linux/fs/xfs/libxfs/xfs_iext_tree.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2017 Christoph Hellwig.
   4 */
   5
   6#include <linux/cache.h>
   7#include <linux/kernel.h>
   8#include <linux/slab.h>
   9#include "xfs.h"
  10#include "xfs_format.h"
  11#include "xfs_bit.h"
  12#include "xfs_log_format.h"
  13#include "xfs_inode.h"
  14#include "xfs_inode_fork.h"
  15#include "xfs_trans_resv.h"
  16#include "xfs_mount.h"
  17#include "xfs_bmap.h"
  18#include "xfs_trace.h"
  19
  20/*
  21 * In-core extent record layout:
  22 *
  23 * +-------+----------------------------+
  24 * | 00:53 | all 54 bits of startoff    |
  25 * | 54:63 | low 10 bits of startblock  |
  26 * +-------+----------------------------+
  27 * | 00:20 | all 21 bits of length      |
  28 * |    21 | unwritten extent bit       |
  29 * | 22:63 | high 42 bits of startblock |
  30 * +-------+----------------------------+
  31 */
  32#define XFS_IEXT_STARTOFF_MASK          xfs_mask64lo(BMBT_STARTOFF_BITLEN)
  33#define XFS_IEXT_LENGTH_MASK            xfs_mask64lo(BMBT_BLOCKCOUNT_BITLEN)
  34#define XFS_IEXT_STARTBLOCK_MASK        xfs_mask64lo(BMBT_STARTBLOCK_BITLEN)
  35
  36struct xfs_iext_rec {
  37        uint64_t                        lo;
  38        uint64_t                        hi;
  39};
  40
  41/*
  42 * Given that the length can't be a zero, only an empty hi value indicates an
  43 * unused record.
  44 */
  45static bool xfs_iext_rec_is_empty(struct xfs_iext_rec *rec)
  46{
  47        return rec->hi == 0;
  48}
  49
  50static inline void xfs_iext_rec_clear(struct xfs_iext_rec *rec)
  51{
  52        rec->lo = 0;
  53        rec->hi = 0;
  54}
  55
  56static void
  57xfs_iext_set(
  58        struct xfs_iext_rec     *rec,
  59        struct xfs_bmbt_irec    *irec)
  60{
  61        ASSERT((irec->br_startoff & ~XFS_IEXT_STARTOFF_MASK) == 0);
  62        ASSERT((irec->br_blockcount & ~XFS_IEXT_LENGTH_MASK) == 0);
  63        ASSERT((irec->br_startblock & ~XFS_IEXT_STARTBLOCK_MASK) == 0);
  64
  65        rec->lo = irec->br_startoff & XFS_IEXT_STARTOFF_MASK;
  66        rec->hi = irec->br_blockcount & XFS_IEXT_LENGTH_MASK;
  67
  68        rec->lo |= (irec->br_startblock << 54);
  69        rec->hi |= ((irec->br_startblock & ~xfs_mask64lo(10)) << (22 - 10));
  70
  71        if (irec->br_state == XFS_EXT_UNWRITTEN)
  72                rec->hi |= (1 << 21);
  73}
  74
  75static void
  76xfs_iext_get(
  77        struct xfs_bmbt_irec    *irec,
  78        struct xfs_iext_rec     *rec)
  79{
  80        irec->br_startoff = rec->lo & XFS_IEXT_STARTOFF_MASK;
  81        irec->br_blockcount = rec->hi & XFS_IEXT_LENGTH_MASK;
  82
  83        irec->br_startblock = rec->lo >> 54;
  84        irec->br_startblock |= (rec->hi & xfs_mask64hi(42)) >> (22 - 10);
  85
  86        if (rec->hi & (1 << 21))
  87                irec->br_state = XFS_EXT_UNWRITTEN;
  88        else
  89                irec->br_state = XFS_EXT_NORM;
  90}
  91
  92enum {
  93        NODE_SIZE       = 256,
  94        KEYS_PER_NODE   = NODE_SIZE / (sizeof(uint64_t) + sizeof(void *)),
  95        RECS_PER_LEAF   = (NODE_SIZE - (2 * sizeof(struct xfs_iext_leaf *))) /
  96                                sizeof(struct xfs_iext_rec),
  97};
  98
  99/*
 100 * In-core extent btree block layout:
 101 *
 102 * There are two types of blocks in the btree: leaf and inner (non-leaf) blocks.
 103 *
 104 * The leaf blocks are made up by %KEYS_PER_NODE extent records, which each
 105 * contain the startoffset, blockcount, startblock and unwritten extent flag.
 106 * See above for the exact format, followed by pointers to the previous and next
 107 * leaf blocks (if there are any).
 108 *
 109 * The inner (non-leaf) blocks first contain KEYS_PER_NODE lookup keys, followed
 110 * by an equal number of pointers to the btree blocks at the next lower level.
 111 *
 112 *              +-------+-------+-------+-------+-------+----------+----------+
 113 * Leaf:        | rec 1 | rec 2 | rec 3 | rec 4 | rec N | prev-ptr | next-ptr |
 114 *              +-------+-------+-------+-------+-------+----------+----------+
 115 *
 116 *              +-------+-------+-------+-------+-------+-------+------+-------+
 117 * Inner:       | key 1 | key 2 | key 3 | key N | ptr 1 | ptr 2 | ptr3 | ptr N |
 118 *              +-------+-------+-------+-------+-------+-------+------+-------+
 119 */
 120struct xfs_iext_node {
 121        uint64_t                keys[KEYS_PER_NODE];
 122#define XFS_IEXT_KEY_INVALID    (1ULL << 63)
 123        void                    *ptrs[KEYS_PER_NODE];
 124};
 125
 126struct xfs_iext_leaf {
 127        struct xfs_iext_rec     recs[RECS_PER_LEAF];
 128        struct xfs_iext_leaf    *prev;
 129        struct xfs_iext_leaf    *next;
 130};
 131
 132inline xfs_extnum_t xfs_iext_count(struct xfs_ifork *ifp)
 133{
 134        return ifp->if_bytes / sizeof(struct xfs_iext_rec);
 135}
 136
 137static inline int xfs_iext_max_recs(struct xfs_ifork *ifp)
 138{
 139        if (ifp->if_height == 1)
 140                return xfs_iext_count(ifp);
 141        return RECS_PER_LEAF;
 142}
 143
 144static inline struct xfs_iext_rec *cur_rec(struct xfs_iext_cursor *cur)
 145{
 146        return &cur->leaf->recs[cur->pos];
 147}
 148
 149static inline bool xfs_iext_valid(struct xfs_ifork *ifp,
 150                struct xfs_iext_cursor *cur)
 151{
 152        if (!cur->leaf)
 153                return false;
 154        if (cur->pos < 0 || cur->pos >= xfs_iext_max_recs(ifp))
 155                return false;
 156        if (xfs_iext_rec_is_empty(cur_rec(cur)))
 157                return false;
 158        return true;
 159}
 160
 161static void *
 162xfs_iext_find_first_leaf(
 163        struct xfs_ifork        *ifp)
 164{
 165        struct xfs_iext_node    *node = ifp->if_u1.if_root;
 166        int                     height;
 167
 168        if (!ifp->if_height)
 169                return NULL;
 170
 171        for (height = ifp->if_height; height > 1; height--) {
 172                node = node->ptrs[0];
 173                ASSERT(node);
 174        }
 175
 176        return node;
 177}
 178
 179static void *
 180xfs_iext_find_last_leaf(
 181        struct xfs_ifork        *ifp)
 182{
 183        struct xfs_iext_node    *node = ifp->if_u1.if_root;
 184        int                     height, i;
 185
 186        if (!ifp->if_height)
 187                return NULL;
 188
 189        for (height = ifp->if_height; height > 1; height--) {
 190                for (i = 1; i < KEYS_PER_NODE; i++)
 191                        if (!node->ptrs[i])
 192                                break;
 193                node = node->ptrs[i - 1];
 194                ASSERT(node);
 195        }
 196
 197        return node;
 198}
 199
 200void
 201xfs_iext_first(
 202        struct xfs_ifork        *ifp,
 203        struct xfs_iext_cursor  *cur)
 204{
 205        cur->pos = 0;
 206        cur->leaf = xfs_iext_find_first_leaf(ifp);
 207}
 208
 209void
 210xfs_iext_last(
 211        struct xfs_ifork        *ifp,
 212        struct xfs_iext_cursor  *cur)
 213{
 214        int                     i;
 215
 216        cur->leaf = xfs_iext_find_last_leaf(ifp);
 217        if (!cur->leaf) {
 218                cur->pos = 0;
 219                return;
 220        }
 221
 222        for (i = 1; i < xfs_iext_max_recs(ifp); i++) {
 223                if (xfs_iext_rec_is_empty(&cur->leaf->recs[i]))
 224                        break;
 225        }
 226        cur->pos = i - 1;
 227}
 228
 229void
 230xfs_iext_next(
 231        struct xfs_ifork        *ifp,
 232        struct xfs_iext_cursor  *cur)
 233{
 234        if (!cur->leaf) {
 235                ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF);
 236                xfs_iext_first(ifp, cur);
 237                return;
 238        }
 239
 240        ASSERT(cur->pos >= 0);
 241        ASSERT(cur->pos < xfs_iext_max_recs(ifp));
 242
 243        cur->pos++;
 244        if (ifp->if_height > 1 && !xfs_iext_valid(ifp, cur) &&
 245            cur->leaf->next) {
 246                cur->leaf = cur->leaf->next;
 247                cur->pos = 0;
 248        }
 249}
 250
 251void
 252xfs_iext_prev(
 253        struct xfs_ifork        *ifp,
 254        struct xfs_iext_cursor  *cur)
 255{
 256        if (!cur->leaf) {
 257                ASSERT(cur->pos <= 0 || cur->pos >= RECS_PER_LEAF);
 258                xfs_iext_last(ifp, cur);
 259                return;
 260        }
 261
 262        ASSERT(cur->pos >= 0);
 263        ASSERT(cur->pos <= RECS_PER_LEAF);
 264
 265recurse:
 266        do {
 267                cur->pos--;
 268                if (xfs_iext_valid(ifp, cur))
 269                        return;
 270        } while (cur->pos > 0);
 271
 272        if (ifp->if_height > 1 && cur->leaf->prev) {
 273                cur->leaf = cur->leaf->prev;
 274                cur->pos = RECS_PER_LEAF;
 275                goto recurse;
 276        }
 277}
 278
 279static inline int
 280xfs_iext_key_cmp(
 281        struct xfs_iext_node    *node,
 282        int                     n,
 283        xfs_fileoff_t           offset)
 284{
 285        if (node->keys[n] > offset)
 286                return 1;
 287        if (node->keys[n] < offset)
 288                return -1;
 289        return 0;
 290}
 291
 292static inline int
 293xfs_iext_rec_cmp(
 294        struct xfs_iext_rec     *rec,
 295        xfs_fileoff_t           offset)
 296{
 297        uint64_t                rec_offset = rec->lo & XFS_IEXT_STARTOFF_MASK;
 298        uint32_t                rec_len = rec->hi & XFS_IEXT_LENGTH_MASK;
 299
 300        if (rec_offset > offset)
 301                return 1;
 302        if (rec_offset + rec_len <= offset)
 303                return -1;
 304        return 0;
 305}
 306
 307static void *
 308xfs_iext_find_level(
 309        struct xfs_ifork        *ifp,
 310        xfs_fileoff_t           offset,
 311        int                     level)
 312{
 313        struct xfs_iext_node    *node = ifp->if_u1.if_root;
 314        int                     height, i;
 315
 316        if (!ifp->if_height)
 317                return NULL;
 318
 319        for (height = ifp->if_height; height > level; height--) {
 320                for (i = 1; i < KEYS_PER_NODE; i++)
 321                        if (xfs_iext_key_cmp(node, i, offset) > 0)
 322                                break;
 323
 324                node = node->ptrs[i - 1];
 325                if (!node)
 326                        break;
 327        }
 328
 329        return node;
 330}
 331
 332static int
 333xfs_iext_node_pos(
 334        struct xfs_iext_node    *node,
 335        xfs_fileoff_t           offset)
 336{
 337        int                     i;
 338
 339        for (i = 1; i < KEYS_PER_NODE; i++) {
 340                if (xfs_iext_key_cmp(node, i, offset) > 0)
 341                        break;
 342        }
 343
 344        return i - 1;
 345}
 346
 347static int
 348xfs_iext_node_insert_pos(
 349        struct xfs_iext_node    *node,
 350        xfs_fileoff_t           offset)
 351{
 352        int                     i;
 353
 354        for (i = 0; i < KEYS_PER_NODE; i++) {
 355                if (xfs_iext_key_cmp(node, i, offset) > 0)
 356                        return i;
 357        }
 358
 359        return KEYS_PER_NODE;
 360}
 361
 362static int
 363xfs_iext_node_nr_entries(
 364        struct xfs_iext_node    *node,
 365        int                     start)
 366{
 367        int                     i;
 368
 369        for (i = start; i < KEYS_PER_NODE; i++) {
 370                if (node->keys[i] == XFS_IEXT_KEY_INVALID)
 371                        break;
 372        }
 373
 374        return i;
 375}
 376
 377static int
 378xfs_iext_leaf_nr_entries(
 379        struct xfs_ifork        *ifp,
 380        struct xfs_iext_leaf    *leaf,
 381        int                     start)
 382{
 383        int                     i;
 384
 385        for (i = start; i < xfs_iext_max_recs(ifp); i++) {
 386                if (xfs_iext_rec_is_empty(&leaf->recs[i]))
 387                        break;
 388        }
 389
 390        return i;
 391}
 392
 393static inline uint64_t
 394xfs_iext_leaf_key(
 395        struct xfs_iext_leaf    *leaf,
 396        int                     n)
 397{
 398        return leaf->recs[n].lo & XFS_IEXT_STARTOFF_MASK;
 399}
 400
 401static void
 402xfs_iext_grow(
 403        struct xfs_ifork        *ifp)
 404{
 405        struct xfs_iext_node    *node = kmem_zalloc(NODE_SIZE, KM_NOFS);
 406        int                     i;
 407
 408        if (ifp->if_height == 1) {
 409                struct xfs_iext_leaf *prev = ifp->if_u1.if_root;
 410
 411                node->keys[0] = xfs_iext_leaf_key(prev, 0);
 412                node->ptrs[0] = prev;
 413        } else  {
 414                struct xfs_iext_node *prev = ifp->if_u1.if_root;
 415
 416                ASSERT(ifp->if_height > 1);
 417
 418                node->keys[0] = prev->keys[0];
 419                node->ptrs[0] = prev;
 420        }
 421
 422        for (i = 1; i < KEYS_PER_NODE; i++)
 423                node->keys[i] = XFS_IEXT_KEY_INVALID;
 424
 425        ifp->if_u1.if_root = node;
 426        ifp->if_height++;
 427}
 428
 429static void
 430xfs_iext_update_node(
 431        struct xfs_ifork        *ifp,
 432        xfs_fileoff_t           old_offset,
 433        xfs_fileoff_t           new_offset,
 434        int                     level,
 435        void                    *ptr)
 436{
 437        struct xfs_iext_node    *node = ifp->if_u1.if_root;
 438        int                     height, i;
 439
 440        for (height = ifp->if_height; height > level; height--) {
 441                for (i = 0; i < KEYS_PER_NODE; i++) {
 442                        if (i > 0 && xfs_iext_key_cmp(node, i, old_offset) > 0)
 443                                break;
 444                        if (node->keys[i] == old_offset)
 445                                node->keys[i] = new_offset;
 446                }
 447                node = node->ptrs[i - 1];
 448                ASSERT(node);
 449        }
 450
 451        ASSERT(node == ptr);
 452}
 453
 454static struct xfs_iext_node *
 455xfs_iext_split_node(
 456        struct xfs_iext_node    **nodep,
 457        int                     *pos,
 458        int                     *nr_entries)
 459{
 460        struct xfs_iext_node    *node = *nodep;
 461        struct xfs_iext_node    *new = kmem_zalloc(NODE_SIZE, KM_NOFS);
 462        const int               nr_move = KEYS_PER_NODE / 2;
 463        int                     nr_keep = nr_move + (KEYS_PER_NODE & 1);
 464        int                     i = 0;
 465
 466        /* for sequential append operations just spill over into the new node */
 467        if (*pos == KEYS_PER_NODE) {
 468                *nodep = new;
 469                *pos = 0;
 470                *nr_entries = 0;
 471                goto done;
 472        }
 473
 474
 475        for (i = 0; i < nr_move; i++) {
 476                new->keys[i] = node->keys[nr_keep + i];
 477                new->ptrs[i] = node->ptrs[nr_keep + i];
 478
 479                node->keys[nr_keep + i] = XFS_IEXT_KEY_INVALID;
 480                node->ptrs[nr_keep + i] = NULL;
 481        }
 482
 483        if (*pos >= nr_keep) {
 484                *nodep = new;
 485                *pos -= nr_keep;
 486                *nr_entries = nr_move;
 487        } else {
 488                *nr_entries = nr_keep;
 489        }
 490done:
 491        for (; i < KEYS_PER_NODE; i++)
 492                new->keys[i] = XFS_IEXT_KEY_INVALID;
 493        return new;
 494}
 495
 496static void
 497xfs_iext_insert_node(
 498        struct xfs_ifork        *ifp,
 499        uint64_t                offset,
 500        void                    *ptr,
 501        int                     level)
 502{
 503        struct xfs_iext_node    *node, *new;
 504        int                     i, pos, nr_entries;
 505
 506again:
 507        if (ifp->if_height < level)
 508                xfs_iext_grow(ifp);
 509
 510        new = NULL;
 511        node = xfs_iext_find_level(ifp, offset, level);
 512        pos = xfs_iext_node_insert_pos(node, offset);
 513        nr_entries = xfs_iext_node_nr_entries(node, pos);
 514
 515        ASSERT(pos >= nr_entries || xfs_iext_key_cmp(node, pos, offset) != 0);
 516        ASSERT(nr_entries <= KEYS_PER_NODE);
 517
 518        if (nr_entries == KEYS_PER_NODE)
 519                new = xfs_iext_split_node(&node, &pos, &nr_entries);
 520
 521        /*
 522         * Update the pointers in higher levels if the first entry changes
 523         * in an existing node.
 524         */
 525        if (node != new && pos == 0 && nr_entries > 0)
 526                xfs_iext_update_node(ifp, node->keys[0], offset, level, node);
 527
 528        for (i = nr_entries; i > pos; i--) {
 529                node->keys[i] = node->keys[i - 1];
 530                node->ptrs[i] = node->ptrs[i - 1];
 531        }
 532        node->keys[pos] = offset;
 533        node->ptrs[pos] = ptr;
 534
 535        if (new) {
 536                offset = new->keys[0];
 537                ptr = new;
 538                level++;
 539                goto again;
 540        }
 541}
 542
 543static struct xfs_iext_leaf *
 544xfs_iext_split_leaf(
 545        struct xfs_iext_cursor  *cur,
 546        int                     *nr_entries)
 547{
 548        struct xfs_iext_leaf    *leaf = cur->leaf;
 549        struct xfs_iext_leaf    *new = kmem_zalloc(NODE_SIZE, KM_NOFS);
 550        const int               nr_move = RECS_PER_LEAF / 2;
 551        int                     nr_keep = nr_move + (RECS_PER_LEAF & 1);
 552        int                     i;
 553
 554        /* for sequential append operations just spill over into the new node */
 555        if (cur->pos == RECS_PER_LEAF) {
 556                cur->leaf = new;
 557                cur->pos = 0;
 558                *nr_entries = 0;
 559                goto done;
 560        }
 561
 562        for (i = 0; i < nr_move; i++) {
 563                new->recs[i] = leaf->recs[nr_keep + i];
 564                xfs_iext_rec_clear(&leaf->recs[nr_keep + i]);
 565        }
 566
 567        if (cur->pos >= nr_keep) {
 568                cur->leaf = new;
 569                cur->pos -= nr_keep;
 570                *nr_entries = nr_move;
 571        } else {
 572                *nr_entries = nr_keep;
 573        }
 574done:
 575        if (leaf->next)
 576                leaf->next->prev = new;
 577        new->next = leaf->next;
 578        new->prev = leaf;
 579        leaf->next = new;
 580        return new;
 581}
 582
 583static void
 584xfs_iext_alloc_root(
 585        struct xfs_ifork        *ifp,
 586        struct xfs_iext_cursor  *cur)
 587{
 588        ASSERT(ifp->if_bytes == 0);
 589
 590        ifp->if_u1.if_root = kmem_zalloc(sizeof(struct xfs_iext_rec), KM_NOFS);
 591        ifp->if_height = 1;
 592
 593        /* now that we have a node step into it */
 594        cur->leaf = ifp->if_u1.if_root;
 595        cur->pos = 0;
 596}
 597
 598static void
 599xfs_iext_realloc_root(
 600        struct xfs_ifork        *ifp,
 601        struct xfs_iext_cursor  *cur)
 602{
 603        size_t new_size = ifp->if_bytes + sizeof(struct xfs_iext_rec);
 604        void *new;
 605
 606        /* account for the prev/next pointers */
 607        if (new_size / sizeof(struct xfs_iext_rec) == RECS_PER_LEAF)
 608                new_size = NODE_SIZE;
 609
 610        new = kmem_realloc(ifp->if_u1.if_root, new_size, KM_NOFS);
 611        memset(new + ifp->if_bytes, 0, new_size - ifp->if_bytes);
 612        ifp->if_u1.if_root = new;
 613        cur->leaf = new;
 614}
 615
 616/*
 617 * Increment the sequence counter on extent tree changes. If we are on a COW
 618 * fork, this allows the writeback code to skip looking for a COW extent if the
 619 * COW fork hasn't changed. We use WRITE_ONCE here to ensure the update to the
 620 * sequence counter is seen before the modifications to the extent tree itself
 621 * take effect.
 622 */
 623static inline void xfs_iext_inc_seq(struct xfs_ifork *ifp, int state)
 624{
 625        WRITE_ONCE(ifp->if_seq, READ_ONCE(ifp->if_seq) + 1);
 626}
 627
 628void
 629xfs_iext_insert(
 630        struct xfs_inode        *ip,
 631        struct xfs_iext_cursor  *cur,
 632        struct xfs_bmbt_irec    *irec,
 633        int                     state)
 634{
 635        struct xfs_ifork        *ifp = xfs_iext_state_to_fork(ip, state);
 636        xfs_fileoff_t           offset = irec->br_startoff;
 637        struct xfs_iext_leaf    *new = NULL;
 638        int                     nr_entries, i;
 639
 640        xfs_iext_inc_seq(ifp, state);
 641
 642        if (ifp->if_height == 0)
 643                xfs_iext_alloc_root(ifp, cur);
 644        else if (ifp->if_height == 1)
 645                xfs_iext_realloc_root(ifp, cur);
 646
 647        nr_entries = xfs_iext_leaf_nr_entries(ifp, cur->leaf, cur->pos);
 648        ASSERT(nr_entries <= RECS_PER_LEAF);
 649        ASSERT(cur->pos >= nr_entries ||
 650               xfs_iext_rec_cmp(cur_rec(cur), irec->br_startoff) != 0);
 651
 652        if (nr_entries == RECS_PER_LEAF)
 653                new = xfs_iext_split_leaf(cur, &nr_entries);
 654
 655        /*
 656         * Update the pointers in higher levels if the first entry changes
 657         * in an existing node.
 658         */
 659        if (cur->leaf != new && cur->pos == 0 && nr_entries > 0) {
 660                xfs_iext_update_node(ifp, xfs_iext_leaf_key(cur->leaf, 0),
 661                                offset, 1, cur->leaf);
 662        }
 663
 664        for (i = nr_entries; i > cur->pos; i--)
 665                cur->leaf->recs[i] = cur->leaf->recs[i - 1];
 666        xfs_iext_set(cur_rec(cur), irec);
 667        ifp->if_bytes += sizeof(struct xfs_iext_rec);
 668
 669        trace_xfs_iext_insert(ip, cur, state, _RET_IP_);
 670
 671        if (new)
 672                xfs_iext_insert_node(ifp, xfs_iext_leaf_key(new, 0), new, 2);
 673}
 674
 675static struct xfs_iext_node *
 676xfs_iext_rebalance_node(
 677        struct xfs_iext_node    *parent,
 678        int                     *pos,
 679        struct xfs_iext_node    *node,
 680        int                     nr_entries)
 681{
 682        /*
 683         * If the neighbouring nodes are completely full, or have different
 684         * parents, we might never be able to merge our node, and will only
 685         * delete it once the number of entries hits zero.
 686         */
 687        if (nr_entries == 0)
 688                return node;
 689
 690        if (*pos > 0) {
 691                struct xfs_iext_node *prev = parent->ptrs[*pos - 1];
 692                int nr_prev = xfs_iext_node_nr_entries(prev, 0), i;
 693
 694                if (nr_prev + nr_entries <= KEYS_PER_NODE) {
 695                        for (i = 0; i < nr_entries; i++) {
 696                                prev->keys[nr_prev + i] = node->keys[i];
 697                                prev->ptrs[nr_prev + i] = node->ptrs[i];
 698                        }
 699                        return node;
 700                }
 701        }
 702
 703        if (*pos + 1 < xfs_iext_node_nr_entries(parent, *pos)) {
 704                struct xfs_iext_node *next = parent->ptrs[*pos + 1];
 705                int nr_next = xfs_iext_node_nr_entries(next, 0), i;
 706
 707                if (nr_entries + nr_next <= KEYS_PER_NODE) {
 708                        /*
 709                         * Merge the next node into this node so that we don't
 710                         * have to do an additional update of the keys in the
 711                         * higher levels.
 712                         */
 713                        for (i = 0; i < nr_next; i++) {
 714                                node->keys[nr_entries + i] = next->keys[i];
 715                                node->ptrs[nr_entries + i] = next->ptrs[i];
 716                        }
 717
 718                        ++*pos;
 719                        return next;
 720                }
 721        }
 722
 723        return NULL;
 724}
 725
 726static void
 727xfs_iext_remove_node(
 728        struct xfs_ifork        *ifp,
 729        xfs_fileoff_t           offset,
 730        void                    *victim)
 731{
 732        struct xfs_iext_node    *node, *parent;
 733        int                     level = 2, pos, nr_entries, i;
 734
 735        ASSERT(level <= ifp->if_height);
 736        node = xfs_iext_find_level(ifp, offset, level);
 737        pos = xfs_iext_node_pos(node, offset);
 738again:
 739        ASSERT(node->ptrs[pos]);
 740        ASSERT(node->ptrs[pos] == victim);
 741        kmem_free(victim);
 742
 743        nr_entries = xfs_iext_node_nr_entries(node, pos) - 1;
 744        offset = node->keys[0];
 745        for (i = pos; i < nr_entries; i++) {
 746                node->keys[i] = node->keys[i + 1];
 747                node->ptrs[i] = node->ptrs[i + 1];
 748        }
 749        node->keys[nr_entries] = XFS_IEXT_KEY_INVALID;
 750        node->ptrs[nr_entries] = NULL;
 751
 752        if (pos == 0 && nr_entries > 0) {
 753                xfs_iext_update_node(ifp, offset, node->keys[0], level, node);
 754                offset = node->keys[0];
 755        }
 756
 757        if (nr_entries >= KEYS_PER_NODE / 2)
 758                return;
 759
 760        if (level < ifp->if_height) {
 761                /*
 762                 * If we aren't at the root yet try to find a neighbour node to
 763                 * merge with (or delete the node if it is empty), and then
 764                 * recurse up to the next level.
 765                 */
 766                level++;
 767                parent = xfs_iext_find_level(ifp, offset, level);
 768                pos = xfs_iext_node_pos(parent, offset);
 769
 770                ASSERT(pos != KEYS_PER_NODE);
 771                ASSERT(parent->ptrs[pos] == node);
 772
 773                node = xfs_iext_rebalance_node(parent, &pos, node, nr_entries);
 774                if (node) {
 775                        victim = node;
 776                        node = parent;
 777                        goto again;
 778                }
 779        } else if (nr_entries == 1) {
 780                /*
 781                 * If we are at the root and only one entry is left we can just
 782                 * free this node and update the root pointer.
 783                 */
 784                ASSERT(node == ifp->if_u1.if_root);
 785                ifp->if_u1.if_root = node->ptrs[0];
 786                ifp->if_height--;
 787                kmem_free(node);
 788        }
 789}
 790
 791static void
 792xfs_iext_rebalance_leaf(
 793        struct xfs_ifork        *ifp,
 794        struct xfs_iext_cursor  *cur,
 795        struct xfs_iext_leaf    *leaf,
 796        xfs_fileoff_t           offset,
 797        int                     nr_entries)
 798{
 799        /*
 800         * If the neighbouring nodes are completely full we might never be able
 801         * to merge our node, and will only delete it once the number of
 802         * entries hits zero.
 803         */
 804        if (nr_entries == 0)
 805                goto remove_node;
 806
 807        if (leaf->prev) {
 808                int nr_prev = xfs_iext_leaf_nr_entries(ifp, leaf->prev, 0), i;
 809
 810                if (nr_prev + nr_entries <= RECS_PER_LEAF) {
 811                        for (i = 0; i < nr_entries; i++)
 812                                leaf->prev->recs[nr_prev + i] = leaf->recs[i];
 813
 814                        if (cur->leaf == leaf) {
 815                                cur->leaf = leaf->prev;
 816                                cur->pos += nr_prev;
 817                        }
 818                        goto remove_node;
 819                }
 820        }
 821
 822        if (leaf->next) {
 823                int nr_next = xfs_iext_leaf_nr_entries(ifp, leaf->next, 0), i;
 824
 825                if (nr_entries + nr_next <= RECS_PER_LEAF) {
 826                        /*
 827                         * Merge the next node into this node so that we don't
 828                         * have to do an additional update of the keys in the
 829                         * higher levels.
 830                         */
 831                        for (i = 0; i < nr_next; i++) {
 832                                leaf->recs[nr_entries + i] =
 833                                        leaf->next->recs[i];
 834                        }
 835
 836                        if (cur->leaf == leaf->next) {
 837                                cur->leaf = leaf;
 838                                cur->pos += nr_entries;
 839                        }
 840
 841                        offset = xfs_iext_leaf_key(leaf->next, 0);
 842                        leaf = leaf->next;
 843                        goto remove_node;
 844                }
 845        }
 846
 847        return;
 848remove_node:
 849        if (leaf->prev)
 850                leaf->prev->next = leaf->next;
 851        if (leaf->next)
 852                leaf->next->prev = leaf->prev;
 853        xfs_iext_remove_node(ifp, offset, leaf);
 854}
 855
 856static void
 857xfs_iext_free_last_leaf(
 858        struct xfs_ifork        *ifp)
 859{
 860        ifp->if_height--;
 861        kmem_free(ifp->if_u1.if_root);
 862        ifp->if_u1.if_root = NULL;
 863}
 864
 865void
 866xfs_iext_remove(
 867        struct xfs_inode        *ip,
 868        struct xfs_iext_cursor  *cur,
 869        int                     state)
 870{
 871        struct xfs_ifork        *ifp = xfs_iext_state_to_fork(ip, state);
 872        struct xfs_iext_leaf    *leaf = cur->leaf;
 873        xfs_fileoff_t           offset = xfs_iext_leaf_key(leaf, 0);
 874        int                     i, nr_entries;
 875
 876        trace_xfs_iext_remove(ip, cur, state, _RET_IP_);
 877
 878        ASSERT(ifp->if_height > 0);
 879        ASSERT(ifp->if_u1.if_root != NULL);
 880        ASSERT(xfs_iext_valid(ifp, cur));
 881
 882        xfs_iext_inc_seq(ifp, state);
 883
 884        nr_entries = xfs_iext_leaf_nr_entries(ifp, leaf, cur->pos) - 1;
 885        for (i = cur->pos; i < nr_entries; i++)
 886                leaf->recs[i] = leaf->recs[i + 1];
 887        xfs_iext_rec_clear(&leaf->recs[nr_entries]);
 888        ifp->if_bytes -= sizeof(struct xfs_iext_rec);
 889
 890        if (cur->pos == 0 && nr_entries > 0) {
 891                xfs_iext_update_node(ifp, offset, xfs_iext_leaf_key(leaf, 0), 1,
 892                                leaf);
 893                offset = xfs_iext_leaf_key(leaf, 0);
 894        } else if (cur->pos == nr_entries) {
 895                if (ifp->if_height > 1 && leaf->next)
 896                        cur->leaf = leaf->next;
 897                else
 898                        cur->leaf = NULL;
 899                cur->pos = 0;
 900        }
 901
 902        if (nr_entries >= RECS_PER_LEAF / 2)
 903                return;
 904
 905        if (ifp->if_height > 1)
 906                xfs_iext_rebalance_leaf(ifp, cur, leaf, offset, nr_entries);
 907        else if (nr_entries == 0)
 908                xfs_iext_free_last_leaf(ifp);
 909}
 910
 911/*
 912 * Lookup the extent covering bno.
 913 *
 914 * If there is an extent covering bno return the extent index, and store the
 915 * expanded extent structure in *gotp, and the extent cursor in *cur.
 916 * If there is no extent covering bno, but there is an extent after it (e.g.
 917 * it lies in a hole) return that extent in *gotp and its cursor in *cur
 918 * instead.
 919 * If bno is beyond the last extent return false, and return an invalid
 920 * cursor value.
 921 */
 922bool
 923xfs_iext_lookup_extent(
 924        struct xfs_inode        *ip,
 925        struct xfs_ifork        *ifp,
 926        xfs_fileoff_t           offset,
 927        struct xfs_iext_cursor  *cur,
 928        struct xfs_bmbt_irec    *gotp)
 929{
 930        XFS_STATS_INC(ip->i_mount, xs_look_exlist);
 931
 932        cur->leaf = xfs_iext_find_level(ifp, offset, 1);
 933        if (!cur->leaf) {
 934                cur->pos = 0;
 935                return false;
 936        }
 937
 938        for (cur->pos = 0; cur->pos < xfs_iext_max_recs(ifp); cur->pos++) {
 939                struct xfs_iext_rec *rec = cur_rec(cur);
 940
 941                if (xfs_iext_rec_is_empty(rec))
 942                        break;
 943                if (xfs_iext_rec_cmp(rec, offset) >= 0)
 944                        goto found;
 945        }
 946
 947        /* Try looking in the next node for an entry > offset */
 948        if (ifp->if_height == 1 || !cur->leaf->next)
 949                return false;
 950        cur->leaf = cur->leaf->next;
 951        cur->pos = 0;
 952        if (!xfs_iext_valid(ifp, cur))
 953                return false;
 954found:
 955        xfs_iext_get(gotp, cur_rec(cur));
 956        return true;
 957}
 958
 959/*
 960 * Returns the last extent before end, and if this extent doesn't cover
 961 * end, update end to the end of the extent.
 962 */
 963bool
 964xfs_iext_lookup_extent_before(
 965        struct xfs_inode        *ip,
 966        struct xfs_ifork        *ifp,
 967        xfs_fileoff_t           *end,
 968        struct xfs_iext_cursor  *cur,
 969        struct xfs_bmbt_irec    *gotp)
 970{
 971        /* could be optimized to not even look up the next on a match.. */
 972        if (xfs_iext_lookup_extent(ip, ifp, *end - 1, cur, gotp) &&
 973            gotp->br_startoff <= *end - 1)
 974                return true;
 975        if (!xfs_iext_prev_extent(ifp, cur, gotp))
 976                return false;
 977        *end = gotp->br_startoff + gotp->br_blockcount;
 978        return true;
 979}
 980
 981void
 982xfs_iext_update_extent(
 983        struct xfs_inode        *ip,
 984        int                     state,
 985        struct xfs_iext_cursor  *cur,
 986        struct xfs_bmbt_irec    *new)
 987{
 988        struct xfs_ifork        *ifp = xfs_iext_state_to_fork(ip, state);
 989
 990        xfs_iext_inc_seq(ifp, state);
 991
 992        if (cur->pos == 0) {
 993                struct xfs_bmbt_irec    old;
 994
 995                xfs_iext_get(&old, cur_rec(cur));
 996                if (new->br_startoff != old.br_startoff) {
 997                        xfs_iext_update_node(ifp, old.br_startoff,
 998                                        new->br_startoff, 1, cur->leaf);
 999                }
1000        }
1001
1002        trace_xfs_bmap_pre_update(ip, cur, state, _RET_IP_);
1003        xfs_iext_set(cur_rec(cur), new);
1004        trace_xfs_bmap_post_update(ip, cur, state, _RET_IP_);
1005}
1006
1007/*
1008 * Return true if the cursor points at an extent and return the extent structure
1009 * in gotp.  Else return false.
1010 */
1011bool
1012xfs_iext_get_extent(
1013        struct xfs_ifork        *ifp,
1014        struct xfs_iext_cursor  *cur,
1015        struct xfs_bmbt_irec    *gotp)
1016{
1017        if (!xfs_iext_valid(ifp, cur))
1018                return false;
1019        xfs_iext_get(gotp, cur_rec(cur));
1020        return true;
1021}
1022
1023/*
1024 * This is a recursive function, because of that we need to be extremely
1025 * careful with stack usage.
1026 */
1027static void
1028xfs_iext_destroy_node(
1029        struct xfs_iext_node    *node,
1030        int                     level)
1031{
1032        int                     i;
1033
1034        if (level > 1) {
1035                for (i = 0; i < KEYS_PER_NODE; i++) {
1036                        if (node->keys[i] == XFS_IEXT_KEY_INVALID)
1037                                break;
1038                        xfs_iext_destroy_node(node->ptrs[i], level - 1);
1039                }
1040        }
1041
1042        kmem_free(node);
1043}
1044
1045void
1046xfs_iext_destroy(
1047        struct xfs_ifork        *ifp)
1048{
1049        xfs_iext_destroy_node(ifp->if_u1.if_root, ifp->if_height);
1050
1051        ifp->if_bytes = 0;
1052        ifp->if_height = 0;
1053        ifp->if_u1.if_root = NULL;
1054}
1055