1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4 * Copyright (c) 2013 Red Hat, Inc.
   5 * All Rights Reserved.
   6 */
   7#include "xfs.h"
   8#include "xfs_fs.h"
   9#include "xfs_shared.h"
  10#include "xfs_format.h"
  11#include "xfs_log_format.h"
  12#include "xfs_trans_resv.h"
  13#include "xfs_sb.h"
  14#include "xfs_mount.h"
  15#include "xfs_da_format.h"
  16#include "xfs_da_btree.h"
  17#include "xfs_inode.h"
  18#include "xfs_trans.h"
  19#include "xfs_bmap_btree.h"
  20#include "xfs_bmap.h"
  21#include "xfs_attr_sf.h"
  22#include "xfs_attr.h"
  23#include "xfs_attr_remote.h"
  24#include "xfs_attr_leaf.h"
  25#include "xfs_error.h"
  26#include "xfs_trace.h"
  27#include "xfs_buf_item.h"
  28#include "xfs_dir2.h"
  29#include "xfs_log.h"
  30#include "xfs_ag.h"
  31#include "xfs_errortag.h"
  32
  33
  34/*
  35 * xfs_attr_leaf.c
  36 *
  37 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
  38 */
  39
  40/*========================================================================
  41 * Function prototypes for the kernel.
  42 *========================================================================*/
  43
  44/*
  45 * Routines used for growing the Btree.
  46 */
  47STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
  48				 xfs_dablk_t which_block, struct xfs_buf **bpp);
  49STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
  50				   struct xfs_attr3_icleaf_hdr *ichdr,
  51				   struct xfs_da_args *args, int freemap_index);
  52STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
  53				   struct xfs_attr3_icleaf_hdr *ichdr,
  54				   struct xfs_buf *leaf_buffer);
  55STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
  56						   xfs_da_state_blk_t *blk1,
  57						   xfs_da_state_blk_t *blk2);
  58STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
  59			xfs_da_state_blk_t *leaf_blk_1,
  60			struct xfs_attr3_icleaf_hdr *ichdr1,
  61			xfs_da_state_blk_t *leaf_blk_2,
  62			struct xfs_attr3_icleaf_hdr *ichdr2,
  63			int *number_entries_in_blk1,
  64			int *number_usedbytes_in_blk1);
  65
  66/*
  67 * Utility routines.
  68 */
  69STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
  70			struct xfs_attr_leafblock *src_leaf,
  71			struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
  72			struct xfs_attr_leafblock *dst_leaf,
  73			struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
  74			int move_count);
  75STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
  76
  77/*
  78 * attr3 block 'firstused' conversion helpers.
  79 *
  80 * firstused refers to the offset of the first used byte of the nameval region
  81 * of an attr leaf block. The region starts at the tail of the block and expands
  82 * backwards towards the middle. As such, firstused is initialized to the block
  83 * size for an empty leaf block and is reduced from there.
  84 *
  85 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
  86 * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
  87 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
  88 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
  89 * the attr block size. The following helpers manage the conversion between the
  90 * in-core and on-disk formats.
  91 */
  92
  93static void
  94xfs_attr3_leaf_firstused_from_disk(
  95	struct xfs_da_geometry		*geo,
  96	struct xfs_attr3_icleaf_hdr	*to,
  97	struct xfs_attr_leafblock	*from)
  98{
  99	struct xfs_attr3_leaf_hdr	*hdr3;
 100
 101	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
 102		hdr3 = (struct xfs_attr3_leaf_hdr *) from;
 103		to->firstused = be16_to_cpu(hdr3->firstused);
 104	} else {
 105		to->firstused = be16_to_cpu(from->hdr.firstused);
 106	}
 107
 108	/*
 109	 * Convert from the magic fsb size value to actual blocksize. This
 110	 * should only occur for empty blocks when the block size overflows
 111	 * 16-bits.
 112	 */
 113	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
 114		ASSERT(!to->count && !to->usedbytes);
 115		ASSERT(geo->blksize > USHRT_MAX);
 116		to->firstused = geo->blksize;
 117	}
 118}
 119
 120static void
 121xfs_attr3_leaf_firstused_to_disk(
 122	struct xfs_da_geometry		*geo,
 123	struct xfs_attr_leafblock	*to,
 124	struct xfs_attr3_icleaf_hdr	*from)
 125{
 126	struct xfs_attr3_leaf_hdr	*hdr3;
 127	uint32_t			firstused;
 128
 129	/* magic value should only be seen on disk */
 130	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
 131
 132	/*
 133	 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
 134	 * value. This only overflows at the max supported value of 64k. Use the
 135	 * magic on-disk value to represent block size in this case.
 136	 */
 137	firstused = from->firstused;
 138	if (firstused > USHRT_MAX) {
 139		ASSERT(from->firstused == geo->blksize);
 140		firstused = XFS_ATTR3_LEAF_NULLOFF;
 141	}
 142
 143	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
 144		hdr3 = (struct xfs_attr3_leaf_hdr *) to;
 145		hdr3->firstused = cpu_to_be16(firstused);
 146	} else {
 147		to->hdr.firstused = cpu_to_be16(firstused);
 148	}
 149}
 150
 151void
 152xfs_attr3_leaf_hdr_from_disk(
 153	struct xfs_da_geometry		*geo,
 154	struct xfs_attr3_icleaf_hdr	*to,
 155	struct xfs_attr_leafblock	*from)
 156{
 157	int	i;
 158
 159	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
 160	       from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
 161
 162	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
 163		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
 164
 165		to->forw = be32_to_cpu(hdr3->info.hdr.forw);
 166		to->back = be32_to_cpu(hdr3->info.hdr.back);
 167		to->magic = be16_to_cpu(hdr3->info.hdr.magic);
 168		to->count = be16_to_cpu(hdr3->count);
 169		to->usedbytes = be16_to_cpu(hdr3->usedbytes);
 170		xfs_attr3_leaf_firstused_from_disk(geo, to, from);
 171		to->holes = hdr3->holes;
 172
 173		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
 174			to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
 175			to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
 176		}
 177		return;
 178	}
 179	to->forw = be32_to_cpu(from->hdr.info.forw);
 180	to->back = be32_to_cpu(from->hdr.info.back);
 181	to->magic = be16_to_cpu(from->hdr.info.magic);
 182	to->count = be16_to_cpu(from->hdr.count);
 183	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
 184	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
 185	to->holes = from->hdr.holes;
 186
 187	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
 188		to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
 189		to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
 190	}
 191}
 192
 193void
 194xfs_attr3_leaf_hdr_to_disk(
 195	struct xfs_da_geometry		*geo,
 196	struct xfs_attr_leafblock	*to,
 197	struct xfs_attr3_icleaf_hdr	*from)
 198{
 199	int				i;
 200
 201	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
 202	       from->magic == XFS_ATTR3_LEAF_MAGIC);
 203
 204	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
 205		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
 206
 207		hdr3->info.hdr.forw = cpu_to_be32(from->forw);
 208		hdr3->info.hdr.back = cpu_to_be32(from->back);
 209		hdr3->info.hdr.magic = cpu_to_be16(from->magic);
 210		hdr3->count = cpu_to_be16(from->count);
 211		hdr3->usedbytes = cpu_to_be16(from->usedbytes);
 212		xfs_attr3_leaf_firstused_to_disk(geo, to, from);
 213		hdr3->holes = from->holes;
 214		hdr3->pad1 = 0;
 215
 216		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
 217			hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
 218			hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
 219		}
 220		return;
 221	}
 222	to->hdr.info.forw = cpu_to_be32(from->forw);
 223	to->hdr.info.back = cpu_to_be32(from->back);
 224	to->hdr.info.magic = cpu_to_be16(from->magic);
 225	to->hdr.count = cpu_to_be16(from->count);
 226	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
 227	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
 228	to->hdr.holes = from->holes;
 229	to->hdr.pad1 = 0;
 230
 231	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
 232		to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
 233		to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
 234	}
 235}
 236
 237static xfs_failaddr_t
 238xfs_attr3_leaf_verify_entry(
 239	struct xfs_mount			*mp,
 240	char					*buf_end,
 241	struct xfs_attr_leafblock		*leaf,
 242	struct xfs_attr3_icleaf_hdr		*leafhdr,
 243	struct xfs_attr_leaf_entry		*ent,
 244	int					idx,
 245	__u32					*last_hashval)
 246{
 247	struct xfs_attr_leaf_name_local		*lentry;
 248	struct xfs_attr_leaf_name_remote	*rentry;
 249	char					*name_end;
 250	unsigned int				nameidx;
 251	unsigned int				namesize;
 252	__u32					hashval;
 253
 254	/* hash order check */
 255	hashval = be32_to_cpu(ent->hashval);
 256	if (hashval < *last_hashval)
 257		return __this_address;
 258	*last_hashval = hashval;
 259
 260	nameidx = be16_to_cpu(ent->nameidx);
 261	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
 262		return __this_address;
 263
 264	/*
 265	 * Check the name information.  The namelen fields are u8 so we can't
 266	 * possibly exceed the maximum name length of 255 bytes.
 267	 */
 268	if (ent->flags & XFS_ATTR_LOCAL) {
 269		lentry = xfs_attr3_leaf_name_local(leaf, idx);
 270		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
 271				be16_to_cpu(lentry->valuelen));
 272		name_end = (char *)lentry + namesize;
 273		if (lentry->namelen == 0)
 274			return __this_address;
 275	} else {
 276		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
 277		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
 278		name_end = (char *)rentry + namesize;
 279		if (rentry->namelen == 0)
 280			return __this_address;
 281		if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
 282		    rentry->valueblk == 0)
 283			return __this_address;
 284	}
 285
 286	if (name_end > buf_end)
 287		return __this_address;
 288
 289	return NULL;
 290}
 291
 292/*
 293 * Validate an attribute leaf block.
 294 *
 295 * Empty leaf blocks can occur under the following circumstances:
 296 *
 297 * 1. setxattr adds a new extended attribute to a file;
 298 * 2. The file has zero existing attributes;
 299 * 3. The attribute is too large to fit in the attribute fork;
 300 * 4. The attribute is small enough to fit in a leaf block;
 301 * 5. A log flush occurs after committing the transaction that creates
 302 *    the (empty) leaf block; and
 303 * 6. The filesystem goes down after the log flush but before the new
 304 *    attribute can be committed to the leaf block.
 305 *
 306 * Hence we need to ensure that we don't fail the validation purely
 307 * because the leaf is empty.
 308 */
 309static xfs_failaddr_t
 310xfs_attr3_leaf_verify(
 311	struct xfs_buf			*bp)
 312{
 313	struct xfs_attr3_icleaf_hdr	ichdr;
 314	struct xfs_mount		*mp = bp->b_mount;
 315	struct xfs_attr_leafblock	*leaf = bp->b_addr;
 316	struct xfs_attr_leaf_entry	*entries;
 317	struct xfs_attr_leaf_entry	*ent;
 318	char				*buf_end;
 319	uint32_t			end;	/* must be 32bit - see below */
 320	__u32				last_hashval = 0;
 321	int				i;
 322	xfs_failaddr_t			fa;
 323
 324	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
 325
 326	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
 327	if (fa)
 328		return fa;
 329
 330	/*
 331	 * firstused is the block offset of the first name info structure.
 332	 * Make sure it doesn't go off the block or crash into the header.
 333	 */
 334	if (ichdr.firstused > mp->m_attr_geo->blksize)
 335		return __this_address;
 336	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
 337		return __this_address;
 338
 339	/* Make sure the entries array doesn't crash into the name info. */
 340	entries = xfs_attr3_leaf_entryp(bp->b_addr);
 341	if ((char *)&entries[ichdr.count] >
 342	    (char *)bp->b_addr + ichdr.firstused)
 343		return __this_address;
 344
 345	/*
 346	 * NOTE: This verifier historically failed empty leaf buffers because
 347	 * we expect the fork to be in another format. Empty attr fork format
 348	 * conversions are possible during xattr set, however, and format
 349	 * conversion is not atomic with the xattr set that triggers it. We
 350	 * cannot assume leaf blocks are non-empty until that is addressed.
 351	*/
 352	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
 353	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
 354		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
 355				ent, i, &last_hashval);
 356		if (fa)
 357			return fa;
 358	}
 359
 360	/*
 361	 * Quickly check the freemap information.  Attribute data has to be
 362	 * aligned to 4-byte boundaries, and likewise for the free space.
 363	 *
 364	 * Note that for 64k block size filesystems, the freemap entries cannot
 365	 * overflow as they are only be16 fields. However, when checking end
 366	 * pointer of the freemap, we have to be careful to detect overflows and
 367	 * so use uint32_t for those checks.
 368	 */
 369	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
 370		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
 371			return __this_address;
 372		if (ichdr.freemap[i].base & 0x3)
 373			return __this_address;
 374		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
 375			return __this_address;
 376		if (ichdr.freemap[i].size & 0x3)
 377			return __this_address;
 378
 379		/* be care of 16 bit overflows here */
 380		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
 381		if (end < ichdr.freemap[i].base)
 382			return __this_address;
 383		if (end > mp->m_attr_geo->blksize)
 384			return __this_address;
 385	}
 386
 387	return NULL;
 388}
 389
 390static void
 391xfs_attr3_leaf_write_verify(
 392	struct xfs_buf	*bp)
 393{
 394	struct xfs_mount	*mp = bp->b_mount;
 395	struct xfs_buf_log_item	*bip = bp->b_log_item;
 396	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
 397	xfs_failaddr_t		fa;
 398
 399	fa = xfs_attr3_leaf_verify(bp);
 400	if (fa) {
 401		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 402		return;
 403	}
 404
 405	if (!xfs_has_crc(mp))
 406		return;
 407
 408	if (bip)
 409		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 410
 411	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
 412}
 413
 414/*
 415 * leaf/node format detection on trees is sketchy, so a node read can be done on
 416 * leaf level blocks when detection identifies the tree as a node format tree
 417 * incorrectly. In this case, we need to swap the verifier to match the correct
 418 * format of the block being read.
 419 */
 420static void
 421xfs_attr3_leaf_read_verify(
 422	struct xfs_buf		*bp)
 423{
 424	struct xfs_mount	*mp = bp->b_mount;
 425	xfs_failaddr_t		fa;
 426
 427	if (xfs_has_crc(mp) &&
 428	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
 429		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 430	else {
 431		fa = xfs_attr3_leaf_verify(bp);
 432		if (fa)
 433			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 434	}
 435}
 436
 437const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
 438	.name = "xfs_attr3_leaf",
 439	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
 440		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
 441	.verify_read = xfs_attr3_leaf_read_verify,
 442	.verify_write = xfs_attr3_leaf_write_verify,
 443	.verify_struct = xfs_attr3_leaf_verify,
 444};
 445
 446int
 447xfs_attr3_leaf_read(
 448	struct xfs_trans	*tp,
 449	struct xfs_inode	*dp,
 450	xfs_dablk_t		bno,
 451	struct xfs_buf		**bpp)
 452{
 453	int			err;
 454
 455	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
 456			&xfs_attr3_leaf_buf_ops);
 457	if (!err && tp && *bpp)
 458		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
 459	return err;
 460}
 461
 462/*========================================================================
 463 * Namespace helper routines
 464 *========================================================================*/
 465
 466/*
 467 * If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
 468 * flag on disk - if there's an incomplete attr then recovery needs to tear it
 469 * down. If there's no incomplete attr, then recovery needs to tear that attr
 470 * down to replace it with the attr that has been logged. In this case, the
 471 * INCOMPLETE flag will not be set in attr->attr_filter, but rather
 472 * XFS_DA_OP_RECOVERY will be set in args->op_flags.
 473 */
 474static bool
 475xfs_attr_match(
 476	struct xfs_da_args	*args,
 477	uint8_t			namelen,
 478	unsigned char		*name,
 479	int			flags)
 480{
 481
 482	if (args->namelen != namelen)
 483		return false;
 484	if (memcmp(args->name, name, namelen) != 0)
 485		return false;
 486
 487	/* Recovery ignores the INCOMPLETE flag. */
 488	if ((args->op_flags & XFS_DA_OP_RECOVERY) &&
 489	    args->attr_filter == (flags & XFS_ATTR_NSP_ONDISK_MASK))
 490		return true;
 491
 492	/* All remaining matches need to be filtered by INCOMPLETE state. */
 493	if (args->attr_filter !=
 494	    (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
 495		return false;
 496	return true;
 497}
 498
 499static int
 500xfs_attr_copy_value(
 501	struct xfs_da_args	*args,
 502	unsigned char		*value,
 503	int			valuelen)
 504{
 505	/*
 506	 * No copy if all we have to do is get the length
 507	 */
 508	if (!args->valuelen) {
 509		args->valuelen = valuelen;
 510		return 0;
 511	}
 512
 513	/*
 514	 * No copy if the length of the existing buffer is too small
 515	 */
 516	if (args->valuelen < valuelen) {
 517		args->valuelen = valuelen;
 518		return -ERANGE;
 519	}
 520
 521	if (!args->value) {
 522		args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
 523		if (!args->value)
 524			return -ENOMEM;
 525	}
 526	args->valuelen = valuelen;
 527
 528	/* remote block xattr requires IO for copy-in */
 529	if (args->rmtblkno)
 530		return xfs_attr_rmtval_get(args);
 531
 532	/*
 533	 * This is to prevent a GCC warning because the remote xattr case
 534	 * doesn't have a value to pass in. In that case, we never reach here,
 535	 * but GCC can't work that out and so throws a "passing NULL to
 536	 * memcpy" warning.
 537	 */
 538	if (!value)
 539		return -EINVAL;
 540	memcpy(args->value, value, valuelen);
 541	return 0;
 542}
 543
 544/*========================================================================
 545 * External routines when attribute fork size < XFS_LITINO(mp).
 546 *========================================================================*/
 547
 548/*
 549 * Query whether the total requested number of attr fork bytes of extended
 550 * attribute space will be able to fit inline.
 551 *
 552 * Returns zero if not, else the i_forkoff fork offset to be used in the
 553 * literal area for attribute data once the new bytes have been added.
 554 *
 555 * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
 556 * special case for dev/uuid inodes, they have fixed size data forks.
 557 */
 558int
 559xfs_attr_shortform_bytesfit(
 560	struct xfs_inode	*dp,
 561	int			bytes)
 562{
 563	struct xfs_mount	*mp = dp->i_mount;
 564	int64_t			dsize;
 565	int			minforkoff;
 566	int			maxforkoff;
 567	int			offset;
 568
 569	/*
 570	 * Check if the new size could fit at all first:
 571	 */
 572	if (bytes > XFS_LITINO(mp))
 573		return 0;
 574
 575	/* rounded down */
 576	offset = (XFS_LITINO(mp) - bytes) >> 3;
 577
 578	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
 579		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
 580		return (offset >= minforkoff) ? minforkoff : 0;
 581	}
 582
 583	/*
 584	 * If the requested numbers of bytes is smaller or equal to the
 585	 * current attribute fork size we can always proceed.
 586	 *
 587	 * Note that if_bytes in the data fork might actually be larger than
 588	 * the current data fork size is due to delalloc extents. In that
 589	 * case either the extent count will go down when they are converted
 590	 * to real extents, or the delalloc conversion will take care of the
 591	 * literal area rebalancing.
 592	 */
 593	if (bytes <= xfs_inode_attr_fork_size(dp))
 594		return dp->i_forkoff;
 595
 596	/*
 597	 * For attr2 we can try to move the forkoff if there is space in the
 598	 * literal area, but for the old format we are done if there is no
 599	 * space in the fixed attribute fork.
 600	 */
 601	if (!xfs_has_attr2(mp))
 602		return 0;
 603
 604	dsize = dp->i_df.if_bytes;
 605
 606	switch (dp->i_df.if_format) {
 607	case XFS_DINODE_FMT_EXTENTS:
 608		/*
 609		 * If there is no attr fork and the data fork is extents,
 610		 * determine if creating the default attr fork will result
 611		 * in the extents form migrating to btree. If so, the
 612		 * minimum offset only needs to be the space required for
 613		 * the btree root.
 614		 */
 615		if (!dp->i_forkoff && dp->i_df.if_bytes >
 616		    xfs_default_attroffset(dp))
 617			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
 618		break;
 619	case XFS_DINODE_FMT_BTREE:
 620		/*
 621		 * If we have a data btree then keep forkoff if we have one,
 622		 * otherwise we are adding a new attr, so then we set
 623		 * minforkoff to where the btree root can finish so we have
 624		 * plenty of room for attrs
 625		 */
 626		if (dp->i_forkoff) {
 627			if (offset < dp->i_forkoff)
 628				return 0;
 629			return dp->i_forkoff;
 630		}
 631		dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
 632		break;
 633	}
 634
 635	/*
 636	 * A data fork btree root must have space for at least
 637	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
 638	 */
 639	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
 640	minforkoff = roundup(minforkoff, 8) >> 3;
 641
 642	/* attr fork btree root can have at least this many key/ptr pairs */
 643	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
 644	maxforkoff = maxforkoff >> 3;	/* rounded down */
 645
 646	if (offset >= maxforkoff)
 647		return maxforkoff;
 648	if (offset >= minforkoff)
 649		return offset;
 650	return 0;
 651}
 652
 653/*
 654 * Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
 655 * - noattr2 mount option is set,
 656 * - on-disk version bit says it is already set, or
 657 * - the attr2 mount option is not set to enable automatic upgrade from attr1.
 658 */
 659STATIC void
 660xfs_sbversion_add_attr2(
 661	struct xfs_mount	*mp,
 662	struct xfs_trans	*tp)
 663{
 664	if (xfs_has_noattr2(mp))
 665		return;
 666	if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
 667		return;
 668	if (!xfs_has_attr2(mp))
 669		return;
 670
 671	spin_lock(&mp->m_sb_lock);
 672	xfs_add_attr2(mp);
 673	spin_unlock(&mp->m_sb_lock);
 674	xfs_log_sb(tp);
 675}
 676
 677/*
 678 * Create the initial contents of a shortform attribute list.
 679 */
 680void
 681xfs_attr_shortform_create(
 682	struct xfs_da_args	*args)
 683{
 684	struct xfs_inode	*dp = args->dp;
 685	struct xfs_ifork	*ifp = &dp->i_af;
 686	struct xfs_attr_sf_hdr	*hdr;
 687
 688	trace_xfs_attr_sf_create(args);
 689
 690	ASSERT(ifp->if_bytes == 0);
 691	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
 692		ifp->if_format = XFS_DINODE_FMT_LOCAL;
 693	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
 694	hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
 695	memset(hdr, 0, sizeof(*hdr));
 696	hdr->totsize = cpu_to_be16(sizeof(*hdr));
 697	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
 698}
 699
 700/*
 701 * Return -EEXIST if attr is found, or -ENOATTR if not
 702 * args:  args containing attribute name and namelen
 703 * sfep:  If not null, pointer will be set to the last attr entry found on
 704	  -EEXIST.  On -ENOATTR pointer is left at the last entry in the list
 705 * basep: If not null, pointer is set to the byte offset of the entry in the
 706 *	  list on -EEXIST.  On -ENOATTR, pointer is left at the byte offset of
 707 *	  the last entry in the list
 708 */
 709int
 710xfs_attr_sf_findname(
 711	struct xfs_da_args	 *args,
 712	struct xfs_attr_sf_entry **sfep,
 713	unsigned int		 *basep)
 714{
 715	struct xfs_attr_shortform *sf;
 716	struct xfs_attr_sf_entry *sfe;
 717	unsigned int		base = sizeof(struct xfs_attr_sf_hdr);
 718	int			size = 0;
 719	int			end;
 720	int			i;
 721
 722	sf = (struct xfs_attr_shortform *)args->dp->i_af.if_u1.if_data;
 723	sfe = &sf->list[0];
 724	end = sf->hdr.count;
 725	for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe),
 726			     base += size, i++) {
 727		size = xfs_attr_sf_entsize(sfe);
 728		if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
 729				    sfe->flags))
 730			continue;
 731		break;
 732	}
 733
 734	if (sfep != NULL)
 735		*sfep = sfe;
 736
 737	if (basep != NULL)
 738		*basep = base;
 739
 740	if (i == end)
 741		return -ENOATTR;
 742	return -EEXIST;
 743}
 744
 745/*
 746 * Add a name/value pair to the shortform attribute list.
 747 * Overflow from the inode has already been checked for.
 748 */
 749void
 750xfs_attr_shortform_add(
 751	struct xfs_da_args		*args,
 752	int				forkoff)
 753{
 754	struct xfs_attr_shortform	*sf;
 755	struct xfs_attr_sf_entry	*sfe;
 756	int				offset, size;
 757	struct xfs_mount		*mp;
 758	struct xfs_inode		*dp;
 759	struct xfs_ifork		*ifp;
 760
 761	trace_xfs_attr_sf_add(args);
 762
 763	dp = args->dp;
 764	mp = dp->i_mount;
 765	dp->i_forkoff = forkoff;
 766
 767	ifp = &dp->i_af;
 768	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
 769	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
 770	if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
 771		ASSERT(0);
 772
 773	offset = (char *)sfe - (char *)sf;
 774	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
 775	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
 776	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
 777	sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset);
 778
 779	sfe->namelen = args->namelen;
 780	sfe->valuelen = args->valuelen;
 781	sfe->flags = args->attr_filter;
 782	memcpy(sfe->nameval, args->name, args->namelen);
 783	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
 784	sf->hdr.count++;
 785	be16_add_cpu(&sf->hdr.totsize, size);
 786	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
 787
 788	xfs_sbversion_add_attr2(mp, args->trans);
 789}
 790
 791/*
 792 * After the last attribute is removed revert to original inode format,
 793 * making all literal area available to the data fork once more.
 794 */
 795void
 796xfs_attr_fork_remove(
 797	struct xfs_inode	*ip,
 798	struct xfs_trans	*tp)
 799{
 800	ASSERT(ip->i_af.if_nextents == 0);
 801
 802	xfs_ifork_zap_attr(ip);
 803	ip->i_forkoff = 0;
 804	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 805}
 806
 807/*
 808 * Remove an attribute from the shortform attribute list structure.
 809 */
 810int
 811xfs_attr_sf_removename(
 812	struct xfs_da_args		*args)
 813{
 814	struct xfs_attr_shortform	*sf;
 815	struct xfs_attr_sf_entry	*sfe;
 816	int				size = 0, end, totsize;
 817	unsigned int			base;
 818	struct xfs_mount		*mp;
 819	struct xfs_inode		*dp;
 820	int				error;
 821
 822	trace_xfs_attr_sf_remove(args);
 823
 824	dp = args->dp;
 825	mp = dp->i_mount;
 826	sf = (struct xfs_attr_shortform *)dp->i_af.if_u1.if_data;
 827
 828	error = xfs_attr_sf_findname(args, &sfe, &base);
 829
 830	/*
 831	 * If we are recovering an operation, finding nothing to
 832	 * remove is not an error - it just means there was nothing
 833	 * to clean up.
 834	 */
 835	if (error == -ENOATTR && (args->op_flags & XFS_DA_OP_RECOVERY))
 836		return 0;
 837	if (error != -EEXIST)
 838		return error;
 839	size = xfs_attr_sf_entsize(sfe);
 840
 841	/*
 842	 * Fix up the attribute fork data, covering the hole
 843	 */
 844	end = base + size;
 845	totsize = be16_to_cpu(sf->hdr.totsize);
 846	if (end != totsize)
 847		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
 848	sf->hdr.count--;
 849	be16_add_cpu(&sf->hdr.totsize, -size);
 850
 851	/*
 852	 * Fix up the start offset of the attribute fork
 853	 */
 854	totsize -= size;
 855	if (totsize == sizeof(xfs_attr_sf_hdr_t) && xfs_has_attr2(mp) &&
 856	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
 857	    !(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE))) {
 858		xfs_attr_fork_remove(dp, args->trans);
 859	} else {
 860		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
 861		dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
 862		ASSERT(dp->i_forkoff);
 863		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
 864				(args->op_flags & XFS_DA_OP_ADDNAME) ||
 865				!xfs_has_attr2(mp) ||
 866				dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
 867		xfs_trans_log_inode(args->trans, dp,
 868					XFS_ILOG_CORE | XFS_ILOG_ADATA);
 869	}
 870
 871	xfs_sbversion_add_attr2(mp, args->trans);
 872
 873	return 0;
 874}
 875
 876/*
 877 * Look up a name in a shortform attribute list structure.
 878 */
 879/*ARGSUSED*/
 880int
 881xfs_attr_shortform_lookup(xfs_da_args_t *args)
 882{
 883	struct xfs_attr_shortform *sf;
 884	struct xfs_attr_sf_entry *sfe;
 885	int i;
 886	struct xfs_ifork *ifp;
 887
 888	trace_xfs_attr_sf_lookup(args);
 889
 890	ifp = &args->dp->i_af;
 891	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
 892	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
 893	sfe = &sf->list[0];
 894	for (i = 0; i < sf->hdr.count;
 895				sfe = xfs_attr_sf_nextentry(sfe), i++) {
 896		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
 897				sfe->flags))
 898			return -EEXIST;
 899	}
 900	return -ENOATTR;
 901}
 902
 903/*
 904 * Retrieve the attribute value and length.
 905 *
 906 * If args->valuelen is zero, only the length needs to be returned.  Unlike a
 907 * lookup, we only return an error if the attribute does not exist or we can't
 908 * retrieve the value.
 909 */
 910int
 911xfs_attr_shortform_getvalue(
 912	struct xfs_da_args	*args)
 913{
 914	struct xfs_attr_shortform *sf;
 915	struct xfs_attr_sf_entry *sfe;
 916	int			i;
 917
 918	ASSERT(args->dp->i_af.if_format == XFS_DINODE_FMT_LOCAL);
 919	sf = (struct xfs_attr_shortform *)args->dp->i_af.if_u1.if_data;
 920	sfe = &sf->list[0];
 921	for (i = 0; i < sf->hdr.count;
 922				sfe = xfs_attr_sf_nextentry(sfe), i++) {
 923		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
 924				sfe->flags))
 925			return xfs_attr_copy_value(args,
 926				&sfe->nameval[args->namelen], sfe->valuelen);
 927	}
 928	return -ENOATTR;
 929}
 930
 931/* Convert from using the shortform to the leaf format. */
 932int
 933xfs_attr_shortform_to_leaf(
 934	struct xfs_da_args		*args)
 935{
 936	struct xfs_inode		*dp;
 937	struct xfs_attr_shortform	*sf;
 938	struct xfs_attr_sf_entry	*sfe;
 939	struct xfs_da_args		nargs;
 940	char				*tmpbuffer;
 941	int				error, i, size;
 942	xfs_dablk_t			blkno;
 943	struct xfs_buf			*bp;
 944	struct xfs_ifork		*ifp;
 945
 946	trace_xfs_attr_sf_to_leaf(args);
 947
 948	dp = args->dp;
 949	ifp = &dp->i_af;
 950	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
 951	size = be16_to_cpu(sf->hdr.totsize);
 952	tmpbuffer = kmem_alloc(size, 0);
 953	ASSERT(tmpbuffer != NULL);
 954	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
 955	sf = (struct xfs_attr_shortform *)tmpbuffer;
 956
 957	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
 958	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
 959
 960	bp = NULL;
 961	error = xfs_da_grow_inode(args, &blkno);
 962	if (error)
 963		goto out;
 964
 965	ASSERT(blkno == 0);
 966	error = xfs_attr3_leaf_create(args, blkno, &bp);
 967	if (error)
 968		goto out;
 969
 970	memset((char *)&nargs, 0, sizeof(nargs));
 971	nargs.dp = dp;
 972	nargs.geo = args->geo;
 973	nargs.total = args->total;
 974	nargs.whichfork = XFS_ATTR_FORK;
 975	nargs.trans = args->trans;
 976	nargs.op_flags = XFS_DA_OP_OKNOENT;
 977
 978	sfe = &sf->list[0];
 979	for (i = 0; i < sf->hdr.count; i++) {
 980		nargs.name = sfe->nameval;
 981		nargs.namelen = sfe->namelen;
 982		nargs.value = &sfe->nameval[nargs.namelen];
 983		nargs.valuelen = sfe->valuelen;
 984		nargs.hashval = xfs_da_hashname(sfe->nameval,
 985						sfe->namelen);
 986		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
 987		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
 988		ASSERT(error == -ENOATTR);
 989		error = xfs_attr3_leaf_add(bp, &nargs);
 990		ASSERT(error != -ENOSPC);
 991		if (error)
 992			goto out;
 993		sfe = xfs_attr_sf_nextentry(sfe);
 994	}
 995	error = 0;
 996out:
 997	kmem_free(tmpbuffer);
 998	return error;
 999}
1000
1001/*
1002 * Check a leaf attribute block to see if all the entries would fit into
1003 * a shortform attribute list.
1004 */
1005int
1006xfs_attr_shortform_allfit(
1007	struct xfs_buf		*bp,
1008	struct xfs_inode	*dp)
1009{
1010	struct xfs_attr_leafblock *leaf;
1011	struct xfs_attr_leaf_entry *entry;
1012	xfs_attr_leaf_name_local_t *name_loc;
1013	struct xfs_attr3_icleaf_hdr leafhdr;
1014	int			bytes;
1015	int			i;
1016	struct xfs_mount	*mp = bp->b_mount;
1017
1018	leaf = bp->b_addr;
1019	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
1020	entry = xfs_attr3_leaf_entryp(leaf);
1021
1022	bytes = sizeof(struct xfs_attr_sf_hdr);
1023	for (i = 0; i < leafhdr.count; entry++, i++) {
1024		if (entry->flags & XFS_ATTR_INCOMPLETE)
1025			continue;		/* don't copy partial entries */
1026		if (!(entry->flags & XFS_ATTR_LOCAL))
1027			return 0;
1028		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1029		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
1030			return 0;
1031		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1032			return 0;
1033		bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1034					be16_to_cpu(name_loc->valuelen));
1035	}
1036	if (xfs_has_attr2(dp->i_mount) &&
1037	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1038	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
1039		return -1;
1040	return xfs_attr_shortform_bytesfit(dp, bytes);
1041}
1042
1043/* Verify the consistency of an inline attribute fork. */
1044xfs_failaddr_t
1045xfs_attr_shortform_verify(
1046	struct xfs_inode		*ip)
1047{
1048	struct xfs_attr_shortform	*sfp;
1049	struct xfs_attr_sf_entry	*sfep;
1050	struct xfs_attr_sf_entry	*next_sfep;
1051	char				*endp;
1052	struct xfs_ifork		*ifp;
1053	int				i;
1054	int64_t				size;
1055
1056	ASSERT(ip->i_af.if_format == XFS_DINODE_FMT_LOCAL);
1057	ifp = xfs_ifork_ptr(ip, XFS_ATTR_FORK);
1058	sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1059	size = ifp->if_bytes;
1060
1061	/*
1062	 * Give up if the attribute is way too short.
1063	 */
1064	if (size < sizeof(struct xfs_attr_sf_hdr))
1065		return __this_address;
1066
1067	endp = (char *)sfp + size;
1068
1069	/* Check all reported entries */
1070	sfep = &sfp->list[0];
1071	for (i = 0; i < sfp->hdr.count; i++) {
1072		/*
1073		 * struct xfs_attr_sf_entry has a variable length.
1074		 * Check the fixed-offset parts of the structure are
1075		 * within the data buffer.
1076		 * xfs_attr_sf_entry is defined with a 1-byte variable
1077		 * array at the end, so we must subtract that off.
1078		 */
1079		if (((char *)sfep + sizeof(*sfep)) >= endp)
1080			return __this_address;
1081
1082		/* Don't allow names with known bad length. */
1083		if (sfep->namelen == 0)
1084			return __this_address;
1085
1086		/*
1087		 * Check that the variable-length part of the structure is
1088		 * within the data buffer.  The next entry starts after the
1089		 * name component, so nextentry is an acceptable test.
1090		 */
1091		next_sfep = xfs_attr_sf_nextentry(sfep);
1092		if ((char *)next_sfep > endp)
1093			return __this_address;
1094
1095		/*
1096		 * Check for unknown flags.  Short form doesn't support
1097		 * the incomplete or local bits, so we can use the namespace
1098		 * mask here.
1099		 */
1100		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1101			return __this_address;
1102
1103		/*
1104		 * Check for invalid namespace combinations.  We only allow
1105		 * one namespace flag per xattr, so we can just count the
1106		 * bits (i.e. hweight) here.
1107		 */
1108		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1109			return __this_address;
1110
1111		sfep = next_sfep;
1112	}
1113	if ((void *)sfep != (void *)endp)
1114		return __this_address;
1115
1116	return NULL;
1117}
1118
1119/*
1120 * Convert a leaf attribute list to shortform attribute list
1121 */
1122int
1123xfs_attr3_leaf_to_shortform(
1124	struct xfs_buf		*bp,
1125	struct xfs_da_args	*args,
1126	int			forkoff)
1127{
1128	struct xfs_attr_leafblock *leaf;
1129	struct xfs_attr3_icleaf_hdr ichdr;
1130	struct xfs_attr_leaf_entry *entry;
1131	struct xfs_attr_leaf_name_local *name_loc;
1132	struct xfs_da_args	nargs;
1133	struct xfs_inode	*dp = args->dp;
1134	char			*tmpbuffer;
1135	int			error;
1136	int			i;
1137
1138	trace_xfs_attr_leaf_to_sf(args);
1139
1140	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1141	if (!tmpbuffer)
1142		return -ENOMEM;
1143
1144	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1145
1146	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1147	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1148	entry = xfs_attr3_leaf_entryp(leaf);
1149
1150	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1151	memset(bp->b_addr, 0, args->geo->blksize);
1152
1153	/*
1154	 * Clean out the prior contents of the attribute list.
1155	 */
1156	error = xfs_da_shrink_inode(args, 0, bp);
1157	if (error)
1158		goto out;
1159
1160	if (forkoff == -1) {
1161		/*
1162		 * Don't remove the attr fork if this operation is the first
1163		 * part of a attr replace operations. We're going to add a new
1164		 * attr immediately, so we need to keep the attr fork around in
1165		 * this case.
1166		 */
1167		if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1168			ASSERT(xfs_has_attr2(dp->i_mount));
1169			ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1170			xfs_attr_fork_remove(dp, args->trans);
1171		}
1172		goto out;
1173	}
1174
1175	xfs_attr_shortform_create(args);
1176
1177	/*
1178	 * Copy the attributes
1179	 */
1180	memset((char *)&nargs, 0, sizeof(nargs));
1181	nargs.geo = args->geo;
1182	nargs.dp = dp;
1183	nargs.total = args->total;
1184	nargs.whichfork = XFS_ATTR_FORK;
1185	nargs.trans = args->trans;
1186	nargs.op_flags = XFS_DA_OP_OKNOENT;
1187
1188	for (i = 0; i < ichdr.count; entry++, i++) {
1189		if (entry->flags & XFS_ATTR_INCOMPLETE)
1190			continue;	/* don't copy partial entries */
1191		if (!entry->nameidx)
1192			continue;
1193		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1194		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1195		nargs.name = name_loc->nameval;
1196		nargs.namelen = name_loc->namelen;
1197		nargs.value = &name_loc->nameval[nargs.namelen];
1198		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1199		nargs.hashval = be32_to_cpu(entry->hashval);
1200		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1201		xfs_attr_shortform_add(&nargs, forkoff);
1202	}
1203	error = 0;
1204
1205out:
1206	kmem_free(tmpbuffer);
1207	return error;
1208}
1209
1210/*
1211 * Convert from using a single leaf to a root node and a leaf.
1212 */
1213int
1214xfs_attr3_leaf_to_node(
1215	struct xfs_da_args	*args)
1216{
1217	struct xfs_attr_leafblock *leaf;
1218	struct xfs_attr3_icleaf_hdr icleafhdr;
1219	struct xfs_attr_leaf_entry *entries;
1220	struct xfs_da3_icnode_hdr icnodehdr;
1221	struct xfs_da_intnode	*node;
1222	struct xfs_inode	*dp = args->dp;
1223	struct xfs_mount	*mp = dp->i_mount;
1224	struct xfs_buf		*bp1 = NULL;
1225	struct xfs_buf		*bp2 = NULL;
1226	xfs_dablk_t		blkno;
1227	int			error;
1228
1229	trace_xfs_attr_leaf_to_node(args);
1230
1231	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1232		error = -EIO;
1233		goto out;
1234	}
1235
1236	error = xfs_da_grow_inode(args, &blkno);
1237	if (error)
1238		goto out;
1239	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1240	if (error)
1241		goto out;
1242
1243	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1244	if (error)
1245		goto out;
1246
1247	/* copy leaf to new buffer, update identifiers */
1248	xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1249	bp2->b_ops = bp1->b_ops;
1250	memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1251	if (xfs_has_crc(mp)) {
1252		struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1253		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp2));
1254	}
1255	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1256
1257	/*
1258	 * Set up the new root node.
1259	 */
1260	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1261	if (error)
1262		goto out;
1263	node = bp1->b_addr;
1264	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1265
1266	leaf = bp2->b_addr;
1267	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1268	entries = xfs_attr3_leaf_entryp(leaf);
1269
1270	/* both on-disk, don't endian-flip twice */
1271	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1272	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1273	icnodehdr.count = 1;
1274	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1275	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1276	error = 0;
1277out:
1278	return error;
1279}
1280
1281/*========================================================================
1282 * Routines used for growing the Btree.
1283 *========================================================================*/
1284
1285/*
1286 * Create the initial contents of a leaf attribute list
1287 * or a leaf in a node attribute list.
1288 */
1289STATIC int
1290xfs_attr3_leaf_create(
1291	struct xfs_da_args	*args,
1292	xfs_dablk_t		blkno,
1293	struct xfs_buf		**bpp)
1294{
1295	struct xfs_attr_leafblock *leaf;
1296	struct xfs_attr3_icleaf_hdr ichdr;
1297	struct xfs_inode	*dp = args->dp;
1298	struct xfs_mount	*mp = dp->i_mount;
1299	struct xfs_buf		*bp;
1300	int			error;
1301
1302	trace_xfs_attr_leaf_create(args);
1303
1304	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1305					    XFS_ATTR_FORK);
1306	if (error)
1307		return error;
1308	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1309	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1310	leaf = bp->b_addr;
1311	memset(leaf, 0, args->geo->blksize);
1312
1313	memset(&ichdr, 0, sizeof(ichdr));
1314	ichdr.firstused = args->geo->blksize;
1315
1316	if (xfs_has_crc(mp)) {
1317		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1318
1319		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1320
1321		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1322		hdr3->owner = cpu_to_be64(dp->i_ino);
1323		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1324
1325		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1326	} else {
1327		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1328		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1329	}
1330	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1331
1332	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1333	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1334
1335	*bpp = bp;
1336	return 0;
1337}
1338
1339/*
1340 * Split the leaf node, rebalance, then add the new entry.
1341 */
1342int
1343xfs_attr3_leaf_split(
1344	struct xfs_da_state	*state,
1345	struct xfs_da_state_blk	*oldblk,
1346	struct xfs_da_state_blk	*newblk)
1347{
1348	xfs_dablk_t blkno;
1349	int error;
1350
1351	trace_xfs_attr_leaf_split(state->args);
1352
1353	/*
1354	 * Allocate space for a new leaf node.
1355	 */
1356	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1357	error = xfs_da_grow_inode(state->args, &blkno);
1358	if (error)
1359		return error;
1360	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1361	if (error)
1362		return error;
1363	newblk->blkno = blkno;
1364	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1365
1366	/*
1367	 * Rebalance the entries across the two leaves.
1368	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1369	 */
1370	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1371	error = xfs_da3_blk_link(state, oldblk, newblk);
1372	if (error)
1373		return error;
1374
1375	/*
1376	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1377	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1378	 * "new" attrs info.  Will need the "old" info to remove it later.
1379	 *
1380	 * Insert the "new" entry in the correct block.
1381	 */
1382	if (state->inleaf) {
1383		trace_xfs_attr_leaf_add_old(state->args);
1384		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1385	} else {
1386		trace_xfs_attr_leaf_add_new(state->args);
1387		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1388	}
1389
1390	/*
1391	 * Update last hashval in each block since we added the name.
1392	 */
1393	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1394	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1395	return error;
1396}
1397
1398/*
1399 * Add a name to the leaf attribute list structure.
1400 */
1401int
1402xfs_attr3_leaf_add(
1403	struct xfs_buf		*bp,
1404	struct xfs_da_args	*args)
1405{
1406	struct xfs_attr_leafblock *leaf;
1407	struct xfs_attr3_icleaf_hdr ichdr;
1408	int			tablesize;
1409	int			entsize;
1410	int			sum;
1411	int			tmp;
1412	int			i;
1413
1414	trace_xfs_attr_leaf_add(args);
1415
1416	leaf = bp->b_addr;
1417	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1418	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1419	entsize = xfs_attr_leaf_newentsize(args, NULL);
1420
1421	/*
1422	 * Search through freemap for first-fit on new name length.
1423	 * (may need to figure in size of entry struct too)
1424	 */
1425	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1426					+ xfs_attr3_leaf_hdr_size(leaf);
1427	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1428		if (tablesize > ichdr.firstused) {
1429			sum += ichdr.freemap[i].size;
1430			continue;
1431		}
1432		if (!ichdr.freemap[i].size)
1433			continue;	/* no space in this map */
1434		tmp = entsize;
1435		if (ichdr.freemap[i].base < ichdr.firstused)
1436			tmp += sizeof(xfs_attr_leaf_entry_t);
1437		if (ichdr.freemap[i].size >= tmp) {
1438			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1439			goto out_log_hdr;
1440		}
1441		sum += ichdr.freemap[i].size;
1442	}
1443
1444	/*
1445	 * If there are no holes in the address space of the block,
1446	 * and we don't have enough freespace, then compaction will do us
1447	 * no good and we should just give up.
1448	 */
1449	if (!ichdr.holes && sum < entsize)
1450		return -ENOSPC;
1451
1452	/*
1453	 * Compact the entries to coalesce free space.
1454	 * This may change the hdr->count via dropping INCOMPLETE entries.
1455	 */
1456	xfs_attr3_leaf_compact(args, &ichdr, bp);
1457
1458	/*
1459	 * After compaction, the block is guaranteed to have only one
1460	 * free region, in freemap[0].  If it is not big enough, give up.
1461	 */
1462	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1463		tmp = -ENOSPC;
1464		goto out_log_hdr;
1465	}
1466
1467	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1468
1469out_log_hdr:
1470	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1471	xfs_trans_log_buf(args->trans, bp,
1472		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1473				xfs_attr3_leaf_hdr_size(leaf)));
1474	return tmp;
1475}
1476
1477/*
1478 * Add a name to a leaf attribute list structure.
1479 */
1480STATIC int
1481xfs_attr3_leaf_add_work(
1482	struct xfs_buf		*bp,
1483	struct xfs_attr3_icleaf_hdr *ichdr,
1484	struct xfs_da_args	*args,
1485	int			mapindex)
1486{
1487	struct xfs_attr_leafblock *leaf;
1488	struct xfs_attr_leaf_entry *entry;
1489	struct xfs_attr_leaf_name_local *name_loc;
1490	struct xfs_attr_leaf_name_remote *name_rmt;
1491	struct xfs_mount	*mp;
1492	int			tmp;
1493	int			i;
1494
1495	trace_xfs_attr_leaf_add_work(args);
1496
1497	leaf = bp->b_addr;
1498	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1499	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1500
1501	/*
1502	 * Force open some space in the entry array and fill it in.
1503	 */
1504	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1505	if (args->index < ichdr->count) {
1506		tmp  = ichdr->count - args->index;
1507		tmp *= sizeof(xfs_attr_leaf_entry_t);
1508		memmove(entry + 1, entry, tmp);
1509		xfs_trans_log_buf(args->trans, bp,
1510		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1511	}
1512	ichdr->count++;
1513
1514	/*
1515	 * Allocate space for the new string (at the end of the run).
1516	 */
1517	mp = args->trans->t_mountp;
1518	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1519	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1520	ASSERT(ichdr->freemap[mapindex].size >=
1521		xfs_attr_leaf_newentsize(args, NULL));
1522	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1523	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1524
1525	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1526
1527	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1528				     ichdr->freemap[mapindex].size);
1529	entry->hashval = cpu_to_be32(args->hashval);
1530	entry->flags = args->attr_filter;
1531	if (tmp)
1532		entry->flags |= XFS_ATTR_LOCAL;
1533	if (args->op_flags & XFS_DA_OP_REPLACE) {
1534		if (!(args->op_flags & XFS_DA_OP_LOGGED))
1535			entry->flags |= XFS_ATTR_INCOMPLETE;
1536		if ((args->blkno2 == args->blkno) &&
1537		    (args->index2 <= args->index)) {
1538			args->index2++;
1539		}
1540	}
1541	xfs_trans_log_buf(args->trans, bp,
1542			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1543	ASSERT((args->index == 0) ||
1544	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1545	ASSERT((args->index == ichdr->count - 1) ||
1546	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1547
1548	/*
1549	 * For "remote" attribute values, simply note that we need to
1550	 * allocate space for the "remote" value.  We can't actually
1551	 * allocate the extents in this transaction, and we can't decide
1552	 * which blocks they should be as we might allocate more blocks
1553	 * as part of this transaction (a split operation for example).
1554	 */
1555	if (entry->flags & XFS_ATTR_LOCAL) {
1556		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1557		name_loc->namelen = args->namelen;
1558		name_loc->valuelen = cpu_to_be16(args->valuelen);
1559		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1560		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1561				   be16_to_cpu(name_loc->valuelen));
1562	} else {
1563		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1564		name_rmt->namelen = args->namelen;
1565		memcpy((char *)name_rmt->name, args->name, args->namelen);
1566		entry->flags |= XFS_ATTR_INCOMPLETE;
1567		/* just in case */
1568		name_rmt->valuelen = 0;
1569		name_rmt->valueblk = 0;
1570		args->rmtblkno = 1;
1571		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1572		args->rmtvaluelen = args->valuelen;
1573	}
1574	xfs_trans_log_buf(args->trans, bp,
1575	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1576				   xfs_attr_leaf_entsize(leaf, args->index)));
1577
1578	/*
1579	 * Update the control info for this leaf node
1580	 */
1581	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1582		ichdr->firstused = be16_to_cpu(entry->nameidx);
1583
1584	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1585					+ xfs_attr3_leaf_hdr_size(leaf));
1586	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1587					+ xfs_attr3_leaf_hdr_size(leaf);
1588
1589	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1590		if (ichdr->freemap[i].base == tmp) {
1591			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1592			ichdr->freemap[i].size -=
1593				min_t(uint16_t, ichdr->freemap[i].size,
1594						sizeof(xfs_attr_leaf_entry_t));
1595		}
1596	}
1597	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1598	return 0;
1599}
1600
1601/*
1602 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1603 */
1604STATIC void
1605xfs_attr3_leaf_compact(
1606	struct xfs_da_args	*args,
1607	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1608	struct xfs_buf		*bp)
1609{
1610	struct xfs_attr_leafblock *leaf_src;
1611	struct xfs_attr_leafblock *leaf_dst;
1612	struct xfs_attr3_icleaf_hdr ichdr_src;
1613	struct xfs_trans	*trans = args->trans;
1614	char			*tmpbuffer;
1615
1616	trace_xfs_attr_leaf_compact(args);
1617
1618	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1619	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1620	memset(bp->b_addr, 0, args->geo->blksize);
1621	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1622	leaf_dst = bp->b_addr;
1623
1624	/*
1625	 * Copy the on-disk header back into the destination buffer to ensure
1626	 * all the information in the header that is not part of the incore
1627	 * header structure is preserved.
1628	 */
1629	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1630
1631	/* Initialise the incore headers */
1632	ichdr_src = *ichdr_dst;	/* struct copy */
1633	ichdr_dst->firstused = args->geo->blksize;
1634	ichdr_dst->usedbytes = 0;
1635	ichdr_dst->count = 0;
1636	ichdr_dst->holes = 0;
1637	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1638	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1639						ichdr_dst->freemap[0].base;
1640
1641	/* write the header back to initialise the underlying buffer */
1642	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1643
1644	/*
1645	 * Copy all entry's in the same (sorted) order,
1646	 * but allocate name/value pairs packed and in sequence.
1647	 */
1648	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1649				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1650	/*
1651	 * this logs the entire buffer, but the caller must write the header
1652	 * back to the buffer when it is finished modifying it.
1653	 */
1654	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1655
1656	kmem_free(tmpbuffer);
1657}
1658
1659/*
1660 * Compare two leaf blocks "order".
1661 * Return 0 unless leaf2 should go before leaf1.
1662 */
1663static int
1664xfs_attr3_leaf_order(
1665	struct xfs_buf	*leaf1_bp,
1666	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1667	struct xfs_buf	*leaf2_bp,
1668	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1669{
1670	struct xfs_attr_leaf_entry *entries1;
1671	struct xfs_attr_leaf_entry *entries2;
1672
1673	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1674	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1675	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1676	    ((be32_to_cpu(entries2[0].hashval) <
1677	      be32_to_cpu(entries1[0].hashval)) ||
1678	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1679	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1680		return 1;
1681	}
1682	return 0;
1683}
1684
1685int
1686xfs_attr_leaf_order(
1687	struct xfs_buf	*leaf1_bp,
1688	struct xfs_buf	*leaf2_bp)
1689{
1690	struct xfs_attr3_icleaf_hdr ichdr1;
1691	struct xfs_attr3_icleaf_hdr ichdr2;
1692	struct xfs_mount *mp = leaf1_bp->b_mount;
1693
1694	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1695	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1696	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1697}
1698
1699/*
1700 * Redistribute the attribute list entries between two leaf nodes,
1701 * taking into account the size of the new entry.
1702 *
1703 * NOTE: if new block is empty, then it will get the upper half of the
1704 * old block.  At present, all (one) callers pass in an empty second block.
1705 *
1706 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1707 * to match what it is doing in splitting the attribute leaf block.  Those
1708 * values are used in "atomic rename" operations on attributes.  Note that
1709 * the "new" and "old" values can end up in different blocks.
1710 */
1711STATIC void
1712xfs_attr3_leaf_rebalance(
1713	struct xfs_da_state	*state,
1714	struct xfs_da_state_blk	*blk1,
1715	struct xfs_da_state_blk	*blk2)
1716{
1717	struct xfs_da_args	*args;
1718	struct xfs_attr_leafblock *leaf1;
1719	struct xfs_attr_leafblock *leaf2;
1720	struct xfs_attr3_icleaf_hdr ichdr1;
1721	struct xfs_attr3_icleaf_hdr ichdr2;
1722	struct xfs_attr_leaf_entry *entries1;
1723	struct xfs_attr_leaf_entry *entries2;
1724	int			count;
1725	int			totallen;
1726	int			max;
1727	int			space;
1728	int			swap;
1729
1730	/*
1731	 * Set up environment.
1732	 */
1733	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1734	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1735	leaf1 = blk1->bp->b_addr;
1736	leaf2 = blk2->bp->b_addr;
1737	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1738	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1739	ASSERT(ichdr2.count == 0);
1740	args = state->args;
1741
1742	trace_xfs_attr_leaf_rebalance(args);
1743
1744	/*
1745	 * Check ordering of blocks, reverse if it makes things simpler.
1746	 *
1747	 * NOTE: Given that all (current) callers pass in an empty
1748	 * second block, this code should never set "swap".
1749	 */
1750	swap = 0;
1751	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1752		swap(blk1, blk2);
1753
1754		/* swap structures rather than reconverting them */
1755		swap(ichdr1, ichdr2);
1756
1757		leaf1 = blk1->bp->b_addr;
1758		leaf2 = blk2->bp->b_addr;
1759		swap = 1;
1760	}
1761
1762	/*
1763	 * Examine entries until we reduce the absolute difference in
1764	 * byte usage between the two blocks to a minimum.  Then get
1765	 * the direction to copy and the number of elements to move.
1766	 *
1767	 * "inleaf" is true if the new entry should be inserted into blk1.
1768	 * If "swap" is also true, then reverse the sense of "inleaf".
1769	 */
1770	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1771						      blk2, &ichdr2,
1772						      &count, &totallen);
1773	if (swap)
1774		state->inleaf = !state->inleaf;
1775
1776	/*
1777	 * Move any entries required from leaf to leaf:
1778	 */
1779	if (count < ichdr1.count) {
1780		/*
1781		 * Figure the total bytes to be added to the destination leaf.
1782		 */
1783		/* number entries being moved */
1784		count = ichdr1.count - count;
1785		space  = ichdr1.usedbytes - totallen;
1786		space += count * sizeof(xfs_attr_leaf_entry_t);
1787
1788		/*
1789		 * leaf2 is the destination, compact it if it looks tight.
1790		 */
1791		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1792		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1793		if (space > max)
1794			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1795
1796		/*
1797		 * Move high entries from leaf1 to low end of leaf2.
1798		 */
1799		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1800				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1801
1802	} else if (count > ichdr1.count) {
1803		/*
1804		 * I assert that since all callers pass in an empty
1805		 * second buffer, this code should never execute.
1806		 */
1807		ASSERT(0);
1808
1809		/*
1810		 * Figure the total bytes to be added to the destination leaf.
1811		 */
1812		/* number entries being moved */
1813		count -= ichdr1.count;
1814		space  = totallen - ichdr1.usedbytes;
1815		space += count * sizeof(xfs_attr_leaf_entry_t);
1816
1817		/*
1818		 * leaf1 is the destination, compact it if it looks tight.
1819		 */
1820		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1821		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1822		if (space > max)
1823			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1824
1825		/*
1826		 * Move low entries from leaf2 to high end of leaf1.
1827		 */
1828		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1829					ichdr1.count, count);
1830	}
1831
1832	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1833	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1834	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1835	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1836
1837	/*
1838	 * Copy out last hashval in each block for B-tree code.
1839	 */
1840	entries1 = xfs_attr3_leaf_entryp(leaf1);
1841	entries2 = xfs_attr3_leaf_entryp(leaf2);
1842	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1843	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1844
1845	/*
1846	 * Adjust the expected index for insertion.
1847	 * NOTE: this code depends on the (current) situation that the
1848	 * second block was originally empty.
1849	 *
1850	 * If the insertion point moved to the 2nd block, we must adjust
1851	 * the index.  We must also track the entry just following the
1852	 * new entry for use in an "atomic rename" operation, that entry
1853	 * is always the "old" entry and the "new" entry is what we are
1854	 * inserting.  The index/blkno fields refer to the "old" entry,
1855	 * while the index2/blkno2 fields refer to the "new" entry.
1856	 */
1857	if (blk1->index > ichdr1.count) {
1858		ASSERT(state->inleaf == 0);
1859		blk2->index = blk1->index - ichdr1.count;
1860		args->index = args->index2 = blk2->index;
1861		args->blkno = args->blkno2 = blk2->blkno;
1862	} else if (blk1->index == ichdr1.count) {
1863		if (state->inleaf) {
1864			args->index = blk1->index;
1865			args->blkno = blk1->blkno;
1866			args->index2 = 0;
1867			args->blkno2 = blk2->blkno;
1868		} else {
1869			/*
1870			 * On a double leaf split, the original attr location
1871			 * is already stored in blkno2/index2, so don't
1872			 * overwrite it overwise we corrupt the tree.
1873			 */
1874			blk2->index = blk1->index - ichdr1.count;
1875			args->index = blk2->index;
1876			args->blkno = blk2->blkno;
1877			if (!state->extravalid) {
1878				/*
1879				 * set the new attr location to match the old
1880				 * one and let the higher level split code
1881				 * decide where in the leaf to place it.
1882				 */
1883				args->index2 = blk2->index;
1884				args->blkno2 = blk2->blkno;
1885			}
1886		}
1887	} else {
1888		ASSERT(state->inleaf == 1);
1889		args->index = args->index2 = blk1->index;
1890		args->blkno = args->blkno2 = blk1->blkno;
1891	}
1892}
1893
1894/*
1895 * Examine entries until we reduce the absolute difference in
1896 * byte usage between the two blocks to a minimum.
1897 * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1898 * GROT: there will always be enough room in either block for a new entry.
1899 * GROT: Do a double-split for this case?
1900 */
1901STATIC int
1902xfs_attr3_leaf_figure_balance(
1903	struct xfs_da_state		*state,
1904	struct xfs_da_state_blk		*blk1,
1905	struct xfs_attr3_icleaf_hdr	*ichdr1,
1906	struct xfs_da_state_blk		*blk2,
1907	struct xfs_attr3_icleaf_hdr	*ichdr2,
1908	int				*countarg,
1909	int				*usedbytesarg)
1910{
1911	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1912	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1913	struct xfs_attr_leaf_entry	*entry;
1914	int				count;
1915	int				max;
1916	int				index;
1917	int				totallen = 0;
1918	int				half;
1919	int				lastdelta;
1920	int				foundit = 0;
1921	int				tmp;
1922
1923	/*
1924	 * Examine entries until we reduce the absolute difference in
1925	 * byte usage between the two blocks to a minimum.
1926	 */
1927	max = ichdr1->count + ichdr2->count;
1928	half = (max + 1) * sizeof(*entry);
1929	half += ichdr1->usedbytes + ichdr2->usedbytes +
1930			xfs_attr_leaf_newentsize(state->args, NULL);
1931	half /= 2;
1932	lastdelta = state->args->geo->blksize;
1933	entry = xfs_attr3_leaf_entryp(leaf1);
1934	for (count = index = 0; count < max; entry++, index++, count++) {
1935
1936#define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1937		/*
1938		 * The new entry is in the first block, account for it.
1939		 */
1940		if (count == blk1->index) {
1941			tmp = totallen + sizeof(*entry) +
1942				xfs_attr_leaf_newentsize(state->args, NULL);
1943			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1944				break;
1945			lastdelta = XFS_ATTR_ABS(half - tmp);
1946			totallen = tmp;
1947			foundit = 1;
1948		}
1949
1950		/*
1951		 * Wrap around into the second block if necessary.
1952		 */
1953		if (count == ichdr1->count) {
1954			leaf1 = leaf2;
1955			entry = xfs_attr3_leaf_entryp(leaf1);
1956			index = 0;
1957		}
1958
1959		/*
1960		 * Figure out if next leaf entry would be too much.
1961		 */
1962		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1963									index);
1964		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1965			break;
1966		lastdelta = XFS_ATTR_ABS(half - tmp);
1967		totallen = tmp;
1968#undef XFS_ATTR_ABS
1969	}
1970
1971	/*
1972	 * Calculate the number of usedbytes that will end up in lower block.
1973	 * If new entry not in lower block, fix up the count.
1974	 */
1975	totallen -= count * sizeof(*entry);
1976	if (foundit) {
1977		totallen -= sizeof(*entry) +
1978				xfs_attr_leaf_newentsize(state->args, NULL);
1979	}
1980
1981	*countarg = count;
1982	*usedbytesarg = totallen;
1983	return foundit;
1984}
1985
1986/*========================================================================
1987 * Routines used for shrinking the Btree.
1988 *========================================================================*/
1989
1990/*
1991 * Check a leaf block and its neighbors to see if the block should be
1992 * collapsed into one or the other neighbor.  Always keep the block
1993 * with the smaller block number.
1994 * If the current block is over 50% full, don't try to join it, return 0.
1995 * If the block is empty, fill in the state structure and return 2.
1996 * If it can be collapsed, fill in the state structure and return 1.
1997 * If nothing can be done, return 0.
1998 *
1999 * GROT: allow for INCOMPLETE entries in calculation.
2000 */
2001int
2002xfs_attr3_leaf_toosmall(
2003	struct xfs_da_state	*state,
2004	int			*action)
2005{
2006	struct xfs_attr_leafblock *leaf;
2007	struct xfs_da_state_blk	*blk;
2008	struct xfs_attr3_icleaf_hdr ichdr;
2009	struct xfs_buf		*bp;
2010	xfs_dablk_t		blkno;
2011	int			bytes;
2012	int			forward;
2013	int			error;
2014	int			retval;
2015	int			i;
2016
2017	trace_xfs_attr_leaf_toosmall(state->args);
2018
2019	/*
2020	 * Check for the degenerate case of the block being over 50% full.
2021	 * If so, it's not worth even looking to see if we might be able
2022	 * to coalesce with a sibling.
2023	 */
2024	blk = &state->path.blk[ state->path.active-1 ];
2025	leaf = blk->bp->b_addr;
2026	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
2027	bytes = xfs_attr3_leaf_hdr_size(leaf) +
2028		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2029		ichdr.usedbytes;
2030	if (bytes > (state->args->geo->blksize >> 1)) {
2031		*action = 0;	/* blk over 50%, don't try to join */
2032		return 0;
2033	}
2034
2035	/*
2036	 * Check for the degenerate case of the block being empty.
2037	 * If the block is empty, we'll simply delete it, no need to
2038	 * coalesce it with a sibling block.  We choose (arbitrarily)
2039	 * to merge with the forward block unless it is NULL.
2040	 */
2041	if (ichdr.count == 0) {
2042		/*
2043		 * Make altpath point to the block we want to keep and
2044		 * path point to the block we want to drop (this one).
2045		 */
2046		forward = (ichdr.forw != 0);
2047		memcpy(&state->altpath, &state->path, sizeof(state->path));
2048		error = xfs_da3_path_shift(state, &state->altpath, forward,
2049						 0, &retval);
2050		if (error)
2051			return error;
2052		if (retval) {
2053			*action = 0;
2054		} else {
2055			*action = 2;
2056		}
2057		return 0;
2058	}
2059
2060	/*
2061	 * Examine each sibling block to see if we can coalesce with
2062	 * at least 25% free space to spare.  We need to figure out
2063	 * whether to merge with the forward or the backward block.
2064	 * We prefer coalescing with the lower numbered sibling so as
2065	 * to shrink an attribute list over time.
2066	 */
2067	/* start with smaller blk num */
2068	forward = ichdr.forw < ichdr.back;
2069	for (i = 0; i < 2; forward = !forward, i++) {
2070		struct xfs_attr3_icleaf_hdr ichdr2;
2071		if (forward)
2072			blkno = ichdr.forw;
2073		else
2074			blkno = ichdr.back;
2075		if (blkno == 0)
2076			continue;
2077		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2078					blkno, &bp);
2079		if (error)
2080			return error;
2081
2082		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2083
2084		bytes = state->args->geo->blksize -
2085			(state->args->geo->blksize >> 2) -
2086			ichdr.usedbytes - ichdr2.usedbytes -
2087			((ichdr.count + ichdr2.count) *
2088					sizeof(xfs_attr_leaf_entry_t)) -
2089			xfs_attr3_leaf_hdr_size(leaf);
2090
2091		xfs_trans_brelse(state->args->trans, bp);
2092		if (bytes >= 0)
2093			break;	/* fits with at least 25% to spare */
2094	}
2095	if (i >= 2) {
2096		*action = 0;
2097		return 0;
2098	}
2099
2100	/*
2101	 * Make altpath point to the block we want to keep (the lower
2102	 * numbered block) and path point to the block we want to drop.
2103	 */
2104	memcpy(&state->altpath, &state->path, sizeof(state->path));
2105	if (blkno < blk->blkno) {
2106		error = xfs_da3_path_shift(state, &state->altpath, forward,
2107						 0, &retval);
2108	} else {
2109		error = xfs_da3_path_shift(state, &state->path, forward,
2110						 0, &retval);
2111	}
2112	if (error)
2113		return error;
2114	if (retval) {
2115		*action = 0;
2116	} else {
2117		*action = 1;
2118	}
2119	return 0;
2120}
2121
2122/*
2123 * Remove a name from the leaf attribute list structure.
2124 *
2125 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2126 * If two leaves are 37% full, when combined they will leave 25% free.
2127 */
2128int
2129xfs_attr3_leaf_remove(
2130	struct xfs_buf		*bp,
2131	struct xfs_da_args	*args)
2132{
2133	struct xfs_attr_leafblock *leaf;
2134	struct xfs_attr3_icleaf_hdr ichdr;
2135	struct xfs_attr_leaf_entry *entry;
2136	int			before;
2137	int			after;
2138	int			smallest;
2139	int			entsize;
2140	int			tablesize;
2141	int			tmp;
2142	int			i;
2143
2144	trace_xfs_attr_leaf_remove(args);
2145
2146	leaf = bp->b_addr;
2147	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2148
2149	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2150	ASSERT(args->index >= 0 && args->index < ichdr.count);
2151	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2152					xfs_attr3_leaf_hdr_size(leaf));
2153
2154	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2155
2156	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2157	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2158
2159	/*
2160	 * Scan through free region table:
2161	 *    check for adjacency of free'd entry with an existing one,
2162	 *    find smallest free region in case we need to replace it,
2163	 *    adjust any map that borders the entry table,
2164	 */
2165	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2166					+ xfs_attr3_leaf_hdr_size(leaf);
2167	tmp = ichdr.freemap[0].size;
2168	before = after = -1;
2169	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2170	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2171	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2172		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2173		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2174		if (ichdr.freemap[i].base == tablesize) {
2175			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2176			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2177		}
2178
2179		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2180				be16_to_cpu(entry->nameidx)) {
2181			before = i;
2182		} else if (ichdr.freemap[i].base ==
2183				(be16_to_cpu(entry->nameidx) + entsize)) {
2184			after = i;
2185		} else if (ichdr.freemap[i].size < tmp) {
2186			tmp = ichdr.freemap[i].size;
2187			smallest = i;
2188		}
2189	}
2190
2191	/*
2192	 * Coalesce adjacent freemap regions,
2193	 * or replace the smallest region.
2194	 */
2195	if ((before >= 0) || (after >= 0)) {
2196		if ((before >= 0) && (after >= 0)) {
2197			ichdr.freemap[before].size += entsize;
2198			ichdr.freemap[before].size += ichdr.freemap[after].size;
2199			ichdr.freemap[after].base = 0;
2200			ichdr.freemap[after].size = 0;
2201		} else if (before >= 0) {
2202			ichdr.freemap[before].size += entsize;
2203		} else {
2204			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2205			ichdr.freemap[after].size += entsize;
2206		}
2207	} else {
2208		/*
2209		 * Replace smallest region (if it is smaller than free'd entry)
2210		 */
2211		if (ichdr.freemap[smallest].size < entsize) {
2212			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2213			ichdr.freemap[smallest].size = entsize;
2214		}
2215	}
2216
2217	/*
2218	 * Did we remove the first entry?
2219	 */
2220	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2221		smallest = 1;
2222	else
2223		smallest = 0;
2224
2225	/*
2226	 * Compress the remaining entries and zero out the removed stuff.
2227	 */
2228	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2229	ichdr.usedbytes -= entsize;
2230	xfs_trans_log_buf(args->trans, bp,
2231	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2232				   entsize));
2233
2234	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2235	memmove(entry, entry + 1, tmp);
2236	ichdr.count--;
2237	xfs_trans_log_buf(args->trans, bp,
2238	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2239
2240	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2241	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2242
2243	/*
2244	 * If we removed the first entry, re-find the first used byte
2245	 * in the name area.  Note that if the entry was the "firstused",
2246	 * then we don't have a "hole" in our block resulting from
2247	 * removing the name.
2248	 */
2249	if (smallest) {
2250		tmp = args->geo->blksize;
2251		entry = xfs_attr3_leaf_entryp(leaf);
2252		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2253			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2254			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2255
2256			if (be16_to_cpu(entry->nameidx) < tmp)
2257				tmp = be16_to_cpu(entry->nameidx);
2258		}
2259		ichdr.firstused = tmp;
2260		ASSERT(ichdr.firstused != 0);
2261	} else {
2262		ichdr.holes = 1;	/* mark as needing compaction */
2263	}
2264	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2265	xfs_trans_log_buf(args->trans, bp,
2266			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2267					  xfs_attr3_leaf_hdr_size(leaf)));
2268
2269	/*
2270	 * Check if leaf is less than 50% full, caller may want to
2271	 * "join" the leaf with a sibling if so.
2272	 */
2273	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2274	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2275
2276	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2277}
2278
2279/*
2280 * Move all the attribute list entries from drop_leaf into save_leaf.
2281 */
2282void
2283xfs_attr3_leaf_unbalance(
2284	struct xfs_da_state	*state,
2285	struct xfs_da_state_blk	*drop_blk,
2286	struct xfs_da_state_blk	*save_blk)
2287{
2288	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2289	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2290	struct xfs_attr3_icleaf_hdr drophdr;
2291	struct xfs_attr3_icleaf_hdr savehdr;
2292	struct xfs_attr_leaf_entry *entry;
2293
2294	trace_xfs_attr_leaf_unbalance(state->args);
2295
2296	drop_leaf = drop_blk->bp->b_addr;
2297	save_leaf = save_blk->bp->b_addr;
2298	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2299	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2300	entry = xfs_attr3_leaf_entryp(drop_leaf);
2301
2302	/*
2303	 * Save last hashval from dying block for later Btree fixup.
2304	 */
2305	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2306
2307	/*
2308	 * Check if we need a temp buffer, or can we do it in place.
2309	 * Note that we don't check "leaf" for holes because we will
2310	 * always be dropping it, toosmall() decided that for us already.
2311	 */
2312	if (savehdr.holes == 0) {
2313		/*
2314		 * dest leaf has no holes, so we add there.  May need
2315		 * to make some room in the entry array.
2316		 */
2317		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2318					 drop_blk->bp, &drophdr)) {
2319			xfs_attr3_leaf_moveents(state->args,
2320						drop_leaf, &drophdr, 0,
2321						save_leaf, &savehdr, 0,
2322						drophdr.count);
2323		} else {
2324			xfs_attr3_leaf_moveents(state->args,
2325						drop_leaf, &drophdr, 0,
2326						save_leaf, &savehdr,
2327						savehdr.count, drophdr.count);
2328		}
2329	} else {
2330		/*
2331		 * Destination has holes, so we make a temporary copy
2332		 * of the leaf and add them both to that.
2333		 */
2334		struct xfs_attr_leafblock *tmp_leaf;
2335		struct xfs_attr3_icleaf_hdr tmphdr;
2336
2337		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2338
2339		/*
2340		 * Copy the header into the temp leaf so that all the stuff
2341		 * not in the incore header is present and gets copied back in
2342		 * once we've moved all the entries.
2343		 */
2344		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2345
2346		memset(&tmphdr, 0, sizeof(tmphdr));
2347		tmphdr.magic = savehdr.magic;
2348		tmphdr.forw = savehdr.forw;
2349		tmphdr.back = savehdr.back;
2350		tmphdr.firstused = state->args->geo->blksize;
2351
2352		/* write the header to the temp buffer to initialise it */
2353		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2354
2355		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2356					 drop_blk->bp, &drophdr)) {
2357			xfs_attr3_leaf_moveents(state->args,
2358						drop_leaf, &drophdr, 0,
2359						tmp_leaf, &tmphdr, 0,
2360						drophdr.count);
2361			xfs_attr3_leaf_moveents(state->args,
2362						save_leaf, &savehdr, 0,
2363						tmp_leaf, &tmphdr, tmphdr.count,
2364						savehdr.count);
2365		} else {
2366			xfs_attr3_leaf_moveents(state->args,
2367						save_leaf, &savehdr, 0,
2368						tmp_leaf, &tmphdr, 0,
2369						savehdr.count);
2370			xfs_attr3_leaf_moveents(state->args,
2371						drop_leaf, &drophdr, 0,
2372						tmp_leaf, &tmphdr, tmphdr.count,
2373						drophdr.count);
2374		}
2375		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2376		savehdr = tmphdr; /* struct copy */
2377		kmem_free(tmp_leaf);
2378	}
2379
2380	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2381	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2382					   state->args->geo->blksize - 1);
2383
2384	/*
2385	 * Copy out last hashval in each block for B-tree code.
2386	 */
2387	entry = xfs_attr3_leaf_entryp(save_leaf);
2388	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2389}
2390
2391/*========================================================================
2392 * Routines used for finding things in the Btree.
2393 *========================================================================*/
2394
2395/*
2396 * Look up a name in a leaf attribute list structure.
2397 * This is the internal routine, it uses the caller's buffer.
2398 *
2399 * Note that duplicate keys are allowed, but only check within the
2400 * current leaf node.  The Btree code must check in adjacent leaf nodes.
2401 *
2402 * Return in args->index the index into the entry[] array of either
2403 * the found entry, or where the entry should have been (insert before
2404 * that entry).
2405 *
2406 * Don't change the args->value unless we find the attribute.
2407 */
2408int
2409xfs_attr3_leaf_lookup_int(
2410	struct xfs_buf		*bp,
2411	struct xfs_da_args	*args)
2412{
2413	struct xfs_attr_leafblock *leaf;
2414	struct xfs_attr3_icleaf_hdr ichdr;
2415	struct xfs_attr_leaf_entry *entry;
2416	struct xfs_attr_leaf_entry *entries;
2417	struct xfs_attr_leaf_name_local *name_loc;
2418	struct xfs_attr_leaf_name_remote *name_rmt;
2419	xfs_dahash_t		hashval;
2420	int			probe;
2421	int			span;
2422
2423	trace_xfs_attr_leaf_lookup(args);
2424
2425	leaf = bp->b_addr;
2426	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2427	entries = xfs_attr3_leaf_entryp(leaf);
2428	if (ichdr.count >= args->geo->blksize / 8) {
2429		xfs_buf_mark_corrupt(bp);
2430		return -EFSCORRUPTED;
2431	}
2432
2433	/*
2434	 * Binary search.  (note: small blocks will skip this loop)
2435	 */
2436	hashval = args->hashval;
2437	probe = span = ichdr.count / 2;
2438	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2439		span /= 2;
2440		if (be32_to_cpu(entry->hashval) < hashval)
2441			probe += span;
2442		else if (be32_to_cpu(entry->hashval) > hashval)
2443			probe -= span;
2444		else
2445			break;
2446	}
2447	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2448		xfs_buf_mark_corrupt(bp);
2449		return -EFSCORRUPTED;
2450	}
2451	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2452		xfs_buf_mark_corrupt(bp);
2453		return -EFSCORRUPTED;
2454	}
2455
2456	/*
2457	 * Since we may have duplicate hashval's, find the first matching
2458	 * hashval in the leaf.
2459	 */
2460	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2461		entry--;
2462		probe--;
2463	}
2464	while (probe < ichdr.count &&
2465	       be32_to_cpu(entry->hashval) < hashval) {
2466		entry++;
2467		probe++;
2468	}
2469	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2470		args->index = probe;
2471		return -ENOATTR;
2472	}
2473
2474	/*
2475	 * Duplicate keys may be present, so search all of them for a match.
2476	 */
2477	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2478			entry++, probe++) {
2479/*
2480 * GROT: Add code to remove incomplete entries.
2481 */
2482		if (entry->flags & XFS_ATTR_LOCAL) {
2483			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2484			if (!xfs_attr_match(args, name_loc->namelen,
2485					name_loc->nameval, entry->flags))
2486				continue;
2487			args->index = probe;
2488			return -EEXIST;
2489		} else {
2490			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2491			if (!xfs_attr_match(args, name_rmt->namelen,
2492					name_rmt->name, entry->flags))
2493				continue;
2494			args->index = probe;
2495			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2496			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2497			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2498							args->dp->i_mount,
2499							args->rmtvaluelen);
2500			return -EEXIST;
2501		}
2502	}
2503	args->index = probe;
2504	return -ENOATTR;
2505}
2506
2507/*
2508 * Get the value associated with an attribute name from a leaf attribute
2509 * list structure.
2510 *
2511 * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2512 * lookup, we only return an error if the attribute does not exist or we can't
2513 * retrieve the value.
2514 */
2515int
2516xfs_attr3_leaf_getvalue(
2517	struct xfs_buf		*bp,
2518	struct xfs_da_args	*args)
2519{
2520	struct xfs_attr_leafblock *leaf;
2521	struct xfs_attr3_icleaf_hdr ichdr;
2522	struct xfs_attr_leaf_entry *entry;
2523	struct xfs_attr_leaf_name_local *name_loc;
2524	struct xfs_attr_leaf_name_remote *name_rmt;
2525
2526	leaf = bp->b_addr;
2527	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2528	ASSERT(ichdr.count < args->geo->blksize / 8);
2529	ASSERT(args->index < ichdr.count);
2530
2531	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2532	if (entry->flags & XFS_ATTR_LOCAL) {
2533		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2534		ASSERT(name_loc->namelen == args->namelen);
2535		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2536		return xfs_attr_copy_value(args,
2537					&name_loc->nameval[args->namelen],
2538					be16_to_cpu(name_loc->valuelen));
2539	}
2540
2541	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2542	ASSERT(name_rmt->namelen == args->namelen);
2543	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2544	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2545	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2546	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2547					       args->rmtvaluelen);
2548	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2549}
2550
2551/*========================================================================
2552 * Utility routines.
2553 *========================================================================*/
2554
2555/*
2556 * Move the indicated entries from one leaf to another.
2557 * NOTE: this routine modifies both source and destination leaves.
2558 */
2559/*ARGSUSED*/
2560STATIC void
2561xfs_attr3_leaf_moveents(
2562	struct xfs_da_args		*args,
2563	struct xfs_attr_leafblock	*leaf_s,
2564	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2565	int				start_s,
2566	struct xfs_attr_leafblock	*leaf_d,
2567	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2568	int				start_d,
2569	int				count)
2570{
2571	struct xfs_attr_leaf_entry	*entry_s;
2572	struct xfs_attr_leaf_entry	*entry_d;
2573	int				desti;
2574	int				tmp;
2575	int				i;
2576
2577	/*
2578	 * Check for nothing to do.
2579	 */
2580	if (count == 0)
2581		return;
2582
2583	/*
2584	 * Set up environment.
2585	 */
2586	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2587	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2588	ASSERT(ichdr_s->magic == ichdr_d->magic);
2589	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2590	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2591					+ xfs_attr3_leaf_hdr_size(leaf_s));
2592	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2593	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2594					+ xfs_attr3_leaf_hdr_size(leaf_d));
2595
2596	ASSERT(start_s < ichdr_s->count);
2597	ASSERT(start_d <= ichdr_d->count);
2598	ASSERT(count <= ichdr_s->count);
2599
2600
2601	/*
2602	 * Move the entries in the destination leaf up to make a hole?
2603	 */
2604	if (start_d < ichdr_d->count) {
2605		tmp  = ichdr_d->count - start_d;
2606		tmp *= sizeof(xfs_attr_leaf_entry_t);
2607		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2608		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2609		memmove(entry_d, entry_s, tmp);
2610	}
2611
2612	/*
2613	 * Copy all entry's in the same (sorted) order,
2614	 * but allocate attribute info packed and in sequence.
2615	 */
2616	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2617	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2618	desti = start_d;
2619	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2620		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2621		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2622#ifdef GROT
2623		/*
2624		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2625		 * may also need to change the insertion index.  Code turned
2626		 * off for 6.2, should be revisited later.
2627		 */
2628		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2629			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2630			ichdr_s->usedbytes -= tmp;
2631			ichdr_s->count -= 1;
2632			entry_d--;	/* to compensate for ++ in loop hdr */
2633			desti--;
2634			if ((start_s + i) < offset)
2635				result++;	/* insertion index adjustment */
2636		} else {
2637#endif /* GROT */
2638			ichdr_d->firstused -= tmp;
2639			/* both on-disk, don't endian flip twice */
2640			entry_d->hashval = entry_s->hashval;
2641			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2642			entry_d->flags = entry_s->flags;
2643			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2644							<= args->geo->blksize);
2645			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2646				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2647			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2648							<= args->geo->blksize);
2649			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2650			ichdr_s->usedbytes -= tmp;
2651			ichdr_d->usedbytes += tmp;
2652			ichdr_s->count -= 1;
2653			ichdr_d->count += 1;
2654			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2655					+ xfs_attr3_leaf_hdr_size(leaf_d);
2656			ASSERT(ichdr_d->firstused >= tmp);
2657#ifdef GROT
2658		}
2659#endif /* GROT */
2660	}
2661
2662	/*
2663	 * Zero out the entries we just copied.
2664	 */
2665	if (start_s == ichdr_s->count) {
2666		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2667		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2668		ASSERT(((char *)entry_s + tmp) <=
2669		       ((char *)leaf_s + args->geo->blksize));
2670		memset(entry_s, 0, tmp);
2671	} else {
2672		/*
2673		 * Move the remaining entries down to fill the hole,
2674		 * then zero the entries at the top.
2675		 */
2676		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2677		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2678		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2679		memmove(entry_d, entry_s, tmp);
2680
2681		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2682		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2683		ASSERT(((char *)entry_s + tmp) <=
2684		       ((char *)leaf_s + args->geo->blksize));
2685		memset(entry_s, 0, tmp);
2686	}
2687
2688	/*
2689	 * Fill in the freemap information
2690	 */
2691	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2692	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2693	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2694	ichdr_d->freemap[1].base = 0;
2695	ichdr_d->freemap[2].base = 0;
2696	ichdr_d->freemap[1].size = 0;
2697	ichdr_d->freemap[2].size = 0;
2698	ichdr_s->holes = 1;	/* leaf may not be compact */
2699}
2700
2701/*
2702 * Pick up the last hashvalue from a leaf block.
2703 */
2704xfs_dahash_t
2705xfs_attr_leaf_lasthash(
2706	struct xfs_buf	*bp,
2707	int		*count)
2708{
2709	struct xfs_attr3_icleaf_hdr ichdr;
2710	struct xfs_attr_leaf_entry *entries;
2711	struct xfs_mount *mp = bp->b_mount;
2712
2713	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2714	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2715	if (count)
2716		*count = ichdr.count;
2717	if (!ichdr.count)
2718		return 0;
2719	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2720}
2721
2722/*
2723 * Calculate the number of bytes used to store the indicated attribute
2724 * (whether local or remote only calculate bytes in this block).
2725 */
2726STATIC int
2727xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2728{
2729	struct xfs_attr_leaf_entry *entries;
2730	xfs_attr_leaf_name_local_t *name_loc;
2731	xfs_attr_leaf_name_remote_t *name_rmt;
2732	int size;
2733
2734	entries = xfs_attr3_leaf_entryp(leaf);
2735	if (entries[index].flags & XFS_ATTR_LOCAL) {
2736		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2737		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2738						   be16_to_cpu(name_loc->valuelen));
2739	} else {
2740		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2741		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2742	}
2743	return size;
2744}
2745
2746/*
2747 * Calculate the number of bytes that would be required to store the new
2748 * attribute (whether local or remote only calculate bytes in this block).
2749 * This routine decides as a side effect whether the attribute will be
2750 * a "local" or a "remote" attribute.
2751 */
2752int
2753xfs_attr_leaf_newentsize(
2754	struct xfs_da_args	*args,
2755	int			*local)
2756{
2757	int			size;
2758
2759	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2760	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2761		if (local)
2762			*local = 1;
2763		return size;
2764	}
2765	if (local)
2766		*local = 0;
2767	return xfs_attr_leaf_entsize_remote(args->namelen);
2768}
2769
2770
2771/*========================================================================
2772 * Manage the INCOMPLETE flag in a leaf entry
2773 *========================================================================*/
2774
2775/*
2776 * Clear the INCOMPLETE flag on an entry in a leaf block.
2777 */
2778int
2779xfs_attr3_leaf_clearflag(
2780	struct xfs_da_args	*args)
2781{
2782	struct xfs_attr_leafblock *leaf;
2783	struct xfs_attr_leaf_entry *entry;
2784	struct xfs_attr_leaf_name_remote *name_rmt;
2785	struct xfs_buf		*bp;
2786	int			error;
2787#ifdef DEBUG
2788	struct xfs_attr3_icleaf_hdr ichdr;
2789	xfs_attr_leaf_name_local_t *name_loc;
2790	int namelen;
2791	char *name;
2792#endif /* DEBUG */
2793
2794	trace_xfs_attr_leaf_clearflag(args);
2795	/*
2796	 * Set up the operation.
2797	 */
2798	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2799	if (error)
2800		return error;
2801
2802	leaf = bp->b_addr;
2803	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2804	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2805
2806#ifdef DEBUG
2807	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2808	ASSERT(args->index < ichdr.count);
2809	ASSERT(args->index >= 0);
2810
2811	if (entry->flags & XFS_ATTR_LOCAL) {
2812		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2813		namelen = name_loc->namelen;
2814		name = (char *)name_loc->nameval;
2815	} else {
2816		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2817		namelen = name_rmt->namelen;
2818		name = (char *)name_rmt->name;
2819	}
2820	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2821	ASSERT(namelen == args->namelen);
2822	ASSERT(memcmp(name, args->name, namelen) == 0);
2823#endif /* DEBUG */
2824
2825	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2826	xfs_trans_log_buf(args->trans, bp,
2827			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2828
2829	if (args->rmtblkno) {
2830		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2831		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2832		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2833		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2834		xfs_trans_log_buf(args->trans, bp,
2835			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2836	}
2837
2838	return 0;
2839}
2840
2841/*
2842 * Set the INCOMPLETE flag on an entry in a leaf block.
2843 */
2844int
2845xfs_attr3_leaf_setflag(
2846	struct xfs_da_args	*args)
2847{
2848	struct xfs_attr_leafblock *leaf;
2849	struct xfs_attr_leaf_entry *entry;
2850	struct xfs_attr_leaf_name_remote *name_rmt;
2851	struct xfs_buf		*bp;
2852	int error;
2853#ifdef DEBUG
2854	struct xfs_attr3_icleaf_hdr ichdr;
2855#endif
2856
2857	trace_xfs_attr_leaf_setflag(args);
2858
2859	/*
2860	 * Set up the operation.
2861	 */
2862	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2863	if (error)
2864		return error;
2865
2866	leaf = bp->b_addr;
2867#ifdef DEBUG
2868	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2869	ASSERT(args->index < ichdr.count);
2870	ASSERT(args->index >= 0);
2871#endif
2872	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2873
2874	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2875	entry->flags |= XFS_ATTR_INCOMPLETE;
2876	xfs_trans_log_buf(args->trans, bp,
2877			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2878	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2879		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2880		name_rmt->valueblk = 0;
2881		name_rmt->valuelen = 0;
2882		xfs_trans_log_buf(args->trans, bp,
2883			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2884	}
2885
2886	return 0;
2887}
2888
2889/*
2890 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2891 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2892 * entry given by args->blkno2/index2.
2893 *
2894 * Note that they could be in different blocks, or in the same block.
2895 */
2896int
2897xfs_attr3_leaf_flipflags(
2898	struct xfs_da_args	*args)
2899{
2900	struct xfs_attr_leafblock *leaf1;
2901	struct xfs_attr_leafblock *leaf2;
2902	struct xfs_attr_leaf_entry *entry1;
2903	struct xfs_attr_leaf_entry *entry2;
2904	struct xfs_attr_leaf_name_remote *name_rmt;
2905	struct xfs_buf		*bp1;
2906	struct xfs_buf		*bp2;
2907	int error;
2908#ifdef DEBUG
2909	struct xfs_attr3_icleaf_hdr ichdr1;
2910	struct xfs_attr3_icleaf_hdr ichdr2;
2911	xfs_attr_leaf_name_local_t *name_loc;
2912	int namelen1, namelen2;
2913	char *name1, *name2;
2914#endif /* DEBUG */
2915
2916	trace_xfs_attr_leaf_flipflags(args);
2917
2918	/*
2919	 * Read the block containing the "old" attr
2920	 */
2921	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2922	if (error)
2923		return error;
2924
2925	/*
2926	 * Read the block containing the "new" attr, if it is different
2927	 */
2928	if (args->blkno2 != args->blkno) {
2929		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2930					   &bp2);
2931		if (error)
2932			return error;
2933	} else {
2934		bp2 = bp1;
2935	}
2936
2937	leaf1 = bp1->b_addr;
2938	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2939
2940	leaf2 = bp2->b_addr;
2941	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2942
2943#ifdef DEBUG
2944	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2945	ASSERT(args->index < ichdr1.count);
2946	ASSERT(args->index >= 0);
2947
2948	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2949	ASSERT(args->index2 < ichdr2.count);
2950	ASSERT(args->index2 >= 0);
2951
2952	if (entry1->flags & XFS_ATTR_LOCAL) {
2953		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2954		namelen1 = name_loc->namelen;
2955		name1 = (char *)name_loc->nameval;
2956	} else {
2957		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2958		namelen1 = name_rmt->namelen;
2959		name1 = (char *)name_rmt->name;
2960	}
2961	if (entry2->flags & XFS_ATTR_LOCAL) {
2962		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2963		namelen2 = name_loc->namelen;
2964		name2 = (char *)name_loc->nameval;
2965	} else {
2966		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2967		namelen2 = name_rmt->namelen;
2968		name2 = (char *)name_rmt->name;
2969	}
2970	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2971	ASSERT(namelen1 == namelen2);
2972	ASSERT(memcmp(name1, name2, namelen1) == 0);
2973#endif /* DEBUG */
2974
2975	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2976	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2977
2978	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2979	xfs_trans_log_buf(args->trans, bp1,
2980			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2981	if (args->rmtblkno) {
2982		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2983		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2984		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2985		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2986		xfs_trans_log_buf(args->trans, bp1,
2987			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2988	}
2989
2990	entry2->flags |= XFS_ATTR_INCOMPLETE;
2991	xfs_trans_log_buf(args->trans, bp2,
2992			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2993	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2994		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2995		name_rmt->valueblk = 0;
2996		name_rmt->valuelen = 0;
2997		xfs_trans_log_buf(args->trans, bp2,
2998			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2999	}
3000
3001	return 0;
3002}