xref: /linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision d195c39052d1da278a00a6744ce59c383b67b191)
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_remote.h"
23 #include "xfs_attr.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 
31 
32 /*
33  * xfs_attr_leaf.c
34  *
35  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
36  */
37 
38 /*========================================================================
39  * Function prototypes for the kernel.
40  *========================================================================*/
41 
42 /*
43  * Routines used for growing the Btree.
44  */
45 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
46 				 xfs_dablk_t which_block, struct xfs_buf **bpp);
47 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
48 				   struct xfs_attr3_icleaf_hdr *ichdr,
49 				   struct xfs_da_args *args, int freemap_index);
50 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
51 				   struct xfs_attr3_icleaf_hdr *ichdr,
52 				   struct xfs_buf *leaf_buffer);
53 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
54 						   xfs_da_state_blk_t *blk1,
55 						   xfs_da_state_blk_t *blk2);
56 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
57 			xfs_da_state_blk_t *leaf_blk_1,
58 			struct xfs_attr3_icleaf_hdr *ichdr1,
59 			xfs_da_state_blk_t *leaf_blk_2,
60 			struct xfs_attr3_icleaf_hdr *ichdr2,
61 			int *number_entries_in_blk1,
62 			int *number_usedbytes_in_blk1);
63 
64 /*
65  * Utility routines.
66  */
67 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
68 			struct xfs_attr_leafblock *src_leaf,
69 			struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
70 			struct xfs_attr_leafblock *dst_leaf,
71 			struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
72 			int move_count);
73 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
74 
75 /*
76  * attr3 block 'firstused' conversion helpers.
77  *
78  * firstused refers to the offset of the first used byte of the nameval region
79  * of an attr leaf block. The region starts at the tail of the block and expands
80  * backwards towards the middle. As such, firstused is initialized to the block
81  * size for an empty leaf block and is reduced from there.
82  *
83  * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
84  * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
85  * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
86  * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
87  * the attr block size. The following helpers manage the conversion between the
88  * in-core and on-disk formats.
89  */
90 
91 static void
92 xfs_attr3_leaf_firstused_from_disk(
93 	struct xfs_da_geometry		*geo,
94 	struct xfs_attr3_icleaf_hdr	*to,
95 	struct xfs_attr_leafblock	*from)
96 {
97 	struct xfs_attr3_leaf_hdr	*hdr3;
98 
99 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
100 		hdr3 = (struct xfs_attr3_leaf_hdr *) from;
101 		to->firstused = be16_to_cpu(hdr3->firstused);
102 	} else {
103 		to->firstused = be16_to_cpu(from->hdr.firstused);
104 	}
105 
106 	/*
107 	 * Convert from the magic fsb size value to actual blocksize. This
108 	 * should only occur for empty blocks when the block size overflows
109 	 * 16-bits.
110 	 */
111 	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
112 		ASSERT(!to->count && !to->usedbytes);
113 		ASSERT(geo->blksize > USHRT_MAX);
114 		to->firstused = geo->blksize;
115 	}
116 }
117 
118 static void
119 xfs_attr3_leaf_firstused_to_disk(
120 	struct xfs_da_geometry		*geo,
121 	struct xfs_attr_leafblock	*to,
122 	struct xfs_attr3_icleaf_hdr	*from)
123 {
124 	struct xfs_attr3_leaf_hdr	*hdr3;
125 	uint32_t			firstused;
126 
127 	/* magic value should only be seen on disk */
128 	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
129 
130 	/*
131 	 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
132 	 * value. This only overflows at the max supported value of 64k. Use the
133 	 * magic on-disk value to represent block size in this case.
134 	 */
135 	firstused = from->firstused;
136 	if (firstused > USHRT_MAX) {
137 		ASSERT(from->firstused == geo->blksize);
138 		firstused = XFS_ATTR3_LEAF_NULLOFF;
139 	}
140 
141 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
142 		hdr3 = (struct xfs_attr3_leaf_hdr *) to;
143 		hdr3->firstused = cpu_to_be16(firstused);
144 	} else {
145 		to->hdr.firstused = cpu_to_be16(firstused);
146 	}
147 }
148 
149 void
150 xfs_attr3_leaf_hdr_from_disk(
151 	struct xfs_da_geometry		*geo,
152 	struct xfs_attr3_icleaf_hdr	*to,
153 	struct xfs_attr_leafblock	*from)
154 {
155 	int	i;
156 
157 	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
158 	       from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
159 
160 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
161 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
162 
163 		to->forw = be32_to_cpu(hdr3->info.hdr.forw);
164 		to->back = be32_to_cpu(hdr3->info.hdr.back);
165 		to->magic = be16_to_cpu(hdr3->info.hdr.magic);
166 		to->count = be16_to_cpu(hdr3->count);
167 		to->usedbytes = be16_to_cpu(hdr3->usedbytes);
168 		xfs_attr3_leaf_firstused_from_disk(geo, to, from);
169 		to->holes = hdr3->holes;
170 
171 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
172 			to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
173 			to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
174 		}
175 		return;
176 	}
177 	to->forw = be32_to_cpu(from->hdr.info.forw);
178 	to->back = be32_to_cpu(from->hdr.info.back);
179 	to->magic = be16_to_cpu(from->hdr.info.magic);
180 	to->count = be16_to_cpu(from->hdr.count);
181 	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
182 	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
183 	to->holes = from->hdr.holes;
184 
185 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
186 		to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
187 		to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
188 	}
189 }
190 
191 void
192 xfs_attr3_leaf_hdr_to_disk(
193 	struct xfs_da_geometry		*geo,
194 	struct xfs_attr_leafblock	*to,
195 	struct xfs_attr3_icleaf_hdr	*from)
196 {
197 	int				i;
198 
199 	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
200 	       from->magic == XFS_ATTR3_LEAF_MAGIC);
201 
202 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
203 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
204 
205 		hdr3->info.hdr.forw = cpu_to_be32(from->forw);
206 		hdr3->info.hdr.back = cpu_to_be32(from->back);
207 		hdr3->info.hdr.magic = cpu_to_be16(from->magic);
208 		hdr3->count = cpu_to_be16(from->count);
209 		hdr3->usedbytes = cpu_to_be16(from->usedbytes);
210 		xfs_attr3_leaf_firstused_to_disk(geo, to, from);
211 		hdr3->holes = from->holes;
212 		hdr3->pad1 = 0;
213 
214 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
215 			hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
216 			hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
217 		}
218 		return;
219 	}
220 	to->hdr.info.forw = cpu_to_be32(from->forw);
221 	to->hdr.info.back = cpu_to_be32(from->back);
222 	to->hdr.info.magic = cpu_to_be16(from->magic);
223 	to->hdr.count = cpu_to_be16(from->count);
224 	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
225 	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
226 	to->hdr.holes = from->holes;
227 	to->hdr.pad1 = 0;
228 
229 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
230 		to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
231 		to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
232 	}
233 }
234 
235 static xfs_failaddr_t
236 xfs_attr3_leaf_verify_entry(
237 	struct xfs_mount			*mp,
238 	char					*buf_end,
239 	struct xfs_attr_leafblock		*leaf,
240 	struct xfs_attr3_icleaf_hdr		*leafhdr,
241 	struct xfs_attr_leaf_entry		*ent,
242 	int					idx,
243 	__u32					*last_hashval)
244 {
245 	struct xfs_attr_leaf_name_local		*lentry;
246 	struct xfs_attr_leaf_name_remote	*rentry;
247 	char					*name_end;
248 	unsigned int				nameidx;
249 	unsigned int				namesize;
250 	__u32					hashval;
251 
252 	/* hash order check */
253 	hashval = be32_to_cpu(ent->hashval);
254 	if (hashval < *last_hashval)
255 		return __this_address;
256 	*last_hashval = hashval;
257 
258 	nameidx = be16_to_cpu(ent->nameidx);
259 	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
260 		return __this_address;
261 
262 	/*
263 	 * Check the name information.  The namelen fields are u8 so we can't
264 	 * possibly exceed the maximum name length of 255 bytes.
265 	 */
266 	if (ent->flags & XFS_ATTR_LOCAL) {
267 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
268 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
269 				be16_to_cpu(lentry->valuelen));
270 		name_end = (char *)lentry + namesize;
271 		if (lentry->namelen == 0)
272 			return __this_address;
273 	} else {
274 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
275 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
276 		name_end = (char *)rentry + namesize;
277 		if (rentry->namelen == 0)
278 			return __this_address;
279 		if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
280 		    rentry->valueblk == 0)
281 			return __this_address;
282 	}
283 
284 	if (name_end > buf_end)
285 		return __this_address;
286 
287 	return NULL;
288 }
289 
290 static xfs_failaddr_t
291 xfs_attr3_leaf_verify(
292 	struct xfs_buf			*bp)
293 {
294 	struct xfs_attr3_icleaf_hdr	ichdr;
295 	struct xfs_mount		*mp = bp->b_mount;
296 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
297 	struct xfs_attr_leaf_entry	*entries;
298 	struct xfs_attr_leaf_entry	*ent;
299 	char				*buf_end;
300 	uint32_t			end;	/* must be 32bit - see below */
301 	__u32				last_hashval = 0;
302 	int				i;
303 	xfs_failaddr_t			fa;
304 
305 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
306 
307 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
308 	if (fa)
309 		return fa;
310 
311 	/*
312 	 * In recovery there is a transient state where count == 0 is valid
313 	 * because we may have transitioned an empty shortform attr to a leaf
314 	 * if the attr didn't fit in shortform.
315 	 */
316 	if (!xfs_log_in_recovery(mp) && ichdr.count == 0)
317 		return __this_address;
318 
319 	/*
320 	 * firstused is the block offset of the first name info structure.
321 	 * Make sure it doesn't go off the block or crash into the header.
322 	 */
323 	if (ichdr.firstused > mp->m_attr_geo->blksize)
324 		return __this_address;
325 	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
326 		return __this_address;
327 
328 	/* Make sure the entries array doesn't crash into the name info. */
329 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
330 	if ((char *)&entries[ichdr.count] >
331 	    (char *)bp->b_addr + ichdr.firstused)
332 		return __this_address;
333 
334 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
335 	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
336 		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
337 				ent, i, &last_hashval);
338 		if (fa)
339 			return fa;
340 	}
341 
342 	/*
343 	 * Quickly check the freemap information.  Attribute data has to be
344 	 * aligned to 4-byte boundaries, and likewise for the free space.
345 	 *
346 	 * Note that for 64k block size filesystems, the freemap entries cannot
347 	 * overflow as they are only be16 fields. However, when checking end
348 	 * pointer of the freemap, we have to be careful to detect overflows and
349 	 * so use uint32_t for those checks.
350 	 */
351 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
352 		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
353 			return __this_address;
354 		if (ichdr.freemap[i].base & 0x3)
355 			return __this_address;
356 		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
357 			return __this_address;
358 		if (ichdr.freemap[i].size & 0x3)
359 			return __this_address;
360 
361 		/* be care of 16 bit overflows here */
362 		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
363 		if (end < ichdr.freemap[i].base)
364 			return __this_address;
365 		if (end > mp->m_attr_geo->blksize)
366 			return __this_address;
367 	}
368 
369 	return NULL;
370 }
371 
372 static void
373 xfs_attr3_leaf_write_verify(
374 	struct xfs_buf	*bp)
375 {
376 	struct xfs_mount	*mp = bp->b_mount;
377 	struct xfs_buf_log_item	*bip = bp->b_log_item;
378 	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
379 	xfs_failaddr_t		fa;
380 
381 	fa = xfs_attr3_leaf_verify(bp);
382 	if (fa) {
383 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
384 		return;
385 	}
386 
387 	if (!xfs_sb_version_hascrc(&mp->m_sb))
388 		return;
389 
390 	if (bip)
391 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
392 
393 	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
394 }
395 
396 /*
397  * leaf/node format detection on trees is sketchy, so a node read can be done on
398  * leaf level blocks when detection identifies the tree as a node format tree
399  * incorrectly. In this case, we need to swap the verifier to match the correct
400  * format of the block being read.
401  */
402 static void
403 xfs_attr3_leaf_read_verify(
404 	struct xfs_buf		*bp)
405 {
406 	struct xfs_mount	*mp = bp->b_mount;
407 	xfs_failaddr_t		fa;
408 
409 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
410 	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
411 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
412 	else {
413 		fa = xfs_attr3_leaf_verify(bp);
414 		if (fa)
415 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
416 	}
417 }
418 
419 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
420 	.name = "xfs_attr3_leaf",
421 	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
422 		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
423 	.verify_read = xfs_attr3_leaf_read_verify,
424 	.verify_write = xfs_attr3_leaf_write_verify,
425 	.verify_struct = xfs_attr3_leaf_verify,
426 };
427 
428 int
429 xfs_attr3_leaf_read(
430 	struct xfs_trans	*tp,
431 	struct xfs_inode	*dp,
432 	xfs_dablk_t		bno,
433 	struct xfs_buf		**bpp)
434 {
435 	int			err;
436 
437 	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
438 			&xfs_attr3_leaf_buf_ops);
439 	if (!err && tp && *bpp)
440 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
441 	return err;
442 }
443 
444 /*========================================================================
445  * Namespace helper routines
446  *========================================================================*/
447 
448 static bool
449 xfs_attr_match(
450 	struct xfs_da_args	*args,
451 	uint8_t			namelen,
452 	unsigned char		*name,
453 	int			flags)
454 {
455 	if (args->namelen != namelen)
456 		return false;
457 	if (memcmp(args->name, name, namelen) != 0)
458 		return false;
459 	/*
460 	 * If we are looking for incomplete entries, show only those, else only
461 	 * show complete entries.
462 	 */
463 	if (args->attr_filter !=
464 	    (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
465 		return false;
466 	return true;
467 }
468 
469 static int
470 xfs_attr_copy_value(
471 	struct xfs_da_args	*args,
472 	unsigned char		*value,
473 	int			valuelen)
474 {
475 	/*
476 	 * No copy if all we have to do is get the length
477 	 */
478 	if (!args->valuelen) {
479 		args->valuelen = valuelen;
480 		return 0;
481 	}
482 
483 	/*
484 	 * No copy if the length of the existing buffer is too small
485 	 */
486 	if (args->valuelen < valuelen) {
487 		args->valuelen = valuelen;
488 		return -ERANGE;
489 	}
490 
491 	if (!args->value) {
492 		args->value = kmem_alloc_large(valuelen, 0);
493 		if (!args->value)
494 			return -ENOMEM;
495 	}
496 	args->valuelen = valuelen;
497 
498 	/* remote block xattr requires IO for copy-in */
499 	if (args->rmtblkno)
500 		return xfs_attr_rmtval_get(args);
501 
502 	/*
503 	 * This is to prevent a GCC warning because the remote xattr case
504 	 * doesn't have a value to pass in. In that case, we never reach here,
505 	 * but GCC can't work that out and so throws a "passing NULL to
506 	 * memcpy" warning.
507 	 */
508 	if (!value)
509 		return -EINVAL;
510 	memcpy(args->value, value, valuelen);
511 	return 0;
512 }
513 
514 /*========================================================================
515  * External routines when attribute fork size < XFS_LITINO(mp).
516  *========================================================================*/
517 
518 /*
519  * Query whether the requested number of additional bytes of extended
520  * attribute space will be able to fit inline.
521  *
522  * Returns zero if not, else the di_forkoff fork offset to be used in the
523  * literal area for attribute data once the new bytes have been added.
524  *
525  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
526  * special case for dev/uuid inodes, they have fixed size data forks.
527  */
528 int
529 xfs_attr_shortform_bytesfit(
530 	struct xfs_inode	*dp,
531 	int			bytes)
532 {
533 	struct xfs_mount	*mp = dp->i_mount;
534 	int64_t			dsize;
535 	int			minforkoff;
536 	int			maxforkoff;
537 	int			offset;
538 
539 	/* rounded down */
540 	offset = (XFS_LITINO(mp) - bytes) >> 3;
541 
542 	if (dp->i_d.di_format == XFS_DINODE_FMT_DEV) {
543 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
544 		return (offset >= minforkoff) ? minforkoff : 0;
545 	}
546 
547 	/*
548 	 * If the requested numbers of bytes is smaller or equal to the
549 	 * current attribute fork size we can always proceed.
550 	 *
551 	 * Note that if_bytes in the data fork might actually be larger than
552 	 * the current data fork size is due to delalloc extents. In that
553 	 * case either the extent count will go down when they are converted
554 	 * to real extents, or the delalloc conversion will take care of the
555 	 * literal area rebalancing.
556 	 */
557 	if (bytes <= XFS_IFORK_ASIZE(dp))
558 		return dp->i_d.di_forkoff;
559 
560 	/*
561 	 * For attr2 we can try to move the forkoff if there is space in the
562 	 * literal area, but for the old format we are done if there is no
563 	 * space in the fixed attribute fork.
564 	 */
565 	if (!(mp->m_flags & XFS_MOUNT_ATTR2))
566 		return 0;
567 
568 	dsize = dp->i_df.if_bytes;
569 
570 	switch (dp->i_d.di_format) {
571 	case XFS_DINODE_FMT_EXTENTS:
572 		/*
573 		 * If there is no attr fork and the data fork is extents,
574 		 * determine if creating the default attr fork will result
575 		 * in the extents form migrating to btree. If so, the
576 		 * minimum offset only needs to be the space required for
577 		 * the btree root.
578 		 */
579 		if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
580 		    xfs_default_attroffset(dp))
581 			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
582 		break;
583 	case XFS_DINODE_FMT_BTREE:
584 		/*
585 		 * If we have a data btree then keep forkoff if we have one,
586 		 * otherwise we are adding a new attr, so then we set
587 		 * minforkoff to where the btree root can finish so we have
588 		 * plenty of room for attrs
589 		 */
590 		if (dp->i_d.di_forkoff) {
591 			if (offset < dp->i_d.di_forkoff)
592 				return 0;
593 			return dp->i_d.di_forkoff;
594 		}
595 		dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
596 		break;
597 	}
598 
599 	/*
600 	 * A data fork btree root must have space for at least
601 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
602 	 */
603 	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
604 	minforkoff = roundup(minforkoff, 8) >> 3;
605 
606 	/* attr fork btree root can have at least this many key/ptr pairs */
607 	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
608 	maxforkoff = maxforkoff >> 3;	/* rounded down */
609 
610 	if (offset >= maxforkoff)
611 		return maxforkoff;
612 	if (offset >= minforkoff)
613 		return offset;
614 	return 0;
615 }
616 
617 /*
618  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
619  */
620 STATIC void
621 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
622 {
623 	if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
624 	    !(xfs_sb_version_hasattr2(&mp->m_sb))) {
625 		spin_lock(&mp->m_sb_lock);
626 		if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
627 			xfs_sb_version_addattr2(&mp->m_sb);
628 			spin_unlock(&mp->m_sb_lock);
629 			xfs_log_sb(tp);
630 		} else
631 			spin_unlock(&mp->m_sb_lock);
632 	}
633 }
634 
635 /*
636  * Create the initial contents of a shortform attribute list.
637  */
638 void
639 xfs_attr_shortform_create(xfs_da_args_t *args)
640 {
641 	xfs_attr_sf_hdr_t *hdr;
642 	xfs_inode_t *dp;
643 	struct xfs_ifork *ifp;
644 
645 	trace_xfs_attr_sf_create(args);
646 
647 	dp = args->dp;
648 	ASSERT(dp != NULL);
649 	ifp = dp->i_afp;
650 	ASSERT(ifp != NULL);
651 	ASSERT(ifp->if_bytes == 0);
652 	if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
653 		ifp->if_flags &= ~XFS_IFEXTENTS;	/* just in case */
654 		dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
655 		ifp->if_flags |= XFS_IFINLINE;
656 	} else {
657 		ASSERT(ifp->if_flags & XFS_IFINLINE);
658 	}
659 	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
660 	hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
661 	hdr->count = 0;
662 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
663 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
664 }
665 
666 /*
667  * Add a name/value pair to the shortform attribute list.
668  * Overflow from the inode has already been checked for.
669  */
670 void
671 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
672 {
673 	xfs_attr_shortform_t *sf;
674 	xfs_attr_sf_entry_t *sfe;
675 	int i, offset, size;
676 	xfs_mount_t *mp;
677 	xfs_inode_t *dp;
678 	struct xfs_ifork *ifp;
679 
680 	trace_xfs_attr_sf_add(args);
681 
682 	dp = args->dp;
683 	mp = dp->i_mount;
684 	dp->i_d.di_forkoff = forkoff;
685 
686 	ifp = dp->i_afp;
687 	ASSERT(ifp->if_flags & XFS_IFINLINE);
688 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
689 	sfe = &sf->list[0];
690 	for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
691 		ASSERT(!xfs_attr_match(args, sfe->namelen, sfe->nameval,
692 			sfe->flags));
693 	}
694 
695 	offset = (char *)sfe - (char *)sf;
696 	size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
697 	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
698 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
699 	sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
700 
701 	sfe->namelen = args->namelen;
702 	sfe->valuelen = args->valuelen;
703 	sfe->flags = args->attr_filter;
704 	memcpy(sfe->nameval, args->name, args->namelen);
705 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
706 	sf->hdr.count++;
707 	be16_add_cpu(&sf->hdr.totsize, size);
708 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
709 
710 	xfs_sbversion_add_attr2(mp, args->trans);
711 }
712 
713 /*
714  * After the last attribute is removed revert to original inode format,
715  * making all literal area available to the data fork once more.
716  */
717 void
718 xfs_attr_fork_remove(
719 	struct xfs_inode	*ip,
720 	struct xfs_trans	*tp)
721 {
722 	xfs_idestroy_fork(ip, XFS_ATTR_FORK);
723 	ip->i_d.di_forkoff = 0;
724 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
725 
726 	ASSERT(ip->i_d.di_anextents == 0);
727 	ASSERT(ip->i_afp == NULL);
728 
729 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
730 }
731 
732 /*
733  * Remove an attribute from the shortform attribute list structure.
734  */
735 int
736 xfs_attr_shortform_remove(xfs_da_args_t *args)
737 {
738 	xfs_attr_shortform_t *sf;
739 	xfs_attr_sf_entry_t *sfe;
740 	int base, size=0, end, totsize, i;
741 	xfs_mount_t *mp;
742 	xfs_inode_t *dp;
743 
744 	trace_xfs_attr_sf_remove(args);
745 
746 	dp = args->dp;
747 	mp = dp->i_mount;
748 	base = sizeof(xfs_attr_sf_hdr_t);
749 	sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
750 	sfe = &sf->list[0];
751 	end = sf->hdr.count;
752 	for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
753 					base += size, i++) {
754 		size = XFS_ATTR_SF_ENTSIZE(sfe);
755 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
756 				sfe->flags))
757 			break;
758 	}
759 	if (i == end)
760 		return -ENOATTR;
761 
762 	/*
763 	 * Fix up the attribute fork data, covering the hole
764 	 */
765 	end = base + size;
766 	totsize = be16_to_cpu(sf->hdr.totsize);
767 	if (end != totsize)
768 		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
769 	sf->hdr.count--;
770 	be16_add_cpu(&sf->hdr.totsize, -size);
771 
772 	/*
773 	 * Fix up the start offset of the attribute fork
774 	 */
775 	totsize -= size;
776 	if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
777 	    (mp->m_flags & XFS_MOUNT_ATTR2) &&
778 	    (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
779 	    !(args->op_flags & XFS_DA_OP_ADDNAME)) {
780 		xfs_attr_fork_remove(dp, args->trans);
781 	} else {
782 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
783 		dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
784 		ASSERT(dp->i_d.di_forkoff);
785 		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
786 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
787 				!(mp->m_flags & XFS_MOUNT_ATTR2) ||
788 				dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
789 		xfs_trans_log_inode(args->trans, dp,
790 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
791 	}
792 
793 	xfs_sbversion_add_attr2(mp, args->trans);
794 
795 	return 0;
796 }
797 
798 /*
799  * Look up a name in a shortform attribute list structure.
800  */
801 /*ARGSUSED*/
802 int
803 xfs_attr_shortform_lookup(xfs_da_args_t *args)
804 {
805 	xfs_attr_shortform_t *sf;
806 	xfs_attr_sf_entry_t *sfe;
807 	int i;
808 	struct xfs_ifork *ifp;
809 
810 	trace_xfs_attr_sf_lookup(args);
811 
812 	ifp = args->dp->i_afp;
813 	ASSERT(ifp->if_flags & XFS_IFINLINE);
814 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
815 	sfe = &sf->list[0];
816 	for (i = 0; i < sf->hdr.count;
817 				sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
818 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
819 				sfe->flags))
820 			return -EEXIST;
821 	}
822 	return -ENOATTR;
823 }
824 
825 /*
826  * Retrieve the attribute value and length.
827  *
828  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
829  * lookup, we only return an error if the attribute does not exist or we can't
830  * retrieve the value.
831  */
832 int
833 xfs_attr_shortform_getvalue(
834 	struct xfs_da_args	*args)
835 {
836 	struct xfs_attr_shortform *sf;
837 	struct xfs_attr_sf_entry *sfe;
838 	int			i;
839 
840 	ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
841 	sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
842 	sfe = &sf->list[0];
843 	for (i = 0; i < sf->hdr.count;
844 				sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
845 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
846 				sfe->flags))
847 			return xfs_attr_copy_value(args,
848 				&sfe->nameval[args->namelen], sfe->valuelen);
849 	}
850 	return -ENOATTR;
851 }
852 
853 /*
854  * Convert from using the shortform to the leaf.  On success, return the
855  * buffer so that we can keep it locked until we're totally done with it.
856  */
857 int
858 xfs_attr_shortform_to_leaf(
859 	struct xfs_da_args		*args,
860 	struct xfs_buf			**leaf_bp)
861 {
862 	struct xfs_inode		*dp;
863 	struct xfs_attr_shortform	*sf;
864 	struct xfs_attr_sf_entry	*sfe;
865 	struct xfs_da_args		nargs;
866 	char				*tmpbuffer;
867 	int				error, i, size;
868 	xfs_dablk_t			blkno;
869 	struct xfs_buf			*bp;
870 	struct xfs_ifork		*ifp;
871 
872 	trace_xfs_attr_sf_to_leaf(args);
873 
874 	dp = args->dp;
875 	ifp = dp->i_afp;
876 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
877 	size = be16_to_cpu(sf->hdr.totsize);
878 	tmpbuffer = kmem_alloc(size, 0);
879 	ASSERT(tmpbuffer != NULL);
880 	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
881 	sf = (xfs_attr_shortform_t *)tmpbuffer;
882 
883 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
884 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
885 
886 	bp = NULL;
887 	error = xfs_da_grow_inode(args, &blkno);
888 	if (error)
889 		goto out;
890 
891 	ASSERT(blkno == 0);
892 	error = xfs_attr3_leaf_create(args, blkno, &bp);
893 	if (error)
894 		goto out;
895 
896 	memset((char *)&nargs, 0, sizeof(nargs));
897 	nargs.dp = dp;
898 	nargs.geo = args->geo;
899 	nargs.total = args->total;
900 	nargs.whichfork = XFS_ATTR_FORK;
901 	nargs.trans = args->trans;
902 	nargs.op_flags = XFS_DA_OP_OKNOENT;
903 
904 	sfe = &sf->list[0];
905 	for (i = 0; i < sf->hdr.count; i++) {
906 		nargs.name = sfe->nameval;
907 		nargs.namelen = sfe->namelen;
908 		nargs.value = &sfe->nameval[nargs.namelen];
909 		nargs.valuelen = sfe->valuelen;
910 		nargs.hashval = xfs_da_hashname(sfe->nameval,
911 						sfe->namelen);
912 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
913 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
914 		ASSERT(error == -ENOATTR);
915 		error = xfs_attr3_leaf_add(bp, &nargs);
916 		ASSERT(error != -ENOSPC);
917 		if (error)
918 			goto out;
919 		sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
920 	}
921 	error = 0;
922 	*leaf_bp = bp;
923 out:
924 	kmem_free(tmpbuffer);
925 	return error;
926 }
927 
928 /*
929  * Check a leaf attribute block to see if all the entries would fit into
930  * a shortform attribute list.
931  */
932 int
933 xfs_attr_shortform_allfit(
934 	struct xfs_buf		*bp,
935 	struct xfs_inode	*dp)
936 {
937 	struct xfs_attr_leafblock *leaf;
938 	struct xfs_attr_leaf_entry *entry;
939 	xfs_attr_leaf_name_local_t *name_loc;
940 	struct xfs_attr3_icleaf_hdr leafhdr;
941 	int			bytes;
942 	int			i;
943 	struct xfs_mount	*mp = bp->b_mount;
944 
945 	leaf = bp->b_addr;
946 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
947 	entry = xfs_attr3_leaf_entryp(leaf);
948 
949 	bytes = sizeof(struct xfs_attr_sf_hdr);
950 	for (i = 0; i < leafhdr.count; entry++, i++) {
951 		if (entry->flags & XFS_ATTR_INCOMPLETE)
952 			continue;		/* don't copy partial entries */
953 		if (!(entry->flags & XFS_ATTR_LOCAL))
954 			return 0;
955 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
956 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
957 			return 0;
958 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
959 			return 0;
960 		bytes += sizeof(struct xfs_attr_sf_entry) - 1
961 				+ name_loc->namelen
962 				+ be16_to_cpu(name_loc->valuelen);
963 	}
964 	if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
965 	    (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
966 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
967 		return -1;
968 	return xfs_attr_shortform_bytesfit(dp, bytes);
969 }
970 
971 /* Verify the consistency of an inline attribute fork. */
972 xfs_failaddr_t
973 xfs_attr_shortform_verify(
974 	struct xfs_inode		*ip)
975 {
976 	struct xfs_attr_shortform	*sfp;
977 	struct xfs_attr_sf_entry	*sfep;
978 	struct xfs_attr_sf_entry	*next_sfep;
979 	char				*endp;
980 	struct xfs_ifork		*ifp;
981 	int				i;
982 	int64_t				size;
983 
984 	ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL);
985 	ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
986 	sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
987 	size = ifp->if_bytes;
988 
989 	/*
990 	 * Give up if the attribute is way too short.
991 	 */
992 	if (size < sizeof(struct xfs_attr_sf_hdr))
993 		return __this_address;
994 
995 	endp = (char *)sfp + size;
996 
997 	/* Check all reported entries */
998 	sfep = &sfp->list[0];
999 	for (i = 0; i < sfp->hdr.count; i++) {
1000 		/*
1001 		 * struct xfs_attr_sf_entry has a variable length.
1002 		 * Check the fixed-offset parts of the structure are
1003 		 * within the data buffer.
1004 		 */
1005 		if (((char *)sfep + sizeof(*sfep)) >= endp)
1006 			return __this_address;
1007 
1008 		/* Don't allow names with known bad length. */
1009 		if (sfep->namelen == 0)
1010 			return __this_address;
1011 
1012 		/*
1013 		 * Check that the variable-length part of the structure is
1014 		 * within the data buffer.  The next entry starts after the
1015 		 * name component, so nextentry is an acceptable test.
1016 		 */
1017 		next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep);
1018 		if ((char *)next_sfep > endp)
1019 			return __this_address;
1020 
1021 		/*
1022 		 * Check for unknown flags.  Short form doesn't support
1023 		 * the incomplete or local bits, so we can use the namespace
1024 		 * mask here.
1025 		 */
1026 		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1027 			return __this_address;
1028 
1029 		/*
1030 		 * Check for invalid namespace combinations.  We only allow
1031 		 * one namespace flag per xattr, so we can just count the
1032 		 * bits (i.e. hweight) here.
1033 		 */
1034 		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1035 			return __this_address;
1036 
1037 		sfep = next_sfep;
1038 	}
1039 	if ((void *)sfep != (void *)endp)
1040 		return __this_address;
1041 
1042 	return NULL;
1043 }
1044 
1045 /*
1046  * Convert a leaf attribute list to shortform attribute list
1047  */
1048 int
1049 xfs_attr3_leaf_to_shortform(
1050 	struct xfs_buf		*bp,
1051 	struct xfs_da_args	*args,
1052 	int			forkoff)
1053 {
1054 	struct xfs_attr_leafblock *leaf;
1055 	struct xfs_attr3_icleaf_hdr ichdr;
1056 	struct xfs_attr_leaf_entry *entry;
1057 	struct xfs_attr_leaf_name_local *name_loc;
1058 	struct xfs_da_args	nargs;
1059 	struct xfs_inode	*dp = args->dp;
1060 	char			*tmpbuffer;
1061 	int			error;
1062 	int			i;
1063 
1064 	trace_xfs_attr_leaf_to_sf(args);
1065 
1066 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1067 	if (!tmpbuffer)
1068 		return -ENOMEM;
1069 
1070 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1071 
1072 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1073 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1074 	entry = xfs_attr3_leaf_entryp(leaf);
1075 
1076 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1077 	memset(bp->b_addr, 0, args->geo->blksize);
1078 
1079 	/*
1080 	 * Clean out the prior contents of the attribute list.
1081 	 */
1082 	error = xfs_da_shrink_inode(args, 0, bp);
1083 	if (error)
1084 		goto out;
1085 
1086 	if (forkoff == -1) {
1087 		ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1088 		ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
1089 		xfs_attr_fork_remove(dp, args->trans);
1090 		goto out;
1091 	}
1092 
1093 	xfs_attr_shortform_create(args);
1094 
1095 	/*
1096 	 * Copy the attributes
1097 	 */
1098 	memset((char *)&nargs, 0, sizeof(nargs));
1099 	nargs.geo = args->geo;
1100 	nargs.dp = dp;
1101 	nargs.total = args->total;
1102 	nargs.whichfork = XFS_ATTR_FORK;
1103 	nargs.trans = args->trans;
1104 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1105 
1106 	for (i = 0; i < ichdr.count; entry++, i++) {
1107 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1108 			continue;	/* don't copy partial entries */
1109 		if (!entry->nameidx)
1110 			continue;
1111 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1112 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1113 		nargs.name = name_loc->nameval;
1114 		nargs.namelen = name_loc->namelen;
1115 		nargs.value = &name_loc->nameval[nargs.namelen];
1116 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1117 		nargs.hashval = be32_to_cpu(entry->hashval);
1118 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1119 		xfs_attr_shortform_add(&nargs, forkoff);
1120 	}
1121 	error = 0;
1122 
1123 out:
1124 	kmem_free(tmpbuffer);
1125 	return error;
1126 }
1127 
1128 /*
1129  * Convert from using a single leaf to a root node and a leaf.
1130  */
1131 int
1132 xfs_attr3_leaf_to_node(
1133 	struct xfs_da_args	*args)
1134 {
1135 	struct xfs_attr_leafblock *leaf;
1136 	struct xfs_attr3_icleaf_hdr icleafhdr;
1137 	struct xfs_attr_leaf_entry *entries;
1138 	struct xfs_da3_icnode_hdr icnodehdr;
1139 	struct xfs_da_intnode	*node;
1140 	struct xfs_inode	*dp = args->dp;
1141 	struct xfs_mount	*mp = dp->i_mount;
1142 	struct xfs_buf		*bp1 = NULL;
1143 	struct xfs_buf		*bp2 = NULL;
1144 	xfs_dablk_t		blkno;
1145 	int			error;
1146 
1147 	trace_xfs_attr_leaf_to_node(args);
1148 
1149 	error = xfs_da_grow_inode(args, &blkno);
1150 	if (error)
1151 		goto out;
1152 	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1153 	if (error)
1154 		goto out;
1155 
1156 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1157 	if (error)
1158 		goto out;
1159 
1160 	/* copy leaf to new buffer, update identifiers */
1161 	xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1162 	bp2->b_ops = bp1->b_ops;
1163 	memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1164 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1165 		struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1166 		hdr3->blkno = cpu_to_be64(bp2->b_bn);
1167 	}
1168 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1169 
1170 	/*
1171 	 * Set up the new root node.
1172 	 */
1173 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1174 	if (error)
1175 		goto out;
1176 	node = bp1->b_addr;
1177 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1178 
1179 	leaf = bp2->b_addr;
1180 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1181 	entries = xfs_attr3_leaf_entryp(leaf);
1182 
1183 	/* both on-disk, don't endian-flip twice */
1184 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1185 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1186 	icnodehdr.count = 1;
1187 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1188 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1189 	error = 0;
1190 out:
1191 	return error;
1192 }
1193 
1194 /*========================================================================
1195  * Routines used for growing the Btree.
1196  *========================================================================*/
1197 
1198 /*
1199  * Create the initial contents of a leaf attribute list
1200  * or a leaf in a node attribute list.
1201  */
1202 STATIC int
1203 xfs_attr3_leaf_create(
1204 	struct xfs_da_args	*args,
1205 	xfs_dablk_t		blkno,
1206 	struct xfs_buf		**bpp)
1207 {
1208 	struct xfs_attr_leafblock *leaf;
1209 	struct xfs_attr3_icleaf_hdr ichdr;
1210 	struct xfs_inode	*dp = args->dp;
1211 	struct xfs_mount	*mp = dp->i_mount;
1212 	struct xfs_buf		*bp;
1213 	int			error;
1214 
1215 	trace_xfs_attr_leaf_create(args);
1216 
1217 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1218 					    XFS_ATTR_FORK);
1219 	if (error)
1220 		return error;
1221 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1222 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1223 	leaf = bp->b_addr;
1224 	memset(leaf, 0, args->geo->blksize);
1225 
1226 	memset(&ichdr, 0, sizeof(ichdr));
1227 	ichdr.firstused = args->geo->blksize;
1228 
1229 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1230 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1231 
1232 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1233 
1234 		hdr3->blkno = cpu_to_be64(bp->b_bn);
1235 		hdr3->owner = cpu_to_be64(dp->i_ino);
1236 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1237 
1238 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1239 	} else {
1240 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1241 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1242 	}
1243 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1244 
1245 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1246 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1247 
1248 	*bpp = bp;
1249 	return 0;
1250 }
1251 
1252 /*
1253  * Split the leaf node, rebalance, then add the new entry.
1254  */
1255 int
1256 xfs_attr3_leaf_split(
1257 	struct xfs_da_state	*state,
1258 	struct xfs_da_state_blk	*oldblk,
1259 	struct xfs_da_state_blk	*newblk)
1260 {
1261 	xfs_dablk_t blkno;
1262 	int error;
1263 
1264 	trace_xfs_attr_leaf_split(state->args);
1265 
1266 	/*
1267 	 * Allocate space for a new leaf node.
1268 	 */
1269 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1270 	error = xfs_da_grow_inode(state->args, &blkno);
1271 	if (error)
1272 		return error;
1273 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1274 	if (error)
1275 		return error;
1276 	newblk->blkno = blkno;
1277 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1278 
1279 	/*
1280 	 * Rebalance the entries across the two leaves.
1281 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1282 	 */
1283 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1284 	error = xfs_da3_blk_link(state, oldblk, newblk);
1285 	if (error)
1286 		return error;
1287 
1288 	/*
1289 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1290 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1291 	 * "new" attrs info.  Will need the "old" info to remove it later.
1292 	 *
1293 	 * Insert the "new" entry in the correct block.
1294 	 */
1295 	if (state->inleaf) {
1296 		trace_xfs_attr_leaf_add_old(state->args);
1297 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1298 	} else {
1299 		trace_xfs_attr_leaf_add_new(state->args);
1300 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1301 	}
1302 
1303 	/*
1304 	 * Update last hashval in each block since we added the name.
1305 	 */
1306 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1307 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1308 	return error;
1309 }
1310 
1311 /*
1312  * Add a name to the leaf attribute list structure.
1313  */
1314 int
1315 xfs_attr3_leaf_add(
1316 	struct xfs_buf		*bp,
1317 	struct xfs_da_args	*args)
1318 {
1319 	struct xfs_attr_leafblock *leaf;
1320 	struct xfs_attr3_icleaf_hdr ichdr;
1321 	int			tablesize;
1322 	int			entsize;
1323 	int			sum;
1324 	int			tmp;
1325 	int			i;
1326 
1327 	trace_xfs_attr_leaf_add(args);
1328 
1329 	leaf = bp->b_addr;
1330 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1331 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1332 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1333 
1334 	/*
1335 	 * Search through freemap for first-fit on new name length.
1336 	 * (may need to figure in size of entry struct too)
1337 	 */
1338 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1339 					+ xfs_attr3_leaf_hdr_size(leaf);
1340 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1341 		if (tablesize > ichdr.firstused) {
1342 			sum += ichdr.freemap[i].size;
1343 			continue;
1344 		}
1345 		if (!ichdr.freemap[i].size)
1346 			continue;	/* no space in this map */
1347 		tmp = entsize;
1348 		if (ichdr.freemap[i].base < ichdr.firstused)
1349 			tmp += sizeof(xfs_attr_leaf_entry_t);
1350 		if (ichdr.freemap[i].size >= tmp) {
1351 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1352 			goto out_log_hdr;
1353 		}
1354 		sum += ichdr.freemap[i].size;
1355 	}
1356 
1357 	/*
1358 	 * If there are no holes in the address space of the block,
1359 	 * and we don't have enough freespace, then compaction will do us
1360 	 * no good and we should just give up.
1361 	 */
1362 	if (!ichdr.holes && sum < entsize)
1363 		return -ENOSPC;
1364 
1365 	/*
1366 	 * Compact the entries to coalesce free space.
1367 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1368 	 */
1369 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1370 
1371 	/*
1372 	 * After compaction, the block is guaranteed to have only one
1373 	 * free region, in freemap[0].  If it is not big enough, give up.
1374 	 */
1375 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1376 		tmp = -ENOSPC;
1377 		goto out_log_hdr;
1378 	}
1379 
1380 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1381 
1382 out_log_hdr:
1383 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1384 	xfs_trans_log_buf(args->trans, bp,
1385 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1386 				xfs_attr3_leaf_hdr_size(leaf)));
1387 	return tmp;
1388 }
1389 
1390 /*
1391  * Add a name to a leaf attribute list structure.
1392  */
1393 STATIC int
1394 xfs_attr3_leaf_add_work(
1395 	struct xfs_buf		*bp,
1396 	struct xfs_attr3_icleaf_hdr *ichdr,
1397 	struct xfs_da_args	*args,
1398 	int			mapindex)
1399 {
1400 	struct xfs_attr_leafblock *leaf;
1401 	struct xfs_attr_leaf_entry *entry;
1402 	struct xfs_attr_leaf_name_local *name_loc;
1403 	struct xfs_attr_leaf_name_remote *name_rmt;
1404 	struct xfs_mount	*mp;
1405 	int			tmp;
1406 	int			i;
1407 
1408 	trace_xfs_attr_leaf_add_work(args);
1409 
1410 	leaf = bp->b_addr;
1411 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1412 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1413 
1414 	/*
1415 	 * Force open some space in the entry array and fill it in.
1416 	 */
1417 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1418 	if (args->index < ichdr->count) {
1419 		tmp  = ichdr->count - args->index;
1420 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1421 		memmove(entry + 1, entry, tmp);
1422 		xfs_trans_log_buf(args->trans, bp,
1423 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1424 	}
1425 	ichdr->count++;
1426 
1427 	/*
1428 	 * Allocate space for the new string (at the end of the run).
1429 	 */
1430 	mp = args->trans->t_mountp;
1431 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1432 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1433 	ASSERT(ichdr->freemap[mapindex].size >=
1434 		xfs_attr_leaf_newentsize(args, NULL));
1435 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1436 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1437 
1438 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1439 
1440 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1441 				     ichdr->freemap[mapindex].size);
1442 	entry->hashval = cpu_to_be32(args->hashval);
1443 	entry->flags = args->attr_filter;
1444 	if (tmp)
1445 		entry->flags |= XFS_ATTR_LOCAL;
1446 	if (args->op_flags & XFS_DA_OP_RENAME) {
1447 		entry->flags |= XFS_ATTR_INCOMPLETE;
1448 		if ((args->blkno2 == args->blkno) &&
1449 		    (args->index2 <= args->index)) {
1450 			args->index2++;
1451 		}
1452 	}
1453 	xfs_trans_log_buf(args->trans, bp,
1454 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1455 	ASSERT((args->index == 0) ||
1456 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1457 	ASSERT((args->index == ichdr->count - 1) ||
1458 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1459 
1460 	/*
1461 	 * For "remote" attribute values, simply note that we need to
1462 	 * allocate space for the "remote" value.  We can't actually
1463 	 * allocate the extents in this transaction, and we can't decide
1464 	 * which blocks they should be as we might allocate more blocks
1465 	 * as part of this transaction (a split operation for example).
1466 	 */
1467 	if (entry->flags & XFS_ATTR_LOCAL) {
1468 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1469 		name_loc->namelen = args->namelen;
1470 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1471 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1472 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1473 				   be16_to_cpu(name_loc->valuelen));
1474 	} else {
1475 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1476 		name_rmt->namelen = args->namelen;
1477 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1478 		entry->flags |= XFS_ATTR_INCOMPLETE;
1479 		/* just in case */
1480 		name_rmt->valuelen = 0;
1481 		name_rmt->valueblk = 0;
1482 		args->rmtblkno = 1;
1483 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1484 		args->rmtvaluelen = args->valuelen;
1485 	}
1486 	xfs_trans_log_buf(args->trans, bp,
1487 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1488 				   xfs_attr_leaf_entsize(leaf, args->index)));
1489 
1490 	/*
1491 	 * Update the control info for this leaf node
1492 	 */
1493 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1494 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1495 
1496 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1497 					+ xfs_attr3_leaf_hdr_size(leaf));
1498 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1499 					+ xfs_attr3_leaf_hdr_size(leaf);
1500 
1501 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1502 		if (ichdr->freemap[i].base == tmp) {
1503 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1504 			ichdr->freemap[i].size -=
1505 				min_t(uint16_t, ichdr->freemap[i].size,
1506 						sizeof(xfs_attr_leaf_entry_t));
1507 		}
1508 	}
1509 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1510 	return 0;
1511 }
1512 
1513 /*
1514  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1515  */
1516 STATIC void
1517 xfs_attr3_leaf_compact(
1518 	struct xfs_da_args	*args,
1519 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1520 	struct xfs_buf		*bp)
1521 {
1522 	struct xfs_attr_leafblock *leaf_src;
1523 	struct xfs_attr_leafblock *leaf_dst;
1524 	struct xfs_attr3_icleaf_hdr ichdr_src;
1525 	struct xfs_trans	*trans = args->trans;
1526 	char			*tmpbuffer;
1527 
1528 	trace_xfs_attr_leaf_compact(args);
1529 
1530 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1531 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1532 	memset(bp->b_addr, 0, args->geo->blksize);
1533 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1534 	leaf_dst = bp->b_addr;
1535 
1536 	/*
1537 	 * Copy the on-disk header back into the destination buffer to ensure
1538 	 * all the information in the header that is not part of the incore
1539 	 * header structure is preserved.
1540 	 */
1541 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1542 
1543 	/* Initialise the incore headers */
1544 	ichdr_src = *ichdr_dst;	/* struct copy */
1545 	ichdr_dst->firstused = args->geo->blksize;
1546 	ichdr_dst->usedbytes = 0;
1547 	ichdr_dst->count = 0;
1548 	ichdr_dst->holes = 0;
1549 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1550 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1551 						ichdr_dst->freemap[0].base;
1552 
1553 	/* write the header back to initialise the underlying buffer */
1554 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1555 
1556 	/*
1557 	 * Copy all entry's in the same (sorted) order,
1558 	 * but allocate name/value pairs packed and in sequence.
1559 	 */
1560 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1561 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1562 	/*
1563 	 * this logs the entire buffer, but the caller must write the header
1564 	 * back to the buffer when it is finished modifying it.
1565 	 */
1566 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1567 
1568 	kmem_free(tmpbuffer);
1569 }
1570 
1571 /*
1572  * Compare two leaf blocks "order".
1573  * Return 0 unless leaf2 should go before leaf1.
1574  */
1575 static int
1576 xfs_attr3_leaf_order(
1577 	struct xfs_buf	*leaf1_bp,
1578 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1579 	struct xfs_buf	*leaf2_bp,
1580 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1581 {
1582 	struct xfs_attr_leaf_entry *entries1;
1583 	struct xfs_attr_leaf_entry *entries2;
1584 
1585 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1586 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1587 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1588 	    ((be32_to_cpu(entries2[0].hashval) <
1589 	      be32_to_cpu(entries1[0].hashval)) ||
1590 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1591 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1592 		return 1;
1593 	}
1594 	return 0;
1595 }
1596 
1597 int
1598 xfs_attr_leaf_order(
1599 	struct xfs_buf	*leaf1_bp,
1600 	struct xfs_buf	*leaf2_bp)
1601 {
1602 	struct xfs_attr3_icleaf_hdr ichdr1;
1603 	struct xfs_attr3_icleaf_hdr ichdr2;
1604 	struct xfs_mount *mp = leaf1_bp->b_mount;
1605 
1606 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1607 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1608 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1609 }
1610 
1611 /*
1612  * Redistribute the attribute list entries between two leaf nodes,
1613  * taking into account the size of the new entry.
1614  *
1615  * NOTE: if new block is empty, then it will get the upper half of the
1616  * old block.  At present, all (one) callers pass in an empty second block.
1617  *
1618  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1619  * to match what it is doing in splitting the attribute leaf block.  Those
1620  * values are used in "atomic rename" operations on attributes.  Note that
1621  * the "new" and "old" values can end up in different blocks.
1622  */
1623 STATIC void
1624 xfs_attr3_leaf_rebalance(
1625 	struct xfs_da_state	*state,
1626 	struct xfs_da_state_blk	*blk1,
1627 	struct xfs_da_state_blk	*blk2)
1628 {
1629 	struct xfs_da_args	*args;
1630 	struct xfs_attr_leafblock *leaf1;
1631 	struct xfs_attr_leafblock *leaf2;
1632 	struct xfs_attr3_icleaf_hdr ichdr1;
1633 	struct xfs_attr3_icleaf_hdr ichdr2;
1634 	struct xfs_attr_leaf_entry *entries1;
1635 	struct xfs_attr_leaf_entry *entries2;
1636 	int			count;
1637 	int			totallen;
1638 	int			max;
1639 	int			space;
1640 	int			swap;
1641 
1642 	/*
1643 	 * Set up environment.
1644 	 */
1645 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1646 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1647 	leaf1 = blk1->bp->b_addr;
1648 	leaf2 = blk2->bp->b_addr;
1649 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1650 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1651 	ASSERT(ichdr2.count == 0);
1652 	args = state->args;
1653 
1654 	trace_xfs_attr_leaf_rebalance(args);
1655 
1656 	/*
1657 	 * Check ordering of blocks, reverse if it makes things simpler.
1658 	 *
1659 	 * NOTE: Given that all (current) callers pass in an empty
1660 	 * second block, this code should never set "swap".
1661 	 */
1662 	swap = 0;
1663 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1664 		swap(blk1, blk2);
1665 
1666 		/* swap structures rather than reconverting them */
1667 		swap(ichdr1, ichdr2);
1668 
1669 		leaf1 = blk1->bp->b_addr;
1670 		leaf2 = blk2->bp->b_addr;
1671 		swap = 1;
1672 	}
1673 
1674 	/*
1675 	 * Examine entries until we reduce the absolute difference in
1676 	 * byte usage between the two blocks to a minimum.  Then get
1677 	 * the direction to copy and the number of elements to move.
1678 	 *
1679 	 * "inleaf" is true if the new entry should be inserted into blk1.
1680 	 * If "swap" is also true, then reverse the sense of "inleaf".
1681 	 */
1682 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1683 						      blk2, &ichdr2,
1684 						      &count, &totallen);
1685 	if (swap)
1686 		state->inleaf = !state->inleaf;
1687 
1688 	/*
1689 	 * Move any entries required from leaf to leaf:
1690 	 */
1691 	if (count < ichdr1.count) {
1692 		/*
1693 		 * Figure the total bytes to be added to the destination leaf.
1694 		 */
1695 		/* number entries being moved */
1696 		count = ichdr1.count - count;
1697 		space  = ichdr1.usedbytes - totallen;
1698 		space += count * sizeof(xfs_attr_leaf_entry_t);
1699 
1700 		/*
1701 		 * leaf2 is the destination, compact it if it looks tight.
1702 		 */
1703 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1704 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1705 		if (space > max)
1706 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1707 
1708 		/*
1709 		 * Move high entries from leaf1 to low end of leaf2.
1710 		 */
1711 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1712 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1713 
1714 	} else if (count > ichdr1.count) {
1715 		/*
1716 		 * I assert that since all callers pass in an empty
1717 		 * second buffer, this code should never execute.
1718 		 */
1719 		ASSERT(0);
1720 
1721 		/*
1722 		 * Figure the total bytes to be added to the destination leaf.
1723 		 */
1724 		/* number entries being moved */
1725 		count -= ichdr1.count;
1726 		space  = totallen - ichdr1.usedbytes;
1727 		space += count * sizeof(xfs_attr_leaf_entry_t);
1728 
1729 		/*
1730 		 * leaf1 is the destination, compact it if it looks tight.
1731 		 */
1732 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1733 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1734 		if (space > max)
1735 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1736 
1737 		/*
1738 		 * Move low entries from leaf2 to high end of leaf1.
1739 		 */
1740 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1741 					ichdr1.count, count);
1742 	}
1743 
1744 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1745 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1746 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1747 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1748 
1749 	/*
1750 	 * Copy out last hashval in each block for B-tree code.
1751 	 */
1752 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1753 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1754 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1755 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1756 
1757 	/*
1758 	 * Adjust the expected index for insertion.
1759 	 * NOTE: this code depends on the (current) situation that the
1760 	 * second block was originally empty.
1761 	 *
1762 	 * If the insertion point moved to the 2nd block, we must adjust
1763 	 * the index.  We must also track the entry just following the
1764 	 * new entry for use in an "atomic rename" operation, that entry
1765 	 * is always the "old" entry and the "new" entry is what we are
1766 	 * inserting.  The index/blkno fields refer to the "old" entry,
1767 	 * while the index2/blkno2 fields refer to the "new" entry.
1768 	 */
1769 	if (blk1->index > ichdr1.count) {
1770 		ASSERT(state->inleaf == 0);
1771 		blk2->index = blk1->index - ichdr1.count;
1772 		args->index = args->index2 = blk2->index;
1773 		args->blkno = args->blkno2 = blk2->blkno;
1774 	} else if (blk1->index == ichdr1.count) {
1775 		if (state->inleaf) {
1776 			args->index = blk1->index;
1777 			args->blkno = blk1->blkno;
1778 			args->index2 = 0;
1779 			args->blkno2 = blk2->blkno;
1780 		} else {
1781 			/*
1782 			 * On a double leaf split, the original attr location
1783 			 * is already stored in blkno2/index2, so don't
1784 			 * overwrite it overwise we corrupt the tree.
1785 			 */
1786 			blk2->index = blk1->index - ichdr1.count;
1787 			args->index = blk2->index;
1788 			args->blkno = blk2->blkno;
1789 			if (!state->extravalid) {
1790 				/*
1791 				 * set the new attr location to match the old
1792 				 * one and let the higher level split code
1793 				 * decide where in the leaf to place it.
1794 				 */
1795 				args->index2 = blk2->index;
1796 				args->blkno2 = blk2->blkno;
1797 			}
1798 		}
1799 	} else {
1800 		ASSERT(state->inleaf == 1);
1801 		args->index = args->index2 = blk1->index;
1802 		args->blkno = args->blkno2 = blk1->blkno;
1803 	}
1804 }
1805 
1806 /*
1807  * Examine entries until we reduce the absolute difference in
1808  * byte usage between the two blocks to a minimum.
1809  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1810  * GROT: there will always be enough room in either block for a new entry.
1811  * GROT: Do a double-split for this case?
1812  */
1813 STATIC int
1814 xfs_attr3_leaf_figure_balance(
1815 	struct xfs_da_state		*state,
1816 	struct xfs_da_state_blk		*blk1,
1817 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1818 	struct xfs_da_state_blk		*blk2,
1819 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1820 	int				*countarg,
1821 	int				*usedbytesarg)
1822 {
1823 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1824 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1825 	struct xfs_attr_leaf_entry	*entry;
1826 	int				count;
1827 	int				max;
1828 	int				index;
1829 	int				totallen = 0;
1830 	int				half;
1831 	int				lastdelta;
1832 	int				foundit = 0;
1833 	int				tmp;
1834 
1835 	/*
1836 	 * Examine entries until we reduce the absolute difference in
1837 	 * byte usage between the two blocks to a minimum.
1838 	 */
1839 	max = ichdr1->count + ichdr2->count;
1840 	half = (max + 1) * sizeof(*entry);
1841 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1842 			xfs_attr_leaf_newentsize(state->args, NULL);
1843 	half /= 2;
1844 	lastdelta = state->args->geo->blksize;
1845 	entry = xfs_attr3_leaf_entryp(leaf1);
1846 	for (count = index = 0; count < max; entry++, index++, count++) {
1847 
1848 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1849 		/*
1850 		 * The new entry is in the first block, account for it.
1851 		 */
1852 		if (count == blk1->index) {
1853 			tmp = totallen + sizeof(*entry) +
1854 				xfs_attr_leaf_newentsize(state->args, NULL);
1855 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1856 				break;
1857 			lastdelta = XFS_ATTR_ABS(half - tmp);
1858 			totallen = tmp;
1859 			foundit = 1;
1860 		}
1861 
1862 		/*
1863 		 * Wrap around into the second block if necessary.
1864 		 */
1865 		if (count == ichdr1->count) {
1866 			leaf1 = leaf2;
1867 			entry = xfs_attr3_leaf_entryp(leaf1);
1868 			index = 0;
1869 		}
1870 
1871 		/*
1872 		 * Figure out if next leaf entry would be too much.
1873 		 */
1874 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1875 									index);
1876 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1877 			break;
1878 		lastdelta = XFS_ATTR_ABS(half - tmp);
1879 		totallen = tmp;
1880 #undef XFS_ATTR_ABS
1881 	}
1882 
1883 	/*
1884 	 * Calculate the number of usedbytes that will end up in lower block.
1885 	 * If new entry not in lower block, fix up the count.
1886 	 */
1887 	totallen -= count * sizeof(*entry);
1888 	if (foundit) {
1889 		totallen -= sizeof(*entry) +
1890 				xfs_attr_leaf_newentsize(state->args, NULL);
1891 	}
1892 
1893 	*countarg = count;
1894 	*usedbytesarg = totallen;
1895 	return foundit;
1896 }
1897 
1898 /*========================================================================
1899  * Routines used for shrinking the Btree.
1900  *========================================================================*/
1901 
1902 /*
1903  * Check a leaf block and its neighbors to see if the block should be
1904  * collapsed into one or the other neighbor.  Always keep the block
1905  * with the smaller block number.
1906  * If the current block is over 50% full, don't try to join it, return 0.
1907  * If the block is empty, fill in the state structure and return 2.
1908  * If it can be collapsed, fill in the state structure and return 1.
1909  * If nothing can be done, return 0.
1910  *
1911  * GROT: allow for INCOMPLETE entries in calculation.
1912  */
1913 int
1914 xfs_attr3_leaf_toosmall(
1915 	struct xfs_da_state	*state,
1916 	int			*action)
1917 {
1918 	struct xfs_attr_leafblock *leaf;
1919 	struct xfs_da_state_blk	*blk;
1920 	struct xfs_attr3_icleaf_hdr ichdr;
1921 	struct xfs_buf		*bp;
1922 	xfs_dablk_t		blkno;
1923 	int			bytes;
1924 	int			forward;
1925 	int			error;
1926 	int			retval;
1927 	int			i;
1928 
1929 	trace_xfs_attr_leaf_toosmall(state->args);
1930 
1931 	/*
1932 	 * Check for the degenerate case of the block being over 50% full.
1933 	 * If so, it's not worth even looking to see if we might be able
1934 	 * to coalesce with a sibling.
1935 	 */
1936 	blk = &state->path.blk[ state->path.active-1 ];
1937 	leaf = blk->bp->b_addr;
1938 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1939 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
1940 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1941 		ichdr.usedbytes;
1942 	if (bytes > (state->args->geo->blksize >> 1)) {
1943 		*action = 0;	/* blk over 50%, don't try to join */
1944 		return 0;
1945 	}
1946 
1947 	/*
1948 	 * Check for the degenerate case of the block being empty.
1949 	 * If the block is empty, we'll simply delete it, no need to
1950 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1951 	 * to merge with the forward block unless it is NULL.
1952 	 */
1953 	if (ichdr.count == 0) {
1954 		/*
1955 		 * Make altpath point to the block we want to keep and
1956 		 * path point to the block we want to drop (this one).
1957 		 */
1958 		forward = (ichdr.forw != 0);
1959 		memcpy(&state->altpath, &state->path, sizeof(state->path));
1960 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1961 						 0, &retval);
1962 		if (error)
1963 			return error;
1964 		if (retval) {
1965 			*action = 0;
1966 		} else {
1967 			*action = 2;
1968 		}
1969 		return 0;
1970 	}
1971 
1972 	/*
1973 	 * Examine each sibling block to see if we can coalesce with
1974 	 * at least 25% free space to spare.  We need to figure out
1975 	 * whether to merge with the forward or the backward block.
1976 	 * We prefer coalescing with the lower numbered sibling so as
1977 	 * to shrink an attribute list over time.
1978 	 */
1979 	/* start with smaller blk num */
1980 	forward = ichdr.forw < ichdr.back;
1981 	for (i = 0; i < 2; forward = !forward, i++) {
1982 		struct xfs_attr3_icleaf_hdr ichdr2;
1983 		if (forward)
1984 			blkno = ichdr.forw;
1985 		else
1986 			blkno = ichdr.back;
1987 		if (blkno == 0)
1988 			continue;
1989 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1990 					blkno, &bp);
1991 		if (error)
1992 			return error;
1993 
1994 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1995 
1996 		bytes = state->args->geo->blksize -
1997 			(state->args->geo->blksize >> 2) -
1998 			ichdr.usedbytes - ichdr2.usedbytes -
1999 			((ichdr.count + ichdr2.count) *
2000 					sizeof(xfs_attr_leaf_entry_t)) -
2001 			xfs_attr3_leaf_hdr_size(leaf);
2002 
2003 		xfs_trans_brelse(state->args->trans, bp);
2004 		if (bytes >= 0)
2005 			break;	/* fits with at least 25% to spare */
2006 	}
2007 	if (i >= 2) {
2008 		*action = 0;
2009 		return 0;
2010 	}
2011 
2012 	/*
2013 	 * Make altpath point to the block we want to keep (the lower
2014 	 * numbered block) and path point to the block we want to drop.
2015 	 */
2016 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2017 	if (blkno < blk->blkno) {
2018 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2019 						 0, &retval);
2020 	} else {
2021 		error = xfs_da3_path_shift(state, &state->path, forward,
2022 						 0, &retval);
2023 	}
2024 	if (error)
2025 		return error;
2026 	if (retval) {
2027 		*action = 0;
2028 	} else {
2029 		*action = 1;
2030 	}
2031 	return 0;
2032 }
2033 
2034 /*
2035  * Remove a name from the leaf attribute list structure.
2036  *
2037  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2038  * If two leaves are 37% full, when combined they will leave 25% free.
2039  */
2040 int
2041 xfs_attr3_leaf_remove(
2042 	struct xfs_buf		*bp,
2043 	struct xfs_da_args	*args)
2044 {
2045 	struct xfs_attr_leafblock *leaf;
2046 	struct xfs_attr3_icleaf_hdr ichdr;
2047 	struct xfs_attr_leaf_entry *entry;
2048 	int			before;
2049 	int			after;
2050 	int			smallest;
2051 	int			entsize;
2052 	int			tablesize;
2053 	int			tmp;
2054 	int			i;
2055 
2056 	trace_xfs_attr_leaf_remove(args);
2057 
2058 	leaf = bp->b_addr;
2059 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2060 
2061 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2062 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2063 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2064 					xfs_attr3_leaf_hdr_size(leaf));
2065 
2066 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2067 
2068 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2069 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2070 
2071 	/*
2072 	 * Scan through free region table:
2073 	 *    check for adjacency of free'd entry with an existing one,
2074 	 *    find smallest free region in case we need to replace it,
2075 	 *    adjust any map that borders the entry table,
2076 	 */
2077 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2078 					+ xfs_attr3_leaf_hdr_size(leaf);
2079 	tmp = ichdr.freemap[0].size;
2080 	before = after = -1;
2081 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2082 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2083 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2084 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2085 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2086 		if (ichdr.freemap[i].base == tablesize) {
2087 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2088 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2089 		}
2090 
2091 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2092 				be16_to_cpu(entry->nameidx)) {
2093 			before = i;
2094 		} else if (ichdr.freemap[i].base ==
2095 				(be16_to_cpu(entry->nameidx) + entsize)) {
2096 			after = i;
2097 		} else if (ichdr.freemap[i].size < tmp) {
2098 			tmp = ichdr.freemap[i].size;
2099 			smallest = i;
2100 		}
2101 	}
2102 
2103 	/*
2104 	 * Coalesce adjacent freemap regions,
2105 	 * or replace the smallest region.
2106 	 */
2107 	if ((before >= 0) || (after >= 0)) {
2108 		if ((before >= 0) && (after >= 0)) {
2109 			ichdr.freemap[before].size += entsize;
2110 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2111 			ichdr.freemap[after].base = 0;
2112 			ichdr.freemap[after].size = 0;
2113 		} else if (before >= 0) {
2114 			ichdr.freemap[before].size += entsize;
2115 		} else {
2116 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2117 			ichdr.freemap[after].size += entsize;
2118 		}
2119 	} else {
2120 		/*
2121 		 * Replace smallest region (if it is smaller than free'd entry)
2122 		 */
2123 		if (ichdr.freemap[smallest].size < entsize) {
2124 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2125 			ichdr.freemap[smallest].size = entsize;
2126 		}
2127 	}
2128 
2129 	/*
2130 	 * Did we remove the first entry?
2131 	 */
2132 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2133 		smallest = 1;
2134 	else
2135 		smallest = 0;
2136 
2137 	/*
2138 	 * Compress the remaining entries and zero out the removed stuff.
2139 	 */
2140 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2141 	ichdr.usedbytes -= entsize;
2142 	xfs_trans_log_buf(args->trans, bp,
2143 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2144 				   entsize));
2145 
2146 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2147 	memmove(entry, entry + 1, tmp);
2148 	ichdr.count--;
2149 	xfs_trans_log_buf(args->trans, bp,
2150 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2151 
2152 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2153 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2154 
2155 	/*
2156 	 * If we removed the first entry, re-find the first used byte
2157 	 * in the name area.  Note that if the entry was the "firstused",
2158 	 * then we don't have a "hole" in our block resulting from
2159 	 * removing the name.
2160 	 */
2161 	if (smallest) {
2162 		tmp = args->geo->blksize;
2163 		entry = xfs_attr3_leaf_entryp(leaf);
2164 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2165 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2166 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2167 
2168 			if (be16_to_cpu(entry->nameidx) < tmp)
2169 				tmp = be16_to_cpu(entry->nameidx);
2170 		}
2171 		ichdr.firstused = tmp;
2172 		ASSERT(ichdr.firstused != 0);
2173 	} else {
2174 		ichdr.holes = 1;	/* mark as needing compaction */
2175 	}
2176 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2177 	xfs_trans_log_buf(args->trans, bp,
2178 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2179 					  xfs_attr3_leaf_hdr_size(leaf)));
2180 
2181 	/*
2182 	 * Check if leaf is less than 50% full, caller may want to
2183 	 * "join" the leaf with a sibling if so.
2184 	 */
2185 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2186 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2187 
2188 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2189 }
2190 
2191 /*
2192  * Move all the attribute list entries from drop_leaf into save_leaf.
2193  */
2194 void
2195 xfs_attr3_leaf_unbalance(
2196 	struct xfs_da_state	*state,
2197 	struct xfs_da_state_blk	*drop_blk,
2198 	struct xfs_da_state_blk	*save_blk)
2199 {
2200 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2201 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2202 	struct xfs_attr3_icleaf_hdr drophdr;
2203 	struct xfs_attr3_icleaf_hdr savehdr;
2204 	struct xfs_attr_leaf_entry *entry;
2205 
2206 	trace_xfs_attr_leaf_unbalance(state->args);
2207 
2208 	drop_leaf = drop_blk->bp->b_addr;
2209 	save_leaf = save_blk->bp->b_addr;
2210 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2211 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2212 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2213 
2214 	/*
2215 	 * Save last hashval from dying block for later Btree fixup.
2216 	 */
2217 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2218 
2219 	/*
2220 	 * Check if we need a temp buffer, or can we do it in place.
2221 	 * Note that we don't check "leaf" for holes because we will
2222 	 * always be dropping it, toosmall() decided that for us already.
2223 	 */
2224 	if (savehdr.holes == 0) {
2225 		/*
2226 		 * dest leaf has no holes, so we add there.  May need
2227 		 * to make some room in the entry array.
2228 		 */
2229 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2230 					 drop_blk->bp, &drophdr)) {
2231 			xfs_attr3_leaf_moveents(state->args,
2232 						drop_leaf, &drophdr, 0,
2233 						save_leaf, &savehdr, 0,
2234 						drophdr.count);
2235 		} else {
2236 			xfs_attr3_leaf_moveents(state->args,
2237 						drop_leaf, &drophdr, 0,
2238 						save_leaf, &savehdr,
2239 						savehdr.count, drophdr.count);
2240 		}
2241 	} else {
2242 		/*
2243 		 * Destination has holes, so we make a temporary copy
2244 		 * of the leaf and add them both to that.
2245 		 */
2246 		struct xfs_attr_leafblock *tmp_leaf;
2247 		struct xfs_attr3_icleaf_hdr tmphdr;
2248 
2249 		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2250 
2251 		/*
2252 		 * Copy the header into the temp leaf so that all the stuff
2253 		 * not in the incore header is present and gets copied back in
2254 		 * once we've moved all the entries.
2255 		 */
2256 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2257 
2258 		memset(&tmphdr, 0, sizeof(tmphdr));
2259 		tmphdr.magic = savehdr.magic;
2260 		tmphdr.forw = savehdr.forw;
2261 		tmphdr.back = savehdr.back;
2262 		tmphdr.firstused = state->args->geo->blksize;
2263 
2264 		/* write the header to the temp buffer to initialise it */
2265 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2266 
2267 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2268 					 drop_blk->bp, &drophdr)) {
2269 			xfs_attr3_leaf_moveents(state->args,
2270 						drop_leaf, &drophdr, 0,
2271 						tmp_leaf, &tmphdr, 0,
2272 						drophdr.count);
2273 			xfs_attr3_leaf_moveents(state->args,
2274 						save_leaf, &savehdr, 0,
2275 						tmp_leaf, &tmphdr, tmphdr.count,
2276 						savehdr.count);
2277 		} else {
2278 			xfs_attr3_leaf_moveents(state->args,
2279 						save_leaf, &savehdr, 0,
2280 						tmp_leaf, &tmphdr, 0,
2281 						savehdr.count);
2282 			xfs_attr3_leaf_moveents(state->args,
2283 						drop_leaf, &drophdr, 0,
2284 						tmp_leaf, &tmphdr, tmphdr.count,
2285 						drophdr.count);
2286 		}
2287 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2288 		savehdr = tmphdr; /* struct copy */
2289 		kmem_free(tmp_leaf);
2290 	}
2291 
2292 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2293 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2294 					   state->args->geo->blksize - 1);
2295 
2296 	/*
2297 	 * Copy out last hashval in each block for B-tree code.
2298 	 */
2299 	entry = xfs_attr3_leaf_entryp(save_leaf);
2300 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2301 }
2302 
2303 /*========================================================================
2304  * Routines used for finding things in the Btree.
2305  *========================================================================*/
2306 
2307 /*
2308  * Look up a name in a leaf attribute list structure.
2309  * This is the internal routine, it uses the caller's buffer.
2310  *
2311  * Note that duplicate keys are allowed, but only check within the
2312  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2313  *
2314  * Return in args->index the index into the entry[] array of either
2315  * the found entry, or where the entry should have been (insert before
2316  * that entry).
2317  *
2318  * Don't change the args->value unless we find the attribute.
2319  */
2320 int
2321 xfs_attr3_leaf_lookup_int(
2322 	struct xfs_buf		*bp,
2323 	struct xfs_da_args	*args)
2324 {
2325 	struct xfs_attr_leafblock *leaf;
2326 	struct xfs_attr3_icleaf_hdr ichdr;
2327 	struct xfs_attr_leaf_entry *entry;
2328 	struct xfs_attr_leaf_entry *entries;
2329 	struct xfs_attr_leaf_name_local *name_loc;
2330 	struct xfs_attr_leaf_name_remote *name_rmt;
2331 	xfs_dahash_t		hashval;
2332 	int			probe;
2333 	int			span;
2334 
2335 	trace_xfs_attr_leaf_lookup(args);
2336 
2337 	leaf = bp->b_addr;
2338 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2339 	entries = xfs_attr3_leaf_entryp(leaf);
2340 	if (ichdr.count >= args->geo->blksize / 8) {
2341 		xfs_buf_mark_corrupt(bp);
2342 		return -EFSCORRUPTED;
2343 	}
2344 
2345 	/*
2346 	 * Binary search.  (note: small blocks will skip this loop)
2347 	 */
2348 	hashval = args->hashval;
2349 	probe = span = ichdr.count / 2;
2350 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2351 		span /= 2;
2352 		if (be32_to_cpu(entry->hashval) < hashval)
2353 			probe += span;
2354 		else if (be32_to_cpu(entry->hashval) > hashval)
2355 			probe -= span;
2356 		else
2357 			break;
2358 	}
2359 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2360 		xfs_buf_mark_corrupt(bp);
2361 		return -EFSCORRUPTED;
2362 	}
2363 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2364 		xfs_buf_mark_corrupt(bp);
2365 		return -EFSCORRUPTED;
2366 	}
2367 
2368 	/*
2369 	 * Since we may have duplicate hashval's, find the first matching
2370 	 * hashval in the leaf.
2371 	 */
2372 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2373 		entry--;
2374 		probe--;
2375 	}
2376 	while (probe < ichdr.count &&
2377 	       be32_to_cpu(entry->hashval) < hashval) {
2378 		entry++;
2379 		probe++;
2380 	}
2381 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2382 		args->index = probe;
2383 		return -ENOATTR;
2384 	}
2385 
2386 	/*
2387 	 * Duplicate keys may be present, so search all of them for a match.
2388 	 */
2389 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2390 			entry++, probe++) {
2391 /*
2392  * GROT: Add code to remove incomplete entries.
2393  */
2394 		if (entry->flags & XFS_ATTR_LOCAL) {
2395 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2396 			if (!xfs_attr_match(args, name_loc->namelen,
2397 					name_loc->nameval, entry->flags))
2398 				continue;
2399 			args->index = probe;
2400 			return -EEXIST;
2401 		} else {
2402 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2403 			if (!xfs_attr_match(args, name_rmt->namelen,
2404 					name_rmt->name, entry->flags))
2405 				continue;
2406 			args->index = probe;
2407 			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2408 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2409 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2410 							args->dp->i_mount,
2411 							args->rmtvaluelen);
2412 			return -EEXIST;
2413 		}
2414 	}
2415 	args->index = probe;
2416 	return -ENOATTR;
2417 }
2418 
2419 /*
2420  * Get the value associated with an attribute name from a leaf attribute
2421  * list structure.
2422  *
2423  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2424  * lookup, we only return an error if the attribute does not exist or we can't
2425  * retrieve the value.
2426  */
2427 int
2428 xfs_attr3_leaf_getvalue(
2429 	struct xfs_buf		*bp,
2430 	struct xfs_da_args	*args)
2431 {
2432 	struct xfs_attr_leafblock *leaf;
2433 	struct xfs_attr3_icleaf_hdr ichdr;
2434 	struct xfs_attr_leaf_entry *entry;
2435 	struct xfs_attr_leaf_name_local *name_loc;
2436 	struct xfs_attr_leaf_name_remote *name_rmt;
2437 
2438 	leaf = bp->b_addr;
2439 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2440 	ASSERT(ichdr.count < args->geo->blksize / 8);
2441 	ASSERT(args->index < ichdr.count);
2442 
2443 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2444 	if (entry->flags & XFS_ATTR_LOCAL) {
2445 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2446 		ASSERT(name_loc->namelen == args->namelen);
2447 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2448 		return xfs_attr_copy_value(args,
2449 					&name_loc->nameval[args->namelen],
2450 					be16_to_cpu(name_loc->valuelen));
2451 	}
2452 
2453 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2454 	ASSERT(name_rmt->namelen == args->namelen);
2455 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2456 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2457 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2458 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2459 					       args->rmtvaluelen);
2460 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2461 }
2462 
2463 /*========================================================================
2464  * Utility routines.
2465  *========================================================================*/
2466 
2467 /*
2468  * Move the indicated entries from one leaf to another.
2469  * NOTE: this routine modifies both source and destination leaves.
2470  */
2471 /*ARGSUSED*/
2472 STATIC void
2473 xfs_attr3_leaf_moveents(
2474 	struct xfs_da_args		*args,
2475 	struct xfs_attr_leafblock	*leaf_s,
2476 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2477 	int				start_s,
2478 	struct xfs_attr_leafblock	*leaf_d,
2479 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2480 	int				start_d,
2481 	int				count)
2482 {
2483 	struct xfs_attr_leaf_entry	*entry_s;
2484 	struct xfs_attr_leaf_entry	*entry_d;
2485 	int				desti;
2486 	int				tmp;
2487 	int				i;
2488 
2489 	/*
2490 	 * Check for nothing to do.
2491 	 */
2492 	if (count == 0)
2493 		return;
2494 
2495 	/*
2496 	 * Set up environment.
2497 	 */
2498 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2499 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2500 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2501 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2502 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2503 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2504 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2505 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2506 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2507 
2508 	ASSERT(start_s < ichdr_s->count);
2509 	ASSERT(start_d <= ichdr_d->count);
2510 	ASSERT(count <= ichdr_s->count);
2511 
2512 
2513 	/*
2514 	 * Move the entries in the destination leaf up to make a hole?
2515 	 */
2516 	if (start_d < ichdr_d->count) {
2517 		tmp  = ichdr_d->count - start_d;
2518 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2519 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2520 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2521 		memmove(entry_d, entry_s, tmp);
2522 	}
2523 
2524 	/*
2525 	 * Copy all entry's in the same (sorted) order,
2526 	 * but allocate attribute info packed and in sequence.
2527 	 */
2528 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2529 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2530 	desti = start_d;
2531 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2532 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2533 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2534 #ifdef GROT
2535 		/*
2536 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2537 		 * may also need to change the insertion index.  Code turned
2538 		 * off for 6.2, should be revisited later.
2539 		 */
2540 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2541 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2542 			ichdr_s->usedbytes -= tmp;
2543 			ichdr_s->count -= 1;
2544 			entry_d--;	/* to compensate for ++ in loop hdr */
2545 			desti--;
2546 			if ((start_s + i) < offset)
2547 				result++;	/* insertion index adjustment */
2548 		} else {
2549 #endif /* GROT */
2550 			ichdr_d->firstused -= tmp;
2551 			/* both on-disk, don't endian flip twice */
2552 			entry_d->hashval = entry_s->hashval;
2553 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2554 			entry_d->flags = entry_s->flags;
2555 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2556 							<= args->geo->blksize);
2557 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2558 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2559 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2560 							<= args->geo->blksize);
2561 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2562 			ichdr_s->usedbytes -= tmp;
2563 			ichdr_d->usedbytes += tmp;
2564 			ichdr_s->count -= 1;
2565 			ichdr_d->count += 1;
2566 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2567 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2568 			ASSERT(ichdr_d->firstused >= tmp);
2569 #ifdef GROT
2570 		}
2571 #endif /* GROT */
2572 	}
2573 
2574 	/*
2575 	 * Zero out the entries we just copied.
2576 	 */
2577 	if (start_s == ichdr_s->count) {
2578 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2579 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2580 		ASSERT(((char *)entry_s + tmp) <=
2581 		       ((char *)leaf_s + args->geo->blksize));
2582 		memset(entry_s, 0, tmp);
2583 	} else {
2584 		/*
2585 		 * Move the remaining entries down to fill the hole,
2586 		 * then zero the entries at the top.
2587 		 */
2588 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2589 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2590 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2591 		memmove(entry_d, entry_s, tmp);
2592 
2593 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2594 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2595 		ASSERT(((char *)entry_s + tmp) <=
2596 		       ((char *)leaf_s + args->geo->blksize));
2597 		memset(entry_s, 0, tmp);
2598 	}
2599 
2600 	/*
2601 	 * Fill in the freemap information
2602 	 */
2603 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2604 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2605 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2606 	ichdr_d->freemap[1].base = 0;
2607 	ichdr_d->freemap[2].base = 0;
2608 	ichdr_d->freemap[1].size = 0;
2609 	ichdr_d->freemap[2].size = 0;
2610 	ichdr_s->holes = 1;	/* leaf may not be compact */
2611 }
2612 
2613 /*
2614  * Pick up the last hashvalue from a leaf block.
2615  */
2616 xfs_dahash_t
2617 xfs_attr_leaf_lasthash(
2618 	struct xfs_buf	*bp,
2619 	int		*count)
2620 {
2621 	struct xfs_attr3_icleaf_hdr ichdr;
2622 	struct xfs_attr_leaf_entry *entries;
2623 	struct xfs_mount *mp = bp->b_mount;
2624 
2625 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2626 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2627 	if (count)
2628 		*count = ichdr.count;
2629 	if (!ichdr.count)
2630 		return 0;
2631 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2632 }
2633 
2634 /*
2635  * Calculate the number of bytes used to store the indicated attribute
2636  * (whether local or remote only calculate bytes in this block).
2637  */
2638 STATIC int
2639 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2640 {
2641 	struct xfs_attr_leaf_entry *entries;
2642 	xfs_attr_leaf_name_local_t *name_loc;
2643 	xfs_attr_leaf_name_remote_t *name_rmt;
2644 	int size;
2645 
2646 	entries = xfs_attr3_leaf_entryp(leaf);
2647 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2648 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2649 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2650 						   be16_to_cpu(name_loc->valuelen));
2651 	} else {
2652 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2653 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2654 	}
2655 	return size;
2656 }
2657 
2658 /*
2659  * Calculate the number of bytes that would be required to store the new
2660  * attribute (whether local or remote only calculate bytes in this block).
2661  * This routine decides as a side effect whether the attribute will be
2662  * a "local" or a "remote" attribute.
2663  */
2664 int
2665 xfs_attr_leaf_newentsize(
2666 	struct xfs_da_args	*args,
2667 	int			*local)
2668 {
2669 	int			size;
2670 
2671 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2672 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2673 		if (local)
2674 			*local = 1;
2675 		return size;
2676 	}
2677 	if (local)
2678 		*local = 0;
2679 	return xfs_attr_leaf_entsize_remote(args->namelen);
2680 }
2681 
2682 
2683 /*========================================================================
2684  * Manage the INCOMPLETE flag in a leaf entry
2685  *========================================================================*/
2686 
2687 /*
2688  * Clear the INCOMPLETE flag on an entry in a leaf block.
2689  */
2690 int
2691 xfs_attr3_leaf_clearflag(
2692 	struct xfs_da_args	*args)
2693 {
2694 	struct xfs_attr_leafblock *leaf;
2695 	struct xfs_attr_leaf_entry *entry;
2696 	struct xfs_attr_leaf_name_remote *name_rmt;
2697 	struct xfs_buf		*bp;
2698 	int			error;
2699 #ifdef DEBUG
2700 	struct xfs_attr3_icleaf_hdr ichdr;
2701 	xfs_attr_leaf_name_local_t *name_loc;
2702 	int namelen;
2703 	char *name;
2704 #endif /* DEBUG */
2705 
2706 	trace_xfs_attr_leaf_clearflag(args);
2707 	/*
2708 	 * Set up the operation.
2709 	 */
2710 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2711 	if (error)
2712 		return error;
2713 
2714 	leaf = bp->b_addr;
2715 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2716 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2717 
2718 #ifdef DEBUG
2719 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2720 	ASSERT(args->index < ichdr.count);
2721 	ASSERT(args->index >= 0);
2722 
2723 	if (entry->flags & XFS_ATTR_LOCAL) {
2724 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2725 		namelen = name_loc->namelen;
2726 		name = (char *)name_loc->nameval;
2727 	} else {
2728 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2729 		namelen = name_rmt->namelen;
2730 		name = (char *)name_rmt->name;
2731 	}
2732 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2733 	ASSERT(namelen == args->namelen);
2734 	ASSERT(memcmp(name, args->name, namelen) == 0);
2735 #endif /* DEBUG */
2736 
2737 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2738 	xfs_trans_log_buf(args->trans, bp,
2739 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2740 
2741 	if (args->rmtblkno) {
2742 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2743 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2744 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2745 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2746 		xfs_trans_log_buf(args->trans, bp,
2747 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2748 	}
2749 
2750 	/*
2751 	 * Commit the flag value change and start the next trans in series.
2752 	 */
2753 	return xfs_trans_roll_inode(&args->trans, args->dp);
2754 }
2755 
2756 /*
2757  * Set the INCOMPLETE flag on an entry in a leaf block.
2758  */
2759 int
2760 xfs_attr3_leaf_setflag(
2761 	struct xfs_da_args	*args)
2762 {
2763 	struct xfs_attr_leafblock *leaf;
2764 	struct xfs_attr_leaf_entry *entry;
2765 	struct xfs_attr_leaf_name_remote *name_rmt;
2766 	struct xfs_buf		*bp;
2767 	int error;
2768 #ifdef DEBUG
2769 	struct xfs_attr3_icleaf_hdr ichdr;
2770 #endif
2771 
2772 	trace_xfs_attr_leaf_setflag(args);
2773 
2774 	/*
2775 	 * Set up the operation.
2776 	 */
2777 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2778 	if (error)
2779 		return error;
2780 
2781 	leaf = bp->b_addr;
2782 #ifdef DEBUG
2783 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2784 	ASSERT(args->index < ichdr.count);
2785 	ASSERT(args->index >= 0);
2786 #endif
2787 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2788 
2789 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2790 	entry->flags |= XFS_ATTR_INCOMPLETE;
2791 	xfs_trans_log_buf(args->trans, bp,
2792 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2793 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2794 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2795 		name_rmt->valueblk = 0;
2796 		name_rmt->valuelen = 0;
2797 		xfs_trans_log_buf(args->trans, bp,
2798 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2799 	}
2800 
2801 	/*
2802 	 * Commit the flag value change and start the next trans in series.
2803 	 */
2804 	return xfs_trans_roll_inode(&args->trans, args->dp);
2805 }
2806 
2807 /*
2808  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2809  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2810  * entry given by args->blkno2/index2.
2811  *
2812  * Note that they could be in different blocks, or in the same block.
2813  */
2814 int
2815 xfs_attr3_leaf_flipflags(
2816 	struct xfs_da_args	*args)
2817 {
2818 	struct xfs_attr_leafblock *leaf1;
2819 	struct xfs_attr_leafblock *leaf2;
2820 	struct xfs_attr_leaf_entry *entry1;
2821 	struct xfs_attr_leaf_entry *entry2;
2822 	struct xfs_attr_leaf_name_remote *name_rmt;
2823 	struct xfs_buf		*bp1;
2824 	struct xfs_buf		*bp2;
2825 	int error;
2826 #ifdef DEBUG
2827 	struct xfs_attr3_icleaf_hdr ichdr1;
2828 	struct xfs_attr3_icleaf_hdr ichdr2;
2829 	xfs_attr_leaf_name_local_t *name_loc;
2830 	int namelen1, namelen2;
2831 	char *name1, *name2;
2832 #endif /* DEBUG */
2833 
2834 	trace_xfs_attr_leaf_flipflags(args);
2835 
2836 	/*
2837 	 * Read the block containing the "old" attr
2838 	 */
2839 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2840 	if (error)
2841 		return error;
2842 
2843 	/*
2844 	 * Read the block containing the "new" attr, if it is different
2845 	 */
2846 	if (args->blkno2 != args->blkno) {
2847 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2848 					   &bp2);
2849 		if (error)
2850 			return error;
2851 	} else {
2852 		bp2 = bp1;
2853 	}
2854 
2855 	leaf1 = bp1->b_addr;
2856 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2857 
2858 	leaf2 = bp2->b_addr;
2859 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2860 
2861 #ifdef DEBUG
2862 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2863 	ASSERT(args->index < ichdr1.count);
2864 	ASSERT(args->index >= 0);
2865 
2866 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2867 	ASSERT(args->index2 < ichdr2.count);
2868 	ASSERT(args->index2 >= 0);
2869 
2870 	if (entry1->flags & XFS_ATTR_LOCAL) {
2871 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2872 		namelen1 = name_loc->namelen;
2873 		name1 = (char *)name_loc->nameval;
2874 	} else {
2875 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2876 		namelen1 = name_rmt->namelen;
2877 		name1 = (char *)name_rmt->name;
2878 	}
2879 	if (entry2->flags & XFS_ATTR_LOCAL) {
2880 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2881 		namelen2 = name_loc->namelen;
2882 		name2 = (char *)name_loc->nameval;
2883 	} else {
2884 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2885 		namelen2 = name_rmt->namelen;
2886 		name2 = (char *)name_rmt->name;
2887 	}
2888 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2889 	ASSERT(namelen1 == namelen2);
2890 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2891 #endif /* DEBUG */
2892 
2893 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2894 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2895 
2896 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2897 	xfs_trans_log_buf(args->trans, bp1,
2898 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2899 	if (args->rmtblkno) {
2900 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2901 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2902 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2903 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2904 		xfs_trans_log_buf(args->trans, bp1,
2905 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2906 	}
2907 
2908 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2909 	xfs_trans_log_buf(args->trans, bp2,
2910 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2911 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2912 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2913 		name_rmt->valueblk = 0;
2914 		name_rmt->valuelen = 0;
2915 		xfs_trans_log_buf(args->trans, bp2,
2916 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2917 	}
2918 
2919 	/*
2920 	 * Commit the flag value change and start the next trans in series.
2921 	 */
2922 	error = xfs_trans_roll_inode(&args->trans, args->dp);
2923 
2924 	return error;
2925 }
2926