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 #include "xfs_ag.h"
  25 #include "xfs_alloc_btree.h"
  26 #include "xfs_ialloc_btree.h"
  27 #include "xfs_bmap_btree.h"
  28 #include "xfs_rmap_btree.h"
  29 #include "xfs_refcount_btree.h"
  30
  31 /*
  32  * Btree magic numbers.
  33  */
  34 static const uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
  35         { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, 0, XFS_BMAP_MAGIC, XFS_IBT_MAGIC,
  36           XFS_FIBT_MAGIC, 0 },
  37         { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC, XFS_RMAP_CRC_MAGIC,
  38           XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC,
  39           XFS_REFC_CRC_MAGIC }
  40 };
  41
  42 uint32_t
  43 xfs_btree_magic(
  44         int                     crc,
  45         xfs_btnum_t             btnum)
  46 {
  47         uint32_t                magic = xfs_magics[crc][btnum];
  48
  49         /* Ensure we asked for crc for crc-only magics. */
  50         ASSERT(magic != 0);
  51         return magic;
  52 }
  53
  54 /*
  55  * These sibling pointer checks are optimised for null sibling pointers. This
  56  * happens a lot, and we don't need to byte swap at runtime if the sibling
  57  * pointer is NULL.
  58  *
  59  * These are explicitly marked at inline because the cost of calling them as
  60  * functions instead of inlining them is about 36 bytes extra code per call site
  61  * on x86-64. Yes, gcc-11 fails to inline them, and explicit inlining of these
  62  * two sibling check functions reduces the compiled code size by over 300
  63  * bytes.
  64  */
  65 static inline xfs_failaddr_t
  66 xfs_btree_check_lblock_siblings(
  67         struct xfs_mount        *mp,
  68         struct xfs_btree_cur    *cur,
  69         int                     level,
  70         xfs_fsblock_t           fsb,
  71         __be64                  dsibling)
  72 {
  73         xfs_fsblock_t           sibling;
  74
  75         if (dsibling == cpu_to_be64(NULLFSBLOCK))
  76                 return NULL;
  77
  78         sibling = be64_to_cpu(dsibling);
  79         if (sibling == fsb)
  80                 return __this_address;
  81         if (level >= 0) {
  82                 if (!xfs_btree_check_lptr(cur, sibling, level + 1))
  83                         return __this_address;
  84         } else {
  85                 if (!xfs_verify_fsbno(mp, sibling))
  86                         return __this_address;
  87         }
  88
  89         return NULL;
  90 }
  91
  92 static inline xfs_failaddr_t
  93 xfs_btree_check_sblock_siblings(
  94         struct xfs_perag        *pag,
  95         struct xfs_btree_cur    *cur,
  96         int                     level,
  97         xfs_agblock_t           agbno,
  98         __be32                  dsibling)
  99 {
 100         xfs_agblock_t           sibling;
 101
 102         if (dsibling == cpu_to_be32(NULLAGBLOCK))
 103                 return NULL;
 104
 105         sibling = be32_to_cpu(dsibling);
 106         if (sibling == agbno)
 107                 return __this_address;
 108         if (level >= 0) {
 109                 if (!xfs_btree_check_sptr(cur, sibling, level + 1))
 110                         return __this_address;
 111         } else {
 112                 if (!xfs_verify_agbno(pag, sibling))
 113                         return __this_address;
 114         }
 115         return NULL;
 116 }
 117
 118 /*
 119  * Check a long 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_lblock(
 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_has_crc(mp);
 132         xfs_failaddr_t          fa;
 133         xfs_fsblock_t           fsb = NULLFSBLOCK;
 134
 135         if (crc) {
 136                 if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
 137                         return __this_address;
 138                 if (block->bb_u.l.bb_blkno !=
 139                     cpu_to_be64(bp ? xfs_buf_daddr(bp) : XFS_BUF_DADDR_NULL))
 140                         return __this_address;
 141                 if (block->bb_u.l.bb_pad != cpu_to_be32(0))
 142                         return __this_address;
 143         }
 144
 145         if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
 146                 return __this_address;
 147         if (be16_to_cpu(block->bb_level) != level)
 148                 return __this_address;
 149         if (be16_to_cpu(block->bb_numrecs) >
 150             cur->bc_ops->get_maxrecs(cur, level))
 151                 return __this_address;
 152
 153         if (bp)
 154                 fsb = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
 155
 156         fa = xfs_btree_check_lblock_siblings(mp, cur, level, fsb,
 157                         block->bb_u.l.bb_leftsib);
 158         if (!fa)
 159                 fa = xfs_btree_check_lblock_siblings(mp, cur, level, fsb,
 160                                 block->bb_u.l.bb_rightsib);
 161         return fa;
 162 }
 163
 164 /* Check a long btree block header. */
 165 static int
 166 xfs_btree_check_lblock(
 167         struct xfs_btree_cur    *cur,
 168         struct xfs_btree_block  *block,
 169         int                     level,
 170         struct xfs_buf          *bp)
 171 {
 172         struct xfs_mount        *mp = cur->bc_mp;
 173         xfs_failaddr_t          fa;
 174
 175         fa = __xfs_btree_check_lblock(cur, block, level, bp);
 176         if (XFS_IS_CORRUPT(mp, fa != NULL) ||
 177             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BTREE_CHECK_LBLOCK)) {
 178                 if (bp)
 179                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 180                 return -EFSCORRUPTED;
 181         }
 182         return 0;
 183 }
 184
 185 /*
 186  * Check a short btree block header.  Return the address of the failing check,
 187  * or NULL if everything is ok.
 188  */
 189 xfs_failaddr_t
 190 __xfs_btree_check_sblock(
 191         struct xfs_btree_cur    *cur,
 192         struct xfs_btree_block  *block,
 193         int                     level,
 194         struct xfs_buf          *bp)
 195 {
 196         struct xfs_mount        *mp = cur->bc_mp;
 197         struct xfs_perag        *pag = cur->bc_ag.pag;
 198         xfs_btnum_t             btnum = cur->bc_btnum;
 199         int                     crc = xfs_has_crc(mp);
 200         xfs_failaddr_t          fa;
 201         xfs_agblock_t           agbno = NULLAGBLOCK;
 202
 203         if (crc) {
 204                 if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
 205                         return __this_address;
 206                 if (block->bb_u.s.bb_blkno !=
 207                     cpu_to_be64(bp ? xfs_buf_daddr(bp) : XFS_BUF_DADDR_NULL))
 208                         return __this_address;
 209         }
 210
 211         if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
 212                 return __this_address;
 213         if (be16_to_cpu(block->bb_level) != level)
 214                 return __this_address;
 215         if (be16_to_cpu(block->bb_numrecs) >
 216             cur->bc_ops->get_maxrecs(cur, level))
 217                 return __this_address;
 218
 219         if (bp)
 220                 agbno = xfs_daddr_to_agbno(mp, xfs_buf_daddr(bp));
 221
 222         fa = xfs_btree_check_sblock_siblings(pag, cur, level, agbno,
 223                         block->bb_u.s.bb_leftsib);
 224         if (!fa)
 225                 fa = xfs_btree_check_sblock_siblings(pag, cur, level, agbno,
 226                                 block->bb_u.s.bb_rightsib);
 227         return fa;
 228 }
 229
 230 /* Check a short btree block header. */
 231 STATIC int
 232 xfs_btree_check_sblock(
 233         struct xfs_btree_cur    *cur,
 234         struct xfs_btree_block  *block,
 235         int                     level,
 236         struct xfs_buf          *bp)
 237 {
 238         struct xfs_mount        *mp = cur->bc_mp;
 239         xfs_failaddr_t          fa;
 240
 241         fa = __xfs_btree_check_sblock(cur, block, level, bp);
 242         if (XFS_IS_CORRUPT(mp, fa != NULL) ||
 243             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BTREE_CHECK_SBLOCK)) {
 244                 if (bp)
 245                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 246                 return -EFSCORRUPTED;
 247         }
 248         return 0;
 249 }
 250
 251 /*
 252  * Debug routine: check that block header is ok.
 253  */
 254 int
 255 xfs_btree_check_block(
 256         struct xfs_btree_cur    *cur,   /* btree cursor */
 257         struct xfs_btree_block  *block, /* generic btree block pointer */
 258         int                     level,  /* level of the btree block */
 259         struct xfs_buf          *bp)    /* buffer containing block, if any */
 260 {
 261         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 262                 return xfs_btree_check_lblock(cur, block, level, bp);
 263         else
 264                 return xfs_btree_check_sblock(cur, block, level, bp);
 265 }
 266
 267 /* Check that this long pointer is valid and points within the fs. */
 268 bool
 269 xfs_btree_check_lptr(
 270         struct xfs_btree_cur    *cur,
 271         xfs_fsblock_t           fsbno,
 272         int                     level)
 273 {
 274         if (level <= 0)
 275                 return false;
 276         return xfs_verify_fsbno(cur->bc_mp, fsbno);
 277 }
 278
 279 /* Check that this short pointer is valid and points within the AG. */
 280 bool
 281 xfs_btree_check_sptr(
 282         struct xfs_btree_cur    *cur,
 283         xfs_agblock_t           agbno,
 284         int                     level)
 285 {
 286         if (level <= 0)
 287                 return false;
 288         return xfs_verify_agbno(cur->bc_ag.pag, agbno);
 289 }
 290
 291 /*
 292  * Check that a given (indexed) btree pointer at a certain level of a
 293  * btree is valid and doesn't point past where it should.
 294  */
 295 static int
 296 xfs_btree_check_ptr(
 297         struct xfs_btree_cur            *cur,
 298         const union xfs_btree_ptr       *ptr,
 299         int                             index,
 300         int                             level)
 301 {
 302         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 303                 if (xfs_btree_check_lptr(cur, be64_to_cpu((&ptr->l)[index]),
 304                                 level))
 305                         return 0;
 306                 xfs_err(cur->bc_mp,
 307 "Inode %llu fork %d: Corrupt btree %d pointer at level %d index %d.",
 308                                 cur->bc_ino.ip->i_ino,
 309                                 cur->bc_ino.whichfork, cur->bc_btnum,
 310                                 level, index);
 311         } else {
 312                 if (xfs_btree_check_sptr(cur, be32_to_cpu((&ptr->s)[index]),
 313                                 level))
 314                         return 0;
 315                 xfs_err(cur->bc_mp,
 316 "AG %u: Corrupt btree %d pointer at level %d index %d.",
 317                                 cur->bc_ag.pag->pag_agno, cur->bc_btnum,
 318                                 level, index);
 319         }
 320
 321         return -EFSCORRUPTED;
 322 }
 323
 324 #ifdef DEBUG
 325 # define xfs_btree_debug_check_ptr      xfs_btree_check_ptr
 326 #else
 327 # define xfs_btree_debug_check_ptr(...) (0)
 328 #endif
 329
 330 /*
 331  * Calculate CRC on the whole btree block and stuff it into the
 332  * long-form btree header.
 333  *
 334  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 335  * it into the buffer so recovery knows what the last modification was that made
 336  * it to disk.
 337  */
 338 void
 339 xfs_btree_lblock_calc_crc(
 340         struct xfs_buf          *bp)
 341 {
 342         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 343         struct xfs_buf_log_item *bip = bp->b_log_item;
 344
 345         if (!xfs_has_crc(bp->b_mount))
 346                 return;
 347         if (bip)
 348                 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 349         xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 350 }
 351
 352 bool
 353 xfs_btree_lblock_verify_crc(
 354         struct xfs_buf          *bp)
 355 {
 356         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 357         struct xfs_mount        *mp = bp->b_mount;
 358
 359         if (xfs_has_crc(mp)) {
 360                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
 361                         return false;
 362                 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 363         }
 364
 365         return true;
 366 }
 367
 368 /*
 369  * Calculate CRC on the whole btree block and stuff it into the
 370  * short-form btree header.
 371  *
 372  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 373  * it into the buffer so recovery knows what the last modification was that made
 374  * it to disk.
 375  */
 376 void
 377 xfs_btree_sblock_calc_crc(
 378         struct xfs_buf          *bp)
 379 {
 380         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 381         struct xfs_buf_log_item *bip = bp->b_log_item;
 382
 383         if (!xfs_has_crc(bp->b_mount))
 384                 return;
 385         if (bip)
 386                 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 387         xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 388 }
 389
 390 bool
 391 xfs_btree_sblock_verify_crc(
 392         struct xfs_buf          *bp)
 393 {
 394         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 395         struct xfs_mount        *mp = bp->b_mount;
 396
 397         if (xfs_has_crc(mp)) {
 398                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
 399                         return false;
 400                 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 401         }
 402
 403         return true;
 404 }
 405
 406 static int
 407 xfs_btree_free_block(
 408         struct xfs_btree_cur    *cur,
 409         struct xfs_buf          *bp)
 410 {
 411         int                     error;
 412
 413         error = cur->bc_ops->free_block(cur, bp);
 414         if (!error) {
 415                 xfs_trans_binval(cur->bc_tp, bp);
 416                 XFS_BTREE_STATS_INC(cur, free);
 417         }
 418         return error;
 419 }
 420
 421 /*
 422  * Delete the btree cursor.
 423  */
 424 void
 425 xfs_btree_del_cursor(
 426         struct xfs_btree_cur    *cur,           /* btree cursor */
 427         int                     error)          /* del because of error */
 428 {
 429         int                     i;              /* btree level */
 430
 431         /*
 432          * Clear the buffer pointers and release the buffers. If we're doing
 433          * this because of an error, inspect all of the entries in the bc_bufs
 434          * array for buffers to be unlocked. This is because some of the btree
 435          * code works from level n down to 0, and if we get an error along the
 436          * way we won't have initialized all the entries down to 0.
 437          */
 438         for (i = 0; i < cur->bc_nlevels; i++) {
 439                 if (cur->bc_levels[i].bp)
 440                         xfs_trans_brelse(cur->bc_tp, cur->bc_levels[i].bp);
 441                 else if (!error)
 442                         break;
 443         }
 444
 445         /*
 446          * If we are doing a BMBT update, the number of unaccounted blocks
 447          * allocated during this cursor life time should be zero. If it's not
 448          * zero, then we should be shut down or on our way to shutdown due to
 449          * cancelling a dirty transaction on error.
 450          */
 451         ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP || cur->bc_ino.allocated == 0 ||
 452                xfs_is_shutdown(cur->bc_mp) || error != 0);
 453         if (unlikely(cur->bc_flags & XFS_BTREE_STAGING))
 454                 kmem_free(cur->bc_ops);
 455         if (!(cur->bc_flags & XFS_BTREE_LONG_PTRS) && cur->bc_ag.pag)
 456                 xfs_perag_put(cur->bc_ag.pag);
 457         kmem_cache_free(cur->bc_cache, cur);
 458 }
 459
 460 /*
 461  * Duplicate the btree cursor.
 462  * Allocate a new one, copy the record, re-get the buffers.
 463  */
 464 int                                     /* error */
 465 xfs_btree_dup_cursor(
 466         struct xfs_btree_cur *cur,              /* input cursor */
 467         struct xfs_btree_cur **ncur)            /* output cursor */
 468 {
 469         struct xfs_buf  *bp;            /* btree block's buffer pointer */
 470         int             error;          /* error return value */
 471         int             i;              /* level number of btree block */
 472         xfs_mount_t     *mp;            /* mount structure for filesystem */
 473         struct xfs_btree_cur *new;              /* new cursor value */
 474         xfs_trans_t     *tp;            /* transaction pointer, can be NULL */
 475
 476         tp = cur->bc_tp;
 477         mp = cur->bc_mp;
 478
 479         /*
 480          * Allocate a new cursor like the old one.
 481          */
 482         new = cur->bc_ops->dup_cursor(cur);
 483
 484         /*
 485          * Copy the record currently in the cursor.
 486          */
 487         new->bc_rec = cur->bc_rec;
 488
 489         /*
 490          * For each level current, re-get the buffer and copy the ptr value.
 491          */
 492         for (i = 0; i < new->bc_nlevels; i++) {
 493                 new->bc_levels[i].ptr = cur->bc_levels[i].ptr;
 494                 new->bc_levels[i].ra = cur->bc_levels[i].ra;
 495                 bp = cur->bc_levels[i].bp;
 496                 if (bp) {
 497                         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
 498                                                    xfs_buf_daddr(bp), mp->m_bsize,
 499                                                    0, &bp,
 500                                                    cur->bc_ops->buf_ops);
 501                         if (error) {
 502                                 xfs_btree_del_cursor(new, error);
 503                                 *ncur = NULL;
 504                                 return error;
 505                         }
 506                 }
 507                 new->bc_levels[i].bp = bp;
 508         }
 509         *ncur = new;
 510         return 0;
 511 }
 512
 513 /*
 514  * XFS btree block layout and addressing:
 515  *
 516  * There are two types of blocks in the btree: leaf and non-leaf blocks.
 517  *
 518  * The leaf record start with a header then followed by records containing
 519  * the values.  A non-leaf block also starts with the same header, and
 520  * then first contains lookup keys followed by an equal number of pointers
 521  * to the btree blocks at the previous level.
 522  *
 523  *              +--------+-------+-------+-------+-------+-------+-------+
 524  * Leaf:        | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
 525  *              +--------+-------+-------+-------+-------+-------+-------+
 526  *
 527  *              +--------+-------+-------+-------+-------+-------+-------+
 528  * Non-Leaf:    | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
 529  *              +--------+-------+-------+-------+-------+-------+-------+
 530  *
 531  * The header is called struct xfs_btree_block for reasons better left unknown
 532  * and comes in different versions for short (32bit) and long (64bit) block
 533  * pointers.  The record and key structures are defined by the btree instances
 534  * and opaque to the btree core.  The block pointers are simple disk endian
 535  * integers, available in a short (32bit) and long (64bit) variant.
 536  *
 537  * The helpers below calculate the offset of a given record, key or pointer
 538  * into a btree block (xfs_btree_*_offset) or return a pointer to the given
 539  * record, key or pointer (xfs_btree_*_addr).  Note that all addressing
 540  * inside the btree block is done using indices starting at one, not zero!
 541  *
 542  * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing
 543  * overlapping intervals.  In such a tree, records are still sorted lowest to
 544  * highest and indexed by the smallest key value that refers to the record.
 545  * However, nodes are different: each pointer has two associated keys -- one
 546  * indexing the lowest key available in the block(s) below (the same behavior
 547  * as the key in a regular btree) and another indexing the highest key
 548  * available in the block(s) below.  Because records are /not/ sorted by the
 549  * highest key, all leaf block updates require us to compute the highest key
 550  * that matches any record in the leaf and to recursively update the high keys
 551  * in the nodes going further up in the tree, if necessary.  Nodes look like
 552  * this:
 553  *
 554  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 555  * Non-Leaf:    | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
 556  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 557  *
 558  * To perform an interval query on an overlapped tree, perform the usual
 559  * depth-first search and use the low and high keys to decide if we can skip
 560  * that particular node.  If a leaf node is reached, return the records that
 561  * intersect the interval.  Note that an interval query may return numerous
 562  * entries.  For a non-overlapped tree, simply search for the record associated
 563  * with the lowest key and iterate forward until a non-matching record is
 564  * found.  Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
 565  * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
 566  * more detail.
 567  *
 568  * Why do we care about overlapping intervals?  Let's say you have a bunch of
 569  * reverse mapping records on a reflink filesystem:
 570  *
 571  * 1: +- file A startblock B offset C length D -----------+
 572  * 2:      +- file E startblock F offset G length H --------------+
 573  * 3:      +- file I startblock F offset J length K --+
 574  * 4:                                                        +- file L... --+
 575  *
 576  * Now say we want to map block (B+D) into file A at offset (C+D).  Ideally,
 577  * we'd simply increment the length of record 1.  But how do we find the record
 578  * that ends at (B+D-1) (i.e. record 1)?  A LE lookup of (B+D-1) would return
 579  * record 3 because the keys are ordered first by startblock.  An interval
 580  * query would return records 1 and 2 because they both overlap (B+D-1), and
 581  * from that we can pick out record 1 as the appropriate left neighbor.
 582  *
 583  * In the non-overlapped case you can do a LE lookup and decrement the cursor
 584  * because a record's interval must end before the next record.
 585  */
 586
 587 /*
 588  * Return size of the btree block header for this btree instance.
 589  */
 590 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
 591 {
 592         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 593                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 594                         return XFS_BTREE_LBLOCK_CRC_LEN;
 595                 return XFS_BTREE_LBLOCK_LEN;
 596         }
 597         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 598                 return XFS_BTREE_SBLOCK_CRC_LEN;
 599         return XFS_BTREE_SBLOCK_LEN;
 600 }
 601
 602 /*
 603  * Return size of btree block pointers for this btree instance.
 604  */
 605 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
 606 {
 607         return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
 608                 sizeof(__be64) : sizeof(__be32);
 609 }
 610
 611 /*
 612  * Calculate offset of the n-th record in a btree block.
 613  */
 614 STATIC size_t
 615 xfs_btree_rec_offset(
 616         struct xfs_btree_cur    *cur,
 617         int                     n)
 618 {
 619         return xfs_btree_block_len(cur) +
 620                 (n - 1) * cur->bc_ops->rec_len;
 621 }
 622
 623 /*
 624  * Calculate offset of the n-th key in a btree block.
 625  */
 626 STATIC size_t
 627 xfs_btree_key_offset(
 628         struct xfs_btree_cur    *cur,
 629         int                     n)
 630 {
 631         return xfs_btree_block_len(cur) +
 632                 (n - 1) * cur->bc_ops->key_len;
 633 }
 634
 635 /*
 636  * Calculate offset of the n-th high key in a btree block.
 637  */
 638 STATIC size_t
 639 xfs_btree_high_key_offset(
 640         struct xfs_btree_cur    *cur,
 641         int                     n)
 642 {
 643         return xfs_btree_block_len(cur) +
 644                 (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2);
 645 }
 646
 647 /*
 648  * Calculate offset of the n-th block pointer in a btree block.
 649  */
 650 STATIC size_t
 651 xfs_btree_ptr_offset(
 652         struct xfs_btree_cur    *cur,
 653         int                     n,
 654         int                     level)
 655 {
 656         return xfs_btree_block_len(cur) +
 657                 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
 658                 (n - 1) * xfs_btree_ptr_len(cur);
 659 }
 660
 661 /*
 662  * Return a pointer to the n-th record in the btree block.
 663  */
 664 union xfs_btree_rec *
 665 xfs_btree_rec_addr(
 666         struct xfs_btree_cur    *cur,
 667         int                     n,
 668         struct xfs_btree_block  *block)
 669 {
 670         return (union xfs_btree_rec *)
 671                 ((char *)block + xfs_btree_rec_offset(cur, n));
 672 }
 673
 674 /*
 675  * Return a pointer to the n-th key in the btree block.
 676  */
 677 union xfs_btree_key *
 678 xfs_btree_key_addr(
 679         struct xfs_btree_cur    *cur,
 680         int                     n,
 681         struct xfs_btree_block  *block)
 682 {
 683         return (union xfs_btree_key *)
 684                 ((char *)block + xfs_btree_key_offset(cur, n));
 685 }
 686
 687 /*
 688  * Return a pointer to the n-th high key in the btree block.
 689  */
 690 union xfs_btree_key *
 691 xfs_btree_high_key_addr(
 692         struct xfs_btree_cur    *cur,
 693         int                     n,
 694         struct xfs_btree_block  *block)
 695 {
 696         return (union xfs_btree_key *)
 697                 ((char *)block + xfs_btree_high_key_offset(cur, n));
 698 }
 699
 700 /*
 701  * Return a pointer to the n-th block pointer in the btree block.
 702  */
 703 union xfs_btree_ptr *
 704 xfs_btree_ptr_addr(
 705         struct xfs_btree_cur    *cur,
 706         int                     n,
 707         struct xfs_btree_block  *block)
 708 {
 709         int                     level = xfs_btree_get_level(block);
 710
 711         ASSERT(block->bb_level != 0);
 712
 713         return (union xfs_btree_ptr *)
 714                 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
 715 }
 716
 717 struct xfs_ifork *
 718 xfs_btree_ifork_ptr(
 719         struct xfs_btree_cur    *cur)
 720 {
 721         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
 722
 723         if (cur->bc_flags & XFS_BTREE_STAGING)
 724                 return cur->bc_ino.ifake->if_fork;
 725         return xfs_ifork_ptr(cur->bc_ino.ip, cur->bc_ino.whichfork);
 726 }
 727
 728 /*
 729  * Get the root block which is stored in the inode.
 730  *
 731  * For now this btree implementation assumes the btree root is always
 732  * stored in the if_broot field of an inode fork.
 733  */
 734 STATIC struct xfs_btree_block *
 735 xfs_btree_get_iroot(
 736         struct xfs_btree_cur    *cur)
 737 {
 738         struct xfs_ifork        *ifp = xfs_btree_ifork_ptr(cur);
 739
 740         return (struct xfs_btree_block *)ifp->if_broot;
 741 }
 742
 743 /*
 744  * Retrieve the block pointer from the cursor at the given level.
 745  * This may be an inode btree root or from a buffer.
 746  */
 747 struct xfs_btree_block *                /* generic btree block pointer */
 748 xfs_btree_get_block(
 749         struct xfs_btree_cur    *cur,   /* btree cursor */
 750         int                     level,  /* level in btree */
 751         struct xfs_buf          **bpp)  /* buffer containing the block */
 752 {
 753         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 754             (level == cur->bc_nlevels - 1)) {
 755                 *bpp = NULL;
 756                 return xfs_btree_get_iroot(cur);
 757         }
 758
 759         *bpp = cur->bc_levels[level].bp;
 760         return XFS_BUF_TO_BLOCK(*bpp);
 761 }
 762
 763 /*
 764  * Change the cursor to point to the first record at the given level.
 765  * Other levels are unaffected.
 766  */
 767 STATIC int                              /* success=1, failure=0 */
 768 xfs_btree_firstrec(
 769         struct xfs_btree_cur    *cur,   /* btree cursor */
 770         int                     level)  /* level to change */
 771 {
 772         struct xfs_btree_block  *block; /* generic btree block pointer */
 773         struct xfs_buf          *bp;    /* buffer containing block */
 774
 775         /*
 776          * Get the block pointer for this level.
 777          */
 778         block = xfs_btree_get_block(cur, level, &bp);
 779         if (xfs_btree_check_block(cur, block, level, bp))
 780                 return 0;
 781         /*
 782          * It's empty, there is no such record.
 783          */
 784         if (!block->bb_numrecs)
 785                 return 0;
 786         /*
 787          * Set the ptr value to 1, that's the first record/key.
 788          */
 789         cur->bc_levels[level].ptr = 1;
 790         return 1;
 791 }
 792
 793 /*
 794  * Change the cursor to point to the last record in the current block
 795  * at the given level.  Other levels are unaffected.
 796  */
 797 STATIC int                              /* success=1, failure=0 */
 798 xfs_btree_lastrec(
 799         struct xfs_btree_cur    *cur,   /* btree cursor */
 800         int                     level)  /* level to change */
 801 {
 802         struct xfs_btree_block  *block; /* generic btree block pointer */
 803         struct xfs_buf          *bp;    /* buffer containing block */
 804
 805         /*
 806          * Get the block pointer for this level.
 807          */
 808         block = xfs_btree_get_block(cur, level, &bp);
 809         if (xfs_btree_check_block(cur, block, level, bp))
 810                 return 0;
 811         /*
 812          * It's empty, there is no such record.
 813          */
 814         if (!block->bb_numrecs)
 815                 return 0;
 816         /*
 817          * Set the ptr value to numrecs, that's the last record/key.
 818          */
 819         cur->bc_levels[level].ptr = be16_to_cpu(block->bb_numrecs);
 820         return 1;
 821 }
 822
 823 /*
 824  * Compute first and last byte offsets for the fields given.
 825  * Interprets the offsets table, which contains struct field offsets.
 826  */
 827 void
 828 xfs_btree_offsets(
 829         uint32_t        fields,         /* bitmask of fields */
 830         const short     *offsets,       /* table of field offsets */
 831         int             nbits,          /* number of bits to inspect */
 832         int             *first,         /* output: first byte offset */
 833         int             *last)          /* output: last byte offset */
 834 {
 835         int             i;              /* current bit number */
 836         uint32_t        imask;          /* mask for current bit number */
 837
 838         ASSERT(fields != 0);
 839         /*
 840          * Find the lowest bit, so the first byte offset.
 841          */
 842         for (i = 0, imask = 1u; ; i++, imask <<= 1) {
 843                 if (imask & fields) {
 844                         *first = offsets[i];
 845                         break;
 846                 }
 847         }
 848         /*
 849          * Find the highest bit, so the last byte offset.
 850          */
 851         for (i = nbits - 1, imask = 1u << i; ; i--, imask >>= 1) {
 852                 if (imask & fields) {
 853                         *last = offsets[i + 1] - 1;
 854                         break;
 855                 }
 856         }
 857 }
 858
 859 /*
 860  * Get a buffer for the block, return it read in.
 861  * Long-form addressing.
 862  */
 863 int
 864 xfs_btree_read_bufl(
 865         struct xfs_mount        *mp,            /* file system mount point */
 866         struct xfs_trans        *tp,            /* transaction pointer */
 867         xfs_fsblock_t           fsbno,          /* file system block number */
 868         struct xfs_buf          **bpp,          /* buffer for fsbno */
 869         int                     refval,         /* ref count value for buffer */
 870         const struct xfs_buf_ops *ops)
 871 {
 872         struct xfs_buf          *bp;            /* return value */
 873         xfs_daddr_t             d;              /* real disk block address */
 874         int                     error;
 875
 876         if (!xfs_verify_fsbno(mp, fsbno))
 877                 return -EFSCORRUPTED;
 878         d = XFS_FSB_TO_DADDR(mp, fsbno);
 879         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
 880                                    mp->m_bsize, 0, &bp, ops);
 881         if (error)
 882                 return error;
 883         if (bp)
 884                 xfs_buf_set_ref(bp, refval);
 885         *bpp = bp;
 886         return 0;
 887 }
 888
 889 /*
 890  * Read-ahead the block, don't wait for it, don't return a buffer.
 891  * Long-form addressing.
 892  */
 893 /* ARGSUSED */
 894 void
 895 xfs_btree_reada_bufl(
 896         struct xfs_mount        *mp,            /* file system mount point */
 897         xfs_fsblock_t           fsbno,          /* file system block number */
 898         xfs_extlen_t            count,          /* count of filesystem blocks */
 899         const struct xfs_buf_ops *ops)
 900 {
 901         xfs_daddr_t             d;
 902
 903         ASSERT(fsbno != NULLFSBLOCK);
 904         d = XFS_FSB_TO_DADDR(mp, fsbno);
 905         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 906 }
 907
 908 /*
 909  * Read-ahead the block, don't wait for it, don't return a buffer.
 910  * Short-form addressing.
 911  */
 912 /* ARGSUSED */
 913 void
 914 xfs_btree_reada_bufs(
 915         struct xfs_mount        *mp,            /* file system mount point */
 916         xfs_agnumber_t          agno,           /* allocation group number */
 917         xfs_agblock_t           agbno,          /* allocation group block number */
 918         xfs_extlen_t            count,          /* count of filesystem blocks */
 919         const struct xfs_buf_ops *ops)
 920 {
 921         xfs_daddr_t             d;
 922
 923         ASSERT(agno != NULLAGNUMBER);
 924         ASSERT(agbno != NULLAGBLOCK);
 925         d = XFS_AGB_TO_DADDR(mp, agno, agbno);
 926         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 927 }
 928
 929 STATIC int
 930 xfs_btree_readahead_lblock(
 931         struct xfs_btree_cur    *cur,
 932         int                     lr,
 933         struct xfs_btree_block  *block)
 934 {
 935         int                     rval = 0;
 936         xfs_fsblock_t           left = be64_to_cpu(block->bb_u.l.bb_leftsib);
 937         xfs_fsblock_t           right = be64_to_cpu(block->bb_u.l.bb_rightsib);
 938
 939         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
 940                 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
 941                                      cur->bc_ops->buf_ops);
 942                 rval++;
 943         }
 944
 945         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
 946                 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
 947                                      cur->bc_ops->buf_ops);
 948                 rval++;
 949         }
 950
 951         return rval;
 952 }
 953
 954 STATIC int
 955 xfs_btree_readahead_sblock(
 956         struct xfs_btree_cur    *cur,
 957         int                     lr,
 958         struct xfs_btree_block *block)
 959 {
 960         int                     rval = 0;
 961         xfs_agblock_t           left = be32_to_cpu(block->bb_u.s.bb_leftsib);
 962         xfs_agblock_t           right = be32_to_cpu(block->bb_u.s.bb_rightsib);
 963
 964
 965         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
 966                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_ag.pag->pag_agno,
 967                                      left, 1, cur->bc_ops->buf_ops);
 968                 rval++;
 969         }
 970
 971         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
 972                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_ag.pag->pag_agno,
 973                                      right, 1, cur->bc_ops->buf_ops);
 974                 rval++;
 975         }
 976
 977         return rval;
 978 }
 979
 980 /*
 981  * Read-ahead btree blocks, at the given level.
 982  * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
 983  */
 984 STATIC int
 985 xfs_btree_readahead(
 986         struct xfs_btree_cur    *cur,           /* btree cursor */
 987         int                     lev,            /* level in btree */
 988         int                     lr)             /* left/right bits */
 989 {
 990         struct xfs_btree_block  *block;
 991
 992         /*
 993          * No readahead needed if we are at the root level and the
 994          * btree root is stored in the inode.
 995          */
 996         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 997             (lev == cur->bc_nlevels - 1))
 998                 return 0;
 999
1000         if ((cur->bc_levels[lev].ra | lr) == cur->bc_levels[lev].ra)
1001                 return 0;
1002
1003         cur->bc_levels[lev].ra |= lr;
1004         block = XFS_BUF_TO_BLOCK(cur->bc_levels[lev].bp);
1005
1006         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1007                 return xfs_btree_readahead_lblock(cur, lr, block);
1008         return xfs_btree_readahead_sblock(cur, lr, block);
1009 }
1010
1011 STATIC int
1012 xfs_btree_ptr_to_daddr(
1013         struct xfs_btree_cur            *cur,
1014         const union xfs_btree_ptr       *ptr,
1015         xfs_daddr_t                     *daddr)
1016 {
1017         xfs_fsblock_t           fsbno;
1018         xfs_agblock_t           agbno;
1019         int                     error;
1020
1021         error = xfs_btree_check_ptr(cur, ptr, 0, 1);
1022         if (error)
1023                 return error;
1024
1025         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1026                 fsbno = be64_to_cpu(ptr->l);
1027                 *daddr = XFS_FSB_TO_DADDR(cur->bc_mp, fsbno);
1028         } else {
1029                 agbno = be32_to_cpu(ptr->s);
1030                 *daddr = XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_ag.pag->pag_agno,
1031                                 agbno);
1032         }
1033
1034         return 0;
1035 }
1036
1037 /*
1038  * Readahead @count btree blocks at the given @ptr location.
1039  *
1040  * We don't need to care about long or short form btrees here as we have a
1041  * method of converting the ptr directly to a daddr available to us.
1042  */
1043 STATIC void
1044 xfs_btree_readahead_ptr(
1045         struct xfs_btree_cur    *cur,
1046         union xfs_btree_ptr     *ptr,
1047         xfs_extlen_t            count)
1048 {
1049         xfs_daddr_t             daddr;
1050
1051         if (xfs_btree_ptr_to_daddr(cur, ptr, &daddr))
1052                 return;
1053         xfs_buf_readahead(cur->bc_mp->m_ddev_targp, daddr,
1054                           cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
1055 }
1056
1057 /*
1058  * Set the buffer for level "lev" in the cursor to bp, releasing
1059  * any previous buffer.
1060  */
1061 STATIC void
1062 xfs_btree_setbuf(
1063         struct xfs_btree_cur    *cur,   /* btree cursor */
1064         int                     lev,    /* level in btree */
1065         struct xfs_buf          *bp)    /* new buffer to set */
1066 {
1067         struct xfs_btree_block  *b;     /* btree block */
1068
1069         if (cur->bc_levels[lev].bp)
1070                 xfs_trans_brelse(cur->bc_tp, cur->bc_levels[lev].bp);
1071         cur->bc_levels[lev].bp = bp;
1072         cur->bc_levels[lev].ra = 0;
1073
1074         b = XFS_BUF_TO_BLOCK(bp);
1075         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1076                 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
1077                         cur->bc_levels[lev].ra |= XFS_BTCUR_LEFTRA;
1078                 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1079                         cur->bc_levels[lev].ra |= XFS_BTCUR_RIGHTRA;
1080         } else {
1081                 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1082                         cur->bc_levels[lev].ra |= XFS_BTCUR_LEFTRA;
1083                 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1084                         cur->bc_levels[lev].ra |= XFS_BTCUR_RIGHTRA;
1085         }
1086 }
1087
1088 bool
1089 xfs_btree_ptr_is_null(
1090         struct xfs_btree_cur            *cur,
1091         const union xfs_btree_ptr       *ptr)
1092 {
1093         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1094                 return ptr->l == cpu_to_be64(NULLFSBLOCK);
1095         else
1096                 return ptr->s == cpu_to_be32(NULLAGBLOCK);
1097 }
1098
1099 void
1100 xfs_btree_set_ptr_null(
1101         struct xfs_btree_cur    *cur,
1102         union xfs_btree_ptr     *ptr)
1103 {
1104         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1105                 ptr->l = cpu_to_be64(NULLFSBLOCK);
1106         else
1107                 ptr->s = cpu_to_be32(NULLAGBLOCK);
1108 }
1109
1110 /*
1111  * Get/set/init sibling pointers
1112  */
1113 void
1114 xfs_btree_get_sibling(
1115         struct xfs_btree_cur    *cur,
1116         struct xfs_btree_block  *block,
1117         union xfs_btree_ptr     *ptr,
1118         int                     lr)
1119 {
1120         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1121
1122         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1123                 if (lr == XFS_BB_RIGHTSIB)
1124                         ptr->l = block->bb_u.l.bb_rightsib;
1125                 else
1126                         ptr->l = block->bb_u.l.bb_leftsib;
1127         } else {
1128                 if (lr == XFS_BB_RIGHTSIB)
1129                         ptr->s = block->bb_u.s.bb_rightsib;
1130                 else
1131                         ptr->s = block->bb_u.s.bb_leftsib;
1132         }
1133 }
1134
1135 void
1136 xfs_btree_set_sibling(
1137         struct xfs_btree_cur            *cur,
1138         struct xfs_btree_block          *block,
1139         const union xfs_btree_ptr       *ptr,
1140         int                             lr)
1141 {
1142         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1143
1144         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1145                 if (lr == XFS_BB_RIGHTSIB)
1146                         block->bb_u.l.bb_rightsib = ptr->l;
1147                 else
1148                         block->bb_u.l.bb_leftsib = ptr->l;
1149         } else {
1150                 if (lr == XFS_BB_RIGHTSIB)
1151                         block->bb_u.s.bb_rightsib = ptr->s;
1152                 else
1153                         block->bb_u.s.bb_leftsib = ptr->s;
1154         }
1155 }
1156
1157 void
1158 xfs_btree_init_block_int(
1159         struct xfs_mount        *mp,
1160         struct xfs_btree_block  *buf,
1161         xfs_daddr_t             blkno,
1162         xfs_btnum_t             btnum,
1163         __u16                   level,
1164         __u16                   numrecs,
1165         __u64                   owner,
1166         unsigned int            flags)
1167 {
1168         int                     crc = xfs_has_crc(mp);
1169         __u32                   magic = xfs_btree_magic(crc, btnum);
1170
1171         buf->bb_magic = cpu_to_be32(magic);
1172         buf->bb_level = cpu_to_be16(level);
1173         buf->bb_numrecs = cpu_to_be16(numrecs);
1174
1175         if (flags & XFS_BTREE_LONG_PTRS) {
1176                 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1177                 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1178                 if (crc) {
1179                         buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1180                         buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1181                         uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1182                         buf->bb_u.l.bb_pad = 0;
1183                         buf->bb_u.l.bb_lsn = 0;
1184                 }
1185         } else {
1186                 /* owner is a 32 bit value on short blocks */
1187                 __u32 __owner = (__u32)owner;
1188
1189                 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1190                 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1191                 if (crc) {
1192                         buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1193                         buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1194                         uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1195                         buf->bb_u.s.bb_lsn = 0;
1196                 }
1197         }
1198 }
1199
1200 void
1201 xfs_btree_init_block(
1202         struct xfs_mount *mp,
1203         struct xfs_buf  *bp,
1204         xfs_btnum_t     btnum,
1205         __u16           level,
1206         __u16           numrecs,
1207         __u64           owner)
1208 {
1209         xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), xfs_buf_daddr(bp),
1210                                  btnum, level, numrecs, owner, 0);
1211 }
1212
1213 void
1214 xfs_btree_init_block_cur(
1215         struct xfs_btree_cur    *cur,
1216         struct xfs_buf          *bp,
1217         int                     level,
1218         int                     numrecs)
1219 {
1220         __u64                   owner;
1221
1222         /*
1223          * we can pull the owner from the cursor right now as the different
1224          * owners align directly with the pointer size of the btree. This may
1225          * change in future, but is safe for current users of the generic btree
1226          * code.
1227          */
1228         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1229                 owner = cur->bc_ino.ip->i_ino;
1230         else
1231                 owner = cur->bc_ag.pag->pag_agno;
1232
1233         xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp),
1234                                 xfs_buf_daddr(bp), cur->bc_btnum, level,
1235                                 numrecs, owner, cur->bc_flags);
1236 }
1237
1238 /*
1239  * Return true if ptr is the last record in the btree and
1240  * we need to track updates to this record.  The decision
1241  * will be further refined in the update_lastrec method.
1242  */
1243 STATIC int
1244 xfs_btree_is_lastrec(
1245         struct xfs_btree_cur    *cur,
1246         struct xfs_btree_block  *block,
1247         int                     level)
1248 {
1249         union xfs_btree_ptr     ptr;
1250
1251         if (level > 0)
1252                 return 0;
1253         if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1254                 return 0;
1255
1256         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1257         if (!xfs_btree_ptr_is_null(cur, &ptr))
1258                 return 0;
1259         return 1;
1260 }
1261
1262 STATIC void
1263 xfs_btree_buf_to_ptr(
1264         struct xfs_btree_cur    *cur,
1265         struct xfs_buf          *bp,
1266         union xfs_btree_ptr     *ptr)
1267 {
1268         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1269                 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1270                                         xfs_buf_daddr(bp)));
1271         else {
1272                 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1273                                         xfs_buf_daddr(bp)));
1274         }
1275 }
1276
1277 STATIC void
1278 xfs_btree_set_refs(
1279         struct xfs_btree_cur    *cur,
1280         struct xfs_buf          *bp)
1281 {
1282         switch (cur->bc_btnum) {
1283         case XFS_BTNUM_BNO:
1284         case XFS_BTNUM_CNT:
1285                 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1286                 break;
1287         case XFS_BTNUM_INO:
1288         case XFS_BTNUM_FINO:
1289                 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1290                 break;
1291         case XFS_BTNUM_BMAP:
1292                 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1293                 break;
1294         case XFS_BTNUM_RMAP:
1295                 xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF);
1296                 break;
1297         case XFS_BTNUM_REFC:
1298                 xfs_buf_set_ref(bp, XFS_REFC_BTREE_REF);
1299                 break;
1300         default:
1301                 ASSERT(0);
1302         }
1303 }
1304
1305 int
1306 xfs_btree_get_buf_block(
1307         struct xfs_btree_cur            *cur,
1308         const union xfs_btree_ptr       *ptr,
1309         struct xfs_btree_block          **block,
1310         struct xfs_buf                  **bpp)
1311 {
1312         struct xfs_mount        *mp = cur->bc_mp;
1313         xfs_daddr_t             d;
1314         int                     error;
1315
1316         error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1317         if (error)
1318                 return error;
1319         error = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d, mp->m_bsize,
1320                         0, bpp);
1321         if (error)
1322                 return error;
1323
1324         (*bpp)->b_ops = cur->bc_ops->buf_ops;
1325         *block = XFS_BUF_TO_BLOCK(*bpp);
1326         return 0;
1327 }
1328
1329 /*
1330  * Read in the buffer at the given ptr and return the buffer and
1331  * the block pointer within the buffer.
1332  */
1333 int
1334 xfs_btree_read_buf_block(
1335         struct xfs_btree_cur            *cur,
1336         const union xfs_btree_ptr       *ptr,
1337         int                             flags,
1338         struct xfs_btree_block          **block,
1339         struct xfs_buf                  **bpp)
1340 {
1341         struct xfs_mount        *mp = cur->bc_mp;
1342         xfs_daddr_t             d;
1343         int                     error;
1344
1345         /* need to sort out how callers deal with failures first */
1346         ASSERT(!(flags & XBF_TRYLOCK));
1347
1348         error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1349         if (error)
1350                 return error;
1351         error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1352                                    mp->m_bsize, flags, bpp,
1353                                    cur->bc_ops->buf_ops);
1354         if (error)
1355                 return error;
1356
1357         xfs_btree_set_refs(cur, *bpp);
1358         *block = XFS_BUF_TO_BLOCK(*bpp);
1359         return 0;
1360 }
1361
1362 /*
1363  * Copy keys from one btree block to another.
1364  */
1365 void
1366 xfs_btree_copy_keys(
1367         struct xfs_btree_cur            *cur,
1368         union xfs_btree_key             *dst_key,
1369         const union xfs_btree_key       *src_key,
1370         int                             numkeys)
1371 {
1372         ASSERT(numkeys >= 0);
1373         memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1374 }
1375
1376 /*
1377  * Copy records from one btree block to another.
1378  */
1379 STATIC void
1380 xfs_btree_copy_recs(
1381         struct xfs_btree_cur    *cur,
1382         union xfs_btree_rec     *dst_rec,
1383         union xfs_btree_rec     *src_rec,
1384         int                     numrecs)
1385 {
1386         ASSERT(numrecs >= 0);
1387         memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1388 }
1389
1390 /*
1391  * Copy block pointers from one btree block to another.
1392  */
1393 void
1394 xfs_btree_copy_ptrs(
1395         struct xfs_btree_cur    *cur,
1396         union xfs_btree_ptr     *dst_ptr,
1397         const union xfs_btree_ptr *src_ptr,
1398         int                     numptrs)
1399 {
1400         ASSERT(numptrs >= 0);
1401         memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1402 }
1403
1404 /*
1405  * Shift keys one index left/right inside a single btree block.
1406  */
1407 STATIC void
1408 xfs_btree_shift_keys(
1409         struct xfs_btree_cur    *cur,
1410         union xfs_btree_key     *key,
1411         int                     dir,
1412         int                     numkeys)
1413 {
1414         char                    *dst_key;
1415
1416         ASSERT(numkeys >= 0);
1417         ASSERT(dir == 1 || dir == -1);
1418
1419         dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1420         memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1421 }
1422
1423 /*
1424  * Shift records one index left/right inside a single btree block.
1425  */
1426 STATIC void
1427 xfs_btree_shift_recs(
1428         struct xfs_btree_cur    *cur,
1429         union xfs_btree_rec     *rec,
1430         int                     dir,
1431         int                     numrecs)
1432 {
1433         char                    *dst_rec;
1434
1435         ASSERT(numrecs >= 0);
1436         ASSERT(dir == 1 || dir == -1);
1437
1438         dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1439         memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1440 }
1441
1442 /*
1443  * Shift block pointers one index left/right inside a single btree block.
1444  */
1445 STATIC void
1446 xfs_btree_shift_ptrs(
1447         struct xfs_btree_cur    *cur,
1448         union xfs_btree_ptr     *ptr,
1449         int                     dir,
1450         int                     numptrs)
1451 {
1452         char                    *dst_ptr;
1453
1454         ASSERT(numptrs >= 0);
1455         ASSERT(dir == 1 || dir == -1);
1456
1457         dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1458         memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1459 }
1460
1461 /*
1462  * Log key values from the btree block.
1463  */
1464 STATIC void
1465 xfs_btree_log_keys(
1466         struct xfs_btree_cur    *cur,
1467         struct xfs_buf          *bp,
1468         int                     first,
1469         int                     last)
1470 {
1471
1472         if (bp) {
1473                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1474                 xfs_trans_log_buf(cur->bc_tp, bp,
1475                                   xfs_btree_key_offset(cur, first),
1476                                   xfs_btree_key_offset(cur, last + 1) - 1);
1477         } else {
1478                 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1479                                 xfs_ilog_fbroot(cur->bc_ino.whichfork));
1480         }
1481 }
1482
1483 /*
1484  * Log record values from the btree block.
1485  */
1486 void
1487 xfs_btree_log_recs(
1488         struct xfs_btree_cur    *cur,
1489         struct xfs_buf          *bp,
1490         int                     first,
1491         int                     last)
1492 {
1493
1494         xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1495         xfs_trans_log_buf(cur->bc_tp, bp,
1496                           xfs_btree_rec_offset(cur, first),
1497                           xfs_btree_rec_offset(cur, last + 1) - 1);
1498
1499 }
1500
1501 /*
1502  * Log block pointer fields from a btree block (nonleaf).
1503  */
1504 STATIC void
1505 xfs_btree_log_ptrs(
1506         struct xfs_btree_cur    *cur,   /* btree cursor */
1507         struct xfs_buf          *bp,    /* buffer containing btree block */
1508         int                     first,  /* index of first pointer to log */
1509         int                     last)   /* index of last pointer to log */
1510 {
1511
1512         if (bp) {
1513                 struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
1514                 int                     level = xfs_btree_get_level(block);
1515
1516                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1517                 xfs_trans_log_buf(cur->bc_tp, bp,
1518                                 xfs_btree_ptr_offset(cur, first, level),
1519                                 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1520         } else {
1521                 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1522                         xfs_ilog_fbroot(cur->bc_ino.whichfork));
1523         }
1524
1525 }
1526
1527 /*
1528  * Log fields from a btree block header.
1529  */
1530 void
1531 xfs_btree_log_block(
1532         struct xfs_btree_cur    *cur,   /* btree cursor */
1533         struct xfs_buf          *bp,    /* buffer containing btree block */
1534         uint32_t                fields) /* mask of fields: XFS_BB_... */
1535 {
1536         int                     first;  /* first byte offset logged */
1537         int                     last;   /* last byte offset logged */
1538         static const short      soffsets[] = {  /* table of offsets (short) */
1539                 offsetof(struct xfs_btree_block, bb_magic),
1540                 offsetof(struct xfs_btree_block, bb_level),
1541                 offsetof(struct xfs_btree_block, bb_numrecs),
1542                 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1543                 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1544                 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1545                 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1546                 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1547                 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1548                 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1549                 XFS_BTREE_SBLOCK_CRC_LEN
1550         };
1551         static const short      loffsets[] = {  /* table of offsets (long) */
1552                 offsetof(struct xfs_btree_block, bb_magic),
1553                 offsetof(struct xfs_btree_block, bb_level),
1554                 offsetof(struct xfs_btree_block, bb_numrecs),
1555                 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1556                 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1557                 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1558                 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1559                 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1560                 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1561                 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1562                 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1563                 XFS_BTREE_LBLOCK_CRC_LEN
1564         };
1565
1566         if (bp) {
1567                 int nbits;
1568
1569                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1570                         /*
1571                          * We don't log the CRC when updating a btree
1572                          * block but instead recreate it during log
1573                          * recovery.  As the log buffers have checksums
1574                          * of their own this is safe and avoids logging a crc
1575                          * update in a lot of places.
1576                          */
1577                         if (fields == XFS_BB_ALL_BITS)
1578                                 fields = XFS_BB_ALL_BITS_CRC;
1579                         nbits = XFS_BB_NUM_BITS_CRC;
1580                 } else {
1581                         nbits = XFS_BB_NUM_BITS;
1582                 }
1583                 xfs_btree_offsets(fields,
1584                                   (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1585                                         loffsets : soffsets,
1586                                   nbits, &first, &last);
1587                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1588                 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1589         } else {
1590                 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1591                         xfs_ilog_fbroot(cur->bc_ino.whichfork));
1592         }
1593 }
1594
1595 /*
1596  * Increment cursor by one record at the level.
1597  * For nonzero levels the leaf-ward information is untouched.
1598  */
1599 int                                             /* error */
1600 xfs_btree_increment(
1601         struct xfs_btree_cur    *cur,
1602         int                     level,
1603         int                     *stat)          /* success/failure */
1604 {
1605         struct xfs_btree_block  *block;
1606         union xfs_btree_ptr     ptr;
1607         struct xfs_buf          *bp;
1608         int                     error;          /* error return value */
1609         int                     lev;
1610
1611         ASSERT(level < cur->bc_nlevels);
1612
1613         /* Read-ahead to the right at this level. */
1614         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1615
1616         /* Get a pointer to the btree block. */
1617         block = xfs_btree_get_block(cur, level, &bp);
1618
1619 #ifdef DEBUG
1620         error = xfs_btree_check_block(cur, block, level, bp);
1621         if (error)
1622                 goto error0;
1623 #endif
1624
1625         /* We're done if we remain in the block after the increment. */
1626         if (++cur->bc_levels[level].ptr <= xfs_btree_get_numrecs(block))
1627                 goto out1;
1628
1629         /* Fail if we just went off the right edge of the tree. */
1630         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1631         if (xfs_btree_ptr_is_null(cur, &ptr))
1632                 goto out0;
1633
1634         XFS_BTREE_STATS_INC(cur, increment);
1635
1636         /*
1637          * March up the tree incrementing pointers.
1638          * Stop when we don't go off the right edge of a block.
1639          */
1640         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1641                 block = xfs_btree_get_block(cur, lev, &bp);
1642
1643 #ifdef DEBUG
1644                 error = xfs_btree_check_block(cur, block, lev, bp);
1645                 if (error)
1646                         goto error0;
1647 #endif
1648
1649                 if (++cur->bc_levels[lev].ptr <= xfs_btree_get_numrecs(block))
1650                         break;
1651
1652                 /* Read-ahead the right block for the next loop. */
1653                 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1654         }
1655
1656         /*
1657          * If we went off the root then we are either seriously
1658          * confused or have the tree root in an inode.
1659          */
1660         if (lev == cur->bc_nlevels) {
1661                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1662                         goto out0;
1663                 ASSERT(0);
1664                 error = -EFSCORRUPTED;
1665                 goto error0;
1666         }
1667         ASSERT(lev < cur->bc_nlevels);
1668
1669         /*
1670          * Now walk back down the tree, fixing up the cursor's buffer
1671          * pointers and key numbers.
1672          */
1673         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1674                 union xfs_btree_ptr     *ptrp;
1675
1676                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_levels[lev].ptr, block);
1677                 --lev;
1678                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1679                 if (error)
1680                         goto error0;
1681
1682                 xfs_btree_setbuf(cur, lev, bp);
1683                 cur->bc_levels[lev].ptr = 1;
1684         }
1685 out1:
1686         *stat = 1;
1687         return 0;
1688
1689 out0:
1690         *stat = 0;
1691         return 0;
1692
1693 error0:
1694         return error;
1695 }
1696
1697 /*
1698  * Decrement cursor by one record at the level.
1699  * For nonzero levels the leaf-ward information is untouched.
1700  */
1701 int                                             /* error */
1702 xfs_btree_decrement(
1703         struct xfs_btree_cur    *cur,
1704         int                     level,
1705         int                     *stat)          /* success/failure */
1706 {
1707         struct xfs_btree_block  *block;
1708         struct xfs_buf          *bp;
1709         int                     error;          /* error return value */
1710         int                     lev;
1711         union xfs_btree_ptr     ptr;
1712
1713         ASSERT(level < cur->bc_nlevels);
1714
1715         /* Read-ahead to the left at this level. */
1716         xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1717
1718         /* We're done if we remain in the block after the decrement. */
1719         if (--cur->bc_levels[level].ptr > 0)
1720                 goto out1;
1721
1722         /* Get a pointer to the btree block. */
1723         block = xfs_btree_get_block(cur, level, &bp);
1724
1725 #ifdef DEBUG
1726         error = xfs_btree_check_block(cur, block, level, bp);
1727         if (error)
1728                 goto error0;
1729 #endif
1730
1731         /* Fail if we just went off the left edge of the tree. */
1732         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1733         if (xfs_btree_ptr_is_null(cur, &ptr))
1734                 goto out0;
1735
1736         XFS_BTREE_STATS_INC(cur, decrement);
1737
1738         /*
1739          * March up the tree decrementing pointers.
1740          * Stop when we don't go off the left edge of a block.
1741          */
1742         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1743                 if (--cur->bc_levels[lev].ptr > 0)
1744                         break;
1745                 /* Read-ahead the left block for the next loop. */
1746                 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1747         }
1748
1749         /*
1750          * If we went off the root then we are seriously confused.
1751          * or the root of the tree is in an inode.
1752          */
1753         if (lev == cur->bc_nlevels) {
1754                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1755                         goto out0;
1756                 ASSERT(0);
1757                 error = -EFSCORRUPTED;
1758                 goto error0;
1759         }
1760         ASSERT(lev < cur->bc_nlevels);
1761
1762         /*
1763          * Now walk back down the tree, fixing up the cursor's buffer
1764          * pointers and key numbers.
1765          */
1766         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1767                 union xfs_btree_ptr     *ptrp;
1768
1769                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_levels[lev].ptr, block);
1770                 --lev;
1771                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1772                 if (error)
1773                         goto error0;
1774                 xfs_btree_setbuf(cur, lev, bp);
1775                 cur->bc_levels[lev].ptr = xfs_btree_get_numrecs(block);
1776         }
1777 out1:
1778         *stat = 1;
1779         return 0;
1780
1781 out0:
1782         *stat = 0;
1783         return 0;
1784
1785 error0:
1786         return error;
1787 }
1788
1789 int
1790 xfs_btree_lookup_get_block(
1791         struct xfs_btree_cur            *cur,   /* btree cursor */
1792         int                             level,  /* level in the btree */
1793         const union xfs_btree_ptr       *pp,    /* ptr to btree block */
1794         struct xfs_btree_block          **blkp) /* return btree block */
1795 {
1796         struct xfs_buf          *bp;    /* buffer pointer for btree block */
1797         xfs_daddr_t             daddr;
1798         int                     error = 0;
1799
1800         /* special case the root block if in an inode */
1801         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1802             (level == cur->bc_nlevels - 1)) {
1803                 *blkp = xfs_btree_get_iroot(cur);
1804                 return 0;
1805         }
1806
1807         /*
1808          * If the old buffer at this level for the disk address we are
1809          * looking for re-use it.
1810          *
1811          * Otherwise throw it away and get a new one.
1812          */
1813         bp = cur->bc_levels[level].bp;
1814         error = xfs_btree_ptr_to_daddr(cur, pp, &daddr);
1815         if (error)
1816                 return error;
1817         if (bp && xfs_buf_daddr(bp) == daddr) {
1818                 *blkp = XFS_BUF_TO_BLOCK(bp);
1819                 return 0;
1820         }
1821
1822         error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1823         if (error)
1824                 return error;
1825
1826         /* Check the inode owner since the verifiers don't. */
1827         if (xfs_has_crc(cur->bc_mp) &&
1828             !(cur->bc_ino.flags & XFS_BTCUR_BMBT_INVALID_OWNER) &&
1829             (cur->bc_flags & XFS_BTREE_LONG_PTRS) &&
1830             be64_to_cpu((*blkp)->bb_u.l.bb_owner) !=
1831                         cur->bc_ino.ip->i_ino)
1832                 goto out_bad;
1833
1834         /* Did we get the level we were looking for? */
1835         if (be16_to_cpu((*blkp)->bb_level) != level)
1836                 goto out_bad;
1837
1838         /* Check that internal nodes have at least one record. */
1839         if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0)
1840                 goto out_bad;
1841
1842         xfs_btree_setbuf(cur, level, bp);
1843         return 0;
1844
1845 out_bad:
1846         *blkp = NULL;
1847         xfs_buf_mark_corrupt(bp);
1848         xfs_trans_brelse(cur->bc_tp, bp);
1849         return -EFSCORRUPTED;
1850 }
1851
1852 /*
1853  * Get current search key.  For level 0 we don't actually have a key
1854  * structure so we make one up from the record.  For all other levels
1855  * we just return the right key.
1856  */
1857 STATIC union xfs_btree_key *
1858 xfs_lookup_get_search_key(
1859         struct xfs_btree_cur    *cur,
1860         int                     level,
1861         int                     keyno,
1862         struct xfs_btree_block  *block,
1863         union xfs_btree_key     *kp)
1864 {
1865         if (level == 0) {
1866                 cur->bc_ops->init_key_from_rec(kp,
1867                                 xfs_btree_rec_addr(cur, keyno, block));
1868                 return kp;
1869         }
1870
1871         return xfs_btree_key_addr(cur, keyno, block);
1872 }
1873
1874 /*
1875  * Lookup the record.  The cursor is made to point to it, based on dir.
1876  * stat is set to 0 if can't find any such record, 1 for success.
1877  */
1878 int                                     /* error */
1879 xfs_btree_lookup(
1880         struct xfs_btree_cur    *cur,   /* btree cursor */
1881         xfs_lookup_t            dir,    /* <=, ==, or >= */
1882         int                     *stat)  /* success/failure */
1883 {
1884         struct xfs_btree_block  *block; /* current btree block */
1885         int64_t                 diff;   /* difference for the current key */
1886         int                     error;  /* error return value */
1887         int                     keyno;  /* current key number */
1888         int                     level;  /* level in the btree */
1889         union xfs_btree_ptr     *pp;    /* ptr to btree block */
1890         union xfs_btree_ptr     ptr;    /* ptr to btree block */
1891
1892         XFS_BTREE_STATS_INC(cur, lookup);
1893
1894         /* No such thing as a zero-level tree. */
1895         if (XFS_IS_CORRUPT(cur->bc_mp, cur->bc_nlevels == 0))
1896                 return -EFSCORRUPTED;
1897
1898         block = NULL;
1899         keyno = 0;
1900
1901         /* initialise start pointer from cursor */
1902         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1903         pp = &ptr;
1904
1905         /*
1906          * Iterate over each level in the btree, starting at the root.
1907          * For each level above the leaves, find the key we need, based
1908          * on the lookup record, then follow the corresponding block
1909          * pointer down to the next level.
1910          */
1911         for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1912                 /* Get the block we need to do the lookup on. */
1913                 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1914                 if (error)
1915                         goto error0;
1916
1917                 if (diff == 0) {
1918                         /*
1919                          * If we already had a key match at a higher level, we
1920                          * know we need to use the first entry in this block.
1921                          */
1922                         keyno = 1;
1923                 } else {
1924                         /* Otherwise search this block. Do a binary search. */
1925
1926                         int     high;   /* high entry number */
1927                         int     low;    /* low entry number */
1928
1929                         /* Set low and high entry numbers, 1-based. */
1930                         low = 1;
1931                         high = xfs_btree_get_numrecs(block);
1932                         if (!high) {
1933                                 /* Block is empty, must be an empty leaf. */
1934                                 if (level != 0 || cur->bc_nlevels != 1) {
1935                                         XFS_CORRUPTION_ERROR(__func__,
1936                                                         XFS_ERRLEVEL_LOW,
1937                                                         cur->bc_mp, block,
1938                                                         sizeof(*block));
1939                                         return -EFSCORRUPTED;
1940                                 }
1941
1942                                 cur->bc_levels[0].ptr = dir != XFS_LOOKUP_LE;
1943                                 *stat = 0;
1944                                 return 0;
1945                         }
1946
1947                         /* Binary search the block. */
1948                         while (low <= high) {
1949                                 union xfs_btree_key     key;
1950                                 union xfs_btree_key     *kp;
1951
1952                                 XFS_BTREE_STATS_INC(cur, compare);
1953
1954                                 /* keyno is average of low and high. */
1955                                 keyno = (low + high) >> 1;
1956
1957                                 /* Get current search key */
1958                                 kp = xfs_lookup_get_search_key(cur, level,
1959                                                 keyno, block, &key);
1960
1961                                 /*
1962                                  * Compute difference to get next direction:
1963                                  *  - less than, move right
1964                                  *  - greater than, move left
1965                                  *  - equal, we're done
1966                                  */
1967                                 diff = cur->bc_ops->key_diff(cur, kp);
1968                                 if (diff < 0)
1969                                         low = keyno + 1;
1970                                 else if (diff > 0)
1971                                         high = keyno - 1;
1972                                 else
1973                                         break;
1974                         }
1975                 }
1976
1977                 /*
1978                  * If there are more levels, set up for the next level
1979                  * by getting the block number and filling in the cursor.
1980                  */
1981                 if (level > 0) {
1982                         /*
1983                          * If we moved left, need the previous key number,
1984                          * unless there isn't one.
1985                          */
1986                         if (diff > 0 && --keyno < 1)
1987                                 keyno = 1;
1988                         pp = xfs_btree_ptr_addr(cur, keyno, block);
1989
1990                         error = xfs_btree_debug_check_ptr(cur, pp, 0, level);
1991                         if (error)
1992                                 goto error0;
1993
1994                         cur->bc_levels[level].ptr = keyno;
1995                 }
1996         }
1997
1998         /* Done with the search. See if we need to adjust the results. */
1999         if (dir != XFS_LOOKUP_LE && diff < 0) {
2000                 keyno++;
2001                 /*
2002                  * If ge search and we went off the end of the block, but it's
2003                  * not the last block, we're in the wrong block.
2004                  */
2005                 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
2006                 if (dir == XFS_LOOKUP_GE &&
2007                     keyno > xfs_btree_get_numrecs(block) &&
2008                     !xfs_btree_ptr_is_null(cur, &ptr)) {
2009                         int     i;
2010
2011                         cur->bc_levels[0].ptr = keyno;
2012                         error = xfs_btree_increment(cur, 0, &i);
2013                         if (error)
2014                                 goto error0;
2015                         if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
2016                                 return -EFSCORRUPTED;
2017                         *stat = 1;
2018                         return 0;
2019                 }
2020         } else if (dir == XFS_LOOKUP_LE && diff > 0)
2021                 keyno--;
2022         cur->bc_levels[0].ptr = keyno;
2023
2024         /* Return if we succeeded or not. */
2025         if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
2026                 *stat = 0;
2027         else if (dir != XFS_LOOKUP_EQ || diff == 0)
2028                 *stat = 1;
2029         else
2030                 *stat = 0;
2031         return 0;
2032
2033 error0:
2034         return error;
2035 }
2036
2037 /* Find the high key storage area from a regular key. */
2038 union xfs_btree_key *
2039 xfs_btree_high_key_from_key(
2040         struct xfs_btree_cur    *cur,
2041         union xfs_btree_key     *key)
2042 {
2043         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2044         return (union xfs_btree_key *)((char *)key +
2045                         (cur->bc_ops->key_len / 2));
2046 }
2047
2048 /* Determine the low (and high if overlapped) keys of a leaf block */
2049 STATIC void
2050 xfs_btree_get_leaf_keys(
2051         struct xfs_btree_cur    *cur,
2052         struct xfs_btree_block  *block,
2053         union xfs_btree_key     *key)
2054 {
2055         union xfs_btree_key     max_hkey;
2056         union xfs_btree_key     hkey;
2057         union xfs_btree_rec     *rec;
2058         union xfs_btree_key     *high;
2059         int                     n;
2060
2061         rec = xfs_btree_rec_addr(cur, 1, block);
2062         cur->bc_ops->init_key_from_rec(key, rec);
2063
2064         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2065
2066                 cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
2067                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2068                         rec = xfs_btree_rec_addr(cur, n, block);
2069                         cur->bc_ops->init_high_key_from_rec(&hkey, rec);
2070                         if (xfs_btree_keycmp_gt(cur, &hkey, &max_hkey))
2071                                 max_hkey = hkey;
2072                 }
2073
2074                 high = xfs_btree_high_key_from_key(cur, key);
2075                 memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
2076         }
2077 }
2078
2079 /* Determine the low (and high if overlapped) keys of a node block */
2080 STATIC void
2081 xfs_btree_get_node_keys(
2082         struct xfs_btree_cur    *cur,
2083         struct xfs_btree_block  *block,
2084         union xfs_btree_key     *key)
2085 {
2086         union xfs_btree_key     *hkey;
2087         union xfs_btree_key     *max_hkey;
2088         union xfs_btree_key     *high;
2089         int                     n;
2090
2091         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2092                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2093                                 cur->bc_ops->key_len / 2);
2094
2095                 max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2096                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2097                         hkey = xfs_btree_high_key_addr(cur, n, block);
2098                         if (xfs_btree_keycmp_gt(cur, hkey, max_hkey))
2099                                 max_hkey = hkey;
2100                 }
2101
2102                 high = xfs_btree_high_key_from_key(cur, key);
2103                 memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2104         } else {
2105                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2106                                 cur->bc_ops->key_len);
2107         }
2108 }
2109
2110 /* Derive the keys for any btree block. */
2111 void
2112 xfs_btree_get_keys(
2113         struct xfs_btree_cur    *cur,
2114         struct xfs_btree_block  *block,
2115         union xfs_btree_key     *key)
2116 {
2117         if (be16_to_cpu(block->bb_level) == 0)
2118                 xfs_btree_get_leaf_keys(cur, block, key);
2119         else
2120                 xfs_btree_get_node_keys(cur, block, key);
2121 }
2122
2123 /*
2124  * Decide if we need to update the parent keys of a btree block.  For
2125  * a standard btree this is only necessary if we're updating the first
2126  * record/key.  For an overlapping btree, we must always update the
2127  * keys because the highest key can be in any of the records or keys
2128  * in the block.
2129  */
2130 static inline bool
2131 xfs_btree_needs_key_update(
2132         struct xfs_btree_cur    *cur,
2133         int                     ptr)
2134 {
2135         return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1;
2136 }
2137
2138 /*
2139  * Update the low and high parent keys of the given level, progressing
2140  * towards the root.  If force_all is false, stop if the keys for a given
2141  * level do not need updating.
2142  */
2143 STATIC int
2144 __xfs_btree_updkeys(
2145         struct xfs_btree_cur    *cur,
2146         int                     level,
2147         struct xfs_btree_block  *block,
2148         struct xfs_buf          *bp0,
2149         bool                    force_all)
2150 {
2151         union xfs_btree_key     key;    /* keys from current level */
2152         union xfs_btree_key     *lkey;  /* keys from the next level up */
2153         union xfs_btree_key     *hkey;
2154         union xfs_btree_key     *nlkey; /* keys from the next level up */
2155         union xfs_btree_key     *nhkey;
2156         struct xfs_buf          *bp;
2157         int                     ptr;
2158
2159         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2160
2161         /* Exit if there aren't any parent levels to update. */
2162         if (level + 1 >= cur->bc_nlevels)
2163                 return 0;
2164
2165         trace_xfs_btree_updkeys(cur, level, bp0);
2166
2167         lkey = &key;
2168         hkey = xfs_btree_high_key_from_key(cur, lkey);
2169         xfs_btree_get_keys(cur, block, lkey);
2170         for (level++; level < cur->bc_nlevels; level++) {
2171 #ifdef DEBUG
2172                 int             error;
2173 #endif
2174                 block = xfs_btree_get_block(cur, level, &bp);
2175                 trace_xfs_btree_updkeys(cur, level, bp);
2176 #ifdef DEBUG
2177                 error = xfs_btree_check_block(cur, block, level, bp);
2178                 if (error)
2179                         return error;
2180 #endif
2181                 ptr = cur->bc_levels[level].ptr;
2182                 nlkey = xfs_btree_key_addr(cur, ptr, block);
2183                 nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2184                 if (!force_all &&
2185                     xfs_btree_keycmp_eq(cur, nlkey, lkey) &&
2186                     xfs_btree_keycmp_eq(cur, nhkey, hkey))
2187                         break;
2188                 xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2189                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2190                 if (level + 1 >= cur->bc_nlevels)
2191                         break;
2192                 xfs_btree_get_node_keys(cur, block, lkey);
2193         }
2194
2195         return 0;
2196 }
2197
2198 /* Update all the keys from some level in cursor back to the root. */
2199 STATIC int
2200 xfs_btree_updkeys_force(
2201         struct xfs_btree_cur    *cur,
2202         int                     level)
2203 {
2204         struct xfs_buf          *bp;
2205         struct xfs_btree_block  *block;
2206
2207         block = xfs_btree_get_block(cur, level, &bp);
2208         return __xfs_btree_updkeys(cur, level, block, bp, true);
2209 }
2210
2211 /*
2212  * Update the parent keys of the given level, progressing towards the root.
2213  */
2214 STATIC int
2215 xfs_btree_update_keys(
2216         struct xfs_btree_cur    *cur,
2217         int                     level)
2218 {
2219         struct xfs_btree_block  *block;
2220         struct xfs_buf          *bp;
2221         union xfs_btree_key     *kp;
2222         union xfs_btree_key     key;
2223         int                     ptr;
2224
2225         ASSERT(level >= 0);
2226
2227         block = xfs_btree_get_block(cur, level, &bp);
2228         if (cur->bc_flags & XFS_BTREE_OVERLAPPING)
2229                 return __xfs_btree_updkeys(cur, level, block, bp, false);
2230
2231         /*
2232          * Go up the tree from this level toward the root.
2233          * At each level, update the key value to the value input.
2234          * Stop when we reach a level where the cursor isn't pointing
2235          * at the first entry in the block.
2236          */
2237         xfs_btree_get_keys(cur, block, &key);
2238         for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2239 #ifdef DEBUG
2240                 int             error;
2241 #endif
2242                 block = xfs_btree_get_block(cur, level, &bp);
2243 #ifdef DEBUG
2244                 error = xfs_btree_check_block(cur, block, level, bp);
2245                 if (error)
2246                         return error;
2247 #endif
2248                 ptr = cur->bc_levels[level].ptr;
2249                 kp = xfs_btree_key_addr(cur, ptr, block);
2250                 xfs_btree_copy_keys(cur, kp, &key, 1);
2251                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2252         }
2253
2254         return 0;
2255 }
2256
2257 /*
2258  * Update the record referred to by cur to the value in the
2259  * given record. This either works (return 0) or gets an
2260  * EFSCORRUPTED error.
2261  */
2262 int
2263 xfs_btree_update(
2264         struct xfs_btree_cur    *cur,
2265         union xfs_btree_rec     *rec)
2266 {
2267         struct xfs_btree_block  *block;
2268         struct xfs_buf          *bp;
2269         int                     error;
2270         int                     ptr;
2271         union xfs_btree_rec     *rp;
2272
2273         /* Pick up the current block. */
2274         block = xfs_btree_get_block(cur, 0, &bp);
2275
2276 #ifdef DEBUG
2277         error = xfs_btree_check_block(cur, block, 0, bp);
2278         if (error)
2279                 goto error0;
2280 #endif
2281         /* Get the address of the rec to be updated. */
2282         ptr = cur->bc_levels[0].ptr;
2283         rp = xfs_btree_rec_addr(cur, ptr, block);
2284
2285         /* Fill in the new contents and log them. */
2286         xfs_btree_copy_recs(cur, rp, rec, 1);
2287         xfs_btree_log_recs(cur, bp, ptr, ptr);
2288
2289         /*
2290          * If we are tracking the last record in the tree and
2291          * we are at the far right edge of the tree, update it.
2292          */
2293         if (xfs_btree_is_lastrec(cur, block, 0)) {
2294                 cur->bc_ops->update_lastrec(cur, block, rec,
2295                                             ptr, LASTREC_UPDATE);
2296         }
2297
2298         /* Pass new key value up to our parent. */
2299         if (xfs_btree_needs_key_update(cur, ptr)) {
2300                 error = xfs_btree_update_keys(cur, 0);
2301                 if (error)
2302                         goto error0;
2303         }
2304
2305         return 0;
2306
2307 error0:
2308         return error;
2309 }
2310
2311 /*
2312  * Move 1 record left from cur/level if possible.
2313  * Update cur to reflect the new path.
2314  */
2315 STATIC int                                      /* error */
2316 xfs_btree_lshift(
2317         struct xfs_btree_cur    *cur,
2318         int                     level,
2319         int                     *stat)          /* success/failure */
2320 {
2321         struct xfs_buf          *lbp;           /* left buffer pointer */
2322         struct xfs_btree_block  *left;          /* left btree block */
2323         int                     lrecs;          /* left record count */
2324         struct xfs_buf          *rbp;           /* right buffer pointer */
2325         struct xfs_btree_block  *right;         /* right btree block */
2326         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2327         int                     rrecs;          /* right record count */
2328         union xfs_btree_ptr     lptr;           /* left btree pointer */
2329         union xfs_btree_key     *rkp = NULL;    /* right btree key */
2330         union xfs_btree_ptr     *rpp = NULL;    /* right address pointer */
2331         union xfs_btree_rec     *rrp = NULL;    /* right record pointer */
2332         int                     error;          /* error return value */
2333         int                     i;
2334
2335         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2336             level == cur->bc_nlevels - 1)
2337                 goto out0;
2338
2339         /* Set up variables for this block as "right". */
2340         right = xfs_btree_get_block(cur, level, &rbp);
2341
2342 #ifdef DEBUG
2343         error = xfs_btree_check_block(cur, right, level, rbp);
2344         if (error)
2345                 goto error0;
2346 #endif
2347
2348         /* If we've got no left sibling then we can't shift an entry left. */
2349         xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2350         if (xfs_btree_ptr_is_null(cur, &lptr))
2351                 goto out0;
2352
2353         /*
2354          * If the cursor entry is the one that would be moved, don't
2355          * do it... it's too complicated.
2356          */
2357         if (cur->bc_levels[level].ptr <= 1)
2358                 goto out0;
2359
2360         /* Set up the left neighbor as "left". */
2361         error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2362         if (error)
2363                 goto error0;
2364
2365         /* If it's full, it can't take another entry. */
2366         lrecs = xfs_btree_get_numrecs(left);
2367         if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2368                 goto out0;
2369
2370         rrecs = xfs_btree_get_numrecs(right);
2371
2372         /*
2373          * We add one entry to the left side and remove one for the right side.
2374          * Account for it here, the changes will be updated on disk and logged
2375          * later.
2376          */
2377         lrecs++;
2378         rrecs--;
2379
2380         XFS_BTREE_STATS_INC(cur, lshift);
2381         XFS_BTREE_STATS_ADD(cur, moves, 1);
2382
2383         /*
2384          * If non-leaf, copy a key and a ptr to the left block.
2385          * Log the changes to the left block.
2386          */
2387         if (level > 0) {
2388                 /* It's a non-leaf.  Move keys and pointers. */
2389                 union xfs_btree_key     *lkp;   /* left btree key */
2390                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2391
2392                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2393                 rkp = xfs_btree_key_addr(cur, 1, right);
2394
2395                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2396                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2397
2398                 error = xfs_btree_debug_check_ptr(cur, rpp, 0, level);
2399                 if (error)
2400                         goto error0;
2401
2402                 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2403                 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2404
2405                 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2406                 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2407
2408                 ASSERT(cur->bc_ops->keys_inorder(cur,
2409                         xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2410         } else {
2411                 /* It's a leaf.  Move records.  */
2412                 union xfs_btree_rec     *lrp;   /* left record pointer */
2413
2414                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2415                 rrp = xfs_btree_rec_addr(cur, 1, right);
2416
2417                 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2418                 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2419
2420                 ASSERT(cur->bc_ops->recs_inorder(cur,
2421                         xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2422         }
2423
2424         xfs_btree_set_numrecs(left, lrecs);
2425         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2426
2427         xfs_btree_set_numrecs(right, rrecs);
2428         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2429
2430         /*
2431          * Slide the contents of right down one entry.
2432          */
2433         XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2434         if (level > 0) {
2435                 /* It's a nonleaf. operate on keys and ptrs */
2436                 for (i = 0; i < rrecs; i++) {
2437                         error = xfs_btree_debug_check_ptr(cur, rpp, i + 1, level);
2438                         if (error)
2439                                 goto error0;
2440                 }
2441
2442                 xfs_btree_shift_keys(cur,
2443                                 xfs_btree_key_addr(cur, 2, right),
2444                                 -1, rrecs);
2445                 xfs_btree_shift_ptrs(cur,
2446                                 xfs_btree_ptr_addr(cur, 2, right),
2447                                 -1, rrecs);
2448
2449                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2450                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2451         } else {
2452                 /* It's a leaf. operate on records */
2453                 xfs_btree_shift_recs(cur,
2454                         xfs_btree_rec_addr(cur, 2, right),
2455                         -1, rrecs);
2456                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2457         }
2458
2459         /*
2460          * Using a temporary cursor, update the parent key values of the
2461          * block on the left.
2462          */
2463         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2464                 error = xfs_btree_dup_cursor(cur, &tcur);
2465                 if (error)
2466                         goto error0;
2467                 i = xfs_btree_firstrec(tcur, level);
2468                 if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2469                         error = -EFSCORRUPTED;
2470                         goto error0;
2471                 }
2472
2473                 error = xfs_btree_decrement(tcur, level, &i);
2474                 if (error)
2475                         goto error1;
2476
2477                 /* Update the parent high keys of the left block, if needed. */
2478                 error = xfs_btree_update_keys(tcur, level);
2479                 if (error)
2480                         goto error1;
2481
2482                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2483         }
2484
2485         /* Update the parent keys of the right block. */
2486         error = xfs_btree_update_keys(cur, level);
2487         if (error)
2488                 goto error0;
2489
2490         /* Slide the cursor value left one. */
2491         cur->bc_levels[level].ptr--;
2492
2493         *stat = 1;
2494         return 0;
2495
2496 out0:
2497         *stat = 0;
2498         return 0;
2499
2500 error0:
2501         return error;
2502
2503 error1:
2504         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2505         return error;
2506 }
2507
2508 /*
2509  * Move 1 record right from cur/level if possible.
2510  * Update cur to reflect the new path.
2511  */
2512 STATIC int                                      /* error */
2513 xfs_btree_rshift(
2514         struct xfs_btree_cur    *cur,
2515         int                     level,
2516         int                     *stat)          /* success/failure */
2517 {
2518         struct xfs_buf          *lbp;           /* left buffer pointer */
2519         struct xfs_btree_block  *left;          /* left btree block */
2520         struct xfs_buf          *rbp;           /* right buffer pointer */
2521         struct xfs_btree_block  *right;         /* right btree block */
2522         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2523         union xfs_btree_ptr     rptr;           /* right block pointer */
2524         union xfs_btree_key     *rkp;           /* right btree key */
2525         int                     rrecs;          /* right record count */
2526         int                     lrecs;          /* left record count */
2527         int                     error;          /* error return value */
2528         int                     i;              /* loop counter */
2529
2530         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2531             (level == cur->bc_nlevels - 1))
2532                 goto out0;
2533
2534         /* Set up variables for this block as "left". */
2535         left = xfs_btree_get_block(cur, level, &lbp);
2536
2537 #ifdef DEBUG
2538         error = xfs_btree_check_block(cur, left, level, lbp);
2539         if (error)
2540                 goto error0;
2541 #endif
2542
2543         /* If we've got no right sibling then we can't shift an entry right. */
2544         xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2545         if (xfs_btree_ptr_is_null(cur, &rptr))
2546                 goto out0;
2547
2548         /*
2549          * If the cursor entry is the one that would be moved, don't
2550          * do it... it's too complicated.
2551          */
2552         lrecs = xfs_btree_get_numrecs(left);
2553         if (cur->bc_levels[level].ptr >= lrecs)
2554                 goto out0;
2555
2556         /* Set up the right neighbor as "right". */
2557         error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2558         if (error)
2559                 goto error0;
2560
2561         /* If it's full, it can't take another entry. */
2562         rrecs = xfs_btree_get_numrecs(right);
2563         if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2564                 goto out0;
2565
2566         XFS_BTREE_STATS_INC(cur, rshift);
2567         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2568
2569         /*
2570          * Make a hole at the start of the right neighbor block, then
2571          * copy the last left block entry to the hole.
2572          */
2573         if (level > 0) {
2574                 /* It's a nonleaf. make a hole in the keys and ptrs */
2575                 union xfs_btree_key     *lkp;
2576                 union xfs_btree_ptr     *lpp;
2577                 union xfs_btree_ptr     *rpp;
2578
2579                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2580                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2581                 rkp = xfs_btree_key_addr(cur, 1, right);
2582                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2583
2584                 for (i = rrecs - 1; i >= 0; i--) {
2585                         error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
2586                         if (error)
2587                                 goto error0;
2588                 }
2589
2590                 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2591                 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2592
2593                 error = xfs_btree_debug_check_ptr(cur, lpp, 0, level);
2594                 if (error)
2595                         goto error0;
2596
2597                 /* Now put the new data in, and log it. */
2598                 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2599                 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2600
2601                 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2602                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2603
2604                 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2605                         xfs_btree_key_addr(cur, 2, right)));
2606         } else {
2607                 /* It's a leaf. make a hole in the records */
2608                 union xfs_btree_rec     *lrp;
2609                 union xfs_btree_rec     *rrp;
2610
2611                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2612                 rrp = xfs_btree_rec_addr(cur, 1, right);
2613
2614                 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2615
2616                 /* Now put the new data in, and log it. */
2617                 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2618                 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2619         }
2620
2621         /*
2622          * Decrement and log left's numrecs, bump and log right's numrecs.
2623          */
2624         xfs_btree_set_numrecs(left, --lrecs);
2625         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2626
2627         xfs_btree_set_numrecs(right, ++rrecs);
2628         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2629
2630         /*
2631          * Using a temporary cursor, update the parent key values of the
2632          * block on the right.
2633          */
2634         error = xfs_btree_dup_cursor(cur, &tcur);
2635         if (error)
2636                 goto error0;
2637         i = xfs_btree_lastrec(tcur, level);
2638         if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2639                 error = -EFSCORRUPTED;
2640                 goto error0;
2641         }
2642
2643         error = xfs_btree_increment(tcur, level, &i);
2644         if (error)
2645                 goto error1;
2646
2647         /* Update the parent high keys of the left block, if needed. */
2648         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2649                 error = xfs_btree_update_keys(cur, level);
2650                 if (error)
2651                         goto error1;
2652         }
2653
2654         /* Update the parent keys of the right block. */
2655         error = xfs_btree_update_keys(tcur, level);
2656         if (error)
2657                 goto error1;
2658
2659         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2660
2661         *stat = 1;
2662         return 0;
2663
2664 out0:
2665         *stat = 0;
2666         return 0;
2667
2668 error0:
2669         return error;
2670
2671 error1:
2672         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2673         return error;
2674 }
2675
2676 /*
2677  * Split cur/level block in half.
2678  * Return new block number and the key to its first
2679  * record (to be inserted into parent).
2680  */
2681 STATIC int                                      /* error */
2682 __xfs_btree_split(
2683         struct xfs_btree_cur    *cur,
2684         int                     level,
2685         union xfs_btree_ptr     *ptrp,
2686         union xfs_btree_key     *key,
2687         struct xfs_btree_cur    **curp,
2688         int                     *stat)          /* success/failure */
2689 {
2690         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
2691         struct xfs_buf          *lbp;           /* left buffer pointer */
2692         struct xfs_btree_block  *left;          /* left btree block */
2693         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
2694         struct xfs_buf          *rbp;           /* right buffer pointer */
2695         struct xfs_btree_block  *right;         /* right btree block */
2696         union xfs_btree_ptr     rrptr;          /* right-right sibling ptr */
2697         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
2698         struct xfs_btree_block  *rrblock;       /* right-right btree block */
2699         int                     lrecs;
2700         int                     rrecs;
2701         int                     src_index;
2702         int                     error;          /* error return value */
2703         int                     i;
2704
2705         XFS_BTREE_STATS_INC(cur, split);
2706
2707         /* Set up left block (current one). */
2708         left = xfs_btree_get_block(cur, level, &lbp);
2709
2710 #ifdef DEBUG
2711         error = xfs_btree_check_block(cur, left, level, lbp);
2712         if (error)
2713                 goto error0;
2714 #endif
2715
2716         xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2717
2718         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2719         error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2720         if (error)
2721                 goto error0;
2722         if (*stat == 0)
2723                 goto out0;
2724         XFS_BTREE_STATS_INC(cur, alloc);
2725
2726         /* Set up the new block as "right". */
2727         error = xfs_btree_get_buf_block(cur, &rptr, &right, &rbp);
2728         if (error)
2729                 goto error0;
2730
2731         /* Fill in the btree header for the new right block. */
2732         xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2733
2734         /*
2735          * Split the entries between the old and the new block evenly.
2736          * Make sure that if there's an odd number of entries now, that
2737          * each new block will have the same number of entries.
2738          */
2739         lrecs = xfs_btree_get_numrecs(left);
2740         rrecs = lrecs / 2;
2741         if ((lrecs & 1) && cur->bc_levels[level].ptr <= rrecs + 1)
2742                 rrecs++;
2743         src_index = (lrecs - rrecs + 1);
2744
2745         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2746
2747         /* Adjust numrecs for the later get_*_keys() calls. */
2748         lrecs -= rrecs;
2749         xfs_btree_set_numrecs(left, lrecs);
2750         xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2751
2752         /*
2753          * Copy btree block entries from the left block over to the
2754          * new block, the right. Update the right block and log the
2755          * changes.
2756          */
2757         if (level > 0) {
2758                 /* It's a non-leaf.  Move keys and pointers. */
2759                 union xfs_btree_key     *lkp;   /* left btree key */
2760                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2761                 union xfs_btree_key     *rkp;   /* right btree key */
2762                 union xfs_btree_ptr     *rpp;   /* right address pointer */
2763
2764                 lkp = xfs_btree_key_addr(cur, src_index, left);
2765                 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2766                 rkp = xfs_btree_key_addr(cur, 1, right);
2767                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2768
2769                 for (i = src_index; i < rrecs; i++) {
2770                         error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
2771                         if (error)
2772                                 goto error0;
2773                 }
2774
2775                 /* Copy the keys & pointers to the new block. */
2776                 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2777                 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2778
2779                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2780                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2781
2782                 /* Stash the keys of the new block for later insertion. */
2783                 xfs_btree_get_node_keys(cur, right, key);
2784         } else {
2785                 /* It's a leaf.  Move records.  */
2786                 union xfs_btree_rec     *lrp;   /* left record pointer */
2787                 union xfs_btree_rec     *rrp;   /* right record pointer */
2788
2789                 lrp = xfs_btree_rec_addr(cur, src_index, left);
2790                 rrp = xfs_btree_rec_addr(cur, 1, right);
2791
2792                 /* Copy records to the new block. */
2793                 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2794                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2795
2796                 /* Stash the keys of the new block for later insertion. */
2797                 xfs_btree_get_leaf_keys(cur, right, key);
2798         }
2799
2800         /*
2801          * Find the left block number by looking in the buffer.
2802          * Adjust sibling pointers.
2803          */
2804         xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2805         xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2806         xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2807         xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2808
2809         xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2810         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2811
2812         /*
2813          * If there's a block to the new block's right, make that block
2814          * point back to right instead of to left.
2815          */
2816         if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2817                 error = xfs_btree_read_buf_block(cur, &rrptr,
2818                                                         0, &rrblock, &rrbp);
2819                 if (error)
2820                         goto error0;
2821                 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2822                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2823         }
2824
2825         /* Update the parent high keys of the left block, if needed. */
2826         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2827                 error = xfs_btree_update_keys(cur, level);
2828                 if (error)
2829                         goto error0;
2830         }
2831
2832         /*
2833          * If the cursor is really in the right block, move it there.
2834          * If it's just pointing past the last entry in left, then we'll
2835          * insert there, so don't change anything in that case.
2836          */
2837         if (cur->bc_levels[level].ptr > lrecs + 1) {
2838                 xfs_btree_setbuf(cur, level, rbp);
2839                 cur->bc_levels[level].ptr -= lrecs;
2840         }
2841         /*
2842          * If there are more levels, we'll need another cursor which refers
2843          * the right block, no matter where this cursor was.
2844          */
2845         if (level + 1 < cur->bc_nlevels) {
2846                 error = xfs_btree_dup_cursor(cur, curp);
2847                 if (error)
2848                         goto error0;
2849                 (*curp)->bc_levels[level + 1].ptr++;
2850         }
2851         *ptrp = rptr;
2852         *stat = 1;
2853         return 0;
2854 out0:
2855         *stat = 0;
2856         return 0;
2857
2858 error0:
2859         return error;
2860 }
2861
2862 #ifdef __KERNEL__
2863 struct xfs_btree_split_args {
2864         struct xfs_btree_cur    *cur;
2865         int                     level;
2866         union xfs_btree_ptr     *ptrp;
2867         union xfs_btree_key     *key;
2868         struct xfs_btree_cur    **curp;
2869         int                     *stat;          /* success/failure */
2870         int                     result;
2871         bool                    kswapd; /* allocation in kswapd context */
2872         struct completion       *done;
2873         struct work_struct      work;
2874 };
2875
2876 /*
2877  * Stack switching interfaces for allocation
2878  */
2879 static void
2880 xfs_btree_split_worker(
2881         struct work_struct      *work)
2882 {
2883         struct xfs_btree_split_args     *args = container_of(work,
2884                                                 struct xfs_btree_split_args, work);
2885         unsigned long           pflags;
2886         unsigned long           new_pflags = 0;
2887
2888         /*
2889          * we are in a transaction context here, but may also be doing work
2890          * in kswapd context, and hence we may need to inherit that state
2891          * temporarily to ensure that we don't block waiting for memory reclaim
2892          * in any way.
2893          */
2894         if (args->kswapd)
2895                 new_pflags |= PF_MEMALLOC | PF_KSWAPD;
2896
2897         current_set_flags_nested(&pflags, new_pflags);
2898         xfs_trans_set_context(args->cur->bc_tp);
2899
2900         args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2901                                          args->key, args->curp, args->stat);
2902
2903         xfs_trans_clear_context(args->cur->bc_tp);
2904         current_restore_flags_nested(&pflags, new_pflags);
2905
2906         /*
2907          * Do not access args after complete() has run here. We don't own args
2908          * and the owner may run and free args before we return here.
2909          */
2910         complete(args->done);
2911
2912 }
2913
2914 /*
2915  * BMBT split requests often come in with little stack to work on so we push
2916  * them off to a worker thread so there is lots of stack to use. For the other
2917  * btree types, just call directly to avoid the context switch overhead here.
2918  *
2919  * Care must be taken here - the work queue rescuer thread introduces potential
2920  * AGF <> worker queue deadlocks if the BMBT block allocation has to lock new
2921  * AGFs to allocate blocks. A task being run by the rescuer could attempt to
2922  * lock an AGF that is already locked by a task queued to run by the rescuer,
2923  * resulting in an ABBA deadlock as the rescuer cannot run the lock holder to
2924  * release it until the current thread it is running gains the lock.
2925  *
2926  * To avoid this issue, we only ever queue BMBT splits that don't have an AGF
2927  * already locked to allocate from. The only place that doesn't hold an AGF
2928  * locked is unwritten extent conversion at IO completion, but that has already
2929  * been offloaded to a worker thread and hence has no stack consumption issues
2930  * we have to worry about.
2931  */
2932 STATIC int                                      /* error */
2933 xfs_btree_split(
2934         struct xfs_btree_cur    *cur,
2935         int                     level,
2936         union xfs_btree_ptr     *ptrp,
2937         union xfs_btree_key     *key,
2938         struct xfs_btree_cur    **curp,
2939         int                     *stat)          /* success/failure */
2940 {
2941         struct xfs_btree_split_args     args;
2942         DECLARE_COMPLETION_ONSTACK(done);
2943
2944         if (cur->bc_btnum != XFS_BTNUM_BMAP ||
2945             cur->bc_tp->t_highest_agno == NULLAGNUMBER)
2946                 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2947
2948         args.cur = cur;
2949         args.level = level;
2950         args.ptrp = ptrp;
2951         args.key = key;
2952         args.curp = curp;
2953         args.stat = stat;
2954         args.done = &done;
2955         args.kswapd = current_is_kswapd();
2956         INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2957         queue_work(xfs_alloc_wq, &args.work);
2958         wait_for_completion(&done);
2959         destroy_work_on_stack(&args.work);
2960         return args.result;
2961 }
2962 #else
2963 #define xfs_btree_split __xfs_btree_split
2964 #endif /* __KERNEL__ */
2965
2966
2967 /*
2968  * Copy the old inode root contents into a real block and make the
2969  * broot point to it.
2970  */
2971 int                                             /* error */
2972 xfs_btree_new_iroot(
2973         struct xfs_btree_cur    *cur,           /* btree cursor */
2974         int                     *logflags,      /* logging flags for inode */
2975         int                     *stat)          /* return status - 0 fail */
2976 {
2977         struct xfs_buf          *cbp;           /* buffer for cblock */
2978         struct xfs_btree_block  *block;         /* btree block */
2979         struct xfs_btree_block  *cblock;        /* child btree block */
2980         union xfs_btree_key     *ckp;           /* child key pointer */
2981         union xfs_btree_ptr     *cpp;           /* child ptr pointer */
2982         union xfs_btree_key     *kp;            /* pointer to btree key */
2983         union xfs_btree_ptr     *pp;            /* pointer to block addr */
2984         union xfs_btree_ptr     nptr;           /* new block addr */
2985         int                     level;          /* btree level */
2986         int                     error;          /* error return code */
2987         int                     i;              /* loop counter */
2988
2989         XFS_BTREE_STATS_INC(cur, newroot);
2990
2991         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2992
2993         level = cur->bc_nlevels - 1;
2994
2995         block = xfs_btree_get_iroot(cur);
2996         pp = xfs_btree_ptr_addr(cur, 1, block);
2997
2998         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2999         error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
3000         if (error)
3001                 goto error0;
3002         if (*stat == 0)
3003                 return 0;
3004
3005         XFS_BTREE_STATS_INC(cur, alloc);
3006
3007         /* Copy the root into a real block. */
3008         error = xfs_btree_get_buf_block(cur, &nptr, &cblock, &cbp);
3009         if (error)
3010                 goto error0;
3011
3012         /*
3013          * we can't just memcpy() the root in for CRC enabled btree blocks.
3014          * In that case have to also ensure the blkno remains correct
3015          */
3016         memcpy(cblock, block, xfs_btree_block_len(cur));
3017         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
3018                 __be64 bno = cpu_to_be64(xfs_buf_daddr(cbp));
3019                 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
3020                         cblock->bb_u.l.bb_blkno = bno;
3021                 else
3022                         cblock->bb_u.s.bb_blkno = bno;
3023         }
3024
3025         be16_add_cpu(&block->bb_level, 1);
3026         xfs_btree_set_numrecs(block, 1);
3027         cur->bc_nlevels++;
3028         ASSERT(cur->bc_nlevels <= cur->bc_maxlevels);
3029         cur->bc_levels[level + 1].ptr = 1;
3030
3031         kp = xfs_btree_key_addr(cur, 1, block);
3032         ckp = xfs_btree_key_addr(cur, 1, cblock);
3033         xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
3034
3035         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3036         for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
3037                 error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3038                 if (error)
3039                         goto error0;
3040         }
3041
3042         xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
3043
3044         error = xfs_btree_debug_check_ptr(cur, &nptr, 0, level);
3045         if (error)
3046                 goto error0;
3047
3048         xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
3049
3050         xfs_iroot_realloc(cur->bc_ino.ip,
3051                           1 - xfs_btree_get_numrecs(cblock),
3052                           cur->bc_ino.whichfork);
3053
3054         xfs_btree_setbuf(cur, level, cbp);
3055
3056         /*
3057          * Do all this logging at the end so that
3058          * the root is at the right level.
3059          */
3060         xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
3061         xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3062         xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3063
3064         *logflags |=
3065                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork);
3066         *stat = 1;
3067         return 0;
3068 error0:
3069         return error;
3070 }
3071
3072 /*
3073  * Allocate a new root block, fill it in.
3074  */
3075 STATIC int                              /* error */
3076 xfs_btree_new_root(
3077         struct xfs_btree_cur    *cur,   /* btree cursor */
3078         int                     *stat)  /* success/failure */
3079 {
3080         struct xfs_btree_block  *block; /* one half of the old root block */
3081         struct xfs_buf          *bp;    /* buffer containing block */
3082         int                     error;  /* error return value */
3083         struct xfs_buf          *lbp;   /* left buffer pointer */
3084         struct xfs_btree_block  *left;  /* left btree block */
3085         struct xfs_buf          *nbp;   /* new (root) buffer */
3086         struct xfs_btree_block  *new;   /* new (root) btree block */
3087         int                     nptr;   /* new value for key index, 1 or 2 */
3088         struct xfs_buf          *rbp;   /* right buffer pointer */
3089         struct xfs_btree_block  *right; /* right btree block */
3090         union xfs_btree_ptr     rptr;
3091         union xfs_btree_ptr     lptr;
3092
3093         XFS_BTREE_STATS_INC(cur, newroot);
3094
3095         /* initialise our start point from the cursor */
3096         cur->bc_ops->init_ptr_from_cur(cur, &rptr);
3097
3098         /* Allocate the new block. If we can't do it, we're toast. Give up. */
3099         error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
3100         if (error)
3101                 goto error0;
3102         if (*stat == 0)
3103                 goto out0;
3104         XFS_BTREE_STATS_INC(cur, alloc);
3105
3106         /* Set up the new block. */
3107         error = xfs_btree_get_buf_block(cur, &lptr, &new, &nbp);
3108         if (error)
3109                 goto error0;
3110
3111         /* Set the root in the holding structure  increasing the level by 1. */
3112         cur->bc_ops->set_root(cur, &lptr, 1);
3113
3114         /*
3115          * At the previous root level there are now two blocks: the old root,
3116          * and the new block generated when it was split.  We don't know which
3117          * one the cursor is pointing at, so we set up variables "left" and
3118          * "right" for each case.
3119          */
3120         block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3121
3122 #ifdef DEBUG
3123         error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3124         if (error)
3125                 goto error0;
3126 #endif
3127
3128         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3129         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3130                 /* Our block is left, pick up the right block. */
3131                 lbp = bp;
3132                 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3133                 left = block;
3134                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3135                 if (error)
3136                         goto error0;
3137                 bp = rbp;
3138                 nptr = 1;
3139         } else {
3140                 /* Our block is right, pick up the left block. */
3141                 rbp = bp;
3142                 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3143                 right = block;
3144                 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3145                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3146                 if (error)
3147                         goto error0;
3148                 bp = lbp;
3149                 nptr = 2;
3150         }
3151
3152         /* Fill in the new block's btree header and log it. */
3153         xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3154         xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3155         ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3156                         !xfs_btree_ptr_is_null(cur, &rptr));
3157
3158         /* Fill in the key data in the new root. */
3159         if (xfs_btree_get_level(left) > 0) {
3160                 /*
3161                  * Get the keys for the left block's keys and put them directly
3162                  * in the parent block.  Do the same for the right block.
3163                  */
3164                 xfs_btree_get_node_keys(cur, left,
3165                                 xfs_btree_key_addr(cur, 1, new));
3166                 xfs_btree_get_node_keys(cur, right,
3167                                 xfs_btree_key_addr(cur, 2, new));
3168         } else {
3169                 /*
3170                  * Get the keys for the left block's records and put them
3171                  * directly in the parent block.  Do the same for the right
3172                  * block.
3173                  */
3174                 xfs_btree_get_leaf_keys(cur, left,
3175                         xfs_btree_key_addr(cur, 1, new));
3176                 xfs_btree_get_leaf_keys(cur, right,
3177                         xfs_btree_key_addr(cur, 2, new));
3178         }
3179         xfs_btree_log_keys(cur, nbp, 1, 2);
3180
3181         /* Fill in the pointer data in the new root. */
3182         xfs_btree_copy_ptrs(cur,
3183                 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3184         xfs_btree_copy_ptrs(cur,
3185                 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3186         xfs_btree_log_ptrs(cur, nbp, 1, 2);
3187
3188         /* Fix up the cursor. */
3189         xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3190         cur->bc_levels[cur->bc_nlevels].ptr = nptr;
3191         cur->bc_nlevels++;
3192         ASSERT(cur->bc_nlevels <= cur->bc_maxlevels);
3193         *stat = 1;
3194         return 0;
3195 error0:
3196         return error;
3197 out0:
3198         *stat = 0;
3199         return 0;
3200 }
3201
3202 STATIC int
3203 xfs_btree_make_block_unfull(
3204         struct xfs_btree_cur    *cur,   /* btree cursor */
3205         int                     level,  /* btree level */
3206         int                     numrecs,/* # of recs in block */
3207         int                     *oindex,/* old tree index */
3208         int                     *index, /* new tree index */
3209         union xfs_btree_ptr     *nptr,  /* new btree ptr */
3210         struct xfs_btree_cur    **ncur, /* new btree cursor */
3211         union xfs_btree_key     *key,   /* key of new block */
3212         int                     *stat)
3213 {
3214         int                     error = 0;
3215
3216         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3217             level == cur->bc_nlevels - 1) {
3218                 struct xfs_inode *ip = cur->bc_ino.ip;
3219
3220                 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3221                         /* A root block that can be made bigger. */
3222                         xfs_iroot_realloc(ip, 1, cur->bc_ino.whichfork);
3223                         *stat = 1;
3224                 } else {
3225                         /* A root block that needs replacing */
3226                         int     logflags = 0;
3227
3228                         error = xfs_btree_new_iroot(cur, &logflags, stat);
3229                         if (error || *stat == 0)
3230                                 return error;
3231
3232                         xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3233                 }
3234
3235                 return 0;
3236         }
3237
3238         /* First, try shifting an entry to the right neighbor. */
3239         error = xfs_btree_rshift(cur, level, stat);
3240         if (error || *stat)
3241                 return error;
3242
3243         /* Next, try shifting an entry to the left neighbor. */
3244         error = xfs_btree_lshift(cur, level, stat);
3245         if (error)
3246                 return error;
3247
3248         if (*stat) {
3249                 *oindex = *index = cur->bc_levels[level].ptr;
3250                 return 0;
3251         }
3252
3253         /*
3254          * Next, try splitting the current block in half.
3255          *
3256          * If this works we have to re-set our variables because we
3257          * could be in a different block now.
3258          */
3259         error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3260         if (error || *stat == 0)
3261                 return error;
3262
3263
3264         *index = cur->bc_levels[level].ptr;
3265         return 0;
3266 }
3267
3268 /*
3269  * Insert one record/level.  Return information to the caller
3270  * allowing the next level up to proceed if necessary.
3271  */
3272 STATIC int
3273 xfs_btree_insrec(
3274         struct xfs_btree_cur    *cur,   /* btree cursor */
3275         int                     level,  /* level to insert record at */
3276         union xfs_btree_ptr     *ptrp,  /* i/o: block number inserted */
3277         union xfs_btree_rec     *rec,   /* record to insert */
3278         union xfs_btree_key     *key,   /* i/o: block key for ptrp */
3279         struct xfs_btree_cur    **curp, /* output: new cursor replacing cur */
3280         int                     *stat)  /* success/failure */
3281 {
3282         struct xfs_btree_block  *block; /* btree block */
3283         struct xfs_buf          *bp;    /* buffer for block */
3284         union xfs_btree_ptr     nptr;   /* new block ptr */
3285         struct xfs_btree_cur    *ncur = NULL;   /* new btree cursor */
3286         union xfs_btree_key     nkey;   /* new block key */
3287         union xfs_btree_key     *lkey;
3288         int                     optr;   /* old key/record index */
3289         int                     ptr;    /* key/record index */
3290         int                     numrecs;/* number of records */
3291         int                     error;  /* error return value */
3292         int                     i;
3293         xfs_daddr_t             old_bn;
3294
3295         ncur = NULL;
3296         lkey = &nkey;
3297
3298         /*
3299          * If we have an external root pointer, and we've made it to the
3300          * root level, allocate a new root block and we're done.
3301          */
3302         if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3303             (level >= cur->bc_nlevels)) {
3304                 error = xfs_btree_new_root(cur, stat);
3305                 xfs_btree_set_ptr_null(cur, ptrp);
3306
3307                 return error;
3308         }
3309
3310         /* If we're off the left edge, return failure. */
3311         ptr = cur->bc_levels[level].ptr;
3312         if (ptr == 0) {
3313                 *stat = 0;
3314                 return 0;
3315         }
3316
3317         optr = ptr;
3318
3319         XFS_BTREE_STATS_INC(cur, insrec);
3320
3321         /* Get pointers to the btree buffer and block. */
3322         block = xfs_btree_get_block(cur, level, &bp);
3323         old_bn = bp ? xfs_buf_daddr(bp) : XFS_BUF_DADDR_NULL;
3324         numrecs = xfs_btree_get_numrecs(block);
3325
3326 #ifdef DEBUG
3327         error = xfs_btree_check_block(cur, block, level, bp);
3328         if (error)
3329                 goto error0;
3330
3331         /* Check that the new entry is being inserted in the right place. */
3332         if (ptr <= numrecs) {
3333                 if (level == 0) {
3334                         ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3335                                 xfs_btree_rec_addr(cur, ptr, block)));
3336                 } else {
3337                         ASSERT(cur->bc_ops->keys_inorder(cur, key,
3338                                 xfs_btree_key_addr(cur, ptr, block)));
3339                 }
3340         }
3341 #endif
3342
3343         /*
3344          * If the block is full, we can't insert the new entry until we
3345          * make the block un-full.
3346          */
3347         xfs_btree_set_ptr_null(cur, &nptr);
3348         if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3349                 error = xfs_btree_make_block_unfull(cur, level, numrecs,
3350                                         &optr, &ptr, &nptr, &ncur, lkey, stat);
3351                 if (error || *stat == 0)
3352                         goto error0;
3353         }
3354
3355         /*
3356          * The current block may have changed if the block was
3357          * previously full and we have just made space in it.
3358          */
3359         block = xfs_btree_get_block(cur, level, &bp);
3360         numrecs = xfs_btree_get_numrecs(block);
3361
3362 #ifdef DEBUG
3363         error = xfs_btree_check_block(cur, block, level, bp);
3364         if (error)
3365                 goto error0;
3366 #endif
3367
3368         /*
3369          * At this point we know there's room for our new entry in the block
3370          * we're pointing at.
3371          */
3372         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3373
3374         if (level > 0) {
3375                 /* It's a nonleaf. make a hole in the keys and ptrs */
3376                 union xfs_btree_key     *kp;
3377                 union xfs_btree_ptr     *pp;
3378
3379                 kp = xfs_btree_key_addr(cur, ptr, block);
3380                 pp = xfs_btree_ptr_addr(cur, ptr, block);
3381
3382                 for (i = numrecs - ptr; i >= 0; i--) {
3383                         error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3384                         if (error)
3385                                 goto error0;
3386                 }
3387
3388                 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3389                 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3390
3391                 error = xfs_btree_debug_check_ptr(cur, ptrp, 0, level);
3392                 if (error)
3393                         goto error0;
3394
3395                 /* Now put the new data in, bump numrecs and log it. */
3396                 xfs_btree_copy_keys(cur, kp, key, 1);
3397                 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3398                 numrecs++;
3399                 xfs_btree_set_numrecs(block, numrecs);
3400                 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3401                 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3402 #ifdef DEBUG
3403                 if (ptr < numrecs) {
3404                         ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3405                                 xfs_btree_key_addr(cur, ptr + 1, block)));
3406                 }
3407 #endif
3408         } else {
3409                 /* It's a leaf. make a hole in the records */
3410                 union xfs_btree_rec             *rp;
3411
3412                 rp = xfs_btree_rec_addr(cur, ptr, block);
3413
3414                 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3415
3416                 /* Now put the new data in, bump numrecs and log it. */
3417                 xfs_btree_copy_recs(cur, rp, rec, 1);
3418                 xfs_btree_set_numrecs(block, ++numrecs);
3419                 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3420 #ifdef DEBUG
3421                 if (ptr < numrecs) {
3422                         ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3423                                 xfs_btree_rec_addr(cur, ptr + 1, block)));
3424                 }
3425 #endif
3426         }
3427
3428         /* Log the new number of records in the btree header. */
3429         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3430
3431         /*
3432          * If we just inserted into a new tree block, we have to
3433          * recalculate nkey here because nkey is out of date.
3434          *
3435          * Otherwise we're just updating an existing block (having shoved
3436          * some records into the new tree block), so use the regular key
3437          * update mechanism.
3438          */
3439         if (bp && xfs_buf_daddr(bp) != old_bn) {
3440                 xfs_btree_get_keys(cur, block, lkey);
3441         } else if (xfs_btree_needs_key_update(cur, optr)) {
3442                 error = xfs_btree_update_keys(cur, level);
3443                 if (error)
3444                         goto error0;
3445         }
3446
3447         /*
3448          * If we are tracking the last record in the tree and
3449          * we are at the far right edge of the tree, update it.
3450          */
3451         if (xfs_btree_is_lastrec(cur, block, level)) {
3452                 cur->bc_ops->update_lastrec(cur, block, rec,
3453                                             ptr, LASTREC_INSREC);
3454         }
3455
3456         /*
3457          * Return the new block number, if any.
3458          * If there is one, give back a record value and a cursor too.
3459          */
3460         *ptrp = nptr;
3461         if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3462                 xfs_btree_copy_keys(cur, key, lkey, 1);
3463                 *curp = ncur;
3464         }
3465
3466         *stat = 1;
3467         return 0;
3468
3469 error0:
3470         if (ncur)
3471                 xfs_btree_del_cursor(ncur, error);
3472         return error;
3473 }
3474
3475 /*
3476  * Insert the record at the point referenced by cur.
3477  *
3478  * A multi-level split of the tree on insert will invalidate the original
3479  * cursor.  All callers of this function should assume that the cursor is
3480  * no longer valid and revalidate it.
3481  */
3482 int
3483 xfs_btree_insert(
3484         struct xfs_btree_cur    *cur,
3485         int                     *stat)
3486 {
3487         int                     error;  /* error return value */
3488         int                     i;      /* result value, 0 for failure */
3489         int                     level;  /* current level number in btree */
3490         union xfs_btree_ptr     nptr;   /* new block number (split result) */
3491         struct xfs_btree_cur    *ncur;  /* new cursor (split result) */
3492         struct xfs_btree_cur    *pcur;  /* previous level's cursor */
3493         union xfs_btree_key     bkey;   /* key of block to insert */
3494         union xfs_btree_key     *key;
3495         union xfs_btree_rec     rec;    /* record to insert */
3496
3497         level = 0;
3498         ncur = NULL;
3499         pcur = cur;
3500         key = &bkey;
3501
3502         xfs_btree_set_ptr_null(cur, &nptr);
3503
3504         /* Make a key out of the record data to be inserted, and save it. */
3505         cur->bc_ops->init_rec_from_cur(cur, &rec);
3506         cur->bc_ops->init_key_from_rec(key, &rec);
3507
3508         /*
3509          * Loop going up the tree, starting at the leaf level.
3510          * Stop when we don't get a split block, that must mean that
3511          * the insert is finished with this level.
3512          */
3513         do {
3514                 /*
3515                  * Insert nrec/nptr into this level of the tree.
3516                  * Note if we fail, nptr will be null.
3517                  */
3518                 error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3519                                 &ncur, &i);
3520                 if (error) {
3521                         if (pcur != cur)
3522                                 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3523                         goto error0;
3524                 }
3525
3526                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3527                         error = -EFSCORRUPTED;
3528                         goto error0;
3529                 }
3530                 level++;
3531
3532                 /*
3533                  * See if the cursor we just used is trash.
3534                  * Can't trash the caller's cursor, but otherwise we should
3535                  * if ncur is a new cursor or we're about to be done.
3536                  */
3537                 if (pcur != cur &&
3538                     (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3539                         /* Save the state from the cursor before we trash it */
3540                         if (cur->bc_ops->update_cursor)
3541                                 cur->bc_ops->update_cursor(pcur, cur);
3542                         cur->bc_nlevels = pcur->bc_nlevels;
3543                         xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3544                 }
3545                 /* If we got a new cursor, switch to it. */
3546                 if (ncur) {
3547                         pcur = ncur;
3548                         ncur = NULL;
3549                 }
3550         } while (!xfs_btree_ptr_is_null(cur, &nptr));
3551
3552         *stat = i;
3553         return 0;
3554 error0:
3555         return error;
3556 }
3557
3558 /*
3559  * Try to merge a non-leaf block back into the inode root.
3560  *
3561  * Note: the killroot names comes from the fact that we're effectively
3562  * killing the old root block.  But because we can't just delete the
3563  * inode we have to copy the single block it was pointing to into the
3564  * inode.
3565  */
3566 STATIC int
3567 xfs_btree_kill_iroot(
3568         struct xfs_btree_cur    *cur)
3569 {
3570         int                     whichfork = cur->bc_ino.whichfork;
3571         struct xfs_inode        *ip = cur->bc_ino.ip;
3572         struct xfs_ifork        *ifp = xfs_ifork_ptr(ip, whichfork);
3573         struct xfs_btree_block  *block;
3574         struct xfs_btree_block  *cblock;
3575         union xfs_btree_key     *kp;
3576         union xfs_btree_key     *ckp;
3577         union xfs_btree_ptr     *pp;
3578         union xfs_btree_ptr     *cpp;
3579         struct xfs_buf          *cbp;
3580         int                     level;
3581         int                     index;
3582         int                     numrecs;
3583         int                     error;
3584 #ifdef DEBUG
3585         union xfs_btree_ptr     ptr;
3586 #endif
3587         int                     i;
3588
3589         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3590         ASSERT(cur->bc_nlevels > 1);
3591
3592         /*
3593          * Don't deal with the root block needs to be a leaf case.
3594          * We're just going to turn the thing back into extents anyway.
3595          */
3596         level = cur->bc_nlevels - 1;
3597         if (level == 1)
3598                 goto out0;
3599
3600         /*
3601          * Give up if the root has multiple children.
3602          */
3603         block = xfs_btree_get_iroot(cur);
3604         if (xfs_btree_get_numrecs(block) != 1)
3605                 goto out0;
3606
3607         cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3608         numrecs = xfs_btree_get_numrecs(cblock);
3609
3610         /*
3611          * Only do this if the next level will fit.
3612          * Then the data must be copied up to the inode,
3613          * instead of freeing the root you free the next level.
3614          */
3615         if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3616                 goto out0;
3617
3618         XFS_BTREE_STATS_INC(cur, killroot);
3619
3620 #ifdef DEBUG
3621         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3622         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3623         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3624         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3625 #endif
3626
3627         index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3628         if (index) {
3629                 xfs_iroot_realloc(cur->bc_ino.ip, index,
3630                                   cur->bc_ino.whichfork);
3631                 block = ifp->if_broot;
3632         }
3633
3634         be16_add_cpu(&block->bb_numrecs, index);
3635         ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3636
3637         kp = xfs_btree_key_addr(cur, 1, block);
3638         ckp = xfs_btree_key_addr(cur, 1, cblock);
3639         xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3640
3641         pp = xfs_btree_ptr_addr(cur, 1, block);
3642         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3643
3644         for (i = 0; i < numrecs; i++) {
3645                 error = xfs_btree_debug_check_ptr(cur, cpp, i, level - 1);
3646                 if (error)
3647                         return error;
3648         }
3649
3650         xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3651
3652         error = xfs_btree_free_block(cur, cbp);
3653         if (error)
3654                 return error;
3655
3656         cur->bc_levels[level - 1].bp = NULL;
3657         be16_add_cpu(&block->bb_level, -1);
3658         xfs_trans_log_inode(cur->bc_tp, ip,
3659                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork));
3660         cur->bc_nlevels--;
3661 out0:
3662         return 0;
3663 }
3664
3665 /*
3666  * Kill the current root node, and replace it with it's only child node.
3667  */
3668 STATIC int
3669 xfs_btree_kill_root(
3670         struct xfs_btree_cur    *cur,
3671         struct xfs_buf          *bp,
3672         int                     level,
3673         union xfs_btree_ptr     *newroot)
3674 {
3675         int                     error;
3676
3677         XFS_BTREE_STATS_INC(cur, killroot);
3678
3679         /*
3680          * Update the root pointer, decreasing the level by 1 and then
3681          * free the old root.
3682          */
3683         cur->bc_ops->set_root(cur, newroot, -1);
3684
3685         error = xfs_btree_free_block(cur, bp);
3686         if (error)
3687                 return error;
3688
3689         cur->bc_levels[level].bp = NULL;
3690         cur->bc_levels[level].ra = 0;
3691         cur->bc_nlevels--;
3692
3693         return 0;
3694 }
3695
3696 STATIC int
3697 xfs_btree_dec_cursor(
3698         struct xfs_btree_cur    *cur,
3699         int                     level,
3700         int                     *stat)
3701 {
3702         int                     error;
3703         int                     i;
3704
3705         if (level > 0) {
3706                 error = xfs_btree_decrement(cur, level, &i);
3707                 if (error)
3708                         return error;
3709         }
3710
3711         *stat = 1;
3712         return 0;
3713 }
3714
3715 /*
3716  * Single level of the btree record deletion routine.
3717  * Delete record pointed to by cur/level.
3718  * Remove the record from its block then rebalance the tree.
3719  * Return 0 for error, 1 for done, 2 to go on to the next level.
3720  */
3721 STATIC int                                      /* error */
3722 xfs_btree_delrec(
3723         struct xfs_btree_cur    *cur,           /* btree cursor */
3724         int                     level,          /* level removing record from */
3725         int                     *stat)          /* fail/done/go-on */
3726 {
3727         struct xfs_btree_block  *block;         /* btree block */
3728         union xfs_btree_ptr     cptr;           /* current block ptr */
3729         struct xfs_buf          *bp;            /* buffer for block */
3730         int                     error;          /* error return value */
3731         int                     i;              /* loop counter */
3732         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
3733         struct xfs_buf          *lbp;           /* left buffer pointer */
3734         struct xfs_btree_block  *left;          /* left btree block */
3735         int                     lrecs = 0;      /* left record count */
3736         int                     ptr;            /* key/record index */
3737         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
3738         struct xfs_buf          *rbp;           /* right buffer pointer */
3739         struct xfs_btree_block  *right;         /* right btree block */
3740         struct xfs_btree_block  *rrblock;       /* right-right btree block */
3741         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
3742         int                     rrecs = 0;      /* right record count */
3743         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
3744         int                     numrecs;        /* temporary numrec count */
3745
3746         tcur = NULL;
3747
3748         /* Get the index of the entry being deleted, check for nothing there. */
3749         ptr = cur->bc_levels[level].ptr;
3750         if (ptr == 0) {
3751                 *stat = 0;
3752                 return 0;
3753         }
3754
3755         /* Get the buffer & block containing the record or key/ptr. */
3756         block = xfs_btree_get_block(cur, level, &bp);
3757         numrecs = xfs_btree_get_numrecs(block);
3758
3759 #ifdef DEBUG
3760         error = xfs_btree_check_block(cur, block, level, bp);
3761         if (error)
3762                 goto error0;
3763 #endif
3764
3765         /* Fail if we're off the end of the block. */
3766         if (ptr > numrecs) {
3767                 *stat = 0;
3768                 return 0;
3769         }
3770
3771         XFS_BTREE_STATS_INC(cur, delrec);
3772         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3773
3774         /* Excise the entries being deleted. */
3775         if (level > 0) {
3776                 /* It's a nonleaf. operate on keys and ptrs */
3777                 union xfs_btree_key     *lkp;
3778                 union xfs_btree_ptr     *lpp;
3779
3780                 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3781                 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3782
3783                 for (i = 0; i < numrecs - ptr; i++) {
3784                         error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
3785                         if (error)
3786                                 goto error0;
3787                 }
3788
3789                 if (ptr < numrecs) {
3790                         xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3791                         xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3792                         xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3793                         xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3794                 }
3795         } else {
3796                 /* It's a leaf. operate on records */
3797                 if (ptr < numrecs) {
3798                         xfs_btree_shift_recs(cur,
3799                                 xfs_btree_rec_addr(cur, ptr + 1, block),
3800                                 -1, numrecs - ptr);
3801                         xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3802                 }
3803         }
3804
3805         /*
3806          * Decrement and log the number of entries in the block.
3807          */
3808         xfs_btree_set_numrecs(block, --numrecs);
3809         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3810
3811         /*
3812          * If we are tracking the last record in the tree and
3813          * we are at the far right edge of the tree, update it.
3814          */
3815         if (xfs_btree_is_lastrec(cur, block, level)) {
3816                 cur->bc_ops->update_lastrec(cur, block, NULL,
3817                                             ptr, LASTREC_DELREC);
3818         }
3819
3820         /*
3821          * We're at the root level.  First, shrink the root block in-memory.
3822          * Try to get rid of the next level down.  If we can't then there's
3823          * nothing left to do.
3824          */
3825         if (level == cur->bc_nlevels - 1) {
3826                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3827                         xfs_iroot_realloc(cur->bc_ino.ip, -1,
3828                                           cur->bc_ino.whichfork);
3829
3830                         error = xfs_btree_kill_iroot(cur);
3831                         if (error)
3832                                 goto error0;
3833
3834                         error = xfs_btree_dec_cursor(cur, level, stat);
3835                         if (error)
3836                                 goto error0;
3837                         *stat = 1;
3838                         return 0;
3839                 }
3840
3841                 /*
3842                  * If this is the root level, and there's only one entry left,
3843                  * and it's NOT the leaf level, then we can get rid of this
3844                  * level.
3845                  */
3846                 if (numrecs == 1 && level > 0) {
3847                         union xfs_btree_ptr     *pp;
3848                         /*
3849                          * pp is still set to the first pointer in the block.
3850                          * Make it the new root of the btree.
3851                          */
3852                         pp = xfs_btree_ptr_addr(cur, 1, block);
3853                         error = xfs_btree_kill_root(cur, bp, level, pp);
3854                         if (error)
3855                                 goto error0;
3856                 } else if (level > 0) {
3857                         error = xfs_btree_dec_cursor(cur, level, stat);
3858                         if (error)
3859                                 goto error0;
3860                 }
3861                 *stat = 1;
3862                 return 0;
3863         }
3864
3865         /*
3866          * If we deleted the leftmost entry in the block, update the
3867          * key values above us in the tree.
3868          */
3869         if (xfs_btree_needs_key_update(cur, ptr)) {
3870                 error = xfs_btree_update_keys(cur, level);
3871                 if (error)
3872                         goto error0;
3873         }
3874
3875         /*
3876          * If the number of records remaining in the block is at least
3877          * the minimum, we're done.
3878          */
3879         if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3880                 error = xfs_btree_dec_cursor(cur, level, stat);
3881                 if (error)
3882                         goto error0;
3883                 return 0;
3884         }
3885
3886         /*
3887          * Otherwise, we have to move some records around to keep the
3888          * tree balanced.  Look at the left and right sibling blocks to
3889          * see if we can re-balance by moving only one record.
3890          */
3891         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3892         xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3893
3894         if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3895                 /*
3896                  * One child of root, need to get a chance to copy its contents
3897                  * into the root and delete it. Can't go up to next level,
3898                  * there's nothing to delete there.
3899                  */
3900                 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3901                     xfs_btree_ptr_is_null(cur, &lptr) &&
3902                     level == cur->bc_nlevels - 2) {
3903                         error = xfs_btree_kill_iroot(cur);
3904                         if (!error)
3905                                 error = xfs_btree_dec_cursor(cur, level, stat);
3906                         if (error)
3907                                 goto error0;
3908                         return 0;
3909                 }
3910         }
3911
3912         ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3913                !xfs_btree_ptr_is_null(cur, &lptr));
3914
3915         /*
3916          * Duplicate the cursor so our btree manipulations here won't
3917          * disrupt the next level up.
3918          */
3919         error = xfs_btree_dup_cursor(cur, &tcur);
3920         if (error)
3921                 goto error0;
3922
3923         /*
3924          * If there's a right sibling, see if it's ok to shift an entry
3925          * out of it.
3926          */
3927         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3928                 /*
3929                  * Move the temp cursor to the last entry in the next block.
3930                  * Actually any entry but the first would suffice.
3931                  */
3932                 i = xfs_btree_lastrec(tcur, level);
3933                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3934                         error = -EFSCORRUPTED;
3935                         goto error0;
3936                 }
3937
3938                 error = xfs_btree_increment(tcur, level, &i);
3939                 if (error)
3940                         goto error0;
3941                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3942                         error = -EFSCORRUPTED;
3943                         goto error0;
3944                 }
3945
3946                 i = xfs_btree_lastrec(tcur, level);
3947                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3948                         error = -EFSCORRUPTED;
3949                         goto error0;
3950                 }
3951
3952                 /* Grab a pointer to the block. */
3953                 right = xfs_btree_get_block(tcur, level, &rbp);
3954 #ifdef DEBUG
3955                 error = xfs_btree_check_block(tcur, right, level, rbp);
3956                 if (error)
3957                         goto error0;
3958 #endif
3959                 /* Grab the current block number, for future use. */
3960                 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3961
3962                 /*
3963                  * If right block is full enough so that removing one entry
3964                  * won't make it too empty, and left-shifting an entry out
3965                  * of right to us works, we're done.
3966                  */
3967                 if (xfs_btree_get_numrecs(right) - 1 >=
3968                     cur->bc_ops->get_minrecs(tcur, level)) {
3969                         error = xfs_btree_lshift(tcur, level, &i);
3970                         if (error)
3971                                 goto error0;
3972                         if (i) {
3973                                 ASSERT(xfs_btree_get_numrecs(block) >=
3974                                        cur->bc_ops->get_minrecs(tcur, level));
3975
3976                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3977                                 tcur = NULL;
3978
3979                                 error = xfs_btree_dec_cursor(cur, level, stat);
3980                                 if (error)
3981                                         goto error0;
3982                                 return 0;
3983                         }
3984                 }
3985
3986                 /*
3987                  * Otherwise, grab the number of records in right for
3988                  * future reference, and fix up the temp cursor to point
3989                  * to our block again (last record).
3990                  */
3991                 rrecs = xfs_btree_get_numrecs(right);
3992                 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3993                         i = xfs_btree_firstrec(tcur, level);
3994                         if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3995                                 error = -EFSCORRUPTED;
3996                                 goto error0;
3997                         }
3998
3999                         error = xfs_btree_decrement(tcur, level, &i);
4000                         if (error)
4001                                 goto error0;
4002                         if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4003                                 error = -EFSCORRUPTED;
4004                                 goto error0;
4005                         }
4006                 }
4007         }
4008
4009         /*
4010          * If there's a left sibling, see if it's ok to shift an entry
4011          * out of it.
4012          */
4013         if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4014                 /*
4015                  * Move the temp cursor to the first entry in the
4016                  * previous block.
4017                  */
4018                 i = xfs_btree_firstrec(tcur, level);
4019                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4020                         error = -EFSCORRUPTED;
4021                         goto error0;
4022                 }
4023
4024                 error = xfs_btree_decrement(tcur, level, &i);
4025                 if (error)
4026                         goto error0;
4027                 i = xfs_btree_firstrec(tcur, level);
4028                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4029                         error = -EFSCORRUPTED;
4030                         goto error0;
4031                 }
4032
4033                 /* Grab a pointer to the block. */
4034                 left = xfs_btree_get_block(tcur, level, &lbp);
4035 #ifdef DEBUG
4036                 error = xfs_btree_check_block(cur, left, level, lbp);
4037                 if (error)
4038                         goto error0;
4039 #endif
4040                 /* Grab the current block number, for future use. */
4041                 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
4042
4043                 /*
4044                  * If left block is full enough so that removing one entry
4045                  * won't make it too empty, and right-shifting an entry out
4046                  * of left to us works, we're done.
4047                  */
4048                 if (xfs_btree_get_numrecs(left) - 1 >=
4049                     cur->bc_ops->get_minrecs(tcur, level)) {
4050                         error = xfs_btree_rshift(tcur, level, &i);
4051                         if (error)
4052                                 goto error0;
4053                         if (i) {
4054                                 ASSERT(xfs_btree_get_numrecs(block) >=
4055                                        cur->bc_ops->get_minrecs(tcur, level));
4056                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4057                                 tcur = NULL;
4058                                 if (level == 0)
4059                                         cur->bc_levels[0].ptr++;
4060
4061                                 *stat = 1;
4062                                 return 0;
4063                         }
4064                 }
4065
4066                 /*
4067                  * Otherwise, grab the number of records in right for
4068                  * future reference.
4069                  */
4070                 lrecs = xfs_btree_get_numrecs(left);
4071         }
4072
4073         /* Delete the temp cursor, we're done with it. */
4074         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4075         tcur = NULL;
4076
4077         /* If here, we need to do a join to keep the tree balanced. */
4078         ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
4079
4080         if (!xfs_btree_ptr_is_null(cur, &lptr) &&
4081             lrecs + xfs_btree_get_numrecs(block) <=
4082                         cur->bc_ops->get_maxrecs(cur, level)) {
4083                 /*
4084                  * Set "right" to be the starting block,
4085                  * "left" to be the left neighbor.
4086                  */
4087                 rptr = cptr;
4088                 right = block;
4089                 rbp = bp;
4090                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
4091                 if (error)
4092                         goto error0;
4093
4094         /*
4095          * If that won't work, see if we can join with the right neighbor block.
4096          */
4097         } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
4098                    rrecs + xfs_btree_get_numrecs(block) <=
4099                         cur->bc_ops->get_maxrecs(cur, level)) {
4100                 /*
4101                  * Set "left" to be the starting block,
4102                  * "right" to be the right neighbor.
4103                  */
4104                 lptr = cptr;
4105                 left = block;
4106                 lbp = bp;
4107                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4108                 if (error)
4109                         goto error0;
4110
4111         /*
4112          * Otherwise, we can't fix the imbalance.
4113          * Just return.  This is probably a logic error, but it's not fatal.
4114          */
4115         } else {
4116                 error = xfs_btree_dec_cursor(cur, level, stat);
4117                 if (error)
4118                         goto error0;
4119                 return 0;
4120         }
4121
4122         rrecs = xfs_btree_get_numrecs(right);
4123         lrecs = xfs_btree_get_numrecs(left);
4124
4125         /*
4126          * We're now going to join "left" and "right" by moving all the stuff
4127          * in "right" to "left" and deleting "right".
4128          */
4129         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4130         if (level > 0) {
4131                 /* It's a non-leaf.  Move keys and pointers. */
4132                 union xfs_btree_key     *lkp;   /* left btree key */
4133                 union xfs_btree_ptr     *lpp;   /* left address pointer */
4134                 union xfs_btree_key     *rkp;   /* right btree key */
4135                 union xfs_btree_ptr     *rpp;   /* right address pointer */
4136
4137                 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4138                 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4139                 rkp = xfs_btree_key_addr(cur, 1, right);
4140                 rpp = xfs_btree_ptr_addr(cur, 1, right);
4141
4142                 for (i = 1; i < rrecs; i++) {
4143                         error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
4144                         if (error)
4145                                 goto error0;
4146                 }
4147
4148                 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4149                 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4150
4151                 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4152                 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4153         } else {
4154                 /* It's a leaf.  Move records.  */
4155                 union xfs_btree_rec     *lrp;   /* left record pointer */
4156                 union xfs_btree_rec     *rrp;   /* right record pointer */
4157
4158                 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4159                 rrp = xfs_btree_rec_addr(cur, 1, right);
4160
4161                 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4162                 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4163         }
4164
4165         XFS_BTREE_STATS_INC(cur, join);
4166
4167         /*
4168          * Fix up the number of records and right block pointer in the
4169          * surviving block, and log it.
4170          */
4171         xfs_btree_set_numrecs(left, lrecs + rrecs);
4172         xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB);
4173         xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4174         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4175
4176         /* If there is a right sibling, point it to the remaining block. */
4177         xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4178         if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4179                 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4180                 if (error)
4181                         goto error0;
4182                 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4183                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4184         }
4185
4186         /* Free the deleted block. */
4187         error = xfs_btree_free_block(cur, rbp);
4188         if (error)
4189                 goto error0;
4190
4191         /*
4192          * If we joined with the left neighbor, set the buffer in the
4193          * cursor to the left block, and fix up the index.
4194          */
4195         if (bp != lbp) {
4196                 cur->bc_levels[level].bp = lbp;
4197                 cur->bc_levels[level].ptr += lrecs;
4198                 cur->bc_levels[level].ra = 0;
4199         }
4200         /*
4201          * If we joined with the right neighbor and there's a level above
4202          * us, increment the cursor at that level.
4203          */
4204         else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
4205                    (level + 1 < cur->bc_nlevels)) {
4206                 error = xfs_btree_increment(cur, level + 1, &i);
4207                 if (error)
4208                         goto error0;
4209         }
4210
4211         /*
4212          * Readjust the ptr at this level if it's not a leaf, since it's
4213          * still pointing at the deletion point, which makes the cursor
4214          * inconsistent.  If this makes the ptr 0, the caller fixes it up.
4215          * We can't use decrement because it would change the next level up.
4216          */
4217         if (level > 0)
4218                 cur->bc_levels[level].ptr--;
4219
4220         /*
4221          * We combined blocks, so we have to update the parent keys if the
4222          * btree supports overlapped intervals.  However,
4223          * bc_levels[level + 1].ptr points to the old block so that the caller
4224          * knows which record to delete.  Therefore, the caller must be savvy
4225          * enough to call updkeys for us if we return stat == 2.  The other
4226          * exit points from this function don't require deletions further up
4227          * the tree, so they can call updkeys directly.
4228          */
4229
4230         /* Return value means the next level up has something to do. */
4231         *stat = 2;
4232         return 0;
4233
4234 error0:
4235         if (tcur)
4236                 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4237         return error;
4238 }
4239
4240 /*
4241  * Delete the record pointed to by cur.
4242  * The cursor refers to the place where the record was (could be inserted)
4243  * when the operation returns.
4244  */
4245 int                                     /* error */
4246 xfs_btree_delete(
4247         struct xfs_btree_cur    *cur,
4248         int                     *stat)  /* success/failure */
4249 {
4250         int                     error;  /* error return value */
4251         int                     level;
4252         int                     i;
4253         bool                    joined = false;
4254
4255         /*
4256          * Go up the tree, starting at leaf level.
4257          *
4258          * If 2 is returned then a join was done; go to the next level.
4259          * Otherwise we are done.
4260          */
4261         for (level = 0, i = 2; i == 2; level++) {
4262                 error = xfs_btree_delrec(cur, level, &i);
4263                 if (error)
4264                         goto error0;
4265                 if (i == 2)
4266                         joined = true;
4267         }
4268
4269         /*
4270          * If we combined blocks as part of deleting the record, delrec won't
4271          * have updated the parent high keys so we have to do that here.
4272          */
4273         if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) {
4274                 error = xfs_btree_updkeys_force(cur, 0);
4275                 if (error)
4276                         goto error0;
4277         }
4278
4279         if (i == 0) {
4280                 for (level = 1; level < cur->bc_nlevels; level++) {
4281                         if (cur->bc_levels[level].ptr == 0) {
4282                                 error = xfs_btree_decrement(cur, level, &i);
4283                                 if (error)
4284                                         goto error0;
4285                                 break;
4286                         }
4287                 }
4288         }
4289
4290         *stat = i;
4291         return 0;
4292 error0:
4293         return error;
4294 }
4295
4296 /*
4297  * Get the data from the pointed-to record.
4298  */
4299 int                                     /* error */
4300 xfs_btree_get_rec(
4301         struct xfs_btree_cur    *cur,   /* btree cursor */
4302         union xfs_btree_rec     **recp, /* output: btree record */
4303         int                     *stat)  /* output: success/failure */
4304 {
4305         struct xfs_btree_block  *block; /* btree block */
4306         struct xfs_buf          *bp;    /* buffer pointer */
4307         int                     ptr;    /* record number */
4308 #ifdef DEBUG
4309         int                     error;  /* error return value */
4310 #endif
4311
4312         ptr = cur->bc_levels[0].ptr;
4313         block = xfs_btree_get_block(cur, 0, &bp);
4314
4315 #ifdef DEBUG
4316         error = xfs_btree_check_block(cur, block, 0, bp);
4317         if (error)
4318                 return error;
4319 #endif
4320
4321         /*
4322          * Off the right end or left end, return failure.
4323          */
4324         if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4325                 *stat = 0;
4326                 return 0;
4327         }
4328
4329         /*
4330          * Point to the record and extract its data.
4331          */
4332         *recp = xfs_btree_rec_addr(cur, ptr, block);
4333         *stat = 1;
4334         return 0;
4335 }
4336
4337 /* Visit a block in a btree. */
4338 STATIC int
4339 xfs_btree_visit_block(
4340         struct xfs_btree_cur            *cur,
4341         int                             level,
4342         xfs_btree_visit_blocks_fn       fn,
4343         void                            *data)
4344 {
4345         struct xfs_btree_block          *block;
4346         struct xfs_buf                  *bp;
4347         union xfs_btree_ptr             rptr;
4348         int                             error;
4349
4350         /* do right sibling readahead */
4351         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4352         block = xfs_btree_get_block(cur, level, &bp);
4353
4354         /* process the block */
4355         error = fn(cur, level, data);
4356         if (error)
4357                 return error;
4358
4359         /* now read rh sibling block for next iteration */
4360         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4361         if (xfs_btree_ptr_is_null(cur, &rptr))
4362                 return -ENOENT;
4363
4364         /*
4365          * We only visit blocks once in this walk, so we have to avoid the
4366          * internal xfs_btree_lookup_get_block() optimisation where it will
4367          * return the same block without checking if the right sibling points
4368          * back to us and creates a cyclic reference in the btree.
4369          */
4370         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
4371                 if (be64_to_cpu(rptr.l) == XFS_DADDR_TO_FSB(cur->bc_mp,
4372                                                         xfs_buf_daddr(bp)))
4373                         return -EFSCORRUPTED;
4374         } else {
4375                 if (be32_to_cpu(rptr.s) == xfs_daddr_to_agbno(cur->bc_mp,
4376                                                         xfs_buf_daddr(bp)))
4377                         return -EFSCORRUPTED;
4378         }
4379         return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4380 }
4381
4382
4383 /* Visit every block in a btree. */
4384 int
4385 xfs_btree_visit_blocks(
4386         struct xfs_btree_cur            *cur,
4387         xfs_btree_visit_blocks_fn       fn,
4388         unsigned int                    flags,
4389         void                            *data)
4390 {
4391         union xfs_btree_ptr             lptr;
4392         int                             level;
4393         struct xfs_btree_block          *block = NULL;
4394         int                             error = 0;
4395
4396         cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4397
4398         /* for each level */
4399         for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4400                 /* grab the left hand block */
4401                 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4402                 if (error)
4403                         return error;
4404
4405                 /* readahead the left most block for the next level down */
4406                 if (level > 0) {
4407                         union xfs_btree_ptr     *ptr;
4408
4409                         ptr = xfs_btree_ptr_addr(cur, 1, block);
4410                         xfs_btree_readahead_ptr(cur, ptr, 1);
4411
4412                         /* save for the next iteration of the loop */
4413                         xfs_btree_copy_ptrs(cur, &lptr, ptr, 1);
4414
4415                         if (!(flags & XFS_BTREE_VISIT_LEAVES))
4416                                 continue;
4417                 } else if (!(flags & XFS_BTREE_VISIT_RECORDS)) {
4418                         continue;
4419                 }
4420
4421                 /* for each buffer in the level */
4422                 do {
4423                         error = xfs_btree_visit_block(cur, level, fn, data);
4424                 } while (!error);
4425
4426                 if (error != -ENOENT)
4427                         return error;
4428         }
4429
4430         return 0;
4431 }
4432
4433 /*
4434  * Change the owner of a btree.
4435  *
4436  * The mechanism we use here is ordered buffer logging. Because we don't know
4437  * how many buffers were are going to need to modify, we don't really want to
4438  * have to make transaction reservations for the worst case of every buffer in a
4439  * full size btree as that may be more space that we can fit in the log....
4440  *
4441  * We do the btree walk in the most optimal manner possible - we have sibling
4442  * pointers so we can just walk all the blocks on each level from left to right
4443  * in a single pass, and then move to the next level and do the same. We can
4444  * also do readahead on the sibling pointers to get IO moving more quickly,
4445  * though for slow disks this is unlikely to make much difference to performance
4446  * as the amount of CPU work we have to do before moving to the next block is
4447  * relatively small.
4448  *
4449  * For each btree block that we load, modify the owner appropriately, set the
4450  * buffer as an ordered buffer and log it appropriately. We need to ensure that
4451  * we mark the region we change dirty so that if the buffer is relogged in
4452  * a subsequent transaction the changes we make here as an ordered buffer are
4453  * correctly relogged in that transaction.  If we are in recovery context, then
4454  * just queue the modified buffer as delayed write buffer so the transaction
4455  * recovery completion writes the changes to disk.
4456  */
4457 struct xfs_btree_block_change_owner_info {
4458         uint64_t                new_owner;
4459         struct list_head        *buffer_list;
4460 };
4461
4462 static int
4463 xfs_btree_block_change_owner(
4464         struct xfs_btree_cur    *cur,
4465         int                     level,
4466         void                    *data)
4467 {
4468         struct xfs_btree_block_change_owner_info        *bbcoi = data;
4469         struct xfs_btree_block  *block;
4470         struct xfs_buf          *bp;
4471
4472         /* modify the owner */
4473         block = xfs_btree_get_block(cur, level, &bp);
4474         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
4475                 if (block->bb_u.l.bb_owner == cpu_to_be64(bbcoi->new_owner))
4476                         return 0;
4477                 block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4478         } else {
4479                 if (block->bb_u.s.bb_owner == cpu_to_be32(bbcoi->new_owner))
4480                         return 0;
4481                 block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4482         }
4483
4484         /*
4485          * If the block is a root block hosted in an inode, we might not have a
4486          * buffer pointer here and we shouldn't attempt to log the change as the
4487          * information is already held in the inode and discarded when the root
4488          * block is formatted into the on-disk inode fork. We still change it,
4489          * though, so everything is consistent in memory.
4490          */
4491         if (!bp) {
4492                 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4493                 ASSERT(level == cur->bc_nlevels - 1);
4494                 return 0;
4495         }
4496
4497         if (cur->bc_tp) {
4498                 if (!xfs_trans_ordered_buf(cur->bc_tp, bp)) {
4499                         xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4500                         return -EAGAIN;
4501                 }
4502         } else {
4503                 xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4504         }
4505
4506         return 0;
4507 }
4508
4509 int
4510 xfs_btree_change_owner(
4511         struct xfs_btree_cur    *cur,
4512         uint64_t                new_owner,
4513         struct list_head        *buffer_list)
4514 {
4515         struct xfs_btree_block_change_owner_info        bbcoi;
4516
4517         bbcoi.new_owner = new_owner;
4518         bbcoi.buffer_list = buffer_list;
4519
4520         return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4521                         XFS_BTREE_VISIT_ALL, &bbcoi);
4522 }
4523
4524 /* Verify the v5 fields of a long-format btree block. */
4525 xfs_failaddr_t
4526 xfs_btree_lblock_v5hdr_verify(
4527         struct xfs_buf          *bp,
4528         uint64_t                owner)
4529 {
4530         struct xfs_mount        *mp = bp->b_mount;
4531         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4532
4533         if (!xfs_has_crc(mp))
4534                 return __this_address;
4535         if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
4536                 return __this_address;
4537         if (block->bb_u.l.bb_blkno != cpu_to_be64(xfs_buf_daddr(bp)))
4538                 return __this_address;
4539         if (owner != XFS_RMAP_OWN_UNKNOWN &&
4540             be64_to_cpu(block->bb_u.l.bb_owner) != owner)
4541                 return __this_address;
4542         return NULL;
4543 }
4544
4545 /* Verify a long-format btree block. */
4546 xfs_failaddr_t
4547 xfs_btree_lblock_verify(
4548         struct xfs_buf          *bp,
4549         unsigned int            max_recs)
4550 {
4551         struct xfs_mount        *mp = bp->b_mount;
4552         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4553         xfs_fsblock_t           fsb;
4554         xfs_failaddr_t          fa;
4555
4556         /* numrecs verification */
4557         if (be16_to_cpu(block->bb_numrecs) > max_recs)
4558                 return __this_address;
4559
4560         /* sibling pointer verification */
4561         fsb = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
4562         fa = xfs_btree_check_lblock_siblings(mp, NULL, -1, fsb,
4563                         block->bb_u.l.bb_leftsib);
4564         if (!fa)
4565                 fa = xfs_btree_check_lblock_siblings(mp, NULL, -1, fsb,
4566                                 block->bb_u.l.bb_rightsib);
4567         return fa;
4568 }
4569
4570 /**
4571  * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4572  *                                    btree block
4573  *
4574  * @bp: buffer containing the btree block
4575  */
4576 xfs_failaddr_t
4577 xfs_btree_sblock_v5hdr_verify(
4578         struct xfs_buf          *bp)
4579 {
4580         struct xfs_mount        *mp = bp->b_mount;
4581         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4582         struct xfs_perag        *pag = bp->b_pag;
4583
4584         if (!xfs_has_crc(mp))
4585                 return __this_address;
4586         if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4587                 return __this_address;
4588         if (block->bb_u.s.bb_blkno != cpu_to_be64(xfs_buf_daddr(bp)))
4589                 return __this_address;
4590         if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4591                 return __this_address;
4592         return NULL;
4593 }
4594
4595 /**
4596  * xfs_btree_sblock_verify() -- verify a short-format btree block
4597  *
4598  * @bp: buffer containing the btree block
4599  * @max_recs: maximum records allowed in this btree node
4600  */
4601 xfs_failaddr_t
4602 xfs_btree_sblock_verify(
4603         struct xfs_buf          *bp,
4604         unsigned int            max_recs)
4605 {
4606         struct xfs_mount        *mp = bp->b_mount;
4607         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4608         xfs_agblock_t           agbno;
4609         xfs_failaddr_t          fa;
4610
4611         /* numrecs verification */
4612         if (be16_to_cpu(block->bb_numrecs) > max_recs)
4613                 return __this_address;
4614
4615         /* sibling pointer verification */
4616         agbno = xfs_daddr_to_agbno(mp, xfs_buf_daddr(bp));
4617         fa = xfs_btree_check_sblock_siblings(bp->b_pag, NULL, -1, agbno,
4618                         block->bb_u.s.bb_leftsib);
4619         if (!fa)
4620                 fa = xfs_btree_check_sblock_siblings(bp->b_pag, NULL, -1, agbno,
4621                                 block->bb_u.s.bb_rightsib);
4622         return fa;
4623 }
4624
4625 /*
4626  * For the given limits on leaf and keyptr records per block, calculate the
4627  * height of the tree needed to index the number of leaf records.
4628  */
4629 unsigned int
4630 xfs_btree_compute_maxlevels(
4631         const unsigned int      *limits,
4632         unsigned long long      records)
4633 {
4634         unsigned long long      level_blocks = howmany_64(records, limits[0]);
4635         unsigned int            height = 1;
4636
4637         while (level_blocks > 1) {
4638                 level_blocks = howmany_64(level_blocks, limits[1]);
4639                 height++;
4640         }
4641
4642         return height;
4643 }
4644
4645 /*
4646  * For the given limits on leaf and keyptr records per block, calculate the
4647  * number of blocks needed to index the given number of leaf records.
4648  */
4649 unsigned long long
4650 xfs_btree_calc_size(
4651         const unsigned int      *limits,
4652         unsigned long long      records)
4653 {
4654         unsigned long long      level_blocks = howmany_64(records, limits[0]);
4655         unsigned long long      blocks = level_blocks;
4656
4657         while (level_blocks > 1) {
4658                 level_blocks = howmany_64(level_blocks, limits[1]);
4659                 blocks += level_blocks;
4660         }
4661
4662         return blocks;
4663 }
4664
4665 /*
4666  * Given a number of available blocks for the btree to consume with records and
4667  * pointers, calculate the height of the tree needed to index all the records
4668  * that space can hold based on the number of pointers each interior node
4669  * holds.
4670  *
4671  * We start by assuming a single level tree consumes a single block, then track
4672  * the number of blocks each node level consumes until we no longer have space
4673  * to store the next node level. At this point, we are indexing all the leaf
4674  * blocks in the space, and there's no more free space to split the tree any
4675  * further. That's our maximum btree height.
4676  */
4677 unsigned int
4678 xfs_btree_space_to_height(
4679         const unsigned int      *limits,
4680         unsigned long long      leaf_blocks)
4681 {
4682         /*
4683          * The root btree block can have fewer than minrecs pointers in it
4684          * because the tree might not be big enough to require that amount of
4685          * fanout. Hence it has a minimum size of 2 pointers, not limits[1].
4686          */
4687         unsigned long long      node_blocks = 2;
4688         unsigned long long      blocks_left = leaf_blocks - 1;
4689         unsigned int            height = 1;
4690
4691         if (leaf_blocks < 1)
4692                 return 0;
4693
4694         while (node_blocks < blocks_left) {
4695                 blocks_left -= node_blocks;
4696                 node_blocks *= limits[1];
4697                 height++;
4698         }
4699
4700         return height;
4701 }
4702
4703 /*
4704  * Query a regular btree for all records overlapping a given interval.
4705  * Start with a LE lookup of the key of low_rec and return all records
4706  * until we find a record with a key greater than the key of high_rec.
4707  */
4708 STATIC int
4709 xfs_btree_simple_query_range(
4710         struct xfs_btree_cur            *cur,
4711         const union xfs_btree_key       *low_key,
4712         const union xfs_btree_key       *high_key,
4713         xfs_btree_query_range_fn        fn,
4714         void                            *priv)
4715 {
4716         union xfs_btree_rec             *recp;
4717         union xfs_btree_key             rec_key;
4718         int                             stat;
4719         bool                            firstrec = true;
4720         int                             error;
4721
4722         ASSERT(cur->bc_ops->init_high_key_from_rec);
4723         ASSERT(cur->bc_ops->diff_two_keys);
4724
4725         /*
4726          * Find the leftmost record.  The btree cursor must be set
4727          * to the low record used to generate low_key.
4728          */
4729         stat = 0;
4730         error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4731         if (error)
4732                 goto out;
4733
4734         /* Nothing?  See if there's anything to the right. */
4735         if (!stat) {
4736                 error = xfs_btree_increment(cur, 0, &stat);
4737                 if (error)
4738                         goto out;
4739         }
4740
4741         while (stat) {
4742                 /* Find the record. */
4743                 error = xfs_btree_get_rec(cur, &recp, &stat);
4744                 if (error || !stat)
4745                         break;
4746
4747                 /* Skip if low_key > high_key(rec). */
4748                 if (firstrec) {
4749                         cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4750                         firstrec = false;
4751                         if (xfs_btree_keycmp_gt(cur, low_key, &rec_key))
4752                                 goto advloop;
4753                 }
4754
4755                 /* Stop if low_key(rec) > high_key. */
4756                 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4757                 if (xfs_btree_keycmp_gt(cur, &rec_key, high_key))
4758                         break;
4759
4760                 /* Callback */
4761                 error = fn(cur, recp, priv);
4762                 if (error)
4763                         break;
4764
4765 advloop:
4766                 /* Move on to the next record. */
4767                 error = xfs_btree_increment(cur, 0, &stat);
4768                 if (error)
4769                         break;
4770         }
4771
4772 out:
4773         return error;
4774 }
4775
4776 /*
4777  * Query an overlapped interval btree for all records overlapping a given
4778  * interval.  This function roughly follows the algorithm given in
4779  * "Interval Trees" of _Introduction to Algorithms_, which is section
4780  * 14.3 in the 2nd and 3rd editions.
4781  *
4782  * First, generate keys for the low and high records passed in.
4783  *
4784  * For any leaf node, generate the high and low keys for the record.
4785  * If the record keys overlap with the query low/high keys, pass the
4786  * record to the function iterator.
4787  *
4788  * For any internal node, compare the low and high keys of each
4789  * pointer against the query low/high keys.  If there's an overlap,
4790  * follow the pointer.
4791  *
4792  * As an optimization, we stop scanning a block when we find a low key
4793  * that is greater than the query's high key.
4794  */
4795 STATIC int
4796 xfs_btree_overlapped_query_range(
4797         struct xfs_btree_cur            *cur,
4798         const union xfs_btree_key       *low_key,
4799         const union xfs_btree_key       *high_key,
4800         xfs_btree_query_range_fn        fn,
4801         void                            *priv)
4802 {
4803         union xfs_btree_ptr             ptr;
4804         union xfs_btree_ptr             *pp;
4805         union xfs_btree_key             rec_key;
4806         union xfs_btree_key             rec_hkey;
4807         union xfs_btree_key             *lkp;
4808         union xfs_btree_key             *hkp;
4809         union xfs_btree_rec             *recp;
4810         struct xfs_btree_block          *block;
4811         int                             level;
4812         struct xfs_buf                  *bp;
4813         int                             i;
4814         int                             error;
4815
4816         /* Load the root of the btree. */
4817         level = cur->bc_nlevels - 1;
4818         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
4819         error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4820         if (error)
4821                 return error;
4822         xfs_btree_get_block(cur, level, &bp);
4823         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4824 #ifdef DEBUG
4825         error = xfs_btree_check_block(cur, block, level, bp);
4826         if (error)
4827                 goto out;
4828 #endif
4829         cur->bc_levels[level].ptr = 1;
4830
4831         while (level < cur->bc_nlevels) {
4832                 block = xfs_btree_get_block(cur, level, &bp);
4833
4834                 /* End of node, pop back towards the root. */
4835                 if (cur->bc_levels[level].ptr >
4836                                         be16_to_cpu(block->bb_numrecs)) {
4837 pop_up:
4838                         if (level < cur->bc_nlevels - 1)
4839                                 cur->bc_levels[level + 1].ptr++;
4840                         level++;
4841                         continue;
4842                 }
4843
4844                 if (level == 0) {
4845                         /* Handle a leaf node. */
4846                         recp = xfs_btree_rec_addr(cur, cur->bc_levels[0].ptr,
4847                                         block);
4848
4849                         cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
4850                         cur->bc_ops->init_key_from_rec(&rec_key, recp);
4851
4852                         /*
4853                          * If (query's high key < record's low key), then there
4854                          * are no more interesting records in this block.  Pop
4855                          * up to the leaf level to find more record blocks.
4856                          *
4857                          * If (record's high key >= query's low key) and
4858                          *    (query's high key >= record's low key), then
4859                          * this record overlaps the query range; callback.
4860                          */
4861                         if (xfs_btree_keycmp_lt(cur, high_key, &rec_key))
4862                                 goto pop_up;
4863                         if (xfs_btree_keycmp_ge(cur, &rec_hkey, low_key)) {
4864                                 error = fn(cur, recp, priv);
4865                                 if (error)
4866                                         break;
4867                         }
4868                         cur->bc_levels[level].ptr++;
4869                         continue;
4870                 }
4871
4872                 /* Handle an internal node. */
4873                 lkp = xfs_btree_key_addr(cur, cur->bc_levels[level].ptr, block);
4874                 hkp = xfs_btree_high_key_addr(cur, cur->bc_levels[level].ptr,
4875                                 block);
4876                 pp = xfs_btree_ptr_addr(cur, cur->bc_levels[level].ptr, block);
4877
4878                 /*
4879                  * If (query's high key < pointer's low key), then there are no
4880                  * more interesting keys in this block.  Pop up one leaf level
4881                  * to continue looking for records.
4882                  *
4883                  * If (pointer's high key >= query's low key) and
4884                  *    (query's high key >= pointer's low key), then
4885                  * this record overlaps the query range; follow pointer.
4886                  */
4887                 if (xfs_btree_keycmp_lt(cur, high_key, lkp))
4888                         goto pop_up;
4889                 if (xfs_btree_keycmp_ge(cur, hkp, low_key)) {
4890                         level--;
4891                         error = xfs_btree_lookup_get_block(cur, level, pp,
4892                                         &block);
4893                         if (error)
4894                                 goto out;
4895                         xfs_btree_get_block(cur, level, &bp);
4896                         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4897 #ifdef DEBUG
4898                         error = xfs_btree_check_block(cur, block, level, bp);
4899                         if (error)
4900                                 goto out;
4901 #endif
4902                         cur->bc_levels[level].ptr = 1;
4903                         continue;
4904                 }
4905                 cur->bc_levels[level].ptr++;
4906         }
4907
4908 out:
4909         /*
4910          * If we don't end this function with the cursor pointing at a record
4911          * block, a subsequent non-error cursor deletion will not release
4912          * node-level buffers, causing a buffer leak.  This is quite possible
4913          * with a zero-results range query, so release the buffers if we
4914          * failed to return any results.
4915          */
4916         if (cur->bc_levels[0].bp == NULL) {
4917                 for (i = 0; i < cur->bc_nlevels; i++) {
4918                         if (cur->bc_levels[i].bp) {
4919                                 xfs_trans_brelse(cur->bc_tp,
4920                                                 cur->bc_levels[i].bp);
4921                                 cur->bc_levels[i].bp = NULL;
4922                                 cur->bc_levels[i].ptr = 0;
4923                                 cur->bc_levels[i].ra = 0;
4924                         }
4925                 }
4926         }
4927
4928         return error;
4929 }
4930
4931 static inline void
4932 xfs_btree_key_from_irec(
4933         struct xfs_btree_cur            *cur,
4934         union xfs_btree_key             *key,
4935         const union xfs_btree_irec      *irec)
4936 {
4937         union xfs_btree_rec             rec;
4938
4939         cur->bc_rec = *irec;
4940         cur->bc_ops->init_rec_from_cur(cur, &rec);
4941         cur->bc_ops->init_key_from_rec(key, &rec);
4942 }
4943
4944 /*
4945  * Query a btree for all records overlapping a given interval of keys.  The
4946  * supplied function will be called with each record found; return one of the
4947  * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4948  * code.  This function returns -ECANCELED, zero, or a negative error code.
4949  */
4950 int
4951 xfs_btree_query_range(
4952         struct xfs_btree_cur            *cur,
4953         const union xfs_btree_irec      *low_rec,
4954         const union xfs_btree_irec      *high_rec,
4955         xfs_btree_query_range_fn        fn,
4956         void                            *priv)
4957 {
4958         union xfs_btree_key             low_key;
4959         union xfs_btree_key             high_key;
4960
4961         /* Find the keys of both ends of the interval. */
4962         xfs_btree_key_from_irec(cur, &high_key, high_rec);
4963         xfs_btree_key_from_irec(cur, &low_key, low_rec);
4964
4965         /* Enforce low key <= high key. */
4966         if (!xfs_btree_keycmp_le(cur, &low_key, &high_key))
4967                 return -EINVAL;
4968
4969         if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
4970                 return xfs_btree_simple_query_range(cur, &low_key,
4971                                 &high_key, fn, priv);
4972         return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
4973                         fn, priv);
4974 }
4975
4976 /* Query a btree for all records. */
4977 int
4978 xfs_btree_query_all(
4979         struct xfs_btree_cur            *cur,
4980         xfs_btree_query_range_fn        fn,
4981         void                            *priv)
4982 {
4983         union xfs_btree_key             low_key;
4984         union xfs_btree_key             high_key;
4985
4986         memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
4987         memset(&low_key, 0, sizeof(low_key));
4988         memset(&high_key, 0xFF, sizeof(high_key));
4989
4990         return xfs_btree_simple_query_range(cur, &low_key, &high_key, fn, priv);
4991 }
4992
4993 static int
4994 xfs_btree_count_blocks_helper(
4995         struct xfs_btree_cur    *cur,
4996         int                     level,
4997         void                    *data)
4998 {
4999         xfs_extlen_t            *blocks = data;
5000         (*blocks)++;
5001
5002         return 0;
5003 }
5004
5005 /* Count the blocks in a btree and return the result in *blocks. */
5006 int
5007 xfs_btree_count_blocks(
5008         struct xfs_btree_cur    *cur,
5009         xfs_extlen_t            *blocks)
5010 {
5011         *blocks = 0;
5012         return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
5013                         XFS_BTREE_VISIT_ALL, blocks);
5014 }
5015
5016 /* Compare two btree pointers. */
5017 int64_t
5018 xfs_btree_diff_two_ptrs(
5019         struct xfs_btree_cur            *cur,
5020         const union xfs_btree_ptr       *a,
5021         const union xfs_btree_ptr       *b)
5022 {
5023         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
5024                 return (int64_t)be64_to_cpu(a->l) - be64_to_cpu(b->l);
5025         return (int64_t)be32_to_cpu(a->s) - be32_to_cpu(b->s);
5026 }
5027
5028 struct xfs_btree_has_records {
5029         /* Keys for the start and end of the range we want to know about. */
5030         union xfs_btree_key             start_key;
5031         union xfs_btree_key             end_key;
5032
5033         /* Mask for key comparisons, if desired. */
5034         const union xfs_btree_key       *key_mask;
5035
5036         /* Highest record key we've seen so far. */
5037         union xfs_btree_key             high_key;
5038
5039         enum xbtree_recpacking          outcome;
5040 };
5041
5042 STATIC int
5043 xfs_btree_has_records_helper(
5044         struct xfs_btree_cur            *cur,
5045         const union xfs_btree_rec       *rec,
5046         void                            *priv)
5047 {
5048         union xfs_btree_key             rec_key;
5049         union xfs_btree_key             rec_high_key;
5050         struct xfs_btree_has_records    *info = priv;
5051         enum xbtree_key_contig          key_contig;
5052
5053         cur->bc_ops->init_key_from_rec(&rec_key, rec);
5054
5055         if (info->outcome == XBTREE_RECPACKING_EMPTY) {
5056                 info->outcome = XBTREE_RECPACKING_SPARSE;
5057
5058                 /*
5059                  * If the first record we find does not overlap the start key,
5060                  * then there is a hole at the start of the search range.
5061                  * Classify this as sparse and stop immediately.
5062                  */
5063                 if (xfs_btree_masked_keycmp_lt(cur, &info->start_key, &rec_key,
5064                                         info->key_mask))
5065                         return -ECANCELED;
5066         } else {
5067                 /*
5068                  * If a subsequent record does not overlap with the any record
5069                  * we've seen so far, there is a hole in the middle of the
5070                  * search range.  Classify this as sparse and stop.
5071                  * If the keys overlap and this btree does not allow overlap,
5072                  * signal corruption.
5073                  */
5074                 key_contig = cur->bc_ops->keys_contiguous(cur, &info->high_key,
5075                                         &rec_key, info->key_mask);
5076                 if (key_contig == XBTREE_KEY_OVERLAP &&
5077                                 !(cur->bc_flags & XFS_BTREE_OVERLAPPING))
5078                         return -EFSCORRUPTED;
5079                 if (key_contig == XBTREE_KEY_GAP)
5080                         return -ECANCELED;
5081         }
5082
5083         /*
5084          * If high_key(rec) is larger than any other high key we've seen,
5085          * remember it for later.
5086          */
5087         cur->bc_ops->init_high_key_from_rec(&rec_high_key, rec);
5088         if (xfs_btree_masked_keycmp_gt(cur, &rec_high_key, &info->high_key,
5089                                 info->key_mask))
5090                 info->high_key = rec_high_key; /* struct copy */
5091
5092         return 0;
5093 }
5094
5095 /*
5096  * Scan part of the keyspace of a btree and tell us if that keyspace does not
5097  * map to any records; is fully mapped to records; or is partially mapped to
5098  * records.  This is the btree record equivalent to determining if a file is
5099  * sparse.
5100  *
5101  * For most btree types, the record scan should use all available btree key
5102  * fields to compare the keys encountered.  These callers should pass NULL for
5103  * @mask.  However, some callers (e.g.  scanning physical space in the rmapbt)
5104  * want to ignore some part of the btree record keyspace when performing the
5105  * comparison.  These callers should pass in a union xfs_btree_key object with
5106  * the fields that *should* be a part of the comparison set to any nonzero
5107  * value, and the rest zeroed.
5108  */
5109 int
5110 xfs_btree_has_records(
5111         struct xfs_btree_cur            *cur,
5112         const union xfs_btree_irec      *low,
5113         const union xfs_btree_irec      *high,
5114         const union xfs_btree_key       *mask,
5115         enum xbtree_recpacking          *outcome)
5116 {
5117         struct xfs_btree_has_records    info = {
5118                 .outcome                = XBTREE_RECPACKING_EMPTY,
5119                 .key_mask               = mask,
5120         };
5121         int                             error;
5122
5123         /* Not all btrees support this operation. */
5124         if (!cur->bc_ops->keys_contiguous) {
5125                 ASSERT(0);
5126                 return -EOPNOTSUPP;
5127         }
5128
5129         xfs_btree_key_from_irec(cur, &info.start_key, low);
5130         xfs_btree_key_from_irec(cur, &info.end_key, high);
5131
5132         error = xfs_btree_query_range(cur, low, high,
5133                         xfs_btree_has_records_helper, &info);
5134         if (error == -ECANCELED)
5135                 goto out;
5136         if (error)
5137                 return error;
5138
5139         if (info.outcome == XBTREE_RECPACKING_EMPTY)
5140                 goto out;
5141
5142         /*
5143          * If the largest high_key(rec) we saw during the walk is greater than
5144          * the end of the search range, classify this as full.  Otherwise,
5145          * there is a hole at the end of the search range.
5146          */
5147         if (xfs_btree_masked_keycmp_ge(cur, &info.high_key, &info.end_key,
5148                                 mask))
5149                 info.outcome = XBTREE_RECPACKING_FULL;
5150
5151 out:
5152         *outcome = info.outcome;
5153         return 0;
5154 }
5155
5156 /* Are there more records in this btree? */
5157 bool
5158 xfs_btree_has_more_records(
5159         struct xfs_btree_cur    *cur)
5160 {
5161         struct xfs_btree_block  *block;
5162         struct xfs_buf          *bp;
5163
5164         block = xfs_btree_get_block(cur, 0, &bp);
5165
5166         /* There are still records in this block. */
5167         if (cur->bc_levels[0].ptr < xfs_btree_get_numrecs(block))
5168                 return true;
5169
5170         /* There are more record blocks. */
5171         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
5172                 return block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK);
5173         else
5174                 return block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK);
5175 }
5176
5177 /* Set up all the btree cursor caches. */
5178 int __init
5179 xfs_btree_init_cur_caches(void)
5180 {
5181         int             error;
5182
5183         error = xfs_allocbt_init_cur_cache();
5184         if (error)
5185                 return error;
5186         error = xfs_inobt_init_cur_cache();
5187         if (error)
5188                 goto err;
5189         error = xfs_bmbt_init_cur_cache();
5190         if (error)
5191                 goto err;
5192         error = xfs_rmapbt_init_cur_cache();
5193         if (error)
5194                 goto err;
5195         error = xfs_refcountbt_init_cur_cache();
5196         if (error)
5197                 goto err;
5198
5199         return 0;
5200 err:
5201         xfs_btree_destroy_cur_caches();
5202         return error;
5203 }
5204
5205 /* Destroy all the btree cursor caches, if they've been allocated. */
5206 void
5207 xfs_btree_destroy_cur_caches(void)
5208 {
5209         xfs_allocbt_destroy_cur_cache();
5210         xfs_inobt_destroy_cur_cache();
5211         xfs_bmbt_destroy_cur_cache();
5212         xfs_rmapbt_destroy_cur_cache();
5213         xfs_refcountbt_destroy_cur_cache();
5214 }
5215
5216 /* Move the btree cursor before the first record. */
5217 int
5218 xfs_btree_goto_left_edge(
5219         struct xfs_btree_cur    *cur)
5220 {
5221         int                     stat = 0;
5222         int                     error;
5223
5224         memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
5225         error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
5226         if (error)
5227                 return error;
5228         if (!stat)
5229                 return 0;
5230
5231         error = xfs_btree_decrement(cur, 0, &stat);
5232         if (error)
5233                 return error;
5234         if (stat != 0) {
5235                 ASSERT(0);
5236                 return -EFSCORRUPTED;
5237         }
5238
5239         return 0;
5240 }