xref: /openbmc/linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision 76a4f7cc)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_da_format.h"
16 #include "xfs_da_btree.h"
17 #include "xfs_inode.h"
18 #include "xfs_trans.h"
19 #include "xfs_bmap_btree.h"
20 #include "xfs_bmap.h"
21 #include "xfs_attr_sf.h"
22 #include "xfs_attr.h"
23 #include "xfs_attr_remote.h"
24 #include "xfs_attr_leaf.h"
25 #include "xfs_error.h"
26 #include "xfs_trace.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_dir2.h"
29 #include "xfs_log.h"
30 #include "xfs_ag.h"
31 
32 
33 /*
34  * xfs_attr_leaf.c
35  *
36  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
37  */
38 
39 /*========================================================================
40  * Function prototypes for the kernel.
41  *========================================================================*/
42 
43 /*
44  * Routines used for growing the Btree.
45  */
46 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
47 				 xfs_dablk_t which_block, struct xfs_buf **bpp);
48 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
49 				   struct xfs_attr3_icleaf_hdr *ichdr,
50 				   struct xfs_da_args *args, int freemap_index);
51 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
52 				   struct xfs_attr3_icleaf_hdr *ichdr,
53 				   struct xfs_buf *leaf_buffer);
54 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
55 						   xfs_da_state_blk_t *blk1,
56 						   xfs_da_state_blk_t *blk2);
57 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
58 			xfs_da_state_blk_t *leaf_blk_1,
59 			struct xfs_attr3_icleaf_hdr *ichdr1,
60 			xfs_da_state_blk_t *leaf_blk_2,
61 			struct xfs_attr3_icleaf_hdr *ichdr2,
62 			int *number_entries_in_blk1,
63 			int *number_usedbytes_in_blk1);
64 
65 /*
66  * Utility routines.
67  */
68 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
69 			struct xfs_attr_leafblock *src_leaf,
70 			struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
71 			struct xfs_attr_leafblock *dst_leaf,
72 			struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
73 			int move_count);
74 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
75 
76 /*
77  * attr3 block 'firstused' conversion helpers.
78  *
79  * firstused refers to the offset of the first used byte of the nameval region
80  * of an attr leaf block. The region starts at the tail of the block and expands
81  * backwards towards the middle. As such, firstused is initialized to the block
82  * size for an empty leaf block and is reduced from there.
83  *
84  * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
85  * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
86  * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
87  * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
88  * the attr block size. The following helpers manage the conversion between the
89  * in-core and on-disk formats.
90  */
91 
92 static void
93 xfs_attr3_leaf_firstused_from_disk(
94 	struct xfs_da_geometry		*geo,
95 	struct xfs_attr3_icleaf_hdr	*to,
96 	struct xfs_attr_leafblock	*from)
97 {
98 	struct xfs_attr3_leaf_hdr	*hdr3;
99 
100 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
101 		hdr3 = (struct xfs_attr3_leaf_hdr *) from;
102 		to->firstused = be16_to_cpu(hdr3->firstused);
103 	} else {
104 		to->firstused = be16_to_cpu(from->hdr.firstused);
105 	}
106 
107 	/*
108 	 * Convert from the magic fsb size value to actual blocksize. This
109 	 * should only occur for empty blocks when the block size overflows
110 	 * 16-bits.
111 	 */
112 	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
113 		ASSERT(!to->count && !to->usedbytes);
114 		ASSERT(geo->blksize > USHRT_MAX);
115 		to->firstused = geo->blksize;
116 	}
117 }
118 
119 static void
120 xfs_attr3_leaf_firstused_to_disk(
121 	struct xfs_da_geometry		*geo,
122 	struct xfs_attr_leafblock	*to,
123 	struct xfs_attr3_icleaf_hdr	*from)
124 {
125 	struct xfs_attr3_leaf_hdr	*hdr3;
126 	uint32_t			firstused;
127 
128 	/* magic value should only be seen on disk */
129 	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
130 
131 	/*
132 	 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
133 	 * value. This only overflows at the max supported value of 64k. Use the
134 	 * magic on-disk value to represent block size in this case.
135 	 */
136 	firstused = from->firstused;
137 	if (firstused > USHRT_MAX) {
138 		ASSERT(from->firstused == geo->blksize);
139 		firstused = XFS_ATTR3_LEAF_NULLOFF;
140 	}
141 
142 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
143 		hdr3 = (struct xfs_attr3_leaf_hdr *) to;
144 		hdr3->firstused = cpu_to_be16(firstused);
145 	} else {
146 		to->hdr.firstused = cpu_to_be16(firstused);
147 	}
148 }
149 
150 void
151 xfs_attr3_leaf_hdr_from_disk(
152 	struct xfs_da_geometry		*geo,
153 	struct xfs_attr3_icleaf_hdr	*to,
154 	struct xfs_attr_leafblock	*from)
155 {
156 	int	i;
157 
158 	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
159 	       from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
160 
161 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
162 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
163 
164 		to->forw = be32_to_cpu(hdr3->info.hdr.forw);
165 		to->back = be32_to_cpu(hdr3->info.hdr.back);
166 		to->magic = be16_to_cpu(hdr3->info.hdr.magic);
167 		to->count = be16_to_cpu(hdr3->count);
168 		to->usedbytes = be16_to_cpu(hdr3->usedbytes);
169 		xfs_attr3_leaf_firstused_from_disk(geo, to, from);
170 		to->holes = hdr3->holes;
171 
172 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
173 			to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
174 			to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
175 		}
176 		return;
177 	}
178 	to->forw = be32_to_cpu(from->hdr.info.forw);
179 	to->back = be32_to_cpu(from->hdr.info.back);
180 	to->magic = be16_to_cpu(from->hdr.info.magic);
181 	to->count = be16_to_cpu(from->hdr.count);
182 	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
183 	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
184 	to->holes = from->hdr.holes;
185 
186 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
187 		to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
188 		to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
189 	}
190 }
191 
192 void
193 xfs_attr3_leaf_hdr_to_disk(
194 	struct xfs_da_geometry		*geo,
195 	struct xfs_attr_leafblock	*to,
196 	struct xfs_attr3_icleaf_hdr	*from)
197 {
198 	int				i;
199 
200 	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
201 	       from->magic == XFS_ATTR3_LEAF_MAGIC);
202 
203 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
204 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
205 
206 		hdr3->info.hdr.forw = cpu_to_be32(from->forw);
207 		hdr3->info.hdr.back = cpu_to_be32(from->back);
208 		hdr3->info.hdr.magic = cpu_to_be16(from->magic);
209 		hdr3->count = cpu_to_be16(from->count);
210 		hdr3->usedbytes = cpu_to_be16(from->usedbytes);
211 		xfs_attr3_leaf_firstused_to_disk(geo, to, from);
212 		hdr3->holes = from->holes;
213 		hdr3->pad1 = 0;
214 
215 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
216 			hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
217 			hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
218 		}
219 		return;
220 	}
221 	to->hdr.info.forw = cpu_to_be32(from->forw);
222 	to->hdr.info.back = cpu_to_be32(from->back);
223 	to->hdr.info.magic = cpu_to_be16(from->magic);
224 	to->hdr.count = cpu_to_be16(from->count);
225 	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
226 	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
227 	to->hdr.holes = from->holes;
228 	to->hdr.pad1 = 0;
229 
230 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
231 		to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
232 		to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
233 	}
234 }
235 
236 static xfs_failaddr_t
237 xfs_attr3_leaf_verify_entry(
238 	struct xfs_mount			*mp,
239 	char					*buf_end,
240 	struct xfs_attr_leafblock		*leaf,
241 	struct xfs_attr3_icleaf_hdr		*leafhdr,
242 	struct xfs_attr_leaf_entry		*ent,
243 	int					idx,
244 	__u32					*last_hashval)
245 {
246 	struct xfs_attr_leaf_name_local		*lentry;
247 	struct xfs_attr_leaf_name_remote	*rentry;
248 	char					*name_end;
249 	unsigned int				nameidx;
250 	unsigned int				namesize;
251 	__u32					hashval;
252 
253 	/* hash order check */
254 	hashval = be32_to_cpu(ent->hashval);
255 	if (hashval < *last_hashval)
256 		return __this_address;
257 	*last_hashval = hashval;
258 
259 	nameidx = be16_to_cpu(ent->nameidx);
260 	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
261 		return __this_address;
262 
263 	/*
264 	 * Check the name information.  The namelen fields are u8 so we can't
265 	 * possibly exceed the maximum name length of 255 bytes.
266 	 */
267 	if (ent->flags & XFS_ATTR_LOCAL) {
268 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
269 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
270 				be16_to_cpu(lentry->valuelen));
271 		name_end = (char *)lentry + namesize;
272 		if (lentry->namelen == 0)
273 			return __this_address;
274 	} else {
275 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
276 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
277 		name_end = (char *)rentry + namesize;
278 		if (rentry->namelen == 0)
279 			return __this_address;
280 		if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
281 		    rentry->valueblk == 0)
282 			return __this_address;
283 	}
284 
285 	if (name_end > buf_end)
286 		return __this_address;
287 
288 	return NULL;
289 }
290 
291 static xfs_failaddr_t
292 xfs_attr3_leaf_verify(
293 	struct xfs_buf			*bp)
294 {
295 	struct xfs_attr3_icleaf_hdr	ichdr;
296 	struct xfs_mount		*mp = bp->b_mount;
297 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
298 	struct xfs_attr_leaf_entry	*entries;
299 	struct xfs_attr_leaf_entry	*ent;
300 	char				*buf_end;
301 	uint32_t			end;	/* must be 32bit - see below */
302 	__u32				last_hashval = 0;
303 	int				i;
304 	xfs_failaddr_t			fa;
305 
306 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
307 
308 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
309 	if (fa)
310 		return fa;
311 
312 	/*
313 	 * firstused is the block offset of the first name info structure.
314 	 * Make sure it doesn't go off the block or crash into the header.
315 	 */
316 	if (ichdr.firstused > mp->m_attr_geo->blksize)
317 		return __this_address;
318 	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
319 		return __this_address;
320 
321 	/* Make sure the entries array doesn't crash into the name info. */
322 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
323 	if ((char *)&entries[ichdr.count] >
324 	    (char *)bp->b_addr + ichdr.firstused)
325 		return __this_address;
326 
327 	/*
328 	 * NOTE: This verifier historically failed empty leaf buffers because
329 	 * we expect the fork to be in another format. Empty attr fork format
330 	 * conversions are possible during xattr set, however, and format
331 	 * conversion is not atomic with the xattr set that triggers it. We
332 	 * cannot assume leaf blocks are non-empty until that is addressed.
333 	*/
334 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
335 	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
336 		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
337 				ent, i, &last_hashval);
338 		if (fa)
339 			return fa;
340 	}
341 
342 	/*
343 	 * Quickly check the freemap information.  Attribute data has to be
344 	 * aligned to 4-byte boundaries, and likewise for the free space.
345 	 *
346 	 * Note that for 64k block size filesystems, the freemap entries cannot
347 	 * overflow as they are only be16 fields. However, when checking end
348 	 * pointer of the freemap, we have to be careful to detect overflows and
349 	 * so use uint32_t for those checks.
350 	 */
351 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
352 		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
353 			return __this_address;
354 		if (ichdr.freemap[i].base & 0x3)
355 			return __this_address;
356 		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
357 			return __this_address;
358 		if (ichdr.freemap[i].size & 0x3)
359 			return __this_address;
360 
361 		/* be care of 16 bit overflows here */
362 		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
363 		if (end < ichdr.freemap[i].base)
364 			return __this_address;
365 		if (end > mp->m_attr_geo->blksize)
366 			return __this_address;
367 	}
368 
369 	return NULL;
370 }
371 
372 static void
373 xfs_attr3_leaf_write_verify(
374 	struct xfs_buf	*bp)
375 {
376 	struct xfs_mount	*mp = bp->b_mount;
377 	struct xfs_buf_log_item	*bip = bp->b_log_item;
378 	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
379 	xfs_failaddr_t		fa;
380 
381 	fa = xfs_attr3_leaf_verify(bp);
382 	if (fa) {
383 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
384 		return;
385 	}
386 
387 	if (!xfs_has_crc(mp))
388 		return;
389 
390 	if (bip)
391 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
392 
393 	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
394 }
395 
396 /*
397  * leaf/node format detection on trees is sketchy, so a node read can be done on
398  * leaf level blocks when detection identifies the tree as a node format tree
399  * incorrectly. In this case, we need to swap the verifier to match the correct
400  * format of the block being read.
401  */
402 static void
403 xfs_attr3_leaf_read_verify(
404 	struct xfs_buf		*bp)
405 {
406 	struct xfs_mount	*mp = bp->b_mount;
407 	xfs_failaddr_t		fa;
408 
409 	if (xfs_has_crc(mp) &&
410 	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
411 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
412 	else {
413 		fa = xfs_attr3_leaf_verify(bp);
414 		if (fa)
415 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
416 	}
417 }
418 
419 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
420 	.name = "xfs_attr3_leaf",
421 	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
422 		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
423 	.verify_read = xfs_attr3_leaf_read_verify,
424 	.verify_write = xfs_attr3_leaf_write_verify,
425 	.verify_struct = xfs_attr3_leaf_verify,
426 };
427 
428 int
429 xfs_attr3_leaf_read(
430 	struct xfs_trans	*tp,
431 	struct xfs_inode	*dp,
432 	xfs_dablk_t		bno,
433 	struct xfs_buf		**bpp)
434 {
435 	int			err;
436 
437 	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
438 			&xfs_attr3_leaf_buf_ops);
439 	if (!err && tp && *bpp)
440 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
441 	return err;
442 }
443 
444 /*========================================================================
445  * Namespace helper routines
446  *========================================================================*/
447 
448 static bool
449 xfs_attr_match(
450 	struct xfs_da_args	*args,
451 	uint8_t			namelen,
452 	unsigned char		*name,
453 	int			flags)
454 {
455 	if (args->namelen != namelen)
456 		return false;
457 	if (memcmp(args->name, name, namelen) != 0)
458 		return false;
459 	/*
460 	 * If we are looking for incomplete entries, show only those, else only
461 	 * show complete entries.
462 	 */
463 	if (args->attr_filter !=
464 	    (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
465 		return false;
466 	return true;
467 }
468 
469 static int
470 xfs_attr_copy_value(
471 	struct xfs_da_args	*args,
472 	unsigned char		*value,
473 	int			valuelen)
474 {
475 	/*
476 	 * No copy if all we have to do is get the length
477 	 */
478 	if (!args->valuelen) {
479 		args->valuelen = valuelen;
480 		return 0;
481 	}
482 
483 	/*
484 	 * No copy if the length of the existing buffer is too small
485 	 */
486 	if (args->valuelen < valuelen) {
487 		args->valuelen = valuelen;
488 		return -ERANGE;
489 	}
490 
491 	if (!args->value) {
492 		args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
493 		if (!args->value)
494 			return -ENOMEM;
495 	}
496 	args->valuelen = valuelen;
497 
498 	/* remote block xattr requires IO for copy-in */
499 	if (args->rmtblkno)
500 		return xfs_attr_rmtval_get(args);
501 
502 	/*
503 	 * This is to prevent a GCC warning because the remote xattr case
504 	 * doesn't have a value to pass in. In that case, we never reach here,
505 	 * but GCC can't work that out and so throws a "passing NULL to
506 	 * memcpy" warning.
507 	 */
508 	if (!value)
509 		return -EINVAL;
510 	memcpy(args->value, value, valuelen);
511 	return 0;
512 }
513 
514 /*========================================================================
515  * External routines when attribute fork size < XFS_LITINO(mp).
516  *========================================================================*/
517 
518 /*
519  * Query whether the total requested number of attr fork bytes of extended
520  * attribute space will be able to fit inline.
521  *
522  * Returns zero if not, else the i_forkoff fork offset to be used in the
523  * literal area for attribute data once the new bytes have been added.
524  *
525  * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
526  * special case for dev/uuid inodes, they have fixed size data forks.
527  */
528 int
529 xfs_attr_shortform_bytesfit(
530 	struct xfs_inode	*dp,
531 	int			bytes)
532 {
533 	struct xfs_mount	*mp = dp->i_mount;
534 	int64_t			dsize;
535 	int			minforkoff;
536 	int			maxforkoff;
537 	int			offset;
538 
539 	/*
540 	 * Check if the new size could fit at all first:
541 	 */
542 	if (bytes > XFS_LITINO(mp))
543 		return 0;
544 
545 	/* rounded down */
546 	offset = (XFS_LITINO(mp) - bytes) >> 3;
547 
548 	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
549 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
550 		return (offset >= minforkoff) ? minforkoff : 0;
551 	}
552 
553 	/*
554 	 * If the requested numbers of bytes is smaller or equal to the
555 	 * current attribute fork size we can always proceed.
556 	 *
557 	 * Note that if_bytes in the data fork might actually be larger than
558 	 * the current data fork size is due to delalloc extents. In that
559 	 * case either the extent count will go down when they are converted
560 	 * to real extents, or the delalloc conversion will take care of the
561 	 * literal area rebalancing.
562 	 */
563 	if (bytes <= XFS_IFORK_ASIZE(dp))
564 		return dp->i_forkoff;
565 
566 	/*
567 	 * For attr2 we can try to move the forkoff if there is space in the
568 	 * literal area, but for the old format we are done if there is no
569 	 * space in the fixed attribute fork.
570 	 */
571 	if (!xfs_has_attr2(mp))
572 		return 0;
573 
574 	dsize = dp->i_df.if_bytes;
575 
576 	switch (dp->i_df.if_format) {
577 	case XFS_DINODE_FMT_EXTENTS:
578 		/*
579 		 * If there is no attr fork and the data fork is extents,
580 		 * determine if creating the default attr fork will result
581 		 * in the extents form migrating to btree. If so, the
582 		 * minimum offset only needs to be the space required for
583 		 * the btree root.
584 		 */
585 		if (!dp->i_forkoff && dp->i_df.if_bytes >
586 		    xfs_default_attroffset(dp))
587 			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
588 		break;
589 	case XFS_DINODE_FMT_BTREE:
590 		/*
591 		 * If we have a data btree then keep forkoff if we have one,
592 		 * otherwise we are adding a new attr, so then we set
593 		 * minforkoff to where the btree root can finish so we have
594 		 * plenty of room for attrs
595 		 */
596 		if (dp->i_forkoff) {
597 			if (offset < dp->i_forkoff)
598 				return 0;
599 			return dp->i_forkoff;
600 		}
601 		dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
602 		break;
603 	}
604 
605 	/*
606 	 * A data fork btree root must have space for at least
607 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
608 	 */
609 	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
610 	minforkoff = roundup(minforkoff, 8) >> 3;
611 
612 	/* attr fork btree root can have at least this many key/ptr pairs */
613 	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
614 	maxforkoff = maxforkoff >> 3;	/* rounded down */
615 
616 	if (offset >= maxforkoff)
617 		return maxforkoff;
618 	if (offset >= minforkoff)
619 		return offset;
620 	return 0;
621 }
622 
623 /*
624  * Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
625  * - noattr2 mount option is set,
626  * - on-disk version bit says it is already set, or
627  * - the attr2 mount option is not set to enable automatic upgrade from attr1.
628  */
629 STATIC void
630 xfs_sbversion_add_attr2(
631 	struct xfs_mount	*mp,
632 	struct xfs_trans	*tp)
633 {
634 	if (xfs_has_noattr2(mp))
635 		return;
636 	if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
637 		return;
638 	if (!xfs_has_attr2(mp))
639 		return;
640 
641 	spin_lock(&mp->m_sb_lock);
642 	xfs_add_attr2(mp);
643 	spin_unlock(&mp->m_sb_lock);
644 	xfs_log_sb(tp);
645 }
646 
647 /*
648  * Create the initial contents of a shortform attribute list.
649  */
650 void
651 xfs_attr_shortform_create(
652 	struct xfs_da_args	*args)
653 {
654 	struct xfs_inode	*dp = args->dp;
655 	struct xfs_ifork	*ifp = dp->i_afp;
656 	struct xfs_attr_sf_hdr	*hdr;
657 
658 	trace_xfs_attr_sf_create(args);
659 
660 	ASSERT(ifp->if_bytes == 0);
661 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
662 		ifp->if_format = XFS_DINODE_FMT_LOCAL;
663 	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
664 	hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
665 	memset(hdr, 0, sizeof(*hdr));
666 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
667 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
668 }
669 
670 /*
671  * Return -EEXIST if attr is found, or -ENOATTR if not
672  * args:  args containing attribute name and namelen
673  * sfep:  If not null, pointer will be set to the last attr entry found on
674 	  -EEXIST.  On -ENOATTR pointer is left at the last entry in the list
675  * basep: If not null, pointer is set to the byte offset of the entry in the
676  *	  list on -EEXIST.  On -ENOATTR, pointer is left at the byte offset of
677  *	  the last entry in the list
678  */
679 int
680 xfs_attr_sf_findname(
681 	struct xfs_da_args	 *args,
682 	struct xfs_attr_sf_entry **sfep,
683 	unsigned int		 *basep)
684 {
685 	struct xfs_attr_shortform *sf;
686 	struct xfs_attr_sf_entry *sfe;
687 	unsigned int		base = sizeof(struct xfs_attr_sf_hdr);
688 	int			size = 0;
689 	int			end;
690 	int			i;
691 
692 	sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
693 	sfe = &sf->list[0];
694 	end = sf->hdr.count;
695 	for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe),
696 			     base += size, i++) {
697 		size = xfs_attr_sf_entsize(sfe);
698 		if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
699 				    sfe->flags))
700 			continue;
701 		break;
702 	}
703 
704 	if (sfep != NULL)
705 		*sfep = sfe;
706 
707 	if (basep != NULL)
708 		*basep = base;
709 
710 	if (i == end)
711 		return -ENOATTR;
712 	return -EEXIST;
713 }
714 
715 /*
716  * Add a name/value pair to the shortform attribute list.
717  * Overflow from the inode has already been checked for.
718  */
719 void
720 xfs_attr_shortform_add(
721 	struct xfs_da_args		*args,
722 	int				forkoff)
723 {
724 	struct xfs_attr_shortform	*sf;
725 	struct xfs_attr_sf_entry	*sfe;
726 	int				offset, size;
727 	struct xfs_mount		*mp;
728 	struct xfs_inode		*dp;
729 	struct xfs_ifork		*ifp;
730 
731 	trace_xfs_attr_sf_add(args);
732 
733 	dp = args->dp;
734 	mp = dp->i_mount;
735 	dp->i_forkoff = forkoff;
736 
737 	ifp = dp->i_afp;
738 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
739 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
740 	if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
741 		ASSERT(0);
742 
743 	offset = (char *)sfe - (char *)sf;
744 	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
745 	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
746 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
747 	sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset);
748 
749 	sfe->namelen = args->namelen;
750 	sfe->valuelen = args->valuelen;
751 	sfe->flags = args->attr_filter;
752 	memcpy(sfe->nameval, args->name, args->namelen);
753 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
754 	sf->hdr.count++;
755 	be16_add_cpu(&sf->hdr.totsize, size);
756 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
757 
758 	xfs_sbversion_add_attr2(mp, args->trans);
759 }
760 
761 /*
762  * After the last attribute is removed revert to original inode format,
763  * making all literal area available to the data fork once more.
764  */
765 void
766 xfs_attr_fork_remove(
767 	struct xfs_inode	*ip,
768 	struct xfs_trans	*tp)
769 {
770 	ASSERT(ip->i_afp->if_nextents == 0);
771 
772 	xfs_idestroy_fork(ip->i_afp);
773 	kmem_cache_free(xfs_ifork_zone, ip->i_afp);
774 	ip->i_afp = NULL;
775 	ip->i_forkoff = 0;
776 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
777 }
778 
779 /*
780  * Remove an attribute from the shortform attribute list structure.
781  */
782 int
783 xfs_attr_sf_removename(
784 	struct xfs_da_args		*args)
785 {
786 	struct xfs_attr_shortform	*sf;
787 	struct xfs_attr_sf_entry	*sfe;
788 	int				size = 0, end, totsize;
789 	unsigned int			base;
790 	struct xfs_mount		*mp;
791 	struct xfs_inode		*dp;
792 	int				error;
793 
794 	trace_xfs_attr_sf_remove(args);
795 
796 	dp = args->dp;
797 	mp = dp->i_mount;
798 	sf = (struct xfs_attr_shortform *)dp->i_afp->if_u1.if_data;
799 
800 	error = xfs_attr_sf_findname(args, &sfe, &base);
801 	if (error != -EEXIST)
802 		return error;
803 	size = xfs_attr_sf_entsize(sfe);
804 
805 	/*
806 	 * Fix up the attribute fork data, covering the hole
807 	 */
808 	end = base + size;
809 	totsize = be16_to_cpu(sf->hdr.totsize);
810 	if (end != totsize)
811 		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
812 	sf->hdr.count--;
813 	be16_add_cpu(&sf->hdr.totsize, -size);
814 
815 	/*
816 	 * Fix up the start offset of the attribute fork
817 	 */
818 	totsize -= size;
819 	if (totsize == sizeof(xfs_attr_sf_hdr_t) && xfs_has_attr2(mp) &&
820 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
821 	    !(args->op_flags & XFS_DA_OP_ADDNAME)) {
822 		xfs_attr_fork_remove(dp, args->trans);
823 	} else {
824 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
825 		dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
826 		ASSERT(dp->i_forkoff);
827 		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
828 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
829 				!xfs_has_attr2(mp) ||
830 				dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
831 		xfs_trans_log_inode(args->trans, dp,
832 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
833 	}
834 
835 	xfs_sbversion_add_attr2(mp, args->trans);
836 
837 	return 0;
838 }
839 
840 /*
841  * Look up a name in a shortform attribute list structure.
842  */
843 /*ARGSUSED*/
844 int
845 xfs_attr_shortform_lookup(xfs_da_args_t *args)
846 {
847 	struct xfs_attr_shortform *sf;
848 	struct xfs_attr_sf_entry *sfe;
849 	int i;
850 	struct xfs_ifork *ifp;
851 
852 	trace_xfs_attr_sf_lookup(args);
853 
854 	ifp = args->dp->i_afp;
855 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
856 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
857 	sfe = &sf->list[0];
858 	for (i = 0; i < sf->hdr.count;
859 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
860 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
861 				sfe->flags))
862 			return -EEXIST;
863 	}
864 	return -ENOATTR;
865 }
866 
867 /*
868  * Retrieve the attribute value and length.
869  *
870  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
871  * lookup, we only return an error if the attribute does not exist or we can't
872  * retrieve the value.
873  */
874 int
875 xfs_attr_shortform_getvalue(
876 	struct xfs_da_args	*args)
877 {
878 	struct xfs_attr_shortform *sf;
879 	struct xfs_attr_sf_entry *sfe;
880 	int			i;
881 
882 	ASSERT(args->dp->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
883 	sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
884 	sfe = &sf->list[0];
885 	for (i = 0; i < sf->hdr.count;
886 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
887 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
888 				sfe->flags))
889 			return xfs_attr_copy_value(args,
890 				&sfe->nameval[args->namelen], sfe->valuelen);
891 	}
892 	return -ENOATTR;
893 }
894 
895 /*
896  * Convert from using the shortform to the leaf.  On success, return the
897  * buffer so that we can keep it locked until we're totally done with it.
898  */
899 int
900 xfs_attr_shortform_to_leaf(
901 	struct xfs_da_args		*args,
902 	struct xfs_buf			**leaf_bp)
903 {
904 	struct xfs_inode		*dp;
905 	struct xfs_attr_shortform	*sf;
906 	struct xfs_attr_sf_entry	*sfe;
907 	struct xfs_da_args		nargs;
908 	char				*tmpbuffer;
909 	int				error, i, size;
910 	xfs_dablk_t			blkno;
911 	struct xfs_buf			*bp;
912 	struct xfs_ifork		*ifp;
913 
914 	trace_xfs_attr_sf_to_leaf(args);
915 
916 	dp = args->dp;
917 	ifp = dp->i_afp;
918 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
919 	size = be16_to_cpu(sf->hdr.totsize);
920 	tmpbuffer = kmem_alloc(size, 0);
921 	ASSERT(tmpbuffer != NULL);
922 	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
923 	sf = (struct xfs_attr_shortform *)tmpbuffer;
924 
925 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
926 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
927 
928 	bp = NULL;
929 	error = xfs_da_grow_inode(args, &blkno);
930 	if (error)
931 		goto out;
932 
933 	ASSERT(blkno == 0);
934 	error = xfs_attr3_leaf_create(args, blkno, &bp);
935 	if (error)
936 		goto out;
937 
938 	memset((char *)&nargs, 0, sizeof(nargs));
939 	nargs.dp = dp;
940 	nargs.geo = args->geo;
941 	nargs.total = args->total;
942 	nargs.whichfork = XFS_ATTR_FORK;
943 	nargs.trans = args->trans;
944 	nargs.op_flags = XFS_DA_OP_OKNOENT;
945 
946 	sfe = &sf->list[0];
947 	for (i = 0; i < sf->hdr.count; i++) {
948 		nargs.name = sfe->nameval;
949 		nargs.namelen = sfe->namelen;
950 		nargs.value = &sfe->nameval[nargs.namelen];
951 		nargs.valuelen = sfe->valuelen;
952 		nargs.hashval = xfs_da_hashname(sfe->nameval,
953 						sfe->namelen);
954 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
955 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
956 		ASSERT(error == -ENOATTR);
957 		error = xfs_attr3_leaf_add(bp, &nargs);
958 		ASSERT(error != -ENOSPC);
959 		if (error)
960 			goto out;
961 		sfe = xfs_attr_sf_nextentry(sfe);
962 	}
963 	error = 0;
964 	*leaf_bp = bp;
965 out:
966 	kmem_free(tmpbuffer);
967 	return error;
968 }
969 
970 /*
971  * Check a leaf attribute block to see if all the entries would fit into
972  * a shortform attribute list.
973  */
974 int
975 xfs_attr_shortform_allfit(
976 	struct xfs_buf		*bp,
977 	struct xfs_inode	*dp)
978 {
979 	struct xfs_attr_leafblock *leaf;
980 	struct xfs_attr_leaf_entry *entry;
981 	xfs_attr_leaf_name_local_t *name_loc;
982 	struct xfs_attr3_icleaf_hdr leafhdr;
983 	int			bytes;
984 	int			i;
985 	struct xfs_mount	*mp = bp->b_mount;
986 
987 	leaf = bp->b_addr;
988 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
989 	entry = xfs_attr3_leaf_entryp(leaf);
990 
991 	bytes = sizeof(struct xfs_attr_sf_hdr);
992 	for (i = 0; i < leafhdr.count; entry++, i++) {
993 		if (entry->flags & XFS_ATTR_INCOMPLETE)
994 			continue;		/* don't copy partial entries */
995 		if (!(entry->flags & XFS_ATTR_LOCAL))
996 			return 0;
997 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
998 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
999 			return 0;
1000 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1001 			return 0;
1002 		bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1003 					be16_to_cpu(name_loc->valuelen));
1004 	}
1005 	if (xfs_has_attr2(dp->i_mount) &&
1006 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1007 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
1008 		return -1;
1009 	return xfs_attr_shortform_bytesfit(dp, bytes);
1010 }
1011 
1012 /* Verify the consistency of an inline attribute fork. */
1013 xfs_failaddr_t
1014 xfs_attr_shortform_verify(
1015 	struct xfs_inode		*ip)
1016 {
1017 	struct xfs_attr_shortform	*sfp;
1018 	struct xfs_attr_sf_entry	*sfep;
1019 	struct xfs_attr_sf_entry	*next_sfep;
1020 	char				*endp;
1021 	struct xfs_ifork		*ifp;
1022 	int				i;
1023 	int64_t				size;
1024 
1025 	ASSERT(ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
1026 	ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
1027 	sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1028 	size = ifp->if_bytes;
1029 
1030 	/*
1031 	 * Give up if the attribute is way too short.
1032 	 */
1033 	if (size < sizeof(struct xfs_attr_sf_hdr))
1034 		return __this_address;
1035 
1036 	endp = (char *)sfp + size;
1037 
1038 	/* Check all reported entries */
1039 	sfep = &sfp->list[0];
1040 	for (i = 0; i < sfp->hdr.count; i++) {
1041 		/*
1042 		 * struct xfs_attr_sf_entry has a variable length.
1043 		 * Check the fixed-offset parts of the structure are
1044 		 * within the data buffer.
1045 		 * xfs_attr_sf_entry is defined with a 1-byte variable
1046 		 * array at the end, so we must subtract that off.
1047 		 */
1048 		if (((char *)sfep + sizeof(*sfep)) >= endp)
1049 			return __this_address;
1050 
1051 		/* Don't allow names with known bad length. */
1052 		if (sfep->namelen == 0)
1053 			return __this_address;
1054 
1055 		/*
1056 		 * Check that the variable-length part of the structure is
1057 		 * within the data buffer.  The next entry starts after the
1058 		 * name component, so nextentry is an acceptable test.
1059 		 */
1060 		next_sfep = xfs_attr_sf_nextentry(sfep);
1061 		if ((char *)next_sfep > endp)
1062 			return __this_address;
1063 
1064 		/*
1065 		 * Check for unknown flags.  Short form doesn't support
1066 		 * the incomplete or local bits, so we can use the namespace
1067 		 * mask here.
1068 		 */
1069 		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1070 			return __this_address;
1071 
1072 		/*
1073 		 * Check for invalid namespace combinations.  We only allow
1074 		 * one namespace flag per xattr, so we can just count the
1075 		 * bits (i.e. hweight) here.
1076 		 */
1077 		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1078 			return __this_address;
1079 
1080 		sfep = next_sfep;
1081 	}
1082 	if ((void *)sfep != (void *)endp)
1083 		return __this_address;
1084 
1085 	return NULL;
1086 }
1087 
1088 /*
1089  * Convert a leaf attribute list to shortform attribute list
1090  */
1091 int
1092 xfs_attr3_leaf_to_shortform(
1093 	struct xfs_buf		*bp,
1094 	struct xfs_da_args	*args,
1095 	int			forkoff)
1096 {
1097 	struct xfs_attr_leafblock *leaf;
1098 	struct xfs_attr3_icleaf_hdr ichdr;
1099 	struct xfs_attr_leaf_entry *entry;
1100 	struct xfs_attr_leaf_name_local *name_loc;
1101 	struct xfs_da_args	nargs;
1102 	struct xfs_inode	*dp = args->dp;
1103 	char			*tmpbuffer;
1104 	int			error;
1105 	int			i;
1106 
1107 	trace_xfs_attr_leaf_to_sf(args);
1108 
1109 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1110 	if (!tmpbuffer)
1111 		return -ENOMEM;
1112 
1113 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1114 
1115 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1116 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1117 	entry = xfs_attr3_leaf_entryp(leaf);
1118 
1119 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1120 	memset(bp->b_addr, 0, args->geo->blksize);
1121 
1122 	/*
1123 	 * Clean out the prior contents of the attribute list.
1124 	 */
1125 	error = xfs_da_shrink_inode(args, 0, bp);
1126 	if (error)
1127 		goto out;
1128 
1129 	if (forkoff == -1) {
1130 		ASSERT(xfs_has_attr2(dp->i_mount));
1131 		ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1132 		xfs_attr_fork_remove(dp, args->trans);
1133 		goto out;
1134 	}
1135 
1136 	xfs_attr_shortform_create(args);
1137 
1138 	/*
1139 	 * Copy the attributes
1140 	 */
1141 	memset((char *)&nargs, 0, sizeof(nargs));
1142 	nargs.geo = args->geo;
1143 	nargs.dp = dp;
1144 	nargs.total = args->total;
1145 	nargs.whichfork = XFS_ATTR_FORK;
1146 	nargs.trans = args->trans;
1147 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1148 
1149 	for (i = 0; i < ichdr.count; entry++, i++) {
1150 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1151 			continue;	/* don't copy partial entries */
1152 		if (!entry->nameidx)
1153 			continue;
1154 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1155 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1156 		nargs.name = name_loc->nameval;
1157 		nargs.namelen = name_loc->namelen;
1158 		nargs.value = &name_loc->nameval[nargs.namelen];
1159 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1160 		nargs.hashval = be32_to_cpu(entry->hashval);
1161 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1162 		xfs_attr_shortform_add(&nargs, forkoff);
1163 	}
1164 	error = 0;
1165 
1166 out:
1167 	kmem_free(tmpbuffer);
1168 	return error;
1169 }
1170 
1171 /*
1172  * Convert from using a single leaf to a root node and a leaf.
1173  */
1174 int
1175 xfs_attr3_leaf_to_node(
1176 	struct xfs_da_args	*args)
1177 {
1178 	struct xfs_attr_leafblock *leaf;
1179 	struct xfs_attr3_icleaf_hdr icleafhdr;
1180 	struct xfs_attr_leaf_entry *entries;
1181 	struct xfs_da3_icnode_hdr icnodehdr;
1182 	struct xfs_da_intnode	*node;
1183 	struct xfs_inode	*dp = args->dp;
1184 	struct xfs_mount	*mp = dp->i_mount;
1185 	struct xfs_buf		*bp1 = NULL;
1186 	struct xfs_buf		*bp2 = NULL;
1187 	xfs_dablk_t		blkno;
1188 	int			error;
1189 
1190 	trace_xfs_attr_leaf_to_node(args);
1191 
1192 	error = xfs_da_grow_inode(args, &blkno);
1193 	if (error)
1194 		goto out;
1195 	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1196 	if (error)
1197 		goto out;
1198 
1199 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1200 	if (error)
1201 		goto out;
1202 
1203 	/* copy leaf to new buffer, update identifiers */
1204 	xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1205 	bp2->b_ops = bp1->b_ops;
1206 	memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1207 	if (xfs_has_crc(mp)) {
1208 		struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1209 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp2));
1210 	}
1211 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1212 
1213 	/*
1214 	 * Set up the new root node.
1215 	 */
1216 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1217 	if (error)
1218 		goto out;
1219 	node = bp1->b_addr;
1220 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1221 
1222 	leaf = bp2->b_addr;
1223 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1224 	entries = xfs_attr3_leaf_entryp(leaf);
1225 
1226 	/* both on-disk, don't endian-flip twice */
1227 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1228 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1229 	icnodehdr.count = 1;
1230 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1231 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1232 	error = 0;
1233 out:
1234 	return error;
1235 }
1236 
1237 /*========================================================================
1238  * Routines used for growing the Btree.
1239  *========================================================================*/
1240 
1241 /*
1242  * Create the initial contents of a leaf attribute list
1243  * or a leaf in a node attribute list.
1244  */
1245 STATIC int
1246 xfs_attr3_leaf_create(
1247 	struct xfs_da_args	*args,
1248 	xfs_dablk_t		blkno,
1249 	struct xfs_buf		**bpp)
1250 {
1251 	struct xfs_attr_leafblock *leaf;
1252 	struct xfs_attr3_icleaf_hdr ichdr;
1253 	struct xfs_inode	*dp = args->dp;
1254 	struct xfs_mount	*mp = dp->i_mount;
1255 	struct xfs_buf		*bp;
1256 	int			error;
1257 
1258 	trace_xfs_attr_leaf_create(args);
1259 
1260 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1261 					    XFS_ATTR_FORK);
1262 	if (error)
1263 		return error;
1264 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1265 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1266 	leaf = bp->b_addr;
1267 	memset(leaf, 0, args->geo->blksize);
1268 
1269 	memset(&ichdr, 0, sizeof(ichdr));
1270 	ichdr.firstused = args->geo->blksize;
1271 
1272 	if (xfs_has_crc(mp)) {
1273 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1274 
1275 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1276 
1277 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1278 		hdr3->owner = cpu_to_be64(dp->i_ino);
1279 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1280 
1281 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1282 	} else {
1283 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1284 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1285 	}
1286 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1287 
1288 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1289 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1290 
1291 	*bpp = bp;
1292 	return 0;
1293 }
1294 
1295 /*
1296  * Split the leaf node, rebalance, then add the new entry.
1297  */
1298 int
1299 xfs_attr3_leaf_split(
1300 	struct xfs_da_state	*state,
1301 	struct xfs_da_state_blk	*oldblk,
1302 	struct xfs_da_state_blk	*newblk)
1303 {
1304 	xfs_dablk_t blkno;
1305 	int error;
1306 
1307 	trace_xfs_attr_leaf_split(state->args);
1308 
1309 	/*
1310 	 * Allocate space for a new leaf node.
1311 	 */
1312 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1313 	error = xfs_da_grow_inode(state->args, &blkno);
1314 	if (error)
1315 		return error;
1316 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1317 	if (error)
1318 		return error;
1319 	newblk->blkno = blkno;
1320 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1321 
1322 	/*
1323 	 * Rebalance the entries across the two leaves.
1324 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1325 	 */
1326 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1327 	error = xfs_da3_blk_link(state, oldblk, newblk);
1328 	if (error)
1329 		return error;
1330 
1331 	/*
1332 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1333 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1334 	 * "new" attrs info.  Will need the "old" info to remove it later.
1335 	 *
1336 	 * Insert the "new" entry in the correct block.
1337 	 */
1338 	if (state->inleaf) {
1339 		trace_xfs_attr_leaf_add_old(state->args);
1340 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1341 	} else {
1342 		trace_xfs_attr_leaf_add_new(state->args);
1343 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1344 	}
1345 
1346 	/*
1347 	 * Update last hashval in each block since we added the name.
1348 	 */
1349 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1350 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1351 	return error;
1352 }
1353 
1354 /*
1355  * Add a name to the leaf attribute list structure.
1356  */
1357 int
1358 xfs_attr3_leaf_add(
1359 	struct xfs_buf		*bp,
1360 	struct xfs_da_args	*args)
1361 {
1362 	struct xfs_attr_leafblock *leaf;
1363 	struct xfs_attr3_icleaf_hdr ichdr;
1364 	int			tablesize;
1365 	int			entsize;
1366 	int			sum;
1367 	int			tmp;
1368 	int			i;
1369 
1370 	trace_xfs_attr_leaf_add(args);
1371 
1372 	leaf = bp->b_addr;
1373 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1374 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1375 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1376 
1377 	/*
1378 	 * Search through freemap for first-fit on new name length.
1379 	 * (may need to figure in size of entry struct too)
1380 	 */
1381 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1382 					+ xfs_attr3_leaf_hdr_size(leaf);
1383 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1384 		if (tablesize > ichdr.firstused) {
1385 			sum += ichdr.freemap[i].size;
1386 			continue;
1387 		}
1388 		if (!ichdr.freemap[i].size)
1389 			continue;	/* no space in this map */
1390 		tmp = entsize;
1391 		if (ichdr.freemap[i].base < ichdr.firstused)
1392 			tmp += sizeof(xfs_attr_leaf_entry_t);
1393 		if (ichdr.freemap[i].size >= tmp) {
1394 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1395 			goto out_log_hdr;
1396 		}
1397 		sum += ichdr.freemap[i].size;
1398 	}
1399 
1400 	/*
1401 	 * If there are no holes in the address space of the block,
1402 	 * and we don't have enough freespace, then compaction will do us
1403 	 * no good and we should just give up.
1404 	 */
1405 	if (!ichdr.holes && sum < entsize)
1406 		return -ENOSPC;
1407 
1408 	/*
1409 	 * Compact the entries to coalesce free space.
1410 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1411 	 */
1412 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1413 
1414 	/*
1415 	 * After compaction, the block is guaranteed to have only one
1416 	 * free region, in freemap[0].  If it is not big enough, give up.
1417 	 */
1418 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1419 		tmp = -ENOSPC;
1420 		goto out_log_hdr;
1421 	}
1422 
1423 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1424 
1425 out_log_hdr:
1426 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1427 	xfs_trans_log_buf(args->trans, bp,
1428 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1429 				xfs_attr3_leaf_hdr_size(leaf)));
1430 	return tmp;
1431 }
1432 
1433 /*
1434  * Add a name to a leaf attribute list structure.
1435  */
1436 STATIC int
1437 xfs_attr3_leaf_add_work(
1438 	struct xfs_buf		*bp,
1439 	struct xfs_attr3_icleaf_hdr *ichdr,
1440 	struct xfs_da_args	*args,
1441 	int			mapindex)
1442 {
1443 	struct xfs_attr_leafblock *leaf;
1444 	struct xfs_attr_leaf_entry *entry;
1445 	struct xfs_attr_leaf_name_local *name_loc;
1446 	struct xfs_attr_leaf_name_remote *name_rmt;
1447 	struct xfs_mount	*mp;
1448 	int			tmp;
1449 	int			i;
1450 
1451 	trace_xfs_attr_leaf_add_work(args);
1452 
1453 	leaf = bp->b_addr;
1454 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1455 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1456 
1457 	/*
1458 	 * Force open some space in the entry array and fill it in.
1459 	 */
1460 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1461 	if (args->index < ichdr->count) {
1462 		tmp  = ichdr->count - args->index;
1463 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1464 		memmove(entry + 1, entry, tmp);
1465 		xfs_trans_log_buf(args->trans, bp,
1466 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1467 	}
1468 	ichdr->count++;
1469 
1470 	/*
1471 	 * Allocate space for the new string (at the end of the run).
1472 	 */
1473 	mp = args->trans->t_mountp;
1474 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1475 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1476 	ASSERT(ichdr->freemap[mapindex].size >=
1477 		xfs_attr_leaf_newentsize(args, NULL));
1478 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1479 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1480 
1481 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1482 
1483 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1484 				     ichdr->freemap[mapindex].size);
1485 	entry->hashval = cpu_to_be32(args->hashval);
1486 	entry->flags = args->attr_filter;
1487 	if (tmp)
1488 		entry->flags |= XFS_ATTR_LOCAL;
1489 	if (args->op_flags & XFS_DA_OP_RENAME) {
1490 		entry->flags |= XFS_ATTR_INCOMPLETE;
1491 		if ((args->blkno2 == args->blkno) &&
1492 		    (args->index2 <= args->index)) {
1493 			args->index2++;
1494 		}
1495 	}
1496 	xfs_trans_log_buf(args->trans, bp,
1497 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1498 	ASSERT((args->index == 0) ||
1499 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1500 	ASSERT((args->index == ichdr->count - 1) ||
1501 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1502 
1503 	/*
1504 	 * For "remote" attribute values, simply note that we need to
1505 	 * allocate space for the "remote" value.  We can't actually
1506 	 * allocate the extents in this transaction, and we can't decide
1507 	 * which blocks they should be as we might allocate more blocks
1508 	 * as part of this transaction (a split operation for example).
1509 	 */
1510 	if (entry->flags & XFS_ATTR_LOCAL) {
1511 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1512 		name_loc->namelen = args->namelen;
1513 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1514 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1515 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1516 				   be16_to_cpu(name_loc->valuelen));
1517 	} else {
1518 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1519 		name_rmt->namelen = args->namelen;
1520 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1521 		entry->flags |= XFS_ATTR_INCOMPLETE;
1522 		/* just in case */
1523 		name_rmt->valuelen = 0;
1524 		name_rmt->valueblk = 0;
1525 		args->rmtblkno = 1;
1526 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1527 		args->rmtvaluelen = args->valuelen;
1528 	}
1529 	xfs_trans_log_buf(args->trans, bp,
1530 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1531 				   xfs_attr_leaf_entsize(leaf, args->index)));
1532 
1533 	/*
1534 	 * Update the control info for this leaf node
1535 	 */
1536 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1537 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1538 
1539 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1540 					+ xfs_attr3_leaf_hdr_size(leaf));
1541 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1542 					+ xfs_attr3_leaf_hdr_size(leaf);
1543 
1544 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1545 		if (ichdr->freemap[i].base == tmp) {
1546 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1547 			ichdr->freemap[i].size -=
1548 				min_t(uint16_t, ichdr->freemap[i].size,
1549 						sizeof(xfs_attr_leaf_entry_t));
1550 		}
1551 	}
1552 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1553 	return 0;
1554 }
1555 
1556 /*
1557  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1558  */
1559 STATIC void
1560 xfs_attr3_leaf_compact(
1561 	struct xfs_da_args	*args,
1562 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1563 	struct xfs_buf		*bp)
1564 {
1565 	struct xfs_attr_leafblock *leaf_src;
1566 	struct xfs_attr_leafblock *leaf_dst;
1567 	struct xfs_attr3_icleaf_hdr ichdr_src;
1568 	struct xfs_trans	*trans = args->trans;
1569 	char			*tmpbuffer;
1570 
1571 	trace_xfs_attr_leaf_compact(args);
1572 
1573 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1574 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1575 	memset(bp->b_addr, 0, args->geo->blksize);
1576 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1577 	leaf_dst = bp->b_addr;
1578 
1579 	/*
1580 	 * Copy the on-disk header back into the destination buffer to ensure
1581 	 * all the information in the header that is not part of the incore
1582 	 * header structure is preserved.
1583 	 */
1584 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1585 
1586 	/* Initialise the incore headers */
1587 	ichdr_src = *ichdr_dst;	/* struct copy */
1588 	ichdr_dst->firstused = args->geo->blksize;
1589 	ichdr_dst->usedbytes = 0;
1590 	ichdr_dst->count = 0;
1591 	ichdr_dst->holes = 0;
1592 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1593 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1594 						ichdr_dst->freemap[0].base;
1595 
1596 	/* write the header back to initialise the underlying buffer */
1597 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1598 
1599 	/*
1600 	 * Copy all entry's in the same (sorted) order,
1601 	 * but allocate name/value pairs packed and in sequence.
1602 	 */
1603 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1604 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1605 	/*
1606 	 * this logs the entire buffer, but the caller must write the header
1607 	 * back to the buffer when it is finished modifying it.
1608 	 */
1609 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1610 
1611 	kmem_free(tmpbuffer);
1612 }
1613 
1614 /*
1615  * Compare two leaf blocks "order".
1616  * Return 0 unless leaf2 should go before leaf1.
1617  */
1618 static int
1619 xfs_attr3_leaf_order(
1620 	struct xfs_buf	*leaf1_bp,
1621 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1622 	struct xfs_buf	*leaf2_bp,
1623 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1624 {
1625 	struct xfs_attr_leaf_entry *entries1;
1626 	struct xfs_attr_leaf_entry *entries2;
1627 
1628 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1629 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1630 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1631 	    ((be32_to_cpu(entries2[0].hashval) <
1632 	      be32_to_cpu(entries1[0].hashval)) ||
1633 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1634 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1635 		return 1;
1636 	}
1637 	return 0;
1638 }
1639 
1640 int
1641 xfs_attr_leaf_order(
1642 	struct xfs_buf	*leaf1_bp,
1643 	struct xfs_buf	*leaf2_bp)
1644 {
1645 	struct xfs_attr3_icleaf_hdr ichdr1;
1646 	struct xfs_attr3_icleaf_hdr ichdr2;
1647 	struct xfs_mount *mp = leaf1_bp->b_mount;
1648 
1649 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1650 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1651 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1652 }
1653 
1654 /*
1655  * Redistribute the attribute list entries between two leaf nodes,
1656  * taking into account the size of the new entry.
1657  *
1658  * NOTE: if new block is empty, then it will get the upper half of the
1659  * old block.  At present, all (one) callers pass in an empty second block.
1660  *
1661  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1662  * to match what it is doing in splitting the attribute leaf block.  Those
1663  * values are used in "atomic rename" operations on attributes.  Note that
1664  * the "new" and "old" values can end up in different blocks.
1665  */
1666 STATIC void
1667 xfs_attr3_leaf_rebalance(
1668 	struct xfs_da_state	*state,
1669 	struct xfs_da_state_blk	*blk1,
1670 	struct xfs_da_state_blk	*blk2)
1671 {
1672 	struct xfs_da_args	*args;
1673 	struct xfs_attr_leafblock *leaf1;
1674 	struct xfs_attr_leafblock *leaf2;
1675 	struct xfs_attr3_icleaf_hdr ichdr1;
1676 	struct xfs_attr3_icleaf_hdr ichdr2;
1677 	struct xfs_attr_leaf_entry *entries1;
1678 	struct xfs_attr_leaf_entry *entries2;
1679 	int			count;
1680 	int			totallen;
1681 	int			max;
1682 	int			space;
1683 	int			swap;
1684 
1685 	/*
1686 	 * Set up environment.
1687 	 */
1688 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1689 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1690 	leaf1 = blk1->bp->b_addr;
1691 	leaf2 = blk2->bp->b_addr;
1692 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1693 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1694 	ASSERT(ichdr2.count == 0);
1695 	args = state->args;
1696 
1697 	trace_xfs_attr_leaf_rebalance(args);
1698 
1699 	/*
1700 	 * Check ordering of blocks, reverse if it makes things simpler.
1701 	 *
1702 	 * NOTE: Given that all (current) callers pass in an empty
1703 	 * second block, this code should never set "swap".
1704 	 */
1705 	swap = 0;
1706 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1707 		swap(blk1, blk2);
1708 
1709 		/* swap structures rather than reconverting them */
1710 		swap(ichdr1, ichdr2);
1711 
1712 		leaf1 = blk1->bp->b_addr;
1713 		leaf2 = blk2->bp->b_addr;
1714 		swap = 1;
1715 	}
1716 
1717 	/*
1718 	 * Examine entries until we reduce the absolute difference in
1719 	 * byte usage between the two blocks to a minimum.  Then get
1720 	 * the direction to copy and the number of elements to move.
1721 	 *
1722 	 * "inleaf" is true if the new entry should be inserted into blk1.
1723 	 * If "swap" is also true, then reverse the sense of "inleaf".
1724 	 */
1725 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1726 						      blk2, &ichdr2,
1727 						      &count, &totallen);
1728 	if (swap)
1729 		state->inleaf = !state->inleaf;
1730 
1731 	/*
1732 	 * Move any entries required from leaf to leaf:
1733 	 */
1734 	if (count < ichdr1.count) {
1735 		/*
1736 		 * Figure the total bytes to be added to the destination leaf.
1737 		 */
1738 		/* number entries being moved */
1739 		count = ichdr1.count - count;
1740 		space  = ichdr1.usedbytes - totallen;
1741 		space += count * sizeof(xfs_attr_leaf_entry_t);
1742 
1743 		/*
1744 		 * leaf2 is the destination, compact it if it looks tight.
1745 		 */
1746 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1747 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1748 		if (space > max)
1749 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1750 
1751 		/*
1752 		 * Move high entries from leaf1 to low end of leaf2.
1753 		 */
1754 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1755 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1756 
1757 	} else if (count > ichdr1.count) {
1758 		/*
1759 		 * I assert that since all callers pass in an empty
1760 		 * second buffer, this code should never execute.
1761 		 */
1762 		ASSERT(0);
1763 
1764 		/*
1765 		 * Figure the total bytes to be added to the destination leaf.
1766 		 */
1767 		/* number entries being moved */
1768 		count -= ichdr1.count;
1769 		space  = totallen - ichdr1.usedbytes;
1770 		space += count * sizeof(xfs_attr_leaf_entry_t);
1771 
1772 		/*
1773 		 * leaf1 is the destination, compact it if it looks tight.
1774 		 */
1775 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1776 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1777 		if (space > max)
1778 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1779 
1780 		/*
1781 		 * Move low entries from leaf2 to high end of leaf1.
1782 		 */
1783 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1784 					ichdr1.count, count);
1785 	}
1786 
1787 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1788 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1789 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1790 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1791 
1792 	/*
1793 	 * Copy out last hashval in each block for B-tree code.
1794 	 */
1795 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1796 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1797 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1798 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1799 
1800 	/*
1801 	 * Adjust the expected index for insertion.
1802 	 * NOTE: this code depends on the (current) situation that the
1803 	 * second block was originally empty.
1804 	 *
1805 	 * If the insertion point moved to the 2nd block, we must adjust
1806 	 * the index.  We must also track the entry just following the
1807 	 * new entry for use in an "atomic rename" operation, that entry
1808 	 * is always the "old" entry and the "new" entry is what we are
1809 	 * inserting.  The index/blkno fields refer to the "old" entry,
1810 	 * while the index2/blkno2 fields refer to the "new" entry.
1811 	 */
1812 	if (blk1->index > ichdr1.count) {
1813 		ASSERT(state->inleaf == 0);
1814 		blk2->index = blk1->index - ichdr1.count;
1815 		args->index = args->index2 = blk2->index;
1816 		args->blkno = args->blkno2 = blk2->blkno;
1817 	} else if (blk1->index == ichdr1.count) {
1818 		if (state->inleaf) {
1819 			args->index = blk1->index;
1820 			args->blkno = blk1->blkno;
1821 			args->index2 = 0;
1822 			args->blkno2 = blk2->blkno;
1823 		} else {
1824 			/*
1825 			 * On a double leaf split, the original attr location
1826 			 * is already stored in blkno2/index2, so don't
1827 			 * overwrite it overwise we corrupt the tree.
1828 			 */
1829 			blk2->index = blk1->index - ichdr1.count;
1830 			args->index = blk2->index;
1831 			args->blkno = blk2->blkno;
1832 			if (!state->extravalid) {
1833 				/*
1834 				 * set the new attr location to match the old
1835 				 * one and let the higher level split code
1836 				 * decide where in the leaf to place it.
1837 				 */
1838 				args->index2 = blk2->index;
1839 				args->blkno2 = blk2->blkno;
1840 			}
1841 		}
1842 	} else {
1843 		ASSERT(state->inleaf == 1);
1844 		args->index = args->index2 = blk1->index;
1845 		args->blkno = args->blkno2 = blk1->blkno;
1846 	}
1847 }
1848 
1849 /*
1850  * Examine entries until we reduce the absolute difference in
1851  * byte usage between the two blocks to a minimum.
1852  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1853  * GROT: there will always be enough room in either block for a new entry.
1854  * GROT: Do a double-split for this case?
1855  */
1856 STATIC int
1857 xfs_attr3_leaf_figure_balance(
1858 	struct xfs_da_state		*state,
1859 	struct xfs_da_state_blk		*blk1,
1860 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1861 	struct xfs_da_state_blk		*blk2,
1862 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1863 	int				*countarg,
1864 	int				*usedbytesarg)
1865 {
1866 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1867 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1868 	struct xfs_attr_leaf_entry	*entry;
1869 	int				count;
1870 	int				max;
1871 	int				index;
1872 	int				totallen = 0;
1873 	int				half;
1874 	int				lastdelta;
1875 	int				foundit = 0;
1876 	int				tmp;
1877 
1878 	/*
1879 	 * Examine entries until we reduce the absolute difference in
1880 	 * byte usage between the two blocks to a minimum.
1881 	 */
1882 	max = ichdr1->count + ichdr2->count;
1883 	half = (max + 1) * sizeof(*entry);
1884 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1885 			xfs_attr_leaf_newentsize(state->args, NULL);
1886 	half /= 2;
1887 	lastdelta = state->args->geo->blksize;
1888 	entry = xfs_attr3_leaf_entryp(leaf1);
1889 	for (count = index = 0; count < max; entry++, index++, count++) {
1890 
1891 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1892 		/*
1893 		 * The new entry is in the first block, account for it.
1894 		 */
1895 		if (count == blk1->index) {
1896 			tmp = totallen + sizeof(*entry) +
1897 				xfs_attr_leaf_newentsize(state->args, NULL);
1898 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1899 				break;
1900 			lastdelta = XFS_ATTR_ABS(half - tmp);
1901 			totallen = tmp;
1902 			foundit = 1;
1903 		}
1904 
1905 		/*
1906 		 * Wrap around into the second block if necessary.
1907 		 */
1908 		if (count == ichdr1->count) {
1909 			leaf1 = leaf2;
1910 			entry = xfs_attr3_leaf_entryp(leaf1);
1911 			index = 0;
1912 		}
1913 
1914 		/*
1915 		 * Figure out if next leaf entry would be too much.
1916 		 */
1917 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1918 									index);
1919 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1920 			break;
1921 		lastdelta = XFS_ATTR_ABS(half - tmp);
1922 		totallen = tmp;
1923 #undef XFS_ATTR_ABS
1924 	}
1925 
1926 	/*
1927 	 * Calculate the number of usedbytes that will end up in lower block.
1928 	 * If new entry not in lower block, fix up the count.
1929 	 */
1930 	totallen -= count * sizeof(*entry);
1931 	if (foundit) {
1932 		totallen -= sizeof(*entry) +
1933 				xfs_attr_leaf_newentsize(state->args, NULL);
1934 	}
1935 
1936 	*countarg = count;
1937 	*usedbytesarg = totallen;
1938 	return foundit;
1939 }
1940 
1941 /*========================================================================
1942  * Routines used for shrinking the Btree.
1943  *========================================================================*/
1944 
1945 /*
1946  * Check a leaf block and its neighbors to see if the block should be
1947  * collapsed into one or the other neighbor.  Always keep the block
1948  * with the smaller block number.
1949  * If the current block is over 50% full, don't try to join it, return 0.
1950  * If the block is empty, fill in the state structure and return 2.
1951  * If it can be collapsed, fill in the state structure and return 1.
1952  * If nothing can be done, return 0.
1953  *
1954  * GROT: allow for INCOMPLETE entries in calculation.
1955  */
1956 int
1957 xfs_attr3_leaf_toosmall(
1958 	struct xfs_da_state	*state,
1959 	int			*action)
1960 {
1961 	struct xfs_attr_leafblock *leaf;
1962 	struct xfs_da_state_blk	*blk;
1963 	struct xfs_attr3_icleaf_hdr ichdr;
1964 	struct xfs_buf		*bp;
1965 	xfs_dablk_t		blkno;
1966 	int			bytes;
1967 	int			forward;
1968 	int			error;
1969 	int			retval;
1970 	int			i;
1971 
1972 	trace_xfs_attr_leaf_toosmall(state->args);
1973 
1974 	/*
1975 	 * Check for the degenerate case of the block being over 50% full.
1976 	 * If so, it's not worth even looking to see if we might be able
1977 	 * to coalesce with a sibling.
1978 	 */
1979 	blk = &state->path.blk[ state->path.active-1 ];
1980 	leaf = blk->bp->b_addr;
1981 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1982 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
1983 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1984 		ichdr.usedbytes;
1985 	if (bytes > (state->args->geo->blksize >> 1)) {
1986 		*action = 0;	/* blk over 50%, don't try to join */
1987 		return 0;
1988 	}
1989 
1990 	/*
1991 	 * Check for the degenerate case of the block being empty.
1992 	 * If the block is empty, we'll simply delete it, no need to
1993 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1994 	 * to merge with the forward block unless it is NULL.
1995 	 */
1996 	if (ichdr.count == 0) {
1997 		/*
1998 		 * Make altpath point to the block we want to keep and
1999 		 * path point to the block we want to drop (this one).
2000 		 */
2001 		forward = (ichdr.forw != 0);
2002 		memcpy(&state->altpath, &state->path, sizeof(state->path));
2003 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2004 						 0, &retval);
2005 		if (error)
2006 			return error;
2007 		if (retval) {
2008 			*action = 0;
2009 		} else {
2010 			*action = 2;
2011 		}
2012 		return 0;
2013 	}
2014 
2015 	/*
2016 	 * Examine each sibling block to see if we can coalesce with
2017 	 * at least 25% free space to spare.  We need to figure out
2018 	 * whether to merge with the forward or the backward block.
2019 	 * We prefer coalescing with the lower numbered sibling so as
2020 	 * to shrink an attribute list over time.
2021 	 */
2022 	/* start with smaller blk num */
2023 	forward = ichdr.forw < ichdr.back;
2024 	for (i = 0; i < 2; forward = !forward, i++) {
2025 		struct xfs_attr3_icleaf_hdr ichdr2;
2026 		if (forward)
2027 			blkno = ichdr.forw;
2028 		else
2029 			blkno = ichdr.back;
2030 		if (blkno == 0)
2031 			continue;
2032 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2033 					blkno, &bp);
2034 		if (error)
2035 			return error;
2036 
2037 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2038 
2039 		bytes = state->args->geo->blksize -
2040 			(state->args->geo->blksize >> 2) -
2041 			ichdr.usedbytes - ichdr2.usedbytes -
2042 			((ichdr.count + ichdr2.count) *
2043 					sizeof(xfs_attr_leaf_entry_t)) -
2044 			xfs_attr3_leaf_hdr_size(leaf);
2045 
2046 		xfs_trans_brelse(state->args->trans, bp);
2047 		if (bytes >= 0)
2048 			break;	/* fits with at least 25% to spare */
2049 	}
2050 	if (i >= 2) {
2051 		*action = 0;
2052 		return 0;
2053 	}
2054 
2055 	/*
2056 	 * Make altpath point to the block we want to keep (the lower
2057 	 * numbered block) and path point to the block we want to drop.
2058 	 */
2059 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2060 	if (blkno < blk->blkno) {
2061 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2062 						 0, &retval);
2063 	} else {
2064 		error = xfs_da3_path_shift(state, &state->path, forward,
2065 						 0, &retval);
2066 	}
2067 	if (error)
2068 		return error;
2069 	if (retval) {
2070 		*action = 0;
2071 	} else {
2072 		*action = 1;
2073 	}
2074 	return 0;
2075 }
2076 
2077 /*
2078  * Remove a name from the leaf attribute list structure.
2079  *
2080  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2081  * If two leaves are 37% full, when combined they will leave 25% free.
2082  */
2083 int
2084 xfs_attr3_leaf_remove(
2085 	struct xfs_buf		*bp,
2086 	struct xfs_da_args	*args)
2087 {
2088 	struct xfs_attr_leafblock *leaf;
2089 	struct xfs_attr3_icleaf_hdr ichdr;
2090 	struct xfs_attr_leaf_entry *entry;
2091 	int			before;
2092 	int			after;
2093 	int			smallest;
2094 	int			entsize;
2095 	int			tablesize;
2096 	int			tmp;
2097 	int			i;
2098 
2099 	trace_xfs_attr_leaf_remove(args);
2100 
2101 	leaf = bp->b_addr;
2102 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2103 
2104 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2105 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2106 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2107 					xfs_attr3_leaf_hdr_size(leaf));
2108 
2109 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2110 
2111 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2112 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2113 
2114 	/*
2115 	 * Scan through free region table:
2116 	 *    check for adjacency of free'd entry with an existing one,
2117 	 *    find smallest free region in case we need to replace it,
2118 	 *    adjust any map that borders the entry table,
2119 	 */
2120 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2121 					+ xfs_attr3_leaf_hdr_size(leaf);
2122 	tmp = ichdr.freemap[0].size;
2123 	before = after = -1;
2124 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2125 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2126 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2127 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2128 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2129 		if (ichdr.freemap[i].base == tablesize) {
2130 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2131 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2132 		}
2133 
2134 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2135 				be16_to_cpu(entry->nameidx)) {
2136 			before = i;
2137 		} else if (ichdr.freemap[i].base ==
2138 				(be16_to_cpu(entry->nameidx) + entsize)) {
2139 			after = i;
2140 		} else if (ichdr.freemap[i].size < tmp) {
2141 			tmp = ichdr.freemap[i].size;
2142 			smallest = i;
2143 		}
2144 	}
2145 
2146 	/*
2147 	 * Coalesce adjacent freemap regions,
2148 	 * or replace the smallest region.
2149 	 */
2150 	if ((before >= 0) || (after >= 0)) {
2151 		if ((before >= 0) && (after >= 0)) {
2152 			ichdr.freemap[before].size += entsize;
2153 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2154 			ichdr.freemap[after].base = 0;
2155 			ichdr.freemap[after].size = 0;
2156 		} else if (before >= 0) {
2157 			ichdr.freemap[before].size += entsize;
2158 		} else {
2159 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2160 			ichdr.freemap[after].size += entsize;
2161 		}
2162 	} else {
2163 		/*
2164 		 * Replace smallest region (if it is smaller than free'd entry)
2165 		 */
2166 		if (ichdr.freemap[smallest].size < entsize) {
2167 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2168 			ichdr.freemap[smallest].size = entsize;
2169 		}
2170 	}
2171 
2172 	/*
2173 	 * Did we remove the first entry?
2174 	 */
2175 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2176 		smallest = 1;
2177 	else
2178 		smallest = 0;
2179 
2180 	/*
2181 	 * Compress the remaining entries and zero out the removed stuff.
2182 	 */
2183 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2184 	ichdr.usedbytes -= entsize;
2185 	xfs_trans_log_buf(args->trans, bp,
2186 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2187 				   entsize));
2188 
2189 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2190 	memmove(entry, entry + 1, tmp);
2191 	ichdr.count--;
2192 	xfs_trans_log_buf(args->trans, bp,
2193 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2194 
2195 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2196 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2197 
2198 	/*
2199 	 * If we removed the first entry, re-find the first used byte
2200 	 * in the name area.  Note that if the entry was the "firstused",
2201 	 * then we don't have a "hole" in our block resulting from
2202 	 * removing the name.
2203 	 */
2204 	if (smallest) {
2205 		tmp = args->geo->blksize;
2206 		entry = xfs_attr3_leaf_entryp(leaf);
2207 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2208 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2209 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2210 
2211 			if (be16_to_cpu(entry->nameidx) < tmp)
2212 				tmp = be16_to_cpu(entry->nameidx);
2213 		}
2214 		ichdr.firstused = tmp;
2215 		ASSERT(ichdr.firstused != 0);
2216 	} else {
2217 		ichdr.holes = 1;	/* mark as needing compaction */
2218 	}
2219 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2220 	xfs_trans_log_buf(args->trans, bp,
2221 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2222 					  xfs_attr3_leaf_hdr_size(leaf)));
2223 
2224 	/*
2225 	 * Check if leaf is less than 50% full, caller may want to
2226 	 * "join" the leaf with a sibling if so.
2227 	 */
2228 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2229 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2230 
2231 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2232 }
2233 
2234 /*
2235  * Move all the attribute list entries from drop_leaf into save_leaf.
2236  */
2237 void
2238 xfs_attr3_leaf_unbalance(
2239 	struct xfs_da_state	*state,
2240 	struct xfs_da_state_blk	*drop_blk,
2241 	struct xfs_da_state_blk	*save_blk)
2242 {
2243 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2244 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2245 	struct xfs_attr3_icleaf_hdr drophdr;
2246 	struct xfs_attr3_icleaf_hdr savehdr;
2247 	struct xfs_attr_leaf_entry *entry;
2248 
2249 	trace_xfs_attr_leaf_unbalance(state->args);
2250 
2251 	drop_leaf = drop_blk->bp->b_addr;
2252 	save_leaf = save_blk->bp->b_addr;
2253 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2254 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2255 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2256 
2257 	/*
2258 	 * Save last hashval from dying block for later Btree fixup.
2259 	 */
2260 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2261 
2262 	/*
2263 	 * Check if we need a temp buffer, or can we do it in place.
2264 	 * Note that we don't check "leaf" for holes because we will
2265 	 * always be dropping it, toosmall() decided that for us already.
2266 	 */
2267 	if (savehdr.holes == 0) {
2268 		/*
2269 		 * dest leaf has no holes, so we add there.  May need
2270 		 * to make some room in the entry array.
2271 		 */
2272 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2273 					 drop_blk->bp, &drophdr)) {
2274 			xfs_attr3_leaf_moveents(state->args,
2275 						drop_leaf, &drophdr, 0,
2276 						save_leaf, &savehdr, 0,
2277 						drophdr.count);
2278 		} else {
2279 			xfs_attr3_leaf_moveents(state->args,
2280 						drop_leaf, &drophdr, 0,
2281 						save_leaf, &savehdr,
2282 						savehdr.count, drophdr.count);
2283 		}
2284 	} else {
2285 		/*
2286 		 * Destination has holes, so we make a temporary copy
2287 		 * of the leaf and add them both to that.
2288 		 */
2289 		struct xfs_attr_leafblock *tmp_leaf;
2290 		struct xfs_attr3_icleaf_hdr tmphdr;
2291 
2292 		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2293 
2294 		/*
2295 		 * Copy the header into the temp leaf so that all the stuff
2296 		 * not in the incore header is present and gets copied back in
2297 		 * once we've moved all the entries.
2298 		 */
2299 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2300 
2301 		memset(&tmphdr, 0, sizeof(tmphdr));
2302 		tmphdr.magic = savehdr.magic;
2303 		tmphdr.forw = savehdr.forw;
2304 		tmphdr.back = savehdr.back;
2305 		tmphdr.firstused = state->args->geo->blksize;
2306 
2307 		/* write the header to the temp buffer to initialise it */
2308 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2309 
2310 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2311 					 drop_blk->bp, &drophdr)) {
2312 			xfs_attr3_leaf_moveents(state->args,
2313 						drop_leaf, &drophdr, 0,
2314 						tmp_leaf, &tmphdr, 0,
2315 						drophdr.count);
2316 			xfs_attr3_leaf_moveents(state->args,
2317 						save_leaf, &savehdr, 0,
2318 						tmp_leaf, &tmphdr, tmphdr.count,
2319 						savehdr.count);
2320 		} else {
2321 			xfs_attr3_leaf_moveents(state->args,
2322 						save_leaf, &savehdr, 0,
2323 						tmp_leaf, &tmphdr, 0,
2324 						savehdr.count);
2325 			xfs_attr3_leaf_moveents(state->args,
2326 						drop_leaf, &drophdr, 0,
2327 						tmp_leaf, &tmphdr, tmphdr.count,
2328 						drophdr.count);
2329 		}
2330 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2331 		savehdr = tmphdr; /* struct copy */
2332 		kmem_free(tmp_leaf);
2333 	}
2334 
2335 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2336 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2337 					   state->args->geo->blksize - 1);
2338 
2339 	/*
2340 	 * Copy out last hashval in each block for B-tree code.
2341 	 */
2342 	entry = xfs_attr3_leaf_entryp(save_leaf);
2343 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2344 }
2345 
2346 /*========================================================================
2347  * Routines used for finding things in the Btree.
2348  *========================================================================*/
2349 
2350 /*
2351  * Look up a name in a leaf attribute list structure.
2352  * This is the internal routine, it uses the caller's buffer.
2353  *
2354  * Note that duplicate keys are allowed, but only check within the
2355  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2356  *
2357  * Return in args->index the index into the entry[] array of either
2358  * the found entry, or where the entry should have been (insert before
2359  * that entry).
2360  *
2361  * Don't change the args->value unless we find the attribute.
2362  */
2363 int
2364 xfs_attr3_leaf_lookup_int(
2365 	struct xfs_buf		*bp,
2366 	struct xfs_da_args	*args)
2367 {
2368 	struct xfs_attr_leafblock *leaf;
2369 	struct xfs_attr3_icleaf_hdr ichdr;
2370 	struct xfs_attr_leaf_entry *entry;
2371 	struct xfs_attr_leaf_entry *entries;
2372 	struct xfs_attr_leaf_name_local *name_loc;
2373 	struct xfs_attr_leaf_name_remote *name_rmt;
2374 	xfs_dahash_t		hashval;
2375 	int			probe;
2376 	int			span;
2377 
2378 	trace_xfs_attr_leaf_lookup(args);
2379 
2380 	leaf = bp->b_addr;
2381 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2382 	entries = xfs_attr3_leaf_entryp(leaf);
2383 	if (ichdr.count >= args->geo->blksize / 8) {
2384 		xfs_buf_mark_corrupt(bp);
2385 		return -EFSCORRUPTED;
2386 	}
2387 
2388 	/*
2389 	 * Binary search.  (note: small blocks will skip this loop)
2390 	 */
2391 	hashval = args->hashval;
2392 	probe = span = ichdr.count / 2;
2393 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2394 		span /= 2;
2395 		if (be32_to_cpu(entry->hashval) < hashval)
2396 			probe += span;
2397 		else if (be32_to_cpu(entry->hashval) > hashval)
2398 			probe -= span;
2399 		else
2400 			break;
2401 	}
2402 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2403 		xfs_buf_mark_corrupt(bp);
2404 		return -EFSCORRUPTED;
2405 	}
2406 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2407 		xfs_buf_mark_corrupt(bp);
2408 		return -EFSCORRUPTED;
2409 	}
2410 
2411 	/*
2412 	 * Since we may have duplicate hashval's, find the first matching
2413 	 * hashval in the leaf.
2414 	 */
2415 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2416 		entry--;
2417 		probe--;
2418 	}
2419 	while (probe < ichdr.count &&
2420 	       be32_to_cpu(entry->hashval) < hashval) {
2421 		entry++;
2422 		probe++;
2423 	}
2424 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2425 		args->index = probe;
2426 		return -ENOATTR;
2427 	}
2428 
2429 	/*
2430 	 * Duplicate keys may be present, so search all of them for a match.
2431 	 */
2432 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2433 			entry++, probe++) {
2434 /*
2435  * GROT: Add code to remove incomplete entries.
2436  */
2437 		if (entry->flags & XFS_ATTR_LOCAL) {
2438 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2439 			if (!xfs_attr_match(args, name_loc->namelen,
2440 					name_loc->nameval, entry->flags))
2441 				continue;
2442 			args->index = probe;
2443 			return -EEXIST;
2444 		} else {
2445 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2446 			if (!xfs_attr_match(args, name_rmt->namelen,
2447 					name_rmt->name, entry->flags))
2448 				continue;
2449 			args->index = probe;
2450 			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2451 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2452 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2453 							args->dp->i_mount,
2454 							args->rmtvaluelen);
2455 			return -EEXIST;
2456 		}
2457 	}
2458 	args->index = probe;
2459 	return -ENOATTR;
2460 }
2461 
2462 /*
2463  * Get the value associated with an attribute name from a leaf attribute
2464  * list structure.
2465  *
2466  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2467  * lookup, we only return an error if the attribute does not exist or we can't
2468  * retrieve the value.
2469  */
2470 int
2471 xfs_attr3_leaf_getvalue(
2472 	struct xfs_buf		*bp,
2473 	struct xfs_da_args	*args)
2474 {
2475 	struct xfs_attr_leafblock *leaf;
2476 	struct xfs_attr3_icleaf_hdr ichdr;
2477 	struct xfs_attr_leaf_entry *entry;
2478 	struct xfs_attr_leaf_name_local *name_loc;
2479 	struct xfs_attr_leaf_name_remote *name_rmt;
2480 
2481 	leaf = bp->b_addr;
2482 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2483 	ASSERT(ichdr.count < args->geo->blksize / 8);
2484 	ASSERT(args->index < ichdr.count);
2485 
2486 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2487 	if (entry->flags & XFS_ATTR_LOCAL) {
2488 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2489 		ASSERT(name_loc->namelen == args->namelen);
2490 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2491 		return xfs_attr_copy_value(args,
2492 					&name_loc->nameval[args->namelen],
2493 					be16_to_cpu(name_loc->valuelen));
2494 	}
2495 
2496 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2497 	ASSERT(name_rmt->namelen == args->namelen);
2498 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2499 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2500 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2501 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2502 					       args->rmtvaluelen);
2503 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2504 }
2505 
2506 /*========================================================================
2507  * Utility routines.
2508  *========================================================================*/
2509 
2510 /*
2511  * Move the indicated entries from one leaf to another.
2512  * NOTE: this routine modifies both source and destination leaves.
2513  */
2514 /*ARGSUSED*/
2515 STATIC void
2516 xfs_attr3_leaf_moveents(
2517 	struct xfs_da_args		*args,
2518 	struct xfs_attr_leafblock	*leaf_s,
2519 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2520 	int				start_s,
2521 	struct xfs_attr_leafblock	*leaf_d,
2522 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2523 	int				start_d,
2524 	int				count)
2525 {
2526 	struct xfs_attr_leaf_entry	*entry_s;
2527 	struct xfs_attr_leaf_entry	*entry_d;
2528 	int				desti;
2529 	int				tmp;
2530 	int				i;
2531 
2532 	/*
2533 	 * Check for nothing to do.
2534 	 */
2535 	if (count == 0)
2536 		return;
2537 
2538 	/*
2539 	 * Set up environment.
2540 	 */
2541 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2542 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2543 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2544 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2545 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2546 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2547 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2548 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2549 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2550 
2551 	ASSERT(start_s < ichdr_s->count);
2552 	ASSERT(start_d <= ichdr_d->count);
2553 	ASSERT(count <= ichdr_s->count);
2554 
2555 
2556 	/*
2557 	 * Move the entries in the destination leaf up to make a hole?
2558 	 */
2559 	if (start_d < ichdr_d->count) {
2560 		tmp  = ichdr_d->count - start_d;
2561 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2562 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2563 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2564 		memmove(entry_d, entry_s, tmp);
2565 	}
2566 
2567 	/*
2568 	 * Copy all entry's in the same (sorted) order,
2569 	 * but allocate attribute info packed and in sequence.
2570 	 */
2571 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2572 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2573 	desti = start_d;
2574 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2575 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2576 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2577 #ifdef GROT
2578 		/*
2579 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2580 		 * may also need to change the insertion index.  Code turned
2581 		 * off for 6.2, should be revisited later.
2582 		 */
2583 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2584 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2585 			ichdr_s->usedbytes -= tmp;
2586 			ichdr_s->count -= 1;
2587 			entry_d--;	/* to compensate for ++ in loop hdr */
2588 			desti--;
2589 			if ((start_s + i) < offset)
2590 				result++;	/* insertion index adjustment */
2591 		} else {
2592 #endif /* GROT */
2593 			ichdr_d->firstused -= tmp;
2594 			/* both on-disk, don't endian flip twice */
2595 			entry_d->hashval = entry_s->hashval;
2596 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2597 			entry_d->flags = entry_s->flags;
2598 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2599 							<= args->geo->blksize);
2600 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2601 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2602 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2603 							<= args->geo->blksize);
2604 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2605 			ichdr_s->usedbytes -= tmp;
2606 			ichdr_d->usedbytes += tmp;
2607 			ichdr_s->count -= 1;
2608 			ichdr_d->count += 1;
2609 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2610 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2611 			ASSERT(ichdr_d->firstused >= tmp);
2612 #ifdef GROT
2613 		}
2614 #endif /* GROT */
2615 	}
2616 
2617 	/*
2618 	 * Zero out the entries we just copied.
2619 	 */
2620 	if (start_s == ichdr_s->count) {
2621 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2622 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2623 		ASSERT(((char *)entry_s + tmp) <=
2624 		       ((char *)leaf_s + args->geo->blksize));
2625 		memset(entry_s, 0, tmp);
2626 	} else {
2627 		/*
2628 		 * Move the remaining entries down to fill the hole,
2629 		 * then zero the entries at the top.
2630 		 */
2631 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2632 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2633 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2634 		memmove(entry_d, entry_s, tmp);
2635 
2636 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2637 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2638 		ASSERT(((char *)entry_s + tmp) <=
2639 		       ((char *)leaf_s + args->geo->blksize));
2640 		memset(entry_s, 0, tmp);
2641 	}
2642 
2643 	/*
2644 	 * Fill in the freemap information
2645 	 */
2646 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2647 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2648 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2649 	ichdr_d->freemap[1].base = 0;
2650 	ichdr_d->freemap[2].base = 0;
2651 	ichdr_d->freemap[1].size = 0;
2652 	ichdr_d->freemap[2].size = 0;
2653 	ichdr_s->holes = 1;	/* leaf may not be compact */
2654 }
2655 
2656 /*
2657  * Pick up the last hashvalue from a leaf block.
2658  */
2659 xfs_dahash_t
2660 xfs_attr_leaf_lasthash(
2661 	struct xfs_buf	*bp,
2662 	int		*count)
2663 {
2664 	struct xfs_attr3_icleaf_hdr ichdr;
2665 	struct xfs_attr_leaf_entry *entries;
2666 	struct xfs_mount *mp = bp->b_mount;
2667 
2668 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2669 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2670 	if (count)
2671 		*count = ichdr.count;
2672 	if (!ichdr.count)
2673 		return 0;
2674 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2675 }
2676 
2677 /*
2678  * Calculate the number of bytes used to store the indicated attribute
2679  * (whether local or remote only calculate bytes in this block).
2680  */
2681 STATIC int
2682 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2683 {
2684 	struct xfs_attr_leaf_entry *entries;
2685 	xfs_attr_leaf_name_local_t *name_loc;
2686 	xfs_attr_leaf_name_remote_t *name_rmt;
2687 	int size;
2688 
2689 	entries = xfs_attr3_leaf_entryp(leaf);
2690 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2691 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2692 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2693 						   be16_to_cpu(name_loc->valuelen));
2694 	} else {
2695 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2696 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2697 	}
2698 	return size;
2699 }
2700 
2701 /*
2702  * Calculate the number of bytes that would be required to store the new
2703  * attribute (whether local or remote only calculate bytes in this block).
2704  * This routine decides as a side effect whether the attribute will be
2705  * a "local" or a "remote" attribute.
2706  */
2707 int
2708 xfs_attr_leaf_newentsize(
2709 	struct xfs_da_args	*args,
2710 	int			*local)
2711 {
2712 	int			size;
2713 
2714 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2715 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2716 		if (local)
2717 			*local = 1;
2718 		return size;
2719 	}
2720 	if (local)
2721 		*local = 0;
2722 	return xfs_attr_leaf_entsize_remote(args->namelen);
2723 }
2724 
2725 
2726 /*========================================================================
2727  * Manage the INCOMPLETE flag in a leaf entry
2728  *========================================================================*/
2729 
2730 /*
2731  * Clear the INCOMPLETE flag on an entry in a leaf block.
2732  */
2733 int
2734 xfs_attr3_leaf_clearflag(
2735 	struct xfs_da_args	*args)
2736 {
2737 	struct xfs_attr_leafblock *leaf;
2738 	struct xfs_attr_leaf_entry *entry;
2739 	struct xfs_attr_leaf_name_remote *name_rmt;
2740 	struct xfs_buf		*bp;
2741 	int			error;
2742 #ifdef DEBUG
2743 	struct xfs_attr3_icleaf_hdr ichdr;
2744 	xfs_attr_leaf_name_local_t *name_loc;
2745 	int namelen;
2746 	char *name;
2747 #endif /* DEBUG */
2748 
2749 	trace_xfs_attr_leaf_clearflag(args);
2750 	/*
2751 	 * Set up the operation.
2752 	 */
2753 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2754 	if (error)
2755 		return error;
2756 
2757 	leaf = bp->b_addr;
2758 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2759 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2760 
2761 #ifdef DEBUG
2762 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2763 	ASSERT(args->index < ichdr.count);
2764 	ASSERT(args->index >= 0);
2765 
2766 	if (entry->flags & XFS_ATTR_LOCAL) {
2767 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2768 		namelen = name_loc->namelen;
2769 		name = (char *)name_loc->nameval;
2770 	} else {
2771 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2772 		namelen = name_rmt->namelen;
2773 		name = (char *)name_rmt->name;
2774 	}
2775 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2776 	ASSERT(namelen == args->namelen);
2777 	ASSERT(memcmp(name, args->name, namelen) == 0);
2778 #endif /* DEBUG */
2779 
2780 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2781 	xfs_trans_log_buf(args->trans, bp,
2782 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2783 
2784 	if (args->rmtblkno) {
2785 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2786 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2787 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2788 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2789 		xfs_trans_log_buf(args->trans, bp,
2790 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2791 	}
2792 
2793 	return 0;
2794 }
2795 
2796 /*
2797  * Set the INCOMPLETE flag on an entry in a leaf block.
2798  */
2799 int
2800 xfs_attr3_leaf_setflag(
2801 	struct xfs_da_args	*args)
2802 {
2803 	struct xfs_attr_leafblock *leaf;
2804 	struct xfs_attr_leaf_entry *entry;
2805 	struct xfs_attr_leaf_name_remote *name_rmt;
2806 	struct xfs_buf		*bp;
2807 	int error;
2808 #ifdef DEBUG
2809 	struct xfs_attr3_icleaf_hdr ichdr;
2810 #endif
2811 
2812 	trace_xfs_attr_leaf_setflag(args);
2813 
2814 	/*
2815 	 * Set up the operation.
2816 	 */
2817 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2818 	if (error)
2819 		return error;
2820 
2821 	leaf = bp->b_addr;
2822 #ifdef DEBUG
2823 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2824 	ASSERT(args->index < ichdr.count);
2825 	ASSERT(args->index >= 0);
2826 #endif
2827 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2828 
2829 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2830 	entry->flags |= XFS_ATTR_INCOMPLETE;
2831 	xfs_trans_log_buf(args->trans, bp,
2832 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2833 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2834 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2835 		name_rmt->valueblk = 0;
2836 		name_rmt->valuelen = 0;
2837 		xfs_trans_log_buf(args->trans, bp,
2838 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2839 	}
2840 
2841 	return 0;
2842 }
2843 
2844 /*
2845  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2846  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2847  * entry given by args->blkno2/index2.
2848  *
2849  * Note that they could be in different blocks, or in the same block.
2850  */
2851 int
2852 xfs_attr3_leaf_flipflags(
2853 	struct xfs_da_args	*args)
2854 {
2855 	struct xfs_attr_leafblock *leaf1;
2856 	struct xfs_attr_leafblock *leaf2;
2857 	struct xfs_attr_leaf_entry *entry1;
2858 	struct xfs_attr_leaf_entry *entry2;
2859 	struct xfs_attr_leaf_name_remote *name_rmt;
2860 	struct xfs_buf		*bp1;
2861 	struct xfs_buf		*bp2;
2862 	int error;
2863 #ifdef DEBUG
2864 	struct xfs_attr3_icleaf_hdr ichdr1;
2865 	struct xfs_attr3_icleaf_hdr ichdr2;
2866 	xfs_attr_leaf_name_local_t *name_loc;
2867 	int namelen1, namelen2;
2868 	char *name1, *name2;
2869 #endif /* DEBUG */
2870 
2871 	trace_xfs_attr_leaf_flipflags(args);
2872 
2873 	/*
2874 	 * Read the block containing the "old" attr
2875 	 */
2876 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2877 	if (error)
2878 		return error;
2879 
2880 	/*
2881 	 * Read the block containing the "new" attr, if it is different
2882 	 */
2883 	if (args->blkno2 != args->blkno) {
2884 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2885 					   &bp2);
2886 		if (error)
2887 			return error;
2888 	} else {
2889 		bp2 = bp1;
2890 	}
2891 
2892 	leaf1 = bp1->b_addr;
2893 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2894 
2895 	leaf2 = bp2->b_addr;
2896 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2897 
2898 #ifdef DEBUG
2899 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2900 	ASSERT(args->index < ichdr1.count);
2901 	ASSERT(args->index >= 0);
2902 
2903 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2904 	ASSERT(args->index2 < ichdr2.count);
2905 	ASSERT(args->index2 >= 0);
2906 
2907 	if (entry1->flags & XFS_ATTR_LOCAL) {
2908 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2909 		namelen1 = name_loc->namelen;
2910 		name1 = (char *)name_loc->nameval;
2911 	} else {
2912 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2913 		namelen1 = name_rmt->namelen;
2914 		name1 = (char *)name_rmt->name;
2915 	}
2916 	if (entry2->flags & XFS_ATTR_LOCAL) {
2917 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2918 		namelen2 = name_loc->namelen;
2919 		name2 = (char *)name_loc->nameval;
2920 	} else {
2921 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2922 		namelen2 = name_rmt->namelen;
2923 		name2 = (char *)name_rmt->name;
2924 	}
2925 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2926 	ASSERT(namelen1 == namelen2);
2927 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2928 #endif /* DEBUG */
2929 
2930 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2931 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2932 
2933 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2934 	xfs_trans_log_buf(args->trans, bp1,
2935 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2936 	if (args->rmtblkno) {
2937 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2938 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2939 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2940 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2941 		xfs_trans_log_buf(args->trans, bp1,
2942 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2943 	}
2944 
2945 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2946 	xfs_trans_log_buf(args->trans, bp2,
2947 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2948 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2949 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2950 		name_rmt->valueblk = 0;
2951 		name_rmt->valuelen = 0;
2952 		xfs_trans_log_buf(args->trans, bp2,
2953 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2954 	}
2955 
2956 	return 0;
2957 }
2958