fs/xfs/libxfs/xfs_btree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_bit.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_trans.h"
  16 #include "xfs_buf_item.h"
  17 #include "xfs_btree.h"
  18 #include "xfs_errortag.h"
  19 #include "xfs_error.h"
  20 #include "xfs_trace.h"
  21 #include "xfs_alloc.h"
  22 #include "xfs_log.h"
  23 #include "xfs_btree_staging.h"
  24
  25 /*
  26  * Cursor allocation zone.
  27  */
  28 kmem_zone_t     *xfs_btree_cur_zone;
  29
  30 /*
  31  * Btree magic numbers.
  32  */
  33 static const uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
  34         { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, 0, XFS_BMAP_MAGIC, XFS_IBT_MAGIC,
  35           XFS_FIBT_MAGIC, 0 },
  36         { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC, XFS_RMAP_CRC_MAGIC,
  37           XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC,
  38           XFS_REFC_CRC_MAGIC }
  39 };
  40
  41 uint32_t
  42 xfs_btree_magic(
  43         int                     crc,
  44         xfs_btnum_t             btnum)
  45 {
  46         uint32_t                magic = xfs_magics[crc][btnum];
  47
  48         /* Ensure we asked for crc for crc-only magics. */
  49         ASSERT(magic != 0);
  50         return magic;
  51 }
  52
  53 /*
  54  * Check a long btree block header.  Return the address of the failing check,
  55  * or NULL if everything is ok.
  56  */
  57 xfs_failaddr_t
  58 __xfs_btree_check_lblock(
  59         struct xfs_btree_cur    *cur,
  60         struct xfs_btree_block  *block,
  61         int                     level,
  62         struct xfs_buf          *bp)
  63 {
  64         struct xfs_mount        *mp = cur->bc_mp;
  65         xfs_btnum_t             btnum = cur->bc_btnum;
  66         int                     crc = xfs_sb_version_hascrc(&mp->m_sb);
  67
  68         if (crc) {
  69                 if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
  70                         return __this_address;
  71                 if (block->bb_u.l.bb_blkno !=
  72                     cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
  73                         return __this_address;
  74                 if (block->bb_u.l.bb_pad != cpu_to_be32(0))
  75                         return __this_address;
  76         }
  77
  78         if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
  79                 return __this_address;
  80         if (be16_to_cpu(block->bb_level) != level)
  81                 return __this_address;
  82         if (be16_to_cpu(block->bb_numrecs) >
  83             cur->bc_ops->get_maxrecs(cur, level))
  84                 return __this_address;
  85         if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) &&
  86             !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_leftsib),
  87                         level + 1))
  88                 return __this_address;
  89         if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) &&
  90             !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_rightsib),
  91                         level + 1))
  92                 return __this_address;
  93
  94         return NULL;
  95 }
  96
  97 /* Check a long btree block header. */
  98 static int
  99 xfs_btree_check_lblock(
 100         struct xfs_btree_cur    *cur,
 101         struct xfs_btree_block  *block,
 102         int                     level,
 103         struct xfs_buf          *bp)
 104 {
 105         struct xfs_mount        *mp = cur->bc_mp;
 106         xfs_failaddr_t          fa;
 107
 108         fa = __xfs_btree_check_lblock(cur, block, level, bp);
 109         if (XFS_IS_CORRUPT(mp, fa != NULL) ||
 110             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BTREE_CHECK_LBLOCK)) {
 111                 if (bp)
 112                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 113                 return -EFSCORRUPTED;
 114         }
 115         return 0;
 116 }
 117
 118 /*
 119  * Check a short btree block header.  Return the address of the failing check,
 120  * or NULL if everything is ok.
 121  */
 122 xfs_failaddr_t
 123 __xfs_btree_check_sblock(
 124         struct xfs_btree_cur    *cur,
 125         struct xfs_btree_block  *block,
 126         int                     level,
 127         struct xfs_buf          *bp)
 128 {
 129         struct xfs_mount        *mp = cur->bc_mp;
 130         xfs_btnum_t             btnum = cur->bc_btnum;
 131         int                     crc = xfs_sb_version_hascrc(&mp->m_sb);
 132
 133         if (crc) {
 134                 if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
 135                         return __this_address;
 136                 if (block->bb_u.s.bb_blkno !=
 137                     cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
 138                         return __this_address;
 139         }
 140
 141         if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
 142                 return __this_address;
 143         if (be16_to_cpu(block->bb_level) != level)
 144                 return __this_address;
 145         if (be16_to_cpu(block->bb_numrecs) >
 146             cur->bc_ops->get_maxrecs(cur, level))
 147                 return __this_address;
 148         if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) &&
 149             !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_leftsib),
 150                         level + 1))
 151                 return __this_address;
 152         if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) &&
 153             !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_rightsib),
 154                         level + 1))
 155                 return __this_address;
 156
 157         return NULL;
 158 }
 159
 160 /* Check a short btree block header. */
 161 STATIC int
 162 xfs_btree_check_sblock(
 163         struct xfs_btree_cur    *cur,
 164         struct xfs_btree_block  *block,
 165         int                     level,
 166         struct xfs_buf          *bp)
 167 {
 168         struct xfs_mount        *mp = cur->bc_mp;
 169         xfs_failaddr_t          fa;
 170
 171         fa = __xfs_btree_check_sblock(cur, block, level, bp);
 172         if (XFS_IS_CORRUPT(mp, fa != NULL) ||
 173             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BTREE_CHECK_SBLOCK)) {
 174                 if (bp)
 175                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 176                 return -EFSCORRUPTED;
 177         }
 178         return 0;
 179 }
 180
 181 /*
 182  * Debug routine: check that block header is ok.
 183  */
 184 int
 185 xfs_btree_check_block(
 186         struct xfs_btree_cur    *cur,   /* btree cursor */
 187         struct xfs_btree_block  *block, /* generic btree block pointer */
 188         int                     level,  /* level of the btree block */
 189         struct xfs_buf          *bp)    /* buffer containing block, if any */
 190 {
 191         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 192                 return xfs_btree_check_lblock(cur, block, level, bp);
 193         else
 194                 return xfs_btree_check_sblock(cur, block, level, bp);
 195 }
 196
 197 /* Check that this long pointer is valid and points within the fs. */
 198 bool
 199 xfs_btree_check_lptr(
 200         struct xfs_btree_cur    *cur,
 201         xfs_fsblock_t           fsbno,
 202         int                     level)
 203 {
 204         if (level <= 0)
 205                 return false;
 206         return xfs_verify_fsbno(cur->bc_mp, fsbno);
 207 }
 208
 209 /* Check that this short pointer is valid and points within the AG. */
 210 bool
 211 xfs_btree_check_sptr(
 212         struct xfs_btree_cur    *cur,
 213         xfs_agblock_t           agbno,
 214         int                     level)
 215 {
 216         if (level <= 0)
 217                 return false;
 218         return xfs_verify_agbno(cur->bc_mp, cur->bc_ag.agno, agbno);
 219 }
 220
 221 /*
 222  * Check that a given (indexed) btree pointer at a certain level of a
 223  * btree is valid and doesn't point past where it should.
 224  */
 225 static int
 226 xfs_btree_check_ptr(
 227         struct xfs_btree_cur    *cur,
 228         union xfs_btree_ptr     *ptr,
 229         int                     index,
 230         int                     level)
 231 {
 232         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 233                 if (xfs_btree_check_lptr(cur, be64_to_cpu((&ptr->l)[index]),
 234                                 level))
 235                         return 0;
 236                 xfs_err(cur->bc_mp,
 237 "Inode %llu fork %d: Corrupt btree %d pointer at level %d index %d.",
 238                                 cur->bc_ino.ip->i_ino,
 239                                 cur->bc_ino.whichfork, cur->bc_btnum,
 240                                 level, index);
 241         } else {
 242                 if (xfs_btree_check_sptr(cur, be32_to_cpu((&ptr->s)[index]),
 243                                 level))
 244                         return 0;
 245                 xfs_err(cur->bc_mp,
 246 "AG %u: Corrupt btree %d pointer at level %d index %d.",
 247                                 cur->bc_ag.agno, cur->bc_btnum,
 248                                 level, index);
 249         }
 250
 251         return -EFSCORRUPTED;
 252 }
 253
 254 #ifdef DEBUG
 255 # define xfs_btree_debug_check_ptr      xfs_btree_check_ptr
 256 #else
 257 # define xfs_btree_debug_check_ptr(...) (0)
 258 #endif
 259
 260 /*
 261  * Calculate CRC on the whole btree block and stuff it into the
 262  * long-form btree header.
 263  *
 264  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 265  * it into the buffer so recovery knows what the last modification was that made
 266  * it to disk.
 267  */
 268 void
 269 xfs_btree_lblock_calc_crc(
 270         struct xfs_buf          *bp)
 271 {
 272         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 273         struct xfs_buf_log_item *bip = bp->b_log_item;
 274
 275         if (!xfs_sb_version_hascrc(&bp->b_mount->m_sb))
 276                 return;
 277         if (bip)
 278                 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 279         xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 280 }
 281
 282 bool
 283 xfs_btree_lblock_verify_crc(
 284         struct xfs_buf          *bp)
 285 {
 286         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 287         struct xfs_mount        *mp = bp->b_mount;
 288
 289         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 290                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
 291                         return false;
 292                 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 293         }
 294
 295         return true;
 296 }
 297
 298 /*
 299  * Calculate CRC on the whole btree block and stuff it into the
 300  * short-form btree header.
 301  *
 302  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 303  * it into the buffer so recovery knows what the last modification was that made
 304  * it to disk.
 305  */
 306 void
 307 xfs_btree_sblock_calc_crc(
 308         struct xfs_buf          *bp)
 309 {
 310         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 311         struct xfs_buf_log_item *bip = bp->b_log_item;
 312
 313         if (!xfs_sb_version_hascrc(&bp->b_mount->m_sb))
 314                 return;
 315         if (bip)
 316                 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 317         xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 318 }
 319
 320 bool
 321 xfs_btree_sblock_verify_crc(
 322         struct xfs_buf          *bp)
 323 {
 324         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 325         struct xfs_mount        *mp = bp->b_mount;
 326
 327         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 328                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
 329                         return false;
 330                 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 331         }
 332
 333         return true;
 334 }
 335
 336 static int
 337 xfs_btree_free_block(
 338         struct xfs_btree_cur    *cur,
 339         struct xfs_buf          *bp)
 340 {
 341         int                     error;
 342
 343         error = cur->bc_ops->free_block(cur, bp);
 344         if (!error) {
 345                 xfs_trans_binval(cur->bc_tp, bp);
 346                 XFS_BTREE_STATS_INC(cur, free);
 347         }
 348         return error;
 349 }
 350
 351 /*
 352  * Delete the btree cursor.
 353  */
 354 void
 355 xfs_btree_del_cursor(
 356         struct xfs_btree_cur    *cur,           /* btree cursor */
 357         int                     error)          /* del because of error */
 358 {
 359         int                     i;              /* btree level */
 360
 361         /*
 362          * Clear the buffer pointers and release the buffers. If we're doing
 363          * this because of an error, inspect all of the entries in the bc_bufs
 364          * array for buffers to be unlocked. This is because some of the btree
 365          * code works from level n down to 0, and if we get an error along the
 366          * way we won't have initialized all the entries down to 0.
 367          */
 368         for (i = 0; i < cur->bc_nlevels; i++) {
 369                 if (cur->bc_bufs[i])
 370                         xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
 371                 else if (!error)
 372                         break;
 373         }
 374
 375         ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP || cur->bc_ino.allocated == 0 ||
 376                XFS_FORCED_SHUTDOWN(cur->bc_mp));
 377         if (unlikely(cur->bc_flags & XFS_BTREE_STAGING))
 378                 kmem_free(cur->bc_ops);
 379         kmem_cache_free(xfs_btree_cur_zone, cur);
 380 }
 381
 382 /*
 383  * Duplicate the btree cursor.
 384  * Allocate a new one, copy the record, re-get the buffers.
 385  */
 386 int                                     /* error */
 387 xfs_btree_dup_cursor(
 388         xfs_btree_cur_t *cur,           /* input cursor */
 389         xfs_btree_cur_t **ncur)         /* output cursor */
 390 {
 391         struct xfs_buf  *bp;            /* btree block's buffer pointer */
 392         int             error;          /* error return value */
 393         int             i;              /* level number of btree block */
 394         xfs_mount_t     *mp;            /* mount structure for filesystem */
 395         xfs_btree_cur_t *new;           /* new cursor value */
 396         xfs_trans_t     *tp;            /* transaction pointer, can be NULL */
 397
 398         tp = cur->bc_tp;
 399         mp = cur->bc_mp;
 400
 401         /*
 402          * Allocate a new cursor like the old one.
 403          */
 404         new = cur->bc_ops->dup_cursor(cur);
 405
 406         /*
 407          * Copy the record currently in the cursor.
 408          */
 409         new->bc_rec = cur->bc_rec;
 410
 411         /*
 412          * For each level current, re-get the buffer and copy the ptr value.
 413          */
 414         for (i = 0; i < new->bc_nlevels; i++) {
 415                 new->bc_ptrs[i] = cur->bc_ptrs[i];
 416                 new->bc_ra[i] = cur->bc_ra[i];
 417                 bp = cur->bc_bufs[i];
 418                 if (bp) {
 419                         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
 420                                                    XFS_BUF_ADDR(bp), mp->m_bsize,
 421                                                    0, &bp,
 422                                                    cur->bc_ops->buf_ops);
 423                         if (error) {
 424                                 xfs_btree_del_cursor(new, error);
 425                                 *ncur = NULL;
 426                                 return error;
 427                         }
 428                 }
 429                 new->bc_bufs[i] = bp;
 430         }
 431         *ncur = new;
 432         return 0;
 433 }
 434
 435 /*
 436  * XFS btree block layout and addressing:
 437  *
 438  * There are two types of blocks in the btree: leaf and non-leaf blocks.
 439  *
 440  * The leaf record start with a header then followed by records containing
 441  * the values.  A non-leaf block also starts with the same header, and
 442  * then first contains lookup keys followed by an equal number of pointers
 443  * to the btree blocks at the previous level.
 444  *
 445  *              +--------+-------+-------+-------+-------+-------+-------+
 446  * Leaf:        | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
 447  *              +--------+-------+-------+-------+-------+-------+-------+
 448  *
 449  *              +--------+-------+-------+-------+-------+-------+-------+
 450  * Non-Leaf:    | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
 451  *              +--------+-------+-------+-------+-------+-------+-------+
 452  *
 453  * The header is called struct xfs_btree_block for reasons better left unknown
 454  * and comes in different versions for short (32bit) and long (64bit) block
 455  * pointers.  The record and key structures are defined by the btree instances
 456  * and opaque to the btree core.  The block pointers are simple disk endian
 457  * integers, available in a short (32bit) and long (64bit) variant.
 458  *
 459  * The helpers below calculate the offset of a given record, key or pointer
 460  * into a btree block (xfs_btree_*_offset) or return a pointer to the given
 461  * record, key or pointer (xfs_btree_*_addr).  Note that all addressing
 462  * inside the btree block is done using indices starting at one, not zero!
 463  *
 464  * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing
 465  * overlapping intervals.  In such a tree, records are still sorted lowest to
 466  * highest and indexed by the smallest key value that refers to the record.
 467  * However, nodes are different: each pointer has two associated keys -- one
 468  * indexing the lowest key available in the block(s) below (the same behavior
 469  * as the key in a regular btree) and another indexing the highest key
 470  * available in the block(s) below.  Because records are /not/ sorted by the
 471  * highest key, all leaf block updates require us to compute the highest key
 472  * that matches any record in the leaf and to recursively update the high keys
 473  * in the nodes going further up in the tree, if necessary.  Nodes look like
 474  * this:
 475  *
 476  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 477  * Non-Leaf:    | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
 478  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 479  *
 480  * To perform an interval query on an overlapped tree, perform the usual
 481  * depth-first search and use the low and high keys to decide if we can skip
 482  * that particular node.  If a leaf node is reached, return the records that
 483  * intersect the interval.  Note that an interval query may return numerous
 484  * entries.  For a non-overlapped tree, simply search for the record associated
 485  * with the lowest key and iterate forward until a non-matching record is
 486  * found.  Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
 487  * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
 488  * more detail.
 489  *
 490  * Why do we care about overlapping intervals?  Let's say you have a bunch of
 491  * reverse mapping records on a reflink filesystem:
 492  *
 493  * 1: +- file A startblock B offset C length D -----------+
 494  * 2:      +- file E startblock F offset G length H --------------+
 495  * 3:      +- file I startblock F offset J length K --+
 496  * 4:                                                        +- file L... --+
 497  *
 498  * Now say we want to map block (B+D) into file A at offset (C+D).  Ideally,
 499  * we'd simply increment the length of record 1.  But how do we find the record
 500  * that ends at (B+D-1) (i.e. record 1)?  A LE lookup of (B+D-1) would return
 501  * record 3 because the keys are ordered first by startblock.  An interval
 502  * query would return records 1 and 2 because they both overlap (B+D-1), and
 503  * from that we can pick out record 1 as the appropriate left neighbor.
 504  *
 505  * In the non-overlapped case you can do a LE lookup and decrement the cursor
 506  * because a record's interval must end before the next record.
 507  */
 508
 509 /*
 510  * Return size of the btree block header for this btree instance.
 511  */
 512 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
 513 {
 514         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 515                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 516                         return XFS_BTREE_LBLOCK_CRC_LEN;
 517                 return XFS_BTREE_LBLOCK_LEN;
 518         }
 519         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 520                 return XFS_BTREE_SBLOCK_CRC_LEN;
 521         return XFS_BTREE_SBLOCK_LEN;
 522 }
 523
 524 /*
 525  * Return size of btree block pointers for this btree instance.
 526  */
 527 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
 528 {
 529         return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
 530                 sizeof(__be64) : sizeof(__be32);
 531 }
 532
 533 /*
 534  * Calculate offset of the n-th record in a btree block.
 535  */
 536 STATIC size_t
 537 xfs_btree_rec_offset(
 538         struct xfs_btree_cur    *cur,
 539         int                     n)
 540 {
 541         return xfs_btree_block_len(cur) +
 542                 (n - 1) * cur->bc_ops->rec_len;
 543 }
 544
 545 /*
 546  * Calculate offset of the n-th key in a btree block.
 547  */
 548 STATIC size_t
 549 xfs_btree_key_offset(
 550         struct xfs_btree_cur    *cur,
 551         int                     n)
 552 {
 553         return xfs_btree_block_len(cur) +
 554                 (n - 1) * cur->bc_ops->key_len;
 555 }
 556
 557 /*
 558  * Calculate offset of the n-th high key in a btree block.
 559  */
 560 STATIC size_t
 561 xfs_btree_high_key_offset(
 562         struct xfs_btree_cur    *cur,
 563         int                     n)
 564 {
 565         return xfs_btree_block_len(cur) +
 566                 (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2);
 567 }
 568
 569 /*
 570  * Calculate offset of the n-th block pointer in a btree block.
 571  */
 572 STATIC size_t
 573 xfs_btree_ptr_offset(
 574         struct xfs_btree_cur    *cur,
 575         int                     n,
 576         int                     level)
 577 {
 578         return xfs_btree_block_len(cur) +
 579                 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
 580                 (n - 1) * xfs_btree_ptr_len(cur);
 581 }
 582
 583 /*
 584  * Return a pointer to the n-th record in the btree block.
 585  */
 586 union xfs_btree_rec *
 587 xfs_btree_rec_addr(
 588         struct xfs_btree_cur    *cur,
 589         int                     n,
 590         struct xfs_btree_block  *block)
 591 {
 592         return (union xfs_btree_rec *)
 593                 ((char *)block + xfs_btree_rec_offset(cur, n));
 594 }
 595
 596 /*
 597  * Return a pointer to the n-th key in the btree block.
 598  */
 599 union xfs_btree_key *
 600 xfs_btree_key_addr(
 601         struct xfs_btree_cur    *cur,
 602         int                     n,
 603         struct xfs_btree_block  *block)
 604 {
 605         return (union xfs_btree_key *)
 606                 ((char *)block + xfs_btree_key_offset(cur, n));
 607 }
 608
 609 /*
 610  * Return a pointer to the n-th high key in the btree block.
 611  */
 612 union xfs_btree_key *
 613 xfs_btree_high_key_addr(
 614         struct xfs_btree_cur    *cur,
 615         int                     n,
 616         struct xfs_btree_block  *block)
 617 {
 618         return (union xfs_btree_key *)
 619                 ((char *)block + xfs_btree_high_key_offset(cur, n));
 620 }
 621
 622 /*
 623  * Return a pointer to the n-th block pointer in the btree block.
 624  */
 625 union xfs_btree_ptr *
 626 xfs_btree_ptr_addr(
 627         struct xfs_btree_cur    *cur,
 628         int                     n,
 629         struct xfs_btree_block  *block)
 630 {
 631         int                     level = xfs_btree_get_level(block);
 632
 633         ASSERT(block->bb_level != 0);
 634
 635         return (union xfs_btree_ptr *)
 636                 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
 637 }
 638
 639 struct xfs_ifork *
 640 xfs_btree_ifork_ptr(
 641         struct xfs_btree_cur    *cur)
 642 {
 643         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
 644
 645         if (cur->bc_flags & XFS_BTREE_STAGING)
 646                 return cur->bc_ino.ifake->if_fork;
 647         return XFS_IFORK_PTR(cur->bc_ino.ip, cur->bc_ino.whichfork);
 648 }
 649
 650 /*
 651  * Get the root block which is stored in the inode.
 652  *
 653  * For now this btree implementation assumes the btree root is always
 654  * stored in the if_broot field of an inode fork.
 655  */
 656 STATIC struct xfs_btree_block *
 657 xfs_btree_get_iroot(
 658         struct xfs_btree_cur    *cur)
 659 {
 660         struct xfs_ifork        *ifp = xfs_btree_ifork_ptr(cur);
 661
 662         return (struct xfs_btree_block *)ifp->if_broot;
 663 }
 664
 665 /*
 666  * Retrieve the block pointer from the cursor at the given level.
 667  * This may be an inode btree root or from a buffer.
 668  */
 669 struct xfs_btree_block *                /* generic btree block pointer */
 670 xfs_btree_get_block(
 671         struct xfs_btree_cur    *cur,   /* btree cursor */
 672         int                     level,  /* level in btree */
 673         struct xfs_buf          **bpp)  /* buffer containing the block */
 674 {
 675         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 676             (level == cur->bc_nlevels - 1)) {
 677                 *bpp = NULL;
 678                 return xfs_btree_get_iroot(cur);
 679         }
 680
 681         *bpp = cur->bc_bufs[level];
 682         return XFS_BUF_TO_BLOCK(*bpp);
 683 }
 684
 685 /*
 686  * Change the cursor to point to the first record at the given level.
 687  * Other levels are unaffected.
 688  */
 689 STATIC int                              /* success=1, failure=0 */
 690 xfs_btree_firstrec(
 691         xfs_btree_cur_t         *cur,   /* btree cursor */
 692         int                     level)  /* level to change */
 693 {
 694         struct xfs_btree_block  *block; /* generic btree block pointer */
 695         struct xfs_buf          *bp;    /* buffer containing block */
 696
 697         /*
 698          * Get the block pointer for this level.
 699          */
 700         block = xfs_btree_get_block(cur, level, &bp);
 701         if (xfs_btree_check_block(cur, block, level, bp))
 702                 return 0;
 703         /*
 704          * It's empty, there is no such record.
 705          */
 706         if (!block->bb_numrecs)
 707                 return 0;
 708         /*
 709          * Set the ptr value to 1, that's the first record/key.
 710          */
 711         cur->bc_ptrs[level] = 1;
 712         return 1;
 713 }
 714
 715 /*
 716  * Change the cursor to point to the last record in the current block
 717  * at the given level.  Other levels are unaffected.
 718  */
 719 STATIC int                              /* success=1, failure=0 */
 720 xfs_btree_lastrec(
 721         xfs_btree_cur_t         *cur,   /* btree cursor */
 722         int                     level)  /* level to change */
 723 {
 724         struct xfs_btree_block  *block; /* generic btree block pointer */
 725         struct xfs_buf          *bp;    /* buffer containing block */
 726
 727         /*
 728          * Get the block pointer for this level.
 729          */
 730         block = xfs_btree_get_block(cur, level, &bp);
 731         if (xfs_btree_check_block(cur, block, level, bp))
 732                 return 0;
 733         /*
 734          * It's empty, there is no such record.
 735          */
 736         if (!block->bb_numrecs)
 737                 return 0;
 738         /*
 739          * Set the ptr value to numrecs, that's the last record/key.
 740          */
 741         cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
 742         return 1;
 743 }
 744
 745 /*
 746  * Compute first and last byte offsets for the fields given.
 747  * Interprets the offsets table, which contains struct field offsets.
 748  */
 749 void
 750 xfs_btree_offsets(
 751         int64_t         fields,         /* bitmask of fields */
 752         const short     *offsets,       /* table of field offsets */
 753         int             nbits,          /* number of bits to inspect */
 754         int             *first,         /* output: first byte offset */
 755         int             *last)          /* output: last byte offset */
 756 {
 757         int             i;              /* current bit number */
 758         int64_t         imask;          /* mask for current bit number */
 759
 760         ASSERT(fields != 0);
 761         /*
 762          * Find the lowest bit, so the first byte offset.
 763          */
 764         for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
 765                 if (imask & fields) {
 766                         *first = offsets[i];
 767                         break;
 768                 }
 769         }
 770         /*
 771          * Find the highest bit, so the last byte offset.
 772          */
 773         for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
 774                 if (imask & fields) {
 775                         *last = offsets[i + 1] - 1;
 776                         break;
 777                 }
 778         }
 779 }
 780
 781 /*
 782  * Get a buffer for the block, return it read in.
 783  * Long-form addressing.
 784  */
 785 int
 786 xfs_btree_read_bufl(
 787         struct xfs_mount        *mp,            /* file system mount point */
 788         struct xfs_trans        *tp,            /* transaction pointer */
 789         xfs_fsblock_t           fsbno,          /* file system block number */
 790         struct xfs_buf          **bpp,          /* buffer for fsbno */
 791         int                     refval,         /* ref count value for buffer */
 792         const struct xfs_buf_ops *ops)
 793 {
 794         struct xfs_buf          *bp;            /* return value */
 795         xfs_daddr_t             d;              /* real disk block address */
 796         int                     error;
 797
 798         if (!xfs_verify_fsbno(mp, fsbno))
 799                 return -EFSCORRUPTED;
 800         d = XFS_FSB_TO_DADDR(mp, fsbno);
 801         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
 802                                    mp->m_bsize, 0, &bp, ops);
 803         if (error)
 804                 return error;
 805         if (bp)
 806                 xfs_buf_set_ref(bp, refval);
 807         *bpp = bp;
 808         return 0;
 809 }
 810
 811 /*
 812  * Read-ahead the block, don't wait for it, don't return a buffer.
 813  * Long-form addressing.
 814  */
 815 /* ARGSUSED */
 816 void
 817 xfs_btree_reada_bufl(
 818         struct xfs_mount        *mp,            /* file system mount point */
 819         xfs_fsblock_t           fsbno,          /* file system block number */
 820         xfs_extlen_t            count,          /* count of filesystem blocks */
 821         const struct xfs_buf_ops *ops)
 822 {
 823         xfs_daddr_t             d;
 824
 825         ASSERT(fsbno != NULLFSBLOCK);
 826         d = XFS_FSB_TO_DADDR(mp, fsbno);
 827         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 828 }
 829
 830 /*
 831  * Read-ahead the block, don't wait for it, don't return a buffer.
 832  * Short-form addressing.
 833  */
 834 /* ARGSUSED */
 835 void
 836 xfs_btree_reada_bufs(
 837         struct xfs_mount        *mp,            /* file system mount point */
 838         xfs_agnumber_t          agno,           /* allocation group number */
 839         xfs_agblock_t           agbno,          /* allocation group block number */
 840         xfs_extlen_t            count,          /* count of filesystem blocks */
 841         const struct xfs_buf_ops *ops)
 842 {
 843         xfs_daddr_t             d;
 844
 845         ASSERT(agno != NULLAGNUMBER);
 846         ASSERT(agbno != NULLAGBLOCK);
 847         d = XFS_AGB_TO_DADDR(mp, agno, agbno);
 848         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 849 }
 850
 851 STATIC int
 852 xfs_btree_readahead_lblock(
 853         struct xfs_btree_cur    *cur,
 854         int                     lr,
 855         struct xfs_btree_block  *block)
 856 {
 857         int                     rval = 0;
 858         xfs_fsblock_t           left = be64_to_cpu(block->bb_u.l.bb_leftsib);
 859         xfs_fsblock_t           right = be64_to_cpu(block->bb_u.l.bb_rightsib);
 860
 861         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
 862                 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
 863                                      cur->bc_ops->buf_ops);
 864                 rval++;
 865         }
 866
 867         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
 868                 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
 869                                      cur->bc_ops->buf_ops);
 870                 rval++;
 871         }
 872
 873         return rval;
 874 }
 875
 876 STATIC int
 877 xfs_btree_readahead_sblock(
 878         struct xfs_btree_cur    *cur,
 879         int                     lr,
 880         struct xfs_btree_block *block)
 881 {
 882         int                     rval = 0;
 883         xfs_agblock_t           left = be32_to_cpu(block->bb_u.s.bb_leftsib);
 884         xfs_agblock_t           right = be32_to_cpu(block->bb_u.s.bb_rightsib);
 885
 886
 887         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
 888                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_ag.agno,
 889                                      left, 1, cur->bc_ops->buf_ops);
 890                 rval++;
 891         }
 892
 893         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
 894                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_ag.agno,
 895                                      right, 1, cur->bc_ops->buf_ops);
 896                 rval++;
 897         }
 898
 899         return rval;
 900 }
 901
 902 /*
 903  * Read-ahead btree blocks, at the given level.
 904  * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
 905  */
 906 STATIC int
 907 xfs_btree_readahead(
 908         struct xfs_btree_cur    *cur,           /* btree cursor */
 909         int                     lev,            /* level in btree */
 910         int                     lr)             /* left/right bits */
 911 {
 912         struct xfs_btree_block  *block;
 913
 914         /*
 915          * No readahead needed if we are at the root level and the
 916          * btree root is stored in the inode.
 917          */
 918         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 919             (lev == cur->bc_nlevels - 1))
 920                 return 0;
 921
 922         if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
 923                 return 0;
 924
 925         cur->bc_ra[lev] |= lr;
 926         block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
 927
 928         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 929                 return xfs_btree_readahead_lblock(cur, lr, block);
 930         return xfs_btree_readahead_sblock(cur, lr, block);
 931 }
 932
 933 STATIC int
 934 xfs_btree_ptr_to_daddr(
 935         struct xfs_btree_cur    *cur,
 936         union xfs_btree_ptr     *ptr,
 937         xfs_daddr_t             *daddr)
 938 {
 939         xfs_fsblock_t           fsbno;
 940         xfs_agblock_t           agbno;
 941         int                     error;
 942
 943         error = xfs_btree_check_ptr(cur, ptr, 0, 1);
 944         if (error)
 945                 return error;
 946
 947         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 948                 fsbno = be64_to_cpu(ptr->l);
 949                 *daddr = XFS_FSB_TO_DADDR(cur->bc_mp, fsbno);
 950         } else {
 951                 agbno = be32_to_cpu(ptr->s);
 952                 *daddr = XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_ag.agno,
 953                                 agbno);
 954         }
 955
 956         return 0;
 957 }
 958
 959 /*
 960  * Readahead @count btree blocks at the given @ptr location.
 961  *
 962  * We don't need to care about long or short form btrees here as we have a
 963  * method of converting the ptr directly to a daddr available to us.
 964  */
 965 STATIC void
 966 xfs_btree_readahead_ptr(
 967         struct xfs_btree_cur    *cur,
 968         union xfs_btree_ptr     *ptr,
 969         xfs_extlen_t            count)
 970 {
 971         xfs_daddr_t             daddr;
 972
 973         if (xfs_btree_ptr_to_daddr(cur, ptr, &daddr))
 974                 return;
 975         xfs_buf_readahead(cur->bc_mp->m_ddev_targp, daddr,
 976                           cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
 977 }
 978
 979 /*
 980  * Set the buffer for level "lev" in the cursor to bp, releasing
 981  * any previous buffer.
 982  */
 983 STATIC void
 984 xfs_btree_setbuf(
 985         xfs_btree_cur_t         *cur,   /* btree cursor */
 986         int                     lev,    /* level in btree */
 987         struct xfs_buf          *bp)    /* new buffer to set */
 988 {
 989         struct xfs_btree_block  *b;     /* btree block */
 990
 991         if (cur->bc_bufs[lev])
 992                 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
 993         cur->bc_bufs[lev] = bp;
 994         cur->bc_ra[lev] = 0;
 995
 996         b = XFS_BUF_TO_BLOCK(bp);
 997         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 998                 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
 999                         cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1000                 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1001                         cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1002         } else {
1003                 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1004                         cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1005                 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1006                         cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1007         }
1008 }
1009
1010 bool
1011 xfs_btree_ptr_is_null(
1012         struct xfs_btree_cur    *cur,
1013         union xfs_btree_ptr     *ptr)
1014 {
1015         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1016                 return ptr->l == cpu_to_be64(NULLFSBLOCK);
1017         else
1018                 return ptr->s == cpu_to_be32(NULLAGBLOCK);
1019 }
1020
1021 void
1022 xfs_btree_set_ptr_null(
1023         struct xfs_btree_cur    *cur,
1024         union xfs_btree_ptr     *ptr)
1025 {
1026         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1027                 ptr->l = cpu_to_be64(NULLFSBLOCK);
1028         else
1029                 ptr->s = cpu_to_be32(NULLAGBLOCK);
1030 }
1031
1032 /*
1033  * Get/set/init sibling pointers
1034  */
1035 void
1036 xfs_btree_get_sibling(
1037         struct xfs_btree_cur    *cur,
1038         struct xfs_btree_block  *block,
1039         union xfs_btree_ptr     *ptr,
1040         int                     lr)
1041 {
1042         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1043
1044         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1045                 if (lr == XFS_BB_RIGHTSIB)
1046                         ptr->l = block->bb_u.l.bb_rightsib;
1047                 else
1048                         ptr->l = block->bb_u.l.bb_leftsib;
1049         } else {
1050                 if (lr == XFS_BB_RIGHTSIB)
1051                         ptr->s = block->bb_u.s.bb_rightsib;
1052                 else
1053                         ptr->s = block->bb_u.s.bb_leftsib;
1054         }
1055 }
1056
1057 void
1058 xfs_btree_set_sibling(
1059         struct xfs_btree_cur    *cur,
1060         struct xfs_btree_block  *block,
1061         union xfs_btree_ptr     *ptr,
1062         int                     lr)
1063 {
1064         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1065
1066         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1067                 if (lr == XFS_BB_RIGHTSIB)
1068                         block->bb_u.l.bb_rightsib = ptr->l;
1069                 else
1070                         block->bb_u.l.bb_leftsib = ptr->l;
1071         } else {
1072                 if (lr == XFS_BB_RIGHTSIB)
1073                         block->bb_u.s.bb_rightsib = ptr->s;
1074                 else
1075                         block->bb_u.s.bb_leftsib = ptr->s;
1076         }
1077 }
1078
1079 void
1080 xfs_btree_init_block_int(
1081         struct xfs_mount        *mp,
1082         struct xfs_btree_block  *buf,
1083         xfs_daddr_t             blkno,
1084         xfs_btnum_t             btnum,
1085         __u16                   level,
1086         __u16                   numrecs,
1087         __u64                   owner,
1088         unsigned int            flags)
1089 {
1090         int                     crc = xfs_sb_version_hascrc(&mp->m_sb);
1091         __u32                   magic = xfs_btree_magic(crc, btnum);
1092
1093         buf->bb_magic = cpu_to_be32(magic);
1094         buf->bb_level = cpu_to_be16(level);
1095         buf->bb_numrecs = cpu_to_be16(numrecs);
1096
1097         if (flags & XFS_BTREE_LONG_PTRS) {
1098                 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1099                 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1100                 if (crc) {
1101                         buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1102                         buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1103                         uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1104                         buf->bb_u.l.bb_pad = 0;
1105                         buf->bb_u.l.bb_lsn = 0;
1106                 }
1107         } else {
1108                 /* owner is a 32 bit value on short blocks */
1109                 __u32 __owner = (__u32)owner;
1110
1111                 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1112                 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1113                 if (crc) {
1114                         buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1115                         buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1116                         uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1117                         buf->bb_u.s.bb_lsn = 0;
1118                 }
1119         }
1120 }
1121
1122 void
1123 xfs_btree_init_block(
1124         struct xfs_mount *mp,
1125         struct xfs_buf  *bp,
1126         xfs_btnum_t     btnum,
1127         __u16           level,
1128         __u16           numrecs,
1129         __u64           owner)
1130 {
1131         xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1132                                  btnum, level, numrecs, owner, 0);
1133 }
1134
1135 void
1136 xfs_btree_init_block_cur(
1137         struct xfs_btree_cur    *cur,
1138         struct xfs_buf          *bp,
1139         int                     level,
1140         int                     numrecs)
1141 {
1142         __u64                   owner;
1143
1144         /*
1145          * we can pull the owner from the cursor right now as the different
1146          * owners align directly with the pointer size of the btree. This may
1147          * change in future, but is safe for current users of the generic btree
1148          * code.
1149          */
1150         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1151                 owner = cur->bc_ino.ip->i_ino;
1152         else
1153                 owner = cur->bc_ag.agno;
1154
1155         xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1156                                  cur->bc_btnum, level, numrecs,
1157                                  owner, cur->bc_flags);
1158 }
1159
1160 /*
1161  * Return true if ptr is the last record in the btree and
1162  * we need to track updates to this record.  The decision
1163  * will be further refined in the update_lastrec method.
1164  */
1165 STATIC int
1166 xfs_btree_is_lastrec(
1167         struct xfs_btree_cur    *cur,
1168         struct xfs_btree_block  *block,
1169         int                     level)
1170 {
1171         union xfs_btree_ptr     ptr;
1172
1173         if (level > 0)
1174                 return 0;
1175         if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1176                 return 0;
1177
1178         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1179         if (!xfs_btree_ptr_is_null(cur, &ptr))
1180                 return 0;
1181         return 1;
1182 }
1183
1184 STATIC void
1185 xfs_btree_buf_to_ptr(
1186         struct xfs_btree_cur    *cur,
1187         struct xfs_buf          *bp,
1188         union xfs_btree_ptr     *ptr)
1189 {
1190         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1191                 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1192                                         XFS_BUF_ADDR(bp)));
1193         else {
1194                 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1195                                         XFS_BUF_ADDR(bp)));
1196         }
1197 }
1198
1199 STATIC void
1200 xfs_btree_set_refs(
1201         struct xfs_btree_cur    *cur,
1202         struct xfs_buf          *bp)
1203 {
1204         switch (cur->bc_btnum) {
1205         case XFS_BTNUM_BNO:
1206         case XFS_BTNUM_CNT:
1207                 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1208                 break;
1209         case XFS_BTNUM_INO:
1210         case XFS_BTNUM_FINO:
1211                 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1212                 break;
1213         case XFS_BTNUM_BMAP:
1214                 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1215                 break;
1216         case XFS_BTNUM_RMAP:
1217                 xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF);
1218                 break;
1219         case XFS_BTNUM_REFC:
1220                 xfs_buf_set_ref(bp, XFS_REFC_BTREE_REF);
1221                 break;
1222         default:
1223                 ASSERT(0);
1224         }
1225 }
1226
1227 int
1228 xfs_btree_get_buf_block(
1229         struct xfs_btree_cur    *cur,
1230         union xfs_btree_ptr     *ptr,
1231         struct xfs_btree_block  **block,
1232         struct xfs_buf          **bpp)
1233 {
1234         struct xfs_mount        *mp = cur->bc_mp;
1235         xfs_daddr_t             d;
1236         int                     error;
1237
1238         error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1239         if (error)
1240                 return error;
1241         error = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d, mp->m_bsize,
1242                         0, bpp);
1243         if (error)
1244                 return error;
1245
1246         (*bpp)->b_ops = cur->bc_ops->buf_ops;
1247         *block = XFS_BUF_TO_BLOCK(*bpp);
1248         return 0;
1249 }
1250
1251 /*
1252  * Read in the buffer at the given ptr and return the buffer and
1253  * the block pointer within the buffer.
1254  */
1255 STATIC int
1256 xfs_btree_read_buf_block(
1257         struct xfs_btree_cur    *cur,
1258         union xfs_btree_ptr     *ptr,
1259         int                     flags,
1260         struct xfs_btree_block  **block,
1261         struct xfs_buf          **bpp)
1262 {
1263         struct xfs_mount        *mp = cur->bc_mp;
1264         xfs_daddr_t             d;
1265         int                     error;
1266
1267         /* need to sort out how callers deal with failures first */
1268         ASSERT(!(flags & XBF_TRYLOCK));
1269
1270         error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1271         if (error)
1272                 return error;
1273         error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1274                                    mp->m_bsize, flags, bpp,
1275                                    cur->bc_ops->buf_ops);
1276         if (error)
1277                 return error;
1278
1279         xfs_btree_set_refs(cur, *bpp);
1280         *block = XFS_BUF_TO_BLOCK(*bpp);
1281         return 0;
1282 }
1283
1284 /*
1285  * Copy keys from one btree block to another.
1286  */
1287 void
1288 xfs_btree_copy_keys(
1289         struct xfs_btree_cur    *cur,
1290         union xfs_btree_key     *dst_key,
1291         union xfs_btree_key     *src_key,
1292         int                     numkeys)
1293 {
1294         ASSERT(numkeys >= 0);
1295         memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1296 }
1297
1298 /*
1299  * Copy records from one btree block to another.
1300  */
1301 STATIC void
1302 xfs_btree_copy_recs(
1303         struct xfs_btree_cur    *cur,
1304         union xfs_btree_rec     *dst_rec,
1305         union xfs_btree_rec     *src_rec,
1306         int                     numrecs)
1307 {
1308         ASSERT(numrecs >= 0);
1309         memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1310 }
1311
1312 /*
1313  * Copy block pointers from one btree block to another.
1314  */
1315 void
1316 xfs_btree_copy_ptrs(
1317         struct xfs_btree_cur    *cur,
1318         union xfs_btree_ptr     *dst_ptr,
1319         const union xfs_btree_ptr *src_ptr,
1320         int                     numptrs)
1321 {
1322         ASSERT(numptrs >= 0);
1323         memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1324 }
1325
1326 /*
1327  * Shift keys one index left/right inside a single btree block.
1328  */
1329 STATIC void
1330 xfs_btree_shift_keys(
1331         struct xfs_btree_cur    *cur,
1332         union xfs_btree_key     *key,
1333         int                     dir,
1334         int                     numkeys)
1335 {
1336         char                    *dst_key;
1337
1338         ASSERT(numkeys >= 0);
1339         ASSERT(dir == 1 || dir == -1);
1340
1341         dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1342         memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1343 }
1344
1345 /*
1346  * Shift records one index left/right inside a single btree block.
1347  */
1348 STATIC void
1349 xfs_btree_shift_recs(
1350         struct xfs_btree_cur    *cur,
1351         union xfs_btree_rec     *rec,
1352         int                     dir,
1353         int                     numrecs)
1354 {
1355         char                    *dst_rec;
1356
1357         ASSERT(numrecs >= 0);
1358         ASSERT(dir == 1 || dir == -1);
1359
1360         dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1361         memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1362 }
1363
1364 /*
1365  * Shift block pointers one index left/right inside a single btree block.
1366  */
1367 STATIC void
1368 xfs_btree_shift_ptrs(
1369         struct xfs_btree_cur    *cur,
1370         union xfs_btree_ptr     *ptr,
1371         int                     dir,
1372         int                     numptrs)
1373 {
1374         char                    *dst_ptr;
1375
1376         ASSERT(numptrs >= 0);
1377         ASSERT(dir == 1 || dir == -1);
1378
1379         dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1380         memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1381 }
1382
1383 /*
1384  * Log key values from the btree block.
1385  */
1386 STATIC void
1387 xfs_btree_log_keys(
1388         struct xfs_btree_cur    *cur,
1389         struct xfs_buf          *bp,
1390         int                     first,
1391         int                     last)
1392 {
1393
1394         if (bp) {
1395                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1396                 xfs_trans_log_buf(cur->bc_tp, bp,
1397                                   xfs_btree_key_offset(cur, first),
1398                                   xfs_btree_key_offset(cur, last + 1) - 1);
1399         } else {
1400                 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1401                                 xfs_ilog_fbroot(cur->bc_ino.whichfork));
1402         }
1403 }
1404
1405 /*
1406  * Log record values from the btree block.
1407  */
1408 void
1409 xfs_btree_log_recs(
1410         struct xfs_btree_cur    *cur,
1411         struct xfs_buf          *bp,
1412         int                     first,
1413         int                     last)
1414 {
1415
1416         xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1417         xfs_trans_log_buf(cur->bc_tp, bp,
1418                           xfs_btree_rec_offset(cur, first),
1419                           xfs_btree_rec_offset(cur, last + 1) - 1);
1420
1421 }
1422
1423 /*
1424  * Log block pointer fields from a btree block (nonleaf).
1425  */
1426 STATIC void
1427 xfs_btree_log_ptrs(
1428         struct xfs_btree_cur    *cur,   /* btree cursor */
1429         struct xfs_buf          *bp,    /* buffer containing btree block */
1430         int                     first,  /* index of first pointer to log */
1431         int                     last)   /* index of last pointer to log */
1432 {
1433
1434         if (bp) {
1435                 struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
1436                 int                     level = xfs_btree_get_level(block);
1437
1438                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1439                 xfs_trans_log_buf(cur->bc_tp, bp,
1440                                 xfs_btree_ptr_offset(cur, first, level),
1441                                 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1442         } else {
1443                 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1444                         xfs_ilog_fbroot(cur->bc_ino.whichfork));
1445         }
1446
1447 }
1448
1449 /*
1450  * Log fields from a btree block header.
1451  */
1452 void
1453 xfs_btree_log_block(
1454         struct xfs_btree_cur    *cur,   /* btree cursor */
1455         struct xfs_buf          *bp,    /* buffer containing btree block */
1456         int                     fields) /* mask of fields: XFS_BB_... */
1457 {
1458         int                     first;  /* first byte offset logged */
1459         int                     last;   /* last byte offset logged */
1460         static const short      soffsets[] = {  /* table of offsets (short) */
1461                 offsetof(struct xfs_btree_block, bb_magic),
1462                 offsetof(struct xfs_btree_block, bb_level),
1463                 offsetof(struct xfs_btree_block, bb_numrecs),
1464                 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1465                 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1466                 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1467                 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1468                 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1469                 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1470                 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1471                 XFS_BTREE_SBLOCK_CRC_LEN
1472         };
1473         static const short      loffsets[] = {  /* table of offsets (long) */
1474                 offsetof(struct xfs_btree_block, bb_magic),
1475                 offsetof(struct xfs_btree_block, bb_level),
1476                 offsetof(struct xfs_btree_block, bb_numrecs),
1477                 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1478                 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1479                 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1480                 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1481                 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1482                 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1483                 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1484                 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1485                 XFS_BTREE_LBLOCK_CRC_LEN
1486         };
1487
1488         if (bp) {
1489                 int nbits;
1490
1491                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1492                         /*
1493                          * We don't log the CRC when updating a btree
1494                          * block but instead recreate it during log
1495                          * recovery.  As the log buffers have checksums
1496                          * of their own this is safe and avoids logging a crc
1497                          * update in a lot of places.
1498                          */
1499                         if (fields == XFS_BB_ALL_BITS)
1500                                 fields = XFS_BB_ALL_BITS_CRC;
1501                         nbits = XFS_BB_NUM_BITS_CRC;
1502                 } else {
1503                         nbits = XFS_BB_NUM_BITS;
1504                 }
1505                 xfs_btree_offsets(fields,
1506                                   (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1507                                         loffsets : soffsets,
1508                                   nbits, &first, &last);
1509                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1510                 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1511         } else {
1512                 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1513                         xfs_ilog_fbroot(cur->bc_ino.whichfork));
1514         }
1515 }
1516
1517 /*
1518  * Increment cursor by one record at the level.
1519  * For nonzero levels the leaf-ward information is untouched.
1520  */
1521 int                                             /* error */
1522 xfs_btree_increment(
1523         struct xfs_btree_cur    *cur,
1524         int                     level,
1525         int                     *stat)          /* success/failure */
1526 {
1527         struct xfs_btree_block  *block;
1528         union xfs_btree_ptr     ptr;
1529         struct xfs_buf          *bp;
1530         int                     error;          /* error return value */
1531         int                     lev;
1532
1533         ASSERT(level < cur->bc_nlevels);
1534
1535         /* Read-ahead to the right at this level. */
1536         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1537
1538         /* Get a pointer to the btree block. */
1539         block = xfs_btree_get_block(cur, level, &bp);
1540
1541 #ifdef DEBUG
1542         error = xfs_btree_check_block(cur, block, level, bp);
1543         if (error)
1544                 goto error0;
1545 #endif
1546
1547         /* We're done if we remain in the block after the increment. */
1548         if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1549                 goto out1;
1550
1551         /* Fail if we just went off the right edge of the tree. */
1552         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1553         if (xfs_btree_ptr_is_null(cur, &ptr))
1554                 goto out0;
1555
1556         XFS_BTREE_STATS_INC(cur, increment);
1557
1558         /*
1559          * March up the tree incrementing pointers.
1560          * Stop when we don't go off the right edge of a block.
1561          */
1562         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1563                 block = xfs_btree_get_block(cur, lev, &bp);
1564
1565 #ifdef DEBUG
1566                 error = xfs_btree_check_block(cur, block, lev, bp);
1567                 if (error)
1568                         goto error0;
1569 #endif
1570
1571                 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1572                         break;
1573
1574                 /* Read-ahead the right block for the next loop. */
1575                 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1576         }
1577
1578         /*
1579          * If we went off the root then we are either seriously
1580          * confused or have the tree root in an inode.
1581          */
1582         if (lev == cur->bc_nlevels) {
1583                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1584                         goto out0;
1585                 ASSERT(0);
1586                 error = -EFSCORRUPTED;
1587                 goto error0;
1588         }
1589         ASSERT(lev < cur->bc_nlevels);
1590
1591         /*
1592          * Now walk back down the tree, fixing up the cursor's buffer
1593          * pointers and key numbers.
1594          */
1595         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1596                 union xfs_btree_ptr     *ptrp;
1597
1598                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1599                 --lev;
1600                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1601                 if (error)
1602                         goto error0;
1603
1604                 xfs_btree_setbuf(cur, lev, bp);
1605                 cur->bc_ptrs[lev] = 1;
1606         }
1607 out1:
1608         *stat = 1;
1609         return 0;
1610
1611 out0:
1612         *stat = 0;
1613         return 0;
1614
1615 error0:
1616         return error;
1617 }
1618
1619 /*
1620  * Decrement cursor by one record at the level.
1621  * For nonzero levels the leaf-ward information is untouched.
1622  */
1623 int                                             /* error */
1624 xfs_btree_decrement(
1625         struct xfs_btree_cur    *cur,
1626         int                     level,
1627         int                     *stat)          /* success/failure */
1628 {
1629         struct xfs_btree_block  *block;
1630         struct xfs_buf          *bp;
1631         int                     error;          /* error return value */
1632         int                     lev;
1633         union xfs_btree_ptr     ptr;
1634
1635         ASSERT(level < cur->bc_nlevels);
1636
1637         /* Read-ahead to the left at this level. */
1638         xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1639
1640         /* We're done if we remain in the block after the decrement. */
1641         if (--cur->bc_ptrs[level] > 0)
1642                 goto out1;
1643
1644         /* Get a pointer to the btree block. */
1645         block = xfs_btree_get_block(cur, level, &bp);
1646
1647 #ifdef DEBUG
1648         error = xfs_btree_check_block(cur, block, level, bp);
1649         if (error)
1650                 goto error0;
1651 #endif
1652
1653         /* Fail if we just went off the left edge of the tree. */
1654         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1655         if (xfs_btree_ptr_is_null(cur, &ptr))
1656                 goto out0;
1657
1658         XFS_BTREE_STATS_INC(cur, decrement);
1659
1660         /*
1661          * March up the tree decrementing pointers.
1662          * Stop when we don't go off the left edge of a block.
1663          */
1664         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1665                 if (--cur->bc_ptrs[lev] > 0)
1666                         break;
1667                 /* Read-ahead the left block for the next loop. */
1668                 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1669         }
1670
1671         /*
1672          * If we went off the root then we are seriously confused.
1673          * or the root of the tree is in an inode.
1674          */
1675         if (lev == cur->bc_nlevels) {
1676                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1677                         goto out0;
1678                 ASSERT(0);
1679                 error = -EFSCORRUPTED;
1680                 goto error0;
1681         }
1682         ASSERT(lev < cur->bc_nlevels);
1683
1684         /*
1685          * Now walk back down the tree, fixing up the cursor's buffer
1686          * pointers and key numbers.
1687          */
1688         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1689                 union xfs_btree_ptr     *ptrp;
1690
1691                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1692                 --lev;
1693                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1694                 if (error)
1695                         goto error0;
1696                 xfs_btree_setbuf(cur, lev, bp);
1697                 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1698         }
1699 out1:
1700         *stat = 1;
1701         return 0;
1702
1703 out0:
1704         *stat = 0;
1705         return 0;
1706
1707 error0:
1708         return error;
1709 }
1710
1711 int
1712 xfs_btree_lookup_get_block(
1713         struct xfs_btree_cur    *cur,   /* btree cursor */
1714         int                     level,  /* level in the btree */
1715         union xfs_btree_ptr     *pp,    /* ptr to btree block */
1716         struct xfs_btree_block  **blkp) /* return btree block */
1717 {
1718         struct xfs_buf          *bp;    /* buffer pointer for btree block */
1719         xfs_daddr_t             daddr;
1720         int                     error = 0;
1721
1722         /* special case the root block if in an inode */
1723         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1724             (level == cur->bc_nlevels - 1)) {
1725                 *blkp = xfs_btree_get_iroot(cur);
1726                 return 0;
1727         }
1728
1729         /*
1730          * If the old buffer at this level for the disk address we are
1731          * looking for re-use it.
1732          *
1733          * Otherwise throw it away and get a new one.
1734          */
1735         bp = cur->bc_bufs[level];
1736         error = xfs_btree_ptr_to_daddr(cur, pp, &daddr);
1737         if (error)
1738                 return error;
1739         if (bp && XFS_BUF_ADDR(bp) == daddr) {
1740                 *blkp = XFS_BUF_TO_BLOCK(bp);
1741                 return 0;
1742         }
1743
1744         error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1745         if (error)
1746                 return error;
1747
1748         /* Check the inode owner since the verifiers don't. */
1749         if (xfs_sb_version_hascrc(&cur->bc_mp->m_sb) &&
1750             !(cur->bc_ino.flags & XFS_BTCUR_BMBT_INVALID_OWNER) &&
1751             (cur->bc_flags & XFS_BTREE_LONG_PTRS) &&
1752             be64_to_cpu((*blkp)->bb_u.l.bb_owner) !=
1753                         cur->bc_ino.ip->i_ino)
1754                 goto out_bad;
1755
1756         /* Did we get the level we were looking for? */
1757         if (be16_to_cpu((*blkp)->bb_level) != level)
1758                 goto out_bad;
1759
1760         /* Check that internal nodes have at least one record. */
1761         if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0)
1762                 goto out_bad;
1763
1764         xfs_btree_setbuf(cur, level, bp);
1765         return 0;
1766
1767 out_bad:
1768         *blkp = NULL;
1769         xfs_buf_mark_corrupt(bp);
1770         xfs_trans_brelse(cur->bc_tp, bp);
1771         return -EFSCORRUPTED;
1772 }
1773
1774 /*
1775  * Get current search key.  For level 0 we don't actually have a key
1776  * structure so we make one up from the record.  For all other levels
1777  * we just return the right key.
1778  */
1779 STATIC union xfs_btree_key *
1780 xfs_lookup_get_search_key(
1781         struct xfs_btree_cur    *cur,
1782         int                     level,
1783         int                     keyno,
1784         struct xfs_btree_block  *block,
1785         union xfs_btree_key     *kp)
1786 {
1787         if (level == 0) {
1788                 cur->bc_ops->init_key_from_rec(kp,
1789                                 xfs_btree_rec_addr(cur, keyno, block));
1790                 return kp;
1791         }
1792
1793         return xfs_btree_key_addr(cur, keyno, block);
1794 }
1795
1796 /*
1797  * Lookup the record.  The cursor is made to point to it, based on dir.
1798  * stat is set to 0 if can't find any such record, 1 for success.
1799  */
1800 int                                     /* error */
1801 xfs_btree_lookup(
1802         struct xfs_btree_cur    *cur,   /* btree cursor */
1803         xfs_lookup_t            dir,    /* <=, ==, or >= */
1804         int                     *stat)  /* success/failure */
1805 {
1806         struct xfs_btree_block  *block; /* current btree block */
1807         int64_t                 diff;   /* difference for the current key */
1808         int                     error;  /* error return value */
1809         int                     keyno;  /* current key number */
1810         int                     level;  /* level in the btree */
1811         union xfs_btree_ptr     *pp;    /* ptr to btree block */
1812         union xfs_btree_ptr     ptr;    /* ptr to btree block */
1813
1814         XFS_BTREE_STATS_INC(cur, lookup);
1815
1816         /* No such thing as a zero-level tree. */
1817         if (XFS_IS_CORRUPT(cur->bc_mp, cur->bc_nlevels == 0))
1818                 return -EFSCORRUPTED;
1819
1820         block = NULL;
1821         keyno = 0;
1822
1823         /* initialise start pointer from cursor */
1824         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1825         pp = &ptr;
1826
1827         /*
1828          * Iterate over each level in the btree, starting at the root.
1829          * For each level above the leaves, find the key we need, based
1830          * on the lookup record, then follow the corresponding block
1831          * pointer down to the next level.
1832          */
1833         for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1834                 /* Get the block we need to do the lookup on. */
1835                 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1836                 if (error)
1837                         goto error0;
1838
1839                 if (diff == 0) {
1840                         /*
1841                          * If we already had a key match at a higher level, we
1842                          * know we need to use the first entry in this block.
1843                          */
1844                         keyno = 1;
1845                 } else {
1846                         /* Otherwise search this block. Do a binary search. */
1847
1848                         int     high;   /* high entry number */
1849                         int     low;    /* low entry number */
1850
1851                         /* Set low and high entry numbers, 1-based. */
1852                         low = 1;
1853                         high = xfs_btree_get_numrecs(block);
1854                         if (!high) {
1855                                 /* Block is empty, must be an empty leaf. */
1856                                 if (level != 0 || cur->bc_nlevels != 1) {
1857                                         XFS_CORRUPTION_ERROR(__func__,
1858                                                         XFS_ERRLEVEL_LOW,
1859                                                         cur->bc_mp, block,
1860                                                         sizeof(*block));
1861                                         return -EFSCORRUPTED;
1862                                 }
1863
1864                                 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1865                                 *stat = 0;
1866                                 return 0;
1867                         }
1868
1869                         /* Binary search the block. */
1870                         while (low <= high) {
1871                                 union xfs_btree_key     key;
1872                                 union xfs_btree_key     *kp;
1873
1874                                 XFS_BTREE_STATS_INC(cur, compare);
1875
1876                                 /* keyno is average of low and high. */
1877                                 keyno = (low + high) >> 1;
1878
1879                                 /* Get current search key */
1880                                 kp = xfs_lookup_get_search_key(cur, level,
1881                                                 keyno, block, &key);
1882
1883                                 /*
1884                                  * Compute difference to get next direction:
1885                                  *  - less than, move right
1886                                  *  - greater than, move left
1887                                  *  - equal, we're done
1888                                  */
1889                                 diff = cur->bc_ops->key_diff(cur, kp);
1890                                 if (diff < 0)
1891                                         low = keyno + 1;
1892                                 else if (diff > 0)
1893                                         high = keyno - 1;
1894                                 else
1895                                         break;
1896                         }
1897                 }
1898
1899                 /*
1900                  * If there are more levels, set up for the next level
1901                  * by getting the block number and filling in the cursor.
1902                  */
1903                 if (level > 0) {
1904                         /*
1905                          * If we moved left, need the previous key number,
1906                          * unless there isn't one.
1907                          */
1908                         if (diff > 0 && --keyno < 1)
1909                                 keyno = 1;
1910                         pp = xfs_btree_ptr_addr(cur, keyno, block);
1911
1912                         error = xfs_btree_debug_check_ptr(cur, pp, 0, level);
1913                         if (error)
1914                                 goto error0;
1915
1916                         cur->bc_ptrs[level] = keyno;
1917                 }
1918         }
1919
1920         /* Done with the search. See if we need to adjust the results. */
1921         if (dir != XFS_LOOKUP_LE && diff < 0) {
1922                 keyno++;
1923                 /*
1924                  * If ge search and we went off the end of the block, but it's
1925                  * not the last block, we're in the wrong block.
1926                  */
1927                 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1928                 if (dir == XFS_LOOKUP_GE &&
1929                     keyno > xfs_btree_get_numrecs(block) &&
1930                     !xfs_btree_ptr_is_null(cur, &ptr)) {
1931                         int     i;
1932
1933                         cur->bc_ptrs[0] = keyno;
1934                         error = xfs_btree_increment(cur, 0, &i);
1935                         if (error)
1936                                 goto error0;
1937                         if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1938                                 return -EFSCORRUPTED;
1939                         *stat = 1;
1940                         return 0;
1941                 }
1942         } else if (dir == XFS_LOOKUP_LE && diff > 0)
1943                 keyno--;
1944         cur->bc_ptrs[0] = keyno;
1945
1946         /* Return if we succeeded or not. */
1947         if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1948                 *stat = 0;
1949         else if (dir != XFS_LOOKUP_EQ || diff == 0)
1950                 *stat = 1;
1951         else
1952                 *stat = 0;
1953         return 0;
1954
1955 error0:
1956         return error;
1957 }
1958
1959 /* Find the high key storage area from a regular key. */
1960 union xfs_btree_key *
1961 xfs_btree_high_key_from_key(
1962         struct xfs_btree_cur    *cur,
1963         union xfs_btree_key     *key)
1964 {
1965         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
1966         return (union xfs_btree_key *)((char *)key +
1967                         (cur->bc_ops->key_len / 2));
1968 }
1969
1970 /* Determine the low (and high if overlapped) keys of a leaf block */
1971 STATIC void
1972 xfs_btree_get_leaf_keys(
1973         struct xfs_btree_cur    *cur,
1974         struct xfs_btree_block  *block,
1975         union xfs_btree_key     *key)
1976 {
1977         union xfs_btree_key     max_hkey;
1978         union xfs_btree_key     hkey;
1979         union xfs_btree_rec     *rec;
1980         union xfs_btree_key     *high;
1981         int                     n;
1982
1983         rec = xfs_btree_rec_addr(cur, 1, block);
1984         cur->bc_ops->init_key_from_rec(key, rec);
1985
1986         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
1987
1988                 cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
1989                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
1990                         rec = xfs_btree_rec_addr(cur, n, block);
1991                         cur->bc_ops->init_high_key_from_rec(&hkey, rec);
1992                         if (cur->bc_ops->diff_two_keys(cur, &hkey, &max_hkey)
1993                                         > 0)
1994                                 max_hkey = hkey;
1995                 }
1996
1997                 high = xfs_btree_high_key_from_key(cur, key);
1998                 memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
1999         }
2000 }
2001
2002 /* Determine the low (and high if overlapped) keys of a node block */
2003 STATIC void
2004 xfs_btree_get_node_keys(
2005         struct xfs_btree_cur    *cur,
2006         struct xfs_btree_block  *block,
2007         union xfs_btree_key     *key)
2008 {
2009         union xfs_btree_key     *hkey;
2010         union xfs_btree_key     *max_hkey;
2011         union xfs_btree_key     *high;
2012         int                     n;
2013
2014         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2015                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2016                                 cur->bc_ops->key_len / 2);
2017
2018                 max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2019                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2020                         hkey = xfs_btree_high_key_addr(cur, n, block);
2021                         if (cur->bc_ops->diff_two_keys(cur, hkey, max_hkey) > 0)
2022                                 max_hkey = hkey;
2023                 }
2024
2025                 high = xfs_btree_high_key_from_key(cur, key);
2026                 memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2027         } else {
2028                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2029                                 cur->bc_ops->key_len);
2030         }
2031 }
2032
2033 /* Derive the keys for any btree block. */
2034 void
2035 xfs_btree_get_keys(
2036         struct xfs_btree_cur    *cur,
2037         struct xfs_btree_block  *block,
2038         union xfs_btree_key     *key)
2039 {
2040         if (be16_to_cpu(block->bb_level) == 0)
2041                 xfs_btree_get_leaf_keys(cur, block, key);
2042         else
2043                 xfs_btree_get_node_keys(cur, block, key);
2044 }
2045
2046 /*
2047  * Decide if we need to update the parent keys of a btree block.  For
2048  * a standard btree this is only necessary if we're updating the first
2049  * record/key.  For an overlapping btree, we must always update the
2050  * keys because the highest key can be in any of the records or keys
2051  * in the block.
2052  */
2053 static inline bool
2054 xfs_btree_needs_key_update(
2055         struct xfs_btree_cur    *cur,
2056         int                     ptr)
2057 {
2058         return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1;
2059 }
2060
2061 /*
2062  * Update the low and high parent keys of the given level, progressing
2063  * towards the root.  If force_all is false, stop if the keys for a given
2064  * level do not need updating.
2065  */
2066 STATIC int
2067 __xfs_btree_updkeys(
2068         struct xfs_btree_cur    *cur,
2069         int                     level,
2070         struct xfs_btree_block  *block,
2071         struct xfs_buf          *bp0,
2072         bool                    force_all)
2073 {
2074         union xfs_btree_key     key;    /* keys from current level */
2075         union xfs_btree_key     *lkey;  /* keys from the next level up */
2076         union xfs_btree_key     *hkey;
2077         union xfs_btree_key     *nlkey; /* keys from the next level up */
2078         union xfs_btree_key     *nhkey;
2079         struct xfs_buf          *bp;
2080         int                     ptr;
2081
2082         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2083
2084         /* Exit if there aren't any parent levels to update. */
2085         if (level + 1 >= cur->bc_nlevels)
2086                 return 0;
2087
2088         trace_xfs_btree_updkeys(cur, level, bp0);
2089
2090         lkey = &key;
2091         hkey = xfs_btree_high_key_from_key(cur, lkey);
2092         xfs_btree_get_keys(cur, block, lkey);
2093         for (level++; level < cur->bc_nlevels; level++) {
2094 #ifdef DEBUG
2095                 int             error;
2096 #endif
2097                 block = xfs_btree_get_block(cur, level, &bp);
2098                 trace_xfs_btree_updkeys(cur, level, bp);
2099 #ifdef DEBUG
2100                 error = xfs_btree_check_block(cur, block, level, bp);
2101                 if (error)
2102                         return error;
2103 #endif
2104                 ptr = cur->bc_ptrs[level];
2105                 nlkey = xfs_btree_key_addr(cur, ptr, block);
2106                 nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2107                 if (!force_all &&
2108                     !(cur->bc_ops->diff_two_keys(cur, nlkey, lkey) != 0 ||
2109                       cur->bc_ops->diff_two_keys(cur, nhkey, hkey) != 0))
2110                         break;
2111                 xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2112                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2113                 if (level + 1 >= cur->bc_nlevels)
2114                         break;
2115                 xfs_btree_get_node_keys(cur, block, lkey);
2116         }
2117
2118         return 0;
2119 }
2120
2121 /* Update all the keys from some level in cursor back to the root. */
2122 STATIC int
2123 xfs_btree_updkeys_force(
2124         struct xfs_btree_cur    *cur,
2125         int                     level)
2126 {
2127         struct xfs_buf          *bp;
2128         struct xfs_btree_block  *block;
2129
2130         block = xfs_btree_get_block(cur, level, &bp);
2131         return __xfs_btree_updkeys(cur, level, block, bp, true);
2132 }
2133
2134 /*
2135  * Update the parent keys of the given level, progressing towards the root.
2136  */
2137 STATIC int
2138 xfs_btree_update_keys(
2139         struct xfs_btree_cur    *cur,
2140         int                     level)
2141 {
2142         struct xfs_btree_block  *block;
2143         struct xfs_buf          *bp;
2144         union xfs_btree_key     *kp;
2145         union xfs_btree_key     key;
2146         int                     ptr;
2147
2148         ASSERT(level >= 0);
2149
2150         block = xfs_btree_get_block(cur, level, &bp);
2151         if (cur->bc_flags & XFS_BTREE_OVERLAPPING)
2152                 return __xfs_btree_updkeys(cur, level, block, bp, false);
2153
2154         /*
2155          * Go up the tree from this level toward the root.
2156          * At each level, update the key value to the value input.
2157          * Stop when we reach a level where the cursor isn't pointing
2158          * at the first entry in the block.
2159          */
2160         xfs_btree_get_keys(cur, block, &key);
2161         for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2162 #ifdef DEBUG
2163                 int             error;
2164 #endif
2165                 block = xfs_btree_get_block(cur, level, &bp);
2166 #ifdef DEBUG
2167                 error = xfs_btree_check_block(cur, block, level, bp);
2168                 if (error)
2169                         return error;
2170 #endif
2171                 ptr = cur->bc_ptrs[level];
2172                 kp = xfs_btree_key_addr(cur, ptr, block);
2173                 xfs_btree_copy_keys(cur, kp, &key, 1);
2174                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2175         }
2176
2177         return 0;
2178 }
2179
2180 /*
2181  * Update the record referred to by cur to the value in the
2182  * given record. This either works (return 0) or gets an
2183  * EFSCORRUPTED error.
2184  */
2185 int
2186 xfs_btree_update(
2187         struct xfs_btree_cur    *cur,
2188         union xfs_btree_rec     *rec)
2189 {
2190         struct xfs_btree_block  *block;
2191         struct xfs_buf          *bp;
2192         int                     error;
2193         int                     ptr;
2194         union xfs_btree_rec     *rp;
2195
2196         /* Pick up the current block. */
2197         block = xfs_btree_get_block(cur, 0, &bp);
2198
2199 #ifdef DEBUG
2200         error = xfs_btree_check_block(cur, block, 0, bp);
2201         if (error)
2202                 goto error0;
2203 #endif
2204         /* Get the address of the rec to be updated. */
2205         ptr = cur->bc_ptrs[0];
2206         rp = xfs_btree_rec_addr(cur, ptr, block);
2207
2208         /* Fill in the new contents and log them. */
2209         xfs_btree_copy_recs(cur, rp, rec, 1);
2210         xfs_btree_log_recs(cur, bp, ptr, ptr);
2211
2212         /*
2213          * If we are tracking the last record in the tree and
2214          * we are at the far right edge of the tree, update it.
2215          */
2216         if (xfs_btree_is_lastrec(cur, block, 0)) {
2217                 cur->bc_ops->update_lastrec(cur, block, rec,
2218                                             ptr, LASTREC_UPDATE);
2219         }
2220
2221         /* Pass new key value up to our parent. */
2222         if (xfs_btree_needs_key_update(cur, ptr)) {
2223                 error = xfs_btree_update_keys(cur, 0);
2224                 if (error)
2225                         goto error0;
2226         }
2227
2228         return 0;
2229
2230 error0:
2231         return error;
2232 }
2233
2234 /*
2235  * Move 1 record left from cur/level if possible.
2236  * Update cur to reflect the new path.
2237  */
2238 STATIC int                                      /* error */
2239 xfs_btree_lshift(
2240         struct xfs_btree_cur    *cur,
2241         int                     level,
2242         int                     *stat)          /* success/failure */
2243 {
2244         struct xfs_buf          *lbp;           /* left buffer pointer */
2245         struct xfs_btree_block  *left;          /* left btree block */
2246         int                     lrecs;          /* left record count */
2247         struct xfs_buf          *rbp;           /* right buffer pointer */
2248         struct xfs_btree_block  *right;         /* right btree block */
2249         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2250         int                     rrecs;          /* right record count */
2251         union xfs_btree_ptr     lptr;           /* left btree pointer */
2252         union xfs_btree_key     *rkp = NULL;    /* right btree key */
2253         union xfs_btree_ptr     *rpp = NULL;    /* right address pointer */
2254         union xfs_btree_rec     *rrp = NULL;    /* right record pointer */
2255         int                     error;          /* error return value */
2256         int                     i;
2257
2258         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2259             level == cur->bc_nlevels - 1)
2260                 goto out0;
2261
2262         /* Set up variables for this block as "right". */
2263         right = xfs_btree_get_block(cur, level, &rbp);
2264
2265 #ifdef DEBUG
2266         error = xfs_btree_check_block(cur, right, level, rbp);
2267         if (error)
2268                 goto error0;
2269 #endif
2270
2271         /* If we've got no left sibling then we can't shift an entry left. */
2272         xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2273         if (xfs_btree_ptr_is_null(cur, &lptr))
2274                 goto out0;
2275
2276         /*
2277          * If the cursor entry is the one that would be moved, don't
2278          * do it... it's too complicated.
2279          */
2280         if (cur->bc_ptrs[level] <= 1)
2281                 goto out0;
2282
2283         /* Set up the left neighbor as "left". */
2284         error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2285         if (error)
2286                 goto error0;
2287
2288         /* If it's full, it can't take another entry. */
2289         lrecs = xfs_btree_get_numrecs(left);
2290         if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2291                 goto out0;
2292
2293         rrecs = xfs_btree_get_numrecs(right);
2294
2295         /*
2296          * We add one entry to the left side and remove one for the right side.
2297          * Account for it here, the changes will be updated on disk and logged
2298          * later.
2299          */
2300         lrecs++;
2301         rrecs--;
2302
2303         XFS_BTREE_STATS_INC(cur, lshift);
2304         XFS_BTREE_STATS_ADD(cur, moves, 1);
2305
2306         /*
2307          * If non-leaf, copy a key and a ptr to the left block.
2308          * Log the changes to the left block.
2309          */
2310         if (level > 0) {
2311                 /* It's a non-leaf.  Move keys and pointers. */
2312                 union xfs_btree_key     *lkp;   /* left btree key */
2313                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2314
2315                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2316                 rkp = xfs_btree_key_addr(cur, 1, right);
2317
2318                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2319                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2320
2321                 error = xfs_btree_debug_check_ptr(cur, rpp, 0, level);
2322                 if (error)
2323                         goto error0;
2324
2325                 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2326                 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2327
2328                 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2329                 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2330
2331                 ASSERT(cur->bc_ops->keys_inorder(cur,
2332                         xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2333         } else {
2334                 /* It's a leaf.  Move records.  */
2335                 union xfs_btree_rec     *lrp;   /* left record pointer */
2336
2337                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2338                 rrp = xfs_btree_rec_addr(cur, 1, right);
2339
2340                 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2341                 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2342
2343                 ASSERT(cur->bc_ops->recs_inorder(cur,
2344                         xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2345         }
2346
2347         xfs_btree_set_numrecs(left, lrecs);
2348         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2349
2350         xfs_btree_set_numrecs(right, rrecs);
2351         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2352
2353         /*
2354          * Slide the contents of right down one entry.
2355          */
2356         XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2357         if (level > 0) {
2358                 /* It's a nonleaf. operate on keys and ptrs */
2359                 for (i = 0; i < rrecs; i++) {
2360                         error = xfs_btree_debug_check_ptr(cur, rpp, i + 1, level);
2361                         if (error)
2362                                 goto error0;
2363                 }
2364
2365                 xfs_btree_shift_keys(cur,
2366                                 xfs_btree_key_addr(cur, 2, right),
2367                                 -1, rrecs);
2368                 xfs_btree_shift_ptrs(cur,
2369                                 xfs_btree_ptr_addr(cur, 2, right),
2370                                 -1, rrecs);
2371
2372                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2373                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2374         } else {
2375                 /* It's a leaf. operate on records */
2376                 xfs_btree_shift_recs(cur,
2377                         xfs_btree_rec_addr(cur, 2, right),
2378                         -1, rrecs);
2379                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2380         }
2381
2382         /*
2383          * Using a temporary cursor, update the parent key values of the
2384          * block on the left.
2385          */
2386         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2387                 error = xfs_btree_dup_cursor(cur, &tcur);
2388                 if (error)
2389                         goto error0;
2390                 i = xfs_btree_firstrec(tcur, level);
2391                 if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2392                         error = -EFSCORRUPTED;
2393                         goto error0;
2394                 }
2395
2396                 error = xfs_btree_decrement(tcur, level, &i);
2397                 if (error)
2398                         goto error1;
2399
2400                 /* Update the parent high keys of the left block, if needed. */
2401                 error = xfs_btree_update_keys(tcur, level);
2402                 if (error)
2403                         goto error1;
2404
2405                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2406         }
2407
2408         /* Update the parent keys of the right block. */
2409         error = xfs_btree_update_keys(cur, level);
2410         if (error)
2411                 goto error0;
2412
2413         /* Slide the cursor value left one. */
2414         cur->bc_ptrs[level]--;
2415
2416         *stat = 1;
2417         return 0;
2418
2419 out0:
2420         *stat = 0;
2421         return 0;
2422
2423 error0:
2424         return error;
2425
2426 error1:
2427         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2428         return error;
2429 }
2430
2431 /*
2432  * Move 1 record right from cur/level if possible.
2433  * Update cur to reflect the new path.
2434  */
2435 STATIC int                                      /* error */
2436 xfs_btree_rshift(
2437         struct xfs_btree_cur    *cur,
2438         int                     level,
2439         int                     *stat)          /* success/failure */
2440 {
2441         struct xfs_buf          *lbp;           /* left buffer pointer */
2442         struct xfs_btree_block  *left;          /* left btree block */
2443         struct xfs_buf          *rbp;           /* right buffer pointer */
2444         struct xfs_btree_block  *right;         /* right btree block */
2445         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2446         union xfs_btree_ptr     rptr;           /* right block pointer */
2447         union xfs_btree_key     *rkp;           /* right btree key */
2448         int                     rrecs;          /* right record count */
2449         int                     lrecs;          /* left record count */
2450         int                     error;          /* error return value */
2451         int                     i;              /* loop counter */
2452
2453         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2454             (level == cur->bc_nlevels - 1))
2455                 goto out0;
2456
2457         /* Set up variables for this block as "left". */
2458         left = xfs_btree_get_block(cur, level, &lbp);
2459
2460 #ifdef DEBUG
2461         error = xfs_btree_check_block(cur, left, level, lbp);
2462         if (error)
2463                 goto error0;
2464 #endif
2465
2466         /* If we've got no right sibling then we can't shift an entry right. */
2467         xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2468         if (xfs_btree_ptr_is_null(cur, &rptr))
2469                 goto out0;
2470
2471         /*
2472          * If the cursor entry is the one that would be moved, don't
2473          * do it... it's too complicated.
2474          */
2475         lrecs = xfs_btree_get_numrecs(left);
2476         if (cur->bc_ptrs[level] >= lrecs)
2477                 goto out0;
2478
2479         /* Set up the right neighbor as "right". */
2480         error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2481         if (error)
2482                 goto error0;
2483
2484         /* If it's full, it can't take another entry. */
2485         rrecs = xfs_btree_get_numrecs(right);
2486         if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2487                 goto out0;
2488
2489         XFS_BTREE_STATS_INC(cur, rshift);
2490         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2491
2492         /*
2493          * Make a hole at the start of the right neighbor block, then
2494          * copy the last left block entry to the hole.
2495          */
2496         if (level > 0) {
2497                 /* It's a nonleaf. make a hole in the keys and ptrs */
2498                 union xfs_btree_key     *lkp;
2499                 union xfs_btree_ptr     *lpp;
2500                 union xfs_btree_ptr     *rpp;
2501
2502                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2503                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2504                 rkp = xfs_btree_key_addr(cur, 1, right);
2505                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2506
2507                 for (i = rrecs - 1; i >= 0; i--) {
2508                         error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
2509                         if (error)
2510                                 goto error0;
2511                 }
2512
2513                 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2514                 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2515
2516                 error = xfs_btree_debug_check_ptr(cur, lpp, 0, level);
2517                 if (error)
2518                         goto error0;
2519
2520                 /* Now put the new data in, and log it. */
2521                 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2522                 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2523
2524                 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2525                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2526
2527                 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2528                         xfs_btree_key_addr(cur, 2, right)));
2529         } else {
2530                 /* It's a leaf. make a hole in the records */
2531                 union xfs_btree_rec     *lrp;
2532                 union xfs_btree_rec     *rrp;
2533
2534                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2535                 rrp = xfs_btree_rec_addr(cur, 1, right);
2536
2537                 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2538
2539                 /* Now put the new data in, and log it. */
2540                 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2541                 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2542         }
2543
2544         /*
2545          * Decrement and log left's numrecs, bump and log right's numrecs.
2546          */
2547         xfs_btree_set_numrecs(left, --lrecs);
2548         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2549
2550         xfs_btree_set_numrecs(right, ++rrecs);
2551         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2552
2553         /*
2554          * Using a temporary cursor, update the parent key values of the
2555          * block on the right.
2556          */
2557         error = xfs_btree_dup_cursor(cur, &tcur);
2558         if (error)
2559                 goto error0;
2560         i = xfs_btree_lastrec(tcur, level);
2561         if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2562                 error = -EFSCORRUPTED;
2563                 goto error0;
2564         }
2565
2566         error = xfs_btree_increment(tcur, level, &i);
2567         if (error)
2568                 goto error1;
2569
2570         /* Update the parent high keys of the left block, if needed. */
2571         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2572                 error = xfs_btree_update_keys(cur, level);
2573                 if (error)
2574                         goto error1;
2575         }
2576
2577         /* Update the parent keys of the right block. */
2578         error = xfs_btree_update_keys(tcur, level);
2579         if (error)
2580                 goto error1;
2581
2582         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2583
2584         *stat = 1;
2585         return 0;
2586
2587 out0:
2588         *stat = 0;
2589         return 0;
2590
2591 error0:
2592         return error;
2593
2594 error1:
2595         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2596         return error;
2597 }
2598
2599 /*
2600  * Split cur/level block in half.
2601  * Return new block number and the key to its first
2602  * record (to be inserted into parent).
2603  */
2604 STATIC int                                      /* error */
2605 __xfs_btree_split(
2606         struct xfs_btree_cur    *cur,
2607         int                     level,
2608         union xfs_btree_ptr     *ptrp,
2609         union xfs_btree_key     *key,
2610         struct xfs_btree_cur    **curp,
2611         int                     *stat)          /* success/failure */
2612 {
2613         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
2614         struct xfs_buf          *lbp;           /* left buffer pointer */
2615         struct xfs_btree_block  *left;          /* left btree block */
2616         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
2617         struct xfs_buf          *rbp;           /* right buffer pointer */
2618         struct xfs_btree_block  *right;         /* right btree block */
2619         union xfs_btree_ptr     rrptr;          /* right-right sibling ptr */
2620         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
2621         struct xfs_btree_block  *rrblock;       /* right-right btree block */
2622         int                     lrecs;
2623         int                     rrecs;
2624         int                     src_index;
2625         int                     error;          /* error return value */
2626         int                     i;
2627
2628         XFS_BTREE_STATS_INC(cur, split);
2629
2630         /* Set up left block (current one). */
2631         left = xfs_btree_get_block(cur, level, &lbp);
2632
2633 #ifdef DEBUG
2634         error = xfs_btree_check_block(cur, left, level, lbp);
2635         if (error)
2636                 goto error0;
2637 #endif
2638
2639         xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2640
2641         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2642         error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2643         if (error)
2644                 goto error0;
2645         if (*stat == 0)
2646                 goto out0;
2647         XFS_BTREE_STATS_INC(cur, alloc);
2648
2649         /* Set up the new block as "right". */
2650         error = xfs_btree_get_buf_block(cur, &rptr, &right, &rbp);
2651         if (error)
2652                 goto error0;
2653
2654         /* Fill in the btree header for the new right block. */
2655         xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2656
2657         /*
2658          * Split the entries between the old and the new block evenly.
2659          * Make sure that if there's an odd number of entries now, that
2660          * each new block will have the same number of entries.
2661          */
2662         lrecs = xfs_btree_get_numrecs(left);
2663         rrecs = lrecs / 2;
2664         if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2665                 rrecs++;
2666         src_index = (lrecs - rrecs + 1);
2667
2668         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2669
2670         /* Adjust numrecs for the later get_*_keys() calls. */
2671         lrecs -= rrecs;
2672         xfs_btree_set_numrecs(left, lrecs);
2673         xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2674
2675         /*
2676          * Copy btree block entries from the left block over to the
2677          * new block, the right. Update the right block and log the
2678          * changes.
2679          */
2680         if (level > 0) {
2681                 /* It's a non-leaf.  Move keys and pointers. */
2682                 union xfs_btree_key     *lkp;   /* left btree key */
2683                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2684                 union xfs_btree_key     *rkp;   /* right btree key */
2685                 union xfs_btree_ptr     *rpp;   /* right address pointer */
2686
2687                 lkp = xfs_btree_key_addr(cur, src_index, left);
2688                 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2689                 rkp = xfs_btree_key_addr(cur, 1, right);
2690                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2691
2692                 for (i = src_index; i < rrecs; i++) {
2693                         error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
2694                         if (error)
2695                                 goto error0;
2696                 }
2697
2698                 /* Copy the keys & pointers to the new block. */
2699                 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2700                 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2701
2702                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2703                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2704
2705                 /* Stash the keys of the new block for later insertion. */
2706                 xfs_btree_get_node_keys(cur, right, key);
2707         } else {
2708                 /* It's a leaf.  Move records.  */
2709                 union xfs_btree_rec     *lrp;   /* left record pointer */
2710                 union xfs_btree_rec     *rrp;   /* right record pointer */
2711
2712                 lrp = xfs_btree_rec_addr(cur, src_index, left);
2713                 rrp = xfs_btree_rec_addr(cur, 1, right);
2714
2715                 /* Copy records to the new block. */
2716                 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2717                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2718
2719                 /* Stash the keys of the new block for later insertion. */
2720                 xfs_btree_get_leaf_keys(cur, right, key);
2721         }
2722
2723         /*
2724          * Find the left block number by looking in the buffer.
2725          * Adjust sibling pointers.
2726          */
2727         xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2728         xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2729         xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2730         xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2731
2732         xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2733         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2734
2735         /*
2736          * If there's a block to the new block's right, make that block
2737          * point back to right instead of to left.
2738          */
2739         if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2740                 error = xfs_btree_read_buf_block(cur, &rrptr,
2741                                                         0, &rrblock, &rrbp);
2742                 if (error)
2743                         goto error0;
2744                 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2745                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2746         }
2747
2748         /* Update the parent high keys of the left block, if needed. */
2749         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2750                 error = xfs_btree_update_keys(cur, level);
2751                 if (error)
2752                         goto error0;
2753         }
2754
2755         /*
2756          * If the cursor is really in the right block, move it there.
2757          * If it's just pointing past the last entry in left, then we'll
2758          * insert there, so don't change anything in that case.
2759          */
2760         if (cur->bc_ptrs[level] > lrecs + 1) {
2761                 xfs_btree_setbuf(cur, level, rbp);
2762                 cur->bc_ptrs[level] -= lrecs;
2763         }
2764         /*
2765          * If there are more levels, we'll need another cursor which refers
2766          * the right block, no matter where this cursor was.
2767          */
2768         if (level + 1 < cur->bc_nlevels) {
2769                 error = xfs_btree_dup_cursor(cur, curp);
2770                 if (error)
2771                         goto error0;
2772                 (*curp)->bc_ptrs[level + 1]++;
2773         }
2774         *ptrp = rptr;
2775         *stat = 1;
2776         return 0;
2777 out0:
2778         *stat = 0;
2779         return 0;
2780
2781 error0:
2782         return error;
2783 }
2784
2785 struct xfs_btree_split_args {
2786         struct xfs_btree_cur    *cur;
2787         int                     level;
2788         union xfs_btree_ptr     *ptrp;
2789         union xfs_btree_key     *key;
2790         struct xfs_btree_cur    **curp;
2791         int                     *stat;          /* success/failure */
2792         int                     result;
2793         bool                    kswapd; /* allocation in kswapd context */
2794         struct completion       *done;
2795         struct work_struct      work;
2796 };
2797
2798 /*
2799  * Stack switching interfaces for allocation
2800  */
2801 static void
2802 xfs_btree_split_worker(
2803         struct work_struct      *work)
2804 {
2805         struct xfs_btree_split_args     *args = container_of(work,
2806                                                 struct xfs_btree_split_args, work);
2807         unsigned long           pflags;
2808         unsigned long           new_pflags = PF_MEMALLOC_NOFS;
2809
2810         /*
2811          * we are in a transaction context here, but may also be doing work
2812          * in kswapd context, and hence we may need to inherit that state
2813          * temporarily to ensure that we don't block waiting for memory reclaim
2814          * in any way.
2815          */
2816         if (args->kswapd)
2817                 new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
2818
2819         current_set_flags_nested(&pflags, new_pflags);
2820
2821         args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2822                                          args->key, args->curp, args->stat);
2823         complete(args->done);
2824
2825         current_restore_flags_nested(&pflags, new_pflags);
2826 }
2827
2828 /*
2829  * BMBT split requests often come in with little stack to work on. Push
2830  * them off to a worker thread so there is lots of stack to use. For the other
2831  * btree types, just call directly to avoid the context switch overhead here.
2832  */
2833 STATIC int                                      /* error */
2834 xfs_btree_split(
2835         struct xfs_btree_cur    *cur,
2836         int                     level,
2837         union xfs_btree_ptr     *ptrp,
2838         union xfs_btree_key     *key,
2839         struct xfs_btree_cur    **curp,
2840         int                     *stat)          /* success/failure */
2841 {
2842         struct xfs_btree_split_args     args;
2843         DECLARE_COMPLETION_ONSTACK(done);
2844
2845         if (cur->bc_btnum != XFS_BTNUM_BMAP)
2846                 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2847
2848         args.cur = cur;
2849         args.level = level;
2850         args.ptrp = ptrp;
2851         args.key = key;
2852         args.curp = curp;
2853         args.stat = stat;
2854         args.done = &done;
2855         args.kswapd = current_is_kswapd();
2856         INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2857         queue_work(xfs_alloc_wq, &args.work);
2858         wait_for_completion(&done);
2859         destroy_work_on_stack(&args.work);
2860         return args.result;
2861 }
2862
2863
2864 /*
2865  * Copy the old inode root contents into a real block and make the
2866  * broot point to it.
2867  */
2868 int                                             /* error */
2869 xfs_btree_new_iroot(
2870         struct xfs_btree_cur    *cur,           /* btree cursor */
2871         int                     *logflags,      /* logging flags for inode */
2872         int                     *stat)          /* return status - 0 fail */
2873 {
2874         struct xfs_buf          *cbp;           /* buffer for cblock */
2875         struct xfs_btree_block  *block;         /* btree block */
2876         struct xfs_btree_block  *cblock;        /* child btree block */
2877         union xfs_btree_key     *ckp;           /* child key pointer */
2878         union xfs_btree_ptr     *cpp;           /* child ptr pointer */
2879         union xfs_btree_key     *kp;            /* pointer to btree key */
2880         union xfs_btree_ptr     *pp;            /* pointer to block addr */
2881         union xfs_btree_ptr     nptr;           /* new block addr */
2882         int                     level;          /* btree level */
2883         int                     error;          /* error return code */
2884         int                     i;              /* loop counter */
2885
2886         XFS_BTREE_STATS_INC(cur, newroot);
2887
2888         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2889
2890         level = cur->bc_nlevels - 1;
2891
2892         block = xfs_btree_get_iroot(cur);
2893         pp = xfs_btree_ptr_addr(cur, 1, block);
2894
2895         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2896         error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
2897         if (error)
2898                 goto error0;
2899         if (*stat == 0)
2900                 return 0;
2901
2902         XFS_BTREE_STATS_INC(cur, alloc);
2903
2904         /* Copy the root into a real block. */
2905         error = xfs_btree_get_buf_block(cur, &nptr, &cblock, &cbp);
2906         if (error)
2907                 goto error0;
2908
2909         /*
2910          * we can't just memcpy() the root in for CRC enabled btree blocks.
2911          * In that case have to also ensure the blkno remains correct
2912          */
2913         memcpy(cblock, block, xfs_btree_block_len(cur));
2914         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
2915                 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
2916                         cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn);
2917                 else
2918                         cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn);
2919         }
2920
2921         be16_add_cpu(&block->bb_level, 1);
2922         xfs_btree_set_numrecs(block, 1);
2923         cur->bc_nlevels++;
2924         cur->bc_ptrs[level + 1] = 1;
2925
2926         kp = xfs_btree_key_addr(cur, 1, block);
2927         ckp = xfs_btree_key_addr(cur, 1, cblock);
2928         xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
2929
2930         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2931         for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
2932                 error = xfs_btree_debug_check_ptr(cur, pp, i, level);
2933                 if (error)
2934                         goto error0;
2935         }
2936
2937         xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
2938
2939         error = xfs_btree_debug_check_ptr(cur, &nptr, 0, level);
2940         if (error)
2941                 goto error0;
2942
2943         xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
2944
2945         xfs_iroot_realloc(cur->bc_ino.ip,
2946                           1 - xfs_btree_get_numrecs(cblock),
2947                           cur->bc_ino.whichfork);
2948
2949         xfs_btree_setbuf(cur, level, cbp);
2950
2951         /*
2952          * Do all this logging at the end so that
2953          * the root is at the right level.
2954          */
2955         xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
2956         xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2957         xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2958
2959         *logflags |=
2960                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork);
2961         *stat = 1;
2962         return 0;
2963 error0:
2964         return error;
2965 }
2966
2967 /*
2968  * Allocate a new root block, fill it in.
2969  */
2970 STATIC int                              /* error */
2971 xfs_btree_new_root(
2972         struct xfs_btree_cur    *cur,   /* btree cursor */
2973         int                     *stat)  /* success/failure */
2974 {
2975         struct xfs_btree_block  *block; /* one half of the old root block */
2976         struct xfs_buf          *bp;    /* buffer containing block */
2977         int                     error;  /* error return value */
2978         struct xfs_buf          *lbp;   /* left buffer pointer */
2979         struct xfs_btree_block  *left;  /* left btree block */
2980         struct xfs_buf          *nbp;   /* new (root) buffer */
2981         struct xfs_btree_block  *new;   /* new (root) btree block */
2982         int                     nptr;   /* new value for key index, 1 or 2 */
2983         struct xfs_buf          *rbp;   /* right buffer pointer */
2984         struct xfs_btree_block  *right; /* right btree block */
2985         union xfs_btree_ptr     rptr;
2986         union xfs_btree_ptr     lptr;
2987
2988         XFS_BTREE_STATS_INC(cur, newroot);
2989
2990         /* initialise our start point from the cursor */
2991         cur->bc_ops->init_ptr_from_cur(cur, &rptr);
2992
2993         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2994         error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
2995         if (error)
2996                 goto error0;
2997         if (*stat == 0)
2998                 goto out0;
2999         XFS_BTREE_STATS_INC(cur, alloc);
3000
3001         /* Set up the new block. */
3002         error = xfs_btree_get_buf_block(cur, &lptr, &new, &nbp);
3003         if (error)
3004                 goto error0;
3005
3006         /* Set the root in the holding structure  increasing the level by 1. */
3007         cur->bc_ops->set_root(cur, &lptr, 1);
3008
3009         /*
3010          * At the previous root level there are now two blocks: the old root,
3011          * and the new block generated when it was split.  We don't know which
3012          * one the cursor is pointing at, so we set up variables "left" and
3013          * "right" for each case.
3014          */
3015         block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3016
3017 #ifdef DEBUG
3018         error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3019         if (error)
3020                 goto error0;
3021 #endif
3022
3023         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3024         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3025                 /* Our block is left, pick up the right block. */
3026                 lbp = bp;
3027                 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3028                 left = block;
3029                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3030                 if (error)
3031                         goto error0;
3032                 bp = rbp;
3033                 nptr = 1;
3034         } else {
3035                 /* Our block is right, pick up the left block. */
3036                 rbp = bp;
3037                 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3038                 right = block;
3039                 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3040                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3041                 if (error)
3042                         goto error0;
3043                 bp = lbp;
3044                 nptr = 2;
3045         }
3046
3047         /* Fill in the new block's btree header and log it. */
3048         xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3049         xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3050         ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3051                         !xfs_btree_ptr_is_null(cur, &rptr));
3052
3053         /* Fill in the key data in the new root. */
3054         if (xfs_btree_get_level(left) > 0) {
3055                 /*
3056                  * Get the keys for the left block's keys and put them directly
3057                  * in the parent block.  Do the same for the right block.
3058                  */
3059                 xfs_btree_get_node_keys(cur, left,
3060                                 xfs_btree_key_addr(cur, 1, new));
3061                 xfs_btree_get_node_keys(cur, right,
3062                                 xfs_btree_key_addr(cur, 2, new));
3063         } else {
3064                 /*
3065                  * Get the keys for the left block's records and put them
3066                  * directly in the parent block.  Do the same for the right
3067                  * block.
3068                  */
3069                 xfs_btree_get_leaf_keys(cur, left,
3070                         xfs_btree_key_addr(cur, 1, new));
3071                 xfs_btree_get_leaf_keys(cur, right,
3072                         xfs_btree_key_addr(cur, 2, new));
3073         }
3074         xfs_btree_log_keys(cur, nbp, 1, 2);
3075
3076         /* Fill in the pointer data in the new root. */
3077         xfs_btree_copy_ptrs(cur,
3078                 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3079         xfs_btree_copy_ptrs(cur,
3080                 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3081         xfs_btree_log_ptrs(cur, nbp, 1, 2);
3082
3083         /* Fix up the cursor. */
3084         xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3085         cur->bc_ptrs[cur->bc_nlevels] = nptr;
3086         cur->bc_nlevels++;
3087         *stat = 1;
3088         return 0;
3089 error0:
3090         return error;
3091 out0:
3092         *stat = 0;
3093         return 0;
3094 }
3095
3096 STATIC int
3097 xfs_btree_make_block_unfull(
3098         struct xfs_btree_cur    *cur,   /* btree cursor */
3099         int                     level,  /* btree level */
3100         int                     numrecs,/* # of recs in block */
3101         int                     *oindex,/* old tree index */
3102         int                     *index, /* new tree index */
3103         union xfs_btree_ptr     *nptr,  /* new btree ptr */
3104         struct xfs_btree_cur    **ncur, /* new btree cursor */
3105         union xfs_btree_key     *key,   /* key of new block */
3106         int                     *stat)
3107 {
3108         int                     error = 0;
3109
3110         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3111             level == cur->bc_nlevels - 1) {
3112                 struct xfs_inode *ip = cur->bc_ino.ip;
3113
3114                 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3115                         /* A root block that can be made bigger. */
3116                         xfs_iroot_realloc(ip, 1, cur->bc_ino.whichfork);
3117                         *stat = 1;
3118                 } else {
3119                         /* A root block that needs replacing */
3120                         int     logflags = 0;
3121
3122                         error = xfs_btree_new_iroot(cur, &logflags, stat);
3123                         if (error || *stat == 0)
3124                                 return error;
3125
3126                         xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3127                 }
3128
3129                 return 0;
3130         }
3131
3132         /* First, try shifting an entry to the right neighbor. */
3133         error = xfs_btree_rshift(cur, level, stat);
3134         if (error || *stat)
3135                 return error;
3136
3137         /* Next, try shifting an entry to the left neighbor. */
3138         error = xfs_btree_lshift(cur, level, stat);
3139         if (error)
3140                 return error;
3141
3142         if (*stat) {
3143                 *oindex = *index = cur->bc_ptrs[level];
3144                 return 0;
3145         }
3146
3147         /*
3148          * Next, try splitting the current block in half.
3149          *
3150          * If this works we have to re-set our variables because we
3151          * could be in a different block now.
3152          */
3153         error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3154         if (error || *stat == 0)
3155                 return error;
3156
3157
3158         *index = cur->bc_ptrs[level];
3159         return 0;
3160 }
3161
3162 /*
3163  * Insert one record/level.  Return information to the caller
3164  * allowing the next level up to proceed if necessary.
3165  */
3166 STATIC int
3167 xfs_btree_insrec(
3168         struct xfs_btree_cur    *cur,   /* btree cursor */
3169         int                     level,  /* level to insert record at */
3170         union xfs_btree_ptr     *ptrp,  /* i/o: block number inserted */
3171         union xfs_btree_rec     *rec,   /* record to insert */
3172         union xfs_btree_key     *key,   /* i/o: block key for ptrp */
3173         struct xfs_btree_cur    **curp, /* output: new cursor replacing cur */
3174         int                     *stat)  /* success/failure */
3175 {
3176         struct xfs_btree_block  *block; /* btree block */
3177         struct xfs_buf          *bp;    /* buffer for block */
3178         union xfs_btree_ptr     nptr;   /* new block ptr */
3179         struct xfs_btree_cur    *ncur;  /* new btree cursor */
3180         union xfs_btree_key     nkey;   /* new block key */
3181         union xfs_btree_key     *lkey;
3182         int                     optr;   /* old key/record index */
3183         int                     ptr;    /* key/record index */
3184         int                     numrecs;/* number of records */
3185         int                     error;  /* error return value */
3186         int                     i;
3187         xfs_daddr_t             old_bn;
3188
3189         ncur = NULL;
3190         lkey = &nkey;
3191
3192         /*
3193          * If we have an external root pointer, and we've made it to the
3194          * root level, allocate a new root block and we're done.
3195          */
3196         if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3197             (level >= cur->bc_nlevels)) {
3198                 error = xfs_btree_new_root(cur, stat);
3199                 xfs_btree_set_ptr_null(cur, ptrp);
3200
3201                 return error;
3202         }
3203
3204         /* If we're off the left edge, return failure. */
3205         ptr = cur->bc_ptrs[level];
3206         if (ptr == 0) {
3207                 *stat = 0;
3208                 return 0;
3209         }
3210
3211         optr = ptr;
3212
3213         XFS_BTREE_STATS_INC(cur, insrec);
3214
3215         /* Get pointers to the btree buffer and block. */
3216         block = xfs_btree_get_block(cur, level, &bp);
3217         old_bn = bp ? bp->b_bn : XFS_BUF_DADDR_NULL;
3218         numrecs = xfs_btree_get_numrecs(block);
3219
3220 #ifdef DEBUG
3221         error = xfs_btree_check_block(cur, block, level, bp);
3222         if (error)
3223                 goto error0;
3224
3225         /* Check that the new entry is being inserted in the right place. */
3226         if (ptr <= numrecs) {
3227                 if (level == 0) {
3228                         ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3229                                 xfs_btree_rec_addr(cur, ptr, block)));
3230                 } else {
3231                         ASSERT(cur->bc_ops->keys_inorder(cur, key,
3232                                 xfs_btree_key_addr(cur, ptr, block)));
3233                 }
3234         }
3235 #endif
3236
3237         /*
3238          * If the block is full, we can't insert the new entry until we
3239          * make the block un-full.
3240          */
3241         xfs_btree_set_ptr_null(cur, &nptr);
3242         if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3243                 error = xfs_btree_make_block_unfull(cur, level, numrecs,
3244                                         &optr, &ptr, &nptr, &ncur, lkey, stat);
3245                 if (error || *stat == 0)
3246                         goto error0;
3247         }
3248
3249         /*
3250          * The current block may have changed if the block was
3251          * previously full and we have just made space in it.
3252          */
3253         block = xfs_btree_get_block(cur, level, &bp);
3254         numrecs = xfs_btree_get_numrecs(block);
3255
3256 #ifdef DEBUG
3257         error = xfs_btree_check_block(cur, block, level, bp);
3258         if (error)
3259                 return error;
3260 #endif
3261
3262         /*
3263          * At this point we know there's room for our new entry in the block
3264          * we're pointing at.
3265          */
3266         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3267
3268         if (level > 0) {
3269                 /* It's a nonleaf. make a hole in the keys and ptrs */
3270                 union xfs_btree_key     *kp;
3271                 union xfs_btree_ptr     *pp;
3272
3273                 kp = xfs_btree_key_addr(cur, ptr, block);
3274                 pp = xfs_btree_ptr_addr(cur, ptr, block);
3275
3276                 for (i = numrecs - ptr; i >= 0; i--) {
3277                         error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3278                         if (error)
3279                                 return error;
3280                 }
3281
3282                 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3283                 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3284
3285                 error = xfs_btree_debug_check_ptr(cur, ptrp, 0, level);
3286                 if (error)
3287                         goto error0;
3288
3289                 /* Now put the new data in, bump numrecs and log it. */
3290                 xfs_btree_copy_keys(cur, kp, key, 1);
3291                 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3292                 numrecs++;
3293                 xfs_btree_set_numrecs(block, numrecs);
3294                 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3295                 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3296 #ifdef DEBUG
3297                 if (ptr < numrecs) {
3298                         ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3299                                 xfs_btree_key_addr(cur, ptr + 1, block)));
3300                 }
3301 #endif
3302         } else {
3303                 /* It's a leaf. make a hole in the records */
3304                 union xfs_btree_rec             *rp;
3305
3306                 rp = xfs_btree_rec_addr(cur, ptr, block);
3307
3308                 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3309
3310                 /* Now put the new data in, bump numrecs and log it. */
3311                 xfs_btree_copy_recs(cur, rp, rec, 1);
3312                 xfs_btree_set_numrecs(block, ++numrecs);
3313                 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3314 #ifdef DEBUG
3315                 if (ptr < numrecs) {
3316                         ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3317                                 xfs_btree_rec_addr(cur, ptr + 1, block)));
3318                 }
3319 #endif
3320         }
3321
3322         /* Log the new number of records in the btree header. */
3323         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3324
3325         /*
3326          * If we just inserted into a new tree block, we have to
3327          * recalculate nkey here because nkey is out of date.
3328          *
3329          * Otherwise we're just updating an existing block (having shoved
3330          * some records into the new tree block), so use the regular key
3331          * update mechanism.
3332          */
3333         if (bp && bp->b_bn != old_bn) {
3334                 xfs_btree_get_keys(cur, block, lkey);
3335         } else if (xfs_btree_needs_key_update(cur, optr)) {
3336                 error = xfs_btree_update_keys(cur, level);
3337                 if (error)
3338                         goto error0;
3339         }
3340
3341         /*
3342          * If we are tracking the last record in the tree and
3343          * we are at the far right edge of the tree, update it.
3344          */
3345         if (xfs_btree_is_lastrec(cur, block, level)) {
3346                 cur->bc_ops->update_lastrec(cur, block, rec,
3347                                             ptr, LASTREC_INSREC);
3348         }
3349
3350         /*
3351          * Return the new block number, if any.
3352          * If there is one, give back a record value and a cursor too.
3353          */
3354         *ptrp = nptr;
3355         if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3356                 xfs_btree_copy_keys(cur, key, lkey, 1);
3357                 *curp = ncur;
3358         }
3359
3360         *stat = 1;
3361         return 0;
3362
3363 error0:
3364         return error;
3365 }
3366
3367 /*
3368  * Insert the record at the point referenced by cur.
3369  *
3370  * A multi-level split of the tree on insert will invalidate the original
3371  * cursor.  All callers of this function should assume that the cursor is
3372  * no longer valid and revalidate it.
3373  */
3374 int
3375 xfs_btree_insert(
3376         struct xfs_btree_cur    *cur,
3377         int                     *stat)
3378 {
3379         int                     error;  /* error return value */
3380         int                     i;      /* result value, 0 for failure */
3381         int                     level;  /* current level number in btree */
3382         union xfs_btree_ptr     nptr;   /* new block number (split result) */
3383         struct xfs_btree_cur    *ncur;  /* new cursor (split result) */
3384         struct xfs_btree_cur    *pcur;  /* previous level's cursor */
3385         union xfs_btree_key     bkey;   /* key of block to insert */
3386         union xfs_btree_key     *key;
3387         union xfs_btree_rec     rec;    /* record to insert */
3388
3389         level = 0;
3390         ncur = NULL;
3391         pcur = cur;
3392         key = &bkey;
3393
3394         xfs_btree_set_ptr_null(cur, &nptr);
3395
3396         /* Make a key out of the record data to be inserted, and save it. */
3397         cur->bc_ops->init_rec_from_cur(cur, &rec);
3398         cur->bc_ops->init_key_from_rec(key, &rec);
3399
3400         /*
3401          * Loop going up the tree, starting at the leaf level.
3402          * Stop when we don't get a split block, that must mean that
3403          * the insert is finished with this level.
3404          */
3405         do {
3406                 /*
3407                  * Insert nrec/nptr into this level of the tree.
3408                  * Note if we fail, nptr will be null.
3409                  */
3410                 error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3411                                 &ncur, &i);
3412                 if (error) {
3413                         if (pcur != cur)
3414                                 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3415                         goto error0;
3416                 }
3417
3418                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3419                         error = -EFSCORRUPTED;
3420                         goto error0;
3421                 }
3422                 level++;
3423
3424                 /*
3425                  * See if the cursor we just used is trash.
3426                  * Can't trash the caller's cursor, but otherwise we should
3427                  * if ncur is a new cursor or we're about to be done.
3428                  */
3429                 if (pcur != cur &&
3430                     (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3431                         /* Save the state from the cursor before we trash it */
3432                         if (cur->bc_ops->update_cursor)
3433                                 cur->bc_ops->update_cursor(pcur, cur);
3434                         cur->bc_nlevels = pcur->bc_nlevels;
3435                         xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3436                 }
3437                 /* If we got a new cursor, switch to it. */
3438                 if (ncur) {
3439                         pcur = ncur;
3440                         ncur = NULL;
3441                 }
3442         } while (!xfs_btree_ptr_is_null(cur, &nptr));
3443
3444         *stat = i;
3445         return 0;
3446 error0:
3447         return error;
3448 }
3449
3450 /*
3451  * Try to merge a non-leaf block back into the inode root.
3452  *
3453  * Note: the killroot names comes from the fact that we're effectively
3454  * killing the old root block.  But because we can't just delete the
3455  * inode we have to copy the single block it was pointing to into the
3456  * inode.
3457  */
3458 STATIC int
3459 xfs_btree_kill_iroot(
3460         struct xfs_btree_cur    *cur)
3461 {
3462         int                     whichfork = cur->bc_ino.whichfork;
3463         struct xfs_inode        *ip = cur->bc_ino.ip;
3464         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3465         struct xfs_btree_block  *block;
3466         struct xfs_btree_block  *cblock;
3467         union xfs_btree_key     *kp;
3468         union xfs_btree_key     *ckp;
3469         union xfs_btree_ptr     *pp;
3470         union xfs_btree_ptr     *cpp;
3471         struct xfs_buf          *cbp;
3472         int                     level;
3473         int                     index;
3474         int                     numrecs;
3475         int                     error;
3476 #ifdef DEBUG
3477         union xfs_btree_ptr     ptr;
3478 #endif
3479         int                     i;
3480
3481         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3482         ASSERT(cur->bc_nlevels > 1);
3483
3484         /*
3485          * Don't deal with the root block needs to be a leaf case.
3486          * We're just going to turn the thing back into extents anyway.
3487          */
3488         level = cur->bc_nlevels - 1;
3489         if (level == 1)
3490                 goto out0;
3491
3492         /*
3493          * Give up if the root has multiple children.
3494          */
3495         block = xfs_btree_get_iroot(cur);
3496         if (xfs_btree_get_numrecs(block) != 1)
3497                 goto out0;
3498
3499         cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3500         numrecs = xfs_btree_get_numrecs(cblock);
3501
3502         /*
3503          * Only do this if the next level will fit.
3504          * Then the data must be copied up to the inode,
3505          * instead of freeing the root you free the next level.
3506          */
3507         if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3508                 goto out0;
3509
3510         XFS_BTREE_STATS_INC(cur, killroot);
3511
3512 #ifdef DEBUG
3513         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3514         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3515         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3516         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3517 #endif
3518
3519         index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3520         if (index) {
3521                 xfs_iroot_realloc(cur->bc_ino.ip, index,
3522                                   cur->bc_ino.whichfork);
3523                 block = ifp->if_broot;
3524         }
3525
3526         be16_add_cpu(&block->bb_numrecs, index);
3527         ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3528
3529         kp = xfs_btree_key_addr(cur, 1, block);
3530         ckp = xfs_btree_key_addr(cur, 1, cblock);
3531         xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3532
3533         pp = xfs_btree_ptr_addr(cur, 1, block);
3534         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3535
3536         for (i = 0; i < numrecs; i++) {
3537                 error = xfs_btree_debug_check_ptr(cur, cpp, i, level - 1);
3538                 if (error)
3539                         return error;
3540         }
3541
3542         xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3543
3544         error = xfs_btree_free_block(cur, cbp);
3545         if (error)
3546                 return error;
3547
3548         cur->bc_bufs[level - 1] = NULL;
3549         be16_add_cpu(&block->bb_level, -1);
3550         xfs_trans_log_inode(cur->bc_tp, ip,
3551                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork));
3552         cur->bc_nlevels--;
3553 out0:
3554         return 0;
3555 }
3556
3557 /*
3558  * Kill the current root node, and replace it with it's only child node.
3559  */
3560 STATIC int
3561 xfs_btree_kill_root(
3562         struct xfs_btree_cur    *cur,
3563         struct xfs_buf          *bp,
3564         int                     level,
3565         union xfs_btree_ptr     *newroot)
3566 {
3567         int                     error;
3568
3569         XFS_BTREE_STATS_INC(cur, killroot);
3570
3571         /*
3572          * Update the root pointer, decreasing the level by 1 and then
3573          * free the old root.
3574          */
3575         cur->bc_ops->set_root(cur, newroot, -1);
3576
3577         error = xfs_btree_free_block(cur, bp);
3578         if (error)
3579                 return error;
3580
3581         cur->bc_bufs[level] = NULL;
3582         cur->bc_ra[level] = 0;
3583         cur->bc_nlevels--;
3584
3585         return 0;
3586 }
3587
3588 STATIC int
3589 xfs_btree_dec_cursor(
3590         struct xfs_btree_cur    *cur,
3591         int                     level,
3592         int                     *stat)
3593 {
3594         int                     error;
3595         int                     i;
3596
3597         if (level > 0) {
3598                 error = xfs_btree_decrement(cur, level, &i);
3599                 if (error)
3600                         return error;
3601         }
3602
3603         *stat = 1;
3604         return 0;
3605 }
3606
3607 /*
3608  * Single level of the btree record deletion routine.
3609  * Delete record pointed to by cur/level.
3610  * Remove the record from its block then rebalance the tree.
3611  * Return 0 for error, 1 for done, 2 to go on to the next level.
3612  */
3613 STATIC int                                      /* error */
3614 xfs_btree_delrec(
3615         struct xfs_btree_cur    *cur,           /* btree cursor */
3616         int                     level,          /* level removing record from */
3617         int                     *stat)          /* fail/done/go-on */
3618 {
3619         struct xfs_btree_block  *block;         /* btree block */
3620         union xfs_btree_ptr     cptr;           /* current block ptr */
3621         struct xfs_buf          *bp;            /* buffer for block */
3622         int                     error;          /* error return value */
3623         int                     i;              /* loop counter */
3624         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
3625         struct xfs_buf          *lbp;           /* left buffer pointer */
3626         struct xfs_btree_block  *left;          /* left btree block */
3627         int                     lrecs = 0;      /* left record count */
3628         int                     ptr;            /* key/record index */
3629         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
3630         struct xfs_buf          *rbp;           /* right buffer pointer */
3631         struct xfs_btree_block  *right;         /* right btree block */
3632         struct xfs_btree_block  *rrblock;       /* right-right btree block */
3633         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
3634         int                     rrecs = 0;      /* right record count */
3635         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
3636         int                     numrecs;        /* temporary numrec count */
3637
3638         tcur = NULL;
3639
3640         /* Get the index of the entry being deleted, check for nothing there. */
3641         ptr = cur->bc_ptrs[level];
3642         if (ptr == 0) {
3643                 *stat = 0;
3644                 return 0;
3645         }
3646
3647         /* Get the buffer & block containing the record or key/ptr. */
3648         block = xfs_btree_get_block(cur, level, &bp);
3649         numrecs = xfs_btree_get_numrecs(block);
3650
3651 #ifdef DEBUG
3652         error = xfs_btree_check_block(cur, block, level, bp);
3653         if (error)
3654                 goto error0;
3655 #endif
3656
3657         /* Fail if we're off the end of the block. */
3658         if (ptr > numrecs) {
3659                 *stat = 0;
3660                 return 0;
3661         }
3662
3663         XFS_BTREE_STATS_INC(cur, delrec);
3664         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3665
3666         /* Excise the entries being deleted. */
3667         if (level > 0) {
3668                 /* It's a nonleaf. operate on keys and ptrs */
3669                 union xfs_btree_key     *lkp;
3670                 union xfs_btree_ptr     *lpp;
3671
3672                 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3673                 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3674
3675                 for (i = 0; i < numrecs - ptr; i++) {
3676                         error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
3677                         if (error)
3678                                 goto error0;
3679                 }
3680
3681                 if (ptr < numrecs) {
3682                         xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3683                         xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3684                         xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3685                         xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3686                 }
3687         } else {
3688                 /* It's a leaf. operate on records */
3689                 if (ptr < numrecs) {
3690                         xfs_btree_shift_recs(cur,
3691                                 xfs_btree_rec_addr(cur, ptr + 1, block),
3692                                 -1, numrecs - ptr);
3693                         xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3694                 }
3695         }
3696
3697         /*
3698          * Decrement and log the number of entries in the block.
3699          */
3700         xfs_btree_set_numrecs(block, --numrecs);
3701         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3702
3703         /*
3704          * If we are tracking the last record in the tree and
3705          * we are at the far right edge of the tree, update it.
3706          */
3707         if (xfs_btree_is_lastrec(cur, block, level)) {
3708                 cur->bc_ops->update_lastrec(cur, block, NULL,
3709                                             ptr, LASTREC_DELREC);
3710         }
3711
3712         /*
3713          * We're at the root level.  First, shrink the root block in-memory.
3714          * Try to get rid of the next level down.  If we can't then there's
3715          * nothing left to do.
3716          */
3717         if (level == cur->bc_nlevels - 1) {
3718                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3719                         xfs_iroot_realloc(cur->bc_ino.ip, -1,
3720                                           cur->bc_ino.whichfork);
3721
3722                         error = xfs_btree_kill_iroot(cur);
3723                         if (error)
3724                                 goto error0;
3725
3726                         error = xfs_btree_dec_cursor(cur, level, stat);
3727                         if (error)
3728                                 goto error0;
3729                         *stat = 1;
3730                         return 0;
3731                 }
3732
3733                 /*
3734                  * If this is the root level, and there's only one entry left,
3735                  * and it's NOT the leaf level, then we can get rid of this
3736                  * level.
3737                  */
3738                 if (numrecs == 1 && level > 0) {
3739                         union xfs_btree_ptr     *pp;
3740                         /*
3741                          * pp is still set to the first pointer in the block.
3742                          * Make it the new root of the btree.
3743                          */
3744                         pp = xfs_btree_ptr_addr(cur, 1, block);
3745                         error = xfs_btree_kill_root(cur, bp, level, pp);
3746                         if (error)
3747                                 goto error0;
3748                 } else if (level > 0) {
3749                         error = xfs_btree_dec_cursor(cur, level, stat);
3750                         if (error)
3751                                 goto error0;
3752                 }
3753                 *stat = 1;
3754                 return 0;
3755         }
3756
3757         /*
3758          * If we deleted the leftmost entry in the block, update the
3759          * key values above us in the tree.
3760          */
3761         if (xfs_btree_needs_key_update(cur, ptr)) {
3762                 error = xfs_btree_update_keys(cur, level);
3763                 if (error)
3764                         goto error0;
3765         }
3766
3767         /*
3768          * If the number of records remaining in the block is at least
3769          * the minimum, we're done.
3770          */
3771         if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3772                 error = xfs_btree_dec_cursor(cur, level, stat);
3773                 if (error)
3774                         goto error0;
3775                 return 0;
3776         }
3777
3778         /*
3779          * Otherwise, we have to move some records around to keep the
3780          * tree balanced.  Look at the left and right sibling blocks to
3781          * see if we can re-balance by moving only one record.
3782          */
3783         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3784         xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3785
3786         if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3787                 /*
3788                  * One child of root, need to get a chance to copy its contents
3789                  * into the root and delete it. Can't go up to next level,
3790                  * there's nothing to delete there.
3791                  */
3792                 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3793                     xfs_btree_ptr_is_null(cur, &lptr) &&
3794                     level == cur->bc_nlevels - 2) {
3795                         error = xfs_btree_kill_iroot(cur);
3796                         if (!error)
3797                                 error = xfs_btree_dec_cursor(cur, level, stat);
3798                         if (error)
3799                                 goto error0;
3800                         return 0;
3801                 }
3802         }
3803
3804         ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3805                !xfs_btree_ptr_is_null(cur, &lptr));
3806
3807         /*
3808          * Duplicate the cursor so our btree manipulations here won't
3809          * disrupt the next level up.
3810          */
3811         error = xfs_btree_dup_cursor(cur, &tcur);
3812         if (error)
3813                 goto error0;
3814
3815         /*
3816          * If there's a right sibling, see if it's ok to shift an entry
3817          * out of it.
3818          */
3819         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3820                 /*
3821                  * Move the temp cursor to the last entry in the next block.
3822                  * Actually any entry but the first would suffice.
3823                  */
3824                 i = xfs_btree_lastrec(tcur, level);
3825                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3826                         error = -EFSCORRUPTED;
3827                         goto error0;
3828                 }
3829
3830                 error = xfs_btree_increment(tcur, level, &i);
3831                 if (error)
3832                         goto error0;
3833                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3834                         error = -EFSCORRUPTED;
3835                         goto error0;
3836                 }
3837
3838                 i = xfs_btree_lastrec(tcur, level);
3839                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3840                         error = -EFSCORRUPTED;
3841                         goto error0;
3842                 }
3843
3844                 /* Grab a pointer to the block. */
3845                 right = xfs_btree_get_block(tcur, level, &rbp);
3846 #ifdef DEBUG
3847                 error = xfs_btree_check_block(tcur, right, level, rbp);
3848                 if (error)
3849                         goto error0;
3850 #endif
3851                 /* Grab the current block number, for future use. */
3852                 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3853
3854                 /*
3855                  * If right block is full enough so that removing one entry
3856                  * won't make it too empty, and left-shifting an entry out
3857                  * of right to us works, we're done.
3858                  */
3859                 if (xfs_btree_get_numrecs(right) - 1 >=
3860                     cur->bc_ops->get_minrecs(tcur, level)) {
3861                         error = xfs_btree_lshift(tcur, level, &i);
3862                         if (error)
3863                                 goto error0;
3864                         if (i) {
3865                                 ASSERT(xfs_btree_get_numrecs(block) >=
3866                                        cur->bc_ops->get_minrecs(tcur, level));
3867
3868                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3869                                 tcur = NULL;
3870
3871                                 error = xfs_btree_dec_cursor(cur, level, stat);
3872                                 if (error)
3873                                         goto error0;
3874                                 return 0;
3875                         }
3876                 }
3877
3878                 /*
3879                  * Otherwise, grab the number of records in right for
3880                  * future reference, and fix up the temp cursor to point
3881                  * to our block again (last record).
3882                  */
3883                 rrecs = xfs_btree_get_numrecs(right);
3884                 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3885                         i = xfs_btree_firstrec(tcur, level);
3886                         if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3887                                 error = -EFSCORRUPTED;
3888                                 goto error0;
3889                         }
3890
3891                         error = xfs_btree_decrement(tcur, level, &i);
3892                         if (error)
3893                                 goto error0;
3894                         if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3895                                 error = -EFSCORRUPTED;
3896                                 goto error0;
3897                         }
3898                 }
3899         }
3900
3901         /*
3902          * If there's a left sibling, see if it's ok to shift an entry
3903          * out of it.
3904          */
3905         if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3906                 /*
3907                  * Move the temp cursor to the first entry in the
3908                  * previous block.
3909                  */
3910                 i = xfs_btree_firstrec(tcur, level);
3911                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3912                         error = -EFSCORRUPTED;
3913                         goto error0;
3914                 }
3915
3916                 error = xfs_btree_decrement(tcur, level, &i);
3917                 if (error)
3918                         goto error0;
3919                 i = xfs_btree_firstrec(tcur, level);
3920                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3921                         error = -EFSCORRUPTED;
3922                         goto error0;
3923                 }
3924
3925                 /* Grab a pointer to the block. */
3926                 left = xfs_btree_get_block(tcur, level, &lbp);
3927 #ifdef DEBUG
3928                 error = xfs_btree_check_block(cur, left, level, lbp);
3929                 if (error)
3930                         goto error0;
3931 #endif
3932                 /* Grab the current block number, for future use. */
3933                 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
3934
3935                 /*
3936                  * If left block is full enough so that removing one entry
3937                  * won't make it too empty, and right-shifting an entry out
3938                  * of left to us works, we're done.
3939                  */
3940                 if (xfs_btree_get_numrecs(left) - 1 >=
3941                     cur->bc_ops->get_minrecs(tcur, level)) {
3942                         error = xfs_btree_rshift(tcur, level, &i);
3943                         if (error)
3944                                 goto error0;
3945                         if (i) {
3946                                 ASSERT(xfs_btree_get_numrecs(block) >=
3947                                        cur->bc_ops->get_minrecs(tcur, level));
3948                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3949                                 tcur = NULL;
3950                                 if (level == 0)
3951                                         cur->bc_ptrs[0]++;
3952
3953                                 *stat = 1;
3954                                 return 0;
3955                         }
3956                 }
3957
3958                 /*
3959                  * Otherwise, grab the number of records in right for
3960                  * future reference.
3961                  */
3962                 lrecs = xfs_btree_get_numrecs(left);
3963         }
3964
3965         /* Delete the temp cursor, we're done with it. */
3966         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3967         tcur = NULL;
3968
3969         /* If here, we need to do a join to keep the tree balanced. */
3970         ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
3971
3972         if (!xfs_btree_ptr_is_null(cur, &lptr) &&
3973             lrecs + xfs_btree_get_numrecs(block) <=
3974                         cur->bc_ops->get_maxrecs(cur, level)) {
3975                 /*
3976                  * Set "right" to be the starting block,
3977                  * "left" to be the left neighbor.
3978                  */
3979                 rptr = cptr;
3980                 right = block;
3981                 rbp = bp;
3982                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3983                 if (error)
3984                         goto error0;
3985
3986         /*
3987          * If that won't work, see if we can join with the right neighbor block.
3988          */
3989         } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
3990                    rrecs + xfs_btree_get_numrecs(block) <=
3991                         cur->bc_ops->get_maxrecs(cur, level)) {
3992                 /*
3993                  * Set "left" to be the starting block,
3994                  * "right" to be the right neighbor.
3995                  */
3996                 lptr = cptr;
3997                 left = block;
3998                 lbp = bp;
3999                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4000                 if (error)
4001                         goto error0;
4002
4003         /*
4004          * Otherwise, we can't fix the imbalance.
4005          * Just return.  This is probably a logic error, but it's not fatal.
4006          */
4007         } else {
4008                 error = xfs_btree_dec_cursor(cur, level, stat);
4009                 if (error)
4010                         goto error0;
4011                 return 0;
4012         }
4013
4014         rrecs = xfs_btree_get_numrecs(right);
4015         lrecs = xfs_btree_get_numrecs(left);
4016
4017         /*
4018          * We're now going to join "left" and "right" by moving all the stuff
4019          * in "right" to "left" and deleting "right".
4020          */
4021         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4022         if (level > 0) {
4023                 /* It's a non-leaf.  Move keys and pointers. */
4024                 union xfs_btree_key     *lkp;   /* left btree key */
4025                 union xfs_btree_ptr     *lpp;   /* left address pointer */
4026                 union xfs_btree_key     *rkp;   /* right btree key */
4027                 union xfs_btree_ptr     *rpp;   /* right address pointer */
4028
4029                 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4030                 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4031                 rkp = xfs_btree_key_addr(cur, 1, right);
4032                 rpp = xfs_btree_ptr_addr(cur, 1, right);
4033
4034                 for (i = 1; i < rrecs; i++) {
4035                         error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
4036                         if (error)
4037                                 goto error0;
4038                 }
4039
4040                 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4041                 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4042
4043                 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4044                 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4045         } else {
4046                 /* It's a leaf.  Move records.  */
4047                 union xfs_btree_rec     *lrp;   /* left record pointer */
4048                 union xfs_btree_rec     *rrp;   /* right record pointer */
4049
4050                 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4051                 rrp = xfs_btree_rec_addr(cur, 1, right);
4052
4053                 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4054                 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4055         }
4056
4057         XFS_BTREE_STATS_INC(cur, join);
4058
4059         /*
4060          * Fix up the number of records and right block pointer in the
4061          * surviving block, and log it.
4062          */
4063         xfs_btree_set_numrecs(left, lrecs + rrecs);
4064         xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB);
4065         xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4066         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4067
4068         /* If there is a right sibling, point it to the remaining block. */
4069         xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4070         if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4071                 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4072                 if (error)
4073                         goto error0;
4074                 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4075                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4076         }
4077
4078         /* Free the deleted block. */
4079         error = xfs_btree_free_block(cur, rbp);
4080         if (error)
4081                 goto error0;
4082
4083         /*
4084          * If we joined with the left neighbor, set the buffer in the
4085          * cursor to the left block, and fix up the index.
4086          */
4087         if (bp != lbp) {
4088                 cur->bc_bufs[level] = lbp;
4089                 cur->bc_ptrs[level] += lrecs;
4090                 cur->bc_ra[level] = 0;
4091         }
4092         /*
4093          * If we joined with the right neighbor and there's a level above
4094          * us, increment the cursor at that level.
4095          */
4096         else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
4097                    (level + 1 < cur->bc_nlevels)) {
4098                 error = xfs_btree_increment(cur, level + 1, &i);
4099                 if (error)
4100                         goto error0;
4101         }
4102
4103         /*
4104          * Readjust the ptr at this level if it's not a leaf, since it's
4105          * still pointing at the deletion point, which makes the cursor
4106          * inconsistent.  If this makes the ptr 0, the caller fixes it up.
4107          * We can't use decrement because it would change the next level up.
4108          */
4109         if (level > 0)
4110                 cur->bc_ptrs[level]--;
4111
4112         /*
4113          * We combined blocks, so we have to update the parent keys if the
4114          * btree supports overlapped intervals.  However, bc_ptrs[level + 1]
4115          * points to the old block so that the caller knows which record to
4116          * delete.  Therefore, the caller must be savvy enough to call updkeys
4117          * for us if we return stat == 2.  The other exit points from this
4118          * function don't require deletions further up the tree, so they can
4119          * call updkeys directly.
4120          */
4121
4122         /* Return value means the next level up has something to do. */
4123         *stat = 2;
4124         return 0;
4125
4126 error0:
4127         if (tcur)
4128                 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4129         return error;
4130 }
4131
4132 /*
4133  * Delete the record pointed to by cur.
4134  * The cursor refers to the place where the record was (could be inserted)
4135  * when the operation returns.
4136  */
4137 int                                     /* error */
4138 xfs_btree_delete(
4139         struct xfs_btree_cur    *cur,
4140         int                     *stat)  /* success/failure */
4141 {
4142         int                     error;  /* error return value */
4143         int                     level;
4144         int                     i;
4145         bool                    joined = false;
4146
4147         /*
4148          * Go up the tree, starting at leaf level.
4149          *
4150          * If 2 is returned then a join was done; go to the next level.
4151          * Otherwise we are done.
4152          */
4153         for (level = 0, i = 2; i == 2; level++) {
4154                 error = xfs_btree_delrec(cur, level, &i);
4155                 if (error)
4156                         goto error0;
4157                 if (i == 2)
4158                         joined = true;
4159         }
4160
4161         /*
4162          * If we combined blocks as part of deleting the record, delrec won't
4163          * have updated the parent high keys so we have to do that here.
4164          */
4165         if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) {
4166                 error = xfs_btree_updkeys_force(cur, 0);
4167                 if (error)
4168                         goto error0;
4169         }
4170
4171         if (i == 0) {
4172                 for (level = 1; level < cur->bc_nlevels; level++) {
4173                         if (cur->bc_ptrs[level] == 0) {
4174                                 error = xfs_btree_decrement(cur, level, &i);
4175                                 if (error)
4176                                         goto error0;
4177                                 break;
4178                         }
4179                 }
4180         }
4181
4182         *stat = i;
4183         return 0;
4184 error0:
4185         return error;
4186 }
4187
4188 /*
4189  * Get the data from the pointed-to record.
4190  */
4191 int                                     /* error */
4192 xfs_btree_get_rec(
4193         struct xfs_btree_cur    *cur,   /* btree cursor */
4194         union xfs_btree_rec     **recp, /* output: btree record */
4195         int                     *stat)  /* output: success/failure */
4196 {
4197         struct xfs_btree_block  *block; /* btree block */
4198         struct xfs_buf          *bp;    /* buffer pointer */
4199         int                     ptr;    /* record number */
4200 #ifdef DEBUG
4201         int                     error;  /* error return value */
4202 #endif
4203
4204         ptr = cur->bc_ptrs[0];
4205         block = xfs_btree_get_block(cur, 0, &bp);
4206
4207 #ifdef DEBUG
4208         error = xfs_btree_check_block(cur, block, 0, bp);
4209         if (error)
4210                 return error;
4211 #endif
4212
4213         /*
4214          * Off the right end or left end, return failure.
4215          */
4216         if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4217                 *stat = 0;
4218                 return 0;
4219         }
4220
4221         /*
4222          * Point to the record and extract its data.
4223          */
4224         *recp = xfs_btree_rec_addr(cur, ptr, block);
4225         *stat = 1;
4226         return 0;
4227 }
4228
4229 /* Visit a block in a btree. */
4230 STATIC int
4231 xfs_btree_visit_block(
4232         struct xfs_btree_cur            *cur,
4233         int                             level,
4234         xfs_btree_visit_blocks_fn       fn,
4235         void                            *data)
4236 {
4237         struct xfs_btree_block          *block;
4238         struct xfs_buf                  *bp;
4239         union xfs_btree_ptr             rptr;
4240         int                             error;
4241
4242         /* do right sibling readahead */
4243         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4244         block = xfs_btree_get_block(cur, level, &bp);
4245
4246         /* process the block */
4247         error = fn(cur, level, data);
4248         if (error)
4249                 return error;
4250
4251         /* now read rh sibling block for next iteration */
4252         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4253         if (xfs_btree_ptr_is_null(cur, &rptr))
4254                 return -ENOENT;
4255
4256         return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4257 }
4258
4259
4260 /* Visit every block in a btree. */
4261 int
4262 xfs_btree_visit_blocks(
4263         struct xfs_btree_cur            *cur,
4264         xfs_btree_visit_blocks_fn       fn,
4265         unsigned int                    flags,
4266         void                            *data)
4267 {
4268         union xfs_btree_ptr             lptr;
4269         int                             level;
4270         struct xfs_btree_block          *block = NULL;
4271         int                             error = 0;
4272
4273         cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4274
4275         /* for each level */
4276         for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4277                 /* grab the left hand block */
4278                 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4279                 if (error)
4280                         return error;
4281
4282                 /* readahead the left most block for the next level down */
4283                 if (level > 0) {
4284                         union xfs_btree_ptr     *ptr;
4285
4286                         ptr = xfs_btree_ptr_addr(cur, 1, block);
4287                         xfs_btree_readahead_ptr(cur, ptr, 1);
4288
4289                         /* save for the next iteration of the loop */
4290                         xfs_btree_copy_ptrs(cur, &lptr, ptr, 1);
4291
4292                         if (!(flags & XFS_BTREE_VISIT_LEAVES))
4293                                 continue;
4294                 } else if (!(flags & XFS_BTREE_VISIT_RECORDS)) {
4295                         continue;
4296                 }
4297
4298                 /* for each buffer in the level */
4299                 do {
4300                         error = xfs_btree_visit_block(cur, level, fn, data);
4301                 } while (!error);
4302
4303                 if (error != -ENOENT)
4304                         return error;
4305         }
4306
4307         return 0;
4308 }
4309
4310 /*
4311  * Change the owner of a btree.
4312  *
4313  * The mechanism we use here is ordered buffer logging. Because we don't know
4314  * how many buffers were are going to need to modify, we don't really want to
4315  * have to make transaction reservations for the worst case of every buffer in a
4316  * full size btree as that may be more space that we can fit in the log....
4317  *
4318  * We do the btree walk in the most optimal manner possible - we have sibling
4319  * pointers so we can just walk all the blocks on each level from left to right
4320  * in a single pass, and then move to the next level and do the same. We can
4321  * also do readahead on the sibling pointers to get IO moving more quickly,
4322  * though for slow disks this is unlikely to make much difference to performance
4323  * as the amount of CPU work we have to do before moving to the next block is
4324  * relatively small.
4325  *
4326  * For each btree block that we load, modify the owner appropriately, set the
4327  * buffer as an ordered buffer and log it appropriately. We need to ensure that
4328  * we mark the region we change dirty so that if the buffer is relogged in
4329  * a subsequent transaction the changes we make here as an ordered buffer are
4330  * correctly relogged in that transaction.  If we are in recovery context, then
4331  * just queue the modified buffer as delayed write buffer so the transaction
4332  * recovery completion writes the changes to disk.
4333  */
4334 struct xfs_btree_block_change_owner_info {
4335         uint64_t                new_owner;
4336         struct list_head        *buffer_list;
4337 };
4338
4339 static int
4340 xfs_btree_block_change_owner(
4341         struct xfs_btree_cur    *cur,
4342         int                     level,
4343         void                    *data)
4344 {
4345         struct xfs_btree_block_change_owner_info        *bbcoi = data;
4346         struct xfs_btree_block  *block;
4347         struct xfs_buf          *bp;
4348
4349         /* modify the owner */
4350         block = xfs_btree_get_block(cur, level, &bp);
4351         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
4352                 if (block->bb_u.l.bb_owner == cpu_to_be64(bbcoi->new_owner))
4353                         return 0;
4354                 block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4355         } else {
4356                 if (block->bb_u.s.bb_owner == cpu_to_be32(bbcoi->new_owner))
4357                         return 0;
4358                 block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4359         }
4360
4361         /*
4362          * If the block is a root block hosted in an inode, we might not have a
4363          * buffer pointer here and we shouldn't attempt to log the change as the
4364          * information is already held in the inode and discarded when the root
4365          * block is formatted into the on-disk inode fork. We still change it,
4366          * though, so everything is consistent in memory.
4367          */
4368         if (!bp) {
4369                 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4370                 ASSERT(level == cur->bc_nlevels - 1);
4371                 return 0;
4372         }
4373
4374         if (cur->bc_tp) {
4375                 if (!xfs_trans_ordered_buf(cur->bc_tp, bp)) {
4376                         xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4377                         return -EAGAIN;
4378                 }
4379         } else {
4380                 xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4381         }
4382
4383         return 0;
4384 }
4385
4386 int
4387 xfs_btree_change_owner(
4388         struct xfs_btree_cur    *cur,
4389         uint64_t                new_owner,
4390         struct list_head        *buffer_list)
4391 {
4392         struct xfs_btree_block_change_owner_info        bbcoi;
4393
4394         bbcoi.new_owner = new_owner;
4395         bbcoi.buffer_list = buffer_list;
4396
4397         return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4398                         XFS_BTREE_VISIT_ALL, &bbcoi);
4399 }
4400
4401 /* Verify the v5 fields of a long-format btree block. */
4402 xfs_failaddr_t
4403 xfs_btree_lblock_v5hdr_verify(
4404         struct xfs_buf          *bp,
4405         uint64_t                owner)
4406 {
4407         struct xfs_mount        *mp = bp->b_mount;
4408         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4409
4410         if (!xfs_sb_version_hascrc(&mp->m_sb))
4411                 return __this_address;
4412         if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
4413                 return __this_address;
4414         if (block->bb_u.l.bb_blkno != cpu_to_be64(bp->b_bn))
4415                 return __this_address;
4416         if (owner != XFS_RMAP_OWN_UNKNOWN &&
4417             be64_to_cpu(block->bb_u.l.bb_owner) != owner)
4418                 return __this_address;
4419         return NULL;
4420 }
4421
4422 /* Verify a long-format btree block. */
4423 xfs_failaddr_t
4424 xfs_btree_lblock_verify(
4425         struct xfs_buf          *bp,
4426         unsigned int            max_recs)
4427 {
4428         struct xfs_mount        *mp = bp->b_mount;
4429         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4430
4431         /* numrecs verification */
4432         if (be16_to_cpu(block->bb_numrecs) > max_recs)
4433                 return __this_address;
4434
4435         /* sibling pointer verification */
4436         if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) &&
4437             !xfs_verify_fsbno(mp, be64_to_cpu(block->bb_u.l.bb_leftsib)))
4438                 return __this_address;
4439         if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) &&
4440             !xfs_verify_fsbno(mp, be64_to_cpu(block->bb_u.l.bb_rightsib)))
4441                 return __this_address;
4442
4443         return NULL;
4444 }
4445
4446 /**
4447  * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4448  *                                    btree block
4449  *
4450  * @bp: buffer containing the btree block
4451  */
4452 xfs_failaddr_t
4453 xfs_btree_sblock_v5hdr_verify(
4454         struct xfs_buf          *bp)
4455 {
4456         struct xfs_mount        *mp = bp->b_mount;
4457         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4458         struct xfs_perag        *pag = bp->b_pag;
4459
4460         if (!xfs_sb_version_hascrc(&mp->m_sb))
4461                 return __this_address;
4462         if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4463                 return __this_address;
4464         if (block->bb_u.s.bb_blkno != cpu_to_be64(bp->b_bn))
4465                 return __this_address;
4466         if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4467                 return __this_address;
4468         return NULL;
4469 }
4470
4471 /**
4472  * xfs_btree_sblock_verify() -- verify a short-format btree block
4473  *
4474  * @bp: buffer containing the btree block
4475  * @max_recs: maximum records allowed in this btree node
4476  */
4477 xfs_failaddr_t
4478 xfs_btree_sblock_verify(
4479         struct xfs_buf          *bp,
4480         unsigned int            max_recs)
4481 {
4482         struct xfs_mount        *mp = bp->b_mount;
4483         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4484         xfs_agblock_t           agno;
4485
4486         /* numrecs verification */
4487         if (be16_to_cpu(block->bb_numrecs) > max_recs)
4488                 return __this_address;
4489
4490         /* sibling pointer verification */
4491         agno = xfs_daddr_to_agno(mp, XFS_BUF_ADDR(bp));
4492         if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) &&
4493             !xfs_verify_agbno(mp, agno, be32_to_cpu(block->bb_u.s.bb_leftsib)))
4494                 return __this_address;
4495         if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) &&
4496             !xfs_verify_agbno(mp, agno, be32_to_cpu(block->bb_u.s.bb_rightsib)))
4497                 return __this_address;
4498
4499         return NULL;
4500 }
4501
4502 /*
4503  * Calculate the number of btree levels needed to store a given number of
4504  * records in a short-format btree.
4505  */
4506 uint
4507 xfs_btree_compute_maxlevels(
4508         uint                    *limits,
4509         unsigned long           len)
4510 {
4511         uint                    level;
4512         unsigned long           maxblocks;
4513
4514         maxblocks = (len + limits[0] - 1) / limits[0];
4515         for (level = 1; maxblocks > 1; level++)
4516                 maxblocks = (maxblocks + limits[1] - 1) / limits[1];
4517         return level;
4518 }
4519
4520 /*
4521  * Query a regular btree for all records overlapping a given interval.
4522  * Start with a LE lookup of the key of low_rec and return all records
4523  * until we find a record with a key greater than the key of high_rec.
4524  */
4525 STATIC int
4526 xfs_btree_simple_query_range(
4527         struct xfs_btree_cur            *cur,
4528         union xfs_btree_key             *low_key,
4529         union xfs_btree_key             *high_key,
4530         xfs_btree_query_range_fn        fn,
4531         void                            *priv)
4532 {
4533         union xfs_btree_rec             *recp;
4534         union xfs_btree_key             rec_key;
4535         int64_t                         diff;
4536         int                             stat;
4537         bool                            firstrec = true;
4538         int                             error;
4539
4540         ASSERT(cur->bc_ops->init_high_key_from_rec);
4541         ASSERT(cur->bc_ops->diff_two_keys);
4542
4543         /*
4544          * Find the leftmost record.  The btree cursor must be set
4545          * to the low record used to generate low_key.
4546          */
4547         stat = 0;
4548         error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4549         if (error)
4550                 goto out;
4551
4552         /* Nothing?  See if there's anything to the right. */
4553         if (!stat) {
4554                 error = xfs_btree_increment(cur, 0, &stat);
4555                 if (error)
4556                         goto out;
4557         }
4558
4559         while (stat) {
4560                 /* Find the record. */
4561                 error = xfs_btree_get_rec(cur, &recp, &stat);
4562                 if (error || !stat)
4563                         break;
4564
4565                 /* Skip if high_key(rec) < low_key. */
4566                 if (firstrec) {
4567                         cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4568                         firstrec = false;
4569                         diff = cur->bc_ops->diff_two_keys(cur, low_key,
4570                                         &rec_key);
4571                         if (diff > 0)
4572                                 goto advloop;
4573                 }
4574
4575                 /* Stop if high_key < low_key(rec). */
4576                 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4577                 diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key);
4578                 if (diff > 0)
4579                         break;
4580
4581                 /* Callback */
4582                 error = fn(cur, recp, priv);
4583                 if (error)
4584                         break;
4585
4586 advloop:
4587                 /* Move on to the next record. */
4588                 error = xfs_btree_increment(cur, 0, &stat);
4589                 if (error)
4590                         break;
4591         }
4592
4593 out:
4594         return error;
4595 }
4596
4597 /*
4598  * Query an overlapped interval btree for all records overlapping a given
4599  * interval.  This function roughly follows the algorithm given in
4600  * "Interval Trees" of _Introduction to Algorithms_, which is section
4601  * 14.3 in the 2nd and 3rd editions.
4602  *
4603  * First, generate keys for the low and high records passed in.
4604  *
4605  * For any leaf node, generate the high and low keys for the record.
4606  * If the record keys overlap with the query low/high keys, pass the
4607  * record to the function iterator.
4608  *
4609  * For any internal node, compare the low and high keys of each
4610  * pointer against the query low/high keys.  If there's an overlap,
4611  * follow the pointer.
4612  *
4613  * As an optimization, we stop scanning a block when we find a low key
4614  * that is greater than the query's high key.
4615  */
4616 STATIC int
4617 xfs_btree_overlapped_query_range(
4618         struct xfs_btree_cur            *cur,
4619         union xfs_btree_key             *low_key,
4620         union xfs_btree_key             *high_key,
4621         xfs_btree_query_range_fn        fn,
4622         void                            *priv)
4623 {
4624         union xfs_btree_ptr             ptr;
4625         union xfs_btree_ptr             *pp;
4626         union xfs_btree_key             rec_key;
4627         union xfs_btree_key             rec_hkey;
4628         union xfs_btree_key             *lkp;
4629         union xfs_btree_key             *hkp;
4630         union xfs_btree_rec             *recp;
4631         struct xfs_btree_block          *block;
4632         int64_t                         ldiff;
4633         int64_t                         hdiff;
4634         int                             level;
4635         struct xfs_buf                  *bp;
4636         int                             i;
4637         int                             error;
4638
4639         /* Load the root of the btree. */
4640         level = cur->bc_nlevels - 1;
4641         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
4642         error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4643         if (error)
4644                 return error;
4645         xfs_btree_get_block(cur, level, &bp);
4646         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4647 #ifdef DEBUG
4648         error = xfs_btree_check_block(cur, block, level, bp);
4649         if (error)
4650                 goto out;
4651 #endif
4652         cur->bc_ptrs[level] = 1;
4653
4654         while (level < cur->bc_nlevels) {
4655                 block = xfs_btree_get_block(cur, level, &bp);
4656
4657                 /* End of node, pop back towards the root. */
4658                 if (cur->bc_ptrs[level] > be16_to_cpu(block->bb_numrecs)) {
4659 pop_up:
4660                         if (level < cur->bc_nlevels - 1)
4661                                 cur->bc_ptrs[level + 1]++;
4662                         level++;
4663                         continue;
4664                 }
4665
4666                 if (level == 0) {
4667                         /* Handle a leaf node. */
4668                         recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
4669
4670                         cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
4671                         ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
4672                                         low_key);
4673
4674                         cur->bc_ops->init_key_from_rec(&rec_key, recp);
4675                         hdiff = cur->bc_ops->diff_two_keys(cur, high_key,
4676                                         &rec_key);
4677
4678                         /*
4679                          * If (record's high key >= query's low key) and
4680                          *    (query's high key >= record's low key), then
4681                          * this record overlaps the query range; callback.
4682                          */
4683                         if (ldiff >= 0 && hdiff >= 0) {
4684                                 error = fn(cur, recp, priv);
4685                                 if (error)
4686                                         break;
4687                         } else if (hdiff < 0) {
4688                                 /* Record is larger than high key; pop. */
4689                                 goto pop_up;
4690                         }
4691                         cur->bc_ptrs[level]++;
4692                         continue;
4693                 }
4694
4695                 /* Handle an internal node. */
4696                 lkp = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block);
4697                 hkp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block);
4698                 pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block);
4699
4700                 ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
4701                 hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
4702
4703                 /*
4704                  * If (pointer's high key >= query's low key) and
4705                  *    (query's high key >= pointer's low key), then
4706                  * this record overlaps the query range; follow pointer.
4707                  */
4708                 if (ldiff >= 0 && hdiff >= 0) {
4709                         level--;
4710                         error = xfs_btree_lookup_get_block(cur, level, pp,
4711                                         &block);
4712                         if (error)
4713                                 goto out;
4714                         xfs_btree_get_block(cur, level, &bp);
4715                         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4716 #ifdef DEBUG
4717                         error = xfs_btree_check_block(cur, block, level, bp);
4718                         if (error)
4719                                 goto out;
4720 #endif
4721                         cur->bc_ptrs[level] = 1;
4722                         continue;
4723                 } else if (hdiff < 0) {
4724                         /* The low key is larger than the upper range; pop. */
4725                         goto pop_up;
4726                 }
4727                 cur->bc_ptrs[level]++;
4728         }
4729
4730 out:
4731         /*
4732          * If we don't end this function with the cursor pointing at a record
4733          * block, a subsequent non-error cursor deletion will not release
4734          * node-level buffers, causing a buffer leak.  This is quite possible
4735          * with a zero-results range query, so release the buffers if we
4736          * failed to return any results.
4737          */
4738         if (cur->bc_bufs[0] == NULL) {
4739                 for (i = 0; i < cur->bc_nlevels; i++) {
4740                         if (cur->bc_bufs[i]) {
4741                                 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
4742                                 cur->bc_bufs[i] = NULL;
4743                                 cur->bc_ptrs[i] = 0;
4744                                 cur->bc_ra[i] = 0;
4745                         }
4746                 }
4747         }
4748
4749         return error;
4750 }
4751
4752 /*
4753  * Query a btree for all records overlapping a given interval of keys.  The
4754  * supplied function will be called with each record found; return one of the
4755  * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4756  * code.  This function returns -ECANCELED, zero, or a negative error code.
4757  */
4758 int
4759 xfs_btree_query_range(
4760         struct xfs_btree_cur            *cur,
4761         union xfs_btree_irec            *low_rec,
4762         union xfs_btree_irec            *high_rec,
4763         xfs_btree_query_range_fn        fn,
4764         void                            *priv)
4765 {
4766         union xfs_btree_rec             rec;
4767         union xfs_btree_key             low_key;
4768         union xfs_btree_key             high_key;
4769
4770         /* Find the keys of both ends of the interval. */
4771         cur->bc_rec = *high_rec;
4772         cur->bc_ops->init_rec_from_cur(cur, &rec);
4773         cur->bc_ops->init_key_from_rec(&high_key, &rec);
4774
4775         cur->bc_rec = *low_rec;
4776         cur->bc_ops->init_rec_from_cur(cur, &rec);
4777         cur->bc_ops->init_key_from_rec(&low_key, &rec);
4778
4779         /* Enforce low key < high key. */
4780         if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0)
4781                 return -EINVAL;
4782
4783         if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
4784                 return xfs_btree_simple_query_range(cur, &low_key,
4785                                 &high_key, fn, priv);
4786         return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
4787                         fn, priv);
4788 }
4789
4790 /* Query a btree for all records. */
4791 int
4792 xfs_btree_query_all(
4793         struct xfs_btree_cur            *cur,
4794         xfs_btree_query_range_fn        fn,
4795         void                            *priv)
4796 {
4797         union xfs_btree_key             low_key;
4798         union xfs_btree_key             high_key;
4799
4800         memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
4801         memset(&low_key, 0, sizeof(low_key));
4802         memset(&high_key, 0xFF, sizeof(high_key));
4803
4804         return xfs_btree_simple_query_range(cur, &low_key, &high_key, fn, priv);
4805 }
4806
4807 /*
4808  * Calculate the number of blocks needed to store a given number of records
4809  * in a short-format (per-AG metadata) btree.
4810  */
4811 unsigned long long
4812 xfs_btree_calc_size(
4813         uint                    *limits,
4814         unsigned long long      len)
4815 {
4816         int                     level;
4817         int                     maxrecs;
4818         unsigned long long      rval;
4819
4820         maxrecs = limits[0];
4821         for (level = 0, rval = 0; len > 1; level++) {
4822                 len += maxrecs - 1;
4823                 do_div(len, maxrecs);
4824                 maxrecs = limits[1];
4825                 rval += len;
4826         }
4827         return rval;
4828 }
4829
4830 static int
4831 xfs_btree_count_blocks_helper(
4832         struct xfs_btree_cur    *cur,
4833         int                     level,
4834         void                    *data)
4835 {
4836         xfs_extlen_t            *blocks = data;
4837         (*blocks)++;
4838
4839         return 0;
4840 }
4841
4842 /* Count the blocks in a btree and return the result in *blocks. */
4843 int
4844 xfs_btree_count_blocks(
4845         struct xfs_btree_cur    *cur,
4846         xfs_extlen_t            *blocks)
4847 {
4848         *blocks = 0;
4849         return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
4850                         XFS_BTREE_VISIT_ALL, blocks);
4851 }
4852
4853 /* Compare two btree pointers. */
4854 int64_t
4855 xfs_btree_diff_two_ptrs(
4856         struct xfs_btree_cur            *cur,
4857         const union xfs_btree_ptr       *a,
4858         const union xfs_btree_ptr       *b)
4859 {
4860         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4861                 return (int64_t)be64_to_cpu(a->l) - be64_to_cpu(b->l);
4862         return (int64_t)be32_to_cpu(a->s) - be32_to_cpu(b->s);
4863 }
4864
4865 /* If there's an extent, we're done. */
4866 STATIC int
4867 xfs_btree_has_record_helper(
4868         struct xfs_btree_cur            *cur,
4869         union xfs_btree_rec             *rec,
4870         void                            *priv)
4871 {
4872         return -ECANCELED;
4873 }
4874
4875 /* Is there a record covering a given range of keys? */
4876 int
4877 xfs_btree_has_record(
4878         struct xfs_btree_cur    *cur,
4879         union xfs_btree_irec    *low,
4880         union xfs_btree_irec    *high,
4881         bool                    *exists)
4882 {
4883         int                     error;
4884
4885         error = xfs_btree_query_range(cur, low, high,
4886                         &xfs_btree_has_record_helper, NULL);
4887         if (error == -ECANCELED) {
4888                 *exists = true;
4889                 return 0;
4890         }
4891         *exists = false;
4892         return error;
4893 }
4894
4895 /* Are there more records in this btree? */
4896 bool
4897 xfs_btree_has_more_records(
4898         struct xfs_btree_cur    *cur)
4899 {
4900         struct xfs_btree_block  *block;
4901         struct xfs_buf          *bp;
4902
4903         block = xfs_btree_get_block(cur, 0, &bp);
4904
4905         /* There are still records in this block. */
4906         if (cur->bc_ptrs[0] < xfs_btree_get_numrecs(block))
4907                 return true;
4908
4909         /* There are more record blocks. */
4910         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4911                 return block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK);
4912         else
4913                 return block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK);
4914 }