xref: /openbmc/linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision 29c37341)
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 	 * firstused is the block offset of the first name info structure.
313 	 * Make sure it doesn't go off the block or crash into the header.
314 	 */
315 	if (ichdr.firstused > mp->m_attr_geo->blksize)
316 		return __this_address;
317 	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
318 		return __this_address;
319 
320 	/* Make sure the entries array doesn't crash into the name info. */
321 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
322 	if ((char *)&entries[ichdr.count] >
323 	    (char *)bp->b_addr + ichdr.firstused)
324 		return __this_address;
325 
326 	/*
327 	 * NOTE: This verifier historically failed empty leaf buffers because
328 	 * we expect the fork to be in another format. Empty attr fork format
329 	 * conversions are possible during xattr set, however, and format
330 	 * conversion is not atomic with the xattr set that triggers it. We
331 	 * cannot assume leaf blocks are non-empty until that is addressed.
332 	*/
333 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
334 	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
335 		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
336 				ent, i, &last_hashval);
337 		if (fa)
338 			return fa;
339 	}
340 
341 	/*
342 	 * Quickly check the freemap information.  Attribute data has to be
343 	 * aligned to 4-byte boundaries, and likewise for the free space.
344 	 *
345 	 * Note that for 64k block size filesystems, the freemap entries cannot
346 	 * overflow as they are only be16 fields. However, when checking end
347 	 * pointer of the freemap, we have to be careful to detect overflows and
348 	 * so use uint32_t for those checks.
349 	 */
350 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
351 		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
352 			return __this_address;
353 		if (ichdr.freemap[i].base & 0x3)
354 			return __this_address;
355 		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
356 			return __this_address;
357 		if (ichdr.freemap[i].size & 0x3)
358 			return __this_address;
359 
360 		/* be care of 16 bit overflows here */
361 		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
362 		if (end < ichdr.freemap[i].base)
363 			return __this_address;
364 		if (end > mp->m_attr_geo->blksize)
365 			return __this_address;
366 	}
367 
368 	return NULL;
369 }
370 
371 static void
372 xfs_attr3_leaf_write_verify(
373 	struct xfs_buf	*bp)
374 {
375 	struct xfs_mount	*mp = bp->b_mount;
376 	struct xfs_buf_log_item	*bip = bp->b_log_item;
377 	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
378 	xfs_failaddr_t		fa;
379 
380 	fa = xfs_attr3_leaf_verify(bp);
381 	if (fa) {
382 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
383 		return;
384 	}
385 
386 	if (!xfs_sb_version_hascrc(&mp->m_sb))
387 		return;
388 
389 	if (bip)
390 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
391 
392 	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
393 }
394 
395 /*
396  * leaf/node format detection on trees is sketchy, so a node read can be done on
397  * leaf level blocks when detection identifies the tree as a node format tree
398  * incorrectly. In this case, we need to swap the verifier to match the correct
399  * format of the block being read.
400  */
401 static void
402 xfs_attr3_leaf_read_verify(
403 	struct xfs_buf		*bp)
404 {
405 	struct xfs_mount	*mp = bp->b_mount;
406 	xfs_failaddr_t		fa;
407 
408 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
409 	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
410 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
411 	else {
412 		fa = xfs_attr3_leaf_verify(bp);
413 		if (fa)
414 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
415 	}
416 }
417 
418 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
419 	.name = "xfs_attr3_leaf",
420 	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
421 		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
422 	.verify_read = xfs_attr3_leaf_read_verify,
423 	.verify_write = xfs_attr3_leaf_write_verify,
424 	.verify_struct = xfs_attr3_leaf_verify,
425 };
426 
427 int
428 xfs_attr3_leaf_read(
429 	struct xfs_trans	*tp,
430 	struct xfs_inode	*dp,
431 	xfs_dablk_t		bno,
432 	struct xfs_buf		**bpp)
433 {
434 	int			err;
435 
436 	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
437 			&xfs_attr3_leaf_buf_ops);
438 	if (!err && tp && *bpp)
439 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
440 	return err;
441 }
442 
443 /*========================================================================
444  * Namespace helper routines
445  *========================================================================*/
446 
447 static bool
448 xfs_attr_match(
449 	struct xfs_da_args	*args,
450 	uint8_t			namelen,
451 	unsigned char		*name,
452 	int			flags)
453 {
454 	if (args->namelen != namelen)
455 		return false;
456 	if (memcmp(args->name, name, namelen) != 0)
457 		return false;
458 	/*
459 	 * If we are looking for incomplete entries, show only those, else only
460 	 * show complete entries.
461 	 */
462 	if (args->attr_filter !=
463 	    (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
464 		return false;
465 	return true;
466 }
467 
468 static int
469 xfs_attr_copy_value(
470 	struct xfs_da_args	*args,
471 	unsigned char		*value,
472 	int			valuelen)
473 {
474 	/*
475 	 * No copy if all we have to do is get the length
476 	 */
477 	if (!args->valuelen) {
478 		args->valuelen = valuelen;
479 		return 0;
480 	}
481 
482 	/*
483 	 * No copy if the length of the existing buffer is too small
484 	 */
485 	if (args->valuelen < valuelen) {
486 		args->valuelen = valuelen;
487 		return -ERANGE;
488 	}
489 
490 	if (!args->value) {
491 		args->value = kmem_alloc_large(valuelen, KM_NOLOCKDEP);
492 		if (!args->value)
493 			return -ENOMEM;
494 	}
495 	args->valuelen = valuelen;
496 
497 	/* remote block xattr requires IO for copy-in */
498 	if (args->rmtblkno)
499 		return xfs_attr_rmtval_get(args);
500 
501 	/*
502 	 * This is to prevent a GCC warning because the remote xattr case
503 	 * doesn't have a value to pass in. In that case, we never reach here,
504 	 * but GCC can't work that out and so throws a "passing NULL to
505 	 * memcpy" warning.
506 	 */
507 	if (!value)
508 		return -EINVAL;
509 	memcpy(args->value, value, valuelen);
510 	return 0;
511 }
512 
513 /*========================================================================
514  * External routines when attribute fork size < XFS_LITINO(mp).
515  *========================================================================*/
516 
517 /*
518  * Query whether the requested number of additional bytes of extended
519  * attribute space will be able to fit inline.
520  *
521  * Returns zero if not, else the di_forkoff fork offset to be used in the
522  * literal area for attribute data once the new bytes have been added.
523  *
524  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
525  * special case for dev/uuid inodes, they have fixed size data forks.
526  */
527 int
528 xfs_attr_shortform_bytesfit(
529 	struct xfs_inode	*dp,
530 	int			bytes)
531 {
532 	struct xfs_mount	*mp = dp->i_mount;
533 	int64_t			dsize;
534 	int			minforkoff;
535 	int			maxforkoff;
536 	int			offset;
537 
538 	/* rounded down */
539 	offset = (XFS_LITINO(mp) - bytes) >> 3;
540 
541 	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
542 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
543 		return (offset >= minforkoff) ? minforkoff : 0;
544 	}
545 
546 	/*
547 	 * If the requested numbers of bytes is smaller or equal to the
548 	 * current attribute fork size we can always proceed.
549 	 *
550 	 * Note that if_bytes in the data fork might actually be larger than
551 	 * the current data fork size is due to delalloc extents. In that
552 	 * case either the extent count will go down when they are converted
553 	 * to real extents, or the delalloc conversion will take care of the
554 	 * literal area rebalancing.
555 	 */
556 	if (bytes <= XFS_IFORK_ASIZE(dp))
557 		return dp->i_d.di_forkoff;
558 
559 	/*
560 	 * For attr2 we can try to move the forkoff if there is space in the
561 	 * literal area, but for the old format we are done if there is no
562 	 * space in the fixed attribute fork.
563 	 */
564 	if (!(mp->m_flags & XFS_MOUNT_ATTR2))
565 		return 0;
566 
567 	dsize = dp->i_df.if_bytes;
568 
569 	switch (dp->i_df.if_format) {
570 	case XFS_DINODE_FMT_EXTENTS:
571 		/*
572 		 * If there is no attr fork and the data fork is extents,
573 		 * determine if creating the default attr fork will result
574 		 * in the extents form migrating to btree. If so, the
575 		 * minimum offset only needs to be the space required for
576 		 * the btree root.
577 		 */
578 		if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
579 		    xfs_default_attroffset(dp))
580 			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
581 		break;
582 	case XFS_DINODE_FMT_BTREE:
583 		/*
584 		 * If we have a data btree then keep forkoff if we have one,
585 		 * otherwise we are adding a new attr, so then we set
586 		 * minforkoff to where the btree root can finish so we have
587 		 * plenty of room for attrs
588 		 */
589 		if (dp->i_d.di_forkoff) {
590 			if (offset < dp->i_d.di_forkoff)
591 				return 0;
592 			return dp->i_d.di_forkoff;
593 		}
594 		dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
595 		break;
596 	}
597 
598 	/*
599 	 * A data fork btree root must have space for at least
600 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
601 	 */
602 	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
603 	minforkoff = roundup(minforkoff, 8) >> 3;
604 
605 	/* attr fork btree root can have at least this many key/ptr pairs */
606 	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
607 	maxforkoff = maxforkoff >> 3;	/* rounded down */
608 
609 	if (offset >= maxforkoff)
610 		return maxforkoff;
611 	if (offset >= minforkoff)
612 		return offset;
613 	return 0;
614 }
615 
616 /*
617  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
618  */
619 STATIC void
620 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
621 {
622 	if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
623 	    !(xfs_sb_version_hasattr2(&mp->m_sb))) {
624 		spin_lock(&mp->m_sb_lock);
625 		if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
626 			xfs_sb_version_addattr2(&mp->m_sb);
627 			spin_unlock(&mp->m_sb_lock);
628 			xfs_log_sb(tp);
629 		} else
630 			spin_unlock(&mp->m_sb_lock);
631 	}
632 }
633 
634 /*
635  * Create the initial contents of a shortform attribute list.
636  */
637 void
638 xfs_attr_shortform_create(
639 	struct xfs_da_args	*args)
640 {
641 	struct xfs_inode	*dp = args->dp;
642 	struct xfs_ifork	*ifp = dp->i_afp;
643 	struct xfs_attr_sf_hdr	*hdr;
644 
645 	trace_xfs_attr_sf_create(args);
646 
647 	ASSERT(ifp->if_bytes == 0);
648 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS) {
649 		ifp->if_flags &= ~XFS_IFEXTENTS;	/* just in case */
650 		ifp->if_format = XFS_DINODE_FMT_LOCAL;
651 		ifp->if_flags |= XFS_IFINLINE;
652 	} else {
653 		ASSERT(ifp->if_flags & XFS_IFINLINE);
654 	}
655 	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
656 	hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
657 	hdr->count = 0;
658 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
659 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
660 }
661 
662 /*
663  * Return -EEXIST if attr is found, or -ENOATTR if not
664  * args:  args containing attribute name and namelen
665  * sfep:  If not null, pointer will be set to the last attr entry found on
666 	  -EEXIST.  On -ENOATTR pointer is left at the last entry in the list
667  * basep: If not null, pointer is set to the byte offset of the entry in the
668  *	  list on -EEXIST.  On -ENOATTR, pointer is left at the byte offset of
669  *	  the last entry in the list
670  */
671 int
672 xfs_attr_sf_findname(
673 	struct xfs_da_args	 *args,
674 	struct xfs_attr_sf_entry **sfep,
675 	unsigned int		 *basep)
676 {
677 	struct xfs_attr_shortform *sf;
678 	struct xfs_attr_sf_entry *sfe;
679 	unsigned int		base = sizeof(struct xfs_attr_sf_hdr);
680 	int			size = 0;
681 	int			end;
682 	int			i;
683 
684 	sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
685 	sfe = &sf->list[0];
686 	end = sf->hdr.count;
687 	for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
688 			     base += size, i++) {
689 		size = XFS_ATTR_SF_ENTSIZE(sfe);
690 		if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
691 				    sfe->flags))
692 			continue;
693 		break;
694 	}
695 
696 	if (sfep != NULL)
697 		*sfep = sfe;
698 
699 	if (basep != NULL)
700 		*basep = base;
701 
702 	if (i == end)
703 		return -ENOATTR;
704 	return -EEXIST;
705 }
706 
707 /*
708  * Add a name/value pair to the shortform attribute list.
709  * Overflow from the inode has already been checked for.
710  */
711 void
712 xfs_attr_shortform_add(
713 	struct xfs_da_args		*args,
714 	int				forkoff)
715 {
716 	struct xfs_attr_shortform	*sf;
717 	struct xfs_attr_sf_entry	*sfe;
718 	int				offset, size;
719 	struct xfs_mount		*mp;
720 	struct xfs_inode		*dp;
721 	struct xfs_ifork		*ifp;
722 
723 	trace_xfs_attr_sf_add(args);
724 
725 	dp = args->dp;
726 	mp = dp->i_mount;
727 	dp->i_d.di_forkoff = forkoff;
728 
729 	ifp = dp->i_afp;
730 	ASSERT(ifp->if_flags & XFS_IFINLINE);
731 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
732 	if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
733 		ASSERT(0);
734 
735 	offset = (char *)sfe - (char *)sf;
736 	size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
737 	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
738 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
739 	sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
740 
741 	sfe->namelen = args->namelen;
742 	sfe->valuelen = args->valuelen;
743 	sfe->flags = args->attr_filter;
744 	memcpy(sfe->nameval, args->name, args->namelen);
745 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
746 	sf->hdr.count++;
747 	be16_add_cpu(&sf->hdr.totsize, size);
748 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
749 
750 	xfs_sbversion_add_attr2(mp, args->trans);
751 }
752 
753 /*
754  * After the last attribute is removed revert to original inode format,
755  * making all literal area available to the data fork once more.
756  */
757 void
758 xfs_attr_fork_remove(
759 	struct xfs_inode	*ip,
760 	struct xfs_trans	*tp)
761 {
762 	ASSERT(ip->i_afp->if_nextents == 0);
763 
764 	xfs_idestroy_fork(ip->i_afp);
765 	kmem_cache_free(xfs_ifork_zone, ip->i_afp);
766 	ip->i_afp = NULL;
767 	ip->i_d.di_forkoff = 0;
768 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
769 }
770 
771 /*
772  * Remove an attribute from the shortform attribute list structure.
773  */
774 int
775 xfs_attr_shortform_remove(
776 	struct xfs_da_args		*args)
777 {
778 	struct xfs_attr_shortform	*sf;
779 	struct xfs_attr_sf_entry	*sfe;
780 	int				size = 0, end, totsize;
781 	unsigned int			base;
782 	struct xfs_mount		*mp;
783 	struct xfs_inode		*dp;
784 	int				error;
785 
786 	trace_xfs_attr_sf_remove(args);
787 
788 	dp = args->dp;
789 	mp = dp->i_mount;
790 	sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
791 
792 	error = xfs_attr_sf_findname(args, &sfe, &base);
793 	if (error != -EEXIST)
794 		return error;
795 	size = XFS_ATTR_SF_ENTSIZE(sfe);
796 
797 	/*
798 	 * Fix up the attribute fork data, covering the hole
799 	 */
800 	end = base + size;
801 	totsize = be16_to_cpu(sf->hdr.totsize);
802 	if (end != totsize)
803 		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
804 	sf->hdr.count--;
805 	be16_add_cpu(&sf->hdr.totsize, -size);
806 
807 	/*
808 	 * Fix up the start offset of the attribute fork
809 	 */
810 	totsize -= size;
811 	if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
812 	    (mp->m_flags & XFS_MOUNT_ATTR2) &&
813 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
814 	    !(args->op_flags & XFS_DA_OP_ADDNAME)) {
815 		xfs_attr_fork_remove(dp, args->trans);
816 	} else {
817 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
818 		dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
819 		ASSERT(dp->i_d.di_forkoff);
820 		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
821 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
822 				!(mp->m_flags & XFS_MOUNT_ATTR2) ||
823 				dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
824 		xfs_trans_log_inode(args->trans, dp,
825 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
826 	}
827 
828 	xfs_sbversion_add_attr2(mp, args->trans);
829 
830 	return 0;
831 }
832 
833 /*
834  * Look up a name in a shortform attribute list structure.
835  */
836 /*ARGSUSED*/
837 int
838 xfs_attr_shortform_lookup(xfs_da_args_t *args)
839 {
840 	xfs_attr_shortform_t *sf;
841 	xfs_attr_sf_entry_t *sfe;
842 	int i;
843 	struct xfs_ifork *ifp;
844 
845 	trace_xfs_attr_sf_lookup(args);
846 
847 	ifp = args->dp->i_afp;
848 	ASSERT(ifp->if_flags & XFS_IFINLINE);
849 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
850 	sfe = &sf->list[0];
851 	for (i = 0; i < sf->hdr.count;
852 				sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
853 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
854 				sfe->flags))
855 			return -EEXIST;
856 	}
857 	return -ENOATTR;
858 }
859 
860 /*
861  * Retrieve the attribute value and length.
862  *
863  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
864  * lookup, we only return an error if the attribute does not exist or we can't
865  * retrieve the value.
866  */
867 int
868 xfs_attr_shortform_getvalue(
869 	struct xfs_da_args	*args)
870 {
871 	struct xfs_attr_shortform *sf;
872 	struct xfs_attr_sf_entry *sfe;
873 	int			i;
874 
875 	ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
876 	sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
877 	sfe = &sf->list[0];
878 	for (i = 0; i < sf->hdr.count;
879 				sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
880 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
881 				sfe->flags))
882 			return xfs_attr_copy_value(args,
883 				&sfe->nameval[args->namelen], sfe->valuelen);
884 	}
885 	return -ENOATTR;
886 }
887 
888 /*
889  * Convert from using the shortform to the leaf.  On success, return the
890  * buffer so that we can keep it locked until we're totally done with it.
891  */
892 int
893 xfs_attr_shortform_to_leaf(
894 	struct xfs_da_args		*args,
895 	struct xfs_buf			**leaf_bp)
896 {
897 	struct xfs_inode		*dp;
898 	struct xfs_attr_shortform	*sf;
899 	struct xfs_attr_sf_entry	*sfe;
900 	struct xfs_da_args		nargs;
901 	char				*tmpbuffer;
902 	int				error, i, size;
903 	xfs_dablk_t			blkno;
904 	struct xfs_buf			*bp;
905 	struct xfs_ifork		*ifp;
906 
907 	trace_xfs_attr_sf_to_leaf(args);
908 
909 	dp = args->dp;
910 	ifp = dp->i_afp;
911 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
912 	size = be16_to_cpu(sf->hdr.totsize);
913 	tmpbuffer = kmem_alloc(size, 0);
914 	ASSERT(tmpbuffer != NULL);
915 	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
916 	sf = (xfs_attr_shortform_t *)tmpbuffer;
917 
918 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
919 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
920 
921 	bp = NULL;
922 	error = xfs_da_grow_inode(args, &blkno);
923 	if (error)
924 		goto out;
925 
926 	ASSERT(blkno == 0);
927 	error = xfs_attr3_leaf_create(args, blkno, &bp);
928 	if (error)
929 		goto out;
930 
931 	memset((char *)&nargs, 0, sizeof(nargs));
932 	nargs.dp = dp;
933 	nargs.geo = args->geo;
934 	nargs.total = args->total;
935 	nargs.whichfork = XFS_ATTR_FORK;
936 	nargs.trans = args->trans;
937 	nargs.op_flags = XFS_DA_OP_OKNOENT;
938 
939 	sfe = &sf->list[0];
940 	for (i = 0; i < sf->hdr.count; i++) {
941 		nargs.name = sfe->nameval;
942 		nargs.namelen = sfe->namelen;
943 		nargs.value = &sfe->nameval[nargs.namelen];
944 		nargs.valuelen = sfe->valuelen;
945 		nargs.hashval = xfs_da_hashname(sfe->nameval,
946 						sfe->namelen);
947 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
948 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
949 		ASSERT(error == -ENOATTR);
950 		error = xfs_attr3_leaf_add(bp, &nargs);
951 		ASSERT(error != -ENOSPC);
952 		if (error)
953 			goto out;
954 		sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
955 	}
956 	error = 0;
957 	*leaf_bp = bp;
958 out:
959 	kmem_free(tmpbuffer);
960 	return error;
961 }
962 
963 /*
964  * Check a leaf attribute block to see if all the entries would fit into
965  * a shortform attribute list.
966  */
967 int
968 xfs_attr_shortform_allfit(
969 	struct xfs_buf		*bp,
970 	struct xfs_inode	*dp)
971 {
972 	struct xfs_attr_leafblock *leaf;
973 	struct xfs_attr_leaf_entry *entry;
974 	xfs_attr_leaf_name_local_t *name_loc;
975 	struct xfs_attr3_icleaf_hdr leafhdr;
976 	int			bytes;
977 	int			i;
978 	struct xfs_mount	*mp = bp->b_mount;
979 
980 	leaf = bp->b_addr;
981 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
982 	entry = xfs_attr3_leaf_entryp(leaf);
983 
984 	bytes = sizeof(struct xfs_attr_sf_hdr);
985 	for (i = 0; i < leafhdr.count; entry++, i++) {
986 		if (entry->flags & XFS_ATTR_INCOMPLETE)
987 			continue;		/* don't copy partial entries */
988 		if (!(entry->flags & XFS_ATTR_LOCAL))
989 			return 0;
990 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
991 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
992 			return 0;
993 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
994 			return 0;
995 		bytes += sizeof(struct xfs_attr_sf_entry) - 1
996 				+ name_loc->namelen
997 				+ be16_to_cpu(name_loc->valuelen);
998 	}
999 	if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
1000 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1001 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
1002 		return -1;
1003 	return xfs_attr_shortform_bytesfit(dp, bytes);
1004 }
1005 
1006 /* Verify the consistency of an inline attribute fork. */
1007 xfs_failaddr_t
1008 xfs_attr_shortform_verify(
1009 	struct xfs_inode		*ip)
1010 {
1011 	struct xfs_attr_shortform	*sfp;
1012 	struct xfs_attr_sf_entry	*sfep;
1013 	struct xfs_attr_sf_entry	*next_sfep;
1014 	char				*endp;
1015 	struct xfs_ifork		*ifp;
1016 	int				i;
1017 	int64_t				size;
1018 
1019 	ASSERT(ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
1020 	ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
1021 	sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1022 	size = ifp->if_bytes;
1023 
1024 	/*
1025 	 * Give up if the attribute is way too short.
1026 	 */
1027 	if (size < sizeof(struct xfs_attr_sf_hdr))
1028 		return __this_address;
1029 
1030 	endp = (char *)sfp + size;
1031 
1032 	/* Check all reported entries */
1033 	sfep = &sfp->list[0];
1034 	for (i = 0; i < sfp->hdr.count; i++) {
1035 		/*
1036 		 * struct xfs_attr_sf_entry has a variable length.
1037 		 * Check the fixed-offset parts of the structure are
1038 		 * within the data buffer.
1039 		 */
1040 		if (((char *)sfep + sizeof(*sfep)) >= endp)
1041 			return __this_address;
1042 
1043 		/* Don't allow names with known bad length. */
1044 		if (sfep->namelen == 0)
1045 			return __this_address;
1046 
1047 		/*
1048 		 * Check that the variable-length part of the structure is
1049 		 * within the data buffer.  The next entry starts after the
1050 		 * name component, so nextentry is an acceptable test.
1051 		 */
1052 		next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep);
1053 		if ((char *)next_sfep > endp)
1054 			return __this_address;
1055 
1056 		/*
1057 		 * Check for unknown flags.  Short form doesn't support
1058 		 * the incomplete or local bits, so we can use the namespace
1059 		 * mask here.
1060 		 */
1061 		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1062 			return __this_address;
1063 
1064 		/*
1065 		 * Check for invalid namespace combinations.  We only allow
1066 		 * one namespace flag per xattr, so we can just count the
1067 		 * bits (i.e. hweight) here.
1068 		 */
1069 		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1070 			return __this_address;
1071 
1072 		sfep = next_sfep;
1073 	}
1074 	if ((void *)sfep != (void *)endp)
1075 		return __this_address;
1076 
1077 	return NULL;
1078 }
1079 
1080 /*
1081  * Convert a leaf attribute list to shortform attribute list
1082  */
1083 int
1084 xfs_attr3_leaf_to_shortform(
1085 	struct xfs_buf		*bp,
1086 	struct xfs_da_args	*args,
1087 	int			forkoff)
1088 {
1089 	struct xfs_attr_leafblock *leaf;
1090 	struct xfs_attr3_icleaf_hdr ichdr;
1091 	struct xfs_attr_leaf_entry *entry;
1092 	struct xfs_attr_leaf_name_local *name_loc;
1093 	struct xfs_da_args	nargs;
1094 	struct xfs_inode	*dp = args->dp;
1095 	char			*tmpbuffer;
1096 	int			error;
1097 	int			i;
1098 
1099 	trace_xfs_attr_leaf_to_sf(args);
1100 
1101 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1102 	if (!tmpbuffer)
1103 		return -ENOMEM;
1104 
1105 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1106 
1107 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1108 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1109 	entry = xfs_attr3_leaf_entryp(leaf);
1110 
1111 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1112 	memset(bp->b_addr, 0, args->geo->blksize);
1113 
1114 	/*
1115 	 * Clean out the prior contents of the attribute list.
1116 	 */
1117 	error = xfs_da_shrink_inode(args, 0, bp);
1118 	if (error)
1119 		goto out;
1120 
1121 	if (forkoff == -1) {
1122 		ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1123 		ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1124 		xfs_attr_fork_remove(dp, args->trans);
1125 		goto out;
1126 	}
1127 
1128 	xfs_attr_shortform_create(args);
1129 
1130 	/*
1131 	 * Copy the attributes
1132 	 */
1133 	memset((char *)&nargs, 0, sizeof(nargs));
1134 	nargs.geo = args->geo;
1135 	nargs.dp = dp;
1136 	nargs.total = args->total;
1137 	nargs.whichfork = XFS_ATTR_FORK;
1138 	nargs.trans = args->trans;
1139 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1140 
1141 	for (i = 0; i < ichdr.count; entry++, i++) {
1142 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1143 			continue;	/* don't copy partial entries */
1144 		if (!entry->nameidx)
1145 			continue;
1146 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1147 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1148 		nargs.name = name_loc->nameval;
1149 		nargs.namelen = name_loc->namelen;
1150 		nargs.value = &name_loc->nameval[nargs.namelen];
1151 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1152 		nargs.hashval = be32_to_cpu(entry->hashval);
1153 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1154 		xfs_attr_shortform_add(&nargs, forkoff);
1155 	}
1156 	error = 0;
1157 
1158 out:
1159 	kmem_free(tmpbuffer);
1160 	return error;
1161 }
1162 
1163 /*
1164  * Convert from using a single leaf to a root node and a leaf.
1165  */
1166 int
1167 xfs_attr3_leaf_to_node(
1168 	struct xfs_da_args	*args)
1169 {
1170 	struct xfs_attr_leafblock *leaf;
1171 	struct xfs_attr3_icleaf_hdr icleafhdr;
1172 	struct xfs_attr_leaf_entry *entries;
1173 	struct xfs_da3_icnode_hdr icnodehdr;
1174 	struct xfs_da_intnode	*node;
1175 	struct xfs_inode	*dp = args->dp;
1176 	struct xfs_mount	*mp = dp->i_mount;
1177 	struct xfs_buf		*bp1 = NULL;
1178 	struct xfs_buf		*bp2 = NULL;
1179 	xfs_dablk_t		blkno;
1180 	int			error;
1181 
1182 	trace_xfs_attr_leaf_to_node(args);
1183 
1184 	error = xfs_da_grow_inode(args, &blkno);
1185 	if (error)
1186 		goto out;
1187 	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1188 	if (error)
1189 		goto out;
1190 
1191 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1192 	if (error)
1193 		goto out;
1194 
1195 	/* copy leaf to new buffer, update identifiers */
1196 	xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1197 	bp2->b_ops = bp1->b_ops;
1198 	memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1199 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1200 		struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1201 		hdr3->blkno = cpu_to_be64(bp2->b_bn);
1202 	}
1203 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1204 
1205 	/*
1206 	 * Set up the new root node.
1207 	 */
1208 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1209 	if (error)
1210 		goto out;
1211 	node = bp1->b_addr;
1212 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1213 
1214 	leaf = bp2->b_addr;
1215 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1216 	entries = xfs_attr3_leaf_entryp(leaf);
1217 
1218 	/* both on-disk, don't endian-flip twice */
1219 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1220 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1221 	icnodehdr.count = 1;
1222 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1223 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1224 	error = 0;
1225 out:
1226 	return error;
1227 }
1228 
1229 /*========================================================================
1230  * Routines used for growing the Btree.
1231  *========================================================================*/
1232 
1233 /*
1234  * Create the initial contents of a leaf attribute list
1235  * or a leaf in a node attribute list.
1236  */
1237 STATIC int
1238 xfs_attr3_leaf_create(
1239 	struct xfs_da_args	*args,
1240 	xfs_dablk_t		blkno,
1241 	struct xfs_buf		**bpp)
1242 {
1243 	struct xfs_attr_leafblock *leaf;
1244 	struct xfs_attr3_icleaf_hdr ichdr;
1245 	struct xfs_inode	*dp = args->dp;
1246 	struct xfs_mount	*mp = dp->i_mount;
1247 	struct xfs_buf		*bp;
1248 	int			error;
1249 
1250 	trace_xfs_attr_leaf_create(args);
1251 
1252 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1253 					    XFS_ATTR_FORK);
1254 	if (error)
1255 		return error;
1256 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1257 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1258 	leaf = bp->b_addr;
1259 	memset(leaf, 0, args->geo->blksize);
1260 
1261 	memset(&ichdr, 0, sizeof(ichdr));
1262 	ichdr.firstused = args->geo->blksize;
1263 
1264 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1265 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1266 
1267 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1268 
1269 		hdr3->blkno = cpu_to_be64(bp->b_bn);
1270 		hdr3->owner = cpu_to_be64(dp->i_ino);
1271 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1272 
1273 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1274 	} else {
1275 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1276 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1277 	}
1278 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1279 
1280 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1281 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1282 
1283 	*bpp = bp;
1284 	return 0;
1285 }
1286 
1287 /*
1288  * Split the leaf node, rebalance, then add the new entry.
1289  */
1290 int
1291 xfs_attr3_leaf_split(
1292 	struct xfs_da_state	*state,
1293 	struct xfs_da_state_blk	*oldblk,
1294 	struct xfs_da_state_blk	*newblk)
1295 {
1296 	xfs_dablk_t blkno;
1297 	int error;
1298 
1299 	trace_xfs_attr_leaf_split(state->args);
1300 
1301 	/*
1302 	 * Allocate space for a new leaf node.
1303 	 */
1304 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1305 	error = xfs_da_grow_inode(state->args, &blkno);
1306 	if (error)
1307 		return error;
1308 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1309 	if (error)
1310 		return error;
1311 	newblk->blkno = blkno;
1312 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1313 
1314 	/*
1315 	 * Rebalance the entries across the two leaves.
1316 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1317 	 */
1318 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1319 	error = xfs_da3_blk_link(state, oldblk, newblk);
1320 	if (error)
1321 		return error;
1322 
1323 	/*
1324 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1325 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1326 	 * "new" attrs info.  Will need the "old" info to remove it later.
1327 	 *
1328 	 * Insert the "new" entry in the correct block.
1329 	 */
1330 	if (state->inleaf) {
1331 		trace_xfs_attr_leaf_add_old(state->args);
1332 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1333 	} else {
1334 		trace_xfs_attr_leaf_add_new(state->args);
1335 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1336 	}
1337 
1338 	/*
1339 	 * Update last hashval in each block since we added the name.
1340 	 */
1341 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1342 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1343 	return error;
1344 }
1345 
1346 /*
1347  * Add a name to the leaf attribute list structure.
1348  */
1349 int
1350 xfs_attr3_leaf_add(
1351 	struct xfs_buf		*bp,
1352 	struct xfs_da_args	*args)
1353 {
1354 	struct xfs_attr_leafblock *leaf;
1355 	struct xfs_attr3_icleaf_hdr ichdr;
1356 	int			tablesize;
1357 	int			entsize;
1358 	int			sum;
1359 	int			tmp;
1360 	int			i;
1361 
1362 	trace_xfs_attr_leaf_add(args);
1363 
1364 	leaf = bp->b_addr;
1365 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1366 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1367 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1368 
1369 	/*
1370 	 * Search through freemap for first-fit on new name length.
1371 	 * (may need to figure in size of entry struct too)
1372 	 */
1373 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1374 					+ xfs_attr3_leaf_hdr_size(leaf);
1375 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1376 		if (tablesize > ichdr.firstused) {
1377 			sum += ichdr.freemap[i].size;
1378 			continue;
1379 		}
1380 		if (!ichdr.freemap[i].size)
1381 			continue;	/* no space in this map */
1382 		tmp = entsize;
1383 		if (ichdr.freemap[i].base < ichdr.firstused)
1384 			tmp += sizeof(xfs_attr_leaf_entry_t);
1385 		if (ichdr.freemap[i].size >= tmp) {
1386 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1387 			goto out_log_hdr;
1388 		}
1389 		sum += ichdr.freemap[i].size;
1390 	}
1391 
1392 	/*
1393 	 * If there are no holes in the address space of the block,
1394 	 * and we don't have enough freespace, then compaction will do us
1395 	 * no good and we should just give up.
1396 	 */
1397 	if (!ichdr.holes && sum < entsize)
1398 		return -ENOSPC;
1399 
1400 	/*
1401 	 * Compact the entries to coalesce free space.
1402 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1403 	 */
1404 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1405 
1406 	/*
1407 	 * After compaction, the block is guaranteed to have only one
1408 	 * free region, in freemap[0].  If it is not big enough, give up.
1409 	 */
1410 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1411 		tmp = -ENOSPC;
1412 		goto out_log_hdr;
1413 	}
1414 
1415 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1416 
1417 out_log_hdr:
1418 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1419 	xfs_trans_log_buf(args->trans, bp,
1420 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1421 				xfs_attr3_leaf_hdr_size(leaf)));
1422 	return tmp;
1423 }
1424 
1425 /*
1426  * Add a name to a leaf attribute list structure.
1427  */
1428 STATIC int
1429 xfs_attr3_leaf_add_work(
1430 	struct xfs_buf		*bp,
1431 	struct xfs_attr3_icleaf_hdr *ichdr,
1432 	struct xfs_da_args	*args,
1433 	int			mapindex)
1434 {
1435 	struct xfs_attr_leafblock *leaf;
1436 	struct xfs_attr_leaf_entry *entry;
1437 	struct xfs_attr_leaf_name_local *name_loc;
1438 	struct xfs_attr_leaf_name_remote *name_rmt;
1439 	struct xfs_mount	*mp;
1440 	int			tmp;
1441 	int			i;
1442 
1443 	trace_xfs_attr_leaf_add_work(args);
1444 
1445 	leaf = bp->b_addr;
1446 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1447 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1448 
1449 	/*
1450 	 * Force open some space in the entry array and fill it in.
1451 	 */
1452 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1453 	if (args->index < ichdr->count) {
1454 		tmp  = ichdr->count - args->index;
1455 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1456 		memmove(entry + 1, entry, tmp);
1457 		xfs_trans_log_buf(args->trans, bp,
1458 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1459 	}
1460 	ichdr->count++;
1461 
1462 	/*
1463 	 * Allocate space for the new string (at the end of the run).
1464 	 */
1465 	mp = args->trans->t_mountp;
1466 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1467 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1468 	ASSERT(ichdr->freemap[mapindex].size >=
1469 		xfs_attr_leaf_newentsize(args, NULL));
1470 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1471 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1472 
1473 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1474 
1475 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1476 				     ichdr->freemap[mapindex].size);
1477 	entry->hashval = cpu_to_be32(args->hashval);
1478 	entry->flags = args->attr_filter;
1479 	if (tmp)
1480 		entry->flags |= XFS_ATTR_LOCAL;
1481 	if (args->op_flags & XFS_DA_OP_RENAME) {
1482 		entry->flags |= XFS_ATTR_INCOMPLETE;
1483 		if ((args->blkno2 == args->blkno) &&
1484 		    (args->index2 <= args->index)) {
1485 			args->index2++;
1486 		}
1487 	}
1488 	xfs_trans_log_buf(args->trans, bp,
1489 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1490 	ASSERT((args->index == 0) ||
1491 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1492 	ASSERT((args->index == ichdr->count - 1) ||
1493 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1494 
1495 	/*
1496 	 * For "remote" attribute values, simply note that we need to
1497 	 * allocate space for the "remote" value.  We can't actually
1498 	 * allocate the extents in this transaction, and we can't decide
1499 	 * which blocks they should be as we might allocate more blocks
1500 	 * as part of this transaction (a split operation for example).
1501 	 */
1502 	if (entry->flags & XFS_ATTR_LOCAL) {
1503 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1504 		name_loc->namelen = args->namelen;
1505 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1506 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1507 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1508 				   be16_to_cpu(name_loc->valuelen));
1509 	} else {
1510 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1511 		name_rmt->namelen = args->namelen;
1512 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1513 		entry->flags |= XFS_ATTR_INCOMPLETE;
1514 		/* just in case */
1515 		name_rmt->valuelen = 0;
1516 		name_rmt->valueblk = 0;
1517 		args->rmtblkno = 1;
1518 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1519 		args->rmtvaluelen = args->valuelen;
1520 	}
1521 	xfs_trans_log_buf(args->trans, bp,
1522 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1523 				   xfs_attr_leaf_entsize(leaf, args->index)));
1524 
1525 	/*
1526 	 * Update the control info for this leaf node
1527 	 */
1528 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1529 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1530 
1531 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1532 					+ xfs_attr3_leaf_hdr_size(leaf));
1533 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1534 					+ xfs_attr3_leaf_hdr_size(leaf);
1535 
1536 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1537 		if (ichdr->freemap[i].base == tmp) {
1538 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1539 			ichdr->freemap[i].size -=
1540 				min_t(uint16_t, ichdr->freemap[i].size,
1541 						sizeof(xfs_attr_leaf_entry_t));
1542 		}
1543 	}
1544 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1545 	return 0;
1546 }
1547 
1548 /*
1549  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1550  */
1551 STATIC void
1552 xfs_attr3_leaf_compact(
1553 	struct xfs_da_args	*args,
1554 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1555 	struct xfs_buf		*bp)
1556 {
1557 	struct xfs_attr_leafblock *leaf_src;
1558 	struct xfs_attr_leafblock *leaf_dst;
1559 	struct xfs_attr3_icleaf_hdr ichdr_src;
1560 	struct xfs_trans	*trans = args->trans;
1561 	char			*tmpbuffer;
1562 
1563 	trace_xfs_attr_leaf_compact(args);
1564 
1565 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1566 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1567 	memset(bp->b_addr, 0, args->geo->blksize);
1568 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1569 	leaf_dst = bp->b_addr;
1570 
1571 	/*
1572 	 * Copy the on-disk header back into the destination buffer to ensure
1573 	 * all the information in the header that is not part of the incore
1574 	 * header structure is preserved.
1575 	 */
1576 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1577 
1578 	/* Initialise the incore headers */
1579 	ichdr_src = *ichdr_dst;	/* struct copy */
1580 	ichdr_dst->firstused = args->geo->blksize;
1581 	ichdr_dst->usedbytes = 0;
1582 	ichdr_dst->count = 0;
1583 	ichdr_dst->holes = 0;
1584 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1585 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1586 						ichdr_dst->freemap[0].base;
1587 
1588 	/* write the header back to initialise the underlying buffer */
1589 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1590 
1591 	/*
1592 	 * Copy all entry's in the same (sorted) order,
1593 	 * but allocate name/value pairs packed and in sequence.
1594 	 */
1595 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1596 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1597 	/*
1598 	 * this logs the entire buffer, but the caller must write the header
1599 	 * back to the buffer when it is finished modifying it.
1600 	 */
1601 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1602 
1603 	kmem_free(tmpbuffer);
1604 }
1605 
1606 /*
1607  * Compare two leaf blocks "order".
1608  * Return 0 unless leaf2 should go before leaf1.
1609  */
1610 static int
1611 xfs_attr3_leaf_order(
1612 	struct xfs_buf	*leaf1_bp,
1613 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1614 	struct xfs_buf	*leaf2_bp,
1615 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1616 {
1617 	struct xfs_attr_leaf_entry *entries1;
1618 	struct xfs_attr_leaf_entry *entries2;
1619 
1620 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1621 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1622 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1623 	    ((be32_to_cpu(entries2[0].hashval) <
1624 	      be32_to_cpu(entries1[0].hashval)) ||
1625 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1626 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1627 		return 1;
1628 	}
1629 	return 0;
1630 }
1631 
1632 int
1633 xfs_attr_leaf_order(
1634 	struct xfs_buf	*leaf1_bp,
1635 	struct xfs_buf	*leaf2_bp)
1636 {
1637 	struct xfs_attr3_icleaf_hdr ichdr1;
1638 	struct xfs_attr3_icleaf_hdr ichdr2;
1639 	struct xfs_mount *mp = leaf1_bp->b_mount;
1640 
1641 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1642 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1643 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1644 }
1645 
1646 /*
1647  * Redistribute the attribute list entries between two leaf nodes,
1648  * taking into account the size of the new entry.
1649  *
1650  * NOTE: if new block is empty, then it will get the upper half of the
1651  * old block.  At present, all (one) callers pass in an empty second block.
1652  *
1653  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1654  * to match what it is doing in splitting the attribute leaf block.  Those
1655  * values are used in "atomic rename" operations on attributes.  Note that
1656  * the "new" and "old" values can end up in different blocks.
1657  */
1658 STATIC void
1659 xfs_attr3_leaf_rebalance(
1660 	struct xfs_da_state	*state,
1661 	struct xfs_da_state_blk	*blk1,
1662 	struct xfs_da_state_blk	*blk2)
1663 {
1664 	struct xfs_da_args	*args;
1665 	struct xfs_attr_leafblock *leaf1;
1666 	struct xfs_attr_leafblock *leaf2;
1667 	struct xfs_attr3_icleaf_hdr ichdr1;
1668 	struct xfs_attr3_icleaf_hdr ichdr2;
1669 	struct xfs_attr_leaf_entry *entries1;
1670 	struct xfs_attr_leaf_entry *entries2;
1671 	int			count;
1672 	int			totallen;
1673 	int			max;
1674 	int			space;
1675 	int			swap;
1676 
1677 	/*
1678 	 * Set up environment.
1679 	 */
1680 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1681 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1682 	leaf1 = blk1->bp->b_addr;
1683 	leaf2 = blk2->bp->b_addr;
1684 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1685 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1686 	ASSERT(ichdr2.count == 0);
1687 	args = state->args;
1688 
1689 	trace_xfs_attr_leaf_rebalance(args);
1690 
1691 	/*
1692 	 * Check ordering of blocks, reverse if it makes things simpler.
1693 	 *
1694 	 * NOTE: Given that all (current) callers pass in an empty
1695 	 * second block, this code should never set "swap".
1696 	 */
1697 	swap = 0;
1698 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1699 		swap(blk1, blk2);
1700 
1701 		/* swap structures rather than reconverting them */
1702 		swap(ichdr1, ichdr2);
1703 
1704 		leaf1 = blk1->bp->b_addr;
1705 		leaf2 = blk2->bp->b_addr;
1706 		swap = 1;
1707 	}
1708 
1709 	/*
1710 	 * Examine entries until we reduce the absolute difference in
1711 	 * byte usage between the two blocks to a minimum.  Then get
1712 	 * the direction to copy and the number of elements to move.
1713 	 *
1714 	 * "inleaf" is true if the new entry should be inserted into blk1.
1715 	 * If "swap" is also true, then reverse the sense of "inleaf".
1716 	 */
1717 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1718 						      blk2, &ichdr2,
1719 						      &count, &totallen);
1720 	if (swap)
1721 		state->inleaf = !state->inleaf;
1722 
1723 	/*
1724 	 * Move any entries required from leaf to leaf:
1725 	 */
1726 	if (count < ichdr1.count) {
1727 		/*
1728 		 * Figure the total bytes to be added to the destination leaf.
1729 		 */
1730 		/* number entries being moved */
1731 		count = ichdr1.count - count;
1732 		space  = ichdr1.usedbytes - totallen;
1733 		space += count * sizeof(xfs_attr_leaf_entry_t);
1734 
1735 		/*
1736 		 * leaf2 is the destination, compact it if it looks tight.
1737 		 */
1738 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1739 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1740 		if (space > max)
1741 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1742 
1743 		/*
1744 		 * Move high entries from leaf1 to low end of leaf2.
1745 		 */
1746 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1747 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1748 
1749 	} else if (count > ichdr1.count) {
1750 		/*
1751 		 * I assert that since all callers pass in an empty
1752 		 * second buffer, this code should never execute.
1753 		 */
1754 		ASSERT(0);
1755 
1756 		/*
1757 		 * Figure the total bytes to be added to the destination leaf.
1758 		 */
1759 		/* number entries being moved */
1760 		count -= ichdr1.count;
1761 		space  = totallen - ichdr1.usedbytes;
1762 		space += count * sizeof(xfs_attr_leaf_entry_t);
1763 
1764 		/*
1765 		 * leaf1 is the destination, compact it if it looks tight.
1766 		 */
1767 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1768 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1769 		if (space > max)
1770 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1771 
1772 		/*
1773 		 * Move low entries from leaf2 to high end of leaf1.
1774 		 */
1775 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1776 					ichdr1.count, count);
1777 	}
1778 
1779 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1780 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1781 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1782 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1783 
1784 	/*
1785 	 * Copy out last hashval in each block for B-tree code.
1786 	 */
1787 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1788 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1789 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1790 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1791 
1792 	/*
1793 	 * Adjust the expected index for insertion.
1794 	 * NOTE: this code depends on the (current) situation that the
1795 	 * second block was originally empty.
1796 	 *
1797 	 * If the insertion point moved to the 2nd block, we must adjust
1798 	 * the index.  We must also track the entry just following the
1799 	 * new entry for use in an "atomic rename" operation, that entry
1800 	 * is always the "old" entry and the "new" entry is what we are
1801 	 * inserting.  The index/blkno fields refer to the "old" entry,
1802 	 * while the index2/blkno2 fields refer to the "new" entry.
1803 	 */
1804 	if (blk1->index > ichdr1.count) {
1805 		ASSERT(state->inleaf == 0);
1806 		blk2->index = blk1->index - ichdr1.count;
1807 		args->index = args->index2 = blk2->index;
1808 		args->blkno = args->blkno2 = blk2->blkno;
1809 	} else if (blk1->index == ichdr1.count) {
1810 		if (state->inleaf) {
1811 			args->index = blk1->index;
1812 			args->blkno = blk1->blkno;
1813 			args->index2 = 0;
1814 			args->blkno2 = blk2->blkno;
1815 		} else {
1816 			/*
1817 			 * On a double leaf split, the original attr location
1818 			 * is already stored in blkno2/index2, so don't
1819 			 * overwrite it overwise we corrupt the tree.
1820 			 */
1821 			blk2->index = blk1->index - ichdr1.count;
1822 			args->index = blk2->index;
1823 			args->blkno = blk2->blkno;
1824 			if (!state->extravalid) {
1825 				/*
1826 				 * set the new attr location to match the old
1827 				 * one and let the higher level split code
1828 				 * decide where in the leaf to place it.
1829 				 */
1830 				args->index2 = blk2->index;
1831 				args->blkno2 = blk2->blkno;
1832 			}
1833 		}
1834 	} else {
1835 		ASSERT(state->inleaf == 1);
1836 		args->index = args->index2 = blk1->index;
1837 		args->blkno = args->blkno2 = blk1->blkno;
1838 	}
1839 }
1840 
1841 /*
1842  * Examine entries until we reduce the absolute difference in
1843  * byte usage between the two blocks to a minimum.
1844  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1845  * GROT: there will always be enough room in either block for a new entry.
1846  * GROT: Do a double-split for this case?
1847  */
1848 STATIC int
1849 xfs_attr3_leaf_figure_balance(
1850 	struct xfs_da_state		*state,
1851 	struct xfs_da_state_blk		*blk1,
1852 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1853 	struct xfs_da_state_blk		*blk2,
1854 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1855 	int				*countarg,
1856 	int				*usedbytesarg)
1857 {
1858 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1859 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1860 	struct xfs_attr_leaf_entry	*entry;
1861 	int				count;
1862 	int				max;
1863 	int				index;
1864 	int				totallen = 0;
1865 	int				half;
1866 	int				lastdelta;
1867 	int				foundit = 0;
1868 	int				tmp;
1869 
1870 	/*
1871 	 * Examine entries until we reduce the absolute difference in
1872 	 * byte usage between the two blocks to a minimum.
1873 	 */
1874 	max = ichdr1->count + ichdr2->count;
1875 	half = (max + 1) * sizeof(*entry);
1876 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1877 			xfs_attr_leaf_newentsize(state->args, NULL);
1878 	half /= 2;
1879 	lastdelta = state->args->geo->blksize;
1880 	entry = xfs_attr3_leaf_entryp(leaf1);
1881 	for (count = index = 0; count < max; entry++, index++, count++) {
1882 
1883 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1884 		/*
1885 		 * The new entry is in the first block, account for it.
1886 		 */
1887 		if (count == blk1->index) {
1888 			tmp = totallen + sizeof(*entry) +
1889 				xfs_attr_leaf_newentsize(state->args, NULL);
1890 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1891 				break;
1892 			lastdelta = XFS_ATTR_ABS(half - tmp);
1893 			totallen = tmp;
1894 			foundit = 1;
1895 		}
1896 
1897 		/*
1898 		 * Wrap around into the second block if necessary.
1899 		 */
1900 		if (count == ichdr1->count) {
1901 			leaf1 = leaf2;
1902 			entry = xfs_attr3_leaf_entryp(leaf1);
1903 			index = 0;
1904 		}
1905 
1906 		/*
1907 		 * Figure out if next leaf entry would be too much.
1908 		 */
1909 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1910 									index);
1911 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1912 			break;
1913 		lastdelta = XFS_ATTR_ABS(half - tmp);
1914 		totallen = tmp;
1915 #undef XFS_ATTR_ABS
1916 	}
1917 
1918 	/*
1919 	 * Calculate the number of usedbytes that will end up in lower block.
1920 	 * If new entry not in lower block, fix up the count.
1921 	 */
1922 	totallen -= count * sizeof(*entry);
1923 	if (foundit) {
1924 		totallen -= sizeof(*entry) +
1925 				xfs_attr_leaf_newentsize(state->args, NULL);
1926 	}
1927 
1928 	*countarg = count;
1929 	*usedbytesarg = totallen;
1930 	return foundit;
1931 }
1932 
1933 /*========================================================================
1934  * Routines used for shrinking the Btree.
1935  *========================================================================*/
1936 
1937 /*
1938  * Check a leaf block and its neighbors to see if the block should be
1939  * collapsed into one or the other neighbor.  Always keep the block
1940  * with the smaller block number.
1941  * If the current block is over 50% full, don't try to join it, return 0.
1942  * If the block is empty, fill in the state structure and return 2.
1943  * If it can be collapsed, fill in the state structure and return 1.
1944  * If nothing can be done, return 0.
1945  *
1946  * GROT: allow for INCOMPLETE entries in calculation.
1947  */
1948 int
1949 xfs_attr3_leaf_toosmall(
1950 	struct xfs_da_state	*state,
1951 	int			*action)
1952 {
1953 	struct xfs_attr_leafblock *leaf;
1954 	struct xfs_da_state_blk	*blk;
1955 	struct xfs_attr3_icleaf_hdr ichdr;
1956 	struct xfs_buf		*bp;
1957 	xfs_dablk_t		blkno;
1958 	int			bytes;
1959 	int			forward;
1960 	int			error;
1961 	int			retval;
1962 	int			i;
1963 
1964 	trace_xfs_attr_leaf_toosmall(state->args);
1965 
1966 	/*
1967 	 * Check for the degenerate case of the block being over 50% full.
1968 	 * If so, it's not worth even looking to see if we might be able
1969 	 * to coalesce with a sibling.
1970 	 */
1971 	blk = &state->path.blk[ state->path.active-1 ];
1972 	leaf = blk->bp->b_addr;
1973 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1974 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
1975 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1976 		ichdr.usedbytes;
1977 	if (bytes > (state->args->geo->blksize >> 1)) {
1978 		*action = 0;	/* blk over 50%, don't try to join */
1979 		return 0;
1980 	}
1981 
1982 	/*
1983 	 * Check for the degenerate case of the block being empty.
1984 	 * If the block is empty, we'll simply delete it, no need to
1985 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1986 	 * to merge with the forward block unless it is NULL.
1987 	 */
1988 	if (ichdr.count == 0) {
1989 		/*
1990 		 * Make altpath point to the block we want to keep and
1991 		 * path point to the block we want to drop (this one).
1992 		 */
1993 		forward = (ichdr.forw != 0);
1994 		memcpy(&state->altpath, &state->path, sizeof(state->path));
1995 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1996 						 0, &retval);
1997 		if (error)
1998 			return error;
1999 		if (retval) {
2000 			*action = 0;
2001 		} else {
2002 			*action = 2;
2003 		}
2004 		return 0;
2005 	}
2006 
2007 	/*
2008 	 * Examine each sibling block to see if we can coalesce with
2009 	 * at least 25% free space to spare.  We need to figure out
2010 	 * whether to merge with the forward or the backward block.
2011 	 * We prefer coalescing with the lower numbered sibling so as
2012 	 * to shrink an attribute list over time.
2013 	 */
2014 	/* start with smaller blk num */
2015 	forward = ichdr.forw < ichdr.back;
2016 	for (i = 0; i < 2; forward = !forward, i++) {
2017 		struct xfs_attr3_icleaf_hdr ichdr2;
2018 		if (forward)
2019 			blkno = ichdr.forw;
2020 		else
2021 			blkno = ichdr.back;
2022 		if (blkno == 0)
2023 			continue;
2024 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2025 					blkno, &bp);
2026 		if (error)
2027 			return error;
2028 
2029 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2030 
2031 		bytes = state->args->geo->blksize -
2032 			(state->args->geo->blksize >> 2) -
2033 			ichdr.usedbytes - ichdr2.usedbytes -
2034 			((ichdr.count + ichdr2.count) *
2035 					sizeof(xfs_attr_leaf_entry_t)) -
2036 			xfs_attr3_leaf_hdr_size(leaf);
2037 
2038 		xfs_trans_brelse(state->args->trans, bp);
2039 		if (bytes >= 0)
2040 			break;	/* fits with at least 25% to spare */
2041 	}
2042 	if (i >= 2) {
2043 		*action = 0;
2044 		return 0;
2045 	}
2046 
2047 	/*
2048 	 * Make altpath point to the block we want to keep (the lower
2049 	 * numbered block) and path point to the block we want to drop.
2050 	 */
2051 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2052 	if (blkno < blk->blkno) {
2053 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2054 						 0, &retval);
2055 	} else {
2056 		error = xfs_da3_path_shift(state, &state->path, forward,
2057 						 0, &retval);
2058 	}
2059 	if (error)
2060 		return error;
2061 	if (retval) {
2062 		*action = 0;
2063 	} else {
2064 		*action = 1;
2065 	}
2066 	return 0;
2067 }
2068 
2069 /*
2070  * Remove a name from the leaf attribute list structure.
2071  *
2072  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2073  * If two leaves are 37% full, when combined they will leave 25% free.
2074  */
2075 int
2076 xfs_attr3_leaf_remove(
2077 	struct xfs_buf		*bp,
2078 	struct xfs_da_args	*args)
2079 {
2080 	struct xfs_attr_leafblock *leaf;
2081 	struct xfs_attr3_icleaf_hdr ichdr;
2082 	struct xfs_attr_leaf_entry *entry;
2083 	int			before;
2084 	int			after;
2085 	int			smallest;
2086 	int			entsize;
2087 	int			tablesize;
2088 	int			tmp;
2089 	int			i;
2090 
2091 	trace_xfs_attr_leaf_remove(args);
2092 
2093 	leaf = bp->b_addr;
2094 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2095 
2096 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2097 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2098 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2099 					xfs_attr3_leaf_hdr_size(leaf));
2100 
2101 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2102 
2103 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2104 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2105 
2106 	/*
2107 	 * Scan through free region table:
2108 	 *    check for adjacency of free'd entry with an existing one,
2109 	 *    find smallest free region in case we need to replace it,
2110 	 *    adjust any map that borders the entry table,
2111 	 */
2112 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2113 					+ xfs_attr3_leaf_hdr_size(leaf);
2114 	tmp = ichdr.freemap[0].size;
2115 	before = after = -1;
2116 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2117 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2118 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2119 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2120 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2121 		if (ichdr.freemap[i].base == tablesize) {
2122 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2123 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2124 		}
2125 
2126 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2127 				be16_to_cpu(entry->nameidx)) {
2128 			before = i;
2129 		} else if (ichdr.freemap[i].base ==
2130 				(be16_to_cpu(entry->nameidx) + entsize)) {
2131 			after = i;
2132 		} else if (ichdr.freemap[i].size < tmp) {
2133 			tmp = ichdr.freemap[i].size;
2134 			smallest = i;
2135 		}
2136 	}
2137 
2138 	/*
2139 	 * Coalesce adjacent freemap regions,
2140 	 * or replace the smallest region.
2141 	 */
2142 	if ((before >= 0) || (after >= 0)) {
2143 		if ((before >= 0) && (after >= 0)) {
2144 			ichdr.freemap[before].size += entsize;
2145 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2146 			ichdr.freemap[after].base = 0;
2147 			ichdr.freemap[after].size = 0;
2148 		} else if (before >= 0) {
2149 			ichdr.freemap[before].size += entsize;
2150 		} else {
2151 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2152 			ichdr.freemap[after].size += entsize;
2153 		}
2154 	} else {
2155 		/*
2156 		 * Replace smallest region (if it is smaller than free'd entry)
2157 		 */
2158 		if (ichdr.freemap[smallest].size < entsize) {
2159 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2160 			ichdr.freemap[smallest].size = entsize;
2161 		}
2162 	}
2163 
2164 	/*
2165 	 * Did we remove the first entry?
2166 	 */
2167 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2168 		smallest = 1;
2169 	else
2170 		smallest = 0;
2171 
2172 	/*
2173 	 * Compress the remaining entries and zero out the removed stuff.
2174 	 */
2175 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2176 	ichdr.usedbytes -= entsize;
2177 	xfs_trans_log_buf(args->trans, bp,
2178 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2179 				   entsize));
2180 
2181 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2182 	memmove(entry, entry + 1, tmp);
2183 	ichdr.count--;
2184 	xfs_trans_log_buf(args->trans, bp,
2185 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2186 
2187 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2188 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2189 
2190 	/*
2191 	 * If we removed the first entry, re-find the first used byte
2192 	 * in the name area.  Note that if the entry was the "firstused",
2193 	 * then we don't have a "hole" in our block resulting from
2194 	 * removing the name.
2195 	 */
2196 	if (smallest) {
2197 		tmp = args->geo->blksize;
2198 		entry = xfs_attr3_leaf_entryp(leaf);
2199 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2200 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2201 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2202 
2203 			if (be16_to_cpu(entry->nameidx) < tmp)
2204 				tmp = be16_to_cpu(entry->nameidx);
2205 		}
2206 		ichdr.firstused = tmp;
2207 		ASSERT(ichdr.firstused != 0);
2208 	} else {
2209 		ichdr.holes = 1;	/* mark as needing compaction */
2210 	}
2211 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2212 	xfs_trans_log_buf(args->trans, bp,
2213 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2214 					  xfs_attr3_leaf_hdr_size(leaf)));
2215 
2216 	/*
2217 	 * Check if leaf is less than 50% full, caller may want to
2218 	 * "join" the leaf with a sibling if so.
2219 	 */
2220 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2221 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2222 
2223 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2224 }
2225 
2226 /*
2227  * Move all the attribute list entries from drop_leaf into save_leaf.
2228  */
2229 void
2230 xfs_attr3_leaf_unbalance(
2231 	struct xfs_da_state	*state,
2232 	struct xfs_da_state_blk	*drop_blk,
2233 	struct xfs_da_state_blk	*save_blk)
2234 {
2235 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2236 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2237 	struct xfs_attr3_icleaf_hdr drophdr;
2238 	struct xfs_attr3_icleaf_hdr savehdr;
2239 	struct xfs_attr_leaf_entry *entry;
2240 
2241 	trace_xfs_attr_leaf_unbalance(state->args);
2242 
2243 	drop_leaf = drop_blk->bp->b_addr;
2244 	save_leaf = save_blk->bp->b_addr;
2245 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2246 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2247 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2248 
2249 	/*
2250 	 * Save last hashval from dying block for later Btree fixup.
2251 	 */
2252 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2253 
2254 	/*
2255 	 * Check if we need a temp buffer, or can we do it in place.
2256 	 * Note that we don't check "leaf" for holes because we will
2257 	 * always be dropping it, toosmall() decided that for us already.
2258 	 */
2259 	if (savehdr.holes == 0) {
2260 		/*
2261 		 * dest leaf has no holes, so we add there.  May need
2262 		 * to make some room in the entry array.
2263 		 */
2264 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2265 					 drop_blk->bp, &drophdr)) {
2266 			xfs_attr3_leaf_moveents(state->args,
2267 						drop_leaf, &drophdr, 0,
2268 						save_leaf, &savehdr, 0,
2269 						drophdr.count);
2270 		} else {
2271 			xfs_attr3_leaf_moveents(state->args,
2272 						drop_leaf, &drophdr, 0,
2273 						save_leaf, &savehdr,
2274 						savehdr.count, drophdr.count);
2275 		}
2276 	} else {
2277 		/*
2278 		 * Destination has holes, so we make a temporary copy
2279 		 * of the leaf and add them both to that.
2280 		 */
2281 		struct xfs_attr_leafblock *tmp_leaf;
2282 		struct xfs_attr3_icleaf_hdr tmphdr;
2283 
2284 		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2285 
2286 		/*
2287 		 * Copy the header into the temp leaf so that all the stuff
2288 		 * not in the incore header is present and gets copied back in
2289 		 * once we've moved all the entries.
2290 		 */
2291 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2292 
2293 		memset(&tmphdr, 0, sizeof(tmphdr));
2294 		tmphdr.magic = savehdr.magic;
2295 		tmphdr.forw = savehdr.forw;
2296 		tmphdr.back = savehdr.back;
2297 		tmphdr.firstused = state->args->geo->blksize;
2298 
2299 		/* write the header to the temp buffer to initialise it */
2300 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2301 
2302 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2303 					 drop_blk->bp, &drophdr)) {
2304 			xfs_attr3_leaf_moveents(state->args,
2305 						drop_leaf, &drophdr, 0,
2306 						tmp_leaf, &tmphdr, 0,
2307 						drophdr.count);
2308 			xfs_attr3_leaf_moveents(state->args,
2309 						save_leaf, &savehdr, 0,
2310 						tmp_leaf, &tmphdr, tmphdr.count,
2311 						savehdr.count);
2312 		} else {
2313 			xfs_attr3_leaf_moveents(state->args,
2314 						save_leaf, &savehdr, 0,
2315 						tmp_leaf, &tmphdr, 0,
2316 						savehdr.count);
2317 			xfs_attr3_leaf_moveents(state->args,
2318 						drop_leaf, &drophdr, 0,
2319 						tmp_leaf, &tmphdr, tmphdr.count,
2320 						drophdr.count);
2321 		}
2322 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2323 		savehdr = tmphdr; /* struct copy */
2324 		kmem_free(tmp_leaf);
2325 	}
2326 
2327 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2328 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2329 					   state->args->geo->blksize - 1);
2330 
2331 	/*
2332 	 * Copy out last hashval in each block for B-tree code.
2333 	 */
2334 	entry = xfs_attr3_leaf_entryp(save_leaf);
2335 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2336 }
2337 
2338 /*========================================================================
2339  * Routines used for finding things in the Btree.
2340  *========================================================================*/
2341 
2342 /*
2343  * Look up a name in a leaf attribute list structure.
2344  * This is the internal routine, it uses the caller's buffer.
2345  *
2346  * Note that duplicate keys are allowed, but only check within the
2347  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2348  *
2349  * Return in args->index the index into the entry[] array of either
2350  * the found entry, or where the entry should have been (insert before
2351  * that entry).
2352  *
2353  * Don't change the args->value unless we find the attribute.
2354  */
2355 int
2356 xfs_attr3_leaf_lookup_int(
2357 	struct xfs_buf		*bp,
2358 	struct xfs_da_args	*args)
2359 {
2360 	struct xfs_attr_leafblock *leaf;
2361 	struct xfs_attr3_icleaf_hdr ichdr;
2362 	struct xfs_attr_leaf_entry *entry;
2363 	struct xfs_attr_leaf_entry *entries;
2364 	struct xfs_attr_leaf_name_local *name_loc;
2365 	struct xfs_attr_leaf_name_remote *name_rmt;
2366 	xfs_dahash_t		hashval;
2367 	int			probe;
2368 	int			span;
2369 
2370 	trace_xfs_attr_leaf_lookup(args);
2371 
2372 	leaf = bp->b_addr;
2373 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2374 	entries = xfs_attr3_leaf_entryp(leaf);
2375 	if (ichdr.count >= args->geo->blksize / 8) {
2376 		xfs_buf_mark_corrupt(bp);
2377 		return -EFSCORRUPTED;
2378 	}
2379 
2380 	/*
2381 	 * Binary search.  (note: small blocks will skip this loop)
2382 	 */
2383 	hashval = args->hashval;
2384 	probe = span = ichdr.count / 2;
2385 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2386 		span /= 2;
2387 		if (be32_to_cpu(entry->hashval) < hashval)
2388 			probe += span;
2389 		else if (be32_to_cpu(entry->hashval) > hashval)
2390 			probe -= span;
2391 		else
2392 			break;
2393 	}
2394 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2395 		xfs_buf_mark_corrupt(bp);
2396 		return -EFSCORRUPTED;
2397 	}
2398 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2399 		xfs_buf_mark_corrupt(bp);
2400 		return -EFSCORRUPTED;
2401 	}
2402 
2403 	/*
2404 	 * Since we may have duplicate hashval's, find the first matching
2405 	 * hashval in the leaf.
2406 	 */
2407 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2408 		entry--;
2409 		probe--;
2410 	}
2411 	while (probe < ichdr.count &&
2412 	       be32_to_cpu(entry->hashval) < hashval) {
2413 		entry++;
2414 		probe++;
2415 	}
2416 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2417 		args->index = probe;
2418 		return -ENOATTR;
2419 	}
2420 
2421 	/*
2422 	 * Duplicate keys may be present, so search all of them for a match.
2423 	 */
2424 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2425 			entry++, probe++) {
2426 /*
2427  * GROT: Add code to remove incomplete entries.
2428  */
2429 		if (entry->flags & XFS_ATTR_LOCAL) {
2430 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2431 			if (!xfs_attr_match(args, name_loc->namelen,
2432 					name_loc->nameval, entry->flags))
2433 				continue;
2434 			args->index = probe;
2435 			return -EEXIST;
2436 		} else {
2437 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2438 			if (!xfs_attr_match(args, name_rmt->namelen,
2439 					name_rmt->name, entry->flags))
2440 				continue;
2441 			args->index = probe;
2442 			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2443 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2444 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2445 							args->dp->i_mount,
2446 							args->rmtvaluelen);
2447 			return -EEXIST;
2448 		}
2449 	}
2450 	args->index = probe;
2451 	return -ENOATTR;
2452 }
2453 
2454 /*
2455  * Get the value associated with an attribute name from a leaf attribute
2456  * list structure.
2457  *
2458  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2459  * lookup, we only return an error if the attribute does not exist or we can't
2460  * retrieve the value.
2461  */
2462 int
2463 xfs_attr3_leaf_getvalue(
2464 	struct xfs_buf		*bp,
2465 	struct xfs_da_args	*args)
2466 {
2467 	struct xfs_attr_leafblock *leaf;
2468 	struct xfs_attr3_icleaf_hdr ichdr;
2469 	struct xfs_attr_leaf_entry *entry;
2470 	struct xfs_attr_leaf_name_local *name_loc;
2471 	struct xfs_attr_leaf_name_remote *name_rmt;
2472 
2473 	leaf = bp->b_addr;
2474 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2475 	ASSERT(ichdr.count < args->geo->blksize / 8);
2476 	ASSERT(args->index < ichdr.count);
2477 
2478 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2479 	if (entry->flags & XFS_ATTR_LOCAL) {
2480 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2481 		ASSERT(name_loc->namelen == args->namelen);
2482 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2483 		return xfs_attr_copy_value(args,
2484 					&name_loc->nameval[args->namelen],
2485 					be16_to_cpu(name_loc->valuelen));
2486 	}
2487 
2488 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2489 	ASSERT(name_rmt->namelen == args->namelen);
2490 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2491 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2492 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2493 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2494 					       args->rmtvaluelen);
2495 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2496 }
2497 
2498 /*========================================================================
2499  * Utility routines.
2500  *========================================================================*/
2501 
2502 /*
2503  * Move the indicated entries from one leaf to another.
2504  * NOTE: this routine modifies both source and destination leaves.
2505  */
2506 /*ARGSUSED*/
2507 STATIC void
2508 xfs_attr3_leaf_moveents(
2509 	struct xfs_da_args		*args,
2510 	struct xfs_attr_leafblock	*leaf_s,
2511 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2512 	int				start_s,
2513 	struct xfs_attr_leafblock	*leaf_d,
2514 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2515 	int				start_d,
2516 	int				count)
2517 {
2518 	struct xfs_attr_leaf_entry	*entry_s;
2519 	struct xfs_attr_leaf_entry	*entry_d;
2520 	int				desti;
2521 	int				tmp;
2522 	int				i;
2523 
2524 	/*
2525 	 * Check for nothing to do.
2526 	 */
2527 	if (count == 0)
2528 		return;
2529 
2530 	/*
2531 	 * Set up environment.
2532 	 */
2533 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2534 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2535 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2536 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2537 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2538 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2539 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2540 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2541 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2542 
2543 	ASSERT(start_s < ichdr_s->count);
2544 	ASSERT(start_d <= ichdr_d->count);
2545 	ASSERT(count <= ichdr_s->count);
2546 
2547 
2548 	/*
2549 	 * Move the entries in the destination leaf up to make a hole?
2550 	 */
2551 	if (start_d < ichdr_d->count) {
2552 		tmp  = ichdr_d->count - start_d;
2553 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2554 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2555 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2556 		memmove(entry_d, entry_s, tmp);
2557 	}
2558 
2559 	/*
2560 	 * Copy all entry's in the same (sorted) order,
2561 	 * but allocate attribute info packed and in sequence.
2562 	 */
2563 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2564 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2565 	desti = start_d;
2566 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2567 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2568 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2569 #ifdef GROT
2570 		/*
2571 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2572 		 * may also need to change the insertion index.  Code turned
2573 		 * off for 6.2, should be revisited later.
2574 		 */
2575 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2576 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2577 			ichdr_s->usedbytes -= tmp;
2578 			ichdr_s->count -= 1;
2579 			entry_d--;	/* to compensate for ++ in loop hdr */
2580 			desti--;
2581 			if ((start_s + i) < offset)
2582 				result++;	/* insertion index adjustment */
2583 		} else {
2584 #endif /* GROT */
2585 			ichdr_d->firstused -= tmp;
2586 			/* both on-disk, don't endian flip twice */
2587 			entry_d->hashval = entry_s->hashval;
2588 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2589 			entry_d->flags = entry_s->flags;
2590 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2591 							<= args->geo->blksize);
2592 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2593 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2594 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2595 							<= args->geo->blksize);
2596 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2597 			ichdr_s->usedbytes -= tmp;
2598 			ichdr_d->usedbytes += tmp;
2599 			ichdr_s->count -= 1;
2600 			ichdr_d->count += 1;
2601 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2602 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2603 			ASSERT(ichdr_d->firstused >= tmp);
2604 #ifdef GROT
2605 		}
2606 #endif /* GROT */
2607 	}
2608 
2609 	/*
2610 	 * Zero out the entries we just copied.
2611 	 */
2612 	if (start_s == ichdr_s->count) {
2613 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2614 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2615 		ASSERT(((char *)entry_s + tmp) <=
2616 		       ((char *)leaf_s + args->geo->blksize));
2617 		memset(entry_s, 0, tmp);
2618 	} else {
2619 		/*
2620 		 * Move the remaining entries down to fill the hole,
2621 		 * then zero the entries at the top.
2622 		 */
2623 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2624 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2625 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2626 		memmove(entry_d, entry_s, tmp);
2627 
2628 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2629 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2630 		ASSERT(((char *)entry_s + tmp) <=
2631 		       ((char *)leaf_s + args->geo->blksize));
2632 		memset(entry_s, 0, tmp);
2633 	}
2634 
2635 	/*
2636 	 * Fill in the freemap information
2637 	 */
2638 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2639 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2640 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2641 	ichdr_d->freemap[1].base = 0;
2642 	ichdr_d->freemap[2].base = 0;
2643 	ichdr_d->freemap[1].size = 0;
2644 	ichdr_d->freemap[2].size = 0;
2645 	ichdr_s->holes = 1;	/* leaf may not be compact */
2646 }
2647 
2648 /*
2649  * Pick up the last hashvalue from a leaf block.
2650  */
2651 xfs_dahash_t
2652 xfs_attr_leaf_lasthash(
2653 	struct xfs_buf	*bp,
2654 	int		*count)
2655 {
2656 	struct xfs_attr3_icleaf_hdr ichdr;
2657 	struct xfs_attr_leaf_entry *entries;
2658 	struct xfs_mount *mp = bp->b_mount;
2659 
2660 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2661 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2662 	if (count)
2663 		*count = ichdr.count;
2664 	if (!ichdr.count)
2665 		return 0;
2666 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2667 }
2668 
2669 /*
2670  * Calculate the number of bytes used to store the indicated attribute
2671  * (whether local or remote only calculate bytes in this block).
2672  */
2673 STATIC int
2674 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2675 {
2676 	struct xfs_attr_leaf_entry *entries;
2677 	xfs_attr_leaf_name_local_t *name_loc;
2678 	xfs_attr_leaf_name_remote_t *name_rmt;
2679 	int size;
2680 
2681 	entries = xfs_attr3_leaf_entryp(leaf);
2682 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2683 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2684 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2685 						   be16_to_cpu(name_loc->valuelen));
2686 	} else {
2687 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2688 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2689 	}
2690 	return size;
2691 }
2692 
2693 /*
2694  * Calculate the number of bytes that would be required to store the new
2695  * attribute (whether local or remote only calculate bytes in this block).
2696  * This routine decides as a side effect whether the attribute will be
2697  * a "local" or a "remote" attribute.
2698  */
2699 int
2700 xfs_attr_leaf_newentsize(
2701 	struct xfs_da_args	*args,
2702 	int			*local)
2703 {
2704 	int			size;
2705 
2706 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2707 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2708 		if (local)
2709 			*local = 1;
2710 		return size;
2711 	}
2712 	if (local)
2713 		*local = 0;
2714 	return xfs_attr_leaf_entsize_remote(args->namelen);
2715 }
2716 
2717 
2718 /*========================================================================
2719  * Manage the INCOMPLETE flag in a leaf entry
2720  *========================================================================*/
2721 
2722 /*
2723  * Clear the INCOMPLETE flag on an entry in a leaf block.
2724  */
2725 int
2726 xfs_attr3_leaf_clearflag(
2727 	struct xfs_da_args	*args)
2728 {
2729 	struct xfs_attr_leafblock *leaf;
2730 	struct xfs_attr_leaf_entry *entry;
2731 	struct xfs_attr_leaf_name_remote *name_rmt;
2732 	struct xfs_buf		*bp;
2733 	int			error;
2734 #ifdef DEBUG
2735 	struct xfs_attr3_icleaf_hdr ichdr;
2736 	xfs_attr_leaf_name_local_t *name_loc;
2737 	int namelen;
2738 	char *name;
2739 #endif /* DEBUG */
2740 
2741 	trace_xfs_attr_leaf_clearflag(args);
2742 	/*
2743 	 * Set up the operation.
2744 	 */
2745 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2746 	if (error)
2747 		return error;
2748 
2749 	leaf = bp->b_addr;
2750 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2751 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2752 
2753 #ifdef DEBUG
2754 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2755 	ASSERT(args->index < ichdr.count);
2756 	ASSERT(args->index >= 0);
2757 
2758 	if (entry->flags & XFS_ATTR_LOCAL) {
2759 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2760 		namelen = name_loc->namelen;
2761 		name = (char *)name_loc->nameval;
2762 	} else {
2763 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2764 		namelen = name_rmt->namelen;
2765 		name = (char *)name_rmt->name;
2766 	}
2767 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2768 	ASSERT(namelen == args->namelen);
2769 	ASSERT(memcmp(name, args->name, namelen) == 0);
2770 #endif /* DEBUG */
2771 
2772 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2773 	xfs_trans_log_buf(args->trans, bp,
2774 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2775 
2776 	if (args->rmtblkno) {
2777 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2778 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2779 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2780 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2781 		xfs_trans_log_buf(args->trans, bp,
2782 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2783 	}
2784 
2785 	return 0;
2786 }
2787 
2788 /*
2789  * Set the INCOMPLETE flag on an entry in a leaf block.
2790  */
2791 int
2792 xfs_attr3_leaf_setflag(
2793 	struct xfs_da_args	*args)
2794 {
2795 	struct xfs_attr_leafblock *leaf;
2796 	struct xfs_attr_leaf_entry *entry;
2797 	struct xfs_attr_leaf_name_remote *name_rmt;
2798 	struct xfs_buf		*bp;
2799 	int error;
2800 #ifdef DEBUG
2801 	struct xfs_attr3_icleaf_hdr ichdr;
2802 #endif
2803 
2804 	trace_xfs_attr_leaf_setflag(args);
2805 
2806 	/*
2807 	 * Set up the operation.
2808 	 */
2809 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2810 	if (error)
2811 		return error;
2812 
2813 	leaf = bp->b_addr;
2814 #ifdef DEBUG
2815 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2816 	ASSERT(args->index < ichdr.count);
2817 	ASSERT(args->index >= 0);
2818 #endif
2819 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2820 
2821 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2822 	entry->flags |= XFS_ATTR_INCOMPLETE;
2823 	xfs_trans_log_buf(args->trans, bp,
2824 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2825 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2826 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2827 		name_rmt->valueblk = 0;
2828 		name_rmt->valuelen = 0;
2829 		xfs_trans_log_buf(args->trans, bp,
2830 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2831 	}
2832 
2833 	return 0;
2834 }
2835 
2836 /*
2837  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2838  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2839  * entry given by args->blkno2/index2.
2840  *
2841  * Note that they could be in different blocks, or in the same block.
2842  */
2843 int
2844 xfs_attr3_leaf_flipflags(
2845 	struct xfs_da_args	*args)
2846 {
2847 	struct xfs_attr_leafblock *leaf1;
2848 	struct xfs_attr_leafblock *leaf2;
2849 	struct xfs_attr_leaf_entry *entry1;
2850 	struct xfs_attr_leaf_entry *entry2;
2851 	struct xfs_attr_leaf_name_remote *name_rmt;
2852 	struct xfs_buf		*bp1;
2853 	struct xfs_buf		*bp2;
2854 	int error;
2855 #ifdef DEBUG
2856 	struct xfs_attr3_icleaf_hdr ichdr1;
2857 	struct xfs_attr3_icleaf_hdr ichdr2;
2858 	xfs_attr_leaf_name_local_t *name_loc;
2859 	int namelen1, namelen2;
2860 	char *name1, *name2;
2861 #endif /* DEBUG */
2862 
2863 	trace_xfs_attr_leaf_flipflags(args);
2864 
2865 	/*
2866 	 * Read the block containing the "old" attr
2867 	 */
2868 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2869 	if (error)
2870 		return error;
2871 
2872 	/*
2873 	 * Read the block containing the "new" attr, if it is different
2874 	 */
2875 	if (args->blkno2 != args->blkno) {
2876 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2877 					   &bp2);
2878 		if (error)
2879 			return error;
2880 	} else {
2881 		bp2 = bp1;
2882 	}
2883 
2884 	leaf1 = bp1->b_addr;
2885 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2886 
2887 	leaf2 = bp2->b_addr;
2888 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2889 
2890 #ifdef DEBUG
2891 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2892 	ASSERT(args->index < ichdr1.count);
2893 	ASSERT(args->index >= 0);
2894 
2895 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2896 	ASSERT(args->index2 < ichdr2.count);
2897 	ASSERT(args->index2 >= 0);
2898 
2899 	if (entry1->flags & XFS_ATTR_LOCAL) {
2900 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2901 		namelen1 = name_loc->namelen;
2902 		name1 = (char *)name_loc->nameval;
2903 	} else {
2904 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2905 		namelen1 = name_rmt->namelen;
2906 		name1 = (char *)name_rmt->name;
2907 	}
2908 	if (entry2->flags & XFS_ATTR_LOCAL) {
2909 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2910 		namelen2 = name_loc->namelen;
2911 		name2 = (char *)name_loc->nameval;
2912 	} else {
2913 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2914 		namelen2 = name_rmt->namelen;
2915 		name2 = (char *)name_rmt->name;
2916 	}
2917 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2918 	ASSERT(namelen1 == namelen2);
2919 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2920 #endif /* DEBUG */
2921 
2922 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2923 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2924 
2925 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2926 	xfs_trans_log_buf(args->trans, bp1,
2927 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2928 	if (args->rmtblkno) {
2929 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2930 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2931 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2932 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2933 		xfs_trans_log_buf(args->trans, bp1,
2934 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2935 	}
2936 
2937 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2938 	xfs_trans_log_buf(args->trans, bp2,
2939 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2940 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2941 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2942 		name_rmt->valueblk = 0;
2943 		name_rmt->valuelen = 0;
2944 		xfs_trans_log_buf(args->trans, bp2,
2945 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2946 	}
2947 
2948 	return 0;
2949 }
2950