xref: /openbmc/linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision f95fc014)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_da_format.h"
16 #include "xfs_da_btree.h"
17 #include "xfs_inode.h"
18 #include "xfs_trans.h"
19 #include "xfs_bmap_btree.h"
20 #include "xfs_bmap.h"
21 #include "xfs_attr_sf.h"
22 #include "xfs_attr_remote.h"
23 #include "xfs_attr.h"
24 #include "xfs_attr_leaf.h"
25 #include "xfs_error.h"
26 #include "xfs_trace.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_dir2.h"
29 #include "xfs_log.h"
30 
31 
32 /*
33  * xfs_attr_leaf.c
34  *
35  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
36  */
37 
38 /*========================================================================
39  * Function prototypes for the kernel.
40  *========================================================================*/
41 
42 /*
43  * Routines used for growing the Btree.
44  */
45 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
46 				 xfs_dablk_t which_block, struct xfs_buf **bpp);
47 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
48 				   struct xfs_attr3_icleaf_hdr *ichdr,
49 				   struct xfs_da_args *args, int freemap_index);
50 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
51 				   struct xfs_attr3_icleaf_hdr *ichdr,
52 				   struct xfs_buf *leaf_buffer);
53 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
54 						   xfs_da_state_blk_t *blk1,
55 						   xfs_da_state_blk_t *blk2);
56 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
57 			xfs_da_state_blk_t *leaf_blk_1,
58 			struct xfs_attr3_icleaf_hdr *ichdr1,
59 			xfs_da_state_blk_t *leaf_blk_2,
60 			struct xfs_attr3_icleaf_hdr *ichdr2,
61 			int *number_entries_in_blk1,
62 			int *number_usedbytes_in_blk1);
63 
64 /*
65  * Utility routines.
66  */
67 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
68 			struct xfs_attr_leafblock *src_leaf,
69 			struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
70 			struct xfs_attr_leafblock *dst_leaf,
71 			struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
72 			int move_count);
73 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
74 
75 /*
76  * attr3 block 'firstused' conversion helpers.
77  *
78  * firstused refers to the offset of the first used byte of the nameval region
79  * of an attr leaf block. The region starts at the tail of the block and expands
80  * backwards towards the middle. As such, firstused is initialized to the block
81  * size for an empty leaf block and is reduced from there.
82  *
83  * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
84  * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
85  * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
86  * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
87  * the attr block size. The following helpers manage the conversion between the
88  * in-core and on-disk formats.
89  */
90 
91 static void
92 xfs_attr3_leaf_firstused_from_disk(
93 	struct xfs_da_geometry		*geo,
94 	struct xfs_attr3_icleaf_hdr	*to,
95 	struct xfs_attr_leafblock	*from)
96 {
97 	struct xfs_attr3_leaf_hdr	*hdr3;
98 
99 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
100 		hdr3 = (struct xfs_attr3_leaf_hdr *) from;
101 		to->firstused = be16_to_cpu(hdr3->firstused);
102 	} else {
103 		to->firstused = be16_to_cpu(from->hdr.firstused);
104 	}
105 
106 	/*
107 	 * Convert from the magic fsb size value to actual blocksize. This
108 	 * should only occur for empty blocks when the block size overflows
109 	 * 16-bits.
110 	 */
111 	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
112 		ASSERT(!to->count && !to->usedbytes);
113 		ASSERT(geo->blksize > USHRT_MAX);
114 		to->firstused = geo->blksize;
115 	}
116 }
117 
118 static void
119 xfs_attr3_leaf_firstused_to_disk(
120 	struct xfs_da_geometry		*geo,
121 	struct xfs_attr_leafblock	*to,
122 	struct xfs_attr3_icleaf_hdr	*from)
123 {
124 	struct xfs_attr3_leaf_hdr	*hdr3;
125 	uint32_t			firstused;
126 
127 	/* magic value should only be seen on disk */
128 	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
129 
130 	/*
131 	 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
132 	 * value. This only overflows at the max supported value of 64k. Use the
133 	 * magic on-disk value to represent block size in this case.
134 	 */
135 	firstused = from->firstused;
136 	if (firstused > USHRT_MAX) {
137 		ASSERT(from->firstused == geo->blksize);
138 		firstused = XFS_ATTR3_LEAF_NULLOFF;
139 	}
140 
141 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
142 		hdr3 = (struct xfs_attr3_leaf_hdr *) to;
143 		hdr3->firstused = cpu_to_be16(firstused);
144 	} else {
145 		to->hdr.firstused = cpu_to_be16(firstused);
146 	}
147 }
148 
149 void
150 xfs_attr3_leaf_hdr_from_disk(
151 	struct xfs_da_geometry		*geo,
152 	struct xfs_attr3_icleaf_hdr	*to,
153 	struct xfs_attr_leafblock	*from)
154 {
155 	int	i;
156 
157 	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
158 	       from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
159 
160 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
161 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
162 
163 		to->forw = be32_to_cpu(hdr3->info.hdr.forw);
164 		to->back = be32_to_cpu(hdr3->info.hdr.back);
165 		to->magic = be16_to_cpu(hdr3->info.hdr.magic);
166 		to->count = be16_to_cpu(hdr3->count);
167 		to->usedbytes = be16_to_cpu(hdr3->usedbytes);
168 		xfs_attr3_leaf_firstused_from_disk(geo, to, from);
169 		to->holes = hdr3->holes;
170 
171 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
172 			to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
173 			to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
174 		}
175 		return;
176 	}
177 	to->forw = be32_to_cpu(from->hdr.info.forw);
178 	to->back = be32_to_cpu(from->hdr.info.back);
179 	to->magic = be16_to_cpu(from->hdr.info.magic);
180 	to->count = be16_to_cpu(from->hdr.count);
181 	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
182 	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
183 	to->holes = from->hdr.holes;
184 
185 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
186 		to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
187 		to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
188 	}
189 }
190 
191 void
192 xfs_attr3_leaf_hdr_to_disk(
193 	struct xfs_da_geometry		*geo,
194 	struct xfs_attr_leafblock	*to,
195 	struct xfs_attr3_icleaf_hdr	*from)
196 {
197 	int				i;
198 
199 	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
200 	       from->magic == XFS_ATTR3_LEAF_MAGIC);
201 
202 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
203 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
204 
205 		hdr3->info.hdr.forw = cpu_to_be32(from->forw);
206 		hdr3->info.hdr.back = cpu_to_be32(from->back);
207 		hdr3->info.hdr.magic = cpu_to_be16(from->magic);
208 		hdr3->count = cpu_to_be16(from->count);
209 		hdr3->usedbytes = cpu_to_be16(from->usedbytes);
210 		xfs_attr3_leaf_firstused_to_disk(geo, to, from);
211 		hdr3->holes = from->holes;
212 		hdr3->pad1 = 0;
213 
214 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
215 			hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
216 			hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
217 		}
218 		return;
219 	}
220 	to->hdr.info.forw = cpu_to_be32(from->forw);
221 	to->hdr.info.back = cpu_to_be32(from->back);
222 	to->hdr.info.magic = cpu_to_be16(from->magic);
223 	to->hdr.count = cpu_to_be16(from->count);
224 	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
225 	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
226 	to->hdr.holes = from->holes;
227 	to->hdr.pad1 = 0;
228 
229 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
230 		to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
231 		to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
232 	}
233 }
234 
235 static xfs_failaddr_t
236 xfs_attr3_leaf_verify_entry(
237 	struct xfs_mount			*mp,
238 	char					*buf_end,
239 	struct xfs_attr_leafblock		*leaf,
240 	struct xfs_attr3_icleaf_hdr		*leafhdr,
241 	struct xfs_attr_leaf_entry		*ent,
242 	int					idx,
243 	__u32					*last_hashval)
244 {
245 	struct xfs_attr_leaf_name_local		*lentry;
246 	struct xfs_attr_leaf_name_remote	*rentry;
247 	char					*name_end;
248 	unsigned int				nameidx;
249 	unsigned int				namesize;
250 	__u32					hashval;
251 
252 	/* hash order check */
253 	hashval = be32_to_cpu(ent->hashval);
254 	if (hashval < *last_hashval)
255 		return __this_address;
256 	*last_hashval = hashval;
257 
258 	nameidx = be16_to_cpu(ent->nameidx);
259 	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
260 		return __this_address;
261 
262 	/*
263 	 * Check the name information.  The namelen fields are u8 so we can't
264 	 * possibly exceed the maximum name length of 255 bytes.
265 	 */
266 	if (ent->flags & XFS_ATTR_LOCAL) {
267 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
268 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
269 				be16_to_cpu(lentry->valuelen));
270 		name_end = (char *)lentry + namesize;
271 		if (lentry->namelen == 0)
272 			return __this_address;
273 	} else {
274 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
275 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
276 		name_end = (char *)rentry + namesize;
277 		if (rentry->namelen == 0)
278 			return __this_address;
279 		if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
280 		    rentry->valueblk == 0)
281 			return __this_address;
282 	}
283 
284 	if (name_end > buf_end)
285 		return __this_address;
286 
287 	return NULL;
288 }
289 
290 static xfs_failaddr_t
291 xfs_attr3_leaf_verify(
292 	struct xfs_buf			*bp)
293 {
294 	struct xfs_attr3_icleaf_hdr	ichdr;
295 	struct xfs_mount		*mp = bp->b_mount;
296 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
297 	struct xfs_attr_leaf_entry	*entries;
298 	struct xfs_attr_leaf_entry	*ent;
299 	char				*buf_end;
300 	uint32_t			end;	/* must be 32bit - see below */
301 	__u32				last_hashval = 0;
302 	int				i;
303 	xfs_failaddr_t			fa;
304 
305 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
306 
307 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
308 	if (fa)
309 		return fa;
310 
311 	/*
312 	 * firstused is the block offset of the first name info structure.
313 	 * Make sure it doesn't go off the block or crash into the header.
314 	 */
315 	if (ichdr.firstused > mp->m_attr_geo->blksize)
316 		return __this_address;
317 	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
318 		return __this_address;
319 
320 	/* Make sure the entries array doesn't crash into the name info. */
321 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
322 	if ((char *)&entries[ichdr.count] >
323 	    (char *)bp->b_addr + ichdr.firstused)
324 		return __this_address;
325 
326 	/*
327 	 * NOTE: This verifier historically failed empty leaf buffers because
328 	 * we expect the fork to be in another format. Empty attr fork format
329 	 * conversions are possible during xattr set, however, and format
330 	 * conversion is not atomic with the xattr set that triggers it. We
331 	 * cannot assume leaf blocks are non-empty until that is addressed.
332 	*/
333 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
334 	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
335 		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
336 				ent, i, &last_hashval);
337 		if (fa)
338 			return fa;
339 	}
340 
341 	/*
342 	 * Quickly check the freemap information.  Attribute data has to be
343 	 * aligned to 4-byte boundaries, and likewise for the free space.
344 	 *
345 	 * Note that for 64k block size filesystems, the freemap entries cannot
346 	 * overflow as they are only be16 fields. However, when checking end
347 	 * pointer of the freemap, we have to be careful to detect overflows and
348 	 * so use uint32_t for those checks.
349 	 */
350 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
351 		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
352 			return __this_address;
353 		if (ichdr.freemap[i].base & 0x3)
354 			return __this_address;
355 		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
356 			return __this_address;
357 		if (ichdr.freemap[i].size & 0x3)
358 			return __this_address;
359 
360 		/* be care of 16 bit overflows here */
361 		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
362 		if (end < ichdr.freemap[i].base)
363 			return __this_address;
364 		if (end > mp->m_attr_geo->blksize)
365 			return __this_address;
366 	}
367 
368 	return NULL;
369 }
370 
371 static void
372 xfs_attr3_leaf_write_verify(
373 	struct xfs_buf	*bp)
374 {
375 	struct xfs_mount	*mp = bp->b_mount;
376 	struct xfs_buf_log_item	*bip = bp->b_log_item;
377 	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
378 	xfs_failaddr_t		fa;
379 
380 	fa = xfs_attr3_leaf_verify(bp);
381 	if (fa) {
382 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
383 		return;
384 	}
385 
386 	if (!xfs_sb_version_hascrc(&mp->m_sb))
387 		return;
388 
389 	if (bip)
390 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
391 
392 	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
393 }
394 
395 /*
396  * leaf/node format detection on trees is sketchy, so a node read can be done on
397  * leaf level blocks when detection identifies the tree as a node format tree
398  * incorrectly. In this case, we need to swap the verifier to match the correct
399  * format of the block being read.
400  */
401 static void
402 xfs_attr3_leaf_read_verify(
403 	struct xfs_buf		*bp)
404 {
405 	struct xfs_mount	*mp = bp->b_mount;
406 	xfs_failaddr_t		fa;
407 
408 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
409 	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
410 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
411 	else {
412 		fa = xfs_attr3_leaf_verify(bp);
413 		if (fa)
414 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
415 	}
416 }
417 
418 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
419 	.name = "xfs_attr3_leaf",
420 	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
421 		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
422 	.verify_read = xfs_attr3_leaf_read_verify,
423 	.verify_write = xfs_attr3_leaf_write_verify,
424 	.verify_struct = xfs_attr3_leaf_verify,
425 };
426 
427 int
428 xfs_attr3_leaf_read(
429 	struct xfs_trans	*tp,
430 	struct xfs_inode	*dp,
431 	xfs_dablk_t		bno,
432 	struct xfs_buf		**bpp)
433 {
434 	int			err;
435 
436 	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
437 			&xfs_attr3_leaf_buf_ops);
438 	if (!err && tp && *bpp)
439 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
440 	return err;
441 }
442 
443 /*========================================================================
444  * Namespace helper routines
445  *========================================================================*/
446 
447 static bool
448 xfs_attr_match(
449 	struct xfs_da_args	*args,
450 	uint8_t			namelen,
451 	unsigned char		*name,
452 	int			flags)
453 {
454 	if (args->namelen != namelen)
455 		return false;
456 	if (memcmp(args->name, name, namelen) != 0)
457 		return false;
458 	/*
459 	 * If we are looking for incomplete entries, show only those, else only
460 	 * show complete entries.
461 	 */
462 	if (args->attr_filter !=
463 	    (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
464 		return false;
465 	return true;
466 }
467 
468 static int
469 xfs_attr_copy_value(
470 	struct xfs_da_args	*args,
471 	unsigned char		*value,
472 	int			valuelen)
473 {
474 	/*
475 	 * No copy if all we have to do is get the length
476 	 */
477 	if (!args->valuelen) {
478 		args->valuelen = valuelen;
479 		return 0;
480 	}
481 
482 	/*
483 	 * No copy if the length of the existing buffer is too small
484 	 */
485 	if (args->valuelen < valuelen) {
486 		args->valuelen = valuelen;
487 		return -ERANGE;
488 	}
489 
490 	if (!args->value) {
491 		args->value = kmem_alloc_large(valuelen, KM_NOLOCKDEP);
492 		if (!args->value)
493 			return -ENOMEM;
494 	}
495 	args->valuelen = valuelen;
496 
497 	/* remote block xattr requires IO for copy-in */
498 	if (args->rmtblkno)
499 		return xfs_attr_rmtval_get(args);
500 
501 	/*
502 	 * This is to prevent a GCC warning because the remote xattr case
503 	 * doesn't have a value to pass in. In that case, we never reach here,
504 	 * but GCC can't work that out and so throws a "passing NULL to
505 	 * memcpy" warning.
506 	 */
507 	if (!value)
508 		return -EINVAL;
509 	memcpy(args->value, value, valuelen);
510 	return 0;
511 }
512 
513 /*========================================================================
514  * External routines when attribute fork size < XFS_LITINO(mp).
515  *========================================================================*/
516 
517 /*
518  * Query whether the requested number of additional bytes of extended
519  * attribute space will be able to fit inline.
520  *
521  * Returns zero if not, else the di_forkoff fork offset to be used in the
522  * literal area for attribute data once the new bytes have been added.
523  *
524  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
525  * special case for dev/uuid inodes, they have fixed size data forks.
526  */
527 int
528 xfs_attr_shortform_bytesfit(
529 	struct xfs_inode	*dp,
530 	int			bytes)
531 {
532 	struct xfs_mount	*mp = dp->i_mount;
533 	int64_t			dsize;
534 	int			minforkoff;
535 	int			maxforkoff;
536 	int			offset;
537 
538 	/* rounded down */
539 	offset = (XFS_LITINO(mp) - bytes) >> 3;
540 
541 	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
542 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
543 		return (offset >= minforkoff) ? minforkoff : 0;
544 	}
545 
546 	/*
547 	 * If the requested numbers of bytes is smaller or equal to the
548 	 * current attribute fork size we can always proceed.
549 	 *
550 	 * Note that if_bytes in the data fork might actually be larger than
551 	 * the current data fork size is due to delalloc extents. In that
552 	 * case either the extent count will go down when they are converted
553 	 * to real extents, or the delalloc conversion will take care of the
554 	 * literal area rebalancing.
555 	 */
556 	if (bytes <= XFS_IFORK_ASIZE(dp))
557 		return dp->i_d.di_forkoff;
558 
559 	/*
560 	 * For attr2 we can try to move the forkoff if there is space in the
561 	 * literal area, but for the old format we are done if there is no
562 	 * space in the fixed attribute fork.
563 	 */
564 	if (!(mp->m_flags & XFS_MOUNT_ATTR2))
565 		return 0;
566 
567 	dsize = dp->i_df.if_bytes;
568 
569 	switch (dp->i_df.if_format) {
570 	case XFS_DINODE_FMT_EXTENTS:
571 		/*
572 		 * If there is no attr fork and the data fork is extents,
573 		 * determine if creating the default attr fork will result
574 		 * in the extents form migrating to btree. If so, the
575 		 * minimum offset only needs to be the space required for
576 		 * the btree root.
577 		 */
578 		if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
579 		    xfs_default_attroffset(dp))
580 			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
581 		break;
582 	case XFS_DINODE_FMT_BTREE:
583 		/*
584 		 * If we have a data btree then keep forkoff if we have one,
585 		 * otherwise we are adding a new attr, so then we set
586 		 * minforkoff to where the btree root can finish so we have
587 		 * plenty of room for attrs
588 		 */
589 		if (dp->i_d.di_forkoff) {
590 			if (offset < dp->i_d.di_forkoff)
591 				return 0;
592 			return dp->i_d.di_forkoff;
593 		}
594 		dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
595 		break;
596 	}
597 
598 	/*
599 	 * A data fork btree root must have space for at least
600 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
601 	 */
602 	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
603 	minforkoff = roundup(minforkoff, 8) >> 3;
604 
605 	/* attr fork btree root can have at least this many key/ptr pairs */
606 	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
607 	maxforkoff = maxforkoff >> 3;	/* rounded down */
608 
609 	if (offset >= maxforkoff)
610 		return maxforkoff;
611 	if (offset >= minforkoff)
612 		return offset;
613 	return 0;
614 }
615 
616 /*
617  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
618  */
619 STATIC void
620 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
621 {
622 	if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
623 	    !(xfs_sb_version_hasattr2(&mp->m_sb))) {
624 		spin_lock(&mp->m_sb_lock);
625 		if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
626 			xfs_sb_version_addattr2(&mp->m_sb);
627 			spin_unlock(&mp->m_sb_lock);
628 			xfs_log_sb(tp);
629 		} else
630 			spin_unlock(&mp->m_sb_lock);
631 	}
632 }
633 
634 /*
635  * Create the initial contents of a shortform attribute list.
636  */
637 void
638 xfs_attr_shortform_create(
639 	struct xfs_da_args	*args)
640 {
641 	struct xfs_inode	*dp = args->dp;
642 	struct xfs_ifork	*ifp = dp->i_afp;
643 	struct xfs_attr_sf_hdr	*hdr;
644 
645 	trace_xfs_attr_sf_create(args);
646 
647 	ASSERT(ifp->if_bytes == 0);
648 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS) {
649 		ifp->if_flags &= ~XFS_IFEXTENTS;	/* just in case */
650 		ifp->if_format = XFS_DINODE_FMT_LOCAL;
651 		ifp->if_flags |= XFS_IFINLINE;
652 	} else {
653 		ASSERT(ifp->if_flags & XFS_IFINLINE);
654 	}
655 	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
656 	hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
657 	memset(hdr, 0, sizeof(*hdr));
658 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
659 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
660 }
661 
662 /*
663  * Return -EEXIST if attr is found, or -ENOATTR if not
664  * args:  args containing attribute name and namelen
665  * sfep:  If not null, pointer will be set to the last attr entry found on
666 	  -EEXIST.  On -ENOATTR pointer is left at the last entry in the list
667  * basep: If not null, pointer is set to the byte offset of the entry in the
668  *	  list on -EEXIST.  On -ENOATTR, pointer is left at the byte offset of
669  *	  the last entry in the list
670  */
671 int
672 xfs_attr_sf_findname(
673 	struct xfs_da_args	 *args,
674 	struct xfs_attr_sf_entry **sfep,
675 	unsigned int		 *basep)
676 {
677 	struct xfs_attr_shortform *sf;
678 	struct xfs_attr_sf_entry *sfe;
679 	unsigned int		base = sizeof(struct xfs_attr_sf_hdr);
680 	int			size = 0;
681 	int			end;
682 	int			i;
683 
684 	sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
685 	sfe = &sf->list[0];
686 	end = sf->hdr.count;
687 	for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
688 			     base += size, i++) {
689 		size = XFS_ATTR_SF_ENTSIZE(sfe);
690 		if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
691 				    sfe->flags))
692 			continue;
693 		break;
694 	}
695 
696 	if (sfep != NULL)
697 		*sfep = sfe;
698 
699 	if (basep != NULL)
700 		*basep = base;
701 
702 	if (i == end)
703 		return -ENOATTR;
704 	return -EEXIST;
705 }
706 
707 /*
708  * Add a name/value pair to the shortform attribute list.
709  * Overflow from the inode has already been checked for.
710  */
711 void
712 xfs_attr_shortform_add(
713 	struct xfs_da_args		*args,
714 	int				forkoff)
715 {
716 	struct xfs_attr_shortform	*sf;
717 	struct xfs_attr_sf_entry	*sfe;
718 	int				offset, size;
719 	struct xfs_mount		*mp;
720 	struct xfs_inode		*dp;
721 	struct xfs_ifork		*ifp;
722 
723 	trace_xfs_attr_sf_add(args);
724 
725 	dp = args->dp;
726 	mp = dp->i_mount;
727 	dp->i_d.di_forkoff = forkoff;
728 
729 	ifp = dp->i_afp;
730 	ASSERT(ifp->if_flags & XFS_IFINLINE);
731 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
732 	if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
733 		ASSERT(0);
734 
735 	offset = (char *)sfe - (char *)sf;
736 	size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
737 	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
738 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
739 	sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
740 
741 	sfe->namelen = args->namelen;
742 	sfe->valuelen = args->valuelen;
743 	sfe->flags = args->attr_filter;
744 	memcpy(sfe->nameval, args->name, args->namelen);
745 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
746 	sf->hdr.count++;
747 	be16_add_cpu(&sf->hdr.totsize, size);
748 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
749 
750 	xfs_sbversion_add_attr2(mp, args->trans);
751 }
752 
753 /*
754  * After the last attribute is removed revert to original inode format,
755  * making all literal area available to the data fork once more.
756  */
757 void
758 xfs_attr_fork_remove(
759 	struct xfs_inode	*ip,
760 	struct xfs_trans	*tp)
761 {
762 	ASSERT(ip->i_afp->if_nextents == 0);
763 
764 	xfs_idestroy_fork(ip->i_afp);
765 	kmem_cache_free(xfs_ifork_zone, ip->i_afp);
766 	ip->i_afp = NULL;
767 	ip->i_d.di_forkoff = 0;
768 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
769 }
770 
771 /*
772  * Remove an attribute from the shortform attribute list structure.
773  */
774 int
775 xfs_attr_shortform_remove(
776 	struct xfs_da_args		*args)
777 {
778 	struct xfs_attr_shortform	*sf;
779 	struct xfs_attr_sf_entry	*sfe;
780 	int				size = 0, end, totsize;
781 	unsigned int			base;
782 	struct xfs_mount		*mp;
783 	struct xfs_inode		*dp;
784 	int				error;
785 
786 	trace_xfs_attr_sf_remove(args);
787 
788 	dp = args->dp;
789 	mp = dp->i_mount;
790 	sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
791 
792 	error = xfs_attr_sf_findname(args, &sfe, &base);
793 	if (error != -EEXIST)
794 		return error;
795 	size = XFS_ATTR_SF_ENTSIZE(sfe);
796 
797 	/*
798 	 * Fix up the attribute fork data, covering the hole
799 	 */
800 	end = base + size;
801 	totsize = be16_to_cpu(sf->hdr.totsize);
802 	if (end != totsize)
803 		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
804 	sf->hdr.count--;
805 	be16_add_cpu(&sf->hdr.totsize, -size);
806 
807 	/*
808 	 * Fix up the start offset of the attribute fork
809 	 */
810 	totsize -= size;
811 	if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
812 	    (mp->m_flags & XFS_MOUNT_ATTR2) &&
813 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
814 	    !(args->op_flags & XFS_DA_OP_ADDNAME)) {
815 		xfs_attr_fork_remove(dp, args->trans);
816 	} else {
817 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
818 		dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
819 		ASSERT(dp->i_d.di_forkoff);
820 		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
821 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
822 				!(mp->m_flags & XFS_MOUNT_ATTR2) ||
823 				dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
824 		xfs_trans_log_inode(args->trans, dp,
825 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
826 	}
827 
828 	xfs_sbversion_add_attr2(mp, args->trans);
829 
830 	return 0;
831 }
832 
833 /*
834  * Look up a name in a shortform attribute list structure.
835  */
836 /*ARGSUSED*/
837 int
838 xfs_attr_shortform_lookup(xfs_da_args_t *args)
839 {
840 	xfs_attr_shortform_t *sf;
841 	xfs_attr_sf_entry_t *sfe;
842 	int i;
843 	struct xfs_ifork *ifp;
844 
845 	trace_xfs_attr_sf_lookup(args);
846 
847 	ifp = args->dp->i_afp;
848 	ASSERT(ifp->if_flags & XFS_IFINLINE);
849 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
850 	sfe = &sf->list[0];
851 	for (i = 0; i < sf->hdr.count;
852 				sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
853 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
854 				sfe->flags))
855 			return -EEXIST;
856 	}
857 	return -ENOATTR;
858 }
859 
860 /*
861  * Retrieve the attribute value and length.
862  *
863  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
864  * lookup, we only return an error if the attribute does not exist or we can't
865  * retrieve the value.
866  */
867 int
868 xfs_attr_shortform_getvalue(
869 	struct xfs_da_args	*args)
870 {
871 	struct xfs_attr_shortform *sf;
872 	struct xfs_attr_sf_entry *sfe;
873 	int			i;
874 
875 	ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
876 	sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
877 	sfe = &sf->list[0];
878 	for (i = 0; i < sf->hdr.count;
879 				sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
880 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
881 				sfe->flags))
882 			return xfs_attr_copy_value(args,
883 				&sfe->nameval[args->namelen], sfe->valuelen);
884 	}
885 	return -ENOATTR;
886 }
887 
888 /*
889  * Convert from using the shortform to the leaf.  On success, return the
890  * buffer so that we can keep it locked until we're totally done with it.
891  */
892 int
893 xfs_attr_shortform_to_leaf(
894 	struct xfs_da_args		*args,
895 	struct xfs_buf			**leaf_bp)
896 {
897 	struct xfs_inode		*dp;
898 	struct xfs_attr_shortform	*sf;
899 	struct xfs_attr_sf_entry	*sfe;
900 	struct xfs_da_args		nargs;
901 	char				*tmpbuffer;
902 	int				error, i, size;
903 	xfs_dablk_t			blkno;
904 	struct xfs_buf			*bp;
905 	struct xfs_ifork		*ifp;
906 
907 	trace_xfs_attr_sf_to_leaf(args);
908 
909 	dp = args->dp;
910 	ifp = dp->i_afp;
911 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
912 	size = be16_to_cpu(sf->hdr.totsize);
913 	tmpbuffer = kmem_alloc(size, 0);
914 	ASSERT(tmpbuffer != NULL);
915 	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
916 	sf = (xfs_attr_shortform_t *)tmpbuffer;
917 
918 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
919 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
920 
921 	bp = NULL;
922 	error = xfs_da_grow_inode(args, &blkno);
923 	if (error)
924 		goto out;
925 
926 	ASSERT(blkno == 0);
927 	error = xfs_attr3_leaf_create(args, blkno, &bp);
928 	if (error)
929 		goto out;
930 
931 	memset((char *)&nargs, 0, sizeof(nargs));
932 	nargs.dp = dp;
933 	nargs.geo = args->geo;
934 	nargs.total = args->total;
935 	nargs.whichfork = XFS_ATTR_FORK;
936 	nargs.trans = args->trans;
937 	nargs.op_flags = XFS_DA_OP_OKNOENT;
938 
939 	sfe = &sf->list[0];
940 	for (i = 0; i < sf->hdr.count; i++) {
941 		nargs.name = sfe->nameval;
942 		nargs.namelen = sfe->namelen;
943 		nargs.value = &sfe->nameval[nargs.namelen];
944 		nargs.valuelen = sfe->valuelen;
945 		nargs.hashval = xfs_da_hashname(sfe->nameval,
946 						sfe->namelen);
947 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
948 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
949 		ASSERT(error == -ENOATTR);
950 		error = xfs_attr3_leaf_add(bp, &nargs);
951 		ASSERT(error != -ENOSPC);
952 		if (error)
953 			goto out;
954 		sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
955 	}
956 	error = 0;
957 	*leaf_bp = bp;
958 out:
959 	kmem_free(tmpbuffer);
960 	return error;
961 }
962 
963 /*
964  * Check a leaf attribute block to see if all the entries would fit into
965  * a shortform attribute list.
966  */
967 int
968 xfs_attr_shortform_allfit(
969 	struct xfs_buf		*bp,
970 	struct xfs_inode	*dp)
971 {
972 	struct xfs_attr_leafblock *leaf;
973 	struct xfs_attr_leaf_entry *entry;
974 	xfs_attr_leaf_name_local_t *name_loc;
975 	struct xfs_attr3_icleaf_hdr leafhdr;
976 	int			bytes;
977 	int			i;
978 	struct xfs_mount	*mp = bp->b_mount;
979 
980 	leaf = bp->b_addr;
981 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
982 	entry = xfs_attr3_leaf_entryp(leaf);
983 
984 	bytes = sizeof(struct xfs_attr_sf_hdr);
985 	for (i = 0; i < leafhdr.count; entry++, i++) {
986 		if (entry->flags & XFS_ATTR_INCOMPLETE)
987 			continue;		/* don't copy partial entries */
988 		if (!(entry->flags & XFS_ATTR_LOCAL))
989 			return 0;
990 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
991 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
992 			return 0;
993 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
994 			return 0;
995 		bytes += sizeof(struct xfs_attr_sf_entry) - 1
996 				+ name_loc->namelen
997 				+ be16_to_cpu(name_loc->valuelen);
998 	}
999 	if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
1000 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1001 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
1002 		return -1;
1003 	return xfs_attr_shortform_bytesfit(dp, bytes);
1004 }
1005 
1006 /* Verify the consistency of an inline attribute fork. */
1007 xfs_failaddr_t
1008 xfs_attr_shortform_verify(
1009 	struct xfs_inode		*ip)
1010 {
1011 	struct xfs_attr_shortform	*sfp;
1012 	struct xfs_attr_sf_entry	*sfep;
1013 	struct xfs_attr_sf_entry	*next_sfep;
1014 	char				*endp;
1015 	struct xfs_ifork		*ifp;
1016 	int				i;
1017 	int64_t				size;
1018 
1019 	ASSERT(ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
1020 	ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
1021 	sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1022 	size = ifp->if_bytes;
1023 
1024 	/*
1025 	 * Give up if the attribute is way too short.
1026 	 */
1027 	if (size < sizeof(struct xfs_attr_sf_hdr))
1028 		return __this_address;
1029 
1030 	endp = (char *)sfp + size;
1031 
1032 	/* Check all reported entries */
1033 	sfep = &sfp->list[0];
1034 	for (i = 0; i < sfp->hdr.count; i++) {
1035 		/*
1036 		 * struct xfs_attr_sf_entry has a variable length.
1037 		 * Check the fixed-offset parts of the structure are
1038 		 * within the data buffer.
1039 		 * xfs_attr_sf_entry is defined with a 1-byte variable
1040 		 * array at the end, so we must subtract that off.
1041 		 */
1042 		if (((char *)sfep + sizeof(*sfep) - 1) >= endp)
1043 			return __this_address;
1044 
1045 		/* Don't allow names with known bad length. */
1046 		if (sfep->namelen == 0)
1047 			return __this_address;
1048 
1049 		/*
1050 		 * Check that the variable-length part of the structure is
1051 		 * within the data buffer.  The next entry starts after the
1052 		 * name component, so nextentry is an acceptable test.
1053 		 */
1054 		next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep);
1055 		if ((char *)next_sfep > endp)
1056 			return __this_address;
1057 
1058 		/*
1059 		 * Check for unknown flags.  Short form doesn't support
1060 		 * the incomplete or local bits, so we can use the namespace
1061 		 * mask here.
1062 		 */
1063 		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1064 			return __this_address;
1065 
1066 		/*
1067 		 * Check for invalid namespace combinations.  We only allow
1068 		 * one namespace flag per xattr, so we can just count the
1069 		 * bits (i.e. hweight) here.
1070 		 */
1071 		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1072 			return __this_address;
1073 
1074 		sfep = next_sfep;
1075 	}
1076 	if ((void *)sfep != (void *)endp)
1077 		return __this_address;
1078 
1079 	return NULL;
1080 }
1081 
1082 /*
1083  * Convert a leaf attribute list to shortform attribute list
1084  */
1085 int
1086 xfs_attr3_leaf_to_shortform(
1087 	struct xfs_buf		*bp,
1088 	struct xfs_da_args	*args,
1089 	int			forkoff)
1090 {
1091 	struct xfs_attr_leafblock *leaf;
1092 	struct xfs_attr3_icleaf_hdr ichdr;
1093 	struct xfs_attr_leaf_entry *entry;
1094 	struct xfs_attr_leaf_name_local *name_loc;
1095 	struct xfs_da_args	nargs;
1096 	struct xfs_inode	*dp = args->dp;
1097 	char			*tmpbuffer;
1098 	int			error;
1099 	int			i;
1100 
1101 	trace_xfs_attr_leaf_to_sf(args);
1102 
1103 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1104 	if (!tmpbuffer)
1105 		return -ENOMEM;
1106 
1107 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1108 
1109 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1110 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1111 	entry = xfs_attr3_leaf_entryp(leaf);
1112 
1113 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1114 	memset(bp->b_addr, 0, args->geo->blksize);
1115 
1116 	/*
1117 	 * Clean out the prior contents of the attribute list.
1118 	 */
1119 	error = xfs_da_shrink_inode(args, 0, bp);
1120 	if (error)
1121 		goto out;
1122 
1123 	if (forkoff == -1) {
1124 		ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1125 		ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1126 		xfs_attr_fork_remove(dp, args->trans);
1127 		goto out;
1128 	}
1129 
1130 	xfs_attr_shortform_create(args);
1131 
1132 	/*
1133 	 * Copy the attributes
1134 	 */
1135 	memset((char *)&nargs, 0, sizeof(nargs));
1136 	nargs.geo = args->geo;
1137 	nargs.dp = dp;
1138 	nargs.total = args->total;
1139 	nargs.whichfork = XFS_ATTR_FORK;
1140 	nargs.trans = args->trans;
1141 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1142 
1143 	for (i = 0; i < ichdr.count; entry++, i++) {
1144 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1145 			continue;	/* don't copy partial entries */
1146 		if (!entry->nameidx)
1147 			continue;
1148 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1149 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1150 		nargs.name = name_loc->nameval;
1151 		nargs.namelen = name_loc->namelen;
1152 		nargs.value = &name_loc->nameval[nargs.namelen];
1153 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1154 		nargs.hashval = be32_to_cpu(entry->hashval);
1155 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1156 		xfs_attr_shortform_add(&nargs, forkoff);
1157 	}
1158 	error = 0;
1159 
1160 out:
1161 	kmem_free(tmpbuffer);
1162 	return error;
1163 }
1164 
1165 /*
1166  * Convert from using a single leaf to a root node and a leaf.
1167  */
1168 int
1169 xfs_attr3_leaf_to_node(
1170 	struct xfs_da_args	*args)
1171 {
1172 	struct xfs_attr_leafblock *leaf;
1173 	struct xfs_attr3_icleaf_hdr icleafhdr;
1174 	struct xfs_attr_leaf_entry *entries;
1175 	struct xfs_da3_icnode_hdr icnodehdr;
1176 	struct xfs_da_intnode	*node;
1177 	struct xfs_inode	*dp = args->dp;
1178 	struct xfs_mount	*mp = dp->i_mount;
1179 	struct xfs_buf		*bp1 = NULL;
1180 	struct xfs_buf		*bp2 = NULL;
1181 	xfs_dablk_t		blkno;
1182 	int			error;
1183 
1184 	trace_xfs_attr_leaf_to_node(args);
1185 
1186 	error = xfs_da_grow_inode(args, &blkno);
1187 	if (error)
1188 		goto out;
1189 	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1190 	if (error)
1191 		goto out;
1192 
1193 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1194 	if (error)
1195 		goto out;
1196 
1197 	/* copy leaf to new buffer, update identifiers */
1198 	xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1199 	bp2->b_ops = bp1->b_ops;
1200 	memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1201 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1202 		struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1203 		hdr3->blkno = cpu_to_be64(bp2->b_bn);
1204 	}
1205 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1206 
1207 	/*
1208 	 * Set up the new root node.
1209 	 */
1210 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1211 	if (error)
1212 		goto out;
1213 	node = bp1->b_addr;
1214 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1215 
1216 	leaf = bp2->b_addr;
1217 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1218 	entries = xfs_attr3_leaf_entryp(leaf);
1219 
1220 	/* both on-disk, don't endian-flip twice */
1221 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1222 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1223 	icnodehdr.count = 1;
1224 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1225 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1226 	error = 0;
1227 out:
1228 	return error;
1229 }
1230 
1231 /*========================================================================
1232  * Routines used for growing the Btree.
1233  *========================================================================*/
1234 
1235 /*
1236  * Create the initial contents of a leaf attribute list
1237  * or a leaf in a node attribute list.
1238  */
1239 STATIC int
1240 xfs_attr3_leaf_create(
1241 	struct xfs_da_args	*args,
1242 	xfs_dablk_t		blkno,
1243 	struct xfs_buf		**bpp)
1244 {
1245 	struct xfs_attr_leafblock *leaf;
1246 	struct xfs_attr3_icleaf_hdr ichdr;
1247 	struct xfs_inode	*dp = args->dp;
1248 	struct xfs_mount	*mp = dp->i_mount;
1249 	struct xfs_buf		*bp;
1250 	int			error;
1251 
1252 	trace_xfs_attr_leaf_create(args);
1253 
1254 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1255 					    XFS_ATTR_FORK);
1256 	if (error)
1257 		return error;
1258 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1259 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1260 	leaf = bp->b_addr;
1261 	memset(leaf, 0, args->geo->blksize);
1262 
1263 	memset(&ichdr, 0, sizeof(ichdr));
1264 	ichdr.firstused = args->geo->blksize;
1265 
1266 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1267 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1268 
1269 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1270 
1271 		hdr3->blkno = cpu_to_be64(bp->b_bn);
1272 		hdr3->owner = cpu_to_be64(dp->i_ino);
1273 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1274 
1275 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1276 	} else {
1277 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1278 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1279 	}
1280 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1281 
1282 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1283 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1284 
1285 	*bpp = bp;
1286 	return 0;
1287 }
1288 
1289 /*
1290  * Split the leaf node, rebalance, then add the new entry.
1291  */
1292 int
1293 xfs_attr3_leaf_split(
1294 	struct xfs_da_state	*state,
1295 	struct xfs_da_state_blk	*oldblk,
1296 	struct xfs_da_state_blk	*newblk)
1297 {
1298 	xfs_dablk_t blkno;
1299 	int error;
1300 
1301 	trace_xfs_attr_leaf_split(state->args);
1302 
1303 	/*
1304 	 * Allocate space for a new leaf node.
1305 	 */
1306 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1307 	error = xfs_da_grow_inode(state->args, &blkno);
1308 	if (error)
1309 		return error;
1310 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1311 	if (error)
1312 		return error;
1313 	newblk->blkno = blkno;
1314 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1315 
1316 	/*
1317 	 * Rebalance the entries across the two leaves.
1318 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1319 	 */
1320 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1321 	error = xfs_da3_blk_link(state, oldblk, newblk);
1322 	if (error)
1323 		return error;
1324 
1325 	/*
1326 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1327 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1328 	 * "new" attrs info.  Will need the "old" info to remove it later.
1329 	 *
1330 	 * Insert the "new" entry in the correct block.
1331 	 */
1332 	if (state->inleaf) {
1333 		trace_xfs_attr_leaf_add_old(state->args);
1334 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1335 	} else {
1336 		trace_xfs_attr_leaf_add_new(state->args);
1337 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1338 	}
1339 
1340 	/*
1341 	 * Update last hashval in each block since we added the name.
1342 	 */
1343 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1344 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1345 	return error;
1346 }
1347 
1348 /*
1349  * Add a name to the leaf attribute list structure.
1350  */
1351 int
1352 xfs_attr3_leaf_add(
1353 	struct xfs_buf		*bp,
1354 	struct xfs_da_args	*args)
1355 {
1356 	struct xfs_attr_leafblock *leaf;
1357 	struct xfs_attr3_icleaf_hdr ichdr;
1358 	int			tablesize;
1359 	int			entsize;
1360 	int			sum;
1361 	int			tmp;
1362 	int			i;
1363 
1364 	trace_xfs_attr_leaf_add(args);
1365 
1366 	leaf = bp->b_addr;
1367 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1368 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1369 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1370 
1371 	/*
1372 	 * Search through freemap for first-fit on new name length.
1373 	 * (may need to figure in size of entry struct too)
1374 	 */
1375 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1376 					+ xfs_attr3_leaf_hdr_size(leaf);
1377 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1378 		if (tablesize > ichdr.firstused) {
1379 			sum += ichdr.freemap[i].size;
1380 			continue;
1381 		}
1382 		if (!ichdr.freemap[i].size)
1383 			continue;	/* no space in this map */
1384 		tmp = entsize;
1385 		if (ichdr.freemap[i].base < ichdr.firstused)
1386 			tmp += sizeof(xfs_attr_leaf_entry_t);
1387 		if (ichdr.freemap[i].size >= tmp) {
1388 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1389 			goto out_log_hdr;
1390 		}
1391 		sum += ichdr.freemap[i].size;
1392 	}
1393 
1394 	/*
1395 	 * If there are no holes in the address space of the block,
1396 	 * and we don't have enough freespace, then compaction will do us
1397 	 * no good and we should just give up.
1398 	 */
1399 	if (!ichdr.holes && sum < entsize)
1400 		return -ENOSPC;
1401 
1402 	/*
1403 	 * Compact the entries to coalesce free space.
1404 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1405 	 */
1406 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1407 
1408 	/*
1409 	 * After compaction, the block is guaranteed to have only one
1410 	 * free region, in freemap[0].  If it is not big enough, give up.
1411 	 */
1412 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1413 		tmp = -ENOSPC;
1414 		goto out_log_hdr;
1415 	}
1416 
1417 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1418 
1419 out_log_hdr:
1420 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1421 	xfs_trans_log_buf(args->trans, bp,
1422 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1423 				xfs_attr3_leaf_hdr_size(leaf)));
1424 	return tmp;
1425 }
1426 
1427 /*
1428  * Add a name to a leaf attribute list structure.
1429  */
1430 STATIC int
1431 xfs_attr3_leaf_add_work(
1432 	struct xfs_buf		*bp,
1433 	struct xfs_attr3_icleaf_hdr *ichdr,
1434 	struct xfs_da_args	*args,
1435 	int			mapindex)
1436 {
1437 	struct xfs_attr_leafblock *leaf;
1438 	struct xfs_attr_leaf_entry *entry;
1439 	struct xfs_attr_leaf_name_local *name_loc;
1440 	struct xfs_attr_leaf_name_remote *name_rmt;
1441 	struct xfs_mount	*mp;
1442 	int			tmp;
1443 	int			i;
1444 
1445 	trace_xfs_attr_leaf_add_work(args);
1446 
1447 	leaf = bp->b_addr;
1448 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1449 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1450 
1451 	/*
1452 	 * Force open some space in the entry array and fill it in.
1453 	 */
1454 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1455 	if (args->index < ichdr->count) {
1456 		tmp  = ichdr->count - args->index;
1457 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1458 		memmove(entry + 1, entry, tmp);
1459 		xfs_trans_log_buf(args->trans, bp,
1460 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1461 	}
1462 	ichdr->count++;
1463 
1464 	/*
1465 	 * Allocate space for the new string (at the end of the run).
1466 	 */
1467 	mp = args->trans->t_mountp;
1468 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1469 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1470 	ASSERT(ichdr->freemap[mapindex].size >=
1471 		xfs_attr_leaf_newentsize(args, NULL));
1472 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1473 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1474 
1475 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1476 
1477 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1478 				     ichdr->freemap[mapindex].size);
1479 	entry->hashval = cpu_to_be32(args->hashval);
1480 	entry->flags = args->attr_filter;
1481 	if (tmp)
1482 		entry->flags |= XFS_ATTR_LOCAL;
1483 	if (args->op_flags & XFS_DA_OP_RENAME) {
1484 		entry->flags |= XFS_ATTR_INCOMPLETE;
1485 		if ((args->blkno2 == args->blkno) &&
1486 		    (args->index2 <= args->index)) {
1487 			args->index2++;
1488 		}
1489 	}
1490 	xfs_trans_log_buf(args->trans, bp,
1491 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1492 	ASSERT((args->index == 0) ||
1493 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1494 	ASSERT((args->index == ichdr->count - 1) ||
1495 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1496 
1497 	/*
1498 	 * For "remote" attribute values, simply note that we need to
1499 	 * allocate space for the "remote" value.  We can't actually
1500 	 * allocate the extents in this transaction, and we can't decide
1501 	 * which blocks they should be as we might allocate more blocks
1502 	 * as part of this transaction (a split operation for example).
1503 	 */
1504 	if (entry->flags & XFS_ATTR_LOCAL) {
1505 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1506 		name_loc->namelen = args->namelen;
1507 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1508 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1509 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1510 				   be16_to_cpu(name_loc->valuelen));
1511 	} else {
1512 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1513 		name_rmt->namelen = args->namelen;
1514 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1515 		entry->flags |= XFS_ATTR_INCOMPLETE;
1516 		/* just in case */
1517 		name_rmt->valuelen = 0;
1518 		name_rmt->valueblk = 0;
1519 		args->rmtblkno = 1;
1520 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1521 		args->rmtvaluelen = args->valuelen;
1522 	}
1523 	xfs_trans_log_buf(args->trans, bp,
1524 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1525 				   xfs_attr_leaf_entsize(leaf, args->index)));
1526 
1527 	/*
1528 	 * Update the control info for this leaf node
1529 	 */
1530 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1531 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1532 
1533 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1534 					+ xfs_attr3_leaf_hdr_size(leaf));
1535 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1536 					+ xfs_attr3_leaf_hdr_size(leaf);
1537 
1538 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1539 		if (ichdr->freemap[i].base == tmp) {
1540 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1541 			ichdr->freemap[i].size -=
1542 				min_t(uint16_t, ichdr->freemap[i].size,
1543 						sizeof(xfs_attr_leaf_entry_t));
1544 		}
1545 	}
1546 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1547 	return 0;
1548 }
1549 
1550 /*
1551  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1552  */
1553 STATIC void
1554 xfs_attr3_leaf_compact(
1555 	struct xfs_da_args	*args,
1556 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1557 	struct xfs_buf		*bp)
1558 {
1559 	struct xfs_attr_leafblock *leaf_src;
1560 	struct xfs_attr_leafblock *leaf_dst;
1561 	struct xfs_attr3_icleaf_hdr ichdr_src;
1562 	struct xfs_trans	*trans = args->trans;
1563 	char			*tmpbuffer;
1564 
1565 	trace_xfs_attr_leaf_compact(args);
1566 
1567 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1568 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1569 	memset(bp->b_addr, 0, args->geo->blksize);
1570 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1571 	leaf_dst = bp->b_addr;
1572 
1573 	/*
1574 	 * Copy the on-disk header back into the destination buffer to ensure
1575 	 * all the information in the header that is not part of the incore
1576 	 * header structure is preserved.
1577 	 */
1578 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1579 
1580 	/* Initialise the incore headers */
1581 	ichdr_src = *ichdr_dst;	/* struct copy */
1582 	ichdr_dst->firstused = args->geo->blksize;
1583 	ichdr_dst->usedbytes = 0;
1584 	ichdr_dst->count = 0;
1585 	ichdr_dst->holes = 0;
1586 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1587 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1588 						ichdr_dst->freemap[0].base;
1589 
1590 	/* write the header back to initialise the underlying buffer */
1591 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1592 
1593 	/*
1594 	 * Copy all entry's in the same (sorted) order,
1595 	 * but allocate name/value pairs packed and in sequence.
1596 	 */
1597 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1598 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1599 	/*
1600 	 * this logs the entire buffer, but the caller must write the header
1601 	 * back to the buffer when it is finished modifying it.
1602 	 */
1603 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1604 
1605 	kmem_free(tmpbuffer);
1606 }
1607 
1608 /*
1609  * Compare two leaf blocks "order".
1610  * Return 0 unless leaf2 should go before leaf1.
1611  */
1612 static int
1613 xfs_attr3_leaf_order(
1614 	struct xfs_buf	*leaf1_bp,
1615 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1616 	struct xfs_buf	*leaf2_bp,
1617 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1618 {
1619 	struct xfs_attr_leaf_entry *entries1;
1620 	struct xfs_attr_leaf_entry *entries2;
1621 
1622 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1623 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1624 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1625 	    ((be32_to_cpu(entries2[0].hashval) <
1626 	      be32_to_cpu(entries1[0].hashval)) ||
1627 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1628 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1629 		return 1;
1630 	}
1631 	return 0;
1632 }
1633 
1634 int
1635 xfs_attr_leaf_order(
1636 	struct xfs_buf	*leaf1_bp,
1637 	struct xfs_buf	*leaf2_bp)
1638 {
1639 	struct xfs_attr3_icleaf_hdr ichdr1;
1640 	struct xfs_attr3_icleaf_hdr ichdr2;
1641 	struct xfs_mount *mp = leaf1_bp->b_mount;
1642 
1643 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1644 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1645 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1646 }
1647 
1648 /*
1649  * Redistribute the attribute list entries between two leaf nodes,
1650  * taking into account the size of the new entry.
1651  *
1652  * NOTE: if new block is empty, then it will get the upper half of the
1653  * old block.  At present, all (one) callers pass in an empty second block.
1654  *
1655  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1656  * to match what it is doing in splitting the attribute leaf block.  Those
1657  * values are used in "atomic rename" operations on attributes.  Note that
1658  * the "new" and "old" values can end up in different blocks.
1659  */
1660 STATIC void
1661 xfs_attr3_leaf_rebalance(
1662 	struct xfs_da_state	*state,
1663 	struct xfs_da_state_blk	*blk1,
1664 	struct xfs_da_state_blk	*blk2)
1665 {
1666 	struct xfs_da_args	*args;
1667 	struct xfs_attr_leafblock *leaf1;
1668 	struct xfs_attr_leafblock *leaf2;
1669 	struct xfs_attr3_icleaf_hdr ichdr1;
1670 	struct xfs_attr3_icleaf_hdr ichdr2;
1671 	struct xfs_attr_leaf_entry *entries1;
1672 	struct xfs_attr_leaf_entry *entries2;
1673 	int			count;
1674 	int			totallen;
1675 	int			max;
1676 	int			space;
1677 	int			swap;
1678 
1679 	/*
1680 	 * Set up environment.
1681 	 */
1682 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1683 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1684 	leaf1 = blk1->bp->b_addr;
1685 	leaf2 = blk2->bp->b_addr;
1686 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1687 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1688 	ASSERT(ichdr2.count == 0);
1689 	args = state->args;
1690 
1691 	trace_xfs_attr_leaf_rebalance(args);
1692 
1693 	/*
1694 	 * Check ordering of blocks, reverse if it makes things simpler.
1695 	 *
1696 	 * NOTE: Given that all (current) callers pass in an empty
1697 	 * second block, this code should never set "swap".
1698 	 */
1699 	swap = 0;
1700 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1701 		swap(blk1, blk2);
1702 
1703 		/* swap structures rather than reconverting them */
1704 		swap(ichdr1, ichdr2);
1705 
1706 		leaf1 = blk1->bp->b_addr;
1707 		leaf2 = blk2->bp->b_addr;
1708 		swap = 1;
1709 	}
1710 
1711 	/*
1712 	 * Examine entries until we reduce the absolute difference in
1713 	 * byte usage between the two blocks to a minimum.  Then get
1714 	 * the direction to copy and the number of elements to move.
1715 	 *
1716 	 * "inleaf" is true if the new entry should be inserted into blk1.
1717 	 * If "swap" is also true, then reverse the sense of "inleaf".
1718 	 */
1719 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1720 						      blk2, &ichdr2,
1721 						      &count, &totallen);
1722 	if (swap)
1723 		state->inleaf = !state->inleaf;
1724 
1725 	/*
1726 	 * Move any entries required from leaf to leaf:
1727 	 */
1728 	if (count < ichdr1.count) {
1729 		/*
1730 		 * Figure the total bytes to be added to the destination leaf.
1731 		 */
1732 		/* number entries being moved */
1733 		count = ichdr1.count - count;
1734 		space  = ichdr1.usedbytes - totallen;
1735 		space += count * sizeof(xfs_attr_leaf_entry_t);
1736 
1737 		/*
1738 		 * leaf2 is the destination, compact it if it looks tight.
1739 		 */
1740 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1741 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1742 		if (space > max)
1743 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1744 
1745 		/*
1746 		 * Move high entries from leaf1 to low end of leaf2.
1747 		 */
1748 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1749 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1750 
1751 	} else if (count > ichdr1.count) {
1752 		/*
1753 		 * I assert that since all callers pass in an empty
1754 		 * second buffer, this code should never execute.
1755 		 */
1756 		ASSERT(0);
1757 
1758 		/*
1759 		 * Figure the total bytes to be added to the destination leaf.
1760 		 */
1761 		/* number entries being moved */
1762 		count -= ichdr1.count;
1763 		space  = totallen - ichdr1.usedbytes;
1764 		space += count * sizeof(xfs_attr_leaf_entry_t);
1765 
1766 		/*
1767 		 * leaf1 is the destination, compact it if it looks tight.
1768 		 */
1769 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1770 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1771 		if (space > max)
1772 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1773 
1774 		/*
1775 		 * Move low entries from leaf2 to high end of leaf1.
1776 		 */
1777 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1778 					ichdr1.count, count);
1779 	}
1780 
1781 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1782 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1783 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1784 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1785 
1786 	/*
1787 	 * Copy out last hashval in each block for B-tree code.
1788 	 */
1789 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1790 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1791 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1792 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1793 
1794 	/*
1795 	 * Adjust the expected index for insertion.
1796 	 * NOTE: this code depends on the (current) situation that the
1797 	 * second block was originally empty.
1798 	 *
1799 	 * If the insertion point moved to the 2nd block, we must adjust
1800 	 * the index.  We must also track the entry just following the
1801 	 * new entry for use in an "atomic rename" operation, that entry
1802 	 * is always the "old" entry and the "new" entry is what we are
1803 	 * inserting.  The index/blkno fields refer to the "old" entry,
1804 	 * while the index2/blkno2 fields refer to the "new" entry.
1805 	 */
1806 	if (blk1->index > ichdr1.count) {
1807 		ASSERT(state->inleaf == 0);
1808 		blk2->index = blk1->index - ichdr1.count;
1809 		args->index = args->index2 = blk2->index;
1810 		args->blkno = args->blkno2 = blk2->blkno;
1811 	} else if (blk1->index == ichdr1.count) {
1812 		if (state->inleaf) {
1813 			args->index = blk1->index;
1814 			args->blkno = blk1->blkno;
1815 			args->index2 = 0;
1816 			args->blkno2 = blk2->blkno;
1817 		} else {
1818 			/*
1819 			 * On a double leaf split, the original attr location
1820 			 * is already stored in blkno2/index2, so don't
1821 			 * overwrite it overwise we corrupt the tree.
1822 			 */
1823 			blk2->index = blk1->index - ichdr1.count;
1824 			args->index = blk2->index;
1825 			args->blkno = blk2->blkno;
1826 			if (!state->extravalid) {
1827 				/*
1828 				 * set the new attr location to match the old
1829 				 * one and let the higher level split code
1830 				 * decide where in the leaf to place it.
1831 				 */
1832 				args->index2 = blk2->index;
1833 				args->blkno2 = blk2->blkno;
1834 			}
1835 		}
1836 	} else {
1837 		ASSERT(state->inleaf == 1);
1838 		args->index = args->index2 = blk1->index;
1839 		args->blkno = args->blkno2 = blk1->blkno;
1840 	}
1841 }
1842 
1843 /*
1844  * Examine entries until we reduce the absolute difference in
1845  * byte usage between the two blocks to a minimum.
1846  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1847  * GROT: there will always be enough room in either block for a new entry.
1848  * GROT: Do a double-split for this case?
1849  */
1850 STATIC int
1851 xfs_attr3_leaf_figure_balance(
1852 	struct xfs_da_state		*state,
1853 	struct xfs_da_state_blk		*blk1,
1854 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1855 	struct xfs_da_state_blk		*blk2,
1856 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1857 	int				*countarg,
1858 	int				*usedbytesarg)
1859 {
1860 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1861 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1862 	struct xfs_attr_leaf_entry	*entry;
1863 	int				count;
1864 	int				max;
1865 	int				index;
1866 	int				totallen = 0;
1867 	int				half;
1868 	int				lastdelta;
1869 	int				foundit = 0;
1870 	int				tmp;
1871 
1872 	/*
1873 	 * Examine entries until we reduce the absolute difference in
1874 	 * byte usage between the two blocks to a minimum.
1875 	 */
1876 	max = ichdr1->count + ichdr2->count;
1877 	half = (max + 1) * sizeof(*entry);
1878 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1879 			xfs_attr_leaf_newentsize(state->args, NULL);
1880 	half /= 2;
1881 	lastdelta = state->args->geo->blksize;
1882 	entry = xfs_attr3_leaf_entryp(leaf1);
1883 	for (count = index = 0; count < max; entry++, index++, count++) {
1884 
1885 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1886 		/*
1887 		 * The new entry is in the first block, account for it.
1888 		 */
1889 		if (count == blk1->index) {
1890 			tmp = totallen + sizeof(*entry) +
1891 				xfs_attr_leaf_newentsize(state->args, NULL);
1892 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1893 				break;
1894 			lastdelta = XFS_ATTR_ABS(half - tmp);
1895 			totallen = tmp;
1896 			foundit = 1;
1897 		}
1898 
1899 		/*
1900 		 * Wrap around into the second block if necessary.
1901 		 */
1902 		if (count == ichdr1->count) {
1903 			leaf1 = leaf2;
1904 			entry = xfs_attr3_leaf_entryp(leaf1);
1905 			index = 0;
1906 		}
1907 
1908 		/*
1909 		 * Figure out if next leaf entry would be too much.
1910 		 */
1911 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1912 									index);
1913 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1914 			break;
1915 		lastdelta = XFS_ATTR_ABS(half - tmp);
1916 		totallen = tmp;
1917 #undef XFS_ATTR_ABS
1918 	}
1919 
1920 	/*
1921 	 * Calculate the number of usedbytes that will end up in lower block.
1922 	 * If new entry not in lower block, fix up the count.
1923 	 */
1924 	totallen -= count * sizeof(*entry);
1925 	if (foundit) {
1926 		totallen -= sizeof(*entry) +
1927 				xfs_attr_leaf_newentsize(state->args, NULL);
1928 	}
1929 
1930 	*countarg = count;
1931 	*usedbytesarg = totallen;
1932 	return foundit;
1933 }
1934 
1935 /*========================================================================
1936  * Routines used for shrinking the Btree.
1937  *========================================================================*/
1938 
1939 /*
1940  * Check a leaf block and its neighbors to see if the block should be
1941  * collapsed into one or the other neighbor.  Always keep the block
1942  * with the smaller block number.
1943  * If the current block is over 50% full, don't try to join it, return 0.
1944  * If the block is empty, fill in the state structure and return 2.
1945  * If it can be collapsed, fill in the state structure and return 1.
1946  * If nothing can be done, return 0.
1947  *
1948  * GROT: allow for INCOMPLETE entries in calculation.
1949  */
1950 int
1951 xfs_attr3_leaf_toosmall(
1952 	struct xfs_da_state	*state,
1953 	int			*action)
1954 {
1955 	struct xfs_attr_leafblock *leaf;
1956 	struct xfs_da_state_blk	*blk;
1957 	struct xfs_attr3_icleaf_hdr ichdr;
1958 	struct xfs_buf		*bp;
1959 	xfs_dablk_t		blkno;
1960 	int			bytes;
1961 	int			forward;
1962 	int			error;
1963 	int			retval;
1964 	int			i;
1965 
1966 	trace_xfs_attr_leaf_toosmall(state->args);
1967 
1968 	/*
1969 	 * Check for the degenerate case of the block being over 50% full.
1970 	 * If so, it's not worth even looking to see if we might be able
1971 	 * to coalesce with a sibling.
1972 	 */
1973 	blk = &state->path.blk[ state->path.active-1 ];
1974 	leaf = blk->bp->b_addr;
1975 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1976 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
1977 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1978 		ichdr.usedbytes;
1979 	if (bytes > (state->args->geo->blksize >> 1)) {
1980 		*action = 0;	/* blk over 50%, don't try to join */
1981 		return 0;
1982 	}
1983 
1984 	/*
1985 	 * Check for the degenerate case of the block being empty.
1986 	 * If the block is empty, we'll simply delete it, no need to
1987 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1988 	 * to merge with the forward block unless it is NULL.
1989 	 */
1990 	if (ichdr.count == 0) {
1991 		/*
1992 		 * Make altpath point to the block we want to keep and
1993 		 * path point to the block we want to drop (this one).
1994 		 */
1995 		forward = (ichdr.forw != 0);
1996 		memcpy(&state->altpath, &state->path, sizeof(state->path));
1997 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1998 						 0, &retval);
1999 		if (error)
2000 			return error;
2001 		if (retval) {
2002 			*action = 0;
2003 		} else {
2004 			*action = 2;
2005 		}
2006 		return 0;
2007 	}
2008 
2009 	/*
2010 	 * Examine each sibling block to see if we can coalesce with
2011 	 * at least 25% free space to spare.  We need to figure out
2012 	 * whether to merge with the forward or the backward block.
2013 	 * We prefer coalescing with the lower numbered sibling so as
2014 	 * to shrink an attribute list over time.
2015 	 */
2016 	/* start with smaller blk num */
2017 	forward = ichdr.forw < ichdr.back;
2018 	for (i = 0; i < 2; forward = !forward, i++) {
2019 		struct xfs_attr3_icleaf_hdr ichdr2;
2020 		if (forward)
2021 			blkno = ichdr.forw;
2022 		else
2023 			blkno = ichdr.back;
2024 		if (blkno == 0)
2025 			continue;
2026 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2027 					blkno, &bp);
2028 		if (error)
2029 			return error;
2030 
2031 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2032 
2033 		bytes = state->args->geo->blksize -
2034 			(state->args->geo->blksize >> 2) -
2035 			ichdr.usedbytes - ichdr2.usedbytes -
2036 			((ichdr.count + ichdr2.count) *
2037 					sizeof(xfs_attr_leaf_entry_t)) -
2038 			xfs_attr3_leaf_hdr_size(leaf);
2039 
2040 		xfs_trans_brelse(state->args->trans, bp);
2041 		if (bytes >= 0)
2042 			break;	/* fits with at least 25% to spare */
2043 	}
2044 	if (i >= 2) {
2045 		*action = 0;
2046 		return 0;
2047 	}
2048 
2049 	/*
2050 	 * Make altpath point to the block we want to keep (the lower
2051 	 * numbered block) and path point to the block we want to drop.
2052 	 */
2053 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2054 	if (blkno < blk->blkno) {
2055 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2056 						 0, &retval);
2057 	} else {
2058 		error = xfs_da3_path_shift(state, &state->path, forward,
2059 						 0, &retval);
2060 	}
2061 	if (error)
2062 		return error;
2063 	if (retval) {
2064 		*action = 0;
2065 	} else {
2066 		*action = 1;
2067 	}
2068 	return 0;
2069 }
2070 
2071 /*
2072  * Remove a name from the leaf attribute list structure.
2073  *
2074  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2075  * If two leaves are 37% full, when combined they will leave 25% free.
2076  */
2077 int
2078 xfs_attr3_leaf_remove(
2079 	struct xfs_buf		*bp,
2080 	struct xfs_da_args	*args)
2081 {
2082 	struct xfs_attr_leafblock *leaf;
2083 	struct xfs_attr3_icleaf_hdr ichdr;
2084 	struct xfs_attr_leaf_entry *entry;
2085 	int			before;
2086 	int			after;
2087 	int			smallest;
2088 	int			entsize;
2089 	int			tablesize;
2090 	int			tmp;
2091 	int			i;
2092 
2093 	trace_xfs_attr_leaf_remove(args);
2094 
2095 	leaf = bp->b_addr;
2096 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2097 
2098 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2099 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2100 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2101 					xfs_attr3_leaf_hdr_size(leaf));
2102 
2103 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2104 
2105 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2106 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2107 
2108 	/*
2109 	 * Scan through free region table:
2110 	 *    check for adjacency of free'd entry with an existing one,
2111 	 *    find smallest free region in case we need to replace it,
2112 	 *    adjust any map that borders the entry table,
2113 	 */
2114 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2115 					+ xfs_attr3_leaf_hdr_size(leaf);
2116 	tmp = ichdr.freemap[0].size;
2117 	before = after = -1;
2118 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2119 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2120 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2121 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2122 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2123 		if (ichdr.freemap[i].base == tablesize) {
2124 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2125 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2126 		}
2127 
2128 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2129 				be16_to_cpu(entry->nameidx)) {
2130 			before = i;
2131 		} else if (ichdr.freemap[i].base ==
2132 				(be16_to_cpu(entry->nameidx) + entsize)) {
2133 			after = i;
2134 		} else if (ichdr.freemap[i].size < tmp) {
2135 			tmp = ichdr.freemap[i].size;
2136 			smallest = i;
2137 		}
2138 	}
2139 
2140 	/*
2141 	 * Coalesce adjacent freemap regions,
2142 	 * or replace the smallest region.
2143 	 */
2144 	if ((before >= 0) || (after >= 0)) {
2145 		if ((before >= 0) && (after >= 0)) {
2146 			ichdr.freemap[before].size += entsize;
2147 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2148 			ichdr.freemap[after].base = 0;
2149 			ichdr.freemap[after].size = 0;
2150 		} else if (before >= 0) {
2151 			ichdr.freemap[before].size += entsize;
2152 		} else {
2153 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2154 			ichdr.freemap[after].size += entsize;
2155 		}
2156 	} else {
2157 		/*
2158 		 * Replace smallest region (if it is smaller than free'd entry)
2159 		 */
2160 		if (ichdr.freemap[smallest].size < entsize) {
2161 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2162 			ichdr.freemap[smallest].size = entsize;
2163 		}
2164 	}
2165 
2166 	/*
2167 	 * Did we remove the first entry?
2168 	 */
2169 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2170 		smallest = 1;
2171 	else
2172 		smallest = 0;
2173 
2174 	/*
2175 	 * Compress the remaining entries and zero out the removed stuff.
2176 	 */
2177 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2178 	ichdr.usedbytes -= entsize;
2179 	xfs_trans_log_buf(args->trans, bp,
2180 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2181 				   entsize));
2182 
2183 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2184 	memmove(entry, entry + 1, tmp);
2185 	ichdr.count--;
2186 	xfs_trans_log_buf(args->trans, bp,
2187 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2188 
2189 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2190 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2191 
2192 	/*
2193 	 * If we removed the first entry, re-find the first used byte
2194 	 * in the name area.  Note that if the entry was the "firstused",
2195 	 * then we don't have a "hole" in our block resulting from
2196 	 * removing the name.
2197 	 */
2198 	if (smallest) {
2199 		tmp = args->geo->blksize;
2200 		entry = xfs_attr3_leaf_entryp(leaf);
2201 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2202 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2203 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2204 
2205 			if (be16_to_cpu(entry->nameidx) < tmp)
2206 				tmp = be16_to_cpu(entry->nameidx);
2207 		}
2208 		ichdr.firstused = tmp;
2209 		ASSERT(ichdr.firstused != 0);
2210 	} else {
2211 		ichdr.holes = 1;	/* mark as needing compaction */
2212 	}
2213 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2214 	xfs_trans_log_buf(args->trans, bp,
2215 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2216 					  xfs_attr3_leaf_hdr_size(leaf)));
2217 
2218 	/*
2219 	 * Check if leaf is less than 50% full, caller may want to
2220 	 * "join" the leaf with a sibling if so.
2221 	 */
2222 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2223 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2224 
2225 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2226 }
2227 
2228 /*
2229  * Move all the attribute list entries from drop_leaf into save_leaf.
2230  */
2231 void
2232 xfs_attr3_leaf_unbalance(
2233 	struct xfs_da_state	*state,
2234 	struct xfs_da_state_blk	*drop_blk,
2235 	struct xfs_da_state_blk	*save_blk)
2236 {
2237 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2238 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2239 	struct xfs_attr3_icleaf_hdr drophdr;
2240 	struct xfs_attr3_icleaf_hdr savehdr;
2241 	struct xfs_attr_leaf_entry *entry;
2242 
2243 	trace_xfs_attr_leaf_unbalance(state->args);
2244 
2245 	drop_leaf = drop_blk->bp->b_addr;
2246 	save_leaf = save_blk->bp->b_addr;
2247 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2248 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2249 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2250 
2251 	/*
2252 	 * Save last hashval from dying block for later Btree fixup.
2253 	 */
2254 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2255 
2256 	/*
2257 	 * Check if we need a temp buffer, or can we do it in place.
2258 	 * Note that we don't check "leaf" for holes because we will
2259 	 * always be dropping it, toosmall() decided that for us already.
2260 	 */
2261 	if (savehdr.holes == 0) {
2262 		/*
2263 		 * dest leaf has no holes, so we add there.  May need
2264 		 * to make some room in the entry array.
2265 		 */
2266 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2267 					 drop_blk->bp, &drophdr)) {
2268 			xfs_attr3_leaf_moveents(state->args,
2269 						drop_leaf, &drophdr, 0,
2270 						save_leaf, &savehdr, 0,
2271 						drophdr.count);
2272 		} else {
2273 			xfs_attr3_leaf_moveents(state->args,
2274 						drop_leaf, &drophdr, 0,
2275 						save_leaf, &savehdr,
2276 						savehdr.count, drophdr.count);
2277 		}
2278 	} else {
2279 		/*
2280 		 * Destination has holes, so we make a temporary copy
2281 		 * of the leaf and add them both to that.
2282 		 */
2283 		struct xfs_attr_leafblock *tmp_leaf;
2284 		struct xfs_attr3_icleaf_hdr tmphdr;
2285 
2286 		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2287 
2288 		/*
2289 		 * Copy the header into the temp leaf so that all the stuff
2290 		 * not in the incore header is present and gets copied back in
2291 		 * once we've moved all the entries.
2292 		 */
2293 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2294 
2295 		memset(&tmphdr, 0, sizeof(tmphdr));
2296 		tmphdr.magic = savehdr.magic;
2297 		tmphdr.forw = savehdr.forw;
2298 		tmphdr.back = savehdr.back;
2299 		tmphdr.firstused = state->args->geo->blksize;
2300 
2301 		/* write the header to the temp buffer to initialise it */
2302 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2303 
2304 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2305 					 drop_blk->bp, &drophdr)) {
2306 			xfs_attr3_leaf_moveents(state->args,
2307 						drop_leaf, &drophdr, 0,
2308 						tmp_leaf, &tmphdr, 0,
2309 						drophdr.count);
2310 			xfs_attr3_leaf_moveents(state->args,
2311 						save_leaf, &savehdr, 0,
2312 						tmp_leaf, &tmphdr, tmphdr.count,
2313 						savehdr.count);
2314 		} else {
2315 			xfs_attr3_leaf_moveents(state->args,
2316 						save_leaf, &savehdr, 0,
2317 						tmp_leaf, &tmphdr, 0,
2318 						savehdr.count);
2319 			xfs_attr3_leaf_moveents(state->args,
2320 						drop_leaf, &drophdr, 0,
2321 						tmp_leaf, &tmphdr, tmphdr.count,
2322 						drophdr.count);
2323 		}
2324 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2325 		savehdr = tmphdr; /* struct copy */
2326 		kmem_free(tmp_leaf);
2327 	}
2328 
2329 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2330 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2331 					   state->args->geo->blksize - 1);
2332 
2333 	/*
2334 	 * Copy out last hashval in each block for B-tree code.
2335 	 */
2336 	entry = xfs_attr3_leaf_entryp(save_leaf);
2337 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2338 }
2339 
2340 /*========================================================================
2341  * Routines used for finding things in the Btree.
2342  *========================================================================*/
2343 
2344 /*
2345  * Look up a name in a leaf attribute list structure.
2346  * This is the internal routine, it uses the caller's buffer.
2347  *
2348  * Note that duplicate keys are allowed, but only check within the
2349  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2350  *
2351  * Return in args->index the index into the entry[] array of either
2352  * the found entry, or where the entry should have been (insert before
2353  * that entry).
2354  *
2355  * Don't change the args->value unless we find the attribute.
2356  */
2357 int
2358 xfs_attr3_leaf_lookup_int(
2359 	struct xfs_buf		*bp,
2360 	struct xfs_da_args	*args)
2361 {
2362 	struct xfs_attr_leafblock *leaf;
2363 	struct xfs_attr3_icleaf_hdr ichdr;
2364 	struct xfs_attr_leaf_entry *entry;
2365 	struct xfs_attr_leaf_entry *entries;
2366 	struct xfs_attr_leaf_name_local *name_loc;
2367 	struct xfs_attr_leaf_name_remote *name_rmt;
2368 	xfs_dahash_t		hashval;
2369 	int			probe;
2370 	int			span;
2371 
2372 	trace_xfs_attr_leaf_lookup(args);
2373 
2374 	leaf = bp->b_addr;
2375 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2376 	entries = xfs_attr3_leaf_entryp(leaf);
2377 	if (ichdr.count >= args->geo->blksize / 8) {
2378 		xfs_buf_mark_corrupt(bp);
2379 		return -EFSCORRUPTED;
2380 	}
2381 
2382 	/*
2383 	 * Binary search.  (note: small blocks will skip this loop)
2384 	 */
2385 	hashval = args->hashval;
2386 	probe = span = ichdr.count / 2;
2387 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2388 		span /= 2;
2389 		if (be32_to_cpu(entry->hashval) < hashval)
2390 			probe += span;
2391 		else if (be32_to_cpu(entry->hashval) > hashval)
2392 			probe -= span;
2393 		else
2394 			break;
2395 	}
2396 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2397 		xfs_buf_mark_corrupt(bp);
2398 		return -EFSCORRUPTED;
2399 	}
2400 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2401 		xfs_buf_mark_corrupt(bp);
2402 		return -EFSCORRUPTED;
2403 	}
2404 
2405 	/*
2406 	 * Since we may have duplicate hashval's, find the first matching
2407 	 * hashval in the leaf.
2408 	 */
2409 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2410 		entry--;
2411 		probe--;
2412 	}
2413 	while (probe < ichdr.count &&
2414 	       be32_to_cpu(entry->hashval) < hashval) {
2415 		entry++;
2416 		probe++;
2417 	}
2418 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2419 		args->index = probe;
2420 		return -ENOATTR;
2421 	}
2422 
2423 	/*
2424 	 * Duplicate keys may be present, so search all of them for a match.
2425 	 */
2426 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2427 			entry++, probe++) {
2428 /*
2429  * GROT: Add code to remove incomplete entries.
2430  */
2431 		if (entry->flags & XFS_ATTR_LOCAL) {
2432 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2433 			if (!xfs_attr_match(args, name_loc->namelen,
2434 					name_loc->nameval, entry->flags))
2435 				continue;
2436 			args->index = probe;
2437 			return -EEXIST;
2438 		} else {
2439 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2440 			if (!xfs_attr_match(args, name_rmt->namelen,
2441 					name_rmt->name, entry->flags))
2442 				continue;
2443 			args->index = probe;
2444 			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2445 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2446 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2447 							args->dp->i_mount,
2448 							args->rmtvaluelen);
2449 			return -EEXIST;
2450 		}
2451 	}
2452 	args->index = probe;
2453 	return -ENOATTR;
2454 }
2455 
2456 /*
2457  * Get the value associated with an attribute name from a leaf attribute
2458  * list structure.
2459  *
2460  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2461  * lookup, we only return an error if the attribute does not exist or we can't
2462  * retrieve the value.
2463  */
2464 int
2465 xfs_attr3_leaf_getvalue(
2466 	struct xfs_buf		*bp,
2467 	struct xfs_da_args	*args)
2468 {
2469 	struct xfs_attr_leafblock *leaf;
2470 	struct xfs_attr3_icleaf_hdr ichdr;
2471 	struct xfs_attr_leaf_entry *entry;
2472 	struct xfs_attr_leaf_name_local *name_loc;
2473 	struct xfs_attr_leaf_name_remote *name_rmt;
2474 
2475 	leaf = bp->b_addr;
2476 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2477 	ASSERT(ichdr.count < args->geo->blksize / 8);
2478 	ASSERT(args->index < ichdr.count);
2479 
2480 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2481 	if (entry->flags & XFS_ATTR_LOCAL) {
2482 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2483 		ASSERT(name_loc->namelen == args->namelen);
2484 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2485 		return xfs_attr_copy_value(args,
2486 					&name_loc->nameval[args->namelen],
2487 					be16_to_cpu(name_loc->valuelen));
2488 	}
2489 
2490 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2491 	ASSERT(name_rmt->namelen == args->namelen);
2492 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2493 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2494 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2495 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2496 					       args->rmtvaluelen);
2497 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2498 }
2499 
2500 /*========================================================================
2501  * Utility routines.
2502  *========================================================================*/
2503 
2504 /*
2505  * Move the indicated entries from one leaf to another.
2506  * NOTE: this routine modifies both source and destination leaves.
2507  */
2508 /*ARGSUSED*/
2509 STATIC void
2510 xfs_attr3_leaf_moveents(
2511 	struct xfs_da_args		*args,
2512 	struct xfs_attr_leafblock	*leaf_s,
2513 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2514 	int				start_s,
2515 	struct xfs_attr_leafblock	*leaf_d,
2516 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2517 	int				start_d,
2518 	int				count)
2519 {
2520 	struct xfs_attr_leaf_entry	*entry_s;
2521 	struct xfs_attr_leaf_entry	*entry_d;
2522 	int				desti;
2523 	int				tmp;
2524 	int				i;
2525 
2526 	/*
2527 	 * Check for nothing to do.
2528 	 */
2529 	if (count == 0)
2530 		return;
2531 
2532 	/*
2533 	 * Set up environment.
2534 	 */
2535 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2536 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2537 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2538 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2539 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2540 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2541 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2542 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2543 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2544 
2545 	ASSERT(start_s < ichdr_s->count);
2546 	ASSERT(start_d <= ichdr_d->count);
2547 	ASSERT(count <= ichdr_s->count);
2548 
2549 
2550 	/*
2551 	 * Move the entries in the destination leaf up to make a hole?
2552 	 */
2553 	if (start_d < ichdr_d->count) {
2554 		tmp  = ichdr_d->count - start_d;
2555 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2556 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2557 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2558 		memmove(entry_d, entry_s, tmp);
2559 	}
2560 
2561 	/*
2562 	 * Copy all entry's in the same (sorted) order,
2563 	 * but allocate attribute info packed and in sequence.
2564 	 */
2565 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2566 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2567 	desti = start_d;
2568 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2569 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2570 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2571 #ifdef GROT
2572 		/*
2573 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2574 		 * may also need to change the insertion index.  Code turned
2575 		 * off for 6.2, should be revisited later.
2576 		 */
2577 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2578 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2579 			ichdr_s->usedbytes -= tmp;
2580 			ichdr_s->count -= 1;
2581 			entry_d--;	/* to compensate for ++ in loop hdr */
2582 			desti--;
2583 			if ((start_s + i) < offset)
2584 				result++;	/* insertion index adjustment */
2585 		} else {
2586 #endif /* GROT */
2587 			ichdr_d->firstused -= tmp;
2588 			/* both on-disk, don't endian flip twice */
2589 			entry_d->hashval = entry_s->hashval;
2590 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2591 			entry_d->flags = entry_s->flags;
2592 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2593 							<= args->geo->blksize);
2594 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2595 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2596 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2597 							<= args->geo->blksize);
2598 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2599 			ichdr_s->usedbytes -= tmp;
2600 			ichdr_d->usedbytes += tmp;
2601 			ichdr_s->count -= 1;
2602 			ichdr_d->count += 1;
2603 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2604 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2605 			ASSERT(ichdr_d->firstused >= tmp);
2606 #ifdef GROT
2607 		}
2608 #endif /* GROT */
2609 	}
2610 
2611 	/*
2612 	 * Zero out the entries we just copied.
2613 	 */
2614 	if (start_s == ichdr_s->count) {
2615 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2616 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2617 		ASSERT(((char *)entry_s + tmp) <=
2618 		       ((char *)leaf_s + args->geo->blksize));
2619 		memset(entry_s, 0, tmp);
2620 	} else {
2621 		/*
2622 		 * Move the remaining entries down to fill the hole,
2623 		 * then zero the entries at the top.
2624 		 */
2625 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2626 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2627 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2628 		memmove(entry_d, entry_s, tmp);
2629 
2630 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2631 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2632 		ASSERT(((char *)entry_s + tmp) <=
2633 		       ((char *)leaf_s + args->geo->blksize));
2634 		memset(entry_s, 0, tmp);
2635 	}
2636 
2637 	/*
2638 	 * Fill in the freemap information
2639 	 */
2640 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2641 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2642 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2643 	ichdr_d->freemap[1].base = 0;
2644 	ichdr_d->freemap[2].base = 0;
2645 	ichdr_d->freemap[1].size = 0;
2646 	ichdr_d->freemap[2].size = 0;
2647 	ichdr_s->holes = 1;	/* leaf may not be compact */
2648 }
2649 
2650 /*
2651  * Pick up the last hashvalue from a leaf block.
2652  */
2653 xfs_dahash_t
2654 xfs_attr_leaf_lasthash(
2655 	struct xfs_buf	*bp,
2656 	int		*count)
2657 {
2658 	struct xfs_attr3_icleaf_hdr ichdr;
2659 	struct xfs_attr_leaf_entry *entries;
2660 	struct xfs_mount *mp = bp->b_mount;
2661 
2662 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2663 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2664 	if (count)
2665 		*count = ichdr.count;
2666 	if (!ichdr.count)
2667 		return 0;
2668 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2669 }
2670 
2671 /*
2672  * Calculate the number of bytes used to store the indicated attribute
2673  * (whether local or remote only calculate bytes in this block).
2674  */
2675 STATIC int
2676 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2677 {
2678 	struct xfs_attr_leaf_entry *entries;
2679 	xfs_attr_leaf_name_local_t *name_loc;
2680 	xfs_attr_leaf_name_remote_t *name_rmt;
2681 	int size;
2682 
2683 	entries = xfs_attr3_leaf_entryp(leaf);
2684 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2685 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2686 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2687 						   be16_to_cpu(name_loc->valuelen));
2688 	} else {
2689 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2690 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2691 	}
2692 	return size;
2693 }
2694 
2695 /*
2696  * Calculate the number of bytes that would be required to store the new
2697  * attribute (whether local or remote only calculate bytes in this block).
2698  * This routine decides as a side effect whether the attribute will be
2699  * a "local" or a "remote" attribute.
2700  */
2701 int
2702 xfs_attr_leaf_newentsize(
2703 	struct xfs_da_args	*args,
2704 	int			*local)
2705 {
2706 	int			size;
2707 
2708 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2709 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2710 		if (local)
2711 			*local = 1;
2712 		return size;
2713 	}
2714 	if (local)
2715 		*local = 0;
2716 	return xfs_attr_leaf_entsize_remote(args->namelen);
2717 }
2718 
2719 
2720 /*========================================================================
2721  * Manage the INCOMPLETE flag in a leaf entry
2722  *========================================================================*/
2723 
2724 /*
2725  * Clear the INCOMPLETE flag on an entry in a leaf block.
2726  */
2727 int
2728 xfs_attr3_leaf_clearflag(
2729 	struct xfs_da_args	*args)
2730 {
2731 	struct xfs_attr_leafblock *leaf;
2732 	struct xfs_attr_leaf_entry *entry;
2733 	struct xfs_attr_leaf_name_remote *name_rmt;
2734 	struct xfs_buf		*bp;
2735 	int			error;
2736 #ifdef DEBUG
2737 	struct xfs_attr3_icleaf_hdr ichdr;
2738 	xfs_attr_leaf_name_local_t *name_loc;
2739 	int namelen;
2740 	char *name;
2741 #endif /* DEBUG */
2742 
2743 	trace_xfs_attr_leaf_clearflag(args);
2744 	/*
2745 	 * Set up the operation.
2746 	 */
2747 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2748 	if (error)
2749 		return error;
2750 
2751 	leaf = bp->b_addr;
2752 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2753 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2754 
2755 #ifdef DEBUG
2756 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2757 	ASSERT(args->index < ichdr.count);
2758 	ASSERT(args->index >= 0);
2759 
2760 	if (entry->flags & XFS_ATTR_LOCAL) {
2761 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2762 		namelen = name_loc->namelen;
2763 		name = (char *)name_loc->nameval;
2764 	} else {
2765 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2766 		namelen = name_rmt->namelen;
2767 		name = (char *)name_rmt->name;
2768 	}
2769 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2770 	ASSERT(namelen == args->namelen);
2771 	ASSERT(memcmp(name, args->name, namelen) == 0);
2772 #endif /* DEBUG */
2773 
2774 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2775 	xfs_trans_log_buf(args->trans, bp,
2776 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2777 
2778 	if (args->rmtblkno) {
2779 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2780 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2781 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2782 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2783 		xfs_trans_log_buf(args->trans, bp,
2784 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2785 	}
2786 
2787 	return 0;
2788 }
2789 
2790 /*
2791  * Set the INCOMPLETE flag on an entry in a leaf block.
2792  */
2793 int
2794 xfs_attr3_leaf_setflag(
2795 	struct xfs_da_args	*args)
2796 {
2797 	struct xfs_attr_leafblock *leaf;
2798 	struct xfs_attr_leaf_entry *entry;
2799 	struct xfs_attr_leaf_name_remote *name_rmt;
2800 	struct xfs_buf		*bp;
2801 	int error;
2802 #ifdef DEBUG
2803 	struct xfs_attr3_icleaf_hdr ichdr;
2804 #endif
2805 
2806 	trace_xfs_attr_leaf_setflag(args);
2807 
2808 	/*
2809 	 * Set up the operation.
2810 	 */
2811 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2812 	if (error)
2813 		return error;
2814 
2815 	leaf = bp->b_addr;
2816 #ifdef DEBUG
2817 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2818 	ASSERT(args->index < ichdr.count);
2819 	ASSERT(args->index >= 0);
2820 #endif
2821 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2822 
2823 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2824 	entry->flags |= XFS_ATTR_INCOMPLETE;
2825 	xfs_trans_log_buf(args->trans, bp,
2826 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2827 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2828 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2829 		name_rmt->valueblk = 0;
2830 		name_rmt->valuelen = 0;
2831 		xfs_trans_log_buf(args->trans, bp,
2832 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2833 	}
2834 
2835 	return 0;
2836 }
2837 
2838 /*
2839  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2840  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2841  * entry given by args->blkno2/index2.
2842  *
2843  * Note that they could be in different blocks, or in the same block.
2844  */
2845 int
2846 xfs_attr3_leaf_flipflags(
2847 	struct xfs_da_args	*args)
2848 {
2849 	struct xfs_attr_leafblock *leaf1;
2850 	struct xfs_attr_leafblock *leaf2;
2851 	struct xfs_attr_leaf_entry *entry1;
2852 	struct xfs_attr_leaf_entry *entry2;
2853 	struct xfs_attr_leaf_name_remote *name_rmt;
2854 	struct xfs_buf		*bp1;
2855 	struct xfs_buf		*bp2;
2856 	int error;
2857 #ifdef DEBUG
2858 	struct xfs_attr3_icleaf_hdr ichdr1;
2859 	struct xfs_attr3_icleaf_hdr ichdr2;
2860 	xfs_attr_leaf_name_local_t *name_loc;
2861 	int namelen1, namelen2;
2862 	char *name1, *name2;
2863 #endif /* DEBUG */
2864 
2865 	trace_xfs_attr_leaf_flipflags(args);
2866 
2867 	/*
2868 	 * Read the block containing the "old" attr
2869 	 */
2870 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2871 	if (error)
2872 		return error;
2873 
2874 	/*
2875 	 * Read the block containing the "new" attr, if it is different
2876 	 */
2877 	if (args->blkno2 != args->blkno) {
2878 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2879 					   &bp2);
2880 		if (error)
2881 			return error;
2882 	} else {
2883 		bp2 = bp1;
2884 	}
2885 
2886 	leaf1 = bp1->b_addr;
2887 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2888 
2889 	leaf2 = bp2->b_addr;
2890 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2891 
2892 #ifdef DEBUG
2893 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2894 	ASSERT(args->index < ichdr1.count);
2895 	ASSERT(args->index >= 0);
2896 
2897 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2898 	ASSERT(args->index2 < ichdr2.count);
2899 	ASSERT(args->index2 >= 0);
2900 
2901 	if (entry1->flags & XFS_ATTR_LOCAL) {
2902 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2903 		namelen1 = name_loc->namelen;
2904 		name1 = (char *)name_loc->nameval;
2905 	} else {
2906 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2907 		namelen1 = name_rmt->namelen;
2908 		name1 = (char *)name_rmt->name;
2909 	}
2910 	if (entry2->flags & XFS_ATTR_LOCAL) {
2911 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2912 		namelen2 = name_loc->namelen;
2913 		name2 = (char *)name_loc->nameval;
2914 	} else {
2915 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2916 		namelen2 = name_rmt->namelen;
2917 		name2 = (char *)name_rmt->name;
2918 	}
2919 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2920 	ASSERT(namelen1 == namelen2);
2921 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2922 #endif /* DEBUG */
2923 
2924 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2925 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2926 
2927 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2928 	xfs_trans_log_buf(args->trans, bp1,
2929 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2930 	if (args->rmtblkno) {
2931 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2932 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2933 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2934 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2935 		xfs_trans_log_buf(args->trans, bp1,
2936 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2937 	}
2938 
2939 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2940 	xfs_trans_log_buf(args->trans, bp2,
2941 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2942 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2943 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2944 		name_rmt->valueblk = 0;
2945 		name_rmt->valuelen = 0;
2946 		xfs_trans_log_buf(args->trans, bp2,
2947 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2948 	}
2949 
2950 	return 0;
2951 }
2952