems(mid) <= orig_slot) {
2189                                 path->nodes[level] = right;
2190                                 path->slots[level + 1] += 1;
2191                                 path->slots[level] = orig_slot -
2192                                         btrfs_header_nritems(mid);
2193                                 btrfs_tree_unlock(mid);
2194                                 free_extent_buffer(mid);
2195                         } else {
2196                                 btrfs_tree_unlock(right);
2197                                 free_extent_buffer(right);
2198                         }
2199                         return 0;
2200                 }
2201                 btrfs_tree_unlock(right);
2202                 free_extent_buffer(right);
2203         }
2204         return 1;
2205 }
2206 
2207 /*
2208  * readahead one full node of leaves, finding things that are close
2209  * to the block in 'slot', and triggering ra on them.
2210  */
2211 static void reada_for_search(struct btrfs_fs_info *fs_info,
2212                              struct btrfs_path *path,
2213                              int level, int slot, u64 objectid)
2214 {
2215         struct extent_buffer *node;
2216         struct btrfs_disk_key disk_key;
2217         u32 nritems;
2218         u64 search;
2219         u64 target;
2220         u64 nread = 0;
2221         struct extent_buffer *eb;
2222         u32 nr;
2223         u32 blocksize;
2224         u32 nscan = 0;
2225 
2226         if (level != 1)
2227                 return;
2228 
2229         if (!path->nodes[level])
2230                 return;
2231 
2232         node = path->nodes[level];
2233 
2234         search = btrfs_node_blockptr(node, slot);
2235         blocksize = fs_info->nodesize;
2236         eb = find_extent_buffer(fs_info, search);
2237         if (eb) {
2238                 free_extent_buffer(eb);
2239                 return;
2240         }
2241 
2242         target = search;
2243 
2244         nritems = btrfs_header_nritems(node);
2245         nr = slot;
2246 
2247         while (1) {
2248                 if (path->reada == READA_BACK) {
2249                         if (nr == 0)
2250                                 break;
2251                         nr--;
2252                 } else if (path->reada == READA_FORWARD) {
2253                         nr++;
2254                         if (nr >= nritems)
2255                                 break;
2256                 }
2257                 if (path->reada == READA_BACK && objectid) {
2258                         btrfs_node_key(node, &disk_key, nr);
2259                         if (btrfs_disk_key_objectid(&disk_key) != objectid)
2260                                 break;
2261                 }
2262                 search = btrfs_node_blockptr(node, nr);
2263                 if ((search <= target && target - search <= 65536) ||
2264                     (search > target && search - target <= 65536)) {
2265                         readahead_tree_block(fs_info, search);
2266                         nread += blocksize;
2267                 }
2268                 nscan++;
2269                 if ((nread > 65536 || nscan > 32))
2270                         break;
2271         }
2272 }
2273 
2274 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2275                                        struct btrfs_path *path, int level)
2276 {
2277         int slot;
2278         int nritems;
2279         struct extent_buffer *parent;
2280         struct extent_buffer *eb;
2281         u64 gen;
2282         u64 block1 = 0;
2283         u64 block2 = 0;
2284 
2285         parent = path->nodes[level + 1];
2286         if (!parent)
2287                 return;
2288 
2289         nritems = btrfs_header_nritems(parent);
2290         slot = path->slots[level + 1];
2291 
2292         if (slot > 0) {
2293                 block1 = btrfs_node_blockptr(parent, slot - 1);
2294                 gen = btrfs_node_ptr_generation(parent, slot - 1);
2295                 eb = find_extent_buffer(fs_info, block1);
2296                 /*
2297                  * if we get -eagain from btrfs_buffer_uptodate, we
2298                  * don't want to return eagain here.  That will loop
2299                  * forever
2300                  */
2301                 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2302                         block1 = 0;
2303                 free_extent_buffer(eb);
2304         }
2305         if (slot + 1 < nritems) {
2306                 block2 = btrfs_node_blockptr(parent, slot + 1);
2307                 gen = btrfs_node_ptr_generation(parent, slot + 1);
2308                 eb = find_extent_buffer(fs_info, block2);
2309                 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2310                         block2 = 0;
2311                 free_extent_buffer(eb);
2312         }
2313 
2314         if (block1)
2315                 readahead_tree_block(fs_info, block1);
2316         if (block2)
2317                 readahead_tree_block(fs_info, block2);
2318 }
2319 
2320 
2321 /*
2322  * when we walk down the tree, it is usually safe to unlock the higher layers
2323  * in the tree.  The exceptions are when our path goes through slot 0, because
2324  * operations on the tree might require changing key pointers higher up in the
2325  * tree.
2326  *
2327  * callers might also have set path->keep_locks, which tells this code to keep
2328  * the lock if the path points to the last slot in the block.  This is part of
2329  * walking through the tree, and selecting the next slot in the higher block.
2330  *
2331  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
2332  * if lowest_unlock is 1, level 0 won't be unlocked
2333  */
2334 static noinline void unlock_up(struct btrfs_path *path, int level,
2335                                int lowest_unlock, int min_write_lock_level,
2336                                int *write_lock_level)
2337 {
2338         int i;
2339         int skip_level = level;
2340         int no_skips = 0;
2341         struct extent_buffer *t;
2342 
2343         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2344                 if (!path->nodes[i])
2345                         break;
2346                 if (!path->locks[i])
2347                         break;
2348                 if (!no_skips && path->slots[i] == 0) {
2349                         skip_level = i + 1;
2350                         continue;
2351                 }
2352                 if (!no_skips && path->keep_locks) {
2353                         u32 nritems;
2354                         t = path->nodes[i];
2355                         nritems = btrfs_header_nritems(t);
2356                         if (nritems < 1 || path->slots[i] >= nritems - 1) {
2357                                 skip_level = i + 1;
2358                                 continue;
2359                         }
2360                 }
2361                 if (skip_level < i && i >= lowest_unlock)
2362                         no_skips = 1;
2363 
2364                 t = path->nodes[i];
2365                 if (i >= lowest_unlock && i > skip_level) {
2366                         btrfs_tree_unlock_rw(t, path->locks[i]);
2367                         path->locks[i] = 0;
2368                         if (write_lock_level &&
2369                             i > min_write_lock_level &&
2370                             i <= *write_lock_level) {
2371                                 *write_lock_level = i - 1;
2372                         }
2373                 }
2374         }
2375 }
2376 
2377 /*
2378  * This releases any locks held in the path starting at level and
2379  * going all the way up to the root.
2380  *
2381  * btrfs_search_slot will keep the lock held on higher nodes in a few
2382  * corner cases, such as COW of the block at slot zero in the node.  This
2383  * ignores those rules, and it should only be called when there are no
2384  * more updates to be done higher up in the tree.
2385  */
2386 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2387 {
2388         int i;
2389 
2390         if (path->keep_locks)
2391                 return;
2392 
2393         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2394                 if (!path->nodes[i])
2395                         continue;
2396                 if (!path->locks[i])
2397                         continue;
2398                 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2399                 path->locks[i] = 0;
2400         }
2401 }
2402 
2403 /*
2404  * helper function for btrfs_search_slot.  The goal is to find a block
2405  * in cache without setting the path to blocking.  If we find the block
2406  * we return zero and the path is unchanged.
2407  *
2408  * If we can't find the block, we set the path blocking and do some
2409  * reada.  -EAGAIN is returned and the search must be repeated.
2410  */
2411 static int
2412 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2413                       struct extent_buffer **eb_ret, int level, int slot,
2414                       const struct btrfs_key *key)
2415 {
2416         struct btrfs_fs_info *fs_info = root->fs_info;
2417         u64 blocknr;
2418         u64 gen;
2419         struct extent_buffer *b = *eb_ret;
2420         struct extent_buffer *tmp;
2421         struct btrfs_key first_key;
2422         int ret;
2423         int parent_level;
2424 
2425         blocknr = btrfs_node_blockptr(b, slot);
2426         gen = btrfs_node_ptr_generation(b, slot);
2427         parent_level = btrfs_header_level(b);
2428         btrfs_node_key_to_cpu(b, &first_key, slot);
2429 
2430         tmp = find_extent_buffer(fs_info, blocknr);
2431         if (tmp) {
2432                 /* first we do an atomic uptodate check */
2433                 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2434                         /*
2435                          * Do extra check for first_key, eb can be stale due to
2436                          * being cached, read from scrub, or have multiple
2437                          * parents (shared tree blocks).
2438                          */
2439                         if (btrfs_verify_level_key(tmp,
2440                                         parent_level - 1, &first_key, gen)) {
2441                                 free_extent_buffer(tmp);
2442                                 return -EUCLEAN;
2443                         }
2444                         *eb_ret = tmp;
2445                         return 0;
2446                 }
2447 
2448                 /* the pages were up to date, but we failed
2449                  * the generation number check.  Do a full
2450                  * read for the generation number that is correct.
2451                  * We must do this without dropping locks so
2452                  * we can trust our generation number
2453                  */
2454                 btrfs_set_path_blocking(p);
2455 
2456                 /* now we're allowed to do a blocking uptodate check */
2457                 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2458                 if (!ret) {
2459                         *eb_ret = tmp;
2460                         return 0;
2461                 }
2462                 free_extent_buffer(tmp);
2463                 btrfs_release_path(p);
2464                 return -EIO;
2465         }
2466 
2467         /*
2468          * reduce lock contention at high levels
2469          * of the btree by dropping locks before
2470          * we read.  Don't release the lock on the current
2471          * level because we need to walk this node to figure
2472          * out which blocks to read.
2473          */
2474         btrfs_unlock_up_safe(p, level + 1);
2475         btrfs_set_path_blocking(p);
2476 
2477         if (p->reada != READA_NONE)
2478                 reada_for_search(fs_info, p, level, slot, key->objectid);
2479 
2480         ret = -EAGAIN;
2481         tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2482                               &first_key);
2483         if (!IS_ERR(tmp)) {
2484                 /*
2485                  * If the read above didn't mark this buffer up to date,
2486                  * it will never end up being up to date.  Set ret to EIO now
2487                  * and give up so that our caller doesn't loop forever
2488                  * on our EAGAINs.
2489                  */
2490                 if (!extent_buffer_uptodate(tmp))
2491                         ret = -EIO;
2492                 free_extent_buffer(tmp);
2493         } else {
2494                 ret = PTR_ERR(tmp);
2495         }
2496 
2497         btrfs_release_path(p);
2498         return ret;
2499 }
2500 
2501 /*
2502  * helper function for btrfs_search_slot.  This does all of the checks
2503  * for node-level blocks and does any balancing required based on
2504  * the ins_len.
2505  *
2506  * If no extra work was required, zero is returned.  If we had to
2507  * drop the path, -EAGAIN is returned and btrfs_search_slot must
2508  * start over
2509  */
2510 static int
2511 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2512                        struct btrfs_root *root, struct btrfs_path *p,
2513                        struct extent_buffer *b, int level, int ins_len,
2514                        int *write_lock_level)
2515 {
2516         struct btrfs_fs_info *fs_info = root->fs_info;
2517         int ret;
2518 
2519         if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2520             BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2521                 int sret;
2522 
2523                 if (*write_lock_level < level + 1) {
2524                         *write_lock_level = level + 1;
2525                         btrfs_release_path(p);
2526                         goto again;
2527                 }
2528 
2529                 btrfs_set_path_blocking(p);
2530                 reada_for_balance(fs_info, p, level);
2531                 sret = split_node(trans, root, p, level);
2532 
2533                 BUG_ON(sret > 0);
2534                 if (sret) {
2535                         ret = sret;
2536                         goto done;
2537                 }
2538                 b = p->nodes[level];
2539         } else if (ins_len < 0 && btrfs_header_nritems(b) <
2540                    BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2541                 int sret;
2542 
2543                 if (*write_lock_level < level + 1) {
2544                         *write_lock_level = level + 1;
2545                         btrfs_release_path(p);
2546                         goto again;
2547                 }
2548 
2549                 btrfs_set_path_blocking(p);
2550                 reada_for_balance(fs_info, p, level);
2551                 sret = balance_level(trans, root, p, level);
2552 
2553                 if (sret) {
2554                         ret = sret;
2555                         goto done;
2556                 }
2557                 b = p->nodes[level];
2558                 if (!b) {
2559                         btrfs_release_path(p);
2560                         goto again;
2561                 }
2562                 BUG_ON(btrfs_header_nritems(b) == 1);
2563         }
2564         return 0;
2565 
2566 again:
2567         ret = -EAGAIN;
2568 done:
2569         return ret;
2570 }
2571 
2572 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2573                       int level, int *prev_cmp, int *slot)
2574 {
2575         if (*prev_cmp != 0) {
2576                 *prev_cmp = btrfs_bin_search(b, key, level, slot);
2577                 return *prev_cmp;
2578         }
2579 
2580         *slot = 0;
2581 
2582         return 0;
2583 }
2584 
2585 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2586                 u64 iobjectid, u64 ioff, u8 key_type,
2587                 struct btrfs_key *found_key)
2588 {
2589         int ret;
2590         struct btrfs_key key;
2591         struct extent_buffer *eb;
2592 
2593         ASSERT(path);
2594         ASSERT(found_key);
2595 
2596         key.type = key_type;
2597         key.objectid = iobjectid;
2598         key.offset = ioff;
2599 
2600         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2601         if (ret < 0)
2602                 return ret;
2603 
2604         eb = path->nodes[0];
2605         if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2606                 ret = btrfs_next_leaf(fs_root, path);
2607                 if (ret)
2608                         return ret;
2609                 eb = path->nodes[0];
2610         }
2611 
2612         btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2613         if (found_key->type != key.type ||
2614                         found_key->objectid != key.objectid)
2615                 return 1;
2616 
2617         return 0;
2618 }
2619 
2620 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2621                                                         struct btrfs_path *p,
2622                                                         int write_lock_level)
2623 {
2624         struct btrfs_fs_info *fs_info = root->fs_info;
2625         struct extent_buffer *b;
2626         int root_lock;
2627         int level = 0;
2628 
2629         /* We try very hard to do read locks on the root */
2630         root_lock = BTRFS_READ_LOCK;
2631 
2632         if (p->search_commit_root) {
2633                 /*
2634                  * The commit roots are read only so we always do read locks,
2635                  * and we always must hold the commit_root_sem when doing
2636                  * searches on them, the only exception is send where we don't
2637                  * want to block transaction commits for a long time, so
2638                  * we need to clone the commit root in order to avoid races
2639                  * with transaction commits that create a snapshot of one of
2640                  * the roots used by a send operation.
2641                  */
2642                 if (p->need_commit_sem) {
2643                         down_read(&fs_info->commit_root_sem);
2644                         b = btrfs_clone_extent_buffer(root->commit_root);
2645                         up_read(&fs_info->commit_root_sem);
2646                         if (!b)
2647                                 return ERR_PTR(-ENOMEM);
2648 
2649                 } else {
2650                         b = root->commit_root;
2651                         extent_buffer_get(b);
2652                 }
2653                 level = btrfs_header_level(b);
2654                 /*
2655                  * Ensure that all callers have set skip_locking when
2656                  * p->search_commit_root = 1.
2657                  */
2658                 ASSERT(p->skip_locking == 1);
2659 
2660                 goto out;
2661         }
2662 
2663         if (p->skip_locking) {
2664                 b = btrfs_root_node(root);
2665                 level = btrfs_header_level(b);
2666                 goto out;
2667         }
2668 
2669         /*
2670          * If the level is set to maximum, we can skip trying to get the read
2671          * lock.
2672          */
2673         if (write_lock_level < BTRFS_MAX_LEVEL) {
2674                 /*
2675                  * We don't know the level of the root node until we actually
2676                  * have it read locked
2677                  */
2678                 b = btrfs_read_lock_root_node(root);
2679                 level = btrfs_header_level(b);
2680                 if (level > write_lock_level)
2681                         goto out;
2682 
2683                 /* Whoops, must trade for write lock */
2684                 btrfs_tree_read_unlock(b);
2685                 free_extent_buffer(b);
2686         }
2687 
2688         b = btrfs_lock_root_node(root);
2689         root_lock = BTRFS_WRITE_LOCK;
2690 
2691         /* The level might have changed, check again */
2692         level = btrfs_header_level(b);
2693 
2694 out:
2695         p->nodes[level] = b;
2696         if (!p->skip_locking)
2697                 p->locks[level] = root_lock;
2698         /*
2699          * Callers are responsible for dropping b's references.
2700          */
2701         return b;
2702 }
2703 
2704 
2705 /*
2706  * btrfs_search_slot - look for a key in a tree and perform necessary
2707  * modifications to preserve tree invariants.
2708  *
2709  * @trans:      Handle of transaction, used when modifying the tree
2710  * @p:          Holds all btree nodes along the search path
2711  * @root:       The root node of the tree
2712  * @key:        The key we are looking for
2713  * @ins_len:    Indicates purpose of search, for inserts it is 1, for
2714  *              deletions it's -1. 0 for plain searches
2715  * @cow:        boolean should CoW operations be performed. Must always be 1
2716  *              when modifying the tree.
2717  *
2718  * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2719  * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2720  *
2721  * If @key is found, 0 is returned and you can find the item in the leaf level
2722  * of the path (level 0)
2723  *
2724  * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2725  * points to the slot where it should be inserted
2726  *
2727  * If an error is encountered while searching the tree a negative error number
2728  * is returned
2729  */
2730 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2731                       const struct btrfs_key *key, struct btrfs_path *p,
2732                       int ins_len, int cow)
2733 {
2734         struct extent_buffer *b;
2735         int slot;
2736         int ret;
2737         int err;
2738         int level;
2739         int lowest_unlock = 1;
2740         /* everything at write_lock_level or lower must be write locked */
2741         int write_lock_level = 0;
2742         u8 lowest_level = 0;
2743         int min_write_lock_level;
2744         int prev_cmp;
2745 
2746         lowest_level = p->lowest_level;
2747         WARN_ON(lowest_level && ins_len > 0);
2748         WARN_ON(p->nodes[0] != NULL);
2749         BUG_ON(!cow && ins_len);
2750 
2751         if (ins_len < 0) {
2752                 lowest_unlock = 2;
2753 
2754                 /* when we are removing items, we might have to go up to level
2755                  * two as we update tree pointers  Make sure we keep write
2756                  * for those levels as well
2757                  */
2758                 write_lock_level = 2;
2759         } else if (ins_len > 0) {
2760                 /*
2761                  * for inserting items, make sure we have a write lock on
2762                  * level 1 so we can update keys
2763                  */
2764                 write_lock_level = 1;
2765         }
2766 
2767         if (!cow)
2768                 write_lock_level = -1;
2769 
2770         if (cow && (p->keep_locks || p->lowest_level))
2771                 write_lock_level = BTRFS_MAX_LEVEL;
2772 
2773         min_write_lock_level = write_lock_level;
2774 
2775 again:
2776         prev_cmp = -1;
2777         b = btrfs_search_slot_get_root(root, p, write_lock_level);
2778         if (IS_ERR(b)) {
2779                 ret = PTR_ERR(b);
2780                 goto done;
2781         }
2782 
2783         while (b) {
2784                 level = btrfs_header_level(b);
2785 
2786                 /*
2787                  * setup the path here so we can release it under lock
2788                  * contention with the cow code
2789                  */
2790                 if (cow) {
2791                         bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2792 
2793                         /*
2794                          * if we don't really need to cow this block
2795                          * then we don't want to set the path blocking,
2796                          * so we test it here
2797                          */
2798                         if (!should_cow_block(trans, root, b)) {
2799                                 trans->dirty = true;
2800                                 goto cow_done;
2801                         }
2802 
2803                         /*
2804                          * must have write locks on this node and the
2805                          * parent
2806                          */
2807                         if (level > write_lock_level ||
2808                             (level + 1 > write_lock_level &&
2809                             level + 1 < BTRFS_MAX_LEVEL &&
2810                             p->nodes[level + 1])) {
2811                                 write_lock_level = level + 1;
2812                                 btrfs_release_path(p);
2813                                 goto again;
2814                         }
2815 
2816                         btrfs_set_path_blocking(p);
2817                         if (last_level)
2818                                 err = btrfs_cow_block(trans, root, b, NULL, 0,
2819                                                       &b);
2820                         else
2821                                 err = btrfs_cow_block(trans, root, b,
2822                                                       p->nodes[level + 1],
2823                                                       p->slots[level + 1], &b);
2824                         if (err) {
2825                                 ret = err;
2826                                 goto done;
2827                         }
2828                 }
2829 cow_done:
2830                 p->nodes[level] = b;
2831                 /*
2832                  * Leave path with blocking locks to avoid massive
2833                  * lock context switch, this is made on purpose.
2834                  */
2835 
2836                 /*
2837                  * we have a lock on b and as long as we aren't changing
2838                  * the tree, there is no way to for the items in b to change.
2839                  * It is safe to drop the lock on our parent before we
2840                  * go through the expensive btree search on b.
2841                  *
2842                  * If we're inserting or deleting (ins_len != 0), then we might
2843                  * be changing slot zero, which may require changing the parent.
2844                  * So, we can't drop the lock until after we know which slot
2845                  * we're operating on.
2846                  */
2847                 if (!ins_len && !p->keep_locks) {
2848                         int u = level + 1;
2849 
2850                         if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2851                                 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2852                                 p->locks[u] = 0;
2853                         }
2854                 }
2855 
2856                 ret = key_search(b, key, level, &prev_cmp, &slot);
2857                 if (ret < 0)
2858                         goto done;
2859 
2860                 if (level != 0) {
2861                         int dec = 0;
2862                         if (ret && slot > 0) {
2863                                 dec = 1;
2864                                 slot -= 1;
2865                         }
2866                         p->slots[level] = slot;
2867                         err = setup_nodes_for_search(trans, root, p, b, level,
2868                                              ins_len, &write_lock_level);
2869                         if (err == -EAGAIN)
2870                                 goto again;
2871                         if (err) {
2872                                 ret = err;
2873                                 goto done;
2874                         }
2875                         b = p->nodes[level];
2876                         slot = p->slots[level];
2877 
2878                         /*
2879                          * slot 0 is special, if we change the key
2880                          * we have to update the parent pointer
2881                          * which means we must have a write lock
2882                          * on the parent
2883                          */
2884                         if (slot == 0 && ins_len &&
2885                             write_lock_level < level + 1) {
2886                                 write_lock_level = level + 1;
2887                                 btrfs_release_path(p);
2888                                 goto again;
2889                         }
2890 
2891                         unlock_up(p, level, lowest_unlock,
2892                                   min_write_lock_level, &write_lock_level);
2893 
2894                         if (level == lowest_level) {
2895                                 if (dec)
2896                                         p->slots[level]++;
2897                                 goto done;
2898                         }
2899 
2900                         err = read_block_for_search(root, p, &b, level,
2901                                                     slot, key);
2902                         if (err == -EAGAIN)
2903                                 goto again;
2904                         if (err) {
2905                                 ret = err;
2906                                 goto done;
2907                         }
2908 
2909                         if (!p->skip_locking) {
2910                                 level = btrfs_header_level(b);
2911                                 if (level <= write_lock_level) {
2912                                         if (!btrfs_try_tree_write_lock(b)) {
2913                                                 btrfs_set_path_blocking(p);
2914                                                 btrfs_tree_lock(b);
2915                                         }
2916                                         p->locks[level] = BTRFS_WRITE_LOCK;
2917                                 } else {
2918                                         if (!btrfs_tree_read_lock_atomic(b)) {
2919                                                 btrfs_set_path_blocking(p);
2920                                                 btrfs_tree_read_lock(b);
2921                                         }
2922                                         p->locks[level] = BTRFS_READ_LOCK;
2923                                 }
2924                                 p->nodes[level] = b;
2925                         }
2926                 } else {
2927                         p->slots[level] = slot;
2928                         if (ins_len > 0 &&
2929                             btrfs_leaf_free_space(b) < ins_len) {
2930                                 if (write_lock_level < 1) {
2931                                         write_lock_level = 1;
2932                                         btrfs_release_path(p);
2933                                         goto again;
2934                                 }
2935 
2936                                 btrfs_set_path_blocking(p);
2937                                 err = split_leaf(trans, root, key,
2938                                                  p, ins_len, ret == 0);
2939 
2940                                 BUG_ON(err > 0);
2941                                 if (err) {
2942                                         ret = err;
2943                                         goto done;
2944                                 }
2945                         }
2946                         if (!p->search_for_split)
2947                                 unlock_up(p, level, lowest_unlock,
2948                                           min_write_lock_level, NULL);
2949                         goto done;
2950                 }
2951         }
2952         ret = 1;
2953 done:
2954         /*
2955          * we don't really know what they plan on doing with the path
2956          * from here on, so for now just mark it as blocking
2957          */
2958         if (!p->leave_spinning)
2959                 btrfs_set_path_blocking(p);
2960         if (ret < 0 && !p->skip_release_on_error)
2961                 btrfs_release_path(p);
2962         return ret;
2963 }
2964 
2965 /*
2966  * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2967  * current state of the tree together with the operations recorded in the tree
2968  * modification log to search for the key in a previous version of this tree, as
2969  * denoted by the time_seq parameter.
2970  *
2971  * Naturally, there is no support for insert, delete or cow operations.
2972  *
2973  * The resulting path and return value will be set up as if we called
2974  * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2975  */
2976 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2977                           struct btrfs_path *p, u64 time_seq)
2978 {
2979         struct btrfs_fs_info *fs_info = root->fs_info;
2980         struct extent_buffer *b;
2981         int slot;
2982         int ret;
2983         int err;
2984         int level;
2985         int lowest_unlock = 1;
2986         u8 lowest_level = 0;
2987         int prev_cmp = -1;
2988 
2989         lowest_level = p->lowest_level;
2990         WARN_ON(p->nodes[0] != NULL);
2991 
2992         if (p->search_commit_root) {
2993                 BUG_ON(time_seq);
2994                 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2995         }
2996 
2997 again:
2998         b = get_old_root(root, time_seq);
2999         if (!b) {
3000                 ret = -EIO;
3001                 goto done;
3002         }
3003         level = btrfs_header_level(b);
3004         p->locks[level] = BTRFS_READ_LOCK;
3005 
3006         while (b) {
3007                 level = btrfs_header_level(b);
3008                 p->nodes[level] = b;
3009 
3010                 /*
3011                  * we have a lock on b and as long as we aren't changing
3012                  * the tree, there is no way to for the items in b to change.
3013                  * It is safe to drop the lock on our parent before we
3014                  * go through the expensive btree search on b.
3015                  */
3016                 btrfs_unlock_up_safe(p, level + 1);
3017 
3018                 /*
3019                  * Since we can unwind ebs we want to do a real search every
3020                  * time.
3021                  */
3022                 prev_cmp = -1;
3023                 ret = key_search(b, key, level, &prev_cmp, &slot);
3024                 if (ret < 0)
3025                         goto done;
3026 
3027                 if (level != 0) {
3028                         int dec = 0;
3029                         if (ret && slot > 0) {
3030                                 dec = 1;
3031                                 slot -= 1;
3032                         }
3033                         p->slots[level] = slot;
3034                         unlock_up(p, level, lowest_unlock, 0, NULL);
3035 
3036                         if (level == lowest_level) {
3037                                 if (dec)
3038                                         p->slots[level]++;
3039                                 goto done;
3040                         }
3041 
3042                         err = read_block_for_search(root, p, &b, level,
3043                                                     slot, key);
3044                         if (err == -EAGAIN)
3045                                 goto again;
3046                         if (err) {
3047                                 ret = err;
3048                                 goto done;
3049                         }
3050 
3051                         level = btrfs_header_level(b);
3052                         if (!btrfs_tree_read_lock_atomic(b)) {
3053                                 btrfs_set_path_blocking(p);
3054                                 btrfs_tree_read_lock(b);
3055                         }
3056                         b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3057                         if (!b) {
3058                                 ret = -ENOMEM;
3059                                 goto done;
3060                         }
3061                         p->locks[level] = BTRFS_READ_LOCK;
3062                         p->nodes[level] = b;
3063                 } else {
3064                         p->slots[level] = slot;
3065                         unlock_up(p, level, lowest_unlock, 0, NULL);
3066                         goto done;
3067                 }
3068         }
3069         ret = 1;
3070 done:
3071         if (!p->leave_spinning)
3072                 btrfs_set_path_blocking(p);
3073         if (ret < 0)
3074                 btrfs_release_path(p);
3075 
3076         return ret;
3077 }
3078 
3079 /*
3080  * helper to use instead of search slot if no exact match is needed but
3081  * instead the next or previous item should be returned.
3082  * When find_higher is true, the next higher item is returned, the next lower
3083  * otherwise.
3084  * When return_any and find_higher are both true, and no higher item is found,
3085  * return the next lower instead.
3086  * When return_any is true and find_higher is false, and no lower item is found,
3087  * return the next higher instead.
3088  * It returns 0 if any item is found, 1 if none is found (tree empty), and
3089  * < 0 on error
3090  */
3091 int btrfs_search_slot_for_read(struct btrfs_root *root,
3092                                const struct btrfs_key *key,
3093                                struct btrfs_path *p, int find_higher,
3094                                int return_any)
3095 {
3096         int ret;
3097         struct extent_buffer *leaf;
3098 
3099 again:
3100         ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3101         if (ret <= 0)
3102                 return ret;
3103         /*
3104          * a return value of 1 means the path is at the position where the
3105          * item should be inserted. Normally this is the next bigger item,
3106          * but in case the previous item is the last in a leaf, path points
3107          * to the first free slot in the previous leaf, i.e. at an invalid
3108          * item.
3109          */
3110         leaf = p->nodes[0];
3111 
3112         if (find_higher) {
3113                 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3114                         ret = btrfs_next_leaf(root, p);
3115                         if (ret <= 0)
3116                                 return ret;
3117                         if (!return_any)
3118                                 return 1;
3119                         /*
3120                          * no higher item found, return the next
3121                          * lower instead
3122                          */
3123                         return_any = 0;
3124                         find_higher = 0;
3125                         btrfs_release_path(p);
3126                         goto again;
3127                 }
3128         } else {
3129                 if (p->slots[0] == 0) {
3130                         ret = btrfs_prev_leaf(root, p);
3131                         if (ret < 0)
3132                                 return ret;
3133                         if (!ret) {
3134                                 leaf = p->nodes[0];
3135                                 if (p->slots[0] == btrfs_header_nritems(leaf))
3136                                         p->slots[0]--;
3137                                 return 0;
3138                         }
3139                         if (!return_any)
3140                                 return 1;
3141                         /*
3142                          * no lower item found, return the next
3143                          * higher instead
3144                          */
3145                         return_any = 0;
3146                         find_higher = 1;
3147                         btrfs_release_path(p);
3148                         goto again;
3149                 } else {
3150                         --p->slots[0];
3151                 }
3152         }
3153         return 0;
3154 }
3155 
3156 /*
3157  * adjust the pointers going up the tree, starting at level
3158  * making sure the right key of each node is points to 'key'.
3159  * This is used after shifting pointers to the left, so it stops
3160  * fixing up pointers when a given leaf/node is not in slot 0 of the
3161  * higher levels
3162  *
3163  */
3164 static void fixup_low_keys(struct btrfs_path *path,
3165                            struct btrfs_disk_key *key, int level)
3166 {
3167         int i;
3168         struct extent_buffer *t;
3169         int ret;
3170 
3171         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3172                 int tslot = path->slots[i];
3173 
3174                 if (!path->nodes[i])
3175                         break;
3176                 t = path->nodes[i];
3177                 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3178                                 GFP_ATOMIC);
3179                 BUG_ON(ret < 0);
3180                 btrfs_set_node_key(t, key, tslot);
3181                 btrfs_mark_buffer_dirty(path->nodes[i]);
3182                 if (tslot != 0)
3183                         break;
3184         }
3185 }
3186 
3187 /*
3188  * update item key.
3189  *
3190  * This function isn't completely safe. It's the caller's responsibility
3191  * that the new key won't break the order
3192  */
3193 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3194                              struct btrfs_path *path,
3195                              const struct btrfs_key *new_key)
3196 {
3197         struct btrfs_disk_key disk_key;
3198         struct extent_buffer *eb;
3199         int slot;
3200 
3201         eb = path->nodes[0];
3202         slot = path->slots[0];
3203         if (slot > 0) {
3204                 btrfs_item_key(eb, &disk_key, slot - 1);
3205                 if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
3206                         btrfs_crit(fs_info,
3207                 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3208                                    slot, btrfs_disk_key_objectid(&disk_key),
3209                                    btrfs_disk_key_type(&disk_key),
3210                                    btrfs_disk_key_offset(&disk_key),
3211                                    new_key->objectid, new_key->type,
3212                                    new_key->offset);
3213                         btrfs_print_leaf(eb);
3214                         BUG();
3215                 }
3216         }
3217         if (slot < btrfs_header_nritems(eb) - 1) {
3218                 btrfs_item_key(eb, &disk_key, slot + 1);
3219                 if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
3220                         btrfs_crit(fs_info,
3221                 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3222                                    slot, btrfs_disk_key_objectid(&disk_key),
3223                                    btrfs_disk_key_type(&disk_key),
3224                                    btrfs_disk_key_offset(&disk_key),
3225                                    new_key->objectid, new_key->type,
3226                                    new_key->offset);
3227                         btrfs_print_leaf(eb);
3228                         BUG();
3229                 }
3230         }
3231 
3232         btrfs_cpu_key_to_disk(&disk_key, new_key);
3233         btrfs_set_item_key(eb, &disk_key, slot);
3234         btrfs_mark_buffer_dirty(eb);
3235         if (slot == 0)
3236                 fixup_low_keys(path, &disk_key, 1);
3237 }
3238 
3239 /*
3240  * try to push data from one node into the next node left in the
3241  * tree.
3242  *
3243  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3244  * error, and > 0 if there was no room in the left hand block.
3245  */
3246 static int push_node_left(struct btrfs_trans_handle *trans,
3247                           struct extent_buffer *dst,
3248                           struct extent_buffer *src, int empty)
3249 {
3250         struct btrfs_fs_info *fs_info = trans->fs_info;
3251         int push_items = 0;
3252         int src_nritems;
3253         int dst_nritems;
3254         int ret = 0;
3255 
3256         src_nritems = btrfs_header_nritems(src);
3257         dst_nritems = btrfs_header_nritems(dst);
3258         push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3259         WARN_ON(btrfs_header_generation(src) != trans->transid);
3260         WARN_ON(btrfs_header_generation(dst) != trans->transid);
3261 
3262         if (!empty && src_nritems <= 8)
3263                 return 1;
3264 
3265         if (push_items <= 0)
3266                 return 1;
3267 
3268         if (empty) {
3269                 push_items = min(src_nritems, push_items);
3270                 if (push_items < src_nritems) {
3271                         /* leave at least 8 pointers in the node if
3272                          * we aren't going to empty it
3273                          */
3274                         if (src_nritems - push_items < 8) {
3275                                 if (push_items <= 8)
3276                                         return 1;
3277                                 push_items -= 8;
3278                         }
3279                 }
3280         } else
3281                 push_items = min(src_nritems - 8, push_items);
3282 
3283         ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
3284         if (ret) {
3285                 btrfs_abort_transaction(trans, ret);
3286                 return ret;
3287         }
3288         copy_extent_buffer(dst, src,
3289                            btrfs_node_key_ptr_offset(dst_nritems),
3290                            btrfs_node_key_ptr_offset(0),
3291                            push_items * sizeof(struct btrfs_key_ptr));
3292 
3293         if (push_items < src_nritems) {
3294                 /*
3295                  * Don't call tree_mod_log_insert_move here, key removal was
3296                  * already fully logged by tree_mod_log_eb_copy above.
3297                  */
3298                 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3299                                       btrfs_node_key_ptr_offset(push_items),
3300                                       (src_nritems - push_items) *
3301                                       sizeof(struct btrfs_key_ptr));
3302         }
3303         btrfs_set_header_nritems(src, src_nritems - push_items);
3304         btrfs_set_header_nritems(dst, dst_nritems + push_items);
3305         btrfs_mark_buffer_dirty(src);
3306         btrfs_mark_buffer_dirty(dst);
3307 
3308         return ret;
3309 }
3310 
3311 /*
3312  * try to push data from one node into the next node right in the
3313  * tree.
3314  *
3315  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3316  * error, and > 0 if there was no room in the right hand block.
3317  *
3318  * this will  only push up to 1/2 the contents of the left node over
3319  */
3320 static int balance_node_right(struct btrfs_trans_handle *trans,
3321                               struct extent_buffer *dst,
3322                               struct extent_buffer *src)
3323 {
3324         struct btrfs_fs_info *fs_info = trans->fs_info;
3325         int push_items = 0;
3326         int max_push;
3327         int src_nritems;
3328         int dst_nritems;
3329         int ret = 0;
3330 
3331         WARN_ON(btrfs_header_generation(src) != trans->transid);
3332         WARN_ON(btrfs_header_generation(dst) != trans->transid);
3333 
3334         src_nritems = btrfs_header_nritems(src);
3335         dst_nritems = btrfs_header_nritems(dst);
3336         push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3337         if (push_items <= 0)
3338                 return 1;
3339 
3340         if (src_nritems < 4)
3341                 return 1;
3342 
3343         max_push = src_nritems / 2 + 1;
3344         /* don't try to empty the node */
3345         if (max_push >= src_nritems)
3346                 return 1;
3347 
3348         if (max_push < push_items)
3349                 push_items = max_push;
3350 
3351         ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3352         BUG_ON(ret < 0);
3353         memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3354                                       btrfs_node_key_ptr_offset(0),
3355                                       (dst_nritems) *
3356                                       sizeof(struct btrfs_key_ptr));
3357 
3358         ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
3359                                    push_items);
3360         if (ret) {
3361                 btrfs_abort_transaction(trans, ret);
3362                 return ret;
3363         }
3364         copy_extent_buffer(dst, src,
3365                            btrfs_node_key_ptr_offset(0),
3366                            btrfs_node_key_ptr_offset(src_nritems - push_items),
3367                            push_items * sizeof(struct btrfs_key_ptr));
3368 
3369         btrfs_set_header_nritems(src, src_nritems - push_items);
3370         btrfs_set_header_nritems(dst, dst_nritems + push_items);
3371 
3372         btrfs_mark_buffer_dirty(src);
3373         btrfs_mark_buffer_dirty(dst);
3374 
3375         return ret;
3376 }
3377 
3378 /*
3379  * helper function to insert a new root level in the tree.
3380  * A new node is allocated, and a single item is inserted to
3381  * point to the existing root
3382  *
3383  * returns zero on success or < 0 on failure.
3384  */
3385 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3386                            struct btrfs_root *root,
3387                            struct btrfs_path *path, int level)
3388 {
3389         struct btrfs_fs_info *fs_info = root->fs_info;
3390         u64 lower_gen;
3391         struct extent_buffer *lower;
3392         struct extent_buffer *c;
3393         struct extent_buffer *old;
3394         struct btrfs_disk_key lower_key;
3395         int ret;
3396 
3397         BUG_ON(path->nodes[level]);
3398         BUG_ON(path->nodes[level-1] != root->node);
3399 
3400         lower = path->nodes[level-1];
3401         if (level == 1)
3402                 btrfs_item_key(lower, &lower_key, 0);
3403         else
3404                 btrfs_node_key(lower, &lower_key, 0);
3405 
3406         c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
3407                                          root->node->start, 0);
3408         if (IS_ERR(c))
3409                 return PTR_ERR(c);
3410 
3411         root_add_used(root, fs_info->nodesize);
3412 
3413         btrfs_set_header_nritems(c, 1);
3414         btrfs_set_node_key(c, &lower_key, 0);
3415         btrfs_set_node_blockptr(c, 0, lower->start);
3416         lower_gen = btrfs_header_generation(lower);
3417         WARN_ON(lower_gen != trans->transid);
3418 
3419         btrfs_set_node_ptr_generation(c, 0, lower_gen);
3420 
3421         btrfs_mark_buffer_dirty(c);
3422 
3423         old = root->node;
3424         ret = tree_mod_log_insert_root(root->node, c, 0);
3425         BUG_ON(ret < 0);
3426         rcu_assign_pointer(root->node, c);
3427 
3428         /* the super has an extra ref to root->node */
3429         free_extent_buffer(old);
3430 
3431         add_root_to_dirty_list(root);
3432         extent_buffer_get(c);
3433         path->nodes[level] = c;
3434         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3435         path->slots[level] = 0;
3436         return 0;
3437 }
3438 
3439 /*
3440  * worker function to insert a single pointer in a node.
3441  * the node should have enough room for the pointer already
3442  *
3443  * slot and level indicate where you want the key to go, and
3444  * blocknr is the block the key points to.
3445  */
3446 static void insert_ptr(struct btrfs_trans_handle *trans,
3447                        struct btrfs_path *path,
3448                        struct btrfs_disk_key *key, u64 bytenr,
3449                        int slot, int level)
3450 {
3451         struct extent_buffer *lower;
3452         int nritems;
3453         int ret;
3454 
3455         BUG_ON(!path->nodes[level]);
3456         btrfs_assert_tree_locked(path->nodes[level]);
3457         lower = path->nodes[level];
3458         nritems = btrfs_header_nritems(lower);
3459         BUG_ON(slot > nritems);
3460         BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
3461         if (slot != nritems) {
3462                 if (level) {
3463                         ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3464                                         nritems - slot);
3465                         BUG_ON(ret < 0);
3466                 }
3467                 memmove_extent_buffer(lower,
3468                               btrfs_node_key_ptr_offset(slot + 1),
3469                               btrfs_node_key_ptr_offset(slot),
3470                               (nritems - slot) * sizeof(struct btrfs_key_ptr));
3471         }
3472         if (level) {
3473                 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3474                                 GFP_NOFS);
3475                 BUG_ON(ret < 0);
3476         }
3477         btrfs_set_node_key(lower, key, slot);
3478         btrfs_set_node_blockptr(lower, slot, bytenr);
3479         WARN_ON(trans->transid == 0);
3480         btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3481         btrfs_set_header_nritems(lower, nritems + 1);
3482         btrfs_mark_buffer_dirty(lower);
3483 }
3484 
3485 /*
3486  * split the node at the specified level in path in two.
3487  * The path is corrected to point to the appropriate node after the split
3488  *
3489  * Before splitting this tries to make some room in the node by pushing
3490  * left and right, if either one works, it returns right away.
3491  *
3492  * returns 0 on success and < 0 on failure
3493  */
3494 static noinline int split_node(struct btrfs_trans_handle *trans,
3495                                struct btrfs_root *root,
3496                                struct btrfs_path *path, int level)
3497 {
3498         struct btrfs_fs_info *fs_info = root->fs_info;
3499         struct extent_buffer *c;
3500         struct extent_buffer *split;
3501         struct btrfs_disk_key disk_key;
3502         int mid;
3503         int ret;
3504         u32 c_nritems;
3505 
3506         c = path->nodes[level];
3507         WARN_ON(btrfs_header_generation(c) != trans->transid);
3508         if (c == root->node) {
3509                 /*
3510                  * trying to split the root, lets make a new one
3511                  *
3512                  * tree mod log: We don't log_removal old root in
3513                  * insert_new_root, because that root buffer will be kept as a
3514                  * normal node. We are going to log removal of half of the
3515                  * elements below with tree_mod_log_eb_copy. We're holding a
3516                  * tree lock on the buffer, which is why we cannot race with
3517                  * other tree_mod_log users.
3518                  */
3519                 ret = insert_new_root(trans, root, path, level + 1);
3520                 if (ret)
3521                         return ret;
3522         } else {
3523                 ret = push_nodes_for_insert(trans, root, path, level);
3524                 c = path->nodes[level];
3525                 if (!ret && btrfs_header_nritems(c) <
3526                     BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3527                         return 0;
3528                 if (ret < 0)
3529                         return ret;
3530         }
3531 
3532         c_nritems = btrfs_header_nritems(c);
3533         mid = (c_nritems + 1) / 2;
3534         btrfs_node_key(c, &disk_key, mid);
3535 
3536         split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
3537                                              c->start, 0);
3538         if (IS_ERR(split))
3539                 return PTR_ERR(split);
3540 
3541         root_add_used(root, fs_info->nodesize);
3542         ASSERT(btrfs_header_level(c) == level);
3543 
3544         ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
3545         if (ret) {
3546                 btrfs_abort_transaction(trans, ret);
3547                 return ret;
3548         }
3549         copy_extent_buffer(split, c,
3550                            btrfs_node_key_ptr_offset(0),
3551                            btrfs_node_key_ptr_offset(mid),
3552                            (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3553         btrfs_set_header_nritems(split, c_nritems - mid);
3554         btrfs_set_header_nritems(c, mid);
3555         ret = 0;
3556 
3557         btrfs_mark_buffer_dirty(c);
3558         btrfs_mark_buffer_dirty(split);
3559 
3560         insert_ptr(trans, path, &disk_key, split->start,
3561                    path->slots[level + 1] + 1, level + 1);
3562 
3563         if (path->slots[level] >= mid) {
3564                 path->slots[level] -= mid;
3565                 btrfs_tree_unlock(c);
3566                 free_extent_buffer(c);
3567                 path->nodes[level] = split;
3568                 path->slots[level + 1] += 1;
3569         } else {
3570                 btrfs_tree_unlock(split);
3571                 free_extent_buffer(split);
3572         }
3573         return ret;
3574 }
3575 
3576 /*
3577  * how many bytes are required to store the items in a leaf.  start
3578  * and nr indicate which items in the leaf to check.  This totals up the
3579  * space used both by the item structs and the item data
3580  */
3581 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3582 {
3583         struct btrfs_item *start_item;
3584         struct btrfs_item *end_item;
3585         struct btrfs_map_token token;
3586         int data_len;
3587         int nritems = btrfs_header_nritems(l);
3588         int end = min(nritems, start + nr) - 1;
3589 
3590         if (!nr)
3591                 return 0;
3592         btrfs_init_map_token(&token, l);
3593         start_item = btrfs_item_nr(start);
3594         end_item = btrfs_item_nr(end);
3595         data_len = btrfs_token_item_offset(l, start_item, &token) +
3596                 btrfs_token_item_size(l, start_item, &token);
3597         data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3598         data_len += sizeof(struct btrfs_item) * nr;
3599         WARN_ON(data_len < 0);
3600         return data_len;
3601 }
3602 
3603 /*
3604  * The space between the end of the leaf items and
3605  * the start of the leaf data.  IOW, how much room
3606  * the leaf has left for both items and data
3607  */
3608 noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
3609 {
3610         struct btrfs_fs_info *fs_info = leaf->fs_info;
3611         int nritems = btrfs_header_nritems(leaf);
3612         int ret;
3613 
3614         ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3615         if (ret < 0) {
3616                 btrfs_crit(fs_info,
3617                            "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3618                            ret,
3619                            (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3620                            leaf_space_used(leaf, 0, nritems), nritems);
3621         }
3622         return ret;
3623 }
3624 
3625 /*
3626  * min slot controls the lowest index we're willing to push to the
3627  * right.  We'll push up to and including min_slot, but no lower
3628  */
3629 static noinline int __push_leaf_right(struct btrfs_path *path,
3630                                       int data_size, int empty,
3631                                       struct extent_buffer *right,
3632                                       int free_space, u32 left_nritems,
3633                                       u32 min_slot)
3634 {
3635         struct btrfs_fs_info *fs_info = right->fs_info;
3636         struct extent_buffer *left = path->nodes[0];
3637         struct extent_buffer *upper = path->nodes[1];
3638         struct btrfs_map_token token;
3639         struct btrfs_disk_key disk_key;
3640         int slot;
3641         u32 i;
3642         int push_space = 0;
3643         int push_items = 0;
3644         struct btrfs_item *item;
3645         u32 nr;
3646         u32 right_nritems;
3647         u32 data_end;
3648         u32 this_item_size;
3649 
3650         if (empty)
3651                 nr = 0;
3652         else
3653                 nr = max_t(u32, 1, min_slot);
3654 
3655         if (path->slots[0] >= left_nritems)
3656                 push_space += data_size;
3657 
3658         slot = path->slots[1];
3659         i = left_nritems - 1;
3660         while (i >= nr) {
3661                 item = btrfs_item_nr(i);
3662 
3663                 if (!empty && push_items > 0) {
3664                         if (path->slots[0] > i)
3665                                 break;
3666                         if (path->slots[0] == i) {
3667                                 int space = btrfs_leaf_free_space(left);
3668 
3669                                 if (space + push_space * 2 > free_space)
3670                                         break;
3671                         }
3672                 }
3673 
3674                 if (path->slots[0] == i)
3675                         push_space += data_size;
3676 
3677                 this_item_size = btrfs_item_size(left, item);
3678                 if (this_item_size + sizeof(*item) + push_space > free_space)
3679                         break;
3680 
3681                 push_items++;
3682                 push_space += this_item_size + sizeof(*item);
3683                 if (i == 0)
3684                         break;
3685                 i--;
3686         }
3687 
3688         if (push_items == 0)
3689                 goto out_unlock;
3690 
3691         WARN_ON(!empty && push_items == left_nritems);
3692 
3693         /* push left to right */
3694         right_nritems = btrfs_header_nritems(right);
3695 
3696         push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3697         push_space -= leaf_data_end(left);
3698 
3699         /* make room in the right data area */
3700         data_end = leaf_data_end(right);
3701         memmove_extent_buffer(right,
3702                               BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3703                               BTRFS_LEAF_DATA_OFFSET + data_end,
3704                               BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3705 
3706         /* copy from the left data area */
3707         copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3708                      BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3709                      BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left),
3710                      push_space);
3711 
3712         memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3713                               btrfs_item_nr_offset(0),
3714                               right_nritems * sizeof(struct btrfs_item));
3715 
3716         /* copy the items from left to right */
3717         copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3718                    btrfs_item_nr_offset(left_nritems - push_items),
3719                    push_items * sizeof(struct btrfs_item));
3720 
3721         /* update the item pointers */
3722         btrfs_init_map_token(&token, right);
3723         right_nritems += push_items;
3724         btrfs_set_header_nritems(right, right_nritems);
3725         push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3726         for (i = 0; i < right_nritems; i++) {
3727                 item = btrfs_item_nr(i);
3728                 push_space -= btrfs_token_item_size(right, item, &token);
3729                 btrfs_set_token_item_offset(right, item, push_space, &token);
3730         }
3731 
3732         left_nritems -= push_items;
3733         btrfs_set_header_nritems(left, left_nritems);
3734 
3735         if (left_nritems)
3736                 btrfs_mark_buffer_dirty(left);
3737         else
3738                 btrfs_clean_tree_block(left);
3739 
3740         btrfs_mark_buffer_dirty(right);
3741 
3742         btrfs_item_key(right, &disk_key, 0);
3743         btrfs_set_node_key(upper, &disk_key, slot + 1);
3744         btrfs_mark_buffer_dirty(upper);
3745 
3746         /* then fixup the leaf pointer in the path */
3747         if (path->slots[0] >= left_nritems) {
3748                 path->slots[0] -= left_nritems;
3749                 if (btrfs_header_nritems(path->nodes[0]) == 0)
3750                         btrfs_clean_tree_block(path->nodes[0]);
3751                 btrfs_tree_unlock(path->nodes[0]);
3752                 free_extent_buffer(path->nodes[0]);
3753                 path->nodes[0] = right;
3754                 path->slots[1] += 1;
3755         } else {
3756                 btrfs_tree_unlock(right);
3757                 free_extent_buffer(right);
3758         }
3759         return 0;
3760 
3761 out_unlock:
3762         btrfs_tree_unlock(right);
3763         free_extent_buffer(right);
3764         return 1;
3765 }
3766 
3767 /*
3768  * push some data in the path leaf to the right, trying to free up at
3769  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3770  *
3771  * returns 1 if the push failed because the other node didn't have enough
3772  * room, 0 if everything worked out and < 0 if there were major errors.
3773  *
3774  * this will push starting from min_slot to the end of the leaf.  It won't
3775  * push any slot lower than min_slot
3776  */
3777 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3778                            *root, struct btrfs_path *path,
3779                            int min_data_size, int data_size,
3780                            int empty, u32 min_slot)
3781 {
3782         struct extent_buffer *left = path->nodes[0];
3783         struct extent_buffer *right;
3784         struct extent_buffer *upper;
3785         int slot;
3786         int free_space;
3787         u32 left_nritems;
3788         int ret;
3789 
3790         if (!path->nodes[1])
3791                 return 1;
3792 
3793         slot = path->slots[1];
3794         upper = path->nodes[1];
3795         if (slot >= btrfs_header_nritems(upper) - 1)
3796                 return 1;
3797 
3798         btrfs_assert_tree_locked(path->nodes[1]);
3799 
3800         right = btrfs_read_node_slot(upper, slot + 1);
3801         /*
3802          * slot + 1 is not valid or we fail to read the right node,
3803          * no big deal, just return.
3804          */
3805         if (IS_ERR(right))
3806                 return 1;
3807 
3808         btrfs_tree_lock(right);
3809         btrfs_set_lock_blocking_write(right);
3810 
3811         free_space = btrfs_leaf_free_space(right);
3812         if (free_space < data_size)
3813                 goto out_unlock;
3814 
3815         /* cow and double check */
3816         ret = btrfs_cow_block(trans, root, right, upper,
3817                               slot + 1, &right);
3818         if (ret)
3819                 goto out_unlock;
3820 
3821         free_space = btrfs_leaf_free_space(right);
3822         if (free_space < data_size)
3823                 goto out_unlock;
3824 
3825         left_nritems = btrfs_header_nritems(left);
3826         if (left_nritems == 0)
3827                 goto out_unlock;
3828 
3829         if (path->slots[0] == left_nritems && !empty) {
3830                 /* Key greater than all keys in the leaf, right neighbor has
3831                  * enough room for it and we're not emptying our leaf to delete
3832                  * it, therefore use right neighbor to insert the new item and
3833                  * no need to touch/dirty our left leaf. */
3834                 btrfs_tree_unlock(left);
3835                 free_extent_buffer(left);
3836                 path->nodes[0] = right;
3837                 path->slots[0] = 0;
3838                 path->slots[1]++;
3839                 return 0;
3840         }
3841 
3842         return __push_leaf_right(path, min_data_size, empty,
3843                                 right, free_space, left_nritems, min_slot);
3844 out_unlock:
3845         btrfs_tree_unlock(right);
3846         free_extent_buffer(right);
3847         return 1;
3848 }
3849 
3850 /*
3851  * push some data in the path leaf to the left, trying to free up at
3852  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3853  *
3854  * max_slot can put a limit on how far into the leaf we'll push items.  The
3855  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us do all the
3856  * items
3857  */
3858 static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
3859                                      int empty, struct extent_buffer *left,
3860                                      int free_space, u32 right_nritems,
3861                                      u32 max_slot)
3862 {
3863         struct btrfs_fs_info *fs_info = left->fs_info;
3864         struct btrfs_disk_key disk_key;
3865         struct extent_buffer *right = path->nodes[0];
3866         int i;
3867         int push_space = 0;
3868         int push_items = 0;
3869         struct btrfs_item *item;
3870         u32 old_left_nritems;
3871         u32 nr;
3872         int ret = 0;
3873         u32 this_item_size;
3874         u32 old_left_item_size;
3875         struct btrfs_map_token token;
3876 
3877         if (empty)
3878                 nr = min(right_nritems, max_slot);
3879         else
3880                 nr = min(right_nritems - 1, max_slot);
3881 
3882         for (i = 0; i < nr; i++) {
3883                 item = btrfs_item_nr(i);
3884 
3885                 if (!empty && push_items > 0) {
3886                         if (path->slots[0] < i)
3887                                 break;
3888                         if (path->slots[0] == i) {
3889                                 int space = btrfs_leaf_free_space(right);
3890 
3891                                 if (space + push_space * 2 > free_space)
3892                                         break;
3893                         }
3894                 }
3895 
3896                 if (path->slots[0] == i)
3897                         push_space += data_size;
3898 
3899                 this_item_size = btrfs_item_size(right, item);
3900                 if (this_item_size + sizeof(*item) + push_space > free_space)
3901                         break;
3902 
3903                 push_items++;
3904                 push_space += this_item_size + sizeof(*item);
3905         }
3906 
3907         if (push_items == 0) {
3908                 ret = 1;
3909                 goto out;
3910         }
3911         WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3912 
3913         /* push data from right to left */
3914         copy_extent_buffer(left, right,
3915                            btrfs_item_nr_offset(btrfs_header_nritems(left)),
3916                            btrfs_item_nr_offset(0),
3917                            push_items * sizeof(struct btrfs_item));
3918 
3919         push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3920                      btrfs_item_offset_nr(right, push_items - 1);
3921 
3922         copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3923                      leaf_data_end(left) - push_space,
3924                      BTRFS_LEAF_DATA_OFFSET +
3925                      btrfs_item_offset_nr(right, push_items - 1),
3926                      push_space);
3927         old_left_nritems = btrfs_header_nritems(left);
3928         BUG_ON(old_left_nritems <= 0);
3929 
3930         btrfs_init_map_token(&token, left);
3931         old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3932         for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3933                 u32 ioff;
3934 
3935                 item = btrfs_item_nr(i);
3936 
3937                 ioff = btrfs_token_item_offset(left, item, &token);
3938                 btrfs_set_token_item_offset(left, item,
3939                       ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3940                       &token);
3941         }
3942         btrfs_set_header_nritems(left, old_left_nritems + push_items);
3943 
3944         /* fixup right node */
3945         if (push_items > right_nritems)
3946                 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3947                        right_nritems);
3948 
3949         if (push_items < right_nritems) {
3950                 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3951                                                   leaf_data_end(right);
3952                 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3953                                       BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3954                                       BTRFS_LEAF_DATA_OFFSET +
3955                                       leaf_data_end(right), push_space);
3956 
3957                 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3958                               btrfs_item_nr_offset(push_items),
3959                              (btrfs_header_nritems(right) - push_items) *
3960                              sizeof(struct btrfs_item));
3961         }
3962 
3963         btrfs_init_map_token(&token, right);
3964         right_nritems -= push_items;
3965         btrfs_set_header_nritems(right, right_nritems);
3966         push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3967         for (i = 0; i < right_nritems; i++) {
3968                 item = btrfs_item_nr(i);
3969 
3970                 push_space = push_space - btrfs_token_item_size(right,
3971                                                                 item, &token);
3972                 btrfs_set_token_item_offset(right, item, push_space, &token);
3973         }
3974 
3975         btrfs_mark_buffer_dirty(left);
3976         if (right_nritems)
3977                 btrfs_mark_buffer_dirty(right);
3978         else
3979                 btrfs_clean_tree_block(right);
3980 
3981         btrfs_item_key(right, &disk_key, 0);
3982         fixup_low_keys(path, &disk_key, 1);
3983 
3984         /* then fixup the leaf pointer in the path */
3985         if (path->slots[0] < push_items) {
3986                 path->slots[0] += old_left_nritems;
3987                 btrfs_tree_unlock(path->nodes[0]);
3988                 free_extent_buffer(path->nodes[0]);
3989                 path->nodes[0] = left;
3990                 path->slots[1] -= 1;
3991         } else {
3992                 btrfs_tree_unlock(left);
3993                 free_extent_buffer(left);
3994                 path->slots[0] -= push_items;
3995         }
3996         BUG_ON(path->slots[0] < 0);
3997         return ret;
3998 out:
3999         btrfs_tree_unlock(left);
4000         free_extent_buffer(left);
4001         return ret;
4002 }
4003 
4004 /*
4005  * push some data in the path leaf to the left, trying to free up at
4006  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
4007  *
4008  * max_slot can put a limit on how far into the leaf we'll push items.  The
4009  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us push all the
4010  * items
4011  */
4012 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
4013                           *root, struct btrfs_path *path, int min_data_size,
4014                           int data_size, int empty, u32 max_slot)
4015 {
4016         struct extent_buffer *right = path->nodes[0];
4017         struct extent_buffer *left;
4018         int slot;
4019         int free_space;
4020         u32 right_nritems;
4021         int ret = 0;
4022 
4023         slot = path->slots[1];
4024         if (slot == 0)
4025                 return 1;
4026         if (!path->nodes[1])
4027                 return 1;
4028 
4029         right_nritems = btrfs_header_nritems(right);
4030         if (right_nritems == 0)
4031                 return 1;
4032 
4033         btrfs_assert_tree_locked(path->nodes[1]);
4034 
4035         left = btrfs_read_node_slot(path->nodes[1], slot - 1);
4036         /*
4037          * slot - 1 is not valid or we fail to read the left node,
4038          * no big deal, just return.
4039          */
4040         if (IS_ERR(left))
4041                 return 1;
4042 
4043         btrfs_tree_lock(left);
4044         btrfs_set_lock_blocking_write(left);
4045 
4046         free_space = btrfs_leaf_free_space(left);
4047         if (free_space < data_size) {
4048                 ret = 1;
4049                 goto out;
4050         }
4051 
4052         /* cow and double check */
4053         ret = btrfs_cow_block(trans, root, left,
4054                               path->nodes[1], slot - 1, &left);
4055         if (ret) {
4056                 /* we hit -ENOSPC, but it isn't fatal here */
4057                 if (ret == -ENOSPC)
4058                         ret = 1;
4059                 goto out;
4060         }
4061 
4062         free_space = btrfs_leaf_free_space(left);
4063         if (free_space < data_size) {
4064                 ret = 1;
4065                 goto out;
4066         }
4067 
4068         return __push_leaf_left(path, min_data_size,
4069                                empty, left, free_space, right_nritems,
4070                                max_slot);
4071 out:
4072         btrfs_tree_unlock(left);
4073         free_extent_buffer(left);
4074         return ret;
4075 }
4076 
4077 /*
4078  * split the path's leaf in two, making sure there is at least data_size
4079  * available for the resulting leaf level of the path.
4080  */
4081 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4082                                     struct btrfs_path *path,
4083                                     struct extent_buffer *l,
4084                                     struct extent_buffer *right,
4085                                     int slot, int mid, int nritems)
4086 {
4087         struct btrfs_fs_info *fs_info = trans->fs_info;
4088         int data_copy_size;
4089         int rt_data_off;
4090         int i;
4091         struct btrfs_disk_key disk_key;
4092         struct btrfs_map_token token;
4093 
4094         nritems = nritems - mid;
4095         btrfs_set_header_nritems(right, nritems);
4096         data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l);
4097 
4098         copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4099                            btrfs_item_nr_offset(mid),
4100                            nritems * sizeof(struct btrfs_item));
4101 
4102         copy_extent_buffer(right, l,
4103                      BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4104                      data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4105                      leaf_data_end(l), data_copy_size);
4106 
4107         rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4108 
4109         btrfs_init_map_token(&token, right);
4110         for (i = 0; i < nritems; i++) {
4111                 struct btrfs_item *item = btrfs_item_nr(i);
4112                 u32 ioff;
4113 
4114                 ioff = btrfs_token_item_offset(right, item, &token);
4115                 btrfs_set_token_item_offset(right, item,
4116                                             ioff + rt_data_off, &token);
4117         }
4118 
4119         btrfs_set_header_nritems(l, mid);
4120         btrfs_item_key(right, &disk_key, 0);
4121         insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
4122 
4123         btrfs_mark_buffer_dirty(right);
4124         btrfs_mark_buffer_dirty(l);
4125         BUG_ON(path->slots[0] != slot);
4126 
4127         if (mid <= slot) {
4128                 btrfs_tree_unlock(path->nodes[0]);
4129                 free_extent_buffer(path->nodes[0]);
4130                 path->nodes[0] = right;
4131                 path->slots[0] -= mid;
4132                 path->slots[1] += 1;
4133         } else {
4134                 btrfs_tree_unlock(right);
4135                 free_extent_buffer(right);
4136         }
4137 
4138         BUG_ON(path->slots[0] < 0);
4139 }
4140 
4141 /*
4142  * double splits happen when we need to insert a big item in the middle
4143  * of a leaf.  A double split can leave us with 3 mostly empty leaves:
4144  * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4145  *          A                 B                 C
4146  *
4147  * We avoid this by trying to push the items on either side of our target
4148  * into the adjacent leaves.  If all goes well we can avoid the double split
4149  * completely.
4150  */
4151 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4152                                           struct btrfs_root *root,
4153                                           struct btrfs_path *path,
4154                                           int data_size)
4155 {
4156         int ret;
4157         int progress = 0;
4158         int slot;
4159         u32 nritems;
4160         int space_needed = data_size;
4161 
4162         slot = path->slots[0];
4163         if (slot < btrfs_header_nritems(path->nodes[0]))
4164                 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4165 
4166         /*
4167          * try to push all the items after our slot into the
4168          * right leaf
4169          */
4170         ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4171         if (ret < 0)
4172                 return ret;
4173 
4174         if (ret == 0)
4175                 progress++;
4176 
4177         nritems = btrfs_header_nritems(path->nodes[0]);
4178         /*
4179          * our goal is to get our slot at the start or end of a leaf.  If
4180          * we've done so we're done
4181          */
4182         if (path->slots[0] == 0 || path->slots[0] == nritems)
4183                 return 0;
4184 
4185         if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4186                 return 0;
4187 
4188         /* try to push all the items before our slot into the next leaf */
4189         slot = path->slots[0];
4190         space_needed = data_size;
4191         if (slot > 0)
4192                 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4193         ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4194         if (ret < 0)
4195                 return ret;
4196 
4197         if (ret == 0)
4198                 progress++;
4199 
4200         if (progress)
4201                 return 0;
4202         return 1;
4203 }
4204 
4205 /*
4206  * split the path's leaf in two, making sure there is at least data_size
4207  * available for the resulting leaf level of the path.
4208  *
4209  * returns 0 if all went well and < 0 on failure.
4210  */
4211 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4212                                struct btrfs_root *root,
4213                                const struct btrfs_key *ins_key,
4214                                struct btrfs_path *path, int data_size,
4215                                int extend)
4216 {
4217         struct btrfs_disk_key disk_key;
4218         struct extent_buffer *l;
4219         u32 nritems;
4220         int mid;
4221         int slot;
4222         struct extent_buffer *right;
4223         struct btrfs_fs_info *fs_info = root->fs_info;
4224         int ret = 0;
4225         int wret;
4226         int split;
4227         int num_doubles = 0;
4228         int tried_avoid_double = 0;
4229 
4230         l = path->nodes[0];
4231         slot = path->slots[0];
4232         if (extend && data_size + btrfs_item_size_nr(l, slot) +
4233             sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4234                 return -EOVERFLOW;
4235 
4236         /* first try to make some room by pushing left and right */
4237         if (data_size && path->nodes[1]) {
4238                 int space_needed = data_size;
4239 
4240                 if (slot < btrfs_header_nritems(l))
4241                         space_needed -= btrfs_leaf_free_space(l);
4242 
4243                 wret = push_leaf_right(trans, root, path, space_needed,
4244                                        space_needed, 0, 0);
4245                 if (wret < 0)
4246                         return wret;
4247                 if (wret) {
4248                         space_needed = data_size;
4249                         if (slot > 0)
4250                                 space_needed -= btrfs_leaf_free_space(l);
4251                         wret = push_leaf_left(trans, root, path, space_needed,
4252                                               space_needed, 0, (u32)-1);
4253                         if (wret < 0)
4254                                 return wret;
4255                 }
4256                 l = path->nodes[0];
4257 
4258                 /* did the pushes work? */
4259                 if (btrfs_leaf_free_space(l) >= data_size)
4260                         return 0;
4261         }
4262 
4263         if (!path->nodes[1]) {
4264                 ret = insert_new_root(trans, root, path, 1);
4265                 if (ret)
4266                         return ret;
4267         }
4268 again:
4269         split = 1;
4270         l = path->nodes[0];
4271         slot = path->slots[0];
4272         nritems = btrfs_header_nritems(l);
4273         mid = (nritems + 1) / 2;
4274 
4275         if (mid <= slot) {
4276                 if (nritems == 1 ||
4277                     leaf_space_used(l, mid, nritems - mid) + data_size >
4278                         BTRFS_LEAF_DATA_SIZE(fs_info)) {
4279                         if (slot >= nritems) {
4280                                 split = 0;
4281                         } else {
4282                                 mid = slot;
4283                                 if (mid != nritems &&
4284                                     leaf_space_used(l, mid, nritems - mid) +
4285                                     data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4286                                         if (data_size && !tried_avoid_double)
4287                                                 goto push_for_double;
4288                                         split = 2;
4289                                 }
4290                         }
4291                 }
4292         } else {
4293                 if (leaf_space_used(l, 0, mid) + data_size >
4294                         BTRFS_LEAF_DATA_SIZE(fs_info)) {
4295                         if (!extend && data_size && slot == 0) {
4296                                 split = 0;
4297                         } else if ((extend || !data_size) && slot == 0) {
4298                                 mid = 1;
4299                         } else {
4300                                 mid = slot;
4301                                 if (mid != nritems &&
4302                                     leaf_space_used(l, mid, nritems - mid) +
4303                                     data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4304                                         if (data_size && !tried_avoid_double)
4305                                                 goto push_for_double;
4306                                         split = 2;
4307                                 }
4308                         }
4309                 }
4310         }
4311 
4312         if (split == 0)
4313                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4314         else
4315                 btrfs_item_key(l, &disk_key, mid);
4316 
4317         right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
4318                                              l->start, 0);
4319         if (IS_ERR(right))
4320                 return PTR_ERR(right);
4321 
4322         root_add_used(root, fs_info->nodesize);
4323 
4324         if (split == 0) {
4325                 if (mid <= slot) {
4326                         btrfs_set_header_nritems(right, 0);
4327                         insert_ptr(trans, path, &disk_key,
4328                                    right->start, path->slots[1] + 1, 1);
4329                         btrfs_tree_unlock(path->nodes[0]);
4330                         free_extent_buffer(path->nodes[0]);
4331                         path->nodes[0] = right;
4332                         path->slots[0] = 0;
4333                         path->slots[1] += 1;
4334                 } else {
4335                         btrfs_set_header_nritems(right, 0);
4336                         insert_ptr(trans, path, &disk_key,
4337                                    right->start, path->slots[1], 1);
4338                         btrfs_tree_unlock(path->nodes[0]);
4339                         free_extent_buffer(path->nodes[0]);
4340                         path->nodes[0] = right;
4341                         path->slots[0] = 0;
4342                         if (path->slots[1] == 0)
4343                                 fixup_low_keys(path, &disk_key, 1);
4344                 }
4345                 /*
4346                  * We create a new leaf 'right' for the required ins_len and
4347                  * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4348                  * the content of ins_len to 'right'.
4349                  */
4350                 return ret;
4351         }
4352 
4353         copy_for_split(trans, path, l, right, slot, mid, nritems);
4354 
4355         if (split == 2) {
4356                 BUG_ON(num_doubles != 0);
4357                 num_doubles++;
4358                 goto again;
4359         }
4360 
4361         return 0;
4362 
4363 push_for_double:
4364         push_for_double_split(trans, root, path, data_size);
4365         tried_avoid_double = 1;
4366         if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4367                 return 0;
4368         goto again;
4369 }
4370 
4371 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4372                                          struct btrfs_root *root,
4373                                          struct btrfs_path *path, int ins_len)
4374 {
4375         struct btrfs_key key;
4376         struct extent_buffer *leaf;
4377         struct btrfs_file_extent_item *fi;
4378         u64 extent_len = 0;
4379         u32 item_size;
4380         int ret;
4381 
4382         leaf = path->nodes[0];
4383         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4384 
4385         BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4386                key.type != BTRFS_EXTENT_CSUM_KEY);
4387 
4388         if (btrfs_leaf_free_space(leaf) >= ins_len)
4389                 return 0;
4390 
4391         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4392         if (key.type == BTRFS_EXTENT_DATA_KEY) {
4393                 fi = btrfs_item_ptr(leaf, path->slots[0],
4394                                     struct btrfs_file_extent_item);
4395                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4396         }
4397         btrfs_release_path(path);
4398 
4399         path->keep_locks = 1;
4400         path->search_for_split = 1;
4401         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4402         path->search_for_split = 0;
4403         if (ret > 0)
4404                 ret = -EAGAIN;
4405         if (ret < 0)
4406                 goto err;
4407 
4408         ret = -EAGAIN;
4409         leaf = path->nodes[0];
4410         /* if our item isn't there, return now */
4411         if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4412                 goto err;
4413 
4414         /* the leaf has  changed, it now has room.  return now */
4415         if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
4416                 goto err;
4417 
4418         if (key.type == BTRFS_EXTENT_DATA_KEY) {
4419                 fi = btrfs_item_ptr(leaf, path->slots[0],
4420                                     struct btrfs_file_extent_item);
4421                 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4422                         goto err;
4423         }
4424 
4425         btrfs_set_path_blocking(path);
4426         ret = split_leaf(trans, root, &key, path, ins_len, 1);
4427         if (ret)
4428                 goto err;
4429 
4430         path->keep_locks = 0;
4431         btrfs_unlock_up_safe(path, 1);
4432         return 0;
4433 err:
4434         path->keep_locks = 0;
4435         return ret;
4436 }
4437 
4438 static noinline int split_item(struct btrfs_path *path,
4439                                const struct btrfs_key *new_key,
4440                                unsigned long split_offset)
4441 {
4442         struct extent_buffer *leaf;
4443         struct btrfs_item *item;
4444         struct btrfs_item *new_item;
4445         int slot;
4446         char *buf;
4447         u32 nritems;
4448         u32 item_size;
4449         u32 orig_offset;
4450         struct btrfs_disk_key disk_key;
4451 
4452         leaf = path->nodes[0];
4453         BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
4454 
4455         btrfs_set_path_blocking(path);
4456 
4457         item = btrfs_item_nr(path->slots[0]);
4458         orig_offset = btrfs_item_offset(leaf, item);
4459         item_size = btrfs_item_size(leaf, item);
4460 
4461         buf = kmalloc(item_size, GFP_NOFS);
4462         if (!buf)
4463                 return -ENOMEM;
4464 
4465         read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4466                             path->slots[0]), item_size);
4467 
4468         slot = path->slots[0] + 1;
4469         nritems = btrfs_header_nritems(leaf);
4470         if (slot != nritems) {
4471                 /* shift the items */
4472                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4473                                 btrfs_item_nr_offset(slot),
4474                                 (nritems - slot) * sizeof(struct btrfs_item));
4475         }
4476 
4477         btrfs_cpu_key_to_disk(&disk_key, new_key);
4478         btrfs_set_item_key(leaf, &disk_key, slot);
4479 
4480         new_item = btrfs_item_nr(slot);
4481 
4482         btrfs_set_item_offset(leaf, new_item, orig_offset);
4483         btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4484 
4485         btrfs_set_item_offset(leaf, item,
4486                               orig_offset + item_size - split_offset);
4487         btrfs_set_item_size(leaf, item, split_offset);
4488 
4489         btrfs_set_header_nritems(leaf, nritems + 1);
4490 
4491         /* write the data for the start of the original item */
4492         write_extent_buffer(leaf, buf,
4493                             btrfs_item_ptr_offset(leaf, path->slots[0]),
4494                             split_offset);
4495 
4496         /* write the data for the new item */
4497         write_extent_buffer(leaf, buf + split_offset,
4498                             btrfs_item_ptr_offset(leaf, slot),
4499                             item_size - split_offset);
4500         btrfs_mark_buffer_dirty(leaf);
4501 
4502         BUG_ON(btrfs_leaf_free_space(leaf) < 0);
4503         kfree(buf);
4504         return 0;
4505 }
4506 
4507 /*
4508  * This function splits a single item into two items,
4509  * giving 'new_key' to the new item and splitting the
4510  * old one at split_offset (from the start of the item).
4511  *
4512  * The path may be released by this operation.  After
4513  * the split, the path is pointing to the old item.  The
4514  * new item is going to be in the same node as the old one.
4515  *
4516  * Note, the item being split must be smaller enough to live alone on
4517  * a tree block with room for one extra struct btrfs_item
4518  *
4519  * This allows us to split the item in place, keeping a lock on the
4520  * leaf the entire time.
4521  */
4522 int btrfs_split_item(struct btrfs_trans_handle *trans,
4523                      struct btrfs_root *root,
4524                      struct btrfs_path *path,
4525                      const struct btrfs_key *new_key,
4526                      unsigned long split_offset)
4527 {
4528         int ret;
4529         ret = setup_leaf_for_split(trans, root, path,
4530                                    sizeof(struct btrfs_item));
4531         if (ret)
4532                 return ret;
4533 
4534         ret = split_item(path, new_key, split_offset);
4535         return ret;
4536 }
4537 
4538 /*
4539  * This function duplicate a item, giving 'new_key' to the new item.
4540  * It guarantees both items live in the same tree leaf and the new item
4541  * is contiguous with the original item.
4542  *
4543  * This allows us to split file extent in place, keeping a lock on the
4544  * leaf the entire time.
4545  */
4546 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4547                          struct btrfs_root *root,
4548                          struct btrfs_path *path,
4549                          const struct btrfs_key *new_key)
4550 {
4551         struct extent_buffer *leaf;
4552         int ret;
4553         u32 item_size;
4554 
4555         leaf = path->nodes[0];
4556         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4557         ret = setup_leaf_for_split(trans, root, path,
4558                                    item_size + sizeof(struct btrfs_item));
4559         if (ret)
4560                 return ret;
4561 
4562         path->slots[0]++;
4563         setup_items_for_insert(root, path, new_key, &item_size,
4564                                item_size, item_size +
4565                                sizeof(struct btrfs_item), 1);
4566         leaf = path->nodes[0];
4567         memcpy_extent_buffer(leaf,
4568                              btrfs_item_ptr_offset(leaf, path->slots[0]),
4569                              btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4570                              item_size);
4571         return 0;
4572 }
4573 
4574 /*
4575  * make the item pointed to by the path smaller.  new_size indicates
4576  * how small to make it, and from_end tells us if we just chop bytes
4577  * off the end of the item or if we shift the item to chop bytes off
4578  * the front.
4579  */
4580 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
4581 {
4582         int slot;
4583         struct extent_buffer *leaf;
4584         struct btrfs_item *item;
4585         u32 nritems;
4586         unsigned int data_end;
4587         unsigned int old_data_start;
4588         unsigned int old_size;
4589         unsigned int size_diff;
4590         int i;
4591         struct btrfs_map_token token;
4592 
4593         leaf = path->nodes[0];
4594         slot = path->slots[0];
4595 
4596         old_size = btrfs_item_size_nr(leaf, slot);
4597         if (old_size == new_size)
4598                 return;
4599 
4600         nritems = btrfs_header_nritems(leaf);
4601         data_end = leaf_data_end(leaf);
4602 
4603         old_data_start = btrfs_item_offset_nr(leaf, slot);
4604 
4605         size_diff = old_size - new_size;
4606 
4607         BUG_ON(slot < 0);
4608         BUG_ON(slot >= nritems);
4609 
4610         /*
4611          * item0..itemN ... dataN.offset..dataN.size .. data0.size
4612          */
4613         /* first correct the data pointers */
4614         btrfs_init_map_token(&token, leaf);
4615         for (i = slot; i < nritems; i++) {
4616                 u32 ioff;
4617                 item = btrfs_item_nr(i);
4618 
4619                 ioff = btrfs_token_item_offset(leaf, item, &token);
4620                 btrfs_set_token_item_offset(leaf, item,
4621                                             ioff + size_diff, &token);
4622         }
4623 
4624         /* shift the data */
4625         if (from_end) {
4626                 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4627                               data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4628                               data_end, old_data_start + new_size - data_end);
4629         } else {
4630                 struct btrfs_disk_key disk_key;
4631                 u64 offset;
4632 
4633                 btrfs_item_key(leaf, &disk_key, slot);
4634 
4635                 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4636                         unsigned long ptr;
4637                         struct btrfs_file_extent_item *fi;
4638 
4639                         fi = btrfs_item_ptr(leaf, slot,
4640                                             struct btrfs_file_extent_item);
4641                         fi = (struct btrfs_file_extent_item *)(
4642                              (unsigned long)fi - size_diff);
4643 
4644                         if (btrfs_file_extent_type(leaf, fi) ==
4645                             BTRFS_FILE_EXTENT_INLINE) {
4646                                 ptr = btrfs_item_ptr_offset(leaf, slot);
4647                                 memmove_extent_buffer(leaf, ptr,
4648                                       (unsigned long)fi,
4649                                       BTRFS_FILE_EXTENT_INLINE_DATA_START);
4650                         }
4651                 }
4652 
4653                 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4654                               data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4655                               data_end, old_data_start - data_end);
4656 
4657                 offset = btrfs_disk_key_offset(&disk_key);
4658                 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4659                 btrfs_set_item_key(leaf, &disk_key, slot);
4660                 if (slot == 0)
4661                         fixup_low_keys(path, &disk_key, 1);
4662         }
4663 
4664         item = btrfs_item_nr(slot);
4665         btrfs_set_item_size(leaf, item, new_size);
4666         btrfs_mark_buffer_dirty(leaf);
4667 
4668         if (btrfs_leaf_free_space(leaf) < 0) {
4669                 btrfs_print_leaf(leaf);
4670                 BUG();
4671         }
4672 }
4673 
4674 /*
4675  * make the item pointed to by the path bigger, data_size is the added size.
4676  */
4677 void btrfs_extend_item(struct btrfs_path *path, u32 data_size)
4678 {
4679         int slot;
4680         struct extent_buffer *leaf;
4681         struct btrfs_item *item;
4682         u32 nritems;
4683         unsigned int data_end;
4684         unsigned int old_data;
4685         unsigned int old_size;
4686         int i;
4687         struct btrfs_map_token token;
4688 
4689         leaf = path->nodes[0];
4690 
4691         nritems = btrfs_header_nritems(leaf);
4692         data_end = leaf_data_end(leaf);
4693 
4694         if (btrfs_leaf_free_space(leaf) < data_size) {
4695                 btrfs_print_leaf(leaf);
4696                 BUG();
4697         }
4698         slot = path->slots[0];
4699         old_data = btrfs_item_end_nr(leaf, slot);
4700 
4701         BUG_ON(slot < 0);
4702         if (slot >= nritems) {
4703                 btrfs_print_leaf(leaf);
4704                 btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
4705                            slot, nritems);
4706                 BUG();
4707         }
4708 
4709         /*
4710          * item0..itemN ... dataN.offset..dataN.size .. data0.size
4711          */
4712         /* first correct the data pointers */
4713         btrfs_init_map_token(&token, leaf);
4714         for (i = slot; i < nritems; i++) {
4715                 u32 ioff;
4716                 item = btrfs_item_nr(i);
4717 
4718                 ioff = btrfs_token_item_offset(leaf, item, &token);
4719                 btrfs_set_token_item_offset(leaf, item,
4720                                             ioff - data_size, &token);
4721         }
4722 
4723         /* shift the data */
4724         memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4725                       data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4726                       data_end, old_data - data_end);
4727 
4728         data_end = old_data;
4729         old_size = btrfs_item_size_nr(leaf, slot);
4730         item = btrfs_item_nr(slot);
4731         btrfs_set_item_size(leaf, item, old_size + data_size);
4732         btrfs_mark_buffer_dirty(leaf);
4733 
4734         if (btrfs_leaf_free_space(leaf) < 0) {
4735                 btrfs_print_leaf(leaf);
4736                 BUG();
4737         }
4738 }
4739 
4740 /*
4741  * this is a helper for btrfs_insert_empty_items, the main goal here is
4742  * to save stack depth by doing the bulk of the work in a function
4743  * that doesn't call btrfs_search_slot
4744  */
4745 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4746                             const struct btrfs_key *cpu_key, u32 *data_size,
4747                             u32 total_data, u32 total_size, int nr)
4748 {
4749         struct btrfs_fs_info *fs_info = root->fs_info;
4750         struct btrfs_item *item;
4751         int i;
4752         u32 nritems;
4753         unsigned int data_end;
4754         struct btrfs_disk_key disk_key;
4755         struct extent_buffer *leaf;
4756         int slot;
4757         struct btrfs_map_token token;
4758 
4759         if (path->slots[0] == 0) {
4760                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4761                 fixup_low_keys(path, &disk_key, 1);
4762         }
4763         btrfs_unlock_up_safe(path, 1);
4764 
4765         leaf = path->nodes[0];
4766         slot = path->slots[0];
4767 
4768         nritems = btrfs_header_nritems(leaf);
4769         data_end = leaf_data_end(leaf);
4770 
4771         if (btrfs_leaf_free_space(leaf) < total_size) {
4772                 btrfs_print_leaf(leaf);
4773                 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4774                            total_size, btrfs_leaf_free_space(leaf));
4775                 BUG();
4776         }
4777 
4778         btrfs_init_map_token(&token, leaf);
4779         if (slot != nritems) {
4780                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4781 
4782                 if (old_data < data_end) {
4783                         btrfs_print_leaf(leaf);
4784                         btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4785                                    slot, old_data, data_end);
4786                         BUG();
4787                 }
4788                 /*
4789                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
4790                  */
4791                 /* first correct the data pointers */
4792                 for (i = slot; i < nritems; i++) {
4793                         u32 ioff;
4794 
4795                         item = btrfs_item_nr(i);
4796                         ioff = btrfs_token_item_offset(leaf, item, &token);
4797                         btrfs_set_token_item_offset(leaf, item,
4798                                                     ioff - total_data, &token);
4799                 }
4800                 /* shift the items */
4801                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4802                               btrfs_item_nr_offset(slot),
4803                               (nritems - slot) * sizeof(struct btrfs_item));
4804 
4805                 /* shift the data */
4806                 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4807                               data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4808                               data_end, old_data - data_end);
4809                 data_end = old_data;
4810         }
4811 
4812         /* setup the item for the new data */
4813         for (i = 0; i < nr; i++) {
4814                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4815                 btrfs_set_item_key(leaf, &disk_key, slot + i);
4816                 item = btrfs_item_nr(slot + i);
4817                 btrfs_set_token_item_offset(leaf, item,
4818                                             data_end - data_size[i], &token);
4819                 data_end -= data_size[i];
4820                 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4821         }
4822 
4823         btrfs_set_header_nritems(leaf, nritems + nr);
4824         btrfs_mark_buffer_dirty(leaf);
4825 
4826         if (btrfs_leaf_free_space(leaf) < 0) {
4827                 btrfs_print_leaf(leaf);
4828                 BUG();
4829         }
4830 }
4831 
4832 /*
4833  * Given a key and some data, insert items into the tree.
4834  * This does all the path init required, making room in the tree if needed.
4835  */
4836 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4837                             struct btrfs_root *root,
4838                             struct btrfs_path *path,
4839                             const struct btrfs_key *cpu_key, u32 *data_size,
4840                             int nr)
4841 {
4842         int ret = 0;
4843         int slot;
4844         int i;
4845         u32 total_size = 0;
4846         u32 total_data = 0;
4847 
4848         for (i = 0; i < nr; i++)
4849                 total_data += data_size[i];
4850 
4851         total_size = total_data + (nr * sizeof(struct btrfs_item));
4852         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4853         if (ret == 0)
4854                 return -EEXIST;
4855         if (ret < 0)
4856                 return ret;
4857 
4858         slot = path->slots[0];
4859         BUG_ON(slot < 0);
4860 
4861         setup_items_for_insert(root, path, cpu_key, data_size,
4862                                total_data, total_size, nr);
4863         return 0;
4864 }
4865 
4866 /*
4867  * Given a key and some data, insert an item into the tree.
4868  * This does all the path init required, making room in the tree if needed.
4869  */
4870 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4871                       const struct btrfs_key *cpu_key, void *data,
4872                       u32 data_size)
4873 {
4874         int ret = 0;
4875         struct btrfs_path *path;
4876         struct extent_buffer *leaf;
4877         unsigned long ptr;
4878 
4879         path = btrfs_alloc_path();
4880         if (!path)
4881                 return -ENOMEM;
4882         ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4883         if (!ret) {
4884                 leaf = path->nodes[0];
4885                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4886                 write_extent_buffer(leaf, data, ptr, data_size);
4887                 btrfs_mark_buffer_dirty(leaf);
4888         }
4889         btrfs_free_path(path);
4890         return ret;
4891 }
4892 
4893 /*
4894  * delete the pointer from a given node.
4895  *
4896  * the tree should have been previously balanced so the deletion does not
4897  * empty a node.
4898  */
4899 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4900                     int level, int slot)
4901 {
4902         struct extent_buffer *parent = path->nodes[level];
4903         u32 nritems;
4904         int ret;
4905 
4906         nritems = btrfs_header_nritems(parent);
4907         if (slot != nritems - 1) {
4908                 if (level) {
4909                         ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4910                                         nritems - slot - 1);
4911                         BUG_ON(ret < 0);
4912                 }
4913                 memmove_extent_buffer(parent,
4914                               btrfs_node_key_ptr_offset(slot),
4915                               btrfs_node_key_ptr_offset(slot + 1),
4916                               sizeof(struct btrfs_key_ptr) *
4917                               (nritems - slot - 1));
4918         } else if (level) {
4919                 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4920                                 GFP_NOFS);
4921                 BUG_ON(ret < 0);
4922         }
4923 
4924         nritems--;
4925         btrfs_set_header_nritems(parent, nritems);
4926         if (nritems == 0 && parent == root->node) {
4927                 BUG_ON(btrfs_header_level(root->node) != 1);
4928                 /* just turn the root into a leaf and break */
4929                 btrfs_set_header_level(root->node, 0);
4930         } else if (slot == 0) {
4931                 struct btrfs_disk_key disk_key;
4932 
4933                 btrfs_node_key(parent, &disk_key, 0);
4934                 fixup_low_keys(path, &disk_key, level + 1);
4935         }
4936         btrfs_mark_buffer_dirty(parent);
4937 }
4938 
4939 /*
4940  * a helper function to delete the leaf pointed to by path->slots[1] and
4941  * path->nodes[1].
4942  *
4943  * This deletes the pointer in path->nodes[1] and frees the leaf
4944  * block extent.  zero is returned if it all worked out, < 0 otherwise.
4945  *
4946  * The path must have already been setup for deleting the leaf, including
4947  * all the proper balancing.  path->nodes[1] must be locked.
4948  */
4949 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4950                                     struct btrfs_root *root,
4951                                     struct btrfs_path *path,
4952                                     struct extent_buffer *leaf)
4953 {
4954         WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4955         del_ptr(root, path, 1, path->slots[1]);
4956 
4957         /*
4958          * btrfs_free_extent is expensive, we want to make sure we
4959          * aren't holding any locks when we call it
4960          */
4961         btrfs_unlock_up_safe(path, 0);
4962 
4963         root_sub_used(root, leaf->len);
4964 
4965         extent_buffer_get(leaf);
4966         btrfs_free_tree_block(trans, root, leaf, 0, 1);
4967         free_extent_buffer_stale(leaf);
4968 }
4969 /*
4970  * delete the item at the leaf level in path.  If that empties
4971  * the leaf, remove it from the tree
4972  */
4973 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4974                     struct btrfs_path *path, int slot, int nr)
4975 {
4976         struct btrfs_fs_info *fs_info = root->fs_info;
4977         struct extent_buffer *leaf;
4978         struct btrfs_item *item;
4979         u32 last_off;
4980         u32 dsize = 0;
4981         int ret = 0;
4982         int wret;
4983         int i;
4984         u32 nritems;
4985 
4986         leaf = path->nodes[0];
4987         last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4988 
4989         for (i = 0; i < nr; i++)
4990                 dsize += btrfs_item_size_nr(leaf, slot + i);
4991 
4992         nritems = btrfs_header_nritems(leaf);
4993 
4994         if (slot + nr != nritems) {
4995                 int data_end = leaf_data_end(leaf);
4996                 struct btrfs_map_token token;
4997 
4998                 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4999                               data_end + dsize,
5000                               BTRFS_LEAF_DATA_OFFSET + data_end,
5001                               last_off - data_end);
5002 
5003                 btrfs_init_map_token(&token, leaf);
5004                 for (i = slot + nr; i < nritems; i++) {
5005                         u32 ioff;
5006 
5007                         item = btrfs_item_nr(i);
5008                         ioff = btrfs_token_item_offset(leaf, item, &token);
5009                         btrfs_set_token_item_offset(leaf, item,
5010                                                     ioff + dsize, &token);
5011                 }
5012 
5013                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
5014                               btrfs_item_nr_offset(slot + nr),
5015                               sizeof(struct btrfs_item) *
5016                               (nritems - slot - nr));
5017         }
5018         btrfs_set_header_nritems(leaf, nritems - nr);
5019         nritems -= nr;
5020 
5021         /* delete the leaf if we've emptied it */
5022         if (nritems == 0) {
5023                 if (leaf == root->node) {
5024                         btrfs_set_header_level(leaf, 0);
5025                 } else {
5026                         btrfs_set_path_blocking(path);
5027                         btrfs_clean_tree_block(leaf);
5028                         btrfs_del_leaf(trans, root, path, leaf);
5029                 }
5030         } else {
5031                 int used = leaf_space_used(leaf, 0, nritems);
5032                 if (slot == 0) {
5033                         struct btrfs_disk_key disk_key;
5034 
5035                         btrfs_item_key(leaf, &disk_key, 0);
5036                         fixup_low_keys(path, &disk_key, 1);
5037                 }
5038 
5039                 /* delete the leaf if it is mostly empty */
5040                 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
5041                         /* push_leaf_left fixes the path.
5042                          * make sure the path still points to our leaf
5043                          * for possible call to del_ptr below
5044                          */
5045                         slot = path->slots[1];
5046                         extent_buffer_get(leaf);
5047 
5048                         btrfs_set_path_blocking(path);
5049                         wret = push_leaf_left(trans, root, path, 1, 1,
5050                                               1, (u32)-1);
5051                         if (wret < 0 && wret != -ENOSPC)
5052                                 ret = wret;
5053 
5054                         if (path->nodes[0] == leaf &&
5055                             btrfs_header_nritems(leaf)) {
5056                                 wret = push_leaf_right(trans, root, path, 1,
5057                                                        1, 1, 0);
5058                                 if (wret < 0 && wret != -ENOSPC)
5059                                         ret = wret;
5060                         }
5061 
5062                         if (btrfs_header_nritems(leaf) == 0) {
5063                                 path->slots[1] = slot;
5064                                 btrfs_del_leaf(trans, root, path, leaf);
5065                                 free_extent_buffer(leaf);
5066                                 ret = 0;
5067                         } else {
5068                                 /* if we're still in the path, make sure
5069                                  * we're dirty.  Otherwise, one of the
5070                                  * push_leaf functions must have already
5071                                  * dirtied this buffer
5072                                  */
5073                                 if (path->nodes[0] == leaf)
5074                                         btrfs_mark_buffer_dirty(leaf);
5075                                 free_extent_buffer(leaf);
5076                         }
5077                 } else {
5078                         btrfs_mark_buffer_dirty(leaf);
5079                 }
5080         }
5081         return ret;
5082 }
5083 
5084 /*
5085  * search the tree again to find a leaf with lesser keys
5086  * returns 0 if it found something or 1 if there are no lesser leaves.
5087  * returns < 0 on io errors.
5088  *
5089  * This may release the path, and so you may lose any locks held at the
5090  * time you call it.
5091  */
5092 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5093 {
5094         struct btrfs_key key;
5095         struct btrfs_disk_key found_key;
5096         int ret;
5097 
5098         btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5099 
5100         if (key.offset > 0) {
5101                 key.offset--;
5102         } else if (key.type > 0) {
5103                 key.type--;
5104                 key.offset = (u64)-1;
5105         } else if (key.objectid > 0) {
5106                 key.objectid--;
5107                 key.type = (u8)-1;
5108                 key.offset = (u64)-1;
5109         } else {
5110                 return 1;
5111         }
5112 
5113         btrfs_release_path(path);
5114         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5115         if (ret < 0)
5116                 return ret;
5117         btrfs_item_key(path->nodes[0], &found_key, 0);
5118         ret = comp_keys(&found_key, &key);
5119         /*
5120          * We might have had an item with the previous key in the tree right
5121          * before we released our path. And after we released our path, that
5122          * item might have been pushed to the first slot (0) of the leaf we
5123          * were holding due to a tree balance. Alternatively, an item with the
5124          * previous key can exist as the only element of a leaf (big fat item).
5125          * Therefore account for these 2 cases, so that our callers (like
5126          * btrfs_previous_item) don't miss an existing item with a key matching
5127          * the previous key we computed above.
5128          */
5129         if (ret <= 0)
5130                 return 0;
5131         return 1;
5132 }
5133 
5134 /*
5135  * A helper function to walk down the tree starting at min_key, and looking
5136  * for nodes or leaves that are have a minimum transaction id.
5137  * This is used by the btree defrag code, and tree logging
5138  *
5139  * This does not cow, but it does stuff the starting key it finds back
5140  * into min_key, so you can call btrfs_search_slot with cow=1 on the
5141  * key and get a writable path.
5142  *
5143  * This honors path->lowest_level to prevent descent past a given level
5144  * of the tree.
5145  *
5146  * min_trans indicates the oldest transaction that you are interested
5147  * in walking through.  Any nodes or leaves older than min_trans are
5148  * skipped over (without reading them).
5149  *
5150  * returns zero if something useful was found, < 0 on error and 1 if there
5151  * was nothing in the tree that matched the search criteria.
5152  */
5153 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5154                          struct btrfs_path *path,
5155                          u64 min_trans)
5156 {
5157         struct extent_buffer *cur;
5158         struct btrfs_key found_key;
5159         int slot;
5160         int sret;
5161         u32 nritems;
5162         int level;
5163         int ret = 1;
5164         int keep_locks = path->keep_locks;
5165 
5166         path->keep_locks = 1;
5167 again:
5168         cur = btrfs_read_lock_root_node(root);
5169         level = btrfs_header_level(cur);
5170         WARN_ON(path->nodes[level]);
5171         path->nodes[level] = cur;
5172         path->locks[level] = BTRFS_READ_LOCK;
5173 
5174         if (btrfs_header_generation(cur) < min_trans) {
5175                 ret = 1;
5176                 goto out;
5177         }
5178         while (1) {
5179                 nritems = btrfs_header_nritems(cur);
5180                 level = btrfs_header_level(cur);
5181                 sret = btrfs_bin_search(cur, min_key, level, &slot);
5182                 if (sret < 0) {
5183                         ret = sret;
5184                         goto out;
5185                 }
5186 
5187                 /* at the lowest level, we're done, setup the path and exit */
5188                 if (level == path->lowest_level) {
5189                         if (slot >= nritems)
5190                                 goto find_next_key;
5191                         ret = 0;
5192                         path->slots[level] = slot;
5193                         btrfs_item_key_to_cpu(cur, &found_key, slot);
5194                         goto out;
5195                 }
5196                 if (sret && slot > 0)
5197                         slot--;
5198                 /*
5199                  * check this node pointer against the min_trans parameters.
5200                  * If it is too old, old, skip to the next one.
5201                  */
5202                 while (slot < nritems) {
5203                         u64 gen;
5204 
5205                         gen = btrfs_node_ptr_generation(cur, slot);
5206                         if (gen < min_trans) {
5207                                 slot++;
5208                                 continue;
5209                         }
5210                         break;
5211                 }
5212 find_next_key:
5213                 /*
5214                  * we didn't find a candidate key in this node, walk forward
5215                  * and find another one
5216                  */
5217                 if (slot >= nritems) {
5218                         path->slots[level] = slot;
5219                         btrfs_set_path_blocking(path);
5220                         sret = btrfs_find_next_key(root, path, min_key, level,
5221                                                   min_trans);
5222                         if (sret == 0) {
5223                                 btrfs_release_path(path);
5224                                 goto again;
5225                         } else {
5226                                 goto out;
5227                         }
5228                 }
5229                 /* save our key for returning back */
5230                 btrfs_node_key_to_cpu(cur, &found_key, slot);
5231                 path->slots[level] = slot;
5232                 if (level == path->lowest_level) {
5233                         ret = 0;
5234                         goto out;
5235                 }
5236                 btrfs_set_path_blocking(path);
5237                 cur = btrfs_read_node_slot(cur, slot);
5238                 if (IS_ERR(cur)) {
5239                         ret = PTR_ERR(cur);
5240                         goto out;
5241                 }
5242 
5243                 btrfs_tree_read_lock(cur);
5244 
5245                 path->locks[level - 1] = BTRFS_READ_LOCK;
5246                 path->nodes[level - 1] = cur;
5247                 unlock_up(path, level, 1, 0, NULL);
5248         }
5249 out:
5250         path->keep_locks = keep_locks;
5251         if (ret == 0) {
5252                 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5253                 btrfs_set_path_blocking(path);
5254                 memcpy(min_key, &found_key, sizeof(found_key));
5255         }
5256         return ret;
5257 }
5258 
5259 /*
5260  * this is similar to btrfs_next_leaf, but does not try to preserve
5261  * and fixup the path.  It looks for and returns the next key in the
5262  * tree based on the current path and the min_trans parameters.
5263  *
5264  * 0 is returned if another key is found, < 0 if there are any errors
5265  * and 1 is returned if there are no higher keys in the tree
5266  *
5267  * path->keep_locks should be set to 1 on the search made before
5268  * calling this function.
5269  */
5270 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5271                         struct btrfs_key *key, int level, u64 min_trans)
5272 {
5273         int slot;
5274         struct extent_buffer *c;
5275 
5276         WARN_ON(!path->keep_locks && !path->skip_locking);
5277         while (level < BTRFS_MAX_LEVEL) {
5278                 if (!path->nodes[level])
5279                         return 1;
5280 
5281                 slot = path->slots[level] + 1;
5282                 c = path->nodes[level];
5283 next:
5284                 if (slot >= btrfs_header_nritems(c)) {
5285                         int ret;
5286                         int orig_lowest;
5287                         struct btrfs_key cur_key;
5288                         if (level + 1 >= BTRFS_MAX_LEVEL ||
5289                             !path->nodes[level + 1])
5290                                 return 1;
5291 
5292                         if (path->locks[level + 1] || path->skip_locking) {
5293                                 level++;
5294                                 continue;
5295                         }
5296 
5297                         slot = btrfs_header_nritems(c) - 1;
5298                         if (level == 0)
5299                                 btrfs_item_key_to_cpu(c, &cur_key, slot);
5300                         else
5301                                 btrfs_node_key_to_cpu(c, &cur_key, slot);
5302 
5303                         orig_lowest = path->lowest_level;
5304                         btrfs_release_path(path);
5305                         path->lowest_level = level;
5306                         ret = btrfs_search_slot(NULL, root, &cur_key, path,
5307                                                 0, 0);
5308                         path->lowest_level = orig_lowest;
5309                         if (ret < 0)
5310                                 return ret;
5311 
5312                         c = path->nodes[level];
5313                         slot = path->slots[level];
5314                         if (ret == 0)
5315                                 slot++;
5316                         goto next;
5317                 }
5318 
5319                 if (level == 0)
5320                         btrfs_item_key_to_cpu(c, key, slot);
5321                 else {
5322                         u64 gen = btrfs_node_ptr_generation(c, slot);
5323 
5324                         if (gen < min_trans) {
5325                                 slot++;
5326                                 goto next;
5327                         }
5328                         btrfs_node_key_to_cpu(c, key, slot);
5329                 }
5330                 return 0;
5331         }
5332         return 1;
5333 }
5334 
5335 /*
5336  * search the tree again to find a leaf with greater keys
5337  * returns 0 if it found something or 1 if there are no greater leaves.
5338  * returns < 0 on io errors.
5339  */
5340 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5341 {
5342         return btrfs_next_old_leaf(root, path, 0);
5343 }
5344 
5345 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5346                         u64 time_seq)
5347 {
5348         int slot;
5349         int level;
5350         struct extent_buffer *c;
5351         struct extent_buffer *next;
5352         struct btrfs_key key;
5353         u32 nritems;
5354         int ret;
5355         int old_spinning = path->leave_spinning;
5356         int next_rw_lock = 0;
5357 
5358         nritems = btrfs_header_nritems(path->nodes[0]);
5359         if (nritems == 0)
5360                 return 1;
5361 
5362         btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5363 again:
5364         level = 1;
5365         next = NULL;
5366         next_rw_lock = 0;
5367         btrfs_release_path(path);
5368 
5369         path->keep_locks = 1;
5370         path->leave_spinning = 1;
5371 
5372         if (time_seq)
5373                 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5374         else
5375                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5376         path->keep_locks = 0;
5377 
5378         if (ret < 0)
5379                 return ret;
5380 
5381         nritems = btrfs_header_nritems(path->nodes[0]);
5382         /*
5383          * by releasing the path above we dropped all our locks.  A balance
5384          * could have added more items next to the key that used to be
5385          * at the very end of the block.  So, check again here and
5386          * advance the path if there are now more items available.
5387          */
5388         if (nritems > 0 && path->slots[0] < nritems - 1) {
5389                 if (ret == 0)
5390                         path->slots[0]++;
5391                 ret = 0;
5392                 goto done;
5393         }
5394         /*
5395          * So the above check misses one case:
5396          * - after releasing the path above, someone has removed the item that
5397          *   used to be at the very end of the block, and balance between leafs
5398          *   gets another one with bigger key.offset to replace it.
5399          *
5400          * This one should be returned as well, or we can get leaf corruption
5401          * later(esp. in __btrfs_drop_extents()).
5402          *
5403          * And a bit more explanation about this check,
5404          * with ret > 0, the key isn't found, the path points to the slot
5405          * where it should be inserted, so the path->slots[0] item must be the
5406          * bigger one.
5407          */
5408         if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5409                 ret = 0;
5410                 goto done;
5411         }
5412 
5413         while (level < BTRFS_MAX_LEVEL) {
5414                 if (!path->nodes[level]) {
5415                         ret = 1;
5416                         goto done;
5417                 }
5418 
5419                 slot = path->slots[level] + 1;
5420                 c = path->nodes[level];
5421                 if (slot >= btrfs_header_nritems(c)) {
5422                         level++;
5423                         if (level == BTRFS_MAX_LEVEL) {
5424                                 ret = 1;
5425                                 goto done;
5426                         }
5427                         continue;
5428                 }
5429 
5430                 if (next) {
5431                         btrfs_tree_unlock_rw(next, next_rw_lock);
5432                         free_extent_buffer(next);
5433                 }
5434 
5435                 next = c;
5436                 next_rw_lock = path->locks[level];
5437                 ret = read_block_for_search(root, path, &next, level,
5438                                             slot, &key);
5439                 if (ret == -EAGAIN)
5440                         goto again;
5441 
5442                 if (ret < 0) {
5443                         btrfs_release_path(path);
5444                         goto done;
5445                 }
5446 
5447                 if (!path->skip_locking) {
5448                         ret = btrfs_try_tree_read_lock(next);
5449                         if (!ret && time_seq) {
5450                                 /*
5451                                  * If we don't get the lock, we may be racing
5452                                  * with push_leaf_left, holding that lock while
5453                                  * itself waiting for the leaf we've currently
5454                                  * locked. To solve this situation, we give up
5455                                  * on our lock and cycle.
5456                                  */
5457                                 free_extent_buffer(next);
5458                                 btrfs_release_path(path);
5459                                 cond_resched();
5460                                 goto again;
5461                         }
5462                         if (!ret) {
5463                                 btrfs_set_path_blocking(path);
5464                                 btrfs_tree_read_lock(next);
5465                         }
5466                         next_rw_lock = BTRFS_READ_LOCK;
5467                 }
5468                 break;
5469         }
5470         path->slots[level] = slot;
5471         while (1) {
5472                 level--;
5473                 c = path->nodes[level];
5474                 if (path->locks[level])
5475                         btrfs_tree_unlock_rw(c, path->locks[level]);
5476 
5477                 free_extent_buffer(c);
5478                 path->nodes[level] = next;
5479                 path->slots[level] = 0;
5480                 if (!path->skip_locking)
5481                         path->locks[level] = next_rw_lock;
5482                 if (!level)
5483                         break;
5484 
5485                 ret = read_block_for_search(root, path, &next, level,
5486                                             0, &key);
5487                 if (ret == -EAGAIN)
5488                         goto again;
5489 
5490                 if (ret < 0) {
5491                         btrfs_release_path(path);
5492                         goto done;
5493                 }
5494 
5495                 if (!path->skip_locking) {
5496                         ret = btrfs_try_tree_read_lock(next);
5497                         if (!ret) {
5498                                 btrfs_set_path_blocking(path);
5499                                 btrfs_tree_read_lock(next);
5500                         }
5501                         next_rw_lock = BTRFS_READ_LOCK;
5502                 }
5503         }
5504         ret = 0;
5505 done:
5506         unlock_up(path, 0, 1, 0, NULL);
5507         path->leave_spinning = old_spinning;
5508         if (!old_spinning)
5509                 btrfs_set_path_blocking(path);
5510 
5511         return ret;
5512 }
5513 
5514 /*
5515  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5516  * searching until it gets past min_objectid or finds an item of 'type'
5517  *
5518  * returns 0 if something is found, 1 if nothing was found and < 0 on error
5519  */
5520 int btrfs_previous_item(struct btrfs_root *root,
5521                         struct btrfs_path *path, u64 min_objectid,
5522                         int type)
5523 {
5524         struct btrfs_key found_key;
5525         struct extent_buffer *leaf;
5526         u32 nritems;
5527         int ret;
5528 
5529         while (1) {
5530                 if (path->slots[0] == 0) {
5531                         btrfs_set_path_blocking(path);
5532                         ret = btrfs_prev_leaf(root, path);
5533                         if (ret != 0)
5534                                 return ret;
5535                 } else {
5536                         path->slots[0]--;
5537                 }
5538                 leaf = path->nodes[0];
5539                 nritems = btrfs_header_nritems(leaf);
5540                 if (nritems == 0)
5541                         return 1;
5542                 if (path->slots[0] == nritems)
5543                         path->slots[0]--;
5544 
5545                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5546                 if (found_key.objectid < min_objectid)
5547                         break;
5548                 if (found_key.type == type)
5549                         return 0;
5550                 if (found_key.objectid == min_objectid &&
5551                     found_key.type < type)
5552                         break;
5553         }
5554         return 1;
5555 }
5556 
5557 /*
5558  * search in extent tree to find a previous Metadata/Data extent item with
5559  * min objecitd.
5560  *
5561  * returns 0 if something is found, 1 if nothing was found and < 0 on error
5562  */
5563 int btrfs_previous_extent_item(struct btrfs_root *root,
5564                         struct btrfs_path *path, u64 min_objectid)
5565 {
5566         struct btrfs_key found_key;
5567         struct extent_buffer *leaf;
5568         u32 nritems;
5569         int ret;
5570 
5571         while (1) {
5572                 if (path->slots[0] == 0) {
5573                         btrfs_set_path_blocking(path);
5574                         ret = btrfs_prev_leaf(root, path);
5575                         if (ret != 0)
5576                                 return ret;
5577                 } else {
5578                         path->slots[0]--;
5579                 }
5580                 leaf = path->nodes[0];
5581                 nritems = btrfs_header_nritems(leaf);
5582                 if (nritems == 0)
5583                         return 1;
5584                 if (path->slots[0] == nritems)
5585                         path->slots[0]--;
5586 
5587                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5588                 if (found_key.objectid < min_objectid)
5589                         break;
5590                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5591                     found_key.type == BTRFS_METADATA_ITEM_KEY)
5592                         return 0;
5593                 if (found_key.objectid == min_objectid &&
5594                     found_key.type < BTRFS_EXTENT_ITEM_KEY)
5595                         break;
5596         }
5597         return 1;
5598 }