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 */
  34static 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
  42uint32_t
  43xfs_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 */
  65static inline xfs_failaddr_t
  66xfs_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
  92static inline xfs_failaddr_t
  93xfs_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 */
 122xfs_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. */
 165static int
 166xfs_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 */
 189xfs_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. */
 231STATIC int
 232xfs_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 */
 254int
 255xfs_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. */
 268bool
 269xfs_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. */
 280bool
 281xfs_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 */
 295static int
 296xfs_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 */
 338void
 339xfs_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
 352bool
 353xfs_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 */
 376void
 377xfs_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
 390bool
 391xfs_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
 406static int
 407xfs_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 */
 424void
 425xfs_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 */
 464int					/* error */
 465xfs_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 */
 590static 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 */
 605static 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 */
 614STATIC size_t
 615xfs_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 */
 626STATIC size_t
 627xfs_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 */
 638STATIC size_t
 639xfs_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 */
 650STATIC size_t
 651xfs_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 */
 664union xfs_btree_rec *
 665xfs_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 */
 677union xfs_btree_key *
 678xfs_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 */
 690union xfs_btree_key *
 691xfs_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 */
 703union xfs_btree_ptr *
 704xfs_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
 717struct xfs_ifork *
 718xfs_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 */
 734STATIC struct xfs_btree_block *
 735xfs_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 */
 747struct xfs_btree_block *		/* generic btree block pointer */
 748xfs_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 */
 767STATIC int				/* success=1, failure=0 */
 768xfs_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 */
 797STATIC int				/* success=1, failure=0 */
 798xfs_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 */
 827void
 828xfs_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 */
 863int
 864xfs_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 */
 894void
 895xfs_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 */
 913void
 914xfs_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
 929STATIC int
 930xfs_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
 954STATIC int
 955xfs_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 */
 984STATIC int
 985xfs_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
1011STATIC int
1012xfs_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 */
1043STATIC void
1044xfs_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 */
1061STATIC void
1062xfs_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
1088bool
1089xfs_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
1099void
1100xfs_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 */
1113void
1114xfs_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
1135void
1136xfs_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
1157void
1158xfs_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
1200void
1201xfs_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
1213void
1214xfs_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 */
1243STATIC int
1244xfs_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
1262STATIC void
1263xfs_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
1277STATIC void
1278xfs_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
1305int
1306xfs_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 */
1333STATIC int
1334xfs_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 */
1365void
1366xfs_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 */
1379STATIC void
1380xfs_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 */
1393void
1394xfs_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 */
1407STATIC void
1408xfs_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 */
1426STATIC void
1427xfs_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 */
1445STATIC void
1446xfs_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 */
1464STATIC void
1465xfs_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 */
1486void
1487xfs_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 */
1504STATIC void
1505xfs_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 */
1530void
1531xfs_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 */
1599int						/* error */
1600xfs_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	}
1685out1:
1686	*stat = 1;
1687	return 0;
1688
1689out0:
1690	*stat = 0;
1691	return 0;
1692
1693error0:
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 */
1701int						/* error */
1702xfs_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	}
1777out1:
1778	*stat = 1;
1779	return 0;
1780
1781out0:
1782	*stat = 0;
1783	return 0;
1784
1785error0:
1786	return error;
1787}
1788
1789int
1790xfs_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
1845out_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 */
1857STATIC union xfs_btree_key *
1858xfs_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 */
1878int					/* error */
1879xfs_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
2033error0:
2034	return error;
2035}
2036
2037/* Find the high key storage area from a regular key. */
2038union xfs_btree_key *
2039xfs_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 */
2049STATIC void
2050xfs_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 (cur->bc_ops->diff_two_keys(cur, &hkey, &max_hkey)
2071					> 0)
2072				max_hkey = hkey;
2073		}
2074
2075		high = xfs_btree_high_key_from_key(cur, key);
2076		memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
2077	}
2078}
2079
2080/* Determine the low (and high if overlapped) keys of a node block */
2081STATIC void
2082xfs_btree_get_node_keys(
2083	struct xfs_btree_cur	*cur,
2084	struct xfs_btree_block	*block,
2085	union xfs_btree_key	*key)
2086{
2087	union xfs_btree_key	*hkey;
2088	union xfs_btree_key	*max_hkey;
2089	union xfs_btree_key	*high;
2090	int			n;
2091
2092	if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2093		memcpy(key, xfs_btree_key_addr(cur, 1, block),
2094				cur->bc_ops->key_len / 2);
2095
2096		max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2097		for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2098			hkey = xfs_btree_high_key_addr(cur, n, block);
2099			if (cur->bc_ops->diff_two_keys(cur, hkey, max_hkey) > 0)
2100				max_hkey = hkey;
2101		}
2102
2103		high = xfs_btree_high_key_from_key(cur, key);
2104		memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2105	} else {
2106		memcpy(key, xfs_btree_key_addr(cur, 1, block),
2107				cur->bc_ops->key_len);
2108	}
2109}
2110
2111/* Derive the keys for any btree block. */
2112void
2113xfs_btree_get_keys(
2114	struct xfs_btree_cur	*cur,
2115	struct xfs_btree_block	*block,
2116	union xfs_btree_key	*key)
2117{
2118	if (be16_to_cpu(block->bb_level) == 0)
2119		xfs_btree_get_leaf_keys(cur, block, key);
2120	else
2121		xfs_btree_get_node_keys(cur, block, key);
2122}
2123
2124/*
2125 * Decide if we need to update the parent keys of a btree block.  For
2126 * a standard btree this is only necessary if we're updating the first
2127 * record/key.  For an overlapping btree, we must always update the
2128 * keys because the highest key can be in any of the records or keys
2129 * in the block.
2130 */
2131static inline bool
2132xfs_btree_needs_key_update(
2133	struct xfs_btree_cur	*cur,
2134	int			ptr)
2135{
2136	return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1;
2137}
2138
2139/*
2140 * Update the low and high parent keys of the given level, progressing
2141 * towards the root.  If force_all is false, stop if the keys for a given
2142 * level do not need updating.
2143 */
2144STATIC int
2145__xfs_btree_updkeys(
2146	struct xfs_btree_cur	*cur,
2147	int			level,
2148	struct xfs_btree_block	*block,
2149	struct xfs_buf		*bp0,
2150	bool			force_all)
2151{
2152	union xfs_btree_key	key;	/* keys from current level */
2153	union xfs_btree_key	*lkey;	/* keys from the next level up */
2154	union xfs_btree_key	*hkey;
2155	union xfs_btree_key	*nlkey;	/* keys from the next level up */
2156	union xfs_btree_key	*nhkey;
2157	struct xfs_buf		*bp;
2158	int			ptr;
2159
2160	ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2161
2162	/* Exit if there aren't any parent levels to update. */
2163	if (level + 1 >= cur->bc_nlevels)
2164		return 0;
2165
2166	trace_xfs_btree_updkeys(cur, level, bp0);
2167
2168	lkey = &key;
2169	hkey = xfs_btree_high_key_from_key(cur, lkey);
2170	xfs_btree_get_keys(cur, block, lkey);
2171	for (level++; level < cur->bc_nlevels; level++) {
2172#ifdef DEBUG
2173		int		error;
2174#endif
2175		block = xfs_btree_get_block(cur, level, &bp);
2176		trace_xfs_btree_updkeys(cur, level, bp);
2177#ifdef DEBUG
2178		error = xfs_btree_check_block(cur, block, level, bp);
2179		if (error)
2180			return error;
2181#endif
2182		ptr = cur->bc_levels[level].ptr;
2183		nlkey = xfs_btree_key_addr(cur, ptr, block);
2184		nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2185		if (!force_all &&
2186		    !(cur->bc_ops->diff_two_keys(cur, nlkey, lkey) != 0 ||
2187		      cur->bc_ops->diff_two_keys(cur, nhkey, hkey) != 0))
2188			break;
2189		xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2190		xfs_btree_log_keys(cur, bp, ptr, ptr);
2191		if (level + 1 >= cur->bc_nlevels)
2192			break;
2193		xfs_btree_get_node_keys(cur, block, lkey);
2194	}
2195
2196	return 0;
2197}
2198
2199/* Update all the keys from some level in cursor back to the root. */
2200STATIC int
2201xfs_btree_updkeys_force(
2202	struct xfs_btree_cur	*cur,
2203	int			level)
2204{
2205	struct xfs_buf		*bp;
2206	struct xfs_btree_block	*block;
2207
2208	block = xfs_btree_get_block(cur, level, &bp);
2209	return __xfs_btree_updkeys(cur, level, block, bp, true);
2210}
2211
2212/*
2213 * Update the parent keys of the given level, progressing towards the root.
2214 */
2215STATIC int
2216xfs_btree_update_keys(
2217	struct xfs_btree_cur	*cur,
2218	int			level)
2219{
2220	struct xfs_btree_block	*block;
2221	struct xfs_buf		*bp;
2222	union xfs_btree_key	*kp;
2223	union xfs_btree_key	key;
2224	int			ptr;
2225
2226	ASSERT(level >= 0);
2227
2228	block = xfs_btree_get_block(cur, level, &bp);
2229	if (cur->bc_flags & XFS_BTREE_OVERLAPPING)
2230		return __xfs_btree_updkeys(cur, level, block, bp, false);
2231
2232	/*
2233	 * Go up the tree from this level toward the root.
2234	 * At each level, update the key value to the value input.
2235	 * Stop when we reach a level where the cursor isn't pointing
2236	 * at the first entry in the block.
2237	 */
2238	xfs_btree_get_keys(cur, block, &key);
2239	for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2240#ifdef DEBUG
2241		int		error;
2242#endif
2243		block = xfs_btree_get_block(cur, level, &bp);
2244#ifdef DEBUG
2245		error = xfs_btree_check_block(cur, block, level, bp);
2246		if (error)
2247			return error;
2248#endif
2249		ptr = cur->bc_levels[level].ptr;
2250		kp = xfs_btree_key_addr(cur, ptr, block);
2251		xfs_btree_copy_keys(cur, kp, &key, 1);
2252		xfs_btree_log_keys(cur, bp, ptr, ptr);
2253	}
2254
2255	return 0;
2256}
2257
2258/*
2259 * Update the record referred to by cur to the value in the
2260 * given record. This either works (return 0) or gets an
2261 * EFSCORRUPTED error.
2262 */
2263int
2264xfs_btree_update(
2265	struct xfs_btree_cur	*cur,
2266	union xfs_btree_rec	*rec)
2267{
2268	struct xfs_btree_block	*block;
2269	struct xfs_buf		*bp;
2270	int			error;
2271	int			ptr;
2272	union xfs_btree_rec	*rp;
2273
2274	/* Pick up the current block. */
2275	block = xfs_btree_get_block(cur, 0, &bp);
2276
2277#ifdef DEBUG
2278	error = xfs_btree_check_block(cur, block, 0, bp);
2279	if (error)
2280		goto error0;
2281#endif
2282	/* Get the address of the rec to be updated. */
2283	ptr = cur->bc_levels[0].ptr;
2284	rp = xfs_btree_rec_addr(cur, ptr, block);
2285
2286	/* Fill in the new contents and log them. */
2287	xfs_btree_copy_recs(cur, rp, rec, 1);
2288	xfs_btree_log_recs(cur, bp, ptr, ptr);
2289
2290	/*
2291	 * If we are tracking the last record in the tree and
2292	 * we are at the far right edge of the tree, update it.
2293	 */
2294	if (xfs_btree_is_lastrec(cur, block, 0)) {
2295		cur->bc_ops->update_lastrec(cur, block, rec,
2296					    ptr, LASTREC_UPDATE);
2297	}
2298
2299	/* Pass new key value up to our parent. */
2300	if (xfs_btree_needs_key_update(cur, ptr)) {
2301		error = xfs_btree_update_keys(cur, 0);
2302		if (error)
2303			goto error0;
2304	}
2305
2306	return 0;
2307
2308error0:
2309	return error;
2310}
2311
2312/*
2313 * Move 1 record left from cur/level if possible.
2314 * Update cur to reflect the new path.
2315 */
2316STATIC int					/* error */
2317xfs_btree_lshift(
2318	struct xfs_btree_cur	*cur,
2319	int			level,
2320	int			*stat)		/* success/failure */
2321{
2322	struct xfs_buf		*lbp;		/* left buffer pointer */
2323	struct xfs_btree_block	*left;		/* left btree block */
2324	int			lrecs;		/* left record count */
2325	struct xfs_buf		*rbp;		/* right buffer pointer */
2326	struct xfs_btree_block	*right;		/* right btree block */
2327	struct xfs_btree_cur	*tcur;		/* temporary btree cursor */
2328	int			rrecs;		/* right record count */
2329	union xfs_btree_ptr	lptr;		/* left btree pointer */
2330	union xfs_btree_key	*rkp = NULL;	/* right btree key */
2331	union xfs_btree_ptr	*rpp = NULL;	/* right address pointer */
2332	union xfs_btree_rec	*rrp = NULL;	/* right record pointer */
2333	int			error;		/* error return value */
2334	int			i;
2335
2336	if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2337	    level == cur->bc_nlevels - 1)
2338		goto out0;
2339
2340	/* Set up variables for this block as "right". */
2341	right = xfs_btree_get_block(cur, level, &rbp);
2342
2343#ifdef DEBUG
2344	error = xfs_btree_check_block(cur, right, level, rbp);
2345	if (error)
2346		goto error0;
2347#endif
2348
2349	/* If we've got no left sibling then we can't shift an entry left. */
2350	xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2351	if (xfs_btree_ptr_is_null(cur, &lptr))
2352		goto out0;
2353
2354	/*
2355	 * If the cursor entry is the one that would be moved, don't
2356	 * do it... it's too complicated.
2357	 */
2358	if (cur->bc_levels[level].ptr <= 1)
2359		goto out0;
2360
2361	/* Set up the left neighbor as "left". */
2362	error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2363	if (error)
2364		goto error0;
2365
2366	/* If it's full, it can't take another entry. */
2367	lrecs = xfs_btree_get_numrecs(left);
2368	if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2369		goto out0;
2370
2371	rrecs = xfs_btree_get_numrecs(right);
2372
2373	/*
2374	 * We add one entry to the left side and remove one for the right side.
2375	 * Account for it here, the changes will be updated on disk and logged
2376	 * later.
2377	 */
2378	lrecs++;
2379	rrecs--;
2380
2381	XFS_BTREE_STATS_INC(cur, lshift);
2382	XFS_BTREE_STATS_ADD(cur, moves, 1);
2383
2384	/*
2385	 * If non-leaf, copy a key and a ptr to the left block.
2386	 * Log the changes to the left block.
2387	 */
2388	if (level > 0) {
2389		/* It's a non-leaf.  Move keys and pointers. */
2390		union xfs_btree_key	*lkp;	/* left btree key */
2391		union xfs_btree_ptr	*lpp;	/* left address pointer */
2392
2393		lkp = xfs_btree_key_addr(cur, lrecs, left);
2394		rkp = xfs_btree_key_addr(cur, 1, right);
2395
2396		lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2397		rpp = xfs_btree_ptr_addr(cur, 1, right);
2398
2399		error = xfs_btree_debug_check_ptr(cur, rpp, 0, level);
2400		if (error)
2401			goto error0;
2402
2403		xfs_btree_copy_keys(cur, lkp, rkp, 1);
2404		xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2405
2406		xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2407		xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2408
2409		ASSERT(cur->bc_ops->keys_inorder(cur,
2410			xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2411	} else {
2412		/* It's a leaf.  Move records.  */
2413		union xfs_btree_rec	*lrp;	/* left record pointer */
2414
2415		lrp = xfs_btree_rec_addr(cur, lrecs, left);
2416		rrp = xfs_btree_rec_addr(cur, 1, right);
2417
2418		xfs_btree_copy_recs(cur, lrp, rrp, 1);
2419		xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2420
2421		ASSERT(cur->bc_ops->recs_inorder(cur,
2422			xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2423	}
2424
2425	xfs_btree_set_numrecs(left, lrecs);
2426	xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2427
2428	xfs_btree_set_numrecs(right, rrecs);
2429	xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2430
2431	/*
2432	 * Slide the contents of right down one entry.
2433	 */
2434	XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2435	if (level > 0) {
2436		/* It's a nonleaf. operate on keys and ptrs */
2437		for (i = 0; i < rrecs; i++) {
2438			error = xfs_btree_debug_check_ptr(cur, rpp, i + 1, level);
2439			if (error)
2440				goto error0;
2441		}
2442
2443		xfs_btree_shift_keys(cur,
2444				xfs_btree_key_addr(cur, 2, right),
2445				-1, rrecs);
2446		xfs_btree_shift_ptrs(cur,
2447				xfs_btree_ptr_addr(cur, 2, right),
2448				-1, rrecs);
2449
2450		xfs_btree_log_keys(cur, rbp, 1, rrecs);
2451		xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2452	} else {
2453		/* It's a leaf. operate on records */
2454		xfs_btree_shift_recs(cur,
2455			xfs_btree_rec_addr(cur, 2, right),
2456			-1, rrecs);
2457		xfs_btree_log_recs(cur, rbp, 1, rrecs);
2458	}
2459
2460	/*
2461	 * Using a temporary cursor, update the parent key values of the
2462	 * block on the left.
2463	 */
2464	if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2465		error = xfs_btree_dup_cursor(cur, &tcur);
2466		if (error)
2467			goto error0;
2468		i = xfs_btree_firstrec(tcur, level);
2469		if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2470			error = -EFSCORRUPTED;
2471			goto error0;
2472		}
2473
2474		error = xfs_btree_decrement(tcur, level, &i);
2475		if (error)
2476			goto error1;
2477
2478		/* Update the parent high keys of the left block, if needed. */
2479		error = xfs_btree_update_keys(tcur, level);
2480		if (error)
2481			goto error1;
2482
2483		xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2484	}
2485
2486	/* Update the parent keys of the right block. */
2487	error = xfs_btree_update_keys(cur, level);
2488	if (error)
2489		goto error0;
2490
2491	/* Slide the cursor value left one. */
2492	cur->bc_levels[level].ptr--;
2493
2494	*stat = 1;
2495	return 0;
2496
2497out0:
2498	*stat = 0;
2499	return 0;
2500
2501error0:
2502	return error;
2503
2504error1:
2505	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2506	return error;
2507}
2508
2509/*
2510 * Move 1 record right from cur/level if possible.
2511 * Update cur to reflect the new path.
2512 */
2513STATIC int					/* error */
2514xfs_btree_rshift(
2515	struct xfs_btree_cur	*cur,
2516	int			level,
2517	int			*stat)		/* success/failure */
2518{
2519	struct xfs_buf		*lbp;		/* left buffer pointer */
2520	struct xfs_btree_block	*left;		/* left btree block */
2521	struct xfs_buf		*rbp;		/* right buffer pointer */
2522	struct xfs_btree_block	*right;		/* right btree block */
2523	struct xfs_btree_cur	*tcur;		/* temporary btree cursor */
2524	union xfs_btree_ptr	rptr;		/* right block pointer */
2525	union xfs_btree_key	*rkp;		/* right btree key */
2526	int			rrecs;		/* right record count */
2527	int			lrecs;		/* left record count */
2528	int			error;		/* error return value */
2529	int			i;		/* loop counter */
2530
2531	if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2532	    (level == cur->bc_nlevels - 1))
2533		goto out0;
2534
2535	/* Set up variables for this block as "left". */
2536	left = xfs_btree_get_block(cur, level, &lbp);
2537
2538#ifdef DEBUG
2539	error = xfs_btree_check_block(cur, left, level, lbp);
2540	if (error)
2541		goto error0;
2542#endif
2543
2544	/* If we've got no right sibling then we can't shift an entry right. */
2545	xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2546	if (xfs_btree_ptr_is_null(cur, &rptr))
2547		goto out0;
2548
2549	/*
2550	 * If the cursor entry is the one that would be moved, don't
2551	 * do it... it's too complicated.
2552	 */
2553	lrecs = xfs_btree_get_numrecs(left);
2554	if (cur->bc_levels[level].ptr >= lrecs)
2555		goto out0;
2556
2557	/* Set up the right neighbor as "right". */
2558	error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2559	if (error)
2560		goto error0;
2561
2562	/* If it's full, it can't take another entry. */
2563	rrecs = xfs_btree_get_numrecs(right);
2564	if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2565		goto out0;
2566
2567	XFS_BTREE_STATS_INC(cur, rshift);
2568	XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2569
2570	/*
2571	 * Make a hole at the start of the right neighbor block, then
2572	 * copy the last left block entry to the hole.
2573	 */
2574	if (level > 0) {
2575		/* It's a nonleaf. make a hole in the keys and ptrs */
2576		union xfs_btree_key	*lkp;
2577		union xfs_btree_ptr	*lpp;
2578		union xfs_btree_ptr	*rpp;
2579
2580		lkp = xfs_btree_key_addr(cur, lrecs, left);
2581		lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2582		rkp = xfs_btree_key_addr(cur, 1, right);
2583		rpp = xfs_btree_ptr_addr(cur, 1, right);
2584
2585		for (i = rrecs - 1; i >= 0; i--) {
2586			error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
2587			if (error)
2588				goto error0;
2589		}
2590
2591		xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2592		xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2593
2594		error = xfs_btree_debug_check_ptr(cur, lpp, 0, level);
2595		if (error)
2596			goto error0;
2597
2598		/* Now put the new data in, and log it. */
2599		xfs_btree_copy_keys(cur, rkp, lkp, 1);
2600		xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2601
2602		xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2603		xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2604
2605		ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2606			xfs_btree_key_addr(cur, 2, right)));
2607	} else {
2608		/* It's a leaf. make a hole in the records */
2609		union xfs_btree_rec	*lrp;
2610		union xfs_btree_rec	*rrp;
2611
2612		lrp = xfs_btree_rec_addr(cur, lrecs, left);
2613		rrp = xfs_btree_rec_addr(cur, 1, right);
2614
2615		xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2616
2617		/* Now put the new data in, and log it. */
2618		xfs_btree_copy_recs(cur, rrp, lrp, 1);
2619		xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2620	}
2621
2622	/*
2623	 * Decrement and log left's numrecs, bump and log right's numrecs.
2624	 */
2625	xfs_btree_set_numrecs(left, --lrecs);
2626	xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2627
2628	xfs_btree_set_numrecs(right, ++rrecs);
2629	xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2630
2631	/*
2632	 * Using a temporary cursor, update the parent key values of the
2633	 * block on the right.
2634	 */
2635	error = xfs_btree_dup_cursor(cur, &tcur);
2636	if (error)
2637		goto error0;
2638	i = xfs_btree_lastrec(tcur, level);
2639	if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2640		error = -EFSCORRUPTED;
2641		goto error0;
2642	}
2643
2644	error = xfs_btree_increment(tcur, level, &i);
2645	if (error)
2646		goto error1;
2647
2648	/* Update the parent high keys of the left block, if needed. */
2649	if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2650		error = xfs_btree_update_keys(cur, level);
2651		if (error)
2652			goto error1;
2653	}
2654
2655	/* Update the parent keys of the right block. */
2656	error = xfs_btree_update_keys(tcur, level);
2657	if (error)
2658		goto error1;
2659
2660	xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2661
2662	*stat = 1;
2663	return 0;
2664
2665out0:
2666	*stat = 0;
2667	return 0;
2668
2669error0:
2670	return error;
2671
2672error1:
2673	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2674	return error;
2675}
2676
2677/*
2678 * Split cur/level block in half.
2679 * Return new block number and the key to its first
2680 * record (to be inserted into parent).
2681 */
2682STATIC int					/* error */
2683__xfs_btree_split(
2684	struct xfs_btree_cur	*cur,
2685	int			level,
2686	union xfs_btree_ptr	*ptrp,
2687	union xfs_btree_key	*key,
2688	struct xfs_btree_cur	**curp,
2689	int			*stat)		/* success/failure */
2690{
2691	union xfs_btree_ptr	lptr;		/* left sibling block ptr */
2692	struct xfs_buf		*lbp;		/* left buffer pointer */
2693	struct xfs_btree_block	*left;		/* left btree block */
2694	union xfs_btree_ptr	rptr;		/* right sibling block ptr */
2695	struct xfs_buf		*rbp;		/* right buffer pointer */
2696	struct xfs_btree_block	*right;		/* right btree block */
2697	union xfs_btree_ptr	rrptr;		/* right-right sibling ptr */
2698	struct xfs_buf		*rrbp;		/* right-right buffer pointer */
2699	struct xfs_btree_block	*rrblock;	/* right-right btree block */
2700	int			lrecs;
2701	int			rrecs;
2702	int			src_index;
2703	int			error;		/* error return value */
2704	int			i;
2705
2706	XFS_BTREE_STATS_INC(cur, split);
2707
2708	/* Set up left block (current one). */
2709	left = xfs_btree_get_block(cur, level, &lbp);
2710
2711#ifdef DEBUG
2712	error = xfs_btree_check_block(cur, left, level, lbp);
2713	if (error)
2714		goto error0;
2715#endif
2716
2717	xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2718
2719	/* Allocate the new block. If we can't do it, we're toast. Give up. */
2720	error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2721	if (error)
2722		goto error0;
2723	if (*stat == 0)
2724		goto out0;
2725	XFS_BTREE_STATS_INC(cur, alloc);
2726
2727	/* Set up the new block as "right". */
2728	error = xfs_btree_get_buf_block(cur, &rptr, &right, &rbp);
2729	if (error)
2730		goto error0;
2731
2732	/* Fill in the btree header for the new right block. */
2733	xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2734
2735	/*
2736	 * Split the entries between the old and the new block evenly.
2737	 * Make sure that if there's an odd number of entries now, that
2738	 * each new block will have the same number of entries.
2739	 */
2740	lrecs = xfs_btree_get_numrecs(left);
2741	rrecs = lrecs / 2;
2742	if ((lrecs & 1) && cur->bc_levels[level].ptr <= rrecs + 1)
2743		rrecs++;
2744	src_index = (lrecs - rrecs + 1);
2745
2746	XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2747
2748	/* Adjust numrecs for the later get_*_keys() calls. */
2749	lrecs -= rrecs;
2750	xfs_btree_set_numrecs(left, lrecs);
2751	xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2752
2753	/*
2754	 * Copy btree block entries from the left block over to the
2755	 * new block, the right. Update the right block and log the
2756	 * changes.
2757	 */
2758	if (level > 0) {
2759		/* It's a non-leaf.  Move keys and pointers. */
2760		union xfs_btree_key	*lkp;	/* left btree key */
2761		union xfs_btree_ptr	*lpp;	/* left address pointer */
2762		union xfs_btree_key	*rkp;	/* right btree key */
2763		union xfs_btree_ptr	*rpp;	/* right address pointer */
2764
2765		lkp = xfs_btree_key_addr(cur, src_index, left);
2766		lpp = xfs_btree_ptr_addr(cur, src_index, left);
2767		rkp = xfs_btree_key_addr(cur, 1, right);
2768		rpp = xfs_btree_ptr_addr(cur, 1, right);
2769
2770		for (i = src_index; i < rrecs; i++) {
2771			error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
2772			if (error)
2773				goto error0;
2774		}
2775
2776		/* Copy the keys & pointers to the new block. */
2777		xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2778		xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2779
2780		xfs_btree_log_keys(cur, rbp, 1, rrecs);
2781		xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2782
2783		/* Stash the keys of the new block for later insertion. */
2784		xfs_btree_get_node_keys(cur, right, key);
2785	} else {
2786		/* It's a leaf.  Move records.  */
2787		union xfs_btree_rec	*lrp;	/* left record pointer */
2788		union xfs_btree_rec	*rrp;	/* right record pointer */
2789
2790		lrp = xfs_btree_rec_addr(cur, src_index, left);
2791		rrp = xfs_btree_rec_addr(cur, 1, right);
2792
2793		/* Copy records to the new block. */
2794		xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2795		xfs_btree_log_recs(cur, rbp, 1, rrecs);
2796
2797		/* Stash the keys of the new block for later insertion. */
2798		xfs_btree_get_leaf_keys(cur, right, key);
2799	}
2800
2801	/*
2802	 * Find the left block number by looking in the buffer.
2803	 * Adjust sibling pointers.
2804	 */
2805	xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2806	xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2807	xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2808	xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2809
2810	xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2811	xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2812
2813	/*
2814	 * If there's a block to the new block's right, make that block
2815	 * point back to right instead of to left.
2816	 */
2817	if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2818		error = xfs_btree_read_buf_block(cur, &rrptr,
2819							0, &rrblock, &rrbp);
2820		if (error)
2821			goto error0;
2822		xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2823		xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2824	}
2825
2826	/* Update the parent high keys of the left block, if needed. */
2827	if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2828		error = xfs_btree_update_keys(cur, level);
2829		if (error)
2830			goto error0;
2831	}
2832
2833	/*
2834	 * If the cursor is really in the right block, move it there.
2835	 * If it's just pointing past the last entry in left, then we'll
2836	 * insert there, so don't change anything in that case.
2837	 */
2838	if (cur->bc_levels[level].ptr > lrecs + 1) {
2839		xfs_btree_setbuf(cur, level, rbp);
2840		cur->bc_levels[level].ptr -= lrecs;
2841	}
2842	/*
2843	 * If there are more levels, we'll need another cursor which refers
2844	 * the right block, no matter where this cursor was.
2845	 */
2846	if (level + 1 < cur->bc_nlevels) {
2847		error = xfs_btree_dup_cursor(cur, curp);
2848		if (error)
2849			goto error0;
2850		(*curp)->bc_levels[level + 1].ptr++;
2851	}
2852	*ptrp = rptr;
2853	*stat = 1;
2854	return 0;
2855out0:
2856	*stat = 0;
2857	return 0;
2858
2859error0:
2860	return error;
2861}
2862
2863#ifdef __KERNEL__
2864struct xfs_btree_split_args {
2865	struct xfs_btree_cur	*cur;
2866	int			level;
2867	union xfs_btree_ptr	*ptrp;
2868	union xfs_btree_key	*key;
2869	struct xfs_btree_cur	**curp;
2870	int			*stat;		/* success/failure */
2871	int			result;
2872	bool			kswapd;	/* allocation in kswapd context */
2873	struct completion	*done;
2874	struct work_struct	work;
2875};
2876
2877/*
2878 * Stack switching interfaces for allocation
2879 */
2880static void
2881xfs_btree_split_worker(
2882	struct work_struct	*work)
2883{
2884	struct xfs_btree_split_args	*args = container_of(work,
2885						struct xfs_btree_split_args, work);
2886	unsigned long		pflags;
2887	unsigned long		new_pflags = 0;
2888
2889	/*
2890	 * we are in a transaction context here, but may also be doing work
2891	 * in kswapd context, and hence we may need to inherit that state
2892	 * temporarily to ensure that we don't block waiting for memory reclaim
2893	 * in any way.
2894	 */
2895	if (args->kswapd)
2896		new_pflags |= PF_MEMALLOC | PF_KSWAPD;
2897
2898	current_set_flags_nested(&pflags, new_pflags);
2899	xfs_trans_set_context(args->cur->bc_tp);
2900
2901	args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2902					 args->key, args->curp, args->stat);
2903
2904	xfs_trans_clear_context(args->cur->bc_tp);
2905	current_restore_flags_nested(&pflags, new_pflags);
2906
2907	/*
2908	 * Do not access args after complete() has run here. We don't own args
2909	 * and the owner may run and free args before we return here.
2910	 */
2911	complete(args->done);
2912
2913}
2914
2915/*
2916 * BMBT split requests often come in with little stack to work on. Push
2917 * them off to a worker thread so there is lots of stack to use. For the other
2918 * btree types, just call directly to avoid the context switch overhead here.
2919 */
2920STATIC int					/* error */
2921xfs_btree_split(
2922	struct xfs_btree_cur	*cur,
2923	int			level,
2924	union xfs_btree_ptr	*ptrp,
2925	union xfs_btree_key	*key,
2926	struct xfs_btree_cur	**curp,
2927	int			*stat)		/* success/failure */
2928{
2929	struct xfs_btree_split_args	args;
2930	DECLARE_COMPLETION_ONSTACK(done);
2931
2932	if (cur->bc_btnum != XFS_BTNUM_BMAP)
2933		return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2934
2935	args.cur = cur;
2936	args.level = level;
2937	args.ptrp = ptrp;
2938	args.key = key;
2939	args.curp = curp;
2940	args.stat = stat;
2941	args.done = &done;
2942	args.kswapd = current_is_kswapd();
2943	INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2944	queue_work(xfs_alloc_wq, &args.work);
2945	wait_for_completion(&done);
2946	destroy_work_on_stack(&args.work);
2947	return args.result;
2948}
2949#else
2950#define xfs_btree_split	__xfs_btree_split
2951#endif /* __KERNEL__ */
2952
2953
2954/*
2955 * Copy the old inode root contents into a real block and make the
2956 * broot point to it.
2957 */
2958int						/* error */
2959xfs_btree_new_iroot(
2960	struct xfs_btree_cur	*cur,		/* btree cursor */
2961	int			*logflags,	/* logging flags for inode */
2962	int			*stat)		/* return status - 0 fail */
2963{
2964	struct xfs_buf		*cbp;		/* buffer for cblock */
2965	struct xfs_btree_block	*block;		/* btree block */
2966	struct xfs_btree_block	*cblock;	/* child btree block */
2967	union xfs_btree_key	*ckp;		/* child key pointer */
2968	union xfs_btree_ptr	*cpp;		/* child ptr pointer */
2969	union xfs_btree_key	*kp;		/* pointer to btree key */
2970	union xfs_btree_ptr	*pp;		/* pointer to block addr */
2971	union xfs_btree_ptr	nptr;		/* new block addr */
2972	int			level;		/* btree level */
2973	int			error;		/* error return code */
2974	int			i;		/* loop counter */
2975
2976	XFS_BTREE_STATS_INC(cur, newroot);
2977
2978	ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2979
2980	level = cur->bc_nlevels - 1;
2981
2982	block = xfs_btree_get_iroot(cur);
2983	pp = xfs_btree_ptr_addr(cur, 1, block);
2984
2985	/* Allocate the new block. If we can't do it, we're toast. Give up. */
2986	error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
2987	if (error)
2988		goto error0;
2989	if (*stat == 0)
2990		return 0;
2991
2992	XFS_BTREE_STATS_INC(cur, alloc);
2993
2994	/* Copy the root into a real block. */
2995	error = xfs_btree_get_buf_block(cur, &nptr, &cblock, &cbp);
2996	if (error)
2997		goto error0;
2998
2999	/*
3000	 * we can't just memcpy() the root in for CRC enabled btree blocks.
3001	 * In that case have to also ensure the blkno remains correct
3002	 */
3003	memcpy(cblock, block, xfs_btree_block_len(cur));
3004	if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
3005		__be64 bno = cpu_to_be64(xfs_buf_daddr(cbp));
3006		if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
3007			cblock->bb_u.l.bb_blkno = bno;
3008		else
3009			cblock->bb_u.s.bb_blkno = bno;
3010	}
3011
3012	be16_add_cpu(&block->bb_level, 1);
3013	xfs_btree_set_numrecs(block, 1);
3014	cur->bc_nlevels++;
3015	ASSERT(cur->bc_nlevels <= cur->bc_maxlevels);
3016	cur->bc_levels[level + 1].ptr = 1;
3017
3018	kp = xfs_btree_key_addr(cur, 1, block);
3019	ckp = xfs_btree_key_addr(cur, 1, cblock);
3020	xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
3021
3022	cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3023	for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
3024		error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3025		if (error)
3026			goto error0;
3027	}
3028
3029	xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
3030
3031	error = xfs_btree_debug_check_ptr(cur, &nptr, 0, level);
3032	if (error)
3033		goto error0;
3034
3035	xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
3036
3037	xfs_iroot_realloc(cur->bc_ino.ip,
3038			  1 - xfs_btree_get_numrecs(cblock),
3039			  cur->bc_ino.whichfork);
3040
3041	xfs_btree_setbuf(cur, level, cbp);
3042
3043	/*
3044	 * Do all this logging at the end so that
3045	 * the root is at the right level.
3046	 */
3047	xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
3048	xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3049	xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3050
3051	*logflags |=
3052		XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork);
3053	*stat = 1;
3054	return 0;
3055error0:
3056	return error;
3057}
3058
3059/*
3060 * Allocate a new root block, fill it in.
3061 */
3062STATIC int				/* error */
3063xfs_btree_new_root(
3064	struct xfs_btree_cur	*cur,	/* btree cursor */
3065	int			*stat)	/* success/failure */
3066{
3067	struct xfs_btree_block	*block;	/* one half of the old root block */
3068	struct xfs_buf		*bp;	/* buffer containing block */
3069	int			error;	/* error return value */
3070	struct xfs_buf		*lbp;	/* left buffer pointer */
3071	struct xfs_btree_block	*left;	/* left btree block */
3072	struct xfs_buf		*nbp;	/* new (root) buffer */
3073	struct xfs_btree_block	*new;	/* new (root) btree block */
3074	int			nptr;	/* new value for key index, 1 or 2 */
3075	struct xfs_buf		*rbp;	/* right buffer pointer */
3076	struct xfs_btree_block	*right;	/* right btree block */
3077	union xfs_btree_ptr	rptr;
3078	union xfs_btree_ptr	lptr;
3079
3080	XFS_BTREE_STATS_INC(cur, newroot);
3081
3082	/* initialise our start point from the cursor */
3083	cur->bc_ops->init_ptr_from_cur(cur, &rptr);
3084
3085	/* Allocate the new block. If we can't do it, we're toast. Give up. */
3086	error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
3087	if (error)
3088		goto error0;
3089	if (*stat == 0)
3090		goto out0;
3091	XFS_BTREE_STATS_INC(cur, alloc);
3092
3093	/* Set up the new block. */
3094	error = xfs_btree_get_buf_block(cur, &lptr, &new, &nbp);
3095	if (error)
3096		goto error0;
3097
3098	/* Set the root in the holding structure  increasing the level by 1. */
3099	cur->bc_ops->set_root(cur, &lptr, 1);
3100
3101	/*
3102	 * At the previous root level there are now two blocks: the old root,
3103	 * and the new block generated when it was split.  We don't know which
3104	 * one the cursor is pointing at, so we set up variables "left" and
3105	 * "right" for each case.
3106	 */
3107	block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3108
3109#ifdef DEBUG
3110	error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3111	if (error)
3112		goto error0;
3113#endif
3114
3115	xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3116	if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3117		/* Our block is left, pick up the right block. */
3118		lbp = bp;
3119		xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3120		left = block;
3121		error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3122		if (error)
3123			goto error0;
3124		bp = rbp;
3125		nptr = 1;
3126	} else {
3127		/* Our block is right, pick up the left block. */
3128		rbp = bp;
3129		xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3130		right = block;
3131		xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3132		error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3133		if (error)
3134			goto error0;
3135		bp = lbp;
3136		nptr = 2;
3137	}
3138
3139	/* Fill in the new block's btree header and log it. */
3140	xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3141	xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3142	ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3143			!xfs_btree_ptr_is_null(cur, &rptr));
3144
3145	/* Fill in the key data in the new root. */
3146	if (xfs_btree_get_level(left) > 0) {
3147		/*
3148		 * Get the keys for the left block's keys and put them directly
3149		 * in the parent block.  Do the same for the right block.
3150		 */
3151		xfs_btree_get_node_keys(cur, left,
3152				xfs_btree_key_addr(cur, 1, new));
3153		xfs_btree_get_node_keys(cur, right,
3154				xfs_btree_key_addr(cur, 2, new));
3155	} else {
3156		/*
3157		 * Get the keys for the left block's records and put them
3158		 * directly in the parent block.  Do the same for the right
3159		 * block.
3160		 */
3161		xfs_btree_get_leaf_keys(cur, left,
3162			xfs_btree_key_addr(cur, 1, new));
3163		xfs_btree_get_leaf_keys(cur, right,
3164			xfs_btree_key_addr(cur, 2, new));
3165	}
3166	xfs_btree_log_keys(cur, nbp, 1, 2);
3167
3168	/* Fill in the pointer data in the new root. */
3169	xfs_btree_copy_ptrs(cur,
3170		xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3171	xfs_btree_copy_ptrs(cur,
3172		xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3173	xfs_btree_log_ptrs(cur, nbp, 1, 2);
3174
3175	/* Fix up the cursor. */
3176	xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3177	cur->bc_levels[cur->bc_nlevels].ptr = nptr;
3178	cur->bc_nlevels++;
3179	ASSERT(cur->bc_nlevels <= cur->bc_maxlevels);
3180	*stat = 1;
3181	return 0;
3182error0:
3183	return error;
3184out0:
3185	*stat = 0;
3186	return 0;
3187}
3188
3189STATIC int
3190xfs_btree_make_block_unfull(
3191	struct xfs_btree_cur	*cur,	/* btree cursor */
3192	int			level,	/* btree level */
3193	int			numrecs,/* # of recs in block */
3194	int			*oindex,/* old tree index */
3195	int			*index,	/* new tree index */
3196	union xfs_btree_ptr	*nptr,	/* new btree ptr */
3197	struct xfs_btree_cur	**ncur,	/* new btree cursor */
3198	union xfs_btree_key	*key,	/* key of new block */
3199	int			*stat)
3200{
3201	int			error = 0;
3202
3203	if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3204	    level == cur->bc_nlevels - 1) {
3205		struct xfs_inode *ip = cur->bc_ino.ip;
3206
3207		if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3208			/* A root block that can be made bigger. */
3209			xfs_iroot_realloc(ip, 1, cur->bc_ino.whichfork);
3210			*stat = 1;
3211		} else {
3212			/* A root block that needs replacing */
3213			int	logflags = 0;
3214
3215			error = xfs_btree_new_iroot(cur, &logflags, stat);
3216			if (error || *stat == 0)
3217				return error;
3218
3219			xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3220		}
3221
3222		return 0;
3223	}
3224
3225	/* First, try shifting an entry to the right neighbor. */
3226	error = xfs_btree_rshift(cur, level, stat);
3227	if (error || *stat)
3228		return error;
3229
3230	/* Next, try shifting an entry to the left neighbor. */
3231	error = xfs_btree_lshift(cur, level, stat);
3232	if (error)
3233		return error;
3234
3235	if (*stat) {
3236		*oindex = *index = cur->bc_levels[level].ptr;
3237		return 0;
3238	}
3239
3240	/*
3241	 * Next, try splitting the current block in half.
3242	 *
3243	 * If this works we have to re-set our variables because we
3244	 * could be in a different block now.
3245	 */
3246	error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3247	if (error || *stat == 0)
3248		return error;
3249
3250
3251	*index = cur->bc_levels[level].ptr;
3252	return 0;
3253}
3254
3255/*
3256 * Insert one record/level.  Return information to the caller
3257 * allowing the next level up to proceed if necessary.
3258 */
3259STATIC int
3260xfs_btree_insrec(
3261	struct xfs_btree_cur	*cur,	/* btree cursor */
3262	int			level,	/* level to insert record at */
3263	union xfs_btree_ptr	*ptrp,	/* i/o: block number inserted */
3264	union xfs_btree_rec	*rec,	/* record to insert */
3265	union xfs_btree_key	*key,	/* i/o: block key for ptrp */
3266	struct xfs_btree_cur	**curp,	/* output: new cursor replacing cur */
3267	int			*stat)	/* success/failure */
3268{
3269	struct xfs_btree_block	*block;	/* btree block */
3270	struct xfs_buf		*bp;	/* buffer for block */
3271	union xfs_btree_ptr	nptr;	/* new block ptr */
3272	struct xfs_btree_cur	*ncur = NULL;	/* new btree cursor */
3273	union xfs_btree_key	nkey;	/* new block key */
3274	union xfs_btree_key	*lkey;
3275	int			optr;	/* old key/record index */
3276	int			ptr;	/* key/record index */
3277	int			numrecs;/* number of records */
3278	int			error;	/* error return value */
3279	int			i;
3280	xfs_daddr_t		old_bn;
3281
3282	ncur = NULL;
3283	lkey = &nkey;
3284
3285	/*
3286	 * If we have an external root pointer, and we've made it to the
3287	 * root level, allocate a new root block and we're done.
3288	 */
3289	if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3290	    (level >= cur->bc_nlevels)) {
3291		error = xfs_btree_new_root(cur, stat);
3292		xfs_btree_set_ptr_null(cur, ptrp);
3293
3294		return error;
3295	}
3296
3297	/* If we're off the left edge, return failure. */
3298	ptr = cur->bc_levels[level].ptr;
3299	if (ptr == 0) {
3300		*stat = 0;
3301		return 0;
3302	}
3303
3304	optr = ptr;
3305
3306	XFS_BTREE_STATS_INC(cur, insrec);
3307
3308	/* Get pointers to the btree buffer and block. */
3309	block = xfs_btree_get_block(cur, level, &bp);
3310	old_bn = bp ? xfs_buf_daddr(bp) : XFS_BUF_DADDR_NULL;
3311	numrecs = xfs_btree_get_numrecs(block);
3312
3313#ifdef DEBUG
3314	error = xfs_btree_check_block(cur, block, level, bp);
3315	if (error)
3316		goto error0;
3317
3318	/* Check that the new entry is being inserted in the right place. */
3319	if (ptr <= numrecs) {
3320		if (level == 0) {
3321			ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3322				xfs_btree_rec_addr(cur, ptr, block)));
3323		} else {
3324			ASSERT(cur->bc_ops->keys_inorder(cur, key,
3325				xfs_btree_key_addr(cur, ptr, block)));
3326		}
3327	}
3328#endif
3329
3330	/*
3331	 * If the block is full, we can't insert the new entry until we
3332	 * make the block un-full.
3333	 */
3334	xfs_btree_set_ptr_null(cur, &nptr);
3335	if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3336		error = xfs_btree_make_block_unfull(cur, level, numrecs,
3337					&optr, &ptr, &nptr, &ncur, lkey, stat);
3338		if (error || *stat == 0)
3339			goto error0;
3340	}
3341
3342	/*
3343	 * The current block may have changed if the block was
3344	 * previously full and we have just made space in it.
3345	 */
3346	block = xfs_btree_get_block(cur, level, &bp);
3347	numrecs = xfs_btree_get_numrecs(block);
3348
3349#ifdef DEBUG
3350	error = xfs_btree_check_block(cur, block, level, bp);
3351	if (error)
3352		goto error0;
3353#endif
3354
3355	/*
3356	 * At this point we know there's room for our new entry in the block
3357	 * we're pointing at.
3358	 */
3359	XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3360
3361	if (level > 0) {
3362		/* It's a nonleaf. make a hole in the keys and ptrs */
3363		union xfs_btree_key	*kp;
3364		union xfs_btree_ptr	*pp;
3365
3366		kp = xfs_btree_key_addr(cur, ptr, block);
3367		pp = xfs_btree_ptr_addr(cur, ptr, block);
3368
3369		for (i = numrecs - ptr; i >= 0; i--) {
3370			error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3371			if (error)
3372				goto error0;
3373		}
3374
3375		xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3376		xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3377
3378		error = xfs_btree_debug_check_ptr(cur, ptrp, 0, level);
3379		if (error)
3380			goto error0;
3381
3382		/* Now put the new data in, bump numrecs and log it. */
3383		xfs_btree_copy_keys(cur, kp, key, 1);
3384		xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3385		numrecs++;
3386		xfs_btree_set_numrecs(block, numrecs);
3387		xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3388		xfs_btree_log_keys(cur, bp, ptr, numrecs);
3389#ifdef DEBUG
3390		if (ptr < numrecs) {
3391			ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3392				xfs_btree_key_addr(cur, ptr + 1, block)));
3393		}
3394#endif
3395	} else {
3396		/* It's a leaf. make a hole in the records */
3397		union xfs_btree_rec             *rp;
3398
3399		rp = xfs_btree_rec_addr(cur, ptr, block);
3400
3401		xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3402
3403		/* Now put the new data in, bump numrecs and log it. */
3404		xfs_btree_copy_recs(cur, rp, rec, 1);
3405		xfs_btree_set_numrecs(block, ++numrecs);
3406		xfs_btree_log_recs(cur, bp, ptr, numrecs);
3407#ifdef DEBUG
3408		if (ptr < numrecs) {
3409			ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3410				xfs_btree_rec_addr(cur, ptr + 1, block)));
3411		}
3412#endif
3413	}
3414
3415	/* Log the new number of records in the btree header. */
3416	xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3417
3418	/*
3419	 * If we just inserted into a new tree block, we have to
3420	 * recalculate nkey here because nkey is out of date.
3421	 *
3422	 * Otherwise we're just updating an existing block (having shoved
3423	 * some records into the new tree block), so use the regular key
3424	 * update mechanism.
3425	 */
3426	if (bp && xfs_buf_daddr(bp) != old_bn) {
3427		xfs_btree_get_keys(cur, block, lkey);
3428	} else if (xfs_btree_needs_key_update(cur, optr)) {
3429		error = xfs_btree_update_keys(cur, level);
3430		if (error)
3431			goto error0;
3432	}
3433
3434	/*
3435	 * If we are tracking the last record in the tree and
3436	 * we are at the far right edge of the tree, update it.
3437	 */
3438	if (xfs_btree_is_lastrec(cur, block, level)) {
3439		cur->bc_ops->update_lastrec(cur, block, rec,
3440					    ptr, LASTREC_INSREC);
3441	}
3442
3443	/*
3444	 * Return the new block number, if any.
3445	 * If there is one, give back a record value and a cursor too.
3446	 */
3447	*ptrp = nptr;
3448	if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3449		xfs_btree_copy_keys(cur, key, lkey, 1);
3450		*curp = ncur;
3451	}
3452
3453	*stat = 1;
3454	return 0;
3455
3456error0:
3457	if (ncur)
3458		xfs_btree_del_cursor(ncur, error);
3459	return error;
3460}
3461
3462/*
3463 * Insert the record at the point referenced by cur.
3464 *
3465 * A multi-level split of the tree on insert will invalidate the original
3466 * cursor.  All callers of this function should assume that the cursor is
3467 * no longer valid and revalidate it.
3468 */
3469int
3470xfs_btree_insert(
3471	struct xfs_btree_cur	*cur,
3472	int			*stat)
3473{
3474	int			error;	/* error return value */
3475	int			i;	/* result value, 0 for failure */
3476	int			level;	/* current level number in btree */
3477	union xfs_btree_ptr	nptr;	/* new block number (split result) */
3478	struct xfs_btree_cur	*ncur;	/* new cursor (split result) */
3479	struct xfs_btree_cur	*pcur;	/* previous level's cursor */
3480	union xfs_btree_key	bkey;	/* key of block to insert */
3481	union xfs_btree_key	*key;
3482	union xfs_btree_rec	rec;	/* record to insert */
3483
3484	level = 0;
3485	ncur = NULL;
3486	pcur = cur;
3487	key = &bkey;
3488
3489	xfs_btree_set_ptr_null(cur, &nptr);
3490
3491	/* Make a key out of the record data to be inserted, and save it. */
3492	cur->bc_ops->init_rec_from_cur(cur, &rec);
3493	cur->bc_ops->init_key_from_rec(key, &rec);
3494
3495	/*
3496	 * Loop going up the tree, starting at the leaf level.
3497	 * Stop when we don't get a split block, that must mean that
3498	 * the insert is finished with this level.
3499	 */
3500	do {
3501		/*
3502		 * Insert nrec/nptr into this level of the tree.
3503		 * Note if we fail, nptr will be null.
3504		 */
3505		error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3506				&ncur, &i);
3507		if (error) {
3508			if (pcur != cur)
3509				xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3510			goto error0;
3511		}
3512
3513		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3514			error = -EFSCORRUPTED;
3515			goto error0;
3516		}
3517		level++;
3518
3519		/*
3520		 * See if the cursor we just used is trash.
3521		 * Can't trash the caller's cursor, but otherwise we should
3522		 * if ncur is a new cursor or we're about to be done.
3523		 */
3524		if (pcur != cur &&
3525		    (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3526			/* Save the state from the cursor before we trash it */
3527			if (cur->bc_ops->update_cursor)
3528				cur->bc_ops->update_cursor(pcur, cur);
3529			cur->bc_nlevels = pcur->bc_nlevels;
3530			xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3531		}
3532		/* If we got a new cursor, switch to it. */
3533		if (ncur) {
3534			pcur = ncur;
3535			ncur = NULL;
3536		}
3537	} while (!xfs_btree_ptr_is_null(cur, &nptr));
3538
3539	*stat = i;
3540	return 0;
3541error0:
3542	return error;
3543}
3544
3545/*
3546 * Try to merge a non-leaf block back into the inode root.
3547 *
3548 * Note: the killroot names comes from the fact that we're effectively
3549 * killing the old root block.  But because we can't just delete the
3550 * inode we have to copy the single block it was pointing to into the
3551 * inode.
3552 */
3553STATIC int
3554xfs_btree_kill_iroot(
3555	struct xfs_btree_cur	*cur)
3556{
3557	int			whichfork = cur->bc_ino.whichfork;
3558	struct xfs_inode	*ip = cur->bc_ino.ip;
3559	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
3560	struct xfs_btree_block	*block;
3561	struct xfs_btree_block	*cblock;
3562	union xfs_btree_key	*kp;
3563	union xfs_btree_key	*ckp;
3564	union xfs_btree_ptr	*pp;
3565	union xfs_btree_ptr	*cpp;
3566	struct xfs_buf		*cbp;
3567	int			level;
3568	int			index;
3569	int			numrecs;
3570	int			error;
3571#ifdef DEBUG
3572	union xfs_btree_ptr	ptr;
3573#endif
3574	int			i;
3575
3576	ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3577	ASSERT(cur->bc_nlevels > 1);
3578
3579	/*
3580	 * Don't deal with the root block needs to be a leaf case.
3581	 * We're just going to turn the thing back into extents anyway.
3582	 */
3583	level = cur->bc_nlevels - 1;
3584	if (level == 1)
3585		goto out0;
3586
3587	/*
3588	 * Give up if the root has multiple children.
3589	 */
3590	block = xfs_btree_get_iroot(cur);
3591	if (xfs_btree_get_numrecs(block) != 1)
3592		goto out0;
3593
3594	cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3595	numrecs = xfs_btree_get_numrecs(cblock);
3596
3597	/*
3598	 * Only do this if the next level will fit.
3599	 * Then the data must be copied up to the inode,
3600	 * instead of freeing the root you free the next level.
3601	 */
3602	if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3603		goto out0;
3604
3605	XFS_BTREE_STATS_INC(cur, killroot);
3606
3607#ifdef DEBUG
3608	xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3609	ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3610	xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3611	ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3612#endif
3613
3614	index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3615	if (index) {
3616		xfs_iroot_realloc(cur->bc_ino.ip, index,
3617				  cur->bc_ino.whichfork);
3618		block = ifp->if_broot;
3619	}
3620
3621	be16_add_cpu(&block->bb_numrecs, index);
3622	ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3623
3624	kp = xfs_btree_key_addr(cur, 1, block);
3625	ckp = xfs_btree_key_addr(cur, 1, cblock);
3626	xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3627
3628	pp = xfs_btree_ptr_addr(cur, 1, block);
3629	cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3630
3631	for (i = 0; i < numrecs; i++) {
3632		error = xfs_btree_debug_check_ptr(cur, cpp, i, level - 1);
3633		if (error)
3634			return error;
3635	}
3636
3637	xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3638
3639	error = xfs_btree_free_block(cur, cbp);
3640	if (error)
3641		return error;
3642
3643	cur->bc_levels[level - 1].bp = NULL;
3644	be16_add_cpu(&block->bb_level, -1);
3645	xfs_trans_log_inode(cur->bc_tp, ip,
3646		XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork));
3647	cur->bc_nlevels--;
3648out0:
3649	return 0;
3650}
3651
3652/*
3653 * Kill the current root node, and replace it with it's only child node.
3654 */
3655STATIC int
3656xfs_btree_kill_root(
3657	struct xfs_btree_cur	*cur,
3658	struct xfs_buf		*bp,
3659	int			level,
3660	union xfs_btree_ptr	*newroot)
3661{
3662	int			error;
3663
3664	XFS_BTREE_STATS_INC(cur, killroot);
3665
3666	/*
3667	 * Update the root pointer, decreasing the level by 1 and then
3668	 * free the old root.
3669	 */
3670	cur->bc_ops->set_root(cur, newroot, -1);
3671
3672	error = xfs_btree_free_block(cur, bp);
3673	if (error)
3674		return error;
3675
3676	cur->bc_levels[level].bp = NULL;
3677	cur->bc_levels[level].ra = 0;
3678	cur->bc_nlevels--;
3679
3680	return 0;
3681}
3682
3683STATIC int
3684xfs_btree_dec_cursor(
3685	struct xfs_btree_cur	*cur,
3686	int			level,
3687	int			*stat)
3688{
3689	int			error;
3690	int			i;
3691
3692	if (level > 0) {
3693		error = xfs_btree_decrement(cur, level, &i);
3694		if (error)
3695			return error;
3696	}
3697
3698	*stat = 1;
3699	return 0;
3700}
3701
3702/*
3703 * Single level of the btree record deletion routine.
3704 * Delete record pointed to by cur/level.
3705 * Remove the record from its block then rebalance the tree.
3706 * Return 0 for error, 1 for done, 2 to go on to the next level.
3707 */
3708STATIC int					/* error */
3709xfs_btree_delrec(
3710	struct xfs_btree_cur	*cur,		/* btree cursor */
3711	int			level,		/* level removing record from */
3712	int			*stat)		/* fail/done/go-on */
3713{
3714	struct xfs_btree_block	*block;		/* btree block */
3715	union xfs_btree_ptr	cptr;		/* current block ptr */
3716	struct xfs_buf		*bp;		/* buffer for block */
3717	int			error;		/* error return value */
3718	int			i;		/* loop counter */
3719	union xfs_btree_ptr	lptr;		/* left sibling block ptr */
3720	struct xfs_buf		*lbp;		/* left buffer pointer */
3721	struct xfs_btree_block	*left;		/* left btree block */
3722	int			lrecs = 0;	/* left record count */
3723	int			ptr;		/* key/record index */
3724	union xfs_btree_ptr	rptr;		/* right sibling block ptr */
3725	struct xfs_buf		*rbp;		/* right buffer pointer */
3726	struct xfs_btree_block	*right;		/* right btree block */
3727	struct xfs_btree_block	*rrblock;	/* right-right btree block */
3728	struct xfs_buf		*rrbp;		/* right-right buffer pointer */
3729	int			rrecs = 0;	/* right record count */
3730	struct xfs_btree_cur	*tcur;		/* temporary btree cursor */
3731	int			numrecs;	/* temporary numrec count */
3732
3733	tcur = NULL;
3734
3735	/* Get the index of the entry being deleted, check for nothing there. */
3736	ptr = cur->bc_levels[level].ptr;
3737	if (ptr == 0) {
3738		*stat = 0;
3739		return 0;
3740	}
3741
3742	/* Get the buffer & block containing the record or key/ptr. */
3743	block = xfs_btree_get_block(cur, level, &bp);
3744	numrecs = xfs_btree_get_numrecs(block);
3745
3746#ifdef DEBUG
3747	error = xfs_btree_check_block(cur, block, level, bp);
3748	if (error)
3749		goto error0;
3750#endif
3751
3752	/* Fail if we're off the end of the block. */
3753	if (ptr > numrecs) {
3754		*stat = 0;
3755		return 0;
3756	}
3757
3758	XFS_BTREE_STATS_INC(cur, delrec);
3759	XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3760
3761	/* Excise the entries being deleted. */
3762	if (level > 0) {
3763		/* It's a nonleaf. operate on keys and ptrs */
3764		union xfs_btree_key	*lkp;
3765		union xfs_btree_ptr	*lpp;
3766
3767		lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3768		lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3769
3770		for (i = 0; i < numrecs - ptr; i++) {
3771			error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
3772			if (error)
3773				goto error0;
3774		}
3775
3776		if (ptr < numrecs) {
3777			xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3778			xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3779			xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3780			xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3781		}
3782	} else {
3783		/* It's a leaf. operate on records */
3784		if (ptr < numrecs) {
3785			xfs_btree_shift_recs(cur,
3786				xfs_btree_rec_addr(cur, ptr + 1, block),
3787				-1, numrecs - ptr);
3788			xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3789		}
3790	}
3791
3792	/*
3793	 * Decrement and log the number of entries in the block.
3794	 */
3795	xfs_btree_set_numrecs(block, --numrecs);
3796	xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3797
3798	/*
3799	 * If we are tracking the last record in the tree and
3800	 * we are at the far right edge of the tree, update it.
3801	 */
3802	if (xfs_btree_is_lastrec(cur, block, level)) {
3803		cur->bc_ops->update_lastrec(cur, block, NULL,
3804					    ptr, LASTREC_DELREC);
3805	}
3806
3807	/*
3808	 * We're at the root level.  First, shrink the root block in-memory.
3809	 * Try to get rid of the next level down.  If we can't then there's
3810	 * nothing left to do.
3811	 */
3812	if (level == cur->bc_nlevels - 1) {
3813		if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3814			xfs_iroot_realloc(cur->bc_ino.ip, -1,
3815					  cur->bc_ino.whichfork);
3816
3817			error = xfs_btree_kill_iroot(cur);
3818			if (error)
3819				goto error0;
3820
3821			error = xfs_btree_dec_cursor(cur, level, stat);
3822			if (error)
3823				goto error0;
3824			*stat = 1;
3825			return 0;
3826		}
3827
3828		/*
3829		 * If this is the root level, and there's only one entry left,
3830		 * and it's NOT the leaf level, then we can get rid of this
3831		 * level.
3832		 */
3833		if (numrecs == 1 && level > 0) {
3834			union xfs_btree_ptr	*pp;
3835			/*
3836			 * pp is still set to the first pointer in the block.
3837			 * Make it the new root of the btree.
3838			 */
3839			pp = xfs_btree_ptr_addr(cur, 1, block);
3840			error = xfs_btree_kill_root(cur, bp, level, pp);
3841			if (error)
3842				goto error0;
3843		} else if (level > 0) {
3844			error = xfs_btree_dec_cursor(cur, level, stat);
3845			if (error)
3846				goto error0;
3847		}
3848		*stat = 1;
3849		return 0;
3850	}
3851
3852	/*
3853	 * If we deleted the leftmost entry in the block, update the
3854	 * key values above us in the tree.
3855	 */
3856	if (xfs_btree_needs_key_update(cur, ptr)) {
3857		error = xfs_btree_update_keys(cur, level);
3858		if (error)
3859			goto error0;
3860	}
3861
3862	/*
3863	 * If the number of records remaining in the block is at least
3864	 * the minimum, we're done.
3865	 */
3866	if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3867		error = xfs_btree_dec_cursor(cur, level, stat);
3868		if (error)
3869			goto error0;
3870		return 0;
3871	}
3872
3873	/*
3874	 * Otherwise, we have to move some records around to keep the
3875	 * tree balanced.  Look at the left and right sibling blocks to
3876	 * see if we can re-balance by moving only one record.
3877	 */
3878	xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3879	xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3880
3881	if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3882		/*
3883		 * One child of root, need to get a chance to copy its contents
3884		 * into the root and delete it. Can't go up to next level,
3885		 * there's nothing to delete there.
3886		 */
3887		if (xfs_btree_ptr_is_null(cur, &rptr) &&
3888		    xfs_btree_ptr_is_null(cur, &lptr) &&
3889		    level == cur->bc_nlevels - 2) {
3890			error = xfs_btree_kill_iroot(cur);
3891			if (!error)
3892				error = xfs_btree_dec_cursor(cur, level, stat);
3893			if (error)
3894				goto error0;
3895			return 0;
3896		}
3897	}
3898
3899	ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3900	       !xfs_btree_ptr_is_null(cur, &lptr));
3901
3902	/*
3903	 * Duplicate the cursor so our btree manipulations here won't
3904	 * disrupt the next level up.
3905	 */
3906	error = xfs_btree_dup_cursor(cur, &tcur);
3907	if (error)
3908		goto error0;
3909
3910	/*
3911	 * If there's a right sibling, see if it's ok to shift an entry
3912	 * out of it.
3913	 */
3914	if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3915		/*
3916		 * Move the temp cursor to the last entry in the next block.
3917		 * Actually any entry but the first would suffice.
3918		 */
3919		i = xfs_btree_lastrec(tcur, level);
3920		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3921			error = -EFSCORRUPTED;
3922			goto error0;
3923		}
3924
3925		error = xfs_btree_increment(tcur, level, &i);
3926		if (error)
3927			goto error0;
3928		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3929			error = -EFSCORRUPTED;
3930			goto error0;
3931		}
3932
3933		i = xfs_btree_lastrec(tcur, level);
3934		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3935			error = -EFSCORRUPTED;
3936			goto error0;
3937		}
3938
3939		/* Grab a pointer to the block. */
3940		right = xfs_btree_get_block(tcur, level, &rbp);
3941#ifdef DEBUG
3942		error = xfs_btree_check_block(tcur, right, level, rbp);
3943		if (error)
3944			goto error0;
3945#endif
3946		/* Grab the current block number, for future use. */
3947		xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3948
3949		/*
3950		 * If right block is full enough so that removing one entry
3951		 * won't make it too empty, and left-shifting an entry out
3952		 * of right to us works, we're done.
3953		 */
3954		if (xfs_btree_get_numrecs(right) - 1 >=
3955		    cur->bc_ops->get_minrecs(tcur, level)) {
3956			error = xfs_btree_lshift(tcur, level, &i);
3957			if (error)
3958				goto error0;
3959			if (i) {
3960				ASSERT(xfs_btree_get_numrecs(block) >=
3961				       cur->bc_ops->get_minrecs(tcur, level));
3962
3963				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3964				tcur = NULL;
3965
3966				error = xfs_btree_dec_cursor(cur, level, stat);
3967				if (error)
3968					goto error0;
3969				return 0;
3970			}
3971		}
3972
3973		/*
3974		 * Otherwise, grab the number of records in right for
3975		 * future reference, and fix up the temp cursor to point
3976		 * to our block again (last record).
3977		 */
3978		rrecs = xfs_btree_get_numrecs(right);
3979		if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3980			i = xfs_btree_firstrec(tcur, level);
3981			if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3982				error = -EFSCORRUPTED;
3983				goto error0;
3984			}
3985
3986			error = xfs_btree_decrement(tcur, level, &i);
3987			if (error)
3988				goto error0;
3989			if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3990				error = -EFSCORRUPTED;
3991				goto error0;
3992			}
3993		}
3994	}
3995
3996	/*
3997	 * If there's a left sibling, see if it's ok to shift an entry
3998	 * out of it.
3999	 */
4000	if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4001		/*
4002		 * Move the temp cursor to the first entry in the
4003		 * previous block.
4004		 */
4005		i = xfs_btree_firstrec(tcur, level);
4006		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4007			error = -EFSCORRUPTED;
4008			goto error0;
4009		}
4010
4011		error = xfs_btree_decrement(tcur, level, &i);
4012		if (error)
4013			goto error0;
4014		i = xfs_btree_firstrec(tcur, level);
4015		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
4016			error = -EFSCORRUPTED;
4017			goto error0;
4018		}
4019
4020		/* Grab a pointer to the block. */
4021		left = xfs_btree_get_block(tcur, level, &lbp);
4022#ifdef DEBUG
4023		error = xfs_btree_check_block(cur, left, level, lbp);
4024		if (error)
4025			goto error0;
4026#endif
4027		/* Grab the current block number, for future use. */
4028		xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
4029
4030		/*
4031		 * If left block is full enough so that removing one entry
4032		 * won't make it too empty, and right-shifting an entry out
4033		 * of left to us works, we're done.
4034		 */
4035		if (xfs_btree_get_numrecs(left) - 1 >=
4036		    cur->bc_ops->get_minrecs(tcur, level)) {
4037			error = xfs_btree_rshift(tcur, level, &i);
4038			if (error)
4039				goto error0;
4040			if (i) {
4041				ASSERT(xfs_btree_get_numrecs(block) >=
4042				       cur->bc_ops->get_minrecs(tcur, level));
4043				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4044				tcur = NULL;
4045				if (level == 0)
4046					cur->bc_levels[0].ptr++;
4047
4048				*stat = 1;
4049				return 0;
4050			}
4051		}
4052
4053		/*
4054		 * Otherwise, grab the number of records in right for
4055		 * future reference.
4056		 */
4057		lrecs = xfs_btree_get_numrecs(left);
4058	}
4059
4060	/* Delete the temp cursor, we're done with it. */
4061	xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4062	tcur = NULL;
4063
4064	/* If here, we need to do a join to keep the tree balanced. */
4065	ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
4066
4067	if (!xfs_btree_ptr_is_null(cur, &lptr) &&
4068	    lrecs + xfs_btree_get_numrecs(block) <=
4069			cur->bc_ops->get_maxrecs(cur, level)) {
4070		/*
4071		 * Set "right" to be the starting block,
4072		 * "left" to be the left neighbor.
4073		 */
4074		rptr = cptr;
4075		right = block;
4076		rbp = bp;
4077		error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
4078		if (error)
4079			goto error0;
4080
4081	/*
4082	 * If that won't work, see if we can join with the right neighbor block.
4083	 */
4084	} else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
4085		   rrecs + xfs_btree_get_numrecs(block) <=
4086			cur->bc_ops->get_maxrecs(cur, level)) {
4087		/*
4088		 * Set "left" to be the starting block,
4089		 * "right" to be the right neighbor.
4090		 */
4091		lptr = cptr;
4092		left = block;
4093		lbp = bp;
4094		error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4095		if (error)
4096			goto error0;
4097
4098	/*
4099	 * Otherwise, we can't fix the imbalance.
4100	 * Just return.  This is probably a logic error, but it's not fatal.
4101	 */
4102	} else {
4103		error = xfs_btree_dec_cursor(cur, level, stat);
4104		if (error)
4105			goto error0;
4106		return 0;
4107	}
4108
4109	rrecs = xfs_btree_get_numrecs(right);
4110	lrecs = xfs_btree_get_numrecs(left);
4111
4112	/*
4113	 * We're now going to join "left" and "right" by moving all the stuff
4114	 * in "right" to "left" and deleting "right".
4115	 */
4116	XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4117	if (level > 0) {
4118		/* It's a non-leaf.  Move keys and pointers. */
4119		union xfs_btree_key	*lkp;	/* left btree key */
4120		union xfs_btree_ptr	*lpp;	/* left address pointer */
4121		union xfs_btree_key	*rkp;	/* right btree key */
4122		union xfs_btree_ptr	*rpp;	/* right address pointer */
4123
4124		lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4125		lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4126		rkp = xfs_btree_key_addr(cur, 1, right);
4127		rpp = xfs_btree_ptr_addr(cur, 1, right);
4128
4129		for (i = 1; i < rrecs; i++) {
4130			error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
4131			if (error)
4132				goto error0;
4133		}
4134
4135		xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4136		xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4137
4138		xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4139		xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4140	} else {
4141		/* It's a leaf.  Move records.  */
4142		union xfs_btree_rec	*lrp;	/* left record pointer */
4143		union xfs_btree_rec	*rrp;	/* right record pointer */
4144
4145		lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4146		rrp = xfs_btree_rec_addr(cur, 1, right);
4147
4148		xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4149		xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4150	}
4151
4152	XFS_BTREE_STATS_INC(cur, join);
4153
4154	/*
4155	 * Fix up the number of records and right block pointer in the
4156	 * surviving block, and log it.
4157	 */
4158	xfs_btree_set_numrecs(left, lrecs + rrecs);
4159	xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB);
4160	xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4161	xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4162
4163	/* If there is a right sibling, point it to the remaining block. */
4164	xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4165	if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4166		error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4167		if (error)
4168			goto error0;
4169		xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4170		xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4171	}
4172
4173	/* Free the deleted block. */
4174	error = xfs_btree_free_block(cur, rbp);
4175	if (error)
4176		goto error0;
4177
4178	/*
4179	 * If we joined with the left neighbor, set the buffer in the
4180	 * cursor to the left block, and fix up the index.
4181	 */
4182	if (bp != lbp) {
4183		cur->bc_levels[level].bp = lbp;
4184		cur->bc_levels[level].ptr += lrecs;
4185		cur->bc_levels[level].ra = 0;
4186	}
4187	/*
4188	 * If we joined with the right neighbor and there's a level above
4189	 * us, increment the cursor at that level.
4190	 */
4191	else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
4192		   (level + 1 < cur->bc_nlevels)) {
4193		error = xfs_btree_increment(cur, level + 1, &i);
4194		if (error)
4195			goto error0;
4196	}
4197
4198	/*
4199	 * Readjust the ptr at this level if it's not a leaf, since it's
4200	 * still pointing at the deletion point, which makes the cursor
4201	 * inconsistent.  If this makes the ptr 0, the caller fixes it up.
4202	 * We can't use decrement because it would change the next level up.
4203	 */
4204	if (level > 0)
4205		cur->bc_levels[level].ptr--;
4206
4207	/*
4208	 * We combined blocks, so we have to update the parent keys if the
4209	 * btree supports overlapped intervals.  However,
4210	 * bc_levels[level + 1].ptr points to the old block so that the caller
4211	 * knows which record to delete.  Therefore, the caller must be savvy
4212	 * enough to call updkeys for us if we return stat == 2.  The other
4213	 * exit points from this function don't require deletions further up
4214	 * the tree, so they can call updkeys directly.
4215	 */
4216
4217	/* Return value means the next level up has something to do. */
4218	*stat = 2;
4219	return 0;
4220
4221error0:
4222	if (tcur)
4223		xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4224	return error;
4225}
4226
4227/*
4228 * Delete the record pointed to by cur.
4229 * The cursor refers to the place where the record was (could be inserted)
4230 * when the operation returns.
4231 */
4232int					/* error */
4233xfs_btree_delete(
4234	struct xfs_btree_cur	*cur,
4235	int			*stat)	/* success/failure */
4236{
4237	int			error;	/* error return value */
4238	int			level;
4239	int			i;
4240	bool			joined = false;
4241
4242	/*
4243	 * Go up the tree, starting at leaf level.
4244	 *
4245	 * If 2 is returned then a join was done; go to the next level.
4246	 * Otherwise we are done.
4247	 */
4248	for (level = 0, i = 2; i == 2; level++) {
4249		error = xfs_btree_delrec(cur, level, &i);
4250		if (error)
4251			goto error0;
4252		if (i == 2)
4253			joined = true;
4254	}
4255
4256	/*
4257	 * If we combined blocks as part of deleting the record, delrec won't
4258	 * have updated the parent high keys so we have to do that here.
4259	 */
4260	if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) {
4261		error = xfs_btree_updkeys_force(cur, 0);
4262		if (error)
4263			goto error0;
4264	}
4265
4266	if (i == 0) {
4267		for (level = 1; level < cur->bc_nlevels; level++) {
4268			if (cur->bc_levels[level].ptr == 0) {
4269				error = xfs_btree_decrement(cur, level, &i);
4270				if (error)
4271					goto error0;
4272				break;
4273			}
4274		}
4275	}
4276
4277	*stat = i;
4278	return 0;
4279error0:
4280	return error;
4281}
4282
4283/*
4284 * Get the data from the pointed-to record.
4285 */
4286int					/* error */
4287xfs_btree_get_rec(
4288	struct xfs_btree_cur	*cur,	/* btree cursor */
4289	union xfs_btree_rec	**recp,	/* output: btree record */
4290	int			*stat)	/* output: success/failure */
4291{
4292	struct xfs_btree_block	*block;	/* btree block */
4293	struct xfs_buf		*bp;	/* buffer pointer */
4294	int			ptr;	/* record number */
4295#ifdef DEBUG
4296	int			error;	/* error return value */
4297#endif
4298
4299	ptr = cur->bc_levels[0].ptr;
4300	block = xfs_btree_get_block(cur, 0, &bp);
4301
4302#ifdef DEBUG
4303	error = xfs_btree_check_block(cur, block, 0, bp);
4304	if (error)
4305		return error;
4306#endif
4307
4308	/*
4309	 * Off the right end or left end, return failure.
4310	 */
4311	if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4312		*stat = 0;
4313		return 0;
4314	}
4315
4316	/*
4317	 * Point to the record and extract its data.
4318	 */
4319	*recp = xfs_btree_rec_addr(cur, ptr, block);
4320	*stat = 1;
4321	return 0;
4322}
4323
4324/* Visit a block in a btree. */
4325STATIC int
4326xfs_btree_visit_block(
4327	struct xfs_btree_cur		*cur,
4328	int				level,
4329	xfs_btree_visit_blocks_fn	fn,
4330	void				*data)
4331{
4332	struct xfs_btree_block		*block;
4333	struct xfs_buf			*bp;
4334	union xfs_btree_ptr		rptr;
4335	int				error;
4336
4337	/* do right sibling readahead */
4338	xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4339	block = xfs_btree_get_block(cur, level, &bp);
4340
4341	/* process the block */
4342	error = fn(cur, level, data);
4343	if (error)
4344		return error;
4345
4346	/* now read rh sibling block for next iteration */
4347	xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4348	if (xfs_btree_ptr_is_null(cur, &rptr))
4349		return -ENOENT;
4350
4351	/*
4352	 * We only visit blocks once in this walk, so we have to avoid the
4353	 * internal xfs_btree_lookup_get_block() optimisation where it will
4354	 * return the same block without checking if the right sibling points
4355	 * back to us and creates a cyclic reference in the btree.
4356	 */
4357	if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
4358		if (be64_to_cpu(rptr.l) == XFS_DADDR_TO_FSB(cur->bc_mp,
4359							xfs_buf_daddr(bp)))
4360			return -EFSCORRUPTED;
4361	} else {
4362		if (be32_to_cpu(rptr.s) == xfs_daddr_to_agbno(cur->bc_mp,
4363							xfs_buf_daddr(bp)))
4364			return -EFSCORRUPTED;
4365	}
4366	return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4367}
4368
4369
4370/* Visit every block in a btree. */
4371int
4372xfs_btree_visit_blocks(
4373	struct xfs_btree_cur		*cur,
4374	xfs_btree_visit_blocks_fn	fn,
4375	unsigned int			flags,
4376	void				*data)
4377{
4378	union xfs_btree_ptr		lptr;
4379	int				level;
4380	struct xfs_btree_block		*block = NULL;
4381	int				error = 0;
4382
4383	cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4384
4385	/* for each level */
4386	for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4387		/* grab the left hand block */
4388		error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4389		if (error)
4390			return error;
4391
4392		/* readahead the left most block for the next level down */
4393		if (level > 0) {
4394			union xfs_btree_ptr     *ptr;
4395
4396			ptr = xfs_btree_ptr_addr(cur, 1, block);
4397			xfs_btree_readahead_ptr(cur, ptr, 1);
4398
4399			/* save for the next iteration of the loop */
4400			xfs_btree_copy_ptrs(cur, &lptr, ptr, 1);
4401
4402			if (!(flags & XFS_BTREE_VISIT_LEAVES))
4403				continue;
4404		} else if (!(flags & XFS_BTREE_VISIT_RECORDS)) {
4405			continue;
4406		}
4407
4408		/* for each buffer in the level */
4409		do {
4410			error = xfs_btree_visit_block(cur, level, fn, data);
4411		} while (!error);
4412
4413		if (error != -ENOENT)
4414			return error;
4415	}
4416
4417	return 0;
4418}
4419
4420/*
4421 * Change the owner of a btree.
4422 *
4423 * The mechanism we use here is ordered buffer logging. Because we don't know
4424 * how many buffers were are going to need to modify, we don't really want to
4425 * have to make transaction reservations for the worst case of every buffer in a
4426 * full size btree as that may be more space that we can fit in the log....
4427 *
4428 * We do the btree walk in the most optimal manner possible - we have sibling
4429 * pointers so we can just walk all the blocks on each level from left to right
4430 * in a single pass, and then move to the next level and do the same. We can
4431 * also do readahead on the sibling pointers to get IO moving more quickly,
4432 * though for slow disks this is unlikely to make much difference to performance
4433 * as the amount of CPU work we have to do before moving to the next block is
4434 * relatively small.
4435 *
4436 * For each btree block that we load, modify the owner appropriately, set the
4437 * buffer as an ordered buffer and log it appropriately. We need to ensure that
4438 * we mark the region we change dirty so that if the buffer is relogged in
4439 * a subsequent transaction the changes we make here as an ordered buffer are
4440 * correctly relogged in that transaction.  If we are in recovery context, then
4441 * just queue the modified buffer as delayed write buffer so the transaction
4442 * recovery completion writes the changes to disk.
4443 */
4444struct xfs_btree_block_change_owner_info {
4445	uint64_t		new_owner;
4446	struct list_head	*buffer_list;
4447};
4448
4449static int
4450xfs_btree_block_change_owner(
4451	struct xfs_btree_cur	*cur,
4452	int			level,
4453	void			*data)
4454{
4455	struct xfs_btree_block_change_owner_info	*bbcoi = data;
4456	struct xfs_btree_block	*block;
4457	struct xfs_buf		*bp;
4458
4459	/* modify the owner */
4460	block = xfs_btree_get_block(cur, level, &bp);
4461	if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
4462		if (block->bb_u.l.bb_owner == cpu_to_be64(bbcoi->new_owner))
4463			return 0;
4464		block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4465	} else {
4466		if (block->bb_u.s.bb_owner == cpu_to_be32(bbcoi->new_owner))
4467			return 0;
4468		block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4469	}
4470
4471	/*
4472	 * If the block is a root block hosted in an inode, we might not have a
4473	 * buffer pointer here and we shouldn't attempt to log the change as the
4474	 * information is already held in the inode and discarded when the root
4475	 * block is formatted into the on-disk inode fork. We still change it,
4476	 * though, so everything is consistent in memory.
4477	 */
4478	if (!bp) {
4479		ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4480		ASSERT(level == cur->bc_nlevels - 1);
4481		return 0;
4482	}
4483
4484	if (cur->bc_tp) {
4485		if (!xfs_trans_ordered_buf(cur->bc_tp, bp)) {
4486			xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4487			return -EAGAIN;
4488		}
4489	} else {
4490		xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4491	}
4492
4493	return 0;
4494}
4495
4496int
4497xfs_btree_change_owner(
4498	struct xfs_btree_cur	*cur,
4499	uint64_t		new_owner,
4500	struct list_head	*buffer_list)
4501{
4502	struct xfs_btree_block_change_owner_info	bbcoi;
4503
4504	bbcoi.new_owner = new_owner;
4505	bbcoi.buffer_list = buffer_list;
4506
4507	return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4508			XFS_BTREE_VISIT_ALL, &bbcoi);
4509}
4510
4511/* Verify the v5 fields of a long-format btree block. */
4512xfs_failaddr_t
4513xfs_btree_lblock_v5hdr_verify(
4514	struct xfs_buf		*bp,
4515	uint64_t		owner)
4516{
4517	struct xfs_mount	*mp = bp->b_mount;
4518	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
4519
4520	if (!xfs_has_crc(mp))
4521		return __this_address;
4522	if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
4523		return __this_address;
4524	if (block->bb_u.l.bb_blkno != cpu_to_be64(xfs_buf_daddr(bp)))
4525		return __this_address;
4526	if (owner != XFS_RMAP_OWN_UNKNOWN &&
4527	    be64_to_cpu(block->bb_u.l.bb_owner) != owner)
4528		return __this_address;
4529	return NULL;
4530}
4531
4532/* Verify a long-format btree block. */
4533xfs_failaddr_t
4534xfs_btree_lblock_verify(
4535	struct xfs_buf		*bp,
4536	unsigned int		max_recs)
4537{
4538	struct xfs_mount	*mp = bp->b_mount;
4539	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
4540	xfs_fsblock_t		fsb;
4541	xfs_failaddr_t		fa;
4542
4543	/* numrecs verification */
4544	if (be16_to_cpu(block->bb_numrecs) > max_recs)
4545		return __this_address;
4546
4547	/* sibling pointer verification */
4548	fsb = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
4549	fa = xfs_btree_check_lblock_siblings(mp, NULL, -1, fsb,
4550			block->bb_u.l.bb_leftsib);
4551	if (!fa)
4552		fa = xfs_btree_check_lblock_siblings(mp, NULL, -1, fsb,
4553				block->bb_u.l.bb_rightsib);
4554	return fa;
 
4555}
4556
4557/**
4558 * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4559 *				      btree block
4560 *
4561 * @bp: buffer containing the btree block
4562 */
4563xfs_failaddr_t
4564xfs_btree_sblock_v5hdr_verify(
4565	struct xfs_buf		*bp)
4566{
4567	struct xfs_mount	*mp = bp->b_mount;
4568	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
4569	struct xfs_perag	*pag = bp->b_pag;
4570
4571	if (!xfs_has_crc(mp))
4572		return __this_address;
4573	if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4574		return __this_address;
4575	if (block->bb_u.s.bb_blkno != cpu_to_be64(xfs_buf_daddr(bp)))
4576		return __this_address;
4577	if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4578		return __this_address;
4579	return NULL;
4580}
4581
4582/**
4583 * xfs_btree_sblock_verify() -- verify a short-format btree block
4584 *
4585 * @bp: buffer containing the btree block
4586 * @max_recs: maximum records allowed in this btree node
4587 */
4588xfs_failaddr_t
4589xfs_btree_sblock_verify(
4590	struct xfs_buf		*bp,
4591	unsigned int		max_recs)
4592{
4593	struct xfs_mount	*mp = bp->b_mount;
4594	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
4595	xfs_agblock_t		agbno;
4596	xfs_failaddr_t		fa;
4597
4598	/* numrecs verification */
4599	if (be16_to_cpu(block->bb_numrecs) > max_recs)
4600		return __this_address;
4601
4602	/* sibling pointer verification */
4603	agbno = xfs_daddr_to_agbno(mp, xfs_buf_daddr(bp));
4604	fa = xfs_btree_check_sblock_siblings(bp->b_pag, NULL, -1, agbno,
4605			block->bb_u.s.bb_leftsib);
4606	if (!fa)
4607		fa = xfs_btree_check_sblock_siblings(bp->b_pag, NULL, -1, agbno,
4608				block->bb_u.s.bb_rightsib);
4609	return fa;
4610}
4611
4612/*
4613 * For the given limits on leaf and keyptr records per block, calculate the
4614 * height of the tree needed to index the number of leaf records.
4615 */
4616unsigned int
4617xfs_btree_compute_maxlevels(
4618	const unsigned int	*limits,
4619	unsigned long long	records)
4620{
4621	unsigned long long	level_blocks = howmany_64(records, limits[0]);
4622	unsigned int		height = 1;
4623
4624	while (level_blocks > 1) {
4625		level_blocks = howmany_64(level_blocks, limits[1]);
4626		height++;
4627	}
4628
4629	return height;
4630}
4631
4632/*
4633 * For the given limits on leaf and keyptr records per block, calculate the
4634 * number of blocks needed to index the given number of leaf records.
4635 */
4636unsigned long long
4637xfs_btree_calc_size(
4638	const unsigned int	*limits,
4639	unsigned long long	records)
4640{
4641	unsigned long long	level_blocks = howmany_64(records, limits[0]);
4642	unsigned long long	blocks = level_blocks;
4643
4644	while (level_blocks > 1) {
4645		level_blocks = howmany_64(level_blocks, limits[1]);
4646		blocks += level_blocks;
4647	}
4648
4649	return blocks;
4650}
4651
4652/*
4653 * Given a number of available blocks for the btree to consume with records and
4654 * pointers, calculate the height of the tree needed to index all the records
4655 * that space can hold based on the number of pointers each interior node
4656 * holds.
4657 *
4658 * We start by assuming a single level tree consumes a single block, then track
4659 * the number of blocks each node level consumes until we no longer have space
4660 * to store the next node level. At this point, we are indexing all the leaf
4661 * blocks in the space, and there's no more free space to split the tree any
4662 * further. That's our maximum btree height.
4663 */
4664unsigned int
4665xfs_btree_space_to_height(
4666	const unsigned int	*limits,
4667	unsigned long long	leaf_blocks)
4668{
4669	/*
4670	 * The root btree block can have fewer than minrecs pointers in it
4671	 * because the tree might not be big enough to require that amount of
4672	 * fanout. Hence it has a minimum size of 2 pointers, not limits[1].
4673	 */
4674	unsigned long long	node_blocks = 2;
4675	unsigned long long	blocks_left = leaf_blocks - 1;
4676	unsigned int		height = 1;
4677
4678	if (leaf_blocks < 1)
4679		return 0;
4680
4681	while (node_blocks < blocks_left) {
4682		blocks_left -= node_blocks;
4683		node_blocks *= limits[1];
4684		height++;
4685	}
4686
4687	return height;
4688}
4689
4690/*
4691 * Query a regular btree for all records overlapping a given interval.
4692 * Start with a LE lookup of the key of low_rec and return all records
4693 * until we find a record with a key greater than the key of high_rec.
4694 */
4695STATIC int
4696xfs_btree_simple_query_range(
4697	struct xfs_btree_cur		*cur,
4698	const union xfs_btree_key	*low_key,
4699	const union xfs_btree_key	*high_key,
4700	xfs_btree_query_range_fn	fn,
4701	void				*priv)
4702{
4703	union xfs_btree_rec		*recp;
4704	union xfs_btree_key		rec_key;
4705	int64_t				diff;
4706	int				stat;
4707	bool				firstrec = true;
4708	int				error;
4709
4710	ASSERT(cur->bc_ops->init_high_key_from_rec);
4711	ASSERT(cur->bc_ops->diff_two_keys);
4712
4713	/*
4714	 * Find the leftmost record.  The btree cursor must be set
4715	 * to the low record used to generate low_key.
4716	 */
4717	stat = 0;
4718	error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4719	if (error)
4720		goto out;
4721
4722	/* Nothing?  See if there's anything to the right. */
4723	if (!stat) {
4724		error = xfs_btree_increment(cur, 0, &stat);
4725		if (error)
4726			goto out;
4727	}
4728
4729	while (stat) {
4730		/* Find the record. */
4731		error = xfs_btree_get_rec(cur, &recp, &stat);
4732		if (error || !stat)
4733			break;
4734
4735		/* Skip if high_key(rec) < low_key. */
4736		if (firstrec) {
4737			cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4738			firstrec = false;
4739			diff = cur->bc_ops->diff_two_keys(cur, low_key,
4740					&rec_key);
4741			if (diff > 0)
4742				goto advloop;
4743		}
4744
4745		/* Stop if high_key < low_key(rec). */
4746		cur->bc_ops->init_key_from_rec(&rec_key, recp);
4747		diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key);
4748		if (diff > 0)
4749			break;
4750
4751		/* Callback */
4752		error = fn(cur, recp, priv);
4753		if (error)
4754			break;
4755
4756advloop:
4757		/* Move on to the next record. */
4758		error = xfs_btree_increment(cur, 0, &stat);
4759		if (error)
4760			break;
4761	}
4762
4763out:
4764	return error;
4765}
4766
4767/*
4768 * Query an overlapped interval btree for all records overlapping a given
4769 * interval.  This function roughly follows the algorithm given in
4770 * "Interval Trees" of _Introduction to Algorithms_, which is section
4771 * 14.3 in the 2nd and 3rd editions.
4772 *
4773 * First, generate keys for the low and high records passed in.
4774 *
4775 * For any leaf node, generate the high and low keys for the record.
4776 * If the record keys overlap with the query low/high keys, pass the
4777 * record to the function iterator.
4778 *
4779 * For any internal node, compare the low and high keys of each
4780 * pointer against the query low/high keys.  If there's an overlap,
4781 * follow the pointer.
4782 *
4783 * As an optimization, we stop scanning a block when we find a low key
4784 * that is greater than the query's high key.
4785 */
4786STATIC int
4787xfs_btree_overlapped_query_range(
4788	struct xfs_btree_cur		*cur,
4789	const union xfs_btree_key	*low_key,
4790	const union xfs_btree_key	*high_key,
4791	xfs_btree_query_range_fn	fn,
4792	void				*priv)
4793{
4794	union xfs_btree_ptr		ptr;
4795	union xfs_btree_ptr		*pp;
4796	union xfs_btree_key		rec_key;
4797	union xfs_btree_key		rec_hkey;
4798	union xfs_btree_key		*lkp;
4799	union xfs_btree_key		*hkp;
4800	union xfs_btree_rec		*recp;
4801	struct xfs_btree_block		*block;
4802	int64_t				ldiff;
4803	int64_t				hdiff;
4804	int				level;
4805	struct xfs_buf			*bp;
4806	int				i;
4807	int				error;
4808
4809	/* Load the root of the btree. */
4810	level = cur->bc_nlevels - 1;
4811	cur->bc_ops->init_ptr_from_cur(cur, &ptr);
4812	error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4813	if (error)
4814		return error;
4815	xfs_btree_get_block(cur, level, &bp);
4816	trace_xfs_btree_overlapped_query_range(cur, level, bp);
4817#ifdef DEBUG
4818	error = xfs_btree_check_block(cur, block, level, bp);
4819	if (error)
4820		goto out;
4821#endif
4822	cur->bc_levels[level].ptr = 1;
4823
4824	while (level < cur->bc_nlevels) {
4825		block = xfs_btree_get_block(cur, level, &bp);
4826
4827		/* End of node, pop back towards the root. */
4828		if (cur->bc_levels[level].ptr >
4829					be16_to_cpu(block->bb_numrecs)) {
4830pop_up:
4831			if (level < cur->bc_nlevels - 1)
4832				cur->bc_levels[level + 1].ptr++;
4833			level++;
4834			continue;
4835		}
4836
4837		if (level == 0) {
4838			/* Handle a leaf node. */
4839			recp = xfs_btree_rec_addr(cur, cur->bc_levels[0].ptr,
4840					block);
4841
4842			cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
4843			ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
4844					low_key);
4845
4846			cur->bc_ops->init_key_from_rec(&rec_key, recp);
4847			hdiff = cur->bc_ops->diff_two_keys(cur, high_key,
4848					&rec_key);
4849
4850			/*
4851			 * If (record's high key >= query's low key) and
4852			 *    (query's high key >= record's low key), then
4853			 * this record overlaps the query range; callback.
4854			 */
4855			if (ldiff >= 0 && hdiff >= 0) {
4856				error = fn(cur, recp, priv);
4857				if (error)
4858					break;
4859			} else if (hdiff < 0) {
4860				/* Record is larger than high key; pop. */
4861				goto pop_up;
4862			}
4863			cur->bc_levels[level].ptr++;
4864			continue;
4865		}
4866
4867		/* Handle an internal node. */
4868		lkp = xfs_btree_key_addr(cur, cur->bc_levels[level].ptr, block);
4869		hkp = xfs_btree_high_key_addr(cur, cur->bc_levels[level].ptr,
4870				block);
4871		pp = xfs_btree_ptr_addr(cur, cur->bc_levels[level].ptr, block);
4872
4873		ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
4874		hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
4875
4876		/*
4877		 * If (pointer's high key >= query's low key) and
4878		 *    (query's high key >= pointer's low key), then
4879		 * this record overlaps the query range; follow pointer.
4880		 */
4881		if (ldiff >= 0 && hdiff >= 0) {
4882			level--;
4883			error = xfs_btree_lookup_get_block(cur, level, pp,
4884					&block);
4885			if (error)
4886				goto out;
4887			xfs_btree_get_block(cur, level, &bp);
4888			trace_xfs_btree_overlapped_query_range(cur, level, bp);
4889#ifdef DEBUG
4890			error = xfs_btree_check_block(cur, block, level, bp);
4891			if (error)
4892				goto out;
4893#endif
4894			cur->bc_levels[level].ptr = 1;
4895			continue;
4896		} else if (hdiff < 0) {
4897			/* The low key is larger than the upper range; pop. */
4898			goto pop_up;
4899		}
4900		cur->bc_levels[level].ptr++;
4901	}
4902
4903out:
4904	/*
4905	 * If we don't end this function with the cursor pointing at a record
4906	 * block, a subsequent non-error cursor deletion will not release
4907	 * node-level buffers, causing a buffer leak.  This is quite possible
4908	 * with a zero-results range query, so release the buffers if we
4909	 * failed to return any results.
4910	 */
4911	if (cur->bc_levels[0].bp == NULL) {
4912		for (i = 0; i < cur->bc_nlevels; i++) {
4913			if (cur->bc_levels[i].bp) {
4914				xfs_trans_brelse(cur->bc_tp,
4915						cur->bc_levels[i].bp);
4916				cur->bc_levels[i].bp = NULL;
4917				cur->bc_levels[i].ptr = 0;
4918				cur->bc_levels[i].ra = 0;
4919			}
4920		}
4921	}
4922
4923	return error;
4924}
4925
4926/*
4927 * Query a btree for all records overlapping a given interval of keys.  The
4928 * supplied function will be called with each record found; return one of the
4929 * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4930 * code.  This function returns -ECANCELED, zero, or a negative error code.
4931 */
4932int
4933xfs_btree_query_range(
4934	struct xfs_btree_cur		*cur,
4935	const union xfs_btree_irec	*low_rec,
4936	const union xfs_btree_irec	*high_rec,
4937	xfs_btree_query_range_fn	fn,
4938	void				*priv)
4939{
4940	union xfs_btree_rec		rec;
4941	union xfs_btree_key		low_key;
4942	union xfs_btree_key		high_key;
4943
4944	/* Find the keys of both ends of the interval. */
4945	cur->bc_rec = *high_rec;
4946	cur->bc_ops->init_rec_from_cur(cur, &rec);
4947	cur->bc_ops->init_key_from_rec(&high_key, &rec);
4948
4949	cur->bc_rec = *low_rec;
4950	cur->bc_ops->init_rec_from_cur(cur, &rec);
4951	cur->bc_ops->init_key_from_rec(&low_key, &rec);
4952
4953	/* Enforce low key < high key. */
4954	if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0)
4955		return -EINVAL;
4956
4957	if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
4958		return xfs_btree_simple_query_range(cur, &low_key,
4959				&high_key, fn, priv);
4960	return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
4961			fn, priv);
4962}
4963
4964/* Query a btree for all records. */
4965int
4966xfs_btree_query_all(
4967	struct xfs_btree_cur		*cur,
4968	xfs_btree_query_range_fn	fn,
4969	void				*priv)
4970{
4971	union xfs_btree_key		low_key;
4972	union xfs_btree_key		high_key;
4973
4974	memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
4975	memset(&low_key, 0, sizeof(low_key));
4976	memset(&high_key, 0xFF, sizeof(high_key));
4977
4978	return xfs_btree_simple_query_range(cur, &low_key, &high_key, fn, priv);
4979}
4980
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4981static int
4982xfs_btree_count_blocks_helper(
4983	struct xfs_btree_cur	*cur,
4984	int			level,
4985	void			*data)
4986{
4987	xfs_extlen_t		*blocks = data;
4988	(*blocks)++;
4989
4990	return 0;
4991}
4992
4993/* Count the blocks in a btree and return the result in *blocks. */
4994int
4995xfs_btree_count_blocks(
4996	struct xfs_btree_cur	*cur,
4997	xfs_extlen_t		*blocks)
4998{
4999	*blocks = 0;
5000	return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
5001			XFS_BTREE_VISIT_ALL, blocks);
5002}
5003
5004/* Compare two btree pointers. */
5005int64_t
5006xfs_btree_diff_two_ptrs(
5007	struct xfs_btree_cur		*cur,
5008	const union xfs_btree_ptr	*a,
5009	const union xfs_btree_ptr	*b)
5010{
5011	if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
5012		return (int64_t)be64_to_cpu(a->l) - be64_to_cpu(b->l);
5013	return (int64_t)be32_to_cpu(a->s) - be32_to_cpu(b->s);
5014}
5015
5016/* If there's an extent, we're done. */
5017STATIC int
5018xfs_btree_has_record_helper(
5019	struct xfs_btree_cur		*cur,
5020	const union xfs_btree_rec	*rec,
5021	void				*priv)
5022{
5023	return -ECANCELED;
5024}
5025
5026/* Is there a record covering a given range of keys? */
5027int
5028xfs_btree_has_record(
5029	struct xfs_btree_cur		*cur,
5030	const union xfs_btree_irec	*low,
5031	const union xfs_btree_irec	*high,
5032	bool				*exists)
5033{
5034	int				error;
5035
5036	error = xfs_btree_query_range(cur, low, high,
5037			&xfs_btree_has_record_helper, NULL);
5038	if (error == -ECANCELED) {
5039		*exists = true;
5040		return 0;
5041	}
5042	*exists = false;
5043	return error;
5044}
5045
5046/* Are there more records in this btree? */
5047bool
5048xfs_btree_has_more_records(
5049	struct xfs_btree_cur	*cur)
5050{
5051	struct xfs_btree_block	*block;
5052	struct xfs_buf		*bp;
5053
5054	block = xfs_btree_get_block(cur, 0, &bp);
5055
5056	/* There are still records in this block. */
5057	if (cur->bc_levels[0].ptr < xfs_btree_get_numrecs(block))
5058		return true;
5059
5060	/* There are more record blocks. */
5061	if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
5062		return block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK);
5063	else
5064		return block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK);
5065}
5066
5067/* Set up all the btree cursor caches. */
5068int __init
5069xfs_btree_init_cur_caches(void)
5070{
5071	int		error;
5072
5073	error = xfs_allocbt_init_cur_cache();
5074	if (error)
5075		return error;
5076	error = xfs_inobt_init_cur_cache();
5077	if (error)
5078		goto err;
5079	error = xfs_bmbt_init_cur_cache();
5080	if (error)
5081		goto err;
5082	error = xfs_rmapbt_init_cur_cache();
5083	if (error)
5084		goto err;
5085	error = xfs_refcountbt_init_cur_cache();
5086	if (error)
5087		goto err;
5088
5089	return 0;
5090err:
5091	xfs_btree_destroy_cur_caches();
5092	return error;
5093}
5094
5095/* Destroy all the btree cursor caches, if they've been allocated. */
5096void
5097xfs_btree_destroy_cur_caches(void)
5098{
5099	xfs_allocbt_destroy_cur_cache();
5100	xfs_inobt_destroy_cur_cache();
5101	xfs_bmbt_destroy_cur_cache();
5102	xfs_rmapbt_destroy_cur_cache();
5103	xfs_refcountbt_destroy_cur_cache();
5104}