ctree.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include "kerncompat.h"
   4 #include "radix-tree.h"
   5 #include "ctree.h"
   6 #include "disk-io.h"
   7 #include "print-tree.h"
   8
   9 static int split_node(struct btrfs_root *root, struct btrfs_path *path,
  10                       int level);
  11 static int split_leaf(struct btrfs_root *root, struct btrfs_path *path,
  12                       int data_size);
  13 static int push_node_left(struct btrfs_root *root, struct btrfs_buffer *dst,
  14                           struct btrfs_buffer *src);
  15 static int balance_node_right(struct btrfs_root *root,
  16                               struct btrfs_buffer *dst_buf,
  17                               struct btrfs_buffer *src_buf);
  18 static int del_ptr(struct btrfs_root *root, struct btrfs_path *path, int level,
  19                    int slot);
  20
  21 inline void btrfs_init_path(struct btrfs_path *p)
  22 {
  23         memset(p, 0, sizeof(*p));
  24 }
  25
  26 void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
  27 {
  28         int i;
  29         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
  30                 if (!p->nodes[i])
  31                         break;
  32                 btrfs_block_release(root, p->nodes[i]);
  33         }
  34         memset(p, 0, sizeof(*p));
  35 }
  36
  37 static int btrfs_cow_block(struct btrfs_root *root,
  38                     struct btrfs_buffer *buf,
  39                     struct btrfs_buffer *parent,
  40                     int parent_slot,
  41                     struct btrfs_buffer **cow_ret)
  42 {
  43         struct btrfs_buffer *cow;
  44
  45         if (!list_empty(&buf->dirty)) {
  46                 *cow_ret = buf;
  47                 return 0;
  48         }
  49         cow = btrfs_alloc_free_block(root);
  50         memcpy(&cow->node, &buf->node, root->blocksize);
  51         btrfs_set_header_blocknr(&cow->node.header, cow->blocknr);
  52         *cow_ret = cow;
  53         btrfs_inc_ref(root, buf);
  54         if (buf == root->node) {
  55                 root->node = cow;
  56                 cow->count++;
  57                 if (buf != root->commit_root)
  58                         btrfs_free_extent(root, buf->blocknr, 1);
  59                 btrfs_block_release(root, buf);
  60         } else {
  61                 btrfs_set_node_blockptr(&parent->node, parent_slot,
  62                                         cow->blocknr);
  63                 BUG_ON(list_empty(&parent->dirty));
  64                 btrfs_free_extent(root, buf->blocknr, 1);
  65         }
  66         btrfs_block_release(root, buf);
  67         return 0;
  68 }
  69
  70 /*
  71  * The leaf data grows from end-to-front in the node.
  72  * this returns the address of the start of the last item,
  73  * which is the stop of the leaf data stack
  74  */
  75 static inline unsigned int leaf_data_end(struct btrfs_root *root,
  76                                          struct btrfs_leaf *leaf)
  77 {
  78         u32 nr = btrfs_header_nritems(&leaf->header);
  79         if (nr == 0)
  80                 return BTRFS_LEAF_DATA_SIZE(root);
  81         return btrfs_item_offset(leaf->items + nr - 1);
  82 }
  83
  84 /*
  85  * The space between the end of the leaf items and
  86  * the start of the leaf data.  IOW, how much room
  87  * the leaf has left for both items and data
  88  */
  89 int btrfs_leaf_free_space(struct btrfs_root *root, struct btrfs_leaf *leaf)
  90 {
  91         int data_end = leaf_data_end(root, leaf);
  92         int nritems = btrfs_header_nritems(&leaf->header);
  93         char *items_end = (char *)(leaf->items + nritems + 1);
  94         return (char *)(btrfs_leaf_data(leaf) + data_end) - (char *)items_end;
  95 }
  96
  97 /*
  98  * compare two keys in a memcmp fashion
  99  */
 100 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
 101 {
 102         struct btrfs_key k1;
 103
 104         btrfs_disk_key_to_cpu(&k1, disk);
 105
 106         if (k1.objectid > k2->objectid)
 107                 return 1;
 108         if (k1.objectid < k2->objectid)
 109                 return -1;
 110         if (k1.offset > k2->offset)
 111                 return 1;
 112         if (k1.offset < k2->offset)
 113                 return -1;
 114         if (k1.flags > k2->flags)
 115                 return 1;
 116         if (k1.flags < k2->flags)
 117                 return -1;
 118         return 0;
 119 }
 120
 121 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
 122                       int level)
 123 {
 124         int i;
 125         struct btrfs_node *parent = NULL;
 126         struct btrfs_node *node = &path->nodes[level]->node;
 127         int parent_slot;
 128         u32 nritems = btrfs_header_nritems(&node->header);
 129
 130         if (path->nodes[level + 1])
 131                 parent = &path->nodes[level + 1]->node;
 132         parent_slot = path->slots[level + 1];
 133         BUG_ON(nritems == 0);
 134         if (parent) {
 135                 struct btrfs_disk_key *parent_key;
 136                 parent_key = &parent->ptrs[parent_slot].key;
 137                 BUG_ON(memcmp(parent_key, &node->ptrs[0].key,
 138                               sizeof(struct btrfs_disk_key)));
 139                 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
 140                        btrfs_header_blocknr(&node->header));
 141         }
 142         BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
 143         for (i = 0; nritems > 1 && i < nritems - 2; i++) {
 144                 struct btrfs_key cpukey;
 145                 btrfs_disk_key_to_cpu(&cpukey, &node->ptrs[i + 1].key);
 146                 BUG_ON(comp_keys(&node->ptrs[i].key, &cpukey) >= 0);
 147         }
 148         return 0;
 149 }
 150
 151 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
 152                       int level)
 153 {
 154         int i;
 155         struct btrfs_leaf *leaf = &path->nodes[level]->leaf;
 156         struct btrfs_node *parent = NULL;
 157         int parent_slot;
 158         u32 nritems = btrfs_header_nritems(&leaf->header);
 159
 160         if (path->nodes[level + 1])
 161                 parent = &path->nodes[level + 1]->node;
 162         parent_slot = path->slots[level + 1];
 163         BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
 164
 165         if (nritems == 0)
 166                 return 0;
 167
 168         if (parent) {
 169                 struct btrfs_disk_key *parent_key;
 170                 parent_key = &parent->ptrs[parent_slot].key;
 171                 BUG_ON(memcmp(parent_key, &leaf->items[0].key,
 172                        sizeof(struct btrfs_disk_key)));
 173                 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
 174                        btrfs_header_blocknr(&leaf->header));
 175         }
 176         for (i = 0; nritems > 1 && i < nritems - 2; i++) {
 177                 struct btrfs_key cpukey;
 178                 btrfs_disk_key_to_cpu(&cpukey, &leaf->items[i + 1].key);
 179                 BUG_ON(comp_keys(&leaf->items[i].key,
 180                                  &cpukey) >= 0);
 181                 BUG_ON(btrfs_item_offset(leaf->items + i) !=
 182                         btrfs_item_end(leaf->items + i + 1));
 183                 if (i == 0) {
 184                         BUG_ON(btrfs_item_offset(leaf->items + i) +
 185                                btrfs_item_size(leaf->items + i) !=
 186                                BTRFS_LEAF_DATA_SIZE(root));
 187                 }
 188         }
 189         return 0;
 190 }
 191
 192 static int check_block(struct btrfs_root *root, struct btrfs_path *path,
 193                         int level)
 194 {
 195         if (level == 0)
 196                 return check_leaf(root, path, level);
 197         return check_node(root, path, level);
 198 }
 199
 200 /*
 201  * search for key in the array p.  items p are item_size apart
 202  * and there are 'max' items in p
 203  * the slot in the array is returned via slot, and it points to
 204  * the place where you would insert key if it is not found in
 205  * the array.
 206  *
 207  * slot may point to max if the key is bigger than all of the keys
 208  */
 209 static int generic_bin_search(char *p, int item_size, struct btrfs_key *key,
 210                        int max, int *slot)
 211 {
 212         int low = 0;
 213         int high = max;
 214         int mid;
 215         int ret;
 216         struct btrfs_disk_key *tmp;
 217
 218         while(low < high) {
 219                 mid = (low + high) / 2;
 220                 tmp = (struct btrfs_disk_key *)(p + mid * item_size);
 221                 ret = comp_keys(tmp, key);
 222
 223                 if (ret < 0)
 224                         low = mid + 1;
 225                 else if (ret > 0)
 226                         high = mid;
 227                 else {
 228                         *slot = mid;
 229                         return 0;
 230                 }
 231         }
 232         *slot = low;
 233         return 1;
 234 }
 235
 236 /*
 237  * simple bin_search frontend that does the right thing for
 238  * leaves vs nodes
 239  */
 240 static int bin_search(struct btrfs_node *c, struct btrfs_key *key, int *slot)
 241 {
 242         if (btrfs_is_leaf(c)) {
 243                 struct btrfs_leaf *l = (struct btrfs_leaf *)c;
 244                 return generic_bin_search((void *)l->items,
 245                                           sizeof(struct btrfs_item),
 246                                           key, btrfs_header_nritems(&c->header),
 247                                           slot);
 248         } else {
 249                 return generic_bin_search((void *)c->ptrs,
 250                                           sizeof(struct btrfs_key_ptr),
 251                                           key, btrfs_header_nritems(&c->header),
 252                                           slot);
 253         }
 254         return -1;
 255 }
 256
 257 static struct btrfs_buffer *read_node_slot(struct btrfs_root *root,
 258                                    struct btrfs_buffer *parent_buf,
 259                                    int slot)
 260 {
 261         struct btrfs_node *node = &parent_buf->node;
 262         if (slot < 0)
 263                 return NULL;
 264         if (slot >= btrfs_header_nritems(&node->header))
 265                 return NULL;
 266         return read_tree_block(root, btrfs_node_blockptr(node, slot));
 267 }
 268
 269 static int balance_level(struct btrfs_root *root, struct btrfs_path *path,
 270                         int level)
 271 {
 272         struct btrfs_buffer *right_buf;
 273         struct btrfs_buffer *mid_buf;
 274         struct btrfs_buffer *left_buf;
 275         struct btrfs_buffer *parent_buf = NULL;
 276         struct btrfs_node *right = NULL;
 277         struct btrfs_node *mid;
 278         struct btrfs_node *left = NULL;
 279         struct btrfs_node *parent = NULL;
 280         int ret = 0;
 281         int wret;
 282         int pslot;
 283         int orig_slot = path->slots[level];
 284         u64 orig_ptr;
 285
 286         if (level == 0)
 287                 return 0;
 288
 289         mid_buf = path->nodes[level];
 290         mid = &mid_buf->node;
 291         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
 292
 293         if (level < BTRFS_MAX_LEVEL - 1)
 294                 parent_buf = path->nodes[level + 1];
 295         pslot = path->slots[level + 1];
 296
 297         if (!parent_buf) {
 298                 struct btrfs_buffer *child;
 299                 u64 blocknr = mid_buf->blocknr;
 300
 301                 if (btrfs_header_nritems(&mid->header) != 1)
 302                         return 0;
 303
 304                 /* promote the child to a root */
 305                 child = read_node_slot(root, mid_buf, 0);
 306                 BUG_ON(!child);
 307                 root->node = child;
 308                 path->nodes[level] = NULL;
 309                 /* once for the path */
 310                 btrfs_block_release(root, mid_buf);
 311                 /* once for the root ptr */
 312                 btrfs_block_release(root, mid_buf);
 313                 clean_tree_block(root, mid_buf);
 314                 return btrfs_free_extent(root, blocknr, 1);
 315         }
 316         parent = &parent_buf->node;
 317
 318         if (btrfs_header_nritems(&mid->header) >
 319             BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
 320                 return 0;
 321
 322         left_buf = read_node_slot(root, parent_buf, pslot - 1);
 323         right_buf = read_node_slot(root, parent_buf, pslot + 1);
 324
 325         /* first, try to make some room in the middle buffer */
 326         if (left_buf) {
 327                 btrfs_cow_block(root, left_buf, parent_buf,
 328                                 pslot - 1, &left_buf);
 329                 left = &left_buf->node;
 330                 orig_slot += btrfs_header_nritems(&left->header);
 331                 wret = push_node_left(root, left_buf, mid_buf);
 332                 if (wret < 0)
 333                         ret = wret;
 334         }
 335
 336         /*
 337          * then try to empty the right most buffer into the middle
 338          */
 339         if (right_buf) {
 340                 btrfs_cow_block(root, right_buf, parent_buf,
 341                                 pslot + 1, &right_buf);
 342                 right = &right_buf->node;
 343                 wret = push_node_left(root, mid_buf, right_buf);
 344                 if (wret < 0)
 345                         ret = wret;
 346                 if (btrfs_header_nritems(&right->header) == 0) {
 347                         u64 blocknr = right_buf->blocknr;
 348                         btrfs_block_release(root, right_buf);
 349                         clean_tree_block(root, right_buf);
 350                         right_buf = NULL;
 351                         right = NULL;
 352                         wret = del_ptr(root, path, level + 1, pslot + 1);
 353                         if (wret)
 354                                 ret = wret;
 355                         wret = btrfs_free_extent(root, blocknr, 1);
 356                         if (wret)
 357                                 ret = wret;
 358                 } else {
 359                         memcpy(&parent->ptrs[pslot + 1].key,
 360                                 &right->ptrs[0].key,
 361                                 sizeof(struct btrfs_disk_key));
 362                         BUG_ON(list_empty(&parent_buf->dirty));
 363                 }
 364         }
 365         if (btrfs_header_nritems(&mid->header) == 1) {
 366                 /*
 367                  * we're not allowed to leave a node with one item in the
 368                  * tree during a delete.  A deletion from lower in the tree
 369                  * could try to delete the only pointer in this node.
 370                  * So, pull some keys from the left.
 371                  * There has to be a left pointer at this point because
 372                  * otherwise we would have pulled some pointers from the
 373                  * right
 374                  */
 375                 BUG_ON(!left_buf);
 376                 wret = balance_node_right(root, mid_buf, left_buf);
 377                 if (wret < 0)
 378                         ret = wret;
 379                 BUG_ON(wret == 1);
 380         }
 381         if (btrfs_header_nritems(&mid->header) == 0) {
 382                 /* we've managed to empty the middle node, drop it */
 383                 u64 blocknr = mid_buf->blocknr;
 384                 btrfs_block_release(root, mid_buf);
 385                 clean_tree_block(root, mid_buf);
 386                 mid_buf = NULL;
 387                 mid = NULL;
 388                 wret = del_ptr(root, path, level + 1, pslot);
 389                 if (wret)
 390                         ret = wret;
 391                 wret = btrfs_free_extent(root, blocknr, 1);
 392                 if (wret)
 393                         ret = wret;
 394         } else {
 395                 /* update the parent key to reflect our changes */
 396                 memcpy(&parent->ptrs[pslot].key, &mid->ptrs[0].key,
 397                        sizeof(struct btrfs_disk_key));
 398                 BUG_ON(list_empty(&parent_buf->dirty));
 399         }
 400
 401         /* update the path */
 402         if (left_buf) {
 403                 if (btrfs_header_nritems(&left->header) > orig_slot) {
 404                         left_buf->count++; // released below
 405                         path->nodes[level] = left_buf;
 406                         path->slots[level + 1] -= 1;
 407                         path->slots[level] = orig_slot;
 408                         if (mid_buf)
 409                                 btrfs_block_release(root, mid_buf);
 410                 } else {
 411                         orig_slot -= btrfs_header_nritems(&left->header);
 412                         path->slots[level] = orig_slot;
 413                 }
 414         }
 415         /* double check we haven't messed things up */
 416         check_block(root, path, level);
 417         if (orig_ptr != btrfs_node_blockptr(&path->nodes[level]->node,
 418                                             path->slots[level]))
 419                 BUG();
 420
 421         if (right_buf)
 422                 btrfs_block_release(root, right_buf);
 423         if (left_buf)
 424                 btrfs_block_release(root, left_buf);
 425         return ret;
 426 }
 427
 428 /*
 429  * look for key in the tree.  path is filled in with nodes along the way
 430  * if key is found, we return zero and you can find the item in the leaf
 431  * level of the path (level 0)
 432  *
 433  * If the key isn't found, the path points to the slot where it should
 434  * be inserted, and 1 is returned.  If there are other errors during the
 435  * search a negative error number is returned.
 436  *
 437  * if ins_len > 0, nodes and leaves will be split as we walk down the
 438  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
 439  * possible)
 440  */
 441 int btrfs_search_slot(struct btrfs_root *root, struct btrfs_key *key,
 442                 struct btrfs_path *p, int ins_len, int cow)
 443 {
 444         struct btrfs_buffer *b;
 445         struct btrfs_buffer *cow_buf;
 446         struct btrfs_node *c;
 447         int slot;
 448         int ret;
 449         int level;
 450
 451 again:
 452         b = root->node;
 453         b->count++;
 454         while (b) {
 455                 level = btrfs_header_level(&b->node.header);
 456                 if (cow) {
 457                         int wret;
 458                         wret = btrfs_cow_block(root, b, p->nodes[level + 1],
 459                                                p->slots[level + 1], &cow_buf);
 460                         b = cow_buf;
 461                 }
 462                 BUG_ON(!cow && ins_len);
 463                 c = &b->node;
 464                 p->nodes[level] = b;
 465                 ret = check_block(root, p, level);
 466                 if (ret)
 467                         return -1;
 468                 ret = bin_search(c, key, &slot);
 469                 if (!btrfs_is_leaf(c)) {
 470                         if (ret && slot > 0)
 471                                 slot -= 1;
 472                         p->slots[level] = slot;
 473                         if (ins_len > 0 && btrfs_header_nritems(&c->header) ==
 474                             BTRFS_NODEPTRS_PER_BLOCK(root)) {
 475                                 int sret = split_node(root, p, level);
 476                                 BUG_ON(sret > 0);
 477                                 if (sret)
 478                                         return sret;
 479                                 b = p->nodes[level];
 480                                 c = &b->node;
 481                                 slot = p->slots[level];
 482                         } else if (ins_len < 0) {
 483                                 int sret = balance_level(root, p, level);
 484                                 if (sret)
 485                                         return sret;
 486                                 b = p->nodes[level];
 487                                 if (!b)
 488                                         goto again;
 489                                 c = &b->node;
 490                                 slot = p->slots[level];
 491                                 BUG_ON(btrfs_header_nritems(&c->header) == 1);
 492                         }
 493                         b = read_tree_block(root, btrfs_node_blockptr(c, slot));
 494                 } else {
 495                         struct btrfs_leaf *l = (struct btrfs_leaf *)c;
 496                         p->slots[level] = slot;
 497                         if (ins_len > 0 && btrfs_leaf_free_space(root, l) <
 498                             sizeof(struct btrfs_item) + ins_len) {
 499                                 int sret = split_leaf(root, p, ins_len);
 500                                 BUG_ON(sret > 0);
 501                                 if (sret)
 502                                         return sret;
 503                         }
 504                         BUG_ON(root->node->count == 1);
 505                         return ret;
 506                 }
 507         }
 508         BUG_ON(root->node->count == 1);
 509         return 1;
 510 }
 511
 512 /*
 513  * adjust the pointers going up the tree, starting at level
 514  * making sure the right key of each node is points to 'key'.
 515  * This is used after shifting pointers to the left, so it stops
 516  * fixing up pointers when a given leaf/node is not in slot 0 of the
 517  * higher levels
 518  *
 519  * If this fails to write a tree block, it returns -1, but continues
 520  * fixing up the blocks in ram so the tree is consistent.
 521  */
 522 static int fixup_low_keys(struct btrfs_root *root,
 523                            struct btrfs_path *path, struct btrfs_disk_key *key,
 524                            int level)
 525 {
 526         int i;
 527         int ret = 0;
 528         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
 529                 struct btrfs_node *t;
 530                 int tslot = path->slots[i];
 531                 if (!path->nodes[i])
 532                         break;
 533                 t = &path->nodes[i]->node;
 534                 memcpy(&t->ptrs[tslot].key, key, sizeof(*key));
 535                 BUG_ON(list_empty(&path->nodes[i]->dirty));
 536                 if (tslot != 0)
 537                         break;
 538         }
 539         return ret;
 540 }
 541
 542 /*
 543  * try to push data from one node into the next node left in the
 544  * tree.
 545  *
 546  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
 547  * error, and > 0 if there was no room in the left hand block.
 548  */
 549 static int push_node_left(struct btrfs_root *root, struct btrfs_buffer *dst_buf,
 550                           struct btrfs_buffer *src_buf)
 551 {
 552         struct btrfs_node *src = &src_buf->node;
 553         struct btrfs_node *dst = &dst_buf->node;
 554         int push_items = 0;
 555         int src_nritems;
 556         int dst_nritems;
 557         int ret = 0;
 558
 559         src_nritems = btrfs_header_nritems(&src->header);
 560         dst_nritems = btrfs_header_nritems(&dst->header);
 561         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
 562         if (push_items <= 0) {
 563                 return 1;
 564         }
 565
 566         if (src_nritems < push_items)
 567                 push_items = src_nritems;
 568
 569         memcpy(dst->ptrs + dst_nritems, src->ptrs,
 570                 push_items * sizeof(struct btrfs_key_ptr));
 571         if (push_items < src_nritems) {
 572                 memmove(src->ptrs, src->ptrs + push_items,
 573                         (src_nritems - push_items) *
 574                         sizeof(struct btrfs_key_ptr));
 575         }
 576         btrfs_set_header_nritems(&src->header, src_nritems - push_items);
 577         btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
 578         BUG_ON(list_empty(&src_buf->dirty));
 579         BUG_ON(list_empty(&dst_buf->dirty));
 580         return ret;
 581 }
 582
 583 /*
 584  * try to push data from one node into the next node right in the
 585  * tree.
 586  *
 587  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
 588  * error, and > 0 if there was no room in the right hand block.
 589  *
 590  * this will  only push up to 1/2 the contents of the left node over
 591  */
 592 static int balance_node_right(struct btrfs_root *root,
 593                               struct btrfs_buffer *dst_buf,
 594                               struct btrfs_buffer *src_buf)
 595 {
 596         struct btrfs_node *src = &src_buf->node;
 597         struct btrfs_node *dst = &dst_buf->node;
 598         int push_items = 0;
 599         int max_push;
 600         int src_nritems;
 601         int dst_nritems;
 602         int ret = 0;
 603
 604         src_nritems = btrfs_header_nritems(&src->header);
 605         dst_nritems = btrfs_header_nritems(&dst->header);
 606         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
 607         if (push_items <= 0) {
 608                 return 1;
 609         }
 610
 611         max_push = src_nritems / 2 + 1;
 612         /* don't try to empty the node */
 613         if (max_push > src_nritems)
 614                 return 1;
 615         if (max_push < push_items)
 616                 push_items = max_push;
 617
 618         memmove(dst->ptrs + push_items, dst->ptrs,
 619                 dst_nritems * sizeof(struct btrfs_key_ptr));
 620         memcpy(dst->ptrs, src->ptrs + src_nritems - push_items,
 621                 push_items * sizeof(struct btrfs_key_ptr));
 622
 623         btrfs_set_header_nritems(&src->header, src_nritems - push_items);
 624         btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
 625
 626         BUG_ON(list_empty(&src_buf->dirty));
 627         BUG_ON(list_empty(&dst_buf->dirty));
 628         return ret;
 629 }
 630
 631 /*
 632  * helper function to insert a new root level in the tree.
 633  * A new node is allocated, and a single item is inserted to
 634  * point to the existing root
 635  *
 636  * returns zero on success or < 0 on failure.
 637  */
 638 static int insert_new_root(struct btrfs_root *root,
 639                            struct btrfs_path *path, int level)
 640 {
 641         struct btrfs_buffer *t;
 642         struct btrfs_node *lower;
 643         struct btrfs_node *c;
 644         struct btrfs_disk_key *lower_key;
 645
 646         BUG_ON(path->nodes[level]);
 647         BUG_ON(path->nodes[level-1] != root->node);
 648
 649         t = btrfs_alloc_free_block(root);
 650         c = &t->node;
 651         memset(c, 0, root->blocksize);
 652         btrfs_set_header_nritems(&c->header, 1);
 653         btrfs_set_header_level(&c->header, level);
 654         btrfs_set_header_blocknr(&c->header, t->blocknr);
 655         btrfs_set_header_parentid(&c->header,
 656                                btrfs_header_parentid(&root->node->node.header));
 657         lower = &path->nodes[level-1]->node;
 658         if (btrfs_is_leaf(lower))
 659                 lower_key = &((struct btrfs_leaf *)lower)->items[0].key;
 660         else
 661                 lower_key = &lower->ptrs[0].key;
 662         memcpy(&c->ptrs[0].key, lower_key, sizeof(struct btrfs_disk_key));
 663         btrfs_set_node_blockptr(c, 0, path->nodes[level - 1]->blocknr);
 664         /* the super has an extra ref to root->node */
 665         btrfs_block_release(root, root->node);
 666         root->node = t;
 667         t->count++;
 668         path->nodes[level] = t;
 669         path->slots[level] = 0;
 670         return 0;
 671 }
 672
 673 /*
 674  * worker function to insert a single pointer in a node.
 675  * the node should have enough room for the pointer already
 676  *
 677  * slot and level indicate where you want the key to go, and
 678  * blocknr is the block the key points to.
 679  *
 680  * returns zero on success and < 0 on any error
 681  */
 682 static int insert_ptr(struct btrfs_root *root,
 683                 struct btrfs_path *path, struct btrfs_disk_key *key,
 684                 u64 blocknr, int slot, int level)
 685 {
 686         struct btrfs_node *lower;
 687         int nritems;
 688
 689         BUG_ON(!path->nodes[level]);
 690         lower = &path->nodes[level]->node;
 691         nritems = btrfs_header_nritems(&lower->header);
 692         if (slot > nritems)
 693                 BUG();
 694         if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
 695                 BUG();
 696         if (slot != nritems) {
 697                 memmove(lower->ptrs + slot + 1, lower->ptrs + slot,
 698                         (nritems - slot) * sizeof(struct btrfs_key_ptr));
 699         }
 700         memcpy(&lower->ptrs[slot].key, key, sizeof(struct btrfs_disk_key));
 701         btrfs_set_node_blockptr(lower, slot, blocknr);
 702         btrfs_set_header_nritems(&lower->header, nritems + 1);
 703         BUG_ON(list_empty(&path->nodes[level]->dirty));
 704         return 0;
 705 }
 706
 707 /*
 708  * split the node at the specified level in path in two.
 709  * The path is corrected to point to the appropriate node after the split
 710  *
 711  * Before splitting this tries to make some room in the node by pushing
 712  * left and right, if either one works, it returns right away.
 713  *
 714  * returns 0 on success and < 0 on failure
 715  */
 716 static int split_node(struct btrfs_root *root, struct btrfs_path *path,
 717                       int level)
 718 {
 719         struct btrfs_buffer *t;
 720         struct btrfs_node *c;
 721         struct btrfs_buffer *split_buffer;
 722         struct btrfs_node *split;
 723         int mid;
 724         int ret;
 725         int wret;
 726         u32 c_nritems;
 727
 728         t = path->nodes[level];
 729         c = &t->node;
 730         if (t == root->node) {
 731                 /* trying to split the root, lets make a new one */
 732                 ret = insert_new_root(root, path, level + 1);
 733                 if (ret)
 734                         return ret;
 735         }
 736         c_nritems = btrfs_header_nritems(&c->header);
 737         split_buffer = btrfs_alloc_free_block(root);
 738         split = &split_buffer->node;
 739         btrfs_set_header_flags(&split->header, btrfs_header_flags(&c->header));
 740         btrfs_set_header_blocknr(&split->header, split_buffer->blocknr);
 741         btrfs_set_header_parentid(&split->header,
 742                                btrfs_header_parentid(&root->node->node.header));
 743         mid = (c_nritems + 1) / 2;
 744         memcpy(split->ptrs, c->ptrs + mid,
 745                 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
 746         btrfs_set_header_nritems(&split->header, c_nritems - mid);
 747         btrfs_set_header_nritems(&c->header, mid);
 748         ret = 0;
 749
 750         BUG_ON(list_empty(&t->dirty));
 751         wret = insert_ptr(root, path, &split->ptrs[0].key,
 752                           split_buffer->blocknr, path->slots[level + 1] + 1,
 753                           level + 1);
 754         if (wret)
 755                 ret = wret;
 756
 757         if (path->slots[level] >= mid) {
 758                 path->slots[level] -= mid;
 759                 btrfs_block_release(root, t);
 760                 path->nodes[level] = split_buffer;
 761                 path->slots[level + 1] += 1;
 762         } else {
 763                 btrfs_block_release(root, split_buffer);
 764         }
 765         return ret;
 766 }
 767
 768 /*
 769  * how many bytes are required to store the items in a leaf.  start
 770  * and nr indicate which items in the leaf to check.  This totals up the
 771  * space used both by the item structs and the item data
 772  */
 773 static int leaf_space_used(struct btrfs_leaf *l, int start, int nr)
 774 {
 775         int data_len;
 776         int end = start + nr - 1;
 777
 778         if (!nr)
 779                 return 0;
 780         data_len = btrfs_item_end(l->items + start);
 781         data_len = data_len - btrfs_item_offset(l->items + end);
 782         data_len += sizeof(struct btrfs_item) * nr;
 783         return data_len;
 784 }
 785
 786 /*
 787  * push some data in the path leaf to the right, trying to free up at
 788  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
 789  *
 790  * returns 1 if the push failed because the other node didn't have enough
 791  * room, 0 if everything worked out and < 0 if there were major errors.
 792  */
 793 static int push_leaf_right(struct btrfs_root *root, struct btrfs_path *path,
 794                            int data_size)
 795 {
 796         struct btrfs_buffer *left_buf = path->nodes[0];
 797         struct btrfs_leaf *left = &left_buf->leaf;
 798         struct btrfs_leaf *right;
 799         struct btrfs_buffer *right_buf;
 800         struct btrfs_buffer *upper;
 801         int slot;
 802         int i;
 803         int free_space;
 804         int push_space = 0;
 805         int push_items = 0;
 806         struct btrfs_item *item;
 807         u32 left_nritems;
 808         u32 right_nritems;
 809
 810         slot = path->slots[1];
 811         if (!path->nodes[1]) {
 812                 return 1;
 813         }
 814         upper = path->nodes[1];
 815         if (slot >= btrfs_header_nritems(&upper->node.header) - 1) {
 816                 return 1;
 817         }
 818         right_buf = read_tree_block(root, btrfs_node_blockptr(&upper->node,
 819                                                               slot + 1));
 820         right = &right_buf->leaf;
 821         free_space = btrfs_leaf_free_space(root, right);
 822         if (free_space < data_size + sizeof(struct btrfs_item)) {
 823                 btrfs_block_release(root, right_buf);
 824                 return 1;
 825         }
 826         /* cow and double check */
 827         btrfs_cow_block(root, right_buf, upper, slot + 1, &right_buf);
 828         right = &right_buf->leaf;
 829         free_space = btrfs_leaf_free_space(root, right);
 830         if (free_space < data_size + sizeof(struct btrfs_item)) {
 831                 btrfs_block_release(root, right_buf);
 832                 return 1;
 833         }
 834
 835         left_nritems = btrfs_header_nritems(&left->header);
 836         for (i = left_nritems - 1; i >= 0; i--) {
 837                 item = left->items + i;
 838                 if (path->slots[0] == i)
 839                         push_space += data_size + sizeof(*item);
 840                 if (btrfs_item_size(item) + sizeof(*item) + push_space >
 841                     free_space)
 842                         break;
 843                 push_items++;
 844                 push_space += btrfs_item_size(item) + sizeof(*item);
 845         }
 846         if (push_items == 0) {
 847                 btrfs_block_release(root, right_buf);
 848                 return 1;
 849         }
 850         right_nritems = btrfs_header_nritems(&right->header);
 851         /* push left to right */
 852         push_space = btrfs_item_end(left->items + left_nritems - push_items);
 853         push_space -= leaf_data_end(root, left);
 854         /* make room in the right data area */
 855         memmove(btrfs_leaf_data(right) + leaf_data_end(root, right) -
 856                 push_space, btrfs_leaf_data(right) + leaf_data_end(root, right),
 857                 BTRFS_LEAF_DATA_SIZE(root) - leaf_data_end(root, right));
 858         /* copy from the left data area */
 859         memcpy(btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - push_space,
 860                 btrfs_leaf_data(left) + leaf_data_end(root, left), push_space);
 861         memmove(right->items + push_items, right->items,
 862                 right_nritems * sizeof(struct btrfs_item));
 863         /* copy the items from left to right */
 864         memcpy(right->items, left->items + left_nritems - push_items,
 865                 push_items * sizeof(struct btrfs_item));
 866
 867         /* update the item pointers */
 868         right_nritems += push_items;
 869         btrfs_set_header_nritems(&right->header, right_nritems);
 870         push_space = BTRFS_LEAF_DATA_SIZE(root);
 871         for (i = 0; i < right_nritems; i++) {
 872                 btrfs_set_item_offset(right->items + i, push_space -
 873                                       btrfs_item_size(right->items + i));
 874                 push_space = btrfs_item_offset(right->items + i);
 875         }
 876         left_nritems -= push_items;
 877         btrfs_set_header_nritems(&left->header, left_nritems);
 878
 879         BUG_ON(list_empty(&left_buf->dirty));
 880         BUG_ON(list_empty(&right_buf->dirty));
 881         memcpy(&upper->node.ptrs[slot + 1].key,
 882                 &right->items[0].key, sizeof(struct btrfs_disk_key));
 883         BUG_ON(list_empty(&upper->dirty));
 884
 885         /* then fixup the leaf pointer in the path */
 886         if (path->slots[0] >= left_nritems) {
 887                 path->slots[0] -= left_nritems;
 888                 btrfs_block_release(root, path->nodes[0]);
 889                 path->nodes[0] = right_buf;
 890                 path->slots[1] += 1;
 891         } else {
 892                 btrfs_block_release(root, right_buf);
 893         }
 894         return 0;
 895 }
 896 /*
 897  * push some data in the path leaf to the left, trying to free up at
 898  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
 899  */
 900 static int push_leaf_left(struct btrfs_root *root, struct btrfs_path *path,
 901                           int data_size)
 902 {
 903         struct btrfs_buffer *right_buf = path->nodes[0];
 904         struct btrfs_leaf *right = &right_buf->leaf;
 905         struct btrfs_buffer *t;
 906         struct btrfs_leaf *left;
 907         int slot;
 908         int i;
 909         int free_space;
 910         int push_space = 0;
 911         int push_items = 0;
 912         struct btrfs_item *item;
 913         u32 old_left_nritems;
 914         int ret = 0;
 915         int wret;
 916
 917         slot = path->slots[1];
 918         if (slot == 0) {
 919                 return 1;
 920         }
 921         if (!path->nodes[1]) {
 922                 return 1;
 923         }
 924         t = read_tree_block(root, btrfs_node_blockptr(&path->nodes[1]->node,
 925                                                       slot - 1));
 926         left = &t->leaf;
 927         free_space = btrfs_leaf_free_space(root, left);
 928         if (free_space < data_size + sizeof(struct btrfs_item)) {
 929                 btrfs_block_release(root, t);
 930                 return 1;
 931         }
 932
 933         /* cow and double check */
 934         btrfs_cow_block(root, t, path->nodes[1], slot - 1, &t);
 935         left = &t->leaf;
 936         free_space = btrfs_leaf_free_space(root, left);
 937         if (free_space < data_size + sizeof(struct btrfs_item)) {
 938                 btrfs_block_release(root, t);
 939                 return 1;
 940         }
 941
 942         for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
 943                 item = right->items + i;
 944                 if (path->slots[0] == i)
 945                         push_space += data_size + sizeof(*item);
 946                 if (btrfs_item_size(item) + sizeof(*item) + push_space >
 947                     free_space)
 948                         break;
 949                 push_items++;
 950                 push_space += btrfs_item_size(item) + sizeof(*item);
 951         }
 952         if (push_items == 0) {
 953                 btrfs_block_release(root, t);
 954                 return 1;
 955         }
 956         /* push data from right to left */
 957         memcpy(left->items + btrfs_header_nritems(&left->header),
 958                 right->items, push_items * sizeof(struct btrfs_item));
 959         push_space = BTRFS_LEAF_DATA_SIZE(root) -
 960                      btrfs_item_offset(right->items + push_items -1);
 961         memcpy(btrfs_leaf_data(left) + leaf_data_end(root, left) - push_space,
 962                 btrfs_leaf_data(right) +
 963                 btrfs_item_offset(right->items + push_items - 1),
 964                 push_space);
 965         old_left_nritems = btrfs_header_nritems(&left->header);
 966         BUG_ON(old_left_nritems < 0);
 967
 968         for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
 969                 u32 ioff = btrfs_item_offset(left->items + i);
 970                 btrfs_set_item_offset(left->items + i, ioff -
 971                                      (BTRFS_LEAF_DATA_SIZE(root) -
 972                                       btrfs_item_offset(left->items +
 973                                                         old_left_nritems - 1)));
 974         }
 975         btrfs_set_header_nritems(&left->header, old_left_nritems + push_items);
 976
 977         /* fixup right node */
 978         push_space = btrfs_item_offset(right->items + push_items - 1) -
 979                      leaf_data_end(root, right);
 980         memmove(btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
 981                 push_space, btrfs_leaf_data(right) +
 982                 leaf_data_end(root, right), push_space);
 983         memmove(right->items, right->items + push_items,
 984                 (btrfs_header_nritems(&right->header) - push_items) *
 985                 sizeof(struct btrfs_item));
 986         btrfs_set_header_nritems(&right->header,
 987                                  btrfs_header_nritems(&right->header) -
 988                                  push_items);
 989         push_space = BTRFS_LEAF_DATA_SIZE(root);
 990
 991         for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
 992                 btrfs_set_item_offset(right->items + i, push_space -
 993                                       btrfs_item_size(right->items + i));
 994                 push_space = btrfs_item_offset(right->items + i);
 995         }
 996
 997         BUG_ON(list_empty(&t->dirty));
 998         BUG_ON(list_empty(&right_buf->dirty));
 999
1000         wret = fixup_low_keys(root, path, &right->items[0].key, 1);
1001         if (wret)
1002                 ret = wret;
1003
1004         /* then fixup the leaf pointer in the path */
1005         if (path->slots[0] < push_items) {
1006                 path->slots[0] += old_left_nritems;
1007                 btrfs_block_release(root, path->nodes[0]);
1008                 path->nodes[0] = t;
1009                 path->slots[1] -= 1;
1010         } else {
1011                 btrfs_block_release(root, t);
1012                 path->slots[0] -= push_items;
1013         }
1014         BUG_ON(path->slots[0] < 0);
1015         return ret;
1016 }
1017
1018 /*
1019  * split the path's leaf in two, making sure there is at least data_size
1020  * available for the resulting leaf level of the path.
1021  *
1022  * returns 0 if all went well and < 0 on failure.
1023  */
1024 static int split_leaf(struct btrfs_root *root, struct btrfs_path *path,
1025                       int data_size)
1026 {
1027         struct btrfs_buffer *l_buf;
1028         struct btrfs_leaf *l;
1029         u32 nritems;
1030         int mid;
1031         int slot;
1032         struct btrfs_leaf *right;
1033         struct btrfs_buffer *right_buffer;
1034         int space_needed = data_size + sizeof(struct btrfs_item);
1035         int data_copy_size;
1036         int rt_data_off;
1037         int i;
1038         int ret;
1039         int wret;
1040
1041         wret = push_leaf_left(root, path, data_size);
1042         if (wret < 0)
1043                 return wret;
1044         if (wret) {
1045                 wret = push_leaf_right(root, path, data_size);
1046                 if (wret < 0)
1047                         return wret;
1048         }
1049         l_buf = path->nodes[0];
1050         l = &l_buf->leaf;
1051
1052         /* did the pushes work? */
1053         if (btrfs_leaf_free_space(root, l) >=
1054             sizeof(struct btrfs_item) + data_size)
1055                 return 0;
1056
1057         if (!path->nodes[1]) {
1058                 ret = insert_new_root(root, path, 1);
1059                 if (ret)
1060                         return ret;
1061         }
1062         slot = path->slots[0];
1063         nritems = btrfs_header_nritems(&l->header);
1064         mid = (nritems + 1)/ 2;
1065         right_buffer = btrfs_alloc_free_block(root);
1066         BUG_ON(!right_buffer);
1067         BUG_ON(mid == nritems);
1068         right = &right_buffer->leaf;
1069         memset(&right->header, 0, sizeof(right->header));
1070         if (mid <= slot) {
1071                 /* FIXME, just alloc a new leaf here */
1072                 if (leaf_space_used(l, mid, nritems - mid) + space_needed >
1073                         BTRFS_LEAF_DATA_SIZE(root))
1074                         BUG();
1075         } else {
1076                 /* FIXME, just alloc a new leaf here */
1077                 if (leaf_space_used(l, 0, mid + 1) + space_needed >
1078                         BTRFS_LEAF_DATA_SIZE(root))
1079                         BUG();
1080         }
1081         btrfs_set_header_nritems(&right->header, nritems - mid);
1082         btrfs_set_header_blocknr(&right->header, right_buffer->blocknr);
1083         btrfs_set_header_level(&right->header, 0);
1084         btrfs_set_header_parentid(&right->header,
1085                                btrfs_header_parentid(&root->node->node.header));
1086         data_copy_size = btrfs_item_end(l->items + mid) -
1087                          leaf_data_end(root, l);
1088         memcpy(right->items, l->items + mid,
1089                (nritems - mid) * sizeof(struct btrfs_item));
1090         memcpy(btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
1091                 data_copy_size, btrfs_leaf_data(l) +
1092                 leaf_data_end(root, l), data_copy_size);
1093         rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
1094                       btrfs_item_end(l->items + mid);
1095
1096         for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
1097                 u32 ioff = btrfs_item_offset(right->items + i);
1098                 btrfs_set_item_offset(right->items + i, ioff + rt_data_off);
1099         }
1100
1101         btrfs_set_header_nritems(&l->header, mid);
1102         ret = 0;
1103         wret = insert_ptr(root, path, &right->items[0].key,
1104                           right_buffer->blocknr, path->slots[1] + 1, 1);
1105         if (wret)
1106                 ret = wret;
1107         BUG_ON(list_empty(&right_buffer->dirty));
1108         BUG_ON(list_empty(&l_buf->dirty));
1109         BUG_ON(path->slots[0] != slot);
1110         if (mid <= slot) {
1111                 btrfs_block_release(root, path->nodes[0]);
1112                 path->nodes[0] = right_buffer;
1113                 path->slots[0] -= mid;
1114                 path->slots[1] += 1;
1115         } else
1116                 btrfs_block_release(root, right_buffer);
1117         BUG_ON(path->slots[0] < 0);
1118         return ret;
1119 }
1120
1121 /*
1122  * Given a key and some data, insert an item into the tree.
1123  * This does all the path init required, making room in the tree if needed.
1124  */
1125 int btrfs_insert_empty_item(struct btrfs_root *root, struct btrfs_path *path,
1126                             struct btrfs_key *cpu_key, u32 data_size)
1127 {
1128         int ret = 0;
1129         int slot;
1130         int slot_orig;
1131         struct btrfs_leaf *leaf;
1132         struct btrfs_buffer *leaf_buf;
1133         u32 nritems;
1134         unsigned int data_end;
1135         struct btrfs_disk_key disk_key;
1136
1137         btrfs_cpu_key_to_disk(&disk_key, cpu_key);
1138
1139         /* create a root if there isn't one */
1140         if (!root->node)
1141                 BUG();
1142         ret = btrfs_search_slot(root, cpu_key, path, data_size, 1);
1143         if (ret == 0) {
1144                 btrfs_release_path(root, path);
1145                 return -EEXIST;
1146         }
1147         if (ret < 0)
1148                 goto out;
1149
1150         slot_orig = path->slots[0];
1151         leaf_buf = path->nodes[0];
1152         leaf = &leaf_buf->leaf;
1153
1154         nritems = btrfs_header_nritems(&leaf->header);
1155         data_end = leaf_data_end(root, leaf);
1156
1157         if (btrfs_leaf_free_space(root, leaf) <
1158             sizeof(struct btrfs_item) + data_size)
1159                 BUG();
1160
1161         slot = path->slots[0];
1162         BUG_ON(slot < 0);
1163         if (slot != nritems) {
1164                 int i;
1165                 unsigned int old_data = btrfs_item_end(leaf->items + slot);
1166
1167                 /*
1168                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
1169                  */
1170                 /* first correct the data pointers */
1171                 for (i = slot; i < nritems; i++) {
1172                         u32 ioff = btrfs_item_offset(leaf->items + i);
1173                         btrfs_set_item_offset(leaf->items + i,
1174                                               ioff - data_size);
1175                 }
1176
1177                 /* shift the items */
1178                 memmove(leaf->items + slot + 1, leaf->items + slot,
1179                         (nritems - slot) * sizeof(struct btrfs_item));
1180
1181                 /* shift the data */
1182                 memmove(btrfs_leaf_data(leaf) + data_end - data_size,
1183                         btrfs_leaf_data(leaf) +
1184                         data_end, old_data - data_end);
1185                 data_end = old_data;
1186         }
1187         /* setup the item for the new data */
1188         memcpy(&leaf->items[slot].key, &disk_key,
1189                 sizeof(struct btrfs_disk_key));
1190         btrfs_set_item_offset(leaf->items + slot, data_end - data_size);
1191         btrfs_set_item_size(leaf->items + slot, data_size);
1192         btrfs_set_header_nritems(&leaf->header, nritems + 1);
1193
1194         ret = 0;
1195         if (slot == 0)
1196                 ret = fixup_low_keys(root, path, &disk_key, 1);
1197
1198         BUG_ON(list_empty(&leaf_buf->dirty));
1199         if (btrfs_leaf_free_space(root, leaf) < 0)
1200                 BUG();
1201         check_leaf(root, path, 0);
1202 out:
1203         return ret;
1204 }
1205
1206 /*
1207  * Given a key and some data, insert an item into the tree.
1208  * This does all the path init required, making room in the tree if needed.
1209  */
1210 int btrfs_insert_item(struct btrfs_root *root, struct btrfs_key *cpu_key,
1211                           void *data, u32 data_size)
1212 {
1213         int ret = 0;
1214         struct btrfs_path path;
1215         u8 *ptr;
1216
1217         btrfs_init_path(&path);
1218         ret = btrfs_insert_empty_item(root, &path, cpu_key, data_size);
1219         if (!ret) {
1220                 ptr = btrfs_item_ptr(&path.nodes[0]->leaf, path.slots[0], u8);
1221                 memcpy(ptr, data, data_size);
1222         }
1223         btrfs_release_path(root, &path);
1224         return ret;
1225 }
1226
1227 /*
1228  * delete the pointer from a given node.
1229  *
1230  * If the delete empties a node, the node is removed from the tree,
1231  * continuing all the way the root if required.  The root is converted into
1232  * a leaf if all the nodes are emptied.
1233  */
1234 static int del_ptr(struct btrfs_root *root, struct btrfs_path *path, int level,
1235                    int slot)
1236 {
1237         struct btrfs_node *node;
1238         struct btrfs_buffer *parent = path->nodes[level];
1239         u32 nritems;
1240         int ret = 0;
1241         int wret;
1242
1243         node = &parent->node;
1244         nritems = btrfs_header_nritems(&node->header);
1245         if (slot != nritems -1) {
1246                 memmove(node->ptrs + slot, node->ptrs + slot + 1,
1247                         sizeof(struct btrfs_key_ptr) * (nritems - slot - 1));
1248         }
1249         nritems--;
1250         btrfs_set_header_nritems(&node->header, nritems);
1251         if (nritems == 0 && parent == root->node) {
1252                 BUG_ON(btrfs_header_level(&root->node->node.header) != 1);
1253                 /* just turn the root into a leaf and break */
1254                 btrfs_set_header_level(&root->node->node.header, 0);
1255         } else if (slot == 0) {
1256                 wret = fixup_low_keys(root, path, &node->ptrs[0].key,
1257                                       level + 1);
1258                 if (wret)
1259                         ret = wret;
1260         }
1261         BUG_ON(list_empty(&parent->dirty));
1262         return ret;
1263 }
1264
1265 /*
1266  * delete the item at the leaf level in path.  If that empties
1267  * the leaf, remove it from the tree
1268  */
1269 int btrfs_del_item(struct btrfs_root *root, struct btrfs_path *path)
1270 {
1271         int slot;
1272         struct btrfs_leaf *leaf;
1273         struct btrfs_buffer *leaf_buf;
1274         int doff;
1275         int dsize;
1276         int ret = 0;
1277         int wret;
1278         u32 nritems;
1279
1280         leaf_buf = path->nodes[0];
1281         leaf = &leaf_buf->leaf;
1282         slot = path->slots[0];
1283         doff = btrfs_item_offset(leaf->items + slot);
1284         dsize = btrfs_item_size(leaf->items + slot);
1285         nritems = btrfs_header_nritems(&leaf->header);
1286
1287         if (slot != nritems - 1) {
1288                 int i;
1289                 int data_end = leaf_data_end(root, leaf);
1290                 memmove(btrfs_leaf_data(leaf) + data_end + dsize,
1291                         btrfs_leaf_data(leaf) + data_end,
1292                         doff - data_end);
1293                 for (i = slot + 1; i < nritems; i++) {
1294                         u32 ioff = btrfs_item_offset(leaf->items + i);
1295                         btrfs_set_item_offset(leaf->items + i, ioff + dsize);
1296                 }
1297                 memmove(leaf->items + slot, leaf->items + slot + 1,
1298                         sizeof(struct btrfs_item) *
1299                         (nritems - slot - 1));
1300         }
1301         btrfs_set_header_nritems(&leaf->header, nritems - 1);
1302         nritems--;
1303         /* delete the leaf if we've emptied it */
1304         if (nritems == 0) {
1305                 if (leaf_buf == root->node) {
1306                         btrfs_set_header_level(&leaf->header, 0);
1307                         BUG_ON(list_empty(&leaf_buf->dirty));
1308                 } else {
1309                         clean_tree_block(root, leaf_buf);
1310                         wret = del_ptr(root, path, 1, path->slots[1]);
1311                         if (wret)
1312                                 ret = wret;
1313                         wret = btrfs_free_extent(root, leaf_buf->blocknr, 1);
1314                         if (wret)
1315                                 ret = wret;
1316                 }
1317         } else {
1318                 int used = leaf_space_used(leaf, 0, nritems);
1319                 if (slot == 0) {
1320                         wret = fixup_low_keys(root, path,
1321                                                    &leaf->items[0].key, 1);
1322                         if (wret)
1323                                 ret = wret;
1324                 }
1325                 BUG_ON(list_empty(&leaf_buf->dirty));
1326
1327                 /* delete the leaf if it is mostly empty */
1328                 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
1329                         /* push_leaf_left fixes the path.
1330                          * make sure the path still points to our leaf
1331                          * for possible call to del_ptr below
1332                          */
1333                         slot = path->slots[1];
1334                         leaf_buf->count++;
1335                         wret = push_leaf_left(root, path, 1);
1336                         if (wret < 0)
1337                                 ret = wret;
1338                         if (path->nodes[0] == leaf_buf &&
1339                             btrfs_header_nritems(&leaf->header)) {
1340                                 wret = push_leaf_right(root, path, 1);
1341                                 if (wret < 0)
1342                                         ret = wret;
1343                         }
1344                         if (btrfs_header_nritems(&leaf->header) == 0) {
1345                                 u64 blocknr = leaf_buf->blocknr;
1346                                 clean_tree_block(root, leaf_buf);
1347                                 wret = del_ptr(root, path, 1, slot);
1348                                 if (wret)
1349                                         ret = wret;
1350                                 btrfs_block_release(root, leaf_buf);
1351                                 wret = btrfs_free_extent(root, blocknr, 1);
1352                                 if (wret)
1353                                         ret = wret;
1354                         } else {
1355                                 btrfs_block_release(root, leaf_buf);
1356                         }
1357                 }
1358         }
1359         return ret;
1360 }
1361
1362 /*
1363  * walk up the tree as far as required to find the next leaf.
1364  * returns 0 if it found something or 1 if there are no greater leaves.
1365  * returns < 0 on io errors.
1366  */
1367 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
1368 {
1369         int slot;
1370         int level = 1;
1371         u64 blocknr;
1372         struct btrfs_buffer *c;
1373         struct btrfs_buffer *next = NULL;
1374
1375         while(level < BTRFS_MAX_LEVEL) {
1376                 if (!path->nodes[level])
1377                         return 1;
1378                 slot = path->slots[level] + 1;
1379                 c = path->nodes[level];
1380                 if (slot >= btrfs_header_nritems(&c->node.header)) {
1381                         level++;
1382                         continue;
1383                 }
1384                 blocknr = btrfs_node_blockptr(&c->node, slot);
1385                 if (next)
1386                         btrfs_block_release(root, next);
1387                 next = read_tree_block(root, blocknr);
1388                 break;
1389         }
1390         path->slots[level] = slot;
1391         while(1) {
1392                 level--;
1393                 c = path->nodes[level];
1394                 btrfs_block_release(root, c);
1395                 path->nodes[level] = next;
1396                 path->slots[level] = 0;
1397                 if (!level)
1398                         break;
1399                 next = read_tree_block(root,
1400                                        btrfs_node_blockptr(&next->node, 0));
1401         }
1402         return 0;
1403 }
1404
1405