1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2017 Christoph Hellwig. 4 */ 5 6 #include "xfs.h" 7 #include "xfs_shared.h" 8 #include "xfs_format.h" 9 #include "xfs_bit.h" 10 #include "xfs_log_format.h" 11 #include "xfs_inode.h" 12 #include "xfs_trans_resv.h" 13 #include "xfs_mount.h" 14 #include "xfs_trace.h" 15 16 /* 17 * In-core extent record layout: 18 * 19 * +-------+----------------------------+ 20 * | 00:53 | all 54 bits of startoff | 21 * | 54:63 | low 10 bits of startblock | 22 * +-------+----------------------------+ 23 * | 00:20 | all 21 bits of length | 24 * | 21 | unwritten extent bit | 25 * | 22:63 | high 42 bits of startblock | 26 * +-------+----------------------------+ 27 */ 28 #define XFS_IEXT_STARTOFF_MASK xfs_mask64lo(BMBT_STARTOFF_BITLEN) 29 #define XFS_IEXT_LENGTH_MASK xfs_mask64lo(BMBT_BLOCKCOUNT_BITLEN) 30 #define XFS_IEXT_STARTBLOCK_MASK xfs_mask64lo(BMBT_STARTBLOCK_BITLEN) 31 32 struct xfs_iext_rec { 33 uint64_t lo; 34 uint64_t hi; 35 }; 36 37 /* 38 * Given that the length can't be a zero, only an empty hi value indicates an 39 * unused record. 40 */ 41 static bool xfs_iext_rec_is_empty(struct xfs_iext_rec *rec) 42 { 43 return rec->hi == 0; 44 } 45 46 static inline void xfs_iext_rec_clear(struct xfs_iext_rec *rec) 47 { 48 rec->lo = 0; 49 rec->hi = 0; 50 } 51 52 static void 53 xfs_iext_set( 54 struct xfs_iext_rec *rec, 55 struct xfs_bmbt_irec *irec) 56 { 57 ASSERT((irec->br_startoff & ~XFS_IEXT_STARTOFF_MASK) == 0); 58 ASSERT((irec->br_blockcount & ~XFS_IEXT_LENGTH_MASK) == 0); 59 ASSERT((irec->br_startblock & ~XFS_IEXT_STARTBLOCK_MASK) == 0); 60 61 rec->lo = irec->br_startoff & XFS_IEXT_STARTOFF_MASK; 62 rec->hi = irec->br_blockcount & XFS_IEXT_LENGTH_MASK; 63 64 rec->lo |= (irec->br_startblock << 54); 65 rec->hi |= ((irec->br_startblock & ~xfs_mask64lo(10)) << (22 - 10)); 66 67 if (irec->br_state == XFS_EXT_UNWRITTEN) 68 rec->hi |= (1 << 21); 69 } 70 71 static void 72 xfs_iext_get( 73 struct xfs_bmbt_irec *irec, 74 struct xfs_iext_rec *rec) 75 { 76 irec->br_startoff = rec->lo & XFS_IEXT_STARTOFF_MASK; 77 irec->br_blockcount = rec->hi & XFS_IEXT_LENGTH_MASK; 78 79 irec->br_startblock = rec->lo >> 54; 80 irec->br_startblock |= (rec->hi & xfs_mask64hi(42)) >> (22 - 10); 81 82 if (rec->hi & (1 << 21)) 83 irec->br_state = XFS_EXT_UNWRITTEN; 84 else 85 irec->br_state = XFS_EXT_NORM; 86 } 87 88 enum { 89 NODE_SIZE = 256, 90 KEYS_PER_NODE = NODE_SIZE / (sizeof(uint64_t) + sizeof(void *)), 91 RECS_PER_LEAF = (NODE_SIZE - (2 * sizeof(struct xfs_iext_leaf *))) / 92 sizeof(struct xfs_iext_rec), 93 }; 94 95 /* 96 * In-core extent btree block layout: 97 * 98 * There are two types of blocks in the btree: leaf and inner (non-leaf) blocks. 99 * 100 * The leaf blocks are made up by %KEYS_PER_NODE extent records, which each 101 * contain the startoffset, blockcount, startblock and unwritten extent flag. 102 * See above for the exact format, followed by pointers to the previous and next 103 * leaf blocks (if there are any). 104 * 105 * The inner (non-leaf) blocks first contain KEYS_PER_NODE lookup keys, followed 106 * by an equal number of pointers to the btree blocks at the next lower level. 107 * 108 * +-------+-------+-------+-------+-------+----------+----------+ 109 * Leaf: | rec 1 | rec 2 | rec 3 | rec 4 | rec N | prev-ptr | next-ptr | 110 * +-------+-------+-------+-------+-------+----------+----------+ 111 * 112 * +-------+-------+-------+-------+-------+-------+------+-------+ 113 * Inner: | key 1 | key 2 | key 3 | key N | ptr 1 | ptr 2 | ptr3 | ptr N | 114 * +-------+-------+-------+-------+-------+-------+------+-------+ 115 */ 116 struct xfs_iext_node { 117 uint64_t keys[KEYS_PER_NODE]; 118 #define XFS_IEXT_KEY_INVALID (1ULL << 63) 119 void *ptrs[KEYS_PER_NODE]; 120 }; 121 122 struct xfs_iext_leaf { 123 struct xfs_iext_rec recs[RECS_PER_LEAF]; 124 struct xfs_iext_leaf *prev; 125 struct xfs_iext_leaf *next; 126 }; 127 128 inline xfs_extnum_t xfs_iext_count(struct xfs_ifork *ifp) 129 { 130 return ifp->if_bytes / sizeof(struct xfs_iext_rec); 131 } 132 133 static inline int xfs_iext_max_recs(struct xfs_ifork *ifp) 134 { 135 if (ifp->if_height == 1) 136 return xfs_iext_count(ifp); 137 return RECS_PER_LEAF; 138 } 139 140 static inline struct xfs_iext_rec *cur_rec(struct xfs_iext_cursor *cur) 141 { 142 return &cur->leaf->recs[cur->pos]; 143 } 144 145 static inline bool xfs_iext_valid(struct xfs_ifork *ifp, 146 struct xfs_iext_cursor *cur) 147 { 148 if (!cur->leaf) 149 return false; 150 if (cur->pos < 0 || cur->pos >= xfs_iext_max_recs(ifp)) 151 return false; 152 if (xfs_iext_rec_is_empty(cur_rec(cur))) 153 return false; 154 return true; 155 } 156 157 static void * 158 xfs_iext_find_first_leaf( 159 struct xfs_ifork *ifp) 160 { 161 struct xfs_iext_node *node = ifp->if_u1.if_root; 162 int height; 163 164 if (!ifp->if_height) 165 return NULL; 166 167 for (height = ifp->if_height; height > 1; height--) { 168 node = node->ptrs[0]; 169 ASSERT(node); 170 } 171 172 return node; 173 } 174 175 static void * 176 xfs_iext_find_last_leaf( 177 struct xfs_ifork *ifp) 178 { 179 struct xfs_iext_node *node = ifp->if_u1.if_root; 180 int height, i; 181 182 if (!ifp->if_height) 183 return NULL; 184 185 for (height = ifp->if_height; height > 1; height--) { 186 for (i = 1; i < KEYS_PER_NODE; i++) 187 if (!node->ptrs[i]) 188 break; 189 node = node->ptrs[i - 1]; 190 ASSERT(node); 191 } 192 193 return node; 194 } 195 196 void 197 xfs_iext_first( 198 struct xfs_ifork *ifp, 199 struct xfs_iext_cursor *cur) 200 { 201 cur->pos = 0; 202 cur->leaf = xfs_iext_find_first_leaf(ifp); 203 } 204 205 void 206 xfs_iext_last( 207 struct xfs_ifork *ifp, 208 struct xfs_iext_cursor *cur) 209 { 210 int i; 211 212 cur->leaf = xfs_iext_find_last_leaf(ifp); 213 if (!cur->leaf) { 214 cur->pos = 0; 215 return; 216 } 217 218 for (i = 1; i < xfs_iext_max_recs(ifp); i++) { 219 if (xfs_iext_rec_is_empty(&cur->leaf->recs[i])) 220 break; 221 } 222 cur->pos = i - 1; 223 } 224 225 void 226 xfs_iext_next( 227 struct xfs_ifork *ifp, 228 struct xfs_iext_cursor *cur) 229 { 230 if (!cur->leaf) { 231 ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF); 232 xfs_iext_first(ifp, cur); 233 return; 234 } 235 236 ASSERT(cur->pos >= 0); 237 ASSERT(cur->pos < xfs_iext_max_recs(ifp)); 238 239 cur->pos++; 240 if (ifp->if_height > 1 && !xfs_iext_valid(ifp, cur) && 241 cur->leaf->next) { 242 cur->leaf = cur->leaf->next; 243 cur->pos = 0; 244 } 245 } 246 247 void 248 xfs_iext_prev( 249 struct xfs_ifork *ifp, 250 struct xfs_iext_cursor *cur) 251 { 252 if (!cur->leaf) { 253 ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF); 254 xfs_iext_last(ifp, cur); 255 return; 256 } 257 258 ASSERT(cur->pos >= 0); 259 ASSERT(cur->pos <= RECS_PER_LEAF); 260 261 recurse: 262 do { 263 cur->pos--; 264 if (xfs_iext_valid(ifp, cur)) 265 return; 266 } while (cur->pos > 0); 267 268 if (ifp->if_height > 1 && cur->leaf->prev) { 269 cur->leaf = cur->leaf->prev; 270 cur->pos = RECS_PER_LEAF; 271 goto recurse; 272 } 273 } 274 275 static inline int 276 xfs_iext_key_cmp( 277 struct xfs_iext_node *node, 278 int n, 279 xfs_fileoff_t offset) 280 { 281 if (node->keys[n] > offset) 282 return 1; 283 if (node->keys[n] < offset) 284 return -1; 285 return 0; 286 } 287 288 static inline int 289 xfs_iext_rec_cmp( 290 struct xfs_iext_rec *rec, 291 xfs_fileoff_t offset) 292 { 293 uint64_t rec_offset = rec->lo & XFS_IEXT_STARTOFF_MASK; 294 uint32_t rec_len = rec->hi & XFS_IEXT_LENGTH_MASK; 295 296 if (rec_offset > offset) 297 return 1; 298 if (rec_offset + rec_len <= offset) 299 return -1; 300 return 0; 301 } 302 303 static void * 304 xfs_iext_find_level( 305 struct xfs_ifork *ifp, 306 xfs_fileoff_t offset, 307 int level) 308 { 309 struct xfs_iext_node *node = ifp->if_u1.if_root; 310 int height, i; 311 312 if (!ifp->if_height) 313 return NULL; 314 315 for (height = ifp->if_height; height > level; height--) { 316 for (i = 1; i < KEYS_PER_NODE; i++) 317 if (xfs_iext_key_cmp(node, i, offset) > 0) 318 break; 319 320 node = node->ptrs[i - 1]; 321 if (!node) 322 break; 323 } 324 325 return node; 326 } 327 328 static int 329 xfs_iext_node_pos( 330 struct xfs_iext_node *node, 331 xfs_fileoff_t offset) 332 { 333 int i; 334 335 for (i = 1; i < KEYS_PER_NODE; i++) { 336 if (xfs_iext_key_cmp(node, i, offset) > 0) 337 break; 338 } 339 340 return i - 1; 341 } 342 343 static int 344 xfs_iext_node_insert_pos( 345 struct xfs_iext_node *node, 346 xfs_fileoff_t offset) 347 { 348 int i; 349 350 for (i = 0; i < KEYS_PER_NODE; i++) { 351 if (xfs_iext_key_cmp(node, i, offset) > 0) 352 return i; 353 } 354 355 return KEYS_PER_NODE; 356 } 357 358 static int 359 xfs_iext_node_nr_entries( 360 struct xfs_iext_node *node, 361 int start) 362 { 363 int i; 364 365 for (i = start; i < KEYS_PER_NODE; i++) { 366 if (node->keys[i] == XFS_IEXT_KEY_INVALID) 367 break; 368 } 369 370 return i; 371 } 372 373 static int 374 xfs_iext_leaf_nr_entries( 375 struct xfs_ifork *ifp, 376 struct xfs_iext_leaf *leaf, 377 int start) 378 { 379 int i; 380 381 for (i = start; i < xfs_iext_max_recs(ifp); i++) { 382 if (xfs_iext_rec_is_empty(&leaf->recs[i])) 383 break; 384 } 385 386 return i; 387 } 388 389 static inline uint64_t 390 xfs_iext_leaf_key( 391 struct xfs_iext_leaf *leaf, 392 int n) 393 { 394 return leaf->recs[n].lo & XFS_IEXT_STARTOFF_MASK; 395 } 396 397 static void 398 xfs_iext_grow( 399 struct xfs_ifork *ifp) 400 { 401 struct xfs_iext_node *node = kmem_zalloc(NODE_SIZE, KM_NOFS); 402 int i; 403 404 if (ifp->if_height == 1) { 405 struct xfs_iext_leaf *prev = ifp->if_u1.if_root; 406 407 node->keys[0] = xfs_iext_leaf_key(prev, 0); 408 node->ptrs[0] = prev; 409 } else { 410 struct xfs_iext_node *prev = ifp->if_u1.if_root; 411 412 ASSERT(ifp->if_height > 1); 413 414 node->keys[0] = prev->keys[0]; 415 node->ptrs[0] = prev; 416 } 417 418 for (i = 1; i < KEYS_PER_NODE; i++) 419 node->keys[i] = XFS_IEXT_KEY_INVALID; 420 421 ifp->if_u1.if_root = node; 422 ifp->if_height++; 423 } 424 425 static void 426 xfs_iext_update_node( 427 struct xfs_ifork *ifp, 428 xfs_fileoff_t old_offset, 429 xfs_fileoff_t new_offset, 430 int level, 431 void *ptr) 432 { 433 struct xfs_iext_node *node = ifp->if_u1.if_root; 434 int height, i; 435 436 for (height = ifp->if_height; height > level; height--) { 437 for (i = 0; i < KEYS_PER_NODE; i++) { 438 if (i > 0 && xfs_iext_key_cmp(node, i, old_offset) > 0) 439 break; 440 if (node->keys[i] == old_offset) 441 node->keys[i] = new_offset; 442 } 443 node = node->ptrs[i - 1]; 444 ASSERT(node); 445 } 446 447 ASSERT(node == ptr); 448 } 449 450 static struct xfs_iext_node * 451 xfs_iext_split_node( 452 struct xfs_iext_node **nodep, 453 int *pos, 454 int *nr_entries) 455 { 456 struct xfs_iext_node *node = *nodep; 457 struct xfs_iext_node *new = kmem_zalloc(NODE_SIZE, KM_NOFS); 458 const int nr_move = KEYS_PER_NODE / 2; 459 int nr_keep = nr_move + (KEYS_PER_NODE & 1); 460 int i = 0; 461 462 /* for sequential append operations just spill over into the new node */ 463 if (*pos == KEYS_PER_NODE) { 464 *nodep = new; 465 *pos = 0; 466 *nr_entries = 0; 467 goto done; 468 } 469 470 471 for (i = 0; i < nr_move; i++) { 472 new->keys[i] = node->keys[nr_keep + i]; 473 new->ptrs[i] = node->ptrs[nr_keep + i]; 474 475 node->keys[nr_keep + i] = XFS_IEXT_KEY_INVALID; 476 node->ptrs[nr_keep + i] = NULL; 477 } 478 479 if (*pos >= nr_keep) { 480 *nodep = new; 481 *pos -= nr_keep; 482 *nr_entries = nr_move; 483 } else { 484 *nr_entries = nr_keep; 485 } 486 done: 487 for (; i < KEYS_PER_NODE; i++) 488 new->keys[i] = XFS_IEXT_KEY_INVALID; 489 return new; 490 } 491 492 static void 493 xfs_iext_insert_node( 494 struct xfs_ifork *ifp, 495 uint64_t offset, 496 void *ptr, 497 int level) 498 { 499 struct xfs_iext_node *node, *new; 500 int i, pos, nr_entries; 501 502 again: 503 if (ifp->if_height < level) 504 xfs_iext_grow(ifp); 505 506 new = NULL; 507 node = xfs_iext_find_level(ifp, offset, level); 508 pos = xfs_iext_node_insert_pos(node, offset); 509 nr_entries = xfs_iext_node_nr_entries(node, pos); 510 511 ASSERT(pos >= nr_entries || xfs_iext_key_cmp(node, pos, offset) != 0); 512 ASSERT(nr_entries <= KEYS_PER_NODE); 513 514 if (nr_entries == KEYS_PER_NODE) 515 new = xfs_iext_split_node(&node, &pos, &nr_entries); 516 517 /* 518 * Update the pointers in higher levels if the first entry changes 519 * in an existing node. 520 */ 521 if (node != new && pos == 0 && nr_entries > 0) 522 xfs_iext_update_node(ifp, node->keys[0], offset, level, node); 523 524 for (i = nr_entries; i > pos; i--) { 525 node->keys[i] = node->keys[i - 1]; 526 node->ptrs[i] = node->ptrs[i - 1]; 527 } 528 node->keys[pos] = offset; 529 node->ptrs[pos] = ptr; 530 531 if (new) { 532 offset = new->keys[0]; 533 ptr = new; 534 level++; 535 goto again; 536 } 537 } 538 539 static struct xfs_iext_leaf * 540 xfs_iext_split_leaf( 541 struct xfs_iext_cursor *cur, 542 int *nr_entries) 543 { 544 struct xfs_iext_leaf *leaf = cur->leaf; 545 struct xfs_iext_leaf *new = kmem_zalloc(NODE_SIZE, KM_NOFS); 546 const int nr_move = RECS_PER_LEAF / 2; 547 int nr_keep = nr_move + (RECS_PER_LEAF & 1); 548 int i; 549 550 /* for sequential append operations just spill over into the new node */ 551 if (cur->pos == RECS_PER_LEAF) { 552 cur->leaf = new; 553 cur->pos = 0; 554 *nr_entries = 0; 555 goto done; 556 } 557 558 for (i = 0; i < nr_move; i++) { 559 new->recs[i] = leaf->recs[nr_keep + i]; 560 xfs_iext_rec_clear(&leaf->recs[nr_keep + i]); 561 } 562 563 if (cur->pos >= nr_keep) { 564 cur->leaf = new; 565 cur->pos -= nr_keep; 566 *nr_entries = nr_move; 567 } else { 568 *nr_entries = nr_keep; 569 } 570 done: 571 if (leaf->next) 572 leaf->next->prev = new; 573 new->next = leaf->next; 574 new->prev = leaf; 575 leaf->next = new; 576 return new; 577 } 578 579 static void 580 xfs_iext_alloc_root( 581 struct xfs_ifork *ifp, 582 struct xfs_iext_cursor *cur) 583 { 584 ASSERT(ifp->if_bytes == 0); 585 586 ifp->if_u1.if_root = kmem_zalloc(sizeof(struct xfs_iext_rec), KM_NOFS); 587 ifp->if_height = 1; 588 589 /* now that we have a node step into it */ 590 cur->leaf = ifp->if_u1.if_root; 591 cur->pos = 0; 592 } 593 594 static void 595 xfs_iext_realloc_root( 596 struct xfs_ifork *ifp, 597 struct xfs_iext_cursor *cur) 598 { 599 int64_t new_size = ifp->if_bytes + sizeof(struct xfs_iext_rec); 600 void *new; 601 602 /* account for the prev/next pointers */ 603 if (new_size / sizeof(struct xfs_iext_rec) == RECS_PER_LEAF) 604 new_size = NODE_SIZE; 605 606 new = kmem_realloc(ifp->if_u1.if_root, new_size, KM_NOFS); 607 memset(new + ifp->if_bytes, 0, new_size - ifp->if_bytes); 608 ifp->if_u1.if_root = new; 609 cur->leaf = new; 610 } 611 612 /* 613 * Increment the sequence counter on extent tree changes. If we are on a COW 614 * fork, this allows the writeback code to skip looking for a COW extent if the 615 * COW fork hasn't changed. We use WRITE_ONCE here to ensure the update to the 616 * sequence counter is seen before the modifications to the extent tree itself 617 * take effect. 618 */ 619 static inline void xfs_iext_inc_seq(struct xfs_ifork *ifp) 620 { 621 WRITE_ONCE(ifp->if_seq, READ_ONCE(ifp->if_seq) + 1); 622 } 623 624 void 625 xfs_iext_insert( 626 struct xfs_inode *ip, 627 struct xfs_iext_cursor *cur, 628 struct xfs_bmbt_irec *irec, 629 int state) 630 { 631 struct xfs_ifork *ifp = xfs_iext_state_to_fork(ip, state); 632 xfs_fileoff_t offset = irec->br_startoff; 633 struct xfs_iext_leaf *new = NULL; 634 int nr_entries, i; 635 636 xfs_iext_inc_seq(ifp); 637 638 if (ifp->if_height == 0) 639 xfs_iext_alloc_root(ifp, cur); 640 else if (ifp->if_height == 1) 641 xfs_iext_realloc_root(ifp, cur); 642 643 nr_entries = xfs_iext_leaf_nr_entries(ifp, cur->leaf, cur->pos); 644 ASSERT(nr_entries <= RECS_PER_LEAF); 645 ASSERT(cur->pos >= nr_entries || 646 xfs_iext_rec_cmp(cur_rec(cur), irec->br_startoff) != 0); 647 648 if (nr_entries == RECS_PER_LEAF) 649 new = xfs_iext_split_leaf(cur, &nr_entries); 650 651 /* 652 * Update the pointers in higher levels if the first entry changes 653 * in an existing node. 654 */ 655 if (cur->leaf != new && cur->pos == 0 && nr_entries > 0) { 656 xfs_iext_update_node(ifp, xfs_iext_leaf_key(cur->leaf, 0), 657 offset, 1, cur->leaf); 658 } 659 660 for (i = nr_entries; i > cur->pos; i--) 661 cur->leaf->recs[i] = cur->leaf->recs[i - 1]; 662 xfs_iext_set(cur_rec(cur), irec); 663 ifp->if_bytes += sizeof(struct xfs_iext_rec); 664 665 trace_xfs_iext_insert(ip, cur, state, _RET_IP_); 666 667 if (new) 668 xfs_iext_insert_node(ifp, xfs_iext_leaf_key(new, 0), new, 2); 669 } 670 671 static struct xfs_iext_node * 672 xfs_iext_rebalance_node( 673 struct xfs_iext_node *parent, 674 int *pos, 675 struct xfs_iext_node *node, 676 int nr_entries) 677 { 678 /* 679 * If the neighbouring nodes are completely full, or have different 680 * parents, we might never be able to merge our node, and will only 681 * delete it once the number of entries hits zero. 682 */ 683 if (nr_entries == 0) 684 return node; 685 686 if (*pos > 0) { 687 struct xfs_iext_node *prev = parent->ptrs[*pos - 1]; 688 int nr_prev = xfs_iext_node_nr_entries(prev, 0), i; 689 690 if (nr_prev + nr_entries <= KEYS_PER_NODE) { 691 for (i = 0; i < nr_entries; i++) { 692 prev->keys[nr_prev + i] = node->keys[i]; 693 prev->ptrs[nr_prev + i] = node->ptrs[i]; 694 } 695 return node; 696 } 697 } 698 699 if (*pos + 1 < xfs_iext_node_nr_entries(parent, *pos)) { 700 struct xfs_iext_node *next = parent->ptrs[*pos + 1]; 701 int nr_next = xfs_iext_node_nr_entries(next, 0), i; 702 703 if (nr_entries + nr_next <= KEYS_PER_NODE) { 704 /* 705 * Merge the next node into this node so that we don't 706 * have to do an additional update of the keys in the 707 * higher levels. 708 */ 709 for (i = 0; i < nr_next; i++) { 710 node->keys[nr_entries + i] = next->keys[i]; 711 node->ptrs[nr_entries + i] = next->ptrs[i]; 712 } 713 714 ++*pos; 715 return next; 716 } 717 } 718 719 return NULL; 720 } 721 722 static void 723 xfs_iext_remove_node( 724 struct xfs_ifork *ifp, 725 xfs_fileoff_t offset, 726 void *victim) 727 { 728 struct xfs_iext_node *node, *parent; 729 int level = 2, pos, nr_entries, i; 730 731 ASSERT(level <= ifp->if_height); 732 node = xfs_iext_find_level(ifp, offset, level); 733 pos = xfs_iext_node_pos(node, offset); 734 again: 735 ASSERT(node->ptrs[pos]); 736 ASSERT(node->ptrs[pos] == victim); 737 kmem_free(victim); 738 739 nr_entries = xfs_iext_node_nr_entries(node, pos) - 1; 740 offset = node->keys[0]; 741 for (i = pos; i < nr_entries; i++) { 742 node->keys[i] = node->keys[i + 1]; 743 node->ptrs[i] = node->ptrs[i + 1]; 744 } 745 node->keys[nr_entries] = XFS_IEXT_KEY_INVALID; 746 node->ptrs[nr_entries] = NULL; 747 748 if (pos == 0 && nr_entries > 0) { 749 xfs_iext_update_node(ifp, offset, node->keys[0], level, node); 750 offset = node->keys[0]; 751 } 752 753 if (nr_entries >= KEYS_PER_NODE / 2) 754 return; 755 756 if (level < ifp->if_height) { 757 /* 758 * If we aren't at the root yet try to find a neighbour node to 759 * merge with (or delete the node if it is empty), and then 760 * recurse up to the next level. 761 */ 762 level++; 763 parent = xfs_iext_find_level(ifp, offset, level); 764 pos = xfs_iext_node_pos(parent, offset); 765 766 ASSERT(pos != KEYS_PER_NODE); 767 ASSERT(parent->ptrs[pos] == node); 768 769 node = xfs_iext_rebalance_node(parent, &pos, node, nr_entries); 770 if (node) { 771 victim = node; 772 node = parent; 773 goto again; 774 } 775 } else if (nr_entries == 1) { 776 /* 777 * If we are at the root and only one entry is left we can just 778 * free this node and update the root pointer. 779 */ 780 ASSERT(node == ifp->if_u1.if_root); 781 ifp->if_u1.if_root = node->ptrs[0]; 782 ifp->if_height--; 783 kmem_free(node); 784 } 785 } 786 787 static void 788 xfs_iext_rebalance_leaf( 789 struct xfs_ifork *ifp, 790 struct xfs_iext_cursor *cur, 791 struct xfs_iext_leaf *leaf, 792 xfs_fileoff_t offset, 793 int nr_entries) 794 { 795 /* 796 * If the neighbouring nodes are completely full we might never be able 797 * to merge our node, and will only delete it once the number of 798 * entries hits zero. 799 */ 800 if (nr_entries == 0) 801 goto remove_node; 802 803 if (leaf->prev) { 804 int nr_prev = xfs_iext_leaf_nr_entries(ifp, leaf->prev, 0), i; 805 806 if (nr_prev + nr_entries <= RECS_PER_LEAF) { 807 for (i = 0; i < nr_entries; i++) 808 leaf->prev->recs[nr_prev + i] = leaf->recs[i]; 809 810 if (cur->leaf == leaf) { 811 cur->leaf = leaf->prev; 812 cur->pos += nr_prev; 813 } 814 goto remove_node; 815 } 816 } 817 818 if (leaf->next) { 819 int nr_next = xfs_iext_leaf_nr_entries(ifp, leaf->next, 0), i; 820 821 if (nr_entries + nr_next <= RECS_PER_LEAF) { 822 /* 823 * Merge the next node into this node so that we don't 824 * have to do an additional update of the keys in the 825 * higher levels. 826 */ 827 for (i = 0; i < nr_next; i++) { 828 leaf->recs[nr_entries + i] = 829 leaf->next->recs[i]; 830 } 831 832 if (cur->leaf == leaf->next) { 833 cur->leaf = leaf; 834 cur->pos += nr_entries; 835 } 836 837 offset = xfs_iext_leaf_key(leaf->next, 0); 838 leaf = leaf->next; 839 goto remove_node; 840 } 841 } 842 843 return; 844 remove_node: 845 if (leaf->prev) 846 leaf->prev->next = leaf->next; 847 if (leaf->next) 848 leaf->next->prev = leaf->prev; 849 xfs_iext_remove_node(ifp, offset, leaf); 850 } 851 852 static void 853 xfs_iext_free_last_leaf( 854 struct xfs_ifork *ifp) 855 { 856 ifp->if_height--; 857 kmem_free(ifp->if_u1.if_root); 858 ifp->if_u1.if_root = NULL; 859 } 860 861 void 862 xfs_iext_remove( 863 struct xfs_inode *ip, 864 struct xfs_iext_cursor *cur, 865 int state) 866 { 867 struct xfs_ifork *ifp = xfs_iext_state_to_fork(ip, state); 868 struct xfs_iext_leaf *leaf = cur->leaf; 869 xfs_fileoff_t offset = xfs_iext_leaf_key(leaf, 0); 870 int i, nr_entries; 871 872 trace_xfs_iext_remove(ip, cur, state, _RET_IP_); 873 874 ASSERT(ifp->if_height > 0); 875 ASSERT(ifp->if_u1.if_root != NULL); 876 ASSERT(xfs_iext_valid(ifp, cur)); 877 878 xfs_iext_inc_seq(ifp); 879 880 nr_entries = xfs_iext_leaf_nr_entries(ifp, leaf, cur->pos) - 1; 881 for (i = cur->pos; i < nr_entries; i++) 882 leaf->recs[i] = leaf->recs[i + 1]; 883 xfs_iext_rec_clear(&leaf->recs[nr_entries]); 884 ifp->if_bytes -= sizeof(struct xfs_iext_rec); 885 886 if (cur->pos == 0 && nr_entries > 0) { 887 xfs_iext_update_node(ifp, offset, xfs_iext_leaf_key(leaf, 0), 1, 888 leaf); 889 offset = xfs_iext_leaf_key(leaf, 0); 890 } else if (cur->pos == nr_entries) { 891 if (ifp->if_height > 1 && leaf->next) 892 cur->leaf = leaf->next; 893 else 894 cur->leaf = NULL; 895 cur->pos = 0; 896 } 897 898 if (nr_entries >= RECS_PER_LEAF / 2) 899 return; 900 901 if (ifp->if_height > 1) 902 xfs_iext_rebalance_leaf(ifp, cur, leaf, offset, nr_entries); 903 else if (nr_entries == 0) 904 xfs_iext_free_last_leaf(ifp); 905 } 906 907 /* 908 * Lookup the extent covering bno. 909 * 910 * If there is an extent covering bno return the extent index, and store the 911 * expanded extent structure in *gotp, and the extent cursor in *cur. 912 * If there is no extent covering bno, but there is an extent after it (e.g. 913 * it lies in a hole) return that extent in *gotp and its cursor in *cur 914 * instead. 915 * If bno is beyond the last extent return false, and return an invalid 916 * cursor value. 917 */ 918 bool 919 xfs_iext_lookup_extent( 920 struct xfs_inode *ip, 921 struct xfs_ifork *ifp, 922 xfs_fileoff_t offset, 923 struct xfs_iext_cursor *cur, 924 struct xfs_bmbt_irec *gotp) 925 { 926 XFS_STATS_INC(ip->i_mount, xs_look_exlist); 927 928 cur->leaf = xfs_iext_find_level(ifp, offset, 1); 929 if (!cur->leaf) { 930 cur->pos = 0; 931 return false; 932 } 933 934 for (cur->pos = 0; cur->pos < xfs_iext_max_recs(ifp); cur->pos++) { 935 struct xfs_iext_rec *rec = cur_rec(cur); 936 937 if (xfs_iext_rec_is_empty(rec)) 938 break; 939 if (xfs_iext_rec_cmp(rec, offset) >= 0) 940 goto found; 941 } 942 943 /* Try looking in the next node for an entry > offset */ 944 if (ifp->if_height == 1 || !cur->leaf->next) 945 return false; 946 cur->leaf = cur->leaf->next; 947 cur->pos = 0; 948 if (!xfs_iext_valid(ifp, cur)) 949 return false; 950 found: 951 xfs_iext_get(gotp, cur_rec(cur)); 952 return true; 953 } 954 955 /* 956 * Returns the last extent before end, and if this extent doesn't cover 957 * end, update end to the end of the extent. 958 */ 959 bool 960 xfs_iext_lookup_extent_before( 961 struct xfs_inode *ip, 962 struct xfs_ifork *ifp, 963 xfs_fileoff_t *end, 964 struct xfs_iext_cursor *cur, 965 struct xfs_bmbt_irec *gotp) 966 { 967 /* could be optimized to not even look up the next on a match.. */ 968 if (xfs_iext_lookup_extent(ip, ifp, *end - 1, cur, gotp) && 969 gotp->br_startoff <= *end - 1) 970 return true; 971 if (!xfs_iext_prev_extent(ifp, cur, gotp)) 972 return false; 973 *end = gotp->br_startoff + gotp->br_blockcount; 974 return true; 975 } 976 977 void 978 xfs_iext_update_extent( 979 struct xfs_inode *ip, 980 int state, 981 struct xfs_iext_cursor *cur, 982 struct xfs_bmbt_irec *new) 983 { 984 struct xfs_ifork *ifp = xfs_iext_state_to_fork(ip, state); 985 986 xfs_iext_inc_seq(ifp); 987 988 if (cur->pos == 0) { 989 struct xfs_bmbt_irec old; 990 991 xfs_iext_get(&old, cur_rec(cur)); 992 if (new->br_startoff != old.br_startoff) { 993 xfs_iext_update_node(ifp, old.br_startoff, 994 new->br_startoff, 1, cur->leaf); 995 } 996 } 997 998 trace_xfs_bmap_pre_update(ip, cur, state, _RET_IP_); 999 xfs_iext_set(cur_rec(cur), new); 1000 trace_xfs_bmap_post_update(ip, cur, state, _RET_IP_); 1001 } 1002 1003 /* 1004 * Return true if the cursor points at an extent and return the extent structure 1005 * in gotp. Else return false. 1006 */ 1007 bool 1008 xfs_iext_get_extent( 1009 struct xfs_ifork *ifp, 1010 struct xfs_iext_cursor *cur, 1011 struct xfs_bmbt_irec *gotp) 1012 { 1013 if (!xfs_iext_valid(ifp, cur)) 1014 return false; 1015 xfs_iext_get(gotp, cur_rec(cur)); 1016 return true; 1017 } 1018 1019 /* 1020 * This is a recursive function, because of that we need to be extremely 1021 * careful with stack usage. 1022 */ 1023 static void 1024 xfs_iext_destroy_node( 1025 struct xfs_iext_node *node, 1026 int level) 1027 { 1028 int i; 1029 1030 if (level > 1) { 1031 for (i = 0; i < KEYS_PER_NODE; i++) { 1032 if (node->keys[i] == XFS_IEXT_KEY_INVALID) 1033 break; 1034 xfs_iext_destroy_node(node->ptrs[i], level - 1); 1035 } 1036 } 1037 1038 kmem_free(node); 1039 } 1040 1041 void 1042 xfs_iext_destroy( 1043 struct xfs_ifork *ifp) 1044 { 1045 xfs_iext_destroy_node(ifp->if_u1.if_root, ifp->if_height); 1046 1047 ifp->if_bytes = 0; 1048 ifp->if_height = 0; 1049 ifp->if_u1.if_root = NULL; 1050 } 1051