1/*
   2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3 * Copyright (c) 2013 Red Hat, Inc.
   4 * All Rights Reserved.
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License as
   8 * published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it would be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write the Free Software Foundation,
  17 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18 */
  19#include "xfs.h"
  20#include "xfs_fs.h"
  21#include "xfs_format.h"
  22#include "xfs_log_format.h"
  23#include "xfs_trans_resv.h"
  24#include "xfs_mount.h"
  25#include "xfs_da_format.h"
  26#include "xfs_da_btree.h"
  27#include "xfs_inode.h"
  28#include "xfs_bmap.h"
  29#include "xfs_dir2.h"
  30#include "xfs_dir2_priv.h"
  31#include "xfs_error.h"
  32#include "xfs_trace.h"
  33#include "xfs_trans.h"
  34#include "xfs_buf_item.h"
  35#include "xfs_cksum.h"
  36#include "xfs_log.h"
  37
  38/*
  39 * Local function declarations.
  40 */
  41static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
  42				    int *indexp, struct xfs_buf **dbpp);
  43static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
  44				    struct xfs_buf *bp, int first, int last);
  45static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
  46				   struct xfs_buf *bp);
  47
  48/*
  49 * Check the internal consistency of a leaf1 block.
  50 * Pop an assert if something is wrong.
  51 */
  52#ifdef DEBUG
  53#define	xfs_dir3_leaf_check(dp, bp) \
  54do { \
  55	if (!xfs_dir3_leaf1_check((dp), (bp))) \
  56		ASSERT(0); \
  57} while (0);
  58
  59STATIC bool
  60xfs_dir3_leaf1_check(
  61	struct xfs_inode	*dp,
  62	struct xfs_buf		*bp)
  63{
  64	struct xfs_dir2_leaf	*leaf = bp->b_addr;
  65	struct xfs_dir3_icleaf_hdr leafhdr;
  66
  67	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
  68
  69	if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
  70		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  71		if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  72			return false;
  73	} else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
  74		return false;
  75
  76	return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
  77}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  78#else
  79#define	xfs_dir3_leaf_check(dp, bp)
  80#endif
  81
  82bool
  83xfs_dir3_leaf_check_int(
  84	struct xfs_mount	*mp,
  85	struct xfs_inode	*dp,
  86	struct xfs_dir3_icleaf_hdr *hdr,
  87	struct xfs_dir2_leaf	*leaf)
  88{
  89	struct xfs_dir2_leaf_entry *ents;
  90	xfs_dir2_leaf_tail_t	*ltp;
  91	int			stale;
  92	int			i;
  93	const struct xfs_dir_ops *ops;
  94	struct xfs_dir3_icleaf_hdr leafhdr;
  95	struct xfs_da_geometry	*geo = mp->m_dir_geo;
  96
  97	/*
  98	 * we can be passed a null dp here from a verifier, so we need to go the
  99	 * hard way to get them.
 100	 */
 101	ops = xfs_dir_get_ops(mp, dp);
 102
 103	if (!hdr) {
 104		ops->leaf_hdr_from_disk(&leafhdr, leaf);
 105		hdr = &leafhdr;
 106	}
 107
 108	ents = ops->leaf_ents_p(leaf);
 109	ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 110
 111	/*
 112	 * XXX (dgc): This value is not restrictive enough.
 113	 * Should factor in the size of the bests table as well.
 114	 * We can deduce a value for that from di_size.
 115	 */
 116	if (hdr->count > ops->leaf_max_ents(geo))
 117		return false;
 118
 119	/* Leaves and bests don't overlap in leaf format. */
 120	if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
 121	     hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
 122	    (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
 123		return false;
 124
 125	/* Check hash value order, count stale entries.  */
 126	for (i = stale = 0; i < hdr->count; i++) {
 127		if (i + 1 < hdr->count) {
 128			if (be32_to_cpu(ents[i].hashval) >
 129					be32_to_cpu(ents[i + 1].hashval))
 130				return false;
 131		}
 132		if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 133			stale++;
 134	}
 135	if (hdr->stale != stale)
 136		return false;
 137	return true;
 138}
 139
 140/*
 141 * We verify the magic numbers before decoding the leaf header so that on debug
 142 * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
 143 * to incorrect magic numbers.
 144 */
 145static bool
 146xfs_dir3_leaf_verify(
 147	struct xfs_buf		*bp,
 148	__uint16_t		magic)
 149{
 150	struct xfs_mount	*mp = bp->b_target->bt_mount;
 151	struct xfs_dir2_leaf	*leaf = bp->b_addr;
 152
 153	ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 154
 155	if (xfs_sb_version_hascrc(&mp->m_sb)) {
 156		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 157		__uint16_t		magic3;
 158
 159		magic3 = (magic == XFS_DIR2_LEAF1_MAGIC) ? XFS_DIR3_LEAF1_MAGIC
 160							 : XFS_DIR3_LEAFN_MAGIC;
 161
 162		if (leaf3->info.hdr.magic != cpu_to_be16(magic3))
 163			return false;
 164		if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid))
 165			return false;
 166		if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
 167			return false;
 168		if (!xfs_log_check_lsn(mp, be64_to_cpu(leaf3->info.lsn)))
 169			return false;
 170	} else {
 171		if (leaf->hdr.info.magic != cpu_to_be16(magic))
 172			return false;
 173	}
 174
 175	return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
 176}
 177
 178static void
 179__read_verify(
 180	struct xfs_buf  *bp,
 181	__uint16_t	magic)
 182{
 183	struct xfs_mount	*mp = bp->b_target->bt_mount;
 
 184
 185	if (xfs_sb_version_hascrc(&mp->m_sb) &&
 186	     !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
 187		xfs_buf_ioerror(bp, -EFSBADCRC);
 188	else if (!xfs_dir3_leaf_verify(bp, magic))
 189		xfs_buf_ioerror(bp, -EFSCORRUPTED);
 190
 191	if (bp->b_error)
 192		xfs_verifier_error(bp);
 193}
 194
 195static void
 196__write_verify(
 197	struct xfs_buf  *bp,
 198	__uint16_t	magic)
 199{
 200	struct xfs_mount	*mp = bp->b_target->bt_mount;
 201	struct xfs_buf_log_item	*bip = bp->b_fspriv;
 202	struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
 
 203
 204	if (!xfs_dir3_leaf_verify(bp, magic)) {
 205		xfs_buf_ioerror(bp, -EFSCORRUPTED);
 206		xfs_verifier_error(bp);
 207		return;
 208	}
 209
 210	if (!xfs_sb_version_hascrc(&mp->m_sb))
 211		return;
 212
 213	if (bip)
 214		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 215
 216	xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
 217}
 218
 
 
 
 
 
 
 
 219static void
 220xfs_dir3_leaf1_read_verify(
 221	struct xfs_buf	*bp)
 222{
 223	__read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 224}
 225
 226static void
 227xfs_dir3_leaf1_write_verify(
 228	struct xfs_buf	*bp)
 229{
 230	__write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 231}
 232
 
 
 
 
 
 
 
 233static void
 234xfs_dir3_leafn_read_verify(
 235	struct xfs_buf	*bp)
 236{
 237	__read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 238}
 239
 240static void
 241xfs_dir3_leafn_write_verify(
 242	struct xfs_buf	*bp)
 243{
 244	__write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 245}
 246
 247const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
 248	.name = "xfs_dir3_leaf1",
 249	.verify_read = xfs_dir3_leaf1_read_verify,
 250	.verify_write = xfs_dir3_leaf1_write_verify,
 
 251};
 252
 253const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
 254	.name = "xfs_dir3_leafn",
 255	.verify_read = xfs_dir3_leafn_read_verify,
 256	.verify_write = xfs_dir3_leafn_write_verify,
 
 257};
 258
 259static int
 260xfs_dir3_leaf_read(
 261	struct xfs_trans	*tp,
 262	struct xfs_inode	*dp,
 263	xfs_dablk_t		fbno,
 264	xfs_daddr_t		mappedbno,
 265	struct xfs_buf		**bpp)
 266{
 267	int			err;
 268
 269	err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 270				XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
 271	if (!err && tp)
 272		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
 273	return err;
 274}
 275
 276int
 277xfs_dir3_leafn_read(
 278	struct xfs_trans	*tp,
 279	struct xfs_inode	*dp,
 280	xfs_dablk_t		fbno,
 281	xfs_daddr_t		mappedbno,
 282	struct xfs_buf		**bpp)
 283{
 284	int			err;
 285
 286	err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 287				XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
 288	if (!err && tp)
 289		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
 290	return err;
 291}
 292
 293/*
 294 * Initialize a new leaf block, leaf1 or leafn magic accepted.
 295 */
 296static void
 297xfs_dir3_leaf_init(
 298	struct xfs_mount	*mp,
 299	struct xfs_trans	*tp,
 300	struct xfs_buf		*bp,
 301	xfs_ino_t		owner,
 302	__uint16_t		type)
 303{
 304	struct xfs_dir2_leaf	*leaf = bp->b_addr;
 305
 306	ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
 307
 308	if (xfs_sb_version_hascrc(&mp->m_sb)) {
 309		struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 310
 311		memset(leaf3, 0, sizeof(*leaf3));
 312
 313		leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
 314					 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
 315					 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 316		leaf3->info.blkno = cpu_to_be64(bp->b_bn);
 317		leaf3->info.owner = cpu_to_be64(owner);
 318		uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
 319	} else {
 320		memset(leaf, 0, sizeof(*leaf));
 321		leaf->hdr.info.magic = cpu_to_be16(type);
 322	}
 323
 324	/*
 325	 * If it's a leaf-format directory initialize the tail.
 326	 * Caller is responsible for initialising the bests table.
 327	 */
 328	if (type == XFS_DIR2_LEAF1_MAGIC) {
 329		struct xfs_dir2_leaf_tail *ltp;
 330
 331		ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
 332		ltp->bestcount = 0;
 333		bp->b_ops = &xfs_dir3_leaf1_buf_ops;
 334		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
 335	} else {
 336		bp->b_ops = &xfs_dir3_leafn_buf_ops;
 337		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 338	}
 339}
 340
 341int
 342xfs_dir3_leaf_get_buf(
 343	xfs_da_args_t		*args,
 344	xfs_dir2_db_t		bno,
 345	struct xfs_buf		**bpp,
 346	__uint16_t		magic)
 347{
 348	struct xfs_inode	*dp = args->dp;
 349	struct xfs_trans	*tp = args->trans;
 350	struct xfs_mount	*mp = dp->i_mount;
 351	struct xfs_buf		*bp;
 352	int			error;
 353
 354	ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 355	ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
 356	       bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 357
 358	error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
 359			       -1, &bp, XFS_DATA_FORK);
 360	if (error)
 361		return error;
 362
 363	xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
 364	xfs_dir3_leaf_log_header(args, bp);
 365	if (magic == XFS_DIR2_LEAF1_MAGIC)
 366		xfs_dir3_leaf_log_tail(args, bp);
 367	*bpp = bp;
 368	return 0;
 369}
 370
 371/*
 372 * Convert a block form directory to a leaf form directory.
 373 */
 374int						/* error */
 375xfs_dir2_block_to_leaf(
 376	xfs_da_args_t		*args,		/* operation arguments */
 377	struct xfs_buf		*dbp)		/* input block's buffer */
 378{
 379	__be16			*bestsp;	/* leaf's bestsp entries */
 380	xfs_dablk_t		blkno;		/* leaf block's bno */
 381	xfs_dir2_data_hdr_t	*hdr;		/* block header */
 382	xfs_dir2_leaf_entry_t	*blp;		/* block's leaf entries */
 383	xfs_dir2_block_tail_t	*btp;		/* block's tail */
 384	xfs_inode_t		*dp;		/* incore directory inode */
 385	int			error;		/* error return code */
 386	struct xfs_buf		*lbp;		/* leaf block's buffer */
 387	xfs_dir2_db_t		ldb;		/* leaf block's bno */
 388	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
 389	xfs_dir2_leaf_tail_t	*ltp;		/* leaf's tail */
 390	int			needlog;	/* need to log block header */
 391	int			needscan;	/* need to rescan bestfree */
 392	xfs_trans_t		*tp;		/* transaction pointer */
 393	struct xfs_dir2_data_free *bf;
 394	struct xfs_dir2_leaf_entry *ents;
 395	struct xfs_dir3_icleaf_hdr leafhdr;
 396
 397	trace_xfs_dir2_block_to_leaf(args);
 398
 399	dp = args->dp;
 400	tp = args->trans;
 401	/*
 402	 * Add the leaf block to the inode.
 403	 * This interface will only put blocks in the leaf/node range.
 404	 * Since that's empty now, we'll get the root (block 0 in range).
 405	 */
 406	if ((error = xfs_da_grow_inode(args, &blkno))) {
 407		return error;
 408	}
 409	ldb = xfs_dir2_da_to_db(args->geo, blkno);
 410	ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
 411	/*
 412	 * Initialize the leaf block, get a buffer for it.
 413	 */
 414	error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
 415	if (error)
 416		return error;
 417
 418	leaf = lbp->b_addr;
 419	hdr = dbp->b_addr;
 420	xfs_dir3_data_check(dp, dbp);
 421	btp = xfs_dir2_block_tail_p(args->geo, hdr);
 422	blp = xfs_dir2_block_leaf_p(btp);
 423	bf = dp->d_ops->data_bestfree_p(hdr);
 424	ents = dp->d_ops->leaf_ents_p(leaf);
 425
 426	/*
 427	 * Set the counts in the leaf header.
 428	 */
 429	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 430	leafhdr.count = be32_to_cpu(btp->count);
 431	leafhdr.stale = be32_to_cpu(btp->stale);
 432	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 433	xfs_dir3_leaf_log_header(args, lbp);
 434
 435	/*
 436	 * Could compact these but I think we always do the conversion
 437	 * after squeezing out stale entries.
 438	 */
 439	memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 440	xfs_dir3_leaf_log_ents(args, lbp, 0, leafhdr.count - 1);
 441	needscan = 0;
 442	needlog = 1;
 443	/*
 444	 * Make the space formerly occupied by the leaf entries and block
 445	 * tail be free.
 446	 */
 447	xfs_dir2_data_make_free(args, dbp,
 448		(xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 449		(xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
 450				       (char *)blp),
 451		&needlog, &needscan);
 452	/*
 453	 * Fix up the block header, make it a data block.
 454	 */
 455	dbp->b_ops = &xfs_dir3_data_buf_ops;
 456	xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
 457	if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
 458		hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 459	else
 460		hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 461
 462	if (needscan)
 463		xfs_dir2_data_freescan(dp, hdr, &needlog);
 464	/*
 465	 * Set up leaf tail and bests table.
 466	 */
 467	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 468	ltp->bestcount = cpu_to_be32(1);
 469	bestsp = xfs_dir2_leaf_bests_p(ltp);
 470	bestsp[0] =  bf[0].length;
 471	/*
 472	 * Log the data header and leaf bests table.
 473	 */
 474	if (needlog)
 475		xfs_dir2_data_log_header(args, dbp);
 476	xfs_dir3_leaf_check(dp, lbp);
 477	xfs_dir3_data_check(dp, dbp);
 478	xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
 479	return 0;
 480}
 481
 482STATIC void
 483xfs_dir3_leaf_find_stale(
 484	struct xfs_dir3_icleaf_hdr *leafhdr,
 485	struct xfs_dir2_leaf_entry *ents,
 486	int			index,
 487	int			*lowstale,
 488	int			*highstale)
 489{
 490	/*
 491	 * Find the first stale entry before our index, if any.
 492	 */
 493	for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 494		if (ents[*lowstale].address ==
 495		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 496			break;
 497	}
 498
 499	/*
 500	 * Find the first stale entry at or after our index, if any.
 501	 * Stop if the result would require moving more entries than using
 502	 * lowstale.
 503	 */
 504	for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
 505		if (ents[*highstale].address ==
 506		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 507			break;
 508		if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 509			break;
 510	}
 511}
 512
 513struct xfs_dir2_leaf_entry *
 514xfs_dir3_leaf_find_entry(
 515	struct xfs_dir3_icleaf_hdr *leafhdr,
 516	struct xfs_dir2_leaf_entry *ents,
 517	int			index,		/* leaf table position */
 518	int			compact,	/* need to compact leaves */
 519	int			lowstale,	/* index of prev stale leaf */
 520	int			highstale,	/* index of next stale leaf */
 521	int			*lfloglow,	/* low leaf logging index */
 522	int			*lfloghigh)	/* high leaf logging index */
 523{
 524	if (!leafhdr->stale) {
 525		xfs_dir2_leaf_entry_t	*lep;	/* leaf entry table pointer */
 526
 527		/*
 528		 * Now we need to make room to insert the leaf entry.
 529		 *
 530		 * If there are no stale entries, just insert a hole at index.
 531		 */
 532		lep = &ents[index];
 533		if (index < leafhdr->count)
 534			memmove(lep + 1, lep,
 535				(leafhdr->count - index) * sizeof(*lep));
 536
 537		/*
 538		 * Record low and high logging indices for the leaf.
 539		 */
 540		*lfloglow = index;
 541		*lfloghigh = leafhdr->count++;
 542		return lep;
 543	}
 544
 545	/*
 546	 * There are stale entries.
 547	 *
 548	 * We will use one of them for the new entry.  It's probably not at
 549	 * the right location, so we'll have to shift some up or down first.
 550	 *
 551	 * If we didn't compact before, we need to find the nearest stale
 552	 * entries before and after our insertion point.
 553	 */
 554	if (compact == 0)
 555		xfs_dir3_leaf_find_stale(leafhdr, ents, index,
 556					 &lowstale, &highstale);
 557
 558	/*
 559	 * If the low one is better, use it.
 560	 */
 561	if (lowstale >= 0 &&
 562	    (highstale == leafhdr->count ||
 563	     index - lowstale - 1 < highstale - index)) {
 564		ASSERT(index - lowstale - 1 >= 0);
 565		ASSERT(ents[lowstale].address ==
 566		       cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 567
 568		/*
 569		 * Copy entries up to cover the stale entry and make room
 570		 * for the new entry.
 571		 */
 572		if (index - lowstale - 1 > 0) {
 573			memmove(&ents[lowstale], &ents[lowstale + 1],
 574				(index - lowstale - 1) *
 575					sizeof(xfs_dir2_leaf_entry_t));
 576		}
 577		*lfloglow = MIN(lowstale, *lfloglow);
 578		*lfloghigh = MAX(index - 1, *lfloghigh);
 579		leafhdr->stale--;
 580		return &ents[index - 1];
 581	}
 582
 583	/*
 584	 * The high one is better, so use that one.
 585	 */
 586	ASSERT(highstale - index >= 0);
 587	ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 588
 589	/*
 590	 * Copy entries down to cover the stale entry and make room for the
 591	 * new entry.
 592	 */
 593	if (highstale - index > 0) {
 594		memmove(&ents[index + 1], &ents[index],
 595			(highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 596	}
 597	*lfloglow = MIN(index, *lfloglow);
 598	*lfloghigh = MAX(highstale, *lfloghigh);
 599	leafhdr->stale--;
 600	return &ents[index];
 601}
 602
 603/*
 604 * Add an entry to a leaf form directory.
 605 */
 606int						/* error */
 607xfs_dir2_leaf_addname(
 608	xfs_da_args_t		*args)		/* operation arguments */
 609{
 610	__be16			*bestsp;	/* freespace table in leaf */
 611	int			compact;	/* need to compact leaves */
 612	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
 613	struct xfs_buf		*dbp;		/* data block buffer */
 614	xfs_dir2_data_entry_t	*dep;		/* data block entry */
 615	xfs_inode_t		*dp;		/* incore directory inode */
 616	xfs_dir2_data_unused_t	*dup;		/* data unused entry */
 617	int			error;		/* error return value */
 618	int			grown;		/* allocated new data block */
 619	int			highstale;	/* index of next stale leaf */
 620	int			i;		/* temporary, index */
 621	int			index;		/* leaf table position */
 622	struct xfs_buf		*lbp;		/* leaf's buffer */
 623	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
 624	int			length;		/* length of new entry */
 625	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry table pointer */
 626	int			lfloglow;	/* low leaf logging index */
 627	int			lfloghigh;	/* high leaf logging index */
 628	int			lowstale;	/* index of prev stale leaf */
 629	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail pointer */
 630	int			needbytes;	/* leaf block bytes needed */
 631	int			needlog;	/* need to log data header */
 632	int			needscan;	/* need to rescan data free */
 633	__be16			*tagp;		/* end of data entry */
 634	xfs_trans_t		*tp;		/* transaction pointer */
 635	xfs_dir2_db_t		use_block;	/* data block number */
 636	struct xfs_dir2_data_free *bf;		/* bestfree table */
 637	struct xfs_dir2_leaf_entry *ents;
 638	struct xfs_dir3_icleaf_hdr leafhdr;
 639
 640	trace_xfs_dir2_leaf_addname(args);
 641
 642	dp = args->dp;
 643	tp = args->trans;
 644
 645	error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
 646	if (error)
 647		return error;
 648
 649	/*
 650	 * Look up the entry by hash value and name.
 651	 * We know it's not there, our caller has already done a lookup.
 652	 * So the index is of the entry to insert in front of.
 653	 * But if there are dup hash values the index is of the first of those.
 654	 */
 655	index = xfs_dir2_leaf_search_hash(args, lbp);
 656	leaf = lbp->b_addr;
 657	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 658	ents = dp->d_ops->leaf_ents_p(leaf);
 659	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 660	bestsp = xfs_dir2_leaf_bests_p(ltp);
 661	length = dp->d_ops->data_entsize(args->namelen);
 662
 663	/*
 664	 * See if there are any entries with the same hash value
 665	 * and space in their block for the new entry.
 666	 * This is good because it puts multiple same-hash value entries
 667	 * in a data block, improving the lookup of those entries.
 668	 */
 669	for (use_block = -1, lep = &ents[index];
 670	     index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 671	     index++, lep++) {
 672		if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 673			continue;
 674		i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
 675		ASSERT(i < be32_to_cpu(ltp->bestcount));
 676		ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 677		if (be16_to_cpu(bestsp[i]) >= length) {
 678			use_block = i;
 679			break;
 680		}
 681	}
 682	/*
 683	 * Didn't find a block yet, linear search all the data blocks.
 684	 */
 685	if (use_block == -1) {
 686		for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 687			/*
 688			 * Remember a block we see that's missing.
 689			 */
 690			if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 691			    use_block == -1)
 692				use_block = i;
 693			else if (be16_to_cpu(bestsp[i]) >= length) {
 694				use_block = i;
 695				break;
 696			}
 697		}
 698	}
 699	/*
 700	 * How many bytes do we need in the leaf block?
 701	 */
 702	needbytes = 0;
 703	if (!leafhdr.stale)
 704		needbytes += sizeof(xfs_dir2_leaf_entry_t);
 705	if (use_block == -1)
 706		needbytes += sizeof(xfs_dir2_data_off_t);
 707
 708	/*
 709	 * Now kill use_block if it refers to a missing block, so we
 710	 * can use it as an indication of allocation needed.
 711	 */
 712	if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 713		use_block = -1;
 714	/*
 715	 * If we don't have enough free bytes but we can make enough
 716	 * by compacting out stale entries, we'll do that.
 717	 */
 718	if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
 719	    leafhdr.stale > 1)
 720		compact = 1;
 721
 722	/*
 723	 * Otherwise if we don't have enough free bytes we need to
 724	 * convert to node form.
 725	 */
 726	else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
 727		/*
 728		 * Just checking or no space reservation, give up.
 729		 */
 730		if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 731							args->total == 0) {
 732			xfs_trans_brelse(tp, lbp);
 733			return -ENOSPC;
 734		}
 735		/*
 736		 * Convert to node form.
 737		 */
 738		error = xfs_dir2_leaf_to_node(args, lbp);
 739		if (error)
 740			return error;
 741		/*
 742		 * Then add the new entry.
 743		 */
 744		return xfs_dir2_node_addname(args);
 745	}
 746	/*
 747	 * Otherwise it will fit without compaction.
 748	 */
 749	else
 750		compact = 0;
 751	/*
 752	 * If just checking, then it will fit unless we needed to allocate
 753	 * a new data block.
 754	 */
 755	if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 756		xfs_trans_brelse(tp, lbp);
 757		return use_block == -1 ? -ENOSPC : 0;
 758	}
 759	/*
 760	 * If no allocations are allowed, return now before we've
 761	 * changed anything.
 762	 */
 763	if (args->total == 0 && use_block == -1) {
 764		xfs_trans_brelse(tp, lbp);
 765		return -ENOSPC;
 766	}
 767	/*
 768	 * Need to compact the leaf entries, removing stale ones.
 769	 * Leave one stale entry behind - the one closest to our
 770	 * insertion index - and we'll shift that one to our insertion
 771	 * point later.
 772	 */
 773	if (compact) {
 774		xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 775			&highstale, &lfloglow, &lfloghigh);
 776	}
 777	/*
 778	 * There are stale entries, so we'll need log-low and log-high
 779	 * impossibly bad values later.
 780	 */
 781	else if (leafhdr.stale) {
 782		lfloglow = leafhdr.count;
 783		lfloghigh = -1;
 784	}
 785	/*
 786	 * If there was no data block space found, we need to allocate
 787	 * a new one.
 788	 */
 789	if (use_block == -1) {
 790		/*
 791		 * Add the new data block.
 792		 */
 793		if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 794				&use_block))) {
 795			xfs_trans_brelse(tp, lbp);
 796			return error;
 797		}
 798		/*
 799		 * Initialize the block.
 800		 */
 801		if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
 802			xfs_trans_brelse(tp, lbp);
 803			return error;
 804		}
 805		/*
 806		 * If we're adding a new data block on the end we need to
 807		 * extend the bests table.  Copy it up one entry.
 808		 */
 809		if (use_block >= be32_to_cpu(ltp->bestcount)) {
 810			bestsp--;
 811			memmove(&bestsp[0], &bestsp[1],
 812				be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 813			be32_add_cpu(&ltp->bestcount, 1);
 814			xfs_dir3_leaf_log_tail(args, lbp);
 815			xfs_dir3_leaf_log_bests(args, lbp, 0,
 816						be32_to_cpu(ltp->bestcount) - 1);
 817		}
 818		/*
 819		 * If we're filling in a previously empty block just log it.
 820		 */
 821		else
 822			xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
 823		hdr = dbp->b_addr;
 824		bf = dp->d_ops->data_bestfree_p(hdr);
 825		bestsp[use_block] = bf[0].length;
 826		grown = 1;
 827	} else {
 828		/*
 829		 * Already had space in some data block.
 830		 * Just read that one in.
 831		 */
 832		error = xfs_dir3_data_read(tp, dp,
 833				   xfs_dir2_db_to_da(args->geo, use_block),
 834				   -1, &dbp);
 835		if (error) {
 836			xfs_trans_brelse(tp, lbp);
 837			return error;
 838		}
 839		hdr = dbp->b_addr;
 840		bf = dp->d_ops->data_bestfree_p(hdr);
 841		grown = 0;
 842	}
 843	/*
 844	 * Point to the biggest freespace in our data block.
 845	 */
 846	dup = (xfs_dir2_data_unused_t *)
 847	      ((char *)hdr + be16_to_cpu(bf[0].offset));
 848	ASSERT(be16_to_cpu(dup->length) >= length);
 849	needscan = needlog = 0;
 850	/*
 851	 * Mark the initial part of our freespace in use for the new entry.
 852	 */
 853	xfs_dir2_data_use_free(args, dbp, dup,
 854		(xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 855		&needlog, &needscan);
 
 
 
 
 856	/*
 857	 * Initialize our new entry (at last).
 858	 */
 859	dep = (xfs_dir2_data_entry_t *)dup;
 860	dep->inumber = cpu_to_be64(args->inumber);
 861	dep->namelen = args->namelen;
 862	memcpy(dep->name, args->name, dep->namelen);
 863	dp->d_ops->data_put_ftype(dep, args->filetype);
 864	tagp = dp->d_ops->data_entry_tag_p(dep);
 865	*tagp = cpu_to_be16((char *)dep - (char *)hdr);
 866	/*
 867	 * Need to scan fix up the bestfree table.
 868	 */
 869	if (needscan)
 870		xfs_dir2_data_freescan(dp, hdr, &needlog);
 871	/*
 872	 * Need to log the data block's header.
 873	 */
 874	if (needlog)
 875		xfs_dir2_data_log_header(args, dbp);
 876	xfs_dir2_data_log_entry(args, dbp, dep);
 877	/*
 878	 * If the bests table needs to be changed, do it.
 879	 * Log the change unless we've already done that.
 880	 */
 881	if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
 882		bestsp[use_block] = bf[0].length;
 883		if (!grown)
 884			xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
 885	}
 886
 887	lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 888				       highstale, &lfloglow, &lfloghigh);
 889
 890	/*
 891	 * Fill in the new leaf entry.
 892	 */
 893	lep->hashval = cpu_to_be32(args->hashval);
 894	lep->address = cpu_to_be32(
 895				xfs_dir2_db_off_to_dataptr(args->geo, use_block,
 896				be16_to_cpu(*tagp)));
 897	/*
 898	 * Log the leaf fields and give up the buffers.
 899	 */
 900	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 901	xfs_dir3_leaf_log_header(args, lbp);
 902	xfs_dir3_leaf_log_ents(args, lbp, lfloglow, lfloghigh);
 903	xfs_dir3_leaf_check(dp, lbp);
 904	xfs_dir3_data_check(dp, dbp);
 905	return 0;
 906}
 907
 908/*
 909 * Compact out any stale entries in the leaf.
 910 * Log the header and changed leaf entries, if any.
 911 */
 912void
 913xfs_dir3_leaf_compact(
 914	xfs_da_args_t	*args,		/* operation arguments */
 915	struct xfs_dir3_icleaf_hdr *leafhdr,
 916	struct xfs_buf	*bp)		/* leaf buffer */
 917{
 918	int		from;		/* source leaf index */
 919	xfs_dir2_leaf_t	*leaf;		/* leaf structure */
 920	int		loglow;		/* first leaf entry to log */
 921	int		to;		/* target leaf index */
 922	struct xfs_dir2_leaf_entry *ents;
 923	struct xfs_inode *dp = args->dp;
 924
 925	leaf = bp->b_addr;
 926	if (!leafhdr->stale)
 927		return;
 928
 929	/*
 930	 * Compress out the stale entries in place.
 931	 */
 932	ents = dp->d_ops->leaf_ents_p(leaf);
 933	for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
 934		if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 935			continue;
 936		/*
 937		 * Only actually copy the entries that are different.
 938		 */
 939		if (from > to) {
 940			if (loglow == -1)
 941				loglow = to;
 942			ents[to] = ents[from];
 943		}
 944		to++;
 945	}
 946	/*
 947	 * Update and log the header, log the leaf entries.
 948	 */
 949	ASSERT(leafhdr->stale == from - to);
 950	leafhdr->count -= leafhdr->stale;
 951	leafhdr->stale = 0;
 952
 953	dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
 954	xfs_dir3_leaf_log_header(args, bp);
 955	if (loglow != -1)
 956		xfs_dir3_leaf_log_ents(args, bp, loglow, to - 1);
 957}
 958
 959/*
 960 * Compact the leaf entries, removing stale ones.
 961 * Leave one stale entry behind - the one closest to our
 962 * insertion index - and the caller will shift that one to our insertion
 963 * point later.
 964 * Return new insertion index, where the remaining stale entry is,
 965 * and leaf logging indices.
 966 */
 967void
 968xfs_dir3_leaf_compact_x1(
 969	struct xfs_dir3_icleaf_hdr *leafhdr,
 970	struct xfs_dir2_leaf_entry *ents,
 971	int		*indexp,	/* insertion index */
 972	int		*lowstalep,	/* out: stale entry before us */
 973	int		*highstalep,	/* out: stale entry after us */
 974	int		*lowlogp,	/* out: low log index */
 975	int		*highlogp)	/* out: high log index */
 976{
 977	int		from;		/* source copy index */
 978	int		highstale;	/* stale entry at/after index */
 979	int		index;		/* insertion index */
 980	int		keepstale;	/* source index of kept stale */
 981	int		lowstale;	/* stale entry before index */
 982	int		newindex=0;	/* new insertion index */
 983	int		to;		/* destination copy index */
 984
 985	ASSERT(leafhdr->stale > 1);
 986	index = *indexp;
 987
 988	xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
 989
 990	/*
 991	 * Pick the better of lowstale and highstale.
 992	 */
 993	if (lowstale >= 0 &&
 994	    (highstale == leafhdr->count ||
 995	     index - lowstale <= highstale - index))
 996		keepstale = lowstale;
 997	else
 998		keepstale = highstale;
 999	/*
1000	 * Copy the entries in place, removing all the stale entries
1001	 * except keepstale.
1002	 */
1003	for (from = to = 0; from < leafhdr->count; from++) {
1004		/*
1005		 * Notice the new value of index.
1006		 */
1007		if (index == from)
1008			newindex = to;
1009		if (from != keepstale &&
1010		    ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1011			if (from == to)
1012				*lowlogp = to;
1013			continue;
1014		}
1015		/*
1016		 * Record the new keepstale value for the insertion.
1017		 */
1018		if (from == keepstale)
1019			lowstale = highstale = to;
1020		/*
1021		 * Copy only the entries that have moved.
1022		 */
1023		if (from > to)
1024			ents[to] = ents[from];
1025		to++;
1026	}
1027	ASSERT(from > to);
1028	/*
1029	 * If the insertion point was past the last entry,
1030	 * set the new insertion point accordingly.
1031	 */
1032	if (index == from)
1033		newindex = to;
1034	*indexp = newindex;
1035	/*
1036	 * Adjust the leaf header values.
1037	 */
1038	leafhdr->count -= from - to;
1039	leafhdr->stale = 1;
1040	/*
1041	 * Remember the low/high stale value only in the "right"
1042	 * direction.
1043	 */
1044	if (lowstale >= newindex)
1045		lowstale = -1;
1046	else
1047		highstale = leafhdr->count;
1048	*highlogp = leafhdr->count - 1;
1049	*lowstalep = lowstale;
1050	*highstalep = highstale;
1051}
1052
1053/*
1054 * Log the bests entries indicated from a leaf1 block.
1055 */
1056static void
1057xfs_dir3_leaf_log_bests(
1058	struct xfs_da_args	*args,
1059	struct xfs_buf		*bp,		/* leaf buffer */
1060	int			first,		/* first entry to log */
1061	int			last)		/* last entry to log */
1062{
1063	__be16			*firstb;	/* pointer to first entry */
1064	__be16			*lastb;		/* pointer to last entry */
1065	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1066	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1067
1068	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1069	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1070
1071	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1072	firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1073	lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1074	xfs_trans_log_buf(args->trans, bp,
1075		(uint)((char *)firstb - (char *)leaf),
1076		(uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1077}
1078
1079/*
1080 * Log the leaf entries indicated from a leaf1 or leafn block.
1081 */
1082void
1083xfs_dir3_leaf_log_ents(
1084	struct xfs_da_args	*args,
1085	struct xfs_buf		*bp,
1086	int			first,
1087	int			last)
1088{
1089	xfs_dir2_leaf_entry_t	*firstlep;	/* pointer to first entry */
1090	xfs_dir2_leaf_entry_t	*lastlep;	/* pointer to last entry */
1091	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1092	struct xfs_dir2_leaf_entry *ents;
1093
1094	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1095	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1096	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1097	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1098
1099	ents = args->dp->d_ops->leaf_ents_p(leaf);
1100	firstlep = &ents[first];
1101	lastlep = &ents[last];
1102	xfs_trans_log_buf(args->trans, bp,
1103		(uint)((char *)firstlep - (char *)leaf),
1104		(uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1105}
1106
1107/*
1108 * Log the header of the leaf1 or leafn block.
1109 */
1110void
1111xfs_dir3_leaf_log_header(
1112	struct xfs_da_args	*args,
1113	struct xfs_buf		*bp)
1114{
1115	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1116
1117	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1118	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1119	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1120	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1121
1122	xfs_trans_log_buf(args->trans, bp,
1123			  (uint)((char *)&leaf->hdr - (char *)leaf),
1124			  args->dp->d_ops->leaf_hdr_size - 1);
1125}
1126
1127/*
1128 * Log the tail of the leaf1 block.
1129 */
1130STATIC void
1131xfs_dir3_leaf_log_tail(
1132	struct xfs_da_args	*args,
1133	struct xfs_buf		*bp)
1134{
1135	struct xfs_dir2_leaf	*leaf = bp->b_addr;
1136	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1137
1138	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1139	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1140	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1141	       leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1142
1143	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1144	xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1145		(uint)(args->geo->blksize - 1));
1146}
1147
1148/*
1149 * Look up the entry referred to by args in the leaf format directory.
1150 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1151 * is also used by the node-format code.
1152 */
1153int
1154xfs_dir2_leaf_lookup(
1155	xfs_da_args_t		*args)		/* operation arguments */
1156{
1157	struct xfs_buf		*dbp;		/* data block buffer */
1158	xfs_dir2_data_entry_t	*dep;		/* data block entry */
1159	xfs_inode_t		*dp;		/* incore directory inode */
1160	int			error;		/* error return code */
1161	int			index;		/* found entry index */
1162	struct xfs_buf		*lbp;		/* leaf buffer */
1163	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1164	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1165	xfs_trans_t		*tp;		/* transaction pointer */
1166	struct xfs_dir2_leaf_entry *ents;
1167
1168	trace_xfs_dir2_leaf_lookup(args);
1169
1170	/*
1171	 * Look up name in the leaf block, returning both buffers and index.
1172	 */
1173	if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1174		return error;
1175	}
1176	tp = args->trans;
1177	dp = args->dp;
1178	xfs_dir3_leaf_check(dp, lbp);
1179	leaf = lbp->b_addr;
1180	ents = dp->d_ops->leaf_ents_p(leaf);
1181	/*
1182	 * Get to the leaf entry and contained data entry address.
1183	 */
1184	lep = &ents[index];
1185
1186	/*
1187	 * Point to the data entry.
1188	 */
1189	dep = (xfs_dir2_data_entry_t *)
1190	      ((char *)dbp->b_addr +
1191	       xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1192	/*
1193	 * Return the found inode number & CI name if appropriate
1194	 */
1195	args->inumber = be64_to_cpu(dep->inumber);
1196	args->filetype = dp->d_ops->data_get_ftype(dep);
1197	error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1198	xfs_trans_brelse(tp, dbp);
1199	xfs_trans_brelse(tp, lbp);
1200	return error;
1201}
1202
1203/*
1204 * Look up name/hash in the leaf block.
1205 * Fill in indexp with the found index, and dbpp with the data buffer.
1206 * If not found dbpp will be NULL, and ENOENT comes back.
1207 * lbpp will always be filled in with the leaf buffer unless there's an error.
1208 */
1209static int					/* error */
1210xfs_dir2_leaf_lookup_int(
1211	xfs_da_args_t		*args,		/* operation arguments */
1212	struct xfs_buf		**lbpp,		/* out: leaf buffer */
1213	int			*indexp,	/* out: index in leaf block */
1214	struct xfs_buf		**dbpp)		/* out: data buffer */
1215{
1216	xfs_dir2_db_t		curdb = -1;	/* current data block number */
1217	struct xfs_buf		*dbp = NULL;	/* data buffer */
1218	xfs_dir2_data_entry_t	*dep;		/* data entry */
1219	xfs_inode_t		*dp;		/* incore directory inode */
1220	int			error;		/* error return code */
1221	int			index;		/* index in leaf block */
1222	struct xfs_buf		*lbp;		/* leaf buffer */
1223	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1224	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1225	xfs_mount_t		*mp;		/* filesystem mount point */
1226	xfs_dir2_db_t		newdb;		/* new data block number */
1227	xfs_trans_t		*tp;		/* transaction pointer */
1228	xfs_dir2_db_t		cidb = -1;	/* case match data block no. */
1229	enum xfs_dacmp		cmp;		/* name compare result */
1230	struct xfs_dir2_leaf_entry *ents;
1231	struct xfs_dir3_icleaf_hdr leafhdr;
1232
1233	dp = args->dp;
1234	tp = args->trans;
1235	mp = dp->i_mount;
1236
1237	error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
1238	if (error)
1239		return error;
1240
1241	*lbpp = lbp;
1242	leaf = lbp->b_addr;
1243	xfs_dir3_leaf_check(dp, lbp);
1244	ents = dp->d_ops->leaf_ents_p(leaf);
1245	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1246
1247	/*
1248	 * Look for the first leaf entry with our hash value.
1249	 */
1250	index = xfs_dir2_leaf_search_hash(args, lbp);
1251	/*
1252	 * Loop over all the entries with the right hash value
1253	 * looking to match the name.
1254	 */
1255	for (lep = &ents[index];
1256	     index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1257	     lep++, index++) {
1258		/*
1259		 * Skip over stale leaf entries.
1260		 */
1261		if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1262			continue;
1263		/*
1264		 * Get the new data block number.
1265		 */
1266		newdb = xfs_dir2_dataptr_to_db(args->geo,
1267					       be32_to_cpu(lep->address));
1268		/*
1269		 * If it's not the same as the old data block number,
1270		 * need to pitch the old one and read the new one.
1271		 */
1272		if (newdb != curdb) {
1273			if (dbp)
1274				xfs_trans_brelse(tp, dbp);
1275			error = xfs_dir3_data_read(tp, dp,
1276					   xfs_dir2_db_to_da(args->geo, newdb),
1277					   -1, &dbp);
1278			if (error) {
1279				xfs_trans_brelse(tp, lbp);
1280				return error;
1281			}
1282			curdb = newdb;
1283		}
1284		/*
1285		 * Point to the data entry.
1286		 */
1287		dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1288			xfs_dir2_dataptr_to_off(args->geo,
1289						be32_to_cpu(lep->address)));
1290		/*
1291		 * Compare name and if it's an exact match, return the index
1292		 * and buffer. If it's the first case-insensitive match, store
1293		 * the index and buffer and continue looking for an exact match.
1294		 */
1295		cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1296		if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1297			args->cmpresult = cmp;
1298			*indexp = index;
1299			/* case exact match: return the current buffer. */
1300			if (cmp == XFS_CMP_EXACT) {
1301				*dbpp = dbp;
1302				return 0;
1303			}
1304			cidb = curdb;
1305		}
1306	}
1307	ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1308	/*
1309	 * Here, we can only be doing a lookup (not a rename or remove).
1310	 * If a case-insensitive match was found earlier, re-read the
1311	 * appropriate data block if required and return it.
1312	 */
1313	if (args->cmpresult == XFS_CMP_CASE) {
1314		ASSERT(cidb != -1);
1315		if (cidb != curdb) {
1316			xfs_trans_brelse(tp, dbp);
1317			error = xfs_dir3_data_read(tp, dp,
1318					   xfs_dir2_db_to_da(args->geo, cidb),
1319					   -1, &dbp);
1320			if (error) {
1321				xfs_trans_brelse(tp, lbp);
1322				return error;
1323			}
1324		}
1325		*dbpp = dbp;
1326		return 0;
1327	}
1328	/*
1329	 * No match found, return -ENOENT.
1330	 */
1331	ASSERT(cidb == -1);
1332	if (dbp)
1333		xfs_trans_brelse(tp, dbp);
1334	xfs_trans_brelse(tp, lbp);
1335	return -ENOENT;
1336}
1337
1338/*
1339 * Remove an entry from a leaf format directory.
1340 */
1341int						/* error */
1342xfs_dir2_leaf_removename(
1343	xfs_da_args_t		*args)		/* operation arguments */
1344{
1345	__be16			*bestsp;	/* leaf block best freespace */
1346	xfs_dir2_data_hdr_t	*hdr;		/* data block header */
1347	xfs_dir2_db_t		db;		/* data block number */
1348	struct xfs_buf		*dbp;		/* data block buffer */
1349	xfs_dir2_data_entry_t	*dep;		/* data entry structure */
1350	xfs_inode_t		*dp;		/* incore directory inode */
1351	int			error;		/* error return code */
1352	xfs_dir2_db_t		i;		/* temporary data block # */
1353	int			index;		/* index into leaf entries */
1354	struct xfs_buf		*lbp;		/* leaf buffer */
1355	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1356	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1357	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1358	int			needlog;	/* need to log data header */
1359	int			needscan;	/* need to rescan data frees */
1360	xfs_dir2_data_off_t	oldbest;	/* old value of best free */
1361	struct xfs_dir2_data_free *bf;		/* bestfree table */
1362	struct xfs_dir2_leaf_entry *ents;
1363	struct xfs_dir3_icleaf_hdr leafhdr;
1364
1365	trace_xfs_dir2_leaf_removename(args);
1366
1367	/*
1368	 * Lookup the leaf entry, get the leaf and data blocks read in.
1369	 */
1370	if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1371		return error;
1372	}
1373	dp = args->dp;
1374	leaf = lbp->b_addr;
1375	hdr = dbp->b_addr;
1376	xfs_dir3_data_check(dp, dbp);
1377	bf = dp->d_ops->data_bestfree_p(hdr);
1378	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1379	ents = dp->d_ops->leaf_ents_p(leaf);
1380	/*
1381	 * Point to the leaf entry, use that to point to the data entry.
1382	 */
1383	lep = &ents[index];
1384	db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1385	dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1386		xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1387	needscan = needlog = 0;
1388	oldbest = be16_to_cpu(bf[0].length);
1389	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1390	bestsp = xfs_dir2_leaf_bests_p(ltp);
1391	ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
 
1392	/*
1393	 * Mark the former data entry unused.
1394	 */
1395	xfs_dir2_data_make_free(args, dbp,
1396		(xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1397		dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1398	/*
1399	 * We just mark the leaf entry stale by putting a null in it.
1400	 */
1401	leafhdr.stale++;
1402	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1403	xfs_dir3_leaf_log_header(args, lbp);
1404
1405	lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1406	xfs_dir3_leaf_log_ents(args, lbp, index, index);
1407
1408	/*
1409	 * Scan the freespace in the data block again if necessary,
1410	 * log the data block header if necessary.
1411	 */
1412	if (needscan)
1413		xfs_dir2_data_freescan(dp, hdr, &needlog);
1414	if (needlog)
1415		xfs_dir2_data_log_header(args, dbp);
1416	/*
1417	 * If the longest freespace in the data block has changed,
1418	 * put the new value in the bests table and log that.
1419	 */
1420	if (be16_to_cpu(bf[0].length) != oldbest) {
1421		bestsp[db] = bf[0].length;
1422		xfs_dir3_leaf_log_bests(args, lbp, db, db);
1423	}
1424	xfs_dir3_data_check(dp, dbp);
1425	/*
1426	 * If the data block is now empty then get rid of the data block.
1427	 */
1428	if (be16_to_cpu(bf[0].length) ==
1429			args->geo->blksize - dp->d_ops->data_entry_offset) {
1430		ASSERT(db != args->geo->datablk);
1431		if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1432			/*
1433			 * Nope, can't get rid of it because it caused
1434			 * allocation of a bmap btree block to do so.
1435			 * Just go on, returning success, leaving the
1436			 * empty block in place.
1437			 */
1438			if (error == -ENOSPC && args->total == 0)
1439				error = 0;
1440			xfs_dir3_leaf_check(dp, lbp);
1441			return error;
1442		}
1443		dbp = NULL;
1444		/*
1445		 * If this is the last data block then compact the
1446		 * bests table by getting rid of entries.
1447		 */
1448		if (db == be32_to_cpu(ltp->bestcount) - 1) {
1449			/*
1450			 * Look for the last active entry (i).
1451			 */
1452			for (i = db - 1; i > 0; i--) {
1453				if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1454					break;
1455			}
1456			/*
1457			 * Copy the table down so inactive entries at the
1458			 * end are removed.
1459			 */
1460			memmove(&bestsp[db - i], bestsp,
1461				(be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1462			be32_add_cpu(&ltp->bestcount, -(db - i));
1463			xfs_dir3_leaf_log_tail(args, lbp);
1464			xfs_dir3_leaf_log_bests(args, lbp, 0,
1465						be32_to_cpu(ltp->bestcount) - 1);
1466		} else
1467			bestsp[db] = cpu_to_be16(NULLDATAOFF);
1468	}
1469	/*
1470	 * If the data block was not the first one, drop it.
1471	 */
1472	else if (db != args->geo->datablk)
1473		dbp = NULL;
1474
1475	xfs_dir3_leaf_check(dp, lbp);
1476	/*
1477	 * See if we can convert to block form.
1478	 */
1479	return xfs_dir2_leaf_to_block(args, lbp, dbp);
1480}
1481
1482/*
1483 * Replace the inode number in a leaf format directory entry.
1484 */
1485int						/* error */
1486xfs_dir2_leaf_replace(
1487	xfs_da_args_t		*args)		/* operation arguments */
1488{
1489	struct xfs_buf		*dbp;		/* data block buffer */
1490	xfs_dir2_data_entry_t	*dep;		/* data block entry */
1491	xfs_inode_t		*dp;		/* incore directory inode */
1492	int			error;		/* error return code */
1493	int			index;		/* index of leaf entry */
1494	struct xfs_buf		*lbp;		/* leaf buffer */
1495	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1496	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1497	xfs_trans_t		*tp;		/* transaction pointer */
1498	struct xfs_dir2_leaf_entry *ents;
1499
1500	trace_xfs_dir2_leaf_replace(args);
1501
1502	/*
1503	 * Look up the entry.
1504	 */
1505	if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1506		return error;
1507	}
1508	dp = args->dp;
1509	leaf = lbp->b_addr;
1510	ents = dp->d_ops->leaf_ents_p(leaf);
1511	/*
1512	 * Point to the leaf entry, get data address from it.
1513	 */
1514	lep = &ents[index];
1515	/*
1516	 * Point to the data entry.
1517	 */
1518	dep = (xfs_dir2_data_entry_t *)
1519	      ((char *)dbp->b_addr +
1520	       xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1521	ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1522	/*
1523	 * Put the new inode number in, log it.
1524	 */
1525	dep->inumber = cpu_to_be64(args->inumber);
1526	dp->d_ops->data_put_ftype(dep, args->filetype);
1527	tp = args->trans;
1528	xfs_dir2_data_log_entry(args, dbp, dep);
1529	xfs_dir3_leaf_check(dp, lbp);
1530	xfs_trans_brelse(tp, lbp);
1531	return 0;
1532}
1533
1534/*
1535 * Return index in the leaf block (lbp) which is either the first
1536 * one with this hash value, or if there are none, the insert point
1537 * for that hash value.
1538 */
1539int						/* index value */
1540xfs_dir2_leaf_search_hash(
1541	xfs_da_args_t		*args,		/* operation arguments */
1542	struct xfs_buf		*lbp)		/* leaf buffer */
1543{
1544	xfs_dahash_t		hash=0;		/* hash from this entry */
1545	xfs_dahash_t		hashwant;	/* hash value looking for */
1546	int			high;		/* high leaf index */
1547	int			low;		/* low leaf index */
1548	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1549	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
1550	int			mid=0;		/* current leaf index */
1551	struct xfs_dir2_leaf_entry *ents;
1552	struct xfs_dir3_icleaf_hdr leafhdr;
1553
1554	leaf = lbp->b_addr;
1555	ents = args->dp->d_ops->leaf_ents_p(leaf);
1556	args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1557
1558	/*
1559	 * Note, the table cannot be empty, so we have to go through the loop.
1560	 * Binary search the leaf entries looking for our hash value.
1561	 */
1562	for (lep = ents, low = 0, high = leafhdr.count - 1,
1563		hashwant = args->hashval;
1564	     low <= high; ) {
1565		mid = (low + high) >> 1;
1566		if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1567			break;
1568		if (hash < hashwant)
1569			low = mid + 1;
1570		else
1571			high = mid - 1;
1572	}
1573	/*
1574	 * Found one, back up through all the equal hash values.
1575	 */
1576	if (hash == hashwant) {
1577		while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1578			mid--;
1579		}
1580	}
1581	/*
1582	 * Need to point to an entry higher than ours.
1583	 */
1584	else if (hash < hashwant)
1585		mid++;
1586	return mid;
1587}
1588
1589/*
1590 * Trim off a trailing data block.  We know it's empty since the leaf
1591 * freespace table says so.
1592 */
1593int						/* error */
1594xfs_dir2_leaf_trim_data(
1595	xfs_da_args_t		*args,		/* operation arguments */
1596	struct xfs_buf		*lbp,		/* leaf buffer */
1597	xfs_dir2_db_t		db)		/* data block number */
1598{
1599	__be16			*bestsp;	/* leaf bests table */
1600	struct xfs_buf		*dbp;		/* data block buffer */
1601	xfs_inode_t		*dp;		/* incore directory inode */
1602	int			error;		/* error return value */
1603	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1604	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
1605	xfs_trans_t		*tp;		/* transaction pointer */
1606
1607	dp = args->dp;
1608	tp = args->trans;
1609	/*
1610	 * Read the offending data block.  We need its buffer.
1611	 */
1612	error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(args->geo, db),
1613				   -1, &dbp);
1614	if (error)
1615		return error;
1616
1617	leaf = lbp->b_addr;
1618	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1619
1620#ifdef DEBUG
1621{
1622	struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1623	struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1624
1625	ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1626	       hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1627	ASSERT(be16_to_cpu(bf[0].length) ==
1628	       args->geo->blksize - dp->d_ops->data_entry_offset);
1629	ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1630}
1631#endif
1632
1633	/*
1634	 * Get rid of the data block.
1635	 */
1636	if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1637		ASSERT(error != -ENOSPC);
1638		xfs_trans_brelse(tp, dbp);
1639		return error;
1640	}
1641	/*
1642	 * Eliminate the last bests entry from the table.
1643	 */
1644	bestsp = xfs_dir2_leaf_bests_p(ltp);
1645	be32_add_cpu(&ltp->bestcount, -1);
1646	memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1647	xfs_dir3_leaf_log_tail(args, lbp);
1648	xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1649	return 0;
1650}
1651
1652static inline size_t
1653xfs_dir3_leaf_size(
1654	struct xfs_dir3_icleaf_hdr	*hdr,
1655	int				counts)
1656{
1657	int	entries;
1658	int	hdrsize;
1659
1660	entries = hdr->count - hdr->stale;
1661	if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1662	    hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1663		hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1664	else
1665		hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1666
1667	return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1668	               + counts * sizeof(xfs_dir2_data_off_t)
1669		       + sizeof(xfs_dir2_leaf_tail_t);
1670}
1671
1672/*
1673 * Convert node form directory to leaf form directory.
1674 * The root of the node form dir needs to already be a LEAFN block.
1675 * Just return if we can't do anything.
1676 */
1677int						/* error */
1678xfs_dir2_node_to_leaf(
1679	xfs_da_state_t		*state)		/* directory operation state */
1680{
1681	xfs_da_args_t		*args;		/* operation arguments */
1682	xfs_inode_t		*dp;		/* incore directory inode */
1683	int			error;		/* error return code */
1684	struct xfs_buf		*fbp;		/* buffer for freespace block */
1685	xfs_fileoff_t		fo;		/* freespace file offset */
1686	xfs_dir2_free_t		*free;		/* freespace structure */
1687	struct xfs_buf		*lbp;		/* buffer for leaf block */
1688	xfs_dir2_leaf_tail_t	*ltp;		/* tail of leaf structure */
1689	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1690	xfs_mount_t		*mp;		/* filesystem mount point */
1691	int			rval;		/* successful free trim? */
1692	xfs_trans_t		*tp;		/* transaction pointer */
1693	struct xfs_dir3_icleaf_hdr leafhdr;
1694	struct xfs_dir3_icfree_hdr freehdr;
1695
1696	/*
1697	 * There's more than a leaf level in the btree, so there must
1698	 * be multiple leafn blocks.  Give up.
1699	 */
1700	if (state->path.active > 1)
1701		return 0;
1702	args = state->args;
1703
1704	trace_xfs_dir2_node_to_leaf(args);
1705
1706	mp = state->mp;
1707	dp = args->dp;
1708	tp = args->trans;
1709	/*
1710	 * Get the last offset in the file.
1711	 */
1712	if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1713		return error;
1714	}
1715	fo -= args->geo->fsbcount;
1716	/*
1717	 * If there are freespace blocks other than the first one,
1718	 * take this opportunity to remove trailing empty freespace blocks
1719	 * that may have been left behind during no-space-reservation
1720	 * operations.
1721	 */
1722	while (fo > args->geo->freeblk) {
1723		if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1724			return error;
1725		}
1726		if (rval)
1727			fo -= args->geo->fsbcount;
1728		else
1729			return 0;
1730	}
1731	/*
1732	 * Now find the block just before the freespace block.
1733	 */
1734	if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1735		return error;
1736	}
1737	/*
1738	 * If it's not the single leaf block, give up.
1739	 */
1740	if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1741		return 0;
1742	lbp = state->path.blk[0].bp;
1743	leaf = lbp->b_addr;
1744	dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1745
1746	ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1747	       leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1748
1749	/*
1750	 * Read the freespace block.
1751	 */
1752	error = xfs_dir2_free_read(tp, dp,  args->geo->freeblk, &fbp);
1753	if (error)
1754		return error;
1755	free = fbp->b_addr;
1756	dp->d_ops->free_hdr_from_disk(&freehdr, free);
1757
1758	ASSERT(!freehdr.firstdb);
1759
1760	/*
1761	 * Now see if the leafn and free data will fit in a leaf1.
1762	 * If not, release the buffer and give up.
1763	 */
1764	if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1765		xfs_trans_brelse(tp, fbp);
1766		return 0;
1767	}
1768
1769	/*
1770	 * If the leaf has any stale entries in it, compress them out.
1771	 */
1772	if (leafhdr.stale)
1773		xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1774
1775	lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1776	xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1777	leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1778					? XFS_DIR2_LEAF1_MAGIC
1779					: XFS_DIR3_LEAF1_MAGIC;
1780
1781	/*
1782	 * Set up the leaf tail from the freespace block.
1783	 */
1784	ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1785	ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1786
1787	/*
1788	 * Set up the leaf bests table.
1789	 */
1790	memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1791		freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1792
1793	dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1794	xfs_dir3_leaf_log_header(args, lbp);
1795	xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1796	xfs_dir3_leaf_log_tail(args, lbp);
1797	xfs_dir3_leaf_check(dp, lbp);
1798
1799	/*
1800	 * Get rid of the freespace block.
1801	 */
1802	error = xfs_dir2_shrink_inode(args,
1803			xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1804			fbp);
1805	if (error) {
1806		/*
1807		 * This can't fail here because it can only happen when
1808		 * punching out the middle of an extent, and this is an
1809		 * isolated block.
1810		 */
1811		ASSERT(error != -ENOSPC);
1812		return error;
1813	}
1814	fbp = NULL;
1815	/*
1816	 * Now see if we can convert the single-leaf directory
1817	 * down to a block form directory.
1818	 * This routine always kills the dabuf for the leaf, so
1819	 * eliminate it from the path.
1820	 */
1821	error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1822	state->path.blk[0].bp = NULL;
1823	return error;
1824}