xref: /openbmc/linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision fdaf1bb3)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_da_format.h"
16 #include "xfs_da_btree.h"
17 #include "xfs_inode.h"
18 #include "xfs_trans.h"
19 #include "xfs_bmap_btree.h"
20 #include "xfs_bmap.h"
21 #include "xfs_attr_sf.h"
22 #include "xfs_attr.h"
23 #include "xfs_attr_remote.h"
24 #include "xfs_attr_leaf.h"
25 #include "xfs_error.h"
26 #include "xfs_trace.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_dir2.h"
29 #include "xfs_log.h"
30 #include "xfs_ag.h"
31 #include "xfs_errortag.h"
32 
33 
34 /*
35  * xfs_attr_leaf.c
36  *
37  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
38  */
39 
40 /*========================================================================
41  * Function prototypes for the kernel.
42  *========================================================================*/
43 
44 /*
45  * Routines used for growing the Btree.
46  */
47 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
48 				 xfs_dablk_t which_block, struct xfs_buf **bpp);
49 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
50 				   struct xfs_attr3_icleaf_hdr *ichdr,
51 				   struct xfs_da_args *args, int freemap_index);
52 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
53 				   struct xfs_attr3_icleaf_hdr *ichdr,
54 				   struct xfs_buf *leaf_buffer);
55 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
56 						   xfs_da_state_blk_t *blk1,
57 						   xfs_da_state_blk_t *blk2);
58 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
59 			xfs_da_state_blk_t *leaf_blk_1,
60 			struct xfs_attr3_icleaf_hdr *ichdr1,
61 			xfs_da_state_blk_t *leaf_blk_2,
62 			struct xfs_attr3_icleaf_hdr *ichdr2,
63 			int *number_entries_in_blk1,
64 			int *number_usedbytes_in_blk1);
65 
66 /*
67  * Utility routines.
68  */
69 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
70 			struct xfs_attr_leafblock *src_leaf,
71 			struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
72 			struct xfs_attr_leafblock *dst_leaf,
73 			struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
74 			int move_count);
75 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
76 
77 /*
78  * attr3 block 'firstused' conversion helpers.
79  *
80  * firstused refers to the offset of the first used byte of the nameval region
81  * of an attr leaf block. The region starts at the tail of the block and expands
82  * backwards towards the middle. As such, firstused is initialized to the block
83  * size for an empty leaf block and is reduced from there.
84  *
85  * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
86  * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
87  * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
88  * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
89  * the attr block size. The following helpers manage the conversion between the
90  * in-core and on-disk formats.
91  */
92 
93 static void
94 xfs_attr3_leaf_firstused_from_disk(
95 	struct xfs_da_geometry		*geo,
96 	struct xfs_attr3_icleaf_hdr	*to,
97 	struct xfs_attr_leafblock	*from)
98 {
99 	struct xfs_attr3_leaf_hdr	*hdr3;
100 
101 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
102 		hdr3 = (struct xfs_attr3_leaf_hdr *) from;
103 		to->firstused = be16_to_cpu(hdr3->firstused);
104 	} else {
105 		to->firstused = be16_to_cpu(from->hdr.firstused);
106 	}
107 
108 	/*
109 	 * Convert from the magic fsb size value to actual blocksize. This
110 	 * should only occur for empty blocks when the block size overflows
111 	 * 16-bits.
112 	 */
113 	if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
114 		ASSERT(!to->count && !to->usedbytes);
115 		ASSERT(geo->blksize > USHRT_MAX);
116 		to->firstused = geo->blksize;
117 	}
118 }
119 
120 static void
121 xfs_attr3_leaf_firstused_to_disk(
122 	struct xfs_da_geometry		*geo,
123 	struct xfs_attr_leafblock	*to,
124 	struct xfs_attr3_icleaf_hdr	*from)
125 {
126 	struct xfs_attr3_leaf_hdr	*hdr3;
127 	uint32_t			firstused;
128 
129 	/* magic value should only be seen on disk */
130 	ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
131 
132 	/*
133 	 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
134 	 * value. This only overflows at the max supported value of 64k. Use the
135 	 * magic on-disk value to represent block size in this case.
136 	 */
137 	firstused = from->firstused;
138 	if (firstused > USHRT_MAX) {
139 		ASSERT(from->firstused == geo->blksize);
140 		firstused = XFS_ATTR3_LEAF_NULLOFF;
141 	}
142 
143 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
144 		hdr3 = (struct xfs_attr3_leaf_hdr *) to;
145 		hdr3->firstused = cpu_to_be16(firstused);
146 	} else {
147 		to->hdr.firstused = cpu_to_be16(firstused);
148 	}
149 }
150 
151 void
152 xfs_attr3_leaf_hdr_from_disk(
153 	struct xfs_da_geometry		*geo,
154 	struct xfs_attr3_icleaf_hdr	*to,
155 	struct xfs_attr_leafblock	*from)
156 {
157 	int	i;
158 
159 	ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
160 	       from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
161 
162 	if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
163 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
164 
165 		to->forw = be32_to_cpu(hdr3->info.hdr.forw);
166 		to->back = be32_to_cpu(hdr3->info.hdr.back);
167 		to->magic = be16_to_cpu(hdr3->info.hdr.magic);
168 		to->count = be16_to_cpu(hdr3->count);
169 		to->usedbytes = be16_to_cpu(hdr3->usedbytes);
170 		xfs_attr3_leaf_firstused_from_disk(geo, to, from);
171 		to->holes = hdr3->holes;
172 
173 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
174 			to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
175 			to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
176 		}
177 		return;
178 	}
179 	to->forw = be32_to_cpu(from->hdr.info.forw);
180 	to->back = be32_to_cpu(from->hdr.info.back);
181 	to->magic = be16_to_cpu(from->hdr.info.magic);
182 	to->count = be16_to_cpu(from->hdr.count);
183 	to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
184 	xfs_attr3_leaf_firstused_from_disk(geo, to, from);
185 	to->holes = from->hdr.holes;
186 
187 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
188 		to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
189 		to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
190 	}
191 }
192 
193 void
194 xfs_attr3_leaf_hdr_to_disk(
195 	struct xfs_da_geometry		*geo,
196 	struct xfs_attr_leafblock	*to,
197 	struct xfs_attr3_icleaf_hdr	*from)
198 {
199 	int				i;
200 
201 	ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
202 	       from->magic == XFS_ATTR3_LEAF_MAGIC);
203 
204 	if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
205 		struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
206 
207 		hdr3->info.hdr.forw = cpu_to_be32(from->forw);
208 		hdr3->info.hdr.back = cpu_to_be32(from->back);
209 		hdr3->info.hdr.magic = cpu_to_be16(from->magic);
210 		hdr3->count = cpu_to_be16(from->count);
211 		hdr3->usedbytes = cpu_to_be16(from->usedbytes);
212 		xfs_attr3_leaf_firstused_to_disk(geo, to, from);
213 		hdr3->holes = from->holes;
214 		hdr3->pad1 = 0;
215 
216 		for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
217 			hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
218 			hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
219 		}
220 		return;
221 	}
222 	to->hdr.info.forw = cpu_to_be32(from->forw);
223 	to->hdr.info.back = cpu_to_be32(from->back);
224 	to->hdr.info.magic = cpu_to_be16(from->magic);
225 	to->hdr.count = cpu_to_be16(from->count);
226 	to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
227 	xfs_attr3_leaf_firstused_to_disk(geo, to, from);
228 	to->hdr.holes = from->holes;
229 	to->hdr.pad1 = 0;
230 
231 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
232 		to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
233 		to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
234 	}
235 }
236 
237 static xfs_failaddr_t
238 xfs_attr3_leaf_verify_entry(
239 	struct xfs_mount			*mp,
240 	char					*buf_end,
241 	struct xfs_attr_leafblock		*leaf,
242 	struct xfs_attr3_icleaf_hdr		*leafhdr,
243 	struct xfs_attr_leaf_entry		*ent,
244 	int					idx,
245 	__u32					*last_hashval)
246 {
247 	struct xfs_attr_leaf_name_local		*lentry;
248 	struct xfs_attr_leaf_name_remote	*rentry;
249 	char					*name_end;
250 	unsigned int				nameidx;
251 	unsigned int				namesize;
252 	__u32					hashval;
253 
254 	/* hash order check */
255 	hashval = be32_to_cpu(ent->hashval);
256 	if (hashval < *last_hashval)
257 		return __this_address;
258 	*last_hashval = hashval;
259 
260 	nameidx = be16_to_cpu(ent->nameidx);
261 	if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
262 		return __this_address;
263 
264 	/*
265 	 * Check the name information.  The namelen fields are u8 so we can't
266 	 * possibly exceed the maximum name length of 255 bytes.
267 	 */
268 	if (ent->flags & XFS_ATTR_LOCAL) {
269 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
270 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
271 				be16_to_cpu(lentry->valuelen));
272 		name_end = (char *)lentry + namesize;
273 		if (lentry->namelen == 0)
274 			return __this_address;
275 	} else {
276 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
277 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
278 		name_end = (char *)rentry + namesize;
279 		if (rentry->namelen == 0)
280 			return __this_address;
281 		if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
282 		    rentry->valueblk == 0)
283 			return __this_address;
284 	}
285 
286 	if (name_end > buf_end)
287 		return __this_address;
288 
289 	return NULL;
290 }
291 
292 static xfs_failaddr_t
293 xfs_attr3_leaf_verify(
294 	struct xfs_buf			*bp)
295 {
296 	struct xfs_attr3_icleaf_hdr	ichdr;
297 	struct xfs_mount		*mp = bp->b_mount;
298 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
299 	struct xfs_attr_leaf_entry	*entries;
300 	struct xfs_attr_leaf_entry	*ent;
301 	char				*buf_end;
302 	uint32_t			end;	/* must be 32bit - see below */
303 	__u32				last_hashval = 0;
304 	int				i;
305 	xfs_failaddr_t			fa;
306 
307 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
308 
309 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
310 	if (fa)
311 		return fa;
312 
313 	/*
314 	 * firstused is the block offset of the first name info structure.
315 	 * Make sure it doesn't go off the block or crash into the header.
316 	 */
317 	if (ichdr.firstused > mp->m_attr_geo->blksize)
318 		return __this_address;
319 	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
320 		return __this_address;
321 
322 	/* Make sure the entries array doesn't crash into the name info. */
323 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
324 	if ((char *)&entries[ichdr.count] >
325 	    (char *)bp->b_addr + ichdr.firstused)
326 		return __this_address;
327 
328 	/*
329 	 * NOTE: This verifier historically failed empty leaf buffers because
330 	 * we expect the fork to be in another format. Empty attr fork format
331 	 * conversions are possible during xattr set, however, and format
332 	 * conversion is not atomic with the xattr set that triggers it. We
333 	 * cannot assume leaf blocks are non-empty until that is addressed.
334 	*/
335 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
336 	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
337 		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
338 				ent, i, &last_hashval);
339 		if (fa)
340 			return fa;
341 	}
342 
343 	/*
344 	 * Quickly check the freemap information.  Attribute data has to be
345 	 * aligned to 4-byte boundaries, and likewise for the free space.
346 	 *
347 	 * Note that for 64k block size filesystems, the freemap entries cannot
348 	 * overflow as they are only be16 fields. However, when checking end
349 	 * pointer of the freemap, we have to be careful to detect overflows and
350 	 * so use uint32_t for those checks.
351 	 */
352 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
353 		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
354 			return __this_address;
355 		if (ichdr.freemap[i].base & 0x3)
356 			return __this_address;
357 		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
358 			return __this_address;
359 		if (ichdr.freemap[i].size & 0x3)
360 			return __this_address;
361 
362 		/* be care of 16 bit overflows here */
363 		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
364 		if (end < ichdr.freemap[i].base)
365 			return __this_address;
366 		if (end > mp->m_attr_geo->blksize)
367 			return __this_address;
368 	}
369 
370 	return NULL;
371 }
372 
373 static void
374 xfs_attr3_leaf_write_verify(
375 	struct xfs_buf	*bp)
376 {
377 	struct xfs_mount	*mp = bp->b_mount;
378 	struct xfs_buf_log_item	*bip = bp->b_log_item;
379 	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
380 	xfs_failaddr_t		fa;
381 
382 	fa = xfs_attr3_leaf_verify(bp);
383 	if (fa) {
384 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
385 		return;
386 	}
387 
388 	if (!xfs_has_crc(mp))
389 		return;
390 
391 	if (bip)
392 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
393 
394 	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
395 }
396 
397 /*
398  * leaf/node format detection on trees is sketchy, so a node read can be done on
399  * leaf level blocks when detection identifies the tree as a node format tree
400  * incorrectly. In this case, we need to swap the verifier to match the correct
401  * format of the block being read.
402  */
403 static void
404 xfs_attr3_leaf_read_verify(
405 	struct xfs_buf		*bp)
406 {
407 	struct xfs_mount	*mp = bp->b_mount;
408 	xfs_failaddr_t		fa;
409 
410 	if (xfs_has_crc(mp) &&
411 	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
412 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
413 	else {
414 		fa = xfs_attr3_leaf_verify(bp);
415 		if (fa)
416 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
417 	}
418 }
419 
420 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
421 	.name = "xfs_attr3_leaf",
422 	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
423 		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
424 	.verify_read = xfs_attr3_leaf_read_verify,
425 	.verify_write = xfs_attr3_leaf_write_verify,
426 	.verify_struct = xfs_attr3_leaf_verify,
427 };
428 
429 int
430 xfs_attr3_leaf_read(
431 	struct xfs_trans	*tp,
432 	struct xfs_inode	*dp,
433 	xfs_dablk_t		bno,
434 	struct xfs_buf		**bpp)
435 {
436 	int			err;
437 
438 	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
439 			&xfs_attr3_leaf_buf_ops);
440 	if (!err && tp && *bpp)
441 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
442 	return err;
443 }
444 
445 /*========================================================================
446  * Namespace helper routines
447  *========================================================================*/
448 
449 /*
450  * If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
451  * flag on disk - if there's an incomplete attr then recovery needs to tear it
452  * down. If there's no incomplete attr, then recovery needs to tear that attr
453  * down to replace it with the attr that has been logged. In this case, the
454  * INCOMPLETE flag will not be set in attr->attr_filter, but rather
455  * XFS_DA_OP_RECOVERY will be set in args->op_flags.
456  */
457 static bool
458 xfs_attr_match(
459 	struct xfs_da_args	*args,
460 	uint8_t			namelen,
461 	unsigned char		*name,
462 	int			flags)
463 {
464 
465 	if (args->namelen != namelen)
466 		return false;
467 	if (memcmp(args->name, name, namelen) != 0)
468 		return false;
469 
470 	/* Recovery ignores the INCOMPLETE flag. */
471 	if ((args->op_flags & XFS_DA_OP_RECOVERY) &&
472 	    args->attr_filter == (flags & XFS_ATTR_NSP_ONDISK_MASK))
473 		return true;
474 
475 	/* All remaining matches need to be filtered by INCOMPLETE state. */
476 	if (args->attr_filter !=
477 	    (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
478 		return false;
479 	return true;
480 }
481 
482 static int
483 xfs_attr_copy_value(
484 	struct xfs_da_args	*args,
485 	unsigned char		*value,
486 	int			valuelen)
487 {
488 	/*
489 	 * No copy if all we have to do is get the length
490 	 */
491 	if (!args->valuelen) {
492 		args->valuelen = valuelen;
493 		return 0;
494 	}
495 
496 	/*
497 	 * No copy if the length of the existing buffer is too small
498 	 */
499 	if (args->valuelen < valuelen) {
500 		args->valuelen = valuelen;
501 		return -ERANGE;
502 	}
503 
504 	if (!args->value) {
505 		args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
506 		if (!args->value)
507 			return -ENOMEM;
508 	}
509 	args->valuelen = valuelen;
510 
511 	/* remote block xattr requires IO for copy-in */
512 	if (args->rmtblkno)
513 		return xfs_attr_rmtval_get(args);
514 
515 	/*
516 	 * This is to prevent a GCC warning because the remote xattr case
517 	 * doesn't have a value to pass in. In that case, we never reach here,
518 	 * but GCC can't work that out and so throws a "passing NULL to
519 	 * memcpy" warning.
520 	 */
521 	if (!value)
522 		return -EINVAL;
523 	memcpy(args->value, value, valuelen);
524 	return 0;
525 }
526 
527 /*========================================================================
528  * External routines when attribute fork size < XFS_LITINO(mp).
529  *========================================================================*/
530 
531 /*
532  * Query whether the total requested number of attr fork bytes of extended
533  * attribute space will be able to fit inline.
534  *
535  * Returns zero if not, else the i_forkoff fork offset to be used in the
536  * literal area for attribute data once the new bytes have been added.
537  *
538  * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
539  * special case for dev/uuid inodes, they have fixed size data forks.
540  */
541 int
542 xfs_attr_shortform_bytesfit(
543 	struct xfs_inode	*dp,
544 	int			bytes)
545 {
546 	struct xfs_mount	*mp = dp->i_mount;
547 	int64_t			dsize;
548 	int			minforkoff;
549 	int			maxforkoff;
550 	int			offset;
551 
552 	/*
553 	 * Check if the new size could fit at all first:
554 	 */
555 	if (bytes > XFS_LITINO(mp))
556 		return 0;
557 
558 	/* rounded down */
559 	offset = (XFS_LITINO(mp) - bytes) >> 3;
560 
561 	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
562 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
563 		return (offset >= minforkoff) ? minforkoff : 0;
564 	}
565 
566 	/*
567 	 * If the requested numbers of bytes is smaller or equal to the
568 	 * current attribute fork size we can always proceed.
569 	 *
570 	 * Note that if_bytes in the data fork might actually be larger than
571 	 * the current data fork size is due to delalloc extents. In that
572 	 * case either the extent count will go down when they are converted
573 	 * to real extents, or the delalloc conversion will take care of the
574 	 * literal area rebalancing.
575 	 */
576 	if (bytes <= XFS_IFORK_ASIZE(dp))
577 		return dp->i_forkoff;
578 
579 	/*
580 	 * For attr2 we can try to move the forkoff if there is space in the
581 	 * literal area, but for the old format we are done if there is no
582 	 * space in the fixed attribute fork.
583 	 */
584 	if (!xfs_has_attr2(mp))
585 		return 0;
586 
587 	dsize = dp->i_df.if_bytes;
588 
589 	switch (dp->i_df.if_format) {
590 	case XFS_DINODE_FMT_EXTENTS:
591 		/*
592 		 * If there is no attr fork and the data fork is extents,
593 		 * determine if creating the default attr fork will result
594 		 * in the extents form migrating to btree. If so, the
595 		 * minimum offset only needs to be the space required for
596 		 * the btree root.
597 		 */
598 		if (!dp->i_forkoff && dp->i_df.if_bytes >
599 		    xfs_default_attroffset(dp))
600 			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
601 		break;
602 	case XFS_DINODE_FMT_BTREE:
603 		/*
604 		 * If we have a data btree then keep forkoff if we have one,
605 		 * otherwise we are adding a new attr, so then we set
606 		 * minforkoff to where the btree root can finish so we have
607 		 * plenty of room for attrs
608 		 */
609 		if (dp->i_forkoff) {
610 			if (offset < dp->i_forkoff)
611 				return 0;
612 			return dp->i_forkoff;
613 		}
614 		dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
615 		break;
616 	}
617 
618 	/*
619 	 * A data fork btree root must have space for at least
620 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
621 	 */
622 	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
623 	minforkoff = roundup(minforkoff, 8) >> 3;
624 
625 	/* attr fork btree root can have at least this many key/ptr pairs */
626 	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
627 	maxforkoff = maxforkoff >> 3;	/* rounded down */
628 
629 	if (offset >= maxforkoff)
630 		return maxforkoff;
631 	if (offset >= minforkoff)
632 		return offset;
633 	return 0;
634 }
635 
636 /*
637  * Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
638  * - noattr2 mount option is set,
639  * - on-disk version bit says it is already set, or
640  * - the attr2 mount option is not set to enable automatic upgrade from attr1.
641  */
642 STATIC void
643 xfs_sbversion_add_attr2(
644 	struct xfs_mount	*mp,
645 	struct xfs_trans	*tp)
646 {
647 	if (xfs_has_noattr2(mp))
648 		return;
649 	if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
650 		return;
651 	if (!xfs_has_attr2(mp))
652 		return;
653 
654 	spin_lock(&mp->m_sb_lock);
655 	xfs_add_attr2(mp);
656 	spin_unlock(&mp->m_sb_lock);
657 	xfs_log_sb(tp);
658 }
659 
660 /*
661  * Create the initial contents of a shortform attribute list.
662  */
663 void
664 xfs_attr_shortform_create(
665 	struct xfs_da_args	*args)
666 {
667 	struct xfs_inode	*dp = args->dp;
668 	struct xfs_ifork	*ifp = dp->i_afp;
669 	struct xfs_attr_sf_hdr	*hdr;
670 
671 	trace_xfs_attr_sf_create(args);
672 
673 	ASSERT(ifp->if_bytes == 0);
674 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
675 		ifp->if_format = XFS_DINODE_FMT_LOCAL;
676 	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
677 	hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
678 	memset(hdr, 0, sizeof(*hdr));
679 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
680 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
681 }
682 
683 /*
684  * Return -EEXIST if attr is found, or -ENOATTR if not
685  * args:  args containing attribute name and namelen
686  * sfep:  If not null, pointer will be set to the last attr entry found on
687 	  -EEXIST.  On -ENOATTR pointer is left at the last entry in the list
688  * basep: If not null, pointer is set to the byte offset of the entry in the
689  *	  list on -EEXIST.  On -ENOATTR, pointer is left at the byte offset of
690  *	  the last entry in the list
691  */
692 int
693 xfs_attr_sf_findname(
694 	struct xfs_da_args	 *args,
695 	struct xfs_attr_sf_entry **sfep,
696 	unsigned int		 *basep)
697 {
698 	struct xfs_attr_shortform *sf;
699 	struct xfs_attr_sf_entry *sfe;
700 	unsigned int		base = sizeof(struct xfs_attr_sf_hdr);
701 	int			size = 0;
702 	int			end;
703 	int			i;
704 
705 	sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
706 	sfe = &sf->list[0];
707 	end = sf->hdr.count;
708 	for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe),
709 			     base += size, i++) {
710 		size = xfs_attr_sf_entsize(sfe);
711 		if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
712 				    sfe->flags))
713 			continue;
714 		break;
715 	}
716 
717 	if (sfep != NULL)
718 		*sfep = sfe;
719 
720 	if (basep != NULL)
721 		*basep = base;
722 
723 	if (i == end)
724 		return -ENOATTR;
725 	return -EEXIST;
726 }
727 
728 /*
729  * Add a name/value pair to the shortform attribute list.
730  * Overflow from the inode has already been checked for.
731  */
732 void
733 xfs_attr_shortform_add(
734 	struct xfs_da_args		*args,
735 	int				forkoff)
736 {
737 	struct xfs_attr_shortform	*sf;
738 	struct xfs_attr_sf_entry	*sfe;
739 	int				offset, size;
740 	struct xfs_mount		*mp;
741 	struct xfs_inode		*dp;
742 	struct xfs_ifork		*ifp;
743 
744 	trace_xfs_attr_sf_add(args);
745 
746 	dp = args->dp;
747 	mp = dp->i_mount;
748 	dp->i_forkoff = forkoff;
749 
750 	ifp = dp->i_afp;
751 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
752 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
753 	if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
754 		ASSERT(0);
755 
756 	offset = (char *)sfe - (char *)sf;
757 	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
758 	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
759 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
760 	sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset);
761 
762 	sfe->namelen = args->namelen;
763 	sfe->valuelen = args->valuelen;
764 	sfe->flags = args->attr_filter;
765 	memcpy(sfe->nameval, args->name, args->namelen);
766 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
767 	sf->hdr.count++;
768 	be16_add_cpu(&sf->hdr.totsize, size);
769 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
770 
771 	xfs_sbversion_add_attr2(mp, args->trans);
772 }
773 
774 /*
775  * After the last attribute is removed revert to original inode format,
776  * making all literal area available to the data fork once more.
777  */
778 void
779 xfs_attr_fork_remove(
780 	struct xfs_inode	*ip,
781 	struct xfs_trans	*tp)
782 {
783 	ASSERT(ip->i_afp->if_nextents == 0);
784 
785 	xfs_idestroy_fork(ip->i_afp);
786 	kmem_cache_free(xfs_ifork_cache, ip->i_afp);
787 	ip->i_afp = NULL;
788 	ip->i_forkoff = 0;
789 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
790 }
791 
792 /*
793  * Remove an attribute from the shortform attribute list structure.
794  */
795 int
796 xfs_attr_sf_removename(
797 	struct xfs_da_args		*args)
798 {
799 	struct xfs_attr_shortform	*sf;
800 	struct xfs_attr_sf_entry	*sfe;
801 	int				size = 0, end, totsize;
802 	unsigned int			base;
803 	struct xfs_mount		*mp;
804 	struct xfs_inode		*dp;
805 	int				error;
806 
807 	trace_xfs_attr_sf_remove(args);
808 
809 	dp = args->dp;
810 	mp = dp->i_mount;
811 	sf = (struct xfs_attr_shortform *)dp->i_afp->if_u1.if_data;
812 
813 	error = xfs_attr_sf_findname(args, &sfe, &base);
814 
815 	/*
816 	 * If we are recovering an operation, finding nothing to
817 	 * remove is not an error - it just means there was nothing
818 	 * to clean up.
819 	 */
820 	if (error == -ENOATTR && (args->op_flags & XFS_DA_OP_RECOVERY))
821 		return 0;
822 	if (error != -EEXIST)
823 		return error;
824 	size = xfs_attr_sf_entsize(sfe);
825 
826 	/*
827 	 * Fix up the attribute fork data, covering the hole
828 	 */
829 	end = base + size;
830 	totsize = be16_to_cpu(sf->hdr.totsize);
831 	if (end != totsize)
832 		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
833 	sf->hdr.count--;
834 	be16_add_cpu(&sf->hdr.totsize, -size);
835 
836 	/*
837 	 * Fix up the start offset of the attribute fork
838 	 */
839 	totsize -= size;
840 	if (totsize == sizeof(xfs_attr_sf_hdr_t) && xfs_has_attr2(mp) &&
841 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
842 	    !(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE))) {
843 		xfs_attr_fork_remove(dp, args->trans);
844 	} else {
845 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
846 		dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
847 		ASSERT(dp->i_forkoff);
848 		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
849 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
850 				!xfs_has_attr2(mp) ||
851 				dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
852 		xfs_trans_log_inode(args->trans, dp,
853 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
854 	}
855 
856 	xfs_sbversion_add_attr2(mp, args->trans);
857 
858 	return 0;
859 }
860 
861 /*
862  * Look up a name in a shortform attribute list structure.
863  */
864 /*ARGSUSED*/
865 int
866 xfs_attr_shortform_lookup(xfs_da_args_t *args)
867 {
868 	struct xfs_attr_shortform *sf;
869 	struct xfs_attr_sf_entry *sfe;
870 	int i;
871 	struct xfs_ifork *ifp;
872 
873 	trace_xfs_attr_sf_lookup(args);
874 
875 	ifp = args->dp->i_afp;
876 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
877 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
878 	sfe = &sf->list[0];
879 	for (i = 0; i < sf->hdr.count;
880 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
881 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
882 				sfe->flags))
883 			return -EEXIST;
884 	}
885 	return -ENOATTR;
886 }
887 
888 /*
889  * Retrieve the attribute value and length.
890  *
891  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
892  * lookup, we only return an error if the attribute does not exist or we can't
893  * retrieve the value.
894  */
895 int
896 xfs_attr_shortform_getvalue(
897 	struct xfs_da_args	*args)
898 {
899 	struct xfs_attr_shortform *sf;
900 	struct xfs_attr_sf_entry *sfe;
901 	int			i;
902 
903 	ASSERT(args->dp->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
904 	sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
905 	sfe = &sf->list[0];
906 	for (i = 0; i < sf->hdr.count;
907 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
908 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
909 				sfe->flags))
910 			return xfs_attr_copy_value(args,
911 				&sfe->nameval[args->namelen], sfe->valuelen);
912 	}
913 	return -ENOATTR;
914 }
915 
916 /*
917  * Convert from using the shortform to the leaf.  On success, return the
918  * buffer so that we can keep it locked until we're totally done with it.
919  */
920 int
921 xfs_attr_shortform_to_leaf(
922 	struct xfs_da_args		*args,
923 	struct xfs_buf			**leaf_bp)
924 {
925 	struct xfs_inode		*dp;
926 	struct xfs_attr_shortform	*sf;
927 	struct xfs_attr_sf_entry	*sfe;
928 	struct xfs_da_args		nargs;
929 	char				*tmpbuffer;
930 	int				error, i, size;
931 	xfs_dablk_t			blkno;
932 	struct xfs_buf			*bp;
933 	struct xfs_ifork		*ifp;
934 
935 	trace_xfs_attr_sf_to_leaf(args);
936 
937 	dp = args->dp;
938 	ifp = dp->i_afp;
939 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
940 	size = be16_to_cpu(sf->hdr.totsize);
941 	tmpbuffer = kmem_alloc(size, 0);
942 	ASSERT(tmpbuffer != NULL);
943 	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
944 	sf = (struct xfs_attr_shortform *)tmpbuffer;
945 
946 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
947 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
948 
949 	bp = NULL;
950 	error = xfs_da_grow_inode(args, &blkno);
951 	if (error)
952 		goto out;
953 
954 	ASSERT(blkno == 0);
955 	error = xfs_attr3_leaf_create(args, blkno, &bp);
956 	if (error)
957 		goto out;
958 
959 	memset((char *)&nargs, 0, sizeof(nargs));
960 	nargs.dp = dp;
961 	nargs.geo = args->geo;
962 	nargs.total = args->total;
963 	nargs.whichfork = XFS_ATTR_FORK;
964 	nargs.trans = args->trans;
965 	nargs.op_flags = XFS_DA_OP_OKNOENT;
966 
967 	sfe = &sf->list[0];
968 	for (i = 0; i < sf->hdr.count; i++) {
969 		nargs.name = sfe->nameval;
970 		nargs.namelen = sfe->namelen;
971 		nargs.value = &sfe->nameval[nargs.namelen];
972 		nargs.valuelen = sfe->valuelen;
973 		nargs.hashval = xfs_da_hashname(sfe->nameval,
974 						sfe->namelen);
975 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
976 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
977 		ASSERT(error == -ENOATTR);
978 		error = xfs_attr3_leaf_add(bp, &nargs);
979 		ASSERT(error != -ENOSPC);
980 		if (error)
981 			goto out;
982 		sfe = xfs_attr_sf_nextentry(sfe);
983 	}
984 	error = 0;
985 	*leaf_bp = bp;
986 out:
987 	kmem_free(tmpbuffer);
988 	return error;
989 }
990 
991 /*
992  * Check a leaf attribute block to see if all the entries would fit into
993  * a shortform attribute list.
994  */
995 int
996 xfs_attr_shortform_allfit(
997 	struct xfs_buf		*bp,
998 	struct xfs_inode	*dp)
999 {
1000 	struct xfs_attr_leafblock *leaf;
1001 	struct xfs_attr_leaf_entry *entry;
1002 	xfs_attr_leaf_name_local_t *name_loc;
1003 	struct xfs_attr3_icleaf_hdr leafhdr;
1004 	int			bytes;
1005 	int			i;
1006 	struct xfs_mount	*mp = bp->b_mount;
1007 
1008 	leaf = bp->b_addr;
1009 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
1010 	entry = xfs_attr3_leaf_entryp(leaf);
1011 
1012 	bytes = sizeof(struct xfs_attr_sf_hdr);
1013 	for (i = 0; i < leafhdr.count; entry++, i++) {
1014 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1015 			continue;		/* don't copy partial entries */
1016 		if (!(entry->flags & XFS_ATTR_LOCAL))
1017 			return 0;
1018 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1019 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
1020 			return 0;
1021 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1022 			return 0;
1023 		bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1024 					be16_to_cpu(name_loc->valuelen));
1025 	}
1026 	if (xfs_has_attr2(dp->i_mount) &&
1027 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1028 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
1029 		return -1;
1030 	return xfs_attr_shortform_bytesfit(dp, bytes);
1031 }
1032 
1033 /* Verify the consistency of an inline attribute fork. */
1034 xfs_failaddr_t
1035 xfs_attr_shortform_verify(
1036 	struct xfs_inode		*ip)
1037 {
1038 	struct xfs_attr_shortform	*sfp;
1039 	struct xfs_attr_sf_entry	*sfep;
1040 	struct xfs_attr_sf_entry	*next_sfep;
1041 	char				*endp;
1042 	struct xfs_ifork		*ifp;
1043 	int				i;
1044 	int64_t				size;
1045 
1046 	ASSERT(ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
1047 	ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
1048 	sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1049 	size = ifp->if_bytes;
1050 
1051 	/*
1052 	 * Give up if the attribute is way too short.
1053 	 */
1054 	if (size < sizeof(struct xfs_attr_sf_hdr))
1055 		return __this_address;
1056 
1057 	endp = (char *)sfp + size;
1058 
1059 	/* Check all reported entries */
1060 	sfep = &sfp->list[0];
1061 	for (i = 0; i < sfp->hdr.count; i++) {
1062 		/*
1063 		 * struct xfs_attr_sf_entry has a variable length.
1064 		 * Check the fixed-offset parts of the structure are
1065 		 * within the data buffer.
1066 		 * xfs_attr_sf_entry is defined with a 1-byte variable
1067 		 * array at the end, so we must subtract that off.
1068 		 */
1069 		if (((char *)sfep + sizeof(*sfep)) >= endp)
1070 			return __this_address;
1071 
1072 		/* Don't allow names with known bad length. */
1073 		if (sfep->namelen == 0)
1074 			return __this_address;
1075 
1076 		/*
1077 		 * Check that the variable-length part of the structure is
1078 		 * within the data buffer.  The next entry starts after the
1079 		 * name component, so nextentry is an acceptable test.
1080 		 */
1081 		next_sfep = xfs_attr_sf_nextentry(sfep);
1082 		if ((char *)next_sfep > endp)
1083 			return __this_address;
1084 
1085 		/*
1086 		 * Check for unknown flags.  Short form doesn't support
1087 		 * the incomplete or local bits, so we can use the namespace
1088 		 * mask here.
1089 		 */
1090 		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1091 			return __this_address;
1092 
1093 		/*
1094 		 * Check for invalid namespace combinations.  We only allow
1095 		 * one namespace flag per xattr, so we can just count the
1096 		 * bits (i.e. hweight) here.
1097 		 */
1098 		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1099 			return __this_address;
1100 
1101 		sfep = next_sfep;
1102 	}
1103 	if ((void *)sfep != (void *)endp)
1104 		return __this_address;
1105 
1106 	return NULL;
1107 }
1108 
1109 /*
1110  * Convert a leaf attribute list to shortform attribute list
1111  */
1112 int
1113 xfs_attr3_leaf_to_shortform(
1114 	struct xfs_buf		*bp,
1115 	struct xfs_da_args	*args,
1116 	int			forkoff)
1117 {
1118 	struct xfs_attr_leafblock *leaf;
1119 	struct xfs_attr3_icleaf_hdr ichdr;
1120 	struct xfs_attr_leaf_entry *entry;
1121 	struct xfs_attr_leaf_name_local *name_loc;
1122 	struct xfs_da_args	nargs;
1123 	struct xfs_inode	*dp = args->dp;
1124 	char			*tmpbuffer;
1125 	int			error;
1126 	int			i;
1127 
1128 	trace_xfs_attr_leaf_to_sf(args);
1129 
1130 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1131 	if (!tmpbuffer)
1132 		return -ENOMEM;
1133 
1134 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1135 
1136 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1137 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1138 	entry = xfs_attr3_leaf_entryp(leaf);
1139 
1140 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1141 	memset(bp->b_addr, 0, args->geo->blksize);
1142 
1143 	/*
1144 	 * Clean out the prior contents of the attribute list.
1145 	 */
1146 	error = xfs_da_shrink_inode(args, 0, bp);
1147 	if (error)
1148 		goto out;
1149 
1150 	if (forkoff == -1) {
1151 		/*
1152 		 * Don't remove the attr fork if this operation is the first
1153 		 * part of a attr replace operations. We're going to add a new
1154 		 * attr immediately, so we need to keep the attr fork around in
1155 		 * this case.
1156 		 */
1157 		if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1158 			ASSERT(xfs_has_attr2(dp->i_mount));
1159 			ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1160 			xfs_attr_fork_remove(dp, args->trans);
1161 		}
1162 		goto out;
1163 	}
1164 
1165 	xfs_attr_shortform_create(args);
1166 
1167 	/*
1168 	 * Copy the attributes
1169 	 */
1170 	memset((char *)&nargs, 0, sizeof(nargs));
1171 	nargs.geo = args->geo;
1172 	nargs.dp = dp;
1173 	nargs.total = args->total;
1174 	nargs.whichfork = XFS_ATTR_FORK;
1175 	nargs.trans = args->trans;
1176 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1177 
1178 	for (i = 0; i < ichdr.count; entry++, i++) {
1179 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1180 			continue;	/* don't copy partial entries */
1181 		if (!entry->nameidx)
1182 			continue;
1183 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1184 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1185 		nargs.name = name_loc->nameval;
1186 		nargs.namelen = name_loc->namelen;
1187 		nargs.value = &name_loc->nameval[nargs.namelen];
1188 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1189 		nargs.hashval = be32_to_cpu(entry->hashval);
1190 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1191 		xfs_attr_shortform_add(&nargs, forkoff);
1192 	}
1193 	error = 0;
1194 
1195 out:
1196 	kmem_free(tmpbuffer);
1197 	return error;
1198 }
1199 
1200 /*
1201  * Convert from using a single leaf to a root node and a leaf.
1202  */
1203 int
1204 xfs_attr3_leaf_to_node(
1205 	struct xfs_da_args	*args)
1206 {
1207 	struct xfs_attr_leafblock *leaf;
1208 	struct xfs_attr3_icleaf_hdr icleafhdr;
1209 	struct xfs_attr_leaf_entry *entries;
1210 	struct xfs_da3_icnode_hdr icnodehdr;
1211 	struct xfs_da_intnode	*node;
1212 	struct xfs_inode	*dp = args->dp;
1213 	struct xfs_mount	*mp = dp->i_mount;
1214 	struct xfs_buf		*bp1 = NULL;
1215 	struct xfs_buf		*bp2 = NULL;
1216 	xfs_dablk_t		blkno;
1217 	int			error;
1218 
1219 	trace_xfs_attr_leaf_to_node(args);
1220 
1221 	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1222 		error = -EIO;
1223 		goto out;
1224 	}
1225 
1226 	error = xfs_da_grow_inode(args, &blkno);
1227 	if (error)
1228 		goto out;
1229 	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1230 	if (error)
1231 		goto out;
1232 
1233 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1234 	if (error)
1235 		goto out;
1236 
1237 	/* copy leaf to new buffer, update identifiers */
1238 	xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1239 	bp2->b_ops = bp1->b_ops;
1240 	memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1241 	if (xfs_has_crc(mp)) {
1242 		struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1243 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp2));
1244 	}
1245 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1246 
1247 	/*
1248 	 * Set up the new root node.
1249 	 */
1250 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1251 	if (error)
1252 		goto out;
1253 	node = bp1->b_addr;
1254 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1255 
1256 	leaf = bp2->b_addr;
1257 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1258 	entries = xfs_attr3_leaf_entryp(leaf);
1259 
1260 	/* both on-disk, don't endian-flip twice */
1261 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1262 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1263 	icnodehdr.count = 1;
1264 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1265 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1266 	error = 0;
1267 out:
1268 	return error;
1269 }
1270 
1271 /*========================================================================
1272  * Routines used for growing the Btree.
1273  *========================================================================*/
1274 
1275 /*
1276  * Create the initial contents of a leaf attribute list
1277  * or a leaf in a node attribute list.
1278  */
1279 STATIC int
1280 xfs_attr3_leaf_create(
1281 	struct xfs_da_args	*args,
1282 	xfs_dablk_t		blkno,
1283 	struct xfs_buf		**bpp)
1284 {
1285 	struct xfs_attr_leafblock *leaf;
1286 	struct xfs_attr3_icleaf_hdr ichdr;
1287 	struct xfs_inode	*dp = args->dp;
1288 	struct xfs_mount	*mp = dp->i_mount;
1289 	struct xfs_buf		*bp;
1290 	int			error;
1291 
1292 	trace_xfs_attr_leaf_create(args);
1293 
1294 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1295 					    XFS_ATTR_FORK);
1296 	if (error)
1297 		return error;
1298 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1299 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1300 	leaf = bp->b_addr;
1301 	memset(leaf, 0, args->geo->blksize);
1302 
1303 	memset(&ichdr, 0, sizeof(ichdr));
1304 	ichdr.firstused = args->geo->blksize;
1305 
1306 	if (xfs_has_crc(mp)) {
1307 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1308 
1309 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1310 
1311 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1312 		hdr3->owner = cpu_to_be64(dp->i_ino);
1313 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1314 
1315 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1316 	} else {
1317 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1318 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1319 	}
1320 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1321 
1322 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1323 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1324 
1325 	*bpp = bp;
1326 	return 0;
1327 }
1328 
1329 /*
1330  * Split the leaf node, rebalance, then add the new entry.
1331  */
1332 int
1333 xfs_attr3_leaf_split(
1334 	struct xfs_da_state	*state,
1335 	struct xfs_da_state_blk	*oldblk,
1336 	struct xfs_da_state_blk	*newblk)
1337 {
1338 	xfs_dablk_t blkno;
1339 	int error;
1340 
1341 	trace_xfs_attr_leaf_split(state->args);
1342 
1343 	/*
1344 	 * Allocate space for a new leaf node.
1345 	 */
1346 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1347 	error = xfs_da_grow_inode(state->args, &blkno);
1348 	if (error)
1349 		return error;
1350 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1351 	if (error)
1352 		return error;
1353 	newblk->blkno = blkno;
1354 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1355 
1356 	/*
1357 	 * Rebalance the entries across the two leaves.
1358 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1359 	 */
1360 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1361 	error = xfs_da3_blk_link(state, oldblk, newblk);
1362 	if (error)
1363 		return error;
1364 
1365 	/*
1366 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1367 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1368 	 * "new" attrs info.  Will need the "old" info to remove it later.
1369 	 *
1370 	 * Insert the "new" entry in the correct block.
1371 	 */
1372 	if (state->inleaf) {
1373 		trace_xfs_attr_leaf_add_old(state->args);
1374 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1375 	} else {
1376 		trace_xfs_attr_leaf_add_new(state->args);
1377 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1378 	}
1379 
1380 	/*
1381 	 * Update last hashval in each block since we added the name.
1382 	 */
1383 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1384 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1385 	return error;
1386 }
1387 
1388 /*
1389  * Add a name to the leaf attribute list structure.
1390  */
1391 int
1392 xfs_attr3_leaf_add(
1393 	struct xfs_buf		*bp,
1394 	struct xfs_da_args	*args)
1395 {
1396 	struct xfs_attr_leafblock *leaf;
1397 	struct xfs_attr3_icleaf_hdr ichdr;
1398 	int			tablesize;
1399 	int			entsize;
1400 	int			sum;
1401 	int			tmp;
1402 	int			i;
1403 
1404 	trace_xfs_attr_leaf_add(args);
1405 
1406 	leaf = bp->b_addr;
1407 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1408 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1409 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1410 
1411 	/*
1412 	 * Search through freemap for first-fit on new name length.
1413 	 * (may need to figure in size of entry struct too)
1414 	 */
1415 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1416 					+ xfs_attr3_leaf_hdr_size(leaf);
1417 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1418 		if (tablesize > ichdr.firstused) {
1419 			sum += ichdr.freemap[i].size;
1420 			continue;
1421 		}
1422 		if (!ichdr.freemap[i].size)
1423 			continue;	/* no space in this map */
1424 		tmp = entsize;
1425 		if (ichdr.freemap[i].base < ichdr.firstused)
1426 			tmp += sizeof(xfs_attr_leaf_entry_t);
1427 		if (ichdr.freemap[i].size >= tmp) {
1428 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1429 			goto out_log_hdr;
1430 		}
1431 		sum += ichdr.freemap[i].size;
1432 	}
1433 
1434 	/*
1435 	 * If there are no holes in the address space of the block,
1436 	 * and we don't have enough freespace, then compaction will do us
1437 	 * no good and we should just give up.
1438 	 */
1439 	if (!ichdr.holes && sum < entsize)
1440 		return -ENOSPC;
1441 
1442 	/*
1443 	 * Compact the entries to coalesce free space.
1444 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1445 	 */
1446 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1447 
1448 	/*
1449 	 * After compaction, the block is guaranteed to have only one
1450 	 * free region, in freemap[0].  If it is not big enough, give up.
1451 	 */
1452 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1453 		tmp = -ENOSPC;
1454 		goto out_log_hdr;
1455 	}
1456 
1457 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1458 
1459 out_log_hdr:
1460 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1461 	xfs_trans_log_buf(args->trans, bp,
1462 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1463 				xfs_attr3_leaf_hdr_size(leaf)));
1464 	return tmp;
1465 }
1466 
1467 /*
1468  * Add a name to a leaf attribute list structure.
1469  */
1470 STATIC int
1471 xfs_attr3_leaf_add_work(
1472 	struct xfs_buf		*bp,
1473 	struct xfs_attr3_icleaf_hdr *ichdr,
1474 	struct xfs_da_args	*args,
1475 	int			mapindex)
1476 {
1477 	struct xfs_attr_leafblock *leaf;
1478 	struct xfs_attr_leaf_entry *entry;
1479 	struct xfs_attr_leaf_name_local *name_loc;
1480 	struct xfs_attr_leaf_name_remote *name_rmt;
1481 	struct xfs_mount	*mp;
1482 	int			tmp;
1483 	int			i;
1484 
1485 	trace_xfs_attr_leaf_add_work(args);
1486 
1487 	leaf = bp->b_addr;
1488 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1489 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1490 
1491 	/*
1492 	 * Force open some space in the entry array and fill it in.
1493 	 */
1494 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1495 	if (args->index < ichdr->count) {
1496 		tmp  = ichdr->count - args->index;
1497 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1498 		memmove(entry + 1, entry, tmp);
1499 		xfs_trans_log_buf(args->trans, bp,
1500 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1501 	}
1502 	ichdr->count++;
1503 
1504 	/*
1505 	 * Allocate space for the new string (at the end of the run).
1506 	 */
1507 	mp = args->trans->t_mountp;
1508 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1509 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1510 	ASSERT(ichdr->freemap[mapindex].size >=
1511 		xfs_attr_leaf_newentsize(args, NULL));
1512 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1513 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1514 
1515 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1516 
1517 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1518 				     ichdr->freemap[mapindex].size);
1519 	entry->hashval = cpu_to_be32(args->hashval);
1520 	entry->flags = args->attr_filter;
1521 	if (tmp)
1522 		entry->flags |= XFS_ATTR_LOCAL;
1523 	if (args->op_flags & XFS_DA_OP_REPLACE) {
1524 		if (!xfs_has_larp(mp))
1525 			entry->flags |= XFS_ATTR_INCOMPLETE;
1526 		if ((args->blkno2 == args->blkno) &&
1527 		    (args->index2 <= args->index)) {
1528 			args->index2++;
1529 		}
1530 	}
1531 	xfs_trans_log_buf(args->trans, bp,
1532 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1533 	ASSERT((args->index == 0) ||
1534 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1535 	ASSERT((args->index == ichdr->count - 1) ||
1536 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1537 
1538 	/*
1539 	 * For "remote" attribute values, simply note that we need to
1540 	 * allocate space for the "remote" value.  We can't actually
1541 	 * allocate the extents in this transaction, and we can't decide
1542 	 * which blocks they should be as we might allocate more blocks
1543 	 * as part of this transaction (a split operation for example).
1544 	 */
1545 	if (entry->flags & XFS_ATTR_LOCAL) {
1546 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1547 		name_loc->namelen = args->namelen;
1548 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1549 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1550 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1551 				   be16_to_cpu(name_loc->valuelen));
1552 	} else {
1553 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1554 		name_rmt->namelen = args->namelen;
1555 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1556 		entry->flags |= XFS_ATTR_INCOMPLETE;
1557 		/* just in case */
1558 		name_rmt->valuelen = 0;
1559 		name_rmt->valueblk = 0;
1560 		args->rmtblkno = 1;
1561 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1562 		args->rmtvaluelen = args->valuelen;
1563 	}
1564 	xfs_trans_log_buf(args->trans, bp,
1565 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1566 				   xfs_attr_leaf_entsize(leaf, args->index)));
1567 
1568 	/*
1569 	 * Update the control info for this leaf node
1570 	 */
1571 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1572 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1573 
1574 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1575 					+ xfs_attr3_leaf_hdr_size(leaf));
1576 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1577 					+ xfs_attr3_leaf_hdr_size(leaf);
1578 
1579 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1580 		if (ichdr->freemap[i].base == tmp) {
1581 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1582 			ichdr->freemap[i].size -=
1583 				min_t(uint16_t, ichdr->freemap[i].size,
1584 						sizeof(xfs_attr_leaf_entry_t));
1585 		}
1586 	}
1587 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1588 	return 0;
1589 }
1590 
1591 /*
1592  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1593  */
1594 STATIC void
1595 xfs_attr3_leaf_compact(
1596 	struct xfs_da_args	*args,
1597 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1598 	struct xfs_buf		*bp)
1599 {
1600 	struct xfs_attr_leafblock *leaf_src;
1601 	struct xfs_attr_leafblock *leaf_dst;
1602 	struct xfs_attr3_icleaf_hdr ichdr_src;
1603 	struct xfs_trans	*trans = args->trans;
1604 	char			*tmpbuffer;
1605 
1606 	trace_xfs_attr_leaf_compact(args);
1607 
1608 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1609 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1610 	memset(bp->b_addr, 0, args->geo->blksize);
1611 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1612 	leaf_dst = bp->b_addr;
1613 
1614 	/*
1615 	 * Copy the on-disk header back into the destination buffer to ensure
1616 	 * all the information in the header that is not part of the incore
1617 	 * header structure is preserved.
1618 	 */
1619 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1620 
1621 	/* Initialise the incore headers */
1622 	ichdr_src = *ichdr_dst;	/* struct copy */
1623 	ichdr_dst->firstused = args->geo->blksize;
1624 	ichdr_dst->usedbytes = 0;
1625 	ichdr_dst->count = 0;
1626 	ichdr_dst->holes = 0;
1627 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1628 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1629 						ichdr_dst->freemap[0].base;
1630 
1631 	/* write the header back to initialise the underlying buffer */
1632 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1633 
1634 	/*
1635 	 * Copy all entry's in the same (sorted) order,
1636 	 * but allocate name/value pairs packed and in sequence.
1637 	 */
1638 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1639 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1640 	/*
1641 	 * this logs the entire buffer, but the caller must write the header
1642 	 * back to the buffer when it is finished modifying it.
1643 	 */
1644 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1645 
1646 	kmem_free(tmpbuffer);
1647 }
1648 
1649 /*
1650  * Compare two leaf blocks "order".
1651  * Return 0 unless leaf2 should go before leaf1.
1652  */
1653 static int
1654 xfs_attr3_leaf_order(
1655 	struct xfs_buf	*leaf1_bp,
1656 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1657 	struct xfs_buf	*leaf2_bp,
1658 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1659 {
1660 	struct xfs_attr_leaf_entry *entries1;
1661 	struct xfs_attr_leaf_entry *entries2;
1662 
1663 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1664 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1665 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1666 	    ((be32_to_cpu(entries2[0].hashval) <
1667 	      be32_to_cpu(entries1[0].hashval)) ||
1668 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1669 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1670 		return 1;
1671 	}
1672 	return 0;
1673 }
1674 
1675 int
1676 xfs_attr_leaf_order(
1677 	struct xfs_buf	*leaf1_bp,
1678 	struct xfs_buf	*leaf2_bp)
1679 {
1680 	struct xfs_attr3_icleaf_hdr ichdr1;
1681 	struct xfs_attr3_icleaf_hdr ichdr2;
1682 	struct xfs_mount *mp = leaf1_bp->b_mount;
1683 
1684 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1685 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1686 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1687 }
1688 
1689 /*
1690  * Redistribute the attribute list entries between two leaf nodes,
1691  * taking into account the size of the new entry.
1692  *
1693  * NOTE: if new block is empty, then it will get the upper half of the
1694  * old block.  At present, all (one) callers pass in an empty second block.
1695  *
1696  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1697  * to match what it is doing in splitting the attribute leaf block.  Those
1698  * values are used in "atomic rename" operations on attributes.  Note that
1699  * the "new" and "old" values can end up in different blocks.
1700  */
1701 STATIC void
1702 xfs_attr3_leaf_rebalance(
1703 	struct xfs_da_state	*state,
1704 	struct xfs_da_state_blk	*blk1,
1705 	struct xfs_da_state_blk	*blk2)
1706 {
1707 	struct xfs_da_args	*args;
1708 	struct xfs_attr_leafblock *leaf1;
1709 	struct xfs_attr_leafblock *leaf2;
1710 	struct xfs_attr3_icleaf_hdr ichdr1;
1711 	struct xfs_attr3_icleaf_hdr ichdr2;
1712 	struct xfs_attr_leaf_entry *entries1;
1713 	struct xfs_attr_leaf_entry *entries2;
1714 	int			count;
1715 	int			totallen;
1716 	int			max;
1717 	int			space;
1718 	int			swap;
1719 
1720 	/*
1721 	 * Set up environment.
1722 	 */
1723 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1724 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1725 	leaf1 = blk1->bp->b_addr;
1726 	leaf2 = blk2->bp->b_addr;
1727 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1728 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1729 	ASSERT(ichdr2.count == 0);
1730 	args = state->args;
1731 
1732 	trace_xfs_attr_leaf_rebalance(args);
1733 
1734 	/*
1735 	 * Check ordering of blocks, reverse if it makes things simpler.
1736 	 *
1737 	 * NOTE: Given that all (current) callers pass in an empty
1738 	 * second block, this code should never set "swap".
1739 	 */
1740 	swap = 0;
1741 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1742 		swap(blk1, blk2);
1743 
1744 		/* swap structures rather than reconverting them */
1745 		swap(ichdr1, ichdr2);
1746 
1747 		leaf1 = blk1->bp->b_addr;
1748 		leaf2 = blk2->bp->b_addr;
1749 		swap = 1;
1750 	}
1751 
1752 	/*
1753 	 * Examine entries until we reduce the absolute difference in
1754 	 * byte usage between the two blocks to a minimum.  Then get
1755 	 * the direction to copy and the number of elements to move.
1756 	 *
1757 	 * "inleaf" is true if the new entry should be inserted into blk1.
1758 	 * If "swap" is also true, then reverse the sense of "inleaf".
1759 	 */
1760 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1761 						      blk2, &ichdr2,
1762 						      &count, &totallen);
1763 	if (swap)
1764 		state->inleaf = !state->inleaf;
1765 
1766 	/*
1767 	 * Move any entries required from leaf to leaf:
1768 	 */
1769 	if (count < ichdr1.count) {
1770 		/*
1771 		 * Figure the total bytes to be added to the destination leaf.
1772 		 */
1773 		/* number entries being moved */
1774 		count = ichdr1.count - count;
1775 		space  = ichdr1.usedbytes - totallen;
1776 		space += count * sizeof(xfs_attr_leaf_entry_t);
1777 
1778 		/*
1779 		 * leaf2 is the destination, compact it if it looks tight.
1780 		 */
1781 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1782 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1783 		if (space > max)
1784 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1785 
1786 		/*
1787 		 * Move high entries from leaf1 to low end of leaf2.
1788 		 */
1789 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1790 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1791 
1792 	} else if (count > ichdr1.count) {
1793 		/*
1794 		 * I assert that since all callers pass in an empty
1795 		 * second buffer, this code should never execute.
1796 		 */
1797 		ASSERT(0);
1798 
1799 		/*
1800 		 * Figure the total bytes to be added to the destination leaf.
1801 		 */
1802 		/* number entries being moved */
1803 		count -= ichdr1.count;
1804 		space  = totallen - ichdr1.usedbytes;
1805 		space += count * sizeof(xfs_attr_leaf_entry_t);
1806 
1807 		/*
1808 		 * leaf1 is the destination, compact it if it looks tight.
1809 		 */
1810 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1811 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1812 		if (space > max)
1813 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1814 
1815 		/*
1816 		 * Move low entries from leaf2 to high end of leaf1.
1817 		 */
1818 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1819 					ichdr1.count, count);
1820 	}
1821 
1822 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1823 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1824 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1825 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1826 
1827 	/*
1828 	 * Copy out last hashval in each block for B-tree code.
1829 	 */
1830 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1831 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1832 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1833 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1834 
1835 	/*
1836 	 * Adjust the expected index for insertion.
1837 	 * NOTE: this code depends on the (current) situation that the
1838 	 * second block was originally empty.
1839 	 *
1840 	 * If the insertion point moved to the 2nd block, we must adjust
1841 	 * the index.  We must also track the entry just following the
1842 	 * new entry for use in an "atomic rename" operation, that entry
1843 	 * is always the "old" entry and the "new" entry is what we are
1844 	 * inserting.  The index/blkno fields refer to the "old" entry,
1845 	 * while the index2/blkno2 fields refer to the "new" entry.
1846 	 */
1847 	if (blk1->index > ichdr1.count) {
1848 		ASSERT(state->inleaf == 0);
1849 		blk2->index = blk1->index - ichdr1.count;
1850 		args->index = args->index2 = blk2->index;
1851 		args->blkno = args->blkno2 = blk2->blkno;
1852 	} else if (blk1->index == ichdr1.count) {
1853 		if (state->inleaf) {
1854 			args->index = blk1->index;
1855 			args->blkno = blk1->blkno;
1856 			args->index2 = 0;
1857 			args->blkno2 = blk2->blkno;
1858 		} else {
1859 			/*
1860 			 * On a double leaf split, the original attr location
1861 			 * is already stored in blkno2/index2, so don't
1862 			 * overwrite it overwise we corrupt the tree.
1863 			 */
1864 			blk2->index = blk1->index - ichdr1.count;
1865 			args->index = blk2->index;
1866 			args->blkno = blk2->blkno;
1867 			if (!state->extravalid) {
1868 				/*
1869 				 * set the new attr location to match the old
1870 				 * one and let the higher level split code
1871 				 * decide where in the leaf to place it.
1872 				 */
1873 				args->index2 = blk2->index;
1874 				args->blkno2 = blk2->blkno;
1875 			}
1876 		}
1877 	} else {
1878 		ASSERT(state->inleaf == 1);
1879 		args->index = args->index2 = blk1->index;
1880 		args->blkno = args->blkno2 = blk1->blkno;
1881 	}
1882 }
1883 
1884 /*
1885  * Examine entries until we reduce the absolute difference in
1886  * byte usage between the two blocks to a minimum.
1887  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1888  * GROT: there will always be enough room in either block for a new entry.
1889  * GROT: Do a double-split for this case?
1890  */
1891 STATIC int
1892 xfs_attr3_leaf_figure_balance(
1893 	struct xfs_da_state		*state,
1894 	struct xfs_da_state_blk		*blk1,
1895 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1896 	struct xfs_da_state_blk		*blk2,
1897 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1898 	int				*countarg,
1899 	int				*usedbytesarg)
1900 {
1901 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1902 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1903 	struct xfs_attr_leaf_entry	*entry;
1904 	int				count;
1905 	int				max;
1906 	int				index;
1907 	int				totallen = 0;
1908 	int				half;
1909 	int				lastdelta;
1910 	int				foundit = 0;
1911 	int				tmp;
1912 
1913 	/*
1914 	 * Examine entries until we reduce the absolute difference in
1915 	 * byte usage between the two blocks to a minimum.
1916 	 */
1917 	max = ichdr1->count + ichdr2->count;
1918 	half = (max + 1) * sizeof(*entry);
1919 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1920 			xfs_attr_leaf_newentsize(state->args, NULL);
1921 	half /= 2;
1922 	lastdelta = state->args->geo->blksize;
1923 	entry = xfs_attr3_leaf_entryp(leaf1);
1924 	for (count = index = 0; count < max; entry++, index++, count++) {
1925 
1926 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1927 		/*
1928 		 * The new entry is in the first block, account for it.
1929 		 */
1930 		if (count == blk1->index) {
1931 			tmp = totallen + sizeof(*entry) +
1932 				xfs_attr_leaf_newentsize(state->args, NULL);
1933 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1934 				break;
1935 			lastdelta = XFS_ATTR_ABS(half - tmp);
1936 			totallen = tmp;
1937 			foundit = 1;
1938 		}
1939 
1940 		/*
1941 		 * Wrap around into the second block if necessary.
1942 		 */
1943 		if (count == ichdr1->count) {
1944 			leaf1 = leaf2;
1945 			entry = xfs_attr3_leaf_entryp(leaf1);
1946 			index = 0;
1947 		}
1948 
1949 		/*
1950 		 * Figure out if next leaf entry would be too much.
1951 		 */
1952 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1953 									index);
1954 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1955 			break;
1956 		lastdelta = XFS_ATTR_ABS(half - tmp);
1957 		totallen = tmp;
1958 #undef XFS_ATTR_ABS
1959 	}
1960 
1961 	/*
1962 	 * Calculate the number of usedbytes that will end up in lower block.
1963 	 * If new entry not in lower block, fix up the count.
1964 	 */
1965 	totallen -= count * sizeof(*entry);
1966 	if (foundit) {
1967 		totallen -= sizeof(*entry) +
1968 				xfs_attr_leaf_newentsize(state->args, NULL);
1969 	}
1970 
1971 	*countarg = count;
1972 	*usedbytesarg = totallen;
1973 	return foundit;
1974 }
1975 
1976 /*========================================================================
1977  * Routines used for shrinking the Btree.
1978  *========================================================================*/
1979 
1980 /*
1981  * Check a leaf block and its neighbors to see if the block should be
1982  * collapsed into one or the other neighbor.  Always keep the block
1983  * with the smaller block number.
1984  * If the current block is over 50% full, don't try to join it, return 0.
1985  * If the block is empty, fill in the state structure and return 2.
1986  * If it can be collapsed, fill in the state structure and return 1.
1987  * If nothing can be done, return 0.
1988  *
1989  * GROT: allow for INCOMPLETE entries in calculation.
1990  */
1991 int
1992 xfs_attr3_leaf_toosmall(
1993 	struct xfs_da_state	*state,
1994 	int			*action)
1995 {
1996 	struct xfs_attr_leafblock *leaf;
1997 	struct xfs_da_state_blk	*blk;
1998 	struct xfs_attr3_icleaf_hdr ichdr;
1999 	struct xfs_buf		*bp;
2000 	xfs_dablk_t		blkno;
2001 	int			bytes;
2002 	int			forward;
2003 	int			error;
2004 	int			retval;
2005 	int			i;
2006 
2007 	trace_xfs_attr_leaf_toosmall(state->args);
2008 
2009 	/*
2010 	 * Check for the degenerate case of the block being over 50% full.
2011 	 * If so, it's not worth even looking to see if we might be able
2012 	 * to coalesce with a sibling.
2013 	 */
2014 	blk = &state->path.blk[ state->path.active-1 ];
2015 	leaf = blk->bp->b_addr;
2016 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
2017 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
2018 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2019 		ichdr.usedbytes;
2020 	if (bytes > (state->args->geo->blksize >> 1)) {
2021 		*action = 0;	/* blk over 50%, don't try to join */
2022 		return 0;
2023 	}
2024 
2025 	/*
2026 	 * Check for the degenerate case of the block being empty.
2027 	 * If the block is empty, we'll simply delete it, no need to
2028 	 * coalesce it with a sibling block.  We choose (arbitrarily)
2029 	 * to merge with the forward block unless it is NULL.
2030 	 */
2031 	if (ichdr.count == 0) {
2032 		/*
2033 		 * Make altpath point to the block we want to keep and
2034 		 * path point to the block we want to drop (this one).
2035 		 */
2036 		forward = (ichdr.forw != 0);
2037 		memcpy(&state->altpath, &state->path, sizeof(state->path));
2038 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2039 						 0, &retval);
2040 		if (error)
2041 			return error;
2042 		if (retval) {
2043 			*action = 0;
2044 		} else {
2045 			*action = 2;
2046 		}
2047 		return 0;
2048 	}
2049 
2050 	/*
2051 	 * Examine each sibling block to see if we can coalesce with
2052 	 * at least 25% free space to spare.  We need to figure out
2053 	 * whether to merge with the forward or the backward block.
2054 	 * We prefer coalescing with the lower numbered sibling so as
2055 	 * to shrink an attribute list over time.
2056 	 */
2057 	/* start with smaller blk num */
2058 	forward = ichdr.forw < ichdr.back;
2059 	for (i = 0; i < 2; forward = !forward, i++) {
2060 		struct xfs_attr3_icleaf_hdr ichdr2;
2061 		if (forward)
2062 			blkno = ichdr.forw;
2063 		else
2064 			blkno = ichdr.back;
2065 		if (blkno == 0)
2066 			continue;
2067 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2068 					blkno, &bp);
2069 		if (error)
2070 			return error;
2071 
2072 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2073 
2074 		bytes = state->args->geo->blksize -
2075 			(state->args->geo->blksize >> 2) -
2076 			ichdr.usedbytes - ichdr2.usedbytes -
2077 			((ichdr.count + ichdr2.count) *
2078 					sizeof(xfs_attr_leaf_entry_t)) -
2079 			xfs_attr3_leaf_hdr_size(leaf);
2080 
2081 		xfs_trans_brelse(state->args->trans, bp);
2082 		if (bytes >= 0)
2083 			break;	/* fits with at least 25% to spare */
2084 	}
2085 	if (i >= 2) {
2086 		*action = 0;
2087 		return 0;
2088 	}
2089 
2090 	/*
2091 	 * Make altpath point to the block we want to keep (the lower
2092 	 * numbered block) and path point to the block we want to drop.
2093 	 */
2094 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2095 	if (blkno < blk->blkno) {
2096 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2097 						 0, &retval);
2098 	} else {
2099 		error = xfs_da3_path_shift(state, &state->path, forward,
2100 						 0, &retval);
2101 	}
2102 	if (error)
2103 		return error;
2104 	if (retval) {
2105 		*action = 0;
2106 	} else {
2107 		*action = 1;
2108 	}
2109 	return 0;
2110 }
2111 
2112 /*
2113  * Remove a name from the leaf attribute list structure.
2114  *
2115  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2116  * If two leaves are 37% full, when combined they will leave 25% free.
2117  */
2118 int
2119 xfs_attr3_leaf_remove(
2120 	struct xfs_buf		*bp,
2121 	struct xfs_da_args	*args)
2122 {
2123 	struct xfs_attr_leafblock *leaf;
2124 	struct xfs_attr3_icleaf_hdr ichdr;
2125 	struct xfs_attr_leaf_entry *entry;
2126 	int			before;
2127 	int			after;
2128 	int			smallest;
2129 	int			entsize;
2130 	int			tablesize;
2131 	int			tmp;
2132 	int			i;
2133 
2134 	trace_xfs_attr_leaf_remove(args);
2135 
2136 	leaf = bp->b_addr;
2137 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2138 
2139 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2140 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2141 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2142 					xfs_attr3_leaf_hdr_size(leaf));
2143 
2144 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2145 
2146 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2147 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2148 
2149 	/*
2150 	 * Scan through free region table:
2151 	 *    check for adjacency of free'd entry with an existing one,
2152 	 *    find smallest free region in case we need to replace it,
2153 	 *    adjust any map that borders the entry table,
2154 	 */
2155 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2156 					+ xfs_attr3_leaf_hdr_size(leaf);
2157 	tmp = ichdr.freemap[0].size;
2158 	before = after = -1;
2159 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2160 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2161 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2162 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2163 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2164 		if (ichdr.freemap[i].base == tablesize) {
2165 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2166 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2167 		}
2168 
2169 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2170 				be16_to_cpu(entry->nameidx)) {
2171 			before = i;
2172 		} else if (ichdr.freemap[i].base ==
2173 				(be16_to_cpu(entry->nameidx) + entsize)) {
2174 			after = i;
2175 		} else if (ichdr.freemap[i].size < tmp) {
2176 			tmp = ichdr.freemap[i].size;
2177 			smallest = i;
2178 		}
2179 	}
2180 
2181 	/*
2182 	 * Coalesce adjacent freemap regions,
2183 	 * or replace the smallest region.
2184 	 */
2185 	if ((before >= 0) || (after >= 0)) {
2186 		if ((before >= 0) && (after >= 0)) {
2187 			ichdr.freemap[before].size += entsize;
2188 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2189 			ichdr.freemap[after].base = 0;
2190 			ichdr.freemap[after].size = 0;
2191 		} else if (before >= 0) {
2192 			ichdr.freemap[before].size += entsize;
2193 		} else {
2194 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2195 			ichdr.freemap[after].size += entsize;
2196 		}
2197 	} else {
2198 		/*
2199 		 * Replace smallest region (if it is smaller than free'd entry)
2200 		 */
2201 		if (ichdr.freemap[smallest].size < entsize) {
2202 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2203 			ichdr.freemap[smallest].size = entsize;
2204 		}
2205 	}
2206 
2207 	/*
2208 	 * Did we remove the first entry?
2209 	 */
2210 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2211 		smallest = 1;
2212 	else
2213 		smallest = 0;
2214 
2215 	/*
2216 	 * Compress the remaining entries and zero out the removed stuff.
2217 	 */
2218 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2219 	ichdr.usedbytes -= entsize;
2220 	xfs_trans_log_buf(args->trans, bp,
2221 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2222 				   entsize));
2223 
2224 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2225 	memmove(entry, entry + 1, tmp);
2226 	ichdr.count--;
2227 	xfs_trans_log_buf(args->trans, bp,
2228 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2229 
2230 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2231 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2232 
2233 	/*
2234 	 * If we removed the first entry, re-find the first used byte
2235 	 * in the name area.  Note that if the entry was the "firstused",
2236 	 * then we don't have a "hole" in our block resulting from
2237 	 * removing the name.
2238 	 */
2239 	if (smallest) {
2240 		tmp = args->geo->blksize;
2241 		entry = xfs_attr3_leaf_entryp(leaf);
2242 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2243 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2244 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2245 
2246 			if (be16_to_cpu(entry->nameidx) < tmp)
2247 				tmp = be16_to_cpu(entry->nameidx);
2248 		}
2249 		ichdr.firstused = tmp;
2250 		ASSERT(ichdr.firstused != 0);
2251 	} else {
2252 		ichdr.holes = 1;	/* mark as needing compaction */
2253 	}
2254 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2255 	xfs_trans_log_buf(args->trans, bp,
2256 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2257 					  xfs_attr3_leaf_hdr_size(leaf)));
2258 
2259 	/*
2260 	 * Check if leaf is less than 50% full, caller may want to
2261 	 * "join" the leaf with a sibling if so.
2262 	 */
2263 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2264 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2265 
2266 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2267 }
2268 
2269 /*
2270  * Move all the attribute list entries from drop_leaf into save_leaf.
2271  */
2272 void
2273 xfs_attr3_leaf_unbalance(
2274 	struct xfs_da_state	*state,
2275 	struct xfs_da_state_blk	*drop_blk,
2276 	struct xfs_da_state_blk	*save_blk)
2277 {
2278 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2279 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2280 	struct xfs_attr3_icleaf_hdr drophdr;
2281 	struct xfs_attr3_icleaf_hdr savehdr;
2282 	struct xfs_attr_leaf_entry *entry;
2283 
2284 	trace_xfs_attr_leaf_unbalance(state->args);
2285 
2286 	drop_leaf = drop_blk->bp->b_addr;
2287 	save_leaf = save_blk->bp->b_addr;
2288 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2289 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2290 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2291 
2292 	/*
2293 	 * Save last hashval from dying block for later Btree fixup.
2294 	 */
2295 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2296 
2297 	/*
2298 	 * Check if we need a temp buffer, or can we do it in place.
2299 	 * Note that we don't check "leaf" for holes because we will
2300 	 * always be dropping it, toosmall() decided that for us already.
2301 	 */
2302 	if (savehdr.holes == 0) {
2303 		/*
2304 		 * dest leaf has no holes, so we add there.  May need
2305 		 * to make some room in the entry array.
2306 		 */
2307 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2308 					 drop_blk->bp, &drophdr)) {
2309 			xfs_attr3_leaf_moveents(state->args,
2310 						drop_leaf, &drophdr, 0,
2311 						save_leaf, &savehdr, 0,
2312 						drophdr.count);
2313 		} else {
2314 			xfs_attr3_leaf_moveents(state->args,
2315 						drop_leaf, &drophdr, 0,
2316 						save_leaf, &savehdr,
2317 						savehdr.count, drophdr.count);
2318 		}
2319 	} else {
2320 		/*
2321 		 * Destination has holes, so we make a temporary copy
2322 		 * of the leaf and add them both to that.
2323 		 */
2324 		struct xfs_attr_leafblock *tmp_leaf;
2325 		struct xfs_attr3_icleaf_hdr tmphdr;
2326 
2327 		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2328 
2329 		/*
2330 		 * Copy the header into the temp leaf so that all the stuff
2331 		 * not in the incore header is present and gets copied back in
2332 		 * once we've moved all the entries.
2333 		 */
2334 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2335 
2336 		memset(&tmphdr, 0, sizeof(tmphdr));
2337 		tmphdr.magic = savehdr.magic;
2338 		tmphdr.forw = savehdr.forw;
2339 		tmphdr.back = savehdr.back;
2340 		tmphdr.firstused = state->args->geo->blksize;
2341 
2342 		/* write the header to the temp buffer to initialise it */
2343 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2344 
2345 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2346 					 drop_blk->bp, &drophdr)) {
2347 			xfs_attr3_leaf_moveents(state->args,
2348 						drop_leaf, &drophdr, 0,
2349 						tmp_leaf, &tmphdr, 0,
2350 						drophdr.count);
2351 			xfs_attr3_leaf_moveents(state->args,
2352 						save_leaf, &savehdr, 0,
2353 						tmp_leaf, &tmphdr, tmphdr.count,
2354 						savehdr.count);
2355 		} else {
2356 			xfs_attr3_leaf_moveents(state->args,
2357 						save_leaf, &savehdr, 0,
2358 						tmp_leaf, &tmphdr, 0,
2359 						savehdr.count);
2360 			xfs_attr3_leaf_moveents(state->args,
2361 						drop_leaf, &drophdr, 0,
2362 						tmp_leaf, &tmphdr, tmphdr.count,
2363 						drophdr.count);
2364 		}
2365 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2366 		savehdr = tmphdr; /* struct copy */
2367 		kmem_free(tmp_leaf);
2368 	}
2369 
2370 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2371 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2372 					   state->args->geo->blksize - 1);
2373 
2374 	/*
2375 	 * Copy out last hashval in each block for B-tree code.
2376 	 */
2377 	entry = xfs_attr3_leaf_entryp(save_leaf);
2378 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2379 }
2380 
2381 /*========================================================================
2382  * Routines used for finding things in the Btree.
2383  *========================================================================*/
2384 
2385 /*
2386  * Look up a name in a leaf attribute list structure.
2387  * This is the internal routine, it uses the caller's buffer.
2388  *
2389  * Note that duplicate keys are allowed, but only check within the
2390  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2391  *
2392  * Return in args->index the index into the entry[] array of either
2393  * the found entry, or where the entry should have been (insert before
2394  * that entry).
2395  *
2396  * Don't change the args->value unless we find the attribute.
2397  */
2398 int
2399 xfs_attr3_leaf_lookup_int(
2400 	struct xfs_buf		*bp,
2401 	struct xfs_da_args	*args)
2402 {
2403 	struct xfs_attr_leafblock *leaf;
2404 	struct xfs_attr3_icleaf_hdr ichdr;
2405 	struct xfs_attr_leaf_entry *entry;
2406 	struct xfs_attr_leaf_entry *entries;
2407 	struct xfs_attr_leaf_name_local *name_loc;
2408 	struct xfs_attr_leaf_name_remote *name_rmt;
2409 	xfs_dahash_t		hashval;
2410 	int			probe;
2411 	int			span;
2412 
2413 	trace_xfs_attr_leaf_lookup(args);
2414 
2415 	leaf = bp->b_addr;
2416 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2417 	entries = xfs_attr3_leaf_entryp(leaf);
2418 	if (ichdr.count >= args->geo->blksize / 8) {
2419 		xfs_buf_mark_corrupt(bp);
2420 		return -EFSCORRUPTED;
2421 	}
2422 
2423 	/*
2424 	 * Binary search.  (note: small blocks will skip this loop)
2425 	 */
2426 	hashval = args->hashval;
2427 	probe = span = ichdr.count / 2;
2428 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2429 		span /= 2;
2430 		if (be32_to_cpu(entry->hashval) < hashval)
2431 			probe += span;
2432 		else if (be32_to_cpu(entry->hashval) > hashval)
2433 			probe -= span;
2434 		else
2435 			break;
2436 	}
2437 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2438 		xfs_buf_mark_corrupt(bp);
2439 		return -EFSCORRUPTED;
2440 	}
2441 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2442 		xfs_buf_mark_corrupt(bp);
2443 		return -EFSCORRUPTED;
2444 	}
2445 
2446 	/*
2447 	 * Since we may have duplicate hashval's, find the first matching
2448 	 * hashval in the leaf.
2449 	 */
2450 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2451 		entry--;
2452 		probe--;
2453 	}
2454 	while (probe < ichdr.count &&
2455 	       be32_to_cpu(entry->hashval) < hashval) {
2456 		entry++;
2457 		probe++;
2458 	}
2459 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2460 		args->index = probe;
2461 		return -ENOATTR;
2462 	}
2463 
2464 	/*
2465 	 * Duplicate keys may be present, so search all of them for a match.
2466 	 */
2467 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2468 			entry++, probe++) {
2469 /*
2470  * GROT: Add code to remove incomplete entries.
2471  */
2472 		if (entry->flags & XFS_ATTR_LOCAL) {
2473 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2474 			if (!xfs_attr_match(args, name_loc->namelen,
2475 					name_loc->nameval, entry->flags))
2476 				continue;
2477 			args->index = probe;
2478 			return -EEXIST;
2479 		} else {
2480 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2481 			if (!xfs_attr_match(args, name_rmt->namelen,
2482 					name_rmt->name, entry->flags))
2483 				continue;
2484 			args->index = probe;
2485 			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2486 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2487 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2488 							args->dp->i_mount,
2489 							args->rmtvaluelen);
2490 			return -EEXIST;
2491 		}
2492 	}
2493 	args->index = probe;
2494 	return -ENOATTR;
2495 }
2496 
2497 /*
2498  * Get the value associated with an attribute name from a leaf attribute
2499  * list structure.
2500  *
2501  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2502  * lookup, we only return an error if the attribute does not exist or we can't
2503  * retrieve the value.
2504  */
2505 int
2506 xfs_attr3_leaf_getvalue(
2507 	struct xfs_buf		*bp,
2508 	struct xfs_da_args	*args)
2509 {
2510 	struct xfs_attr_leafblock *leaf;
2511 	struct xfs_attr3_icleaf_hdr ichdr;
2512 	struct xfs_attr_leaf_entry *entry;
2513 	struct xfs_attr_leaf_name_local *name_loc;
2514 	struct xfs_attr_leaf_name_remote *name_rmt;
2515 
2516 	leaf = bp->b_addr;
2517 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2518 	ASSERT(ichdr.count < args->geo->blksize / 8);
2519 	ASSERT(args->index < ichdr.count);
2520 
2521 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2522 	if (entry->flags & XFS_ATTR_LOCAL) {
2523 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2524 		ASSERT(name_loc->namelen == args->namelen);
2525 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2526 		return xfs_attr_copy_value(args,
2527 					&name_loc->nameval[args->namelen],
2528 					be16_to_cpu(name_loc->valuelen));
2529 	}
2530 
2531 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2532 	ASSERT(name_rmt->namelen == args->namelen);
2533 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2534 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2535 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2536 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2537 					       args->rmtvaluelen);
2538 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2539 }
2540 
2541 /*========================================================================
2542  * Utility routines.
2543  *========================================================================*/
2544 
2545 /*
2546  * Move the indicated entries from one leaf to another.
2547  * NOTE: this routine modifies both source and destination leaves.
2548  */
2549 /*ARGSUSED*/
2550 STATIC void
2551 xfs_attr3_leaf_moveents(
2552 	struct xfs_da_args		*args,
2553 	struct xfs_attr_leafblock	*leaf_s,
2554 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2555 	int				start_s,
2556 	struct xfs_attr_leafblock	*leaf_d,
2557 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2558 	int				start_d,
2559 	int				count)
2560 {
2561 	struct xfs_attr_leaf_entry	*entry_s;
2562 	struct xfs_attr_leaf_entry	*entry_d;
2563 	int				desti;
2564 	int				tmp;
2565 	int				i;
2566 
2567 	/*
2568 	 * Check for nothing to do.
2569 	 */
2570 	if (count == 0)
2571 		return;
2572 
2573 	/*
2574 	 * Set up environment.
2575 	 */
2576 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2577 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2578 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2579 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2580 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2581 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2582 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2583 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2584 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2585 
2586 	ASSERT(start_s < ichdr_s->count);
2587 	ASSERT(start_d <= ichdr_d->count);
2588 	ASSERT(count <= ichdr_s->count);
2589 
2590 
2591 	/*
2592 	 * Move the entries in the destination leaf up to make a hole?
2593 	 */
2594 	if (start_d < ichdr_d->count) {
2595 		tmp  = ichdr_d->count - start_d;
2596 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2597 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2598 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2599 		memmove(entry_d, entry_s, tmp);
2600 	}
2601 
2602 	/*
2603 	 * Copy all entry's in the same (sorted) order,
2604 	 * but allocate attribute info packed and in sequence.
2605 	 */
2606 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2607 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2608 	desti = start_d;
2609 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2610 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2611 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2612 #ifdef GROT
2613 		/*
2614 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2615 		 * may also need to change the insertion index.  Code turned
2616 		 * off for 6.2, should be revisited later.
2617 		 */
2618 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2619 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2620 			ichdr_s->usedbytes -= tmp;
2621 			ichdr_s->count -= 1;
2622 			entry_d--;	/* to compensate for ++ in loop hdr */
2623 			desti--;
2624 			if ((start_s + i) < offset)
2625 				result++;	/* insertion index adjustment */
2626 		} else {
2627 #endif /* GROT */
2628 			ichdr_d->firstused -= tmp;
2629 			/* both on-disk, don't endian flip twice */
2630 			entry_d->hashval = entry_s->hashval;
2631 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2632 			entry_d->flags = entry_s->flags;
2633 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2634 							<= args->geo->blksize);
2635 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2636 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2637 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2638 							<= args->geo->blksize);
2639 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2640 			ichdr_s->usedbytes -= tmp;
2641 			ichdr_d->usedbytes += tmp;
2642 			ichdr_s->count -= 1;
2643 			ichdr_d->count += 1;
2644 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2645 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2646 			ASSERT(ichdr_d->firstused >= tmp);
2647 #ifdef GROT
2648 		}
2649 #endif /* GROT */
2650 	}
2651 
2652 	/*
2653 	 * Zero out the entries we just copied.
2654 	 */
2655 	if (start_s == ichdr_s->count) {
2656 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2657 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2658 		ASSERT(((char *)entry_s + tmp) <=
2659 		       ((char *)leaf_s + args->geo->blksize));
2660 		memset(entry_s, 0, tmp);
2661 	} else {
2662 		/*
2663 		 * Move the remaining entries down to fill the hole,
2664 		 * then zero the entries at the top.
2665 		 */
2666 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2667 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2668 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2669 		memmove(entry_d, entry_s, tmp);
2670 
2671 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2672 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2673 		ASSERT(((char *)entry_s + tmp) <=
2674 		       ((char *)leaf_s + args->geo->blksize));
2675 		memset(entry_s, 0, tmp);
2676 	}
2677 
2678 	/*
2679 	 * Fill in the freemap information
2680 	 */
2681 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2682 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2683 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2684 	ichdr_d->freemap[1].base = 0;
2685 	ichdr_d->freemap[2].base = 0;
2686 	ichdr_d->freemap[1].size = 0;
2687 	ichdr_d->freemap[2].size = 0;
2688 	ichdr_s->holes = 1;	/* leaf may not be compact */
2689 }
2690 
2691 /*
2692  * Pick up the last hashvalue from a leaf block.
2693  */
2694 xfs_dahash_t
2695 xfs_attr_leaf_lasthash(
2696 	struct xfs_buf	*bp,
2697 	int		*count)
2698 {
2699 	struct xfs_attr3_icleaf_hdr ichdr;
2700 	struct xfs_attr_leaf_entry *entries;
2701 	struct xfs_mount *mp = bp->b_mount;
2702 
2703 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2704 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2705 	if (count)
2706 		*count = ichdr.count;
2707 	if (!ichdr.count)
2708 		return 0;
2709 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2710 }
2711 
2712 /*
2713  * Calculate the number of bytes used to store the indicated attribute
2714  * (whether local or remote only calculate bytes in this block).
2715  */
2716 STATIC int
2717 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2718 {
2719 	struct xfs_attr_leaf_entry *entries;
2720 	xfs_attr_leaf_name_local_t *name_loc;
2721 	xfs_attr_leaf_name_remote_t *name_rmt;
2722 	int size;
2723 
2724 	entries = xfs_attr3_leaf_entryp(leaf);
2725 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2726 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2727 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2728 						   be16_to_cpu(name_loc->valuelen));
2729 	} else {
2730 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2731 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2732 	}
2733 	return size;
2734 }
2735 
2736 /*
2737  * Calculate the number of bytes that would be required to store the new
2738  * attribute (whether local or remote only calculate bytes in this block).
2739  * This routine decides as a side effect whether the attribute will be
2740  * a "local" or a "remote" attribute.
2741  */
2742 int
2743 xfs_attr_leaf_newentsize(
2744 	struct xfs_da_args	*args,
2745 	int			*local)
2746 {
2747 	int			size;
2748 
2749 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2750 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2751 		if (local)
2752 			*local = 1;
2753 		return size;
2754 	}
2755 	if (local)
2756 		*local = 0;
2757 	return xfs_attr_leaf_entsize_remote(args->namelen);
2758 }
2759 
2760 
2761 /*========================================================================
2762  * Manage the INCOMPLETE flag in a leaf entry
2763  *========================================================================*/
2764 
2765 /*
2766  * Clear the INCOMPLETE flag on an entry in a leaf block.
2767  */
2768 int
2769 xfs_attr3_leaf_clearflag(
2770 	struct xfs_da_args	*args)
2771 {
2772 	struct xfs_attr_leafblock *leaf;
2773 	struct xfs_attr_leaf_entry *entry;
2774 	struct xfs_attr_leaf_name_remote *name_rmt;
2775 	struct xfs_buf		*bp;
2776 	int			error;
2777 #ifdef DEBUG
2778 	struct xfs_attr3_icleaf_hdr ichdr;
2779 	xfs_attr_leaf_name_local_t *name_loc;
2780 	int namelen;
2781 	char *name;
2782 #endif /* DEBUG */
2783 
2784 	trace_xfs_attr_leaf_clearflag(args);
2785 	/*
2786 	 * Set up the operation.
2787 	 */
2788 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2789 	if (error)
2790 		return error;
2791 
2792 	leaf = bp->b_addr;
2793 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2794 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2795 
2796 #ifdef DEBUG
2797 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2798 	ASSERT(args->index < ichdr.count);
2799 	ASSERT(args->index >= 0);
2800 
2801 	if (entry->flags & XFS_ATTR_LOCAL) {
2802 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2803 		namelen = name_loc->namelen;
2804 		name = (char *)name_loc->nameval;
2805 	} else {
2806 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2807 		namelen = name_rmt->namelen;
2808 		name = (char *)name_rmt->name;
2809 	}
2810 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2811 	ASSERT(namelen == args->namelen);
2812 	ASSERT(memcmp(name, args->name, namelen) == 0);
2813 #endif /* DEBUG */
2814 
2815 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2816 	xfs_trans_log_buf(args->trans, bp,
2817 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2818 
2819 	if (args->rmtblkno) {
2820 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2821 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2822 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2823 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2824 		xfs_trans_log_buf(args->trans, bp,
2825 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2826 	}
2827 
2828 	return 0;
2829 }
2830 
2831 /*
2832  * Set the INCOMPLETE flag on an entry in a leaf block.
2833  */
2834 int
2835 xfs_attr3_leaf_setflag(
2836 	struct xfs_da_args	*args)
2837 {
2838 	struct xfs_attr_leafblock *leaf;
2839 	struct xfs_attr_leaf_entry *entry;
2840 	struct xfs_attr_leaf_name_remote *name_rmt;
2841 	struct xfs_buf		*bp;
2842 	int error;
2843 #ifdef DEBUG
2844 	struct xfs_attr3_icleaf_hdr ichdr;
2845 #endif
2846 
2847 	trace_xfs_attr_leaf_setflag(args);
2848 
2849 	/*
2850 	 * Set up the operation.
2851 	 */
2852 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2853 	if (error)
2854 		return error;
2855 
2856 	leaf = bp->b_addr;
2857 #ifdef DEBUG
2858 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2859 	ASSERT(args->index < ichdr.count);
2860 	ASSERT(args->index >= 0);
2861 #endif
2862 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2863 
2864 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2865 	entry->flags |= XFS_ATTR_INCOMPLETE;
2866 	xfs_trans_log_buf(args->trans, bp,
2867 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2868 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2869 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2870 		name_rmt->valueblk = 0;
2871 		name_rmt->valuelen = 0;
2872 		xfs_trans_log_buf(args->trans, bp,
2873 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2874 	}
2875 
2876 	return 0;
2877 }
2878 
2879 /*
2880  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2881  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2882  * entry given by args->blkno2/index2.
2883  *
2884  * Note that they could be in different blocks, or in the same block.
2885  */
2886 int
2887 xfs_attr3_leaf_flipflags(
2888 	struct xfs_da_args	*args)
2889 {
2890 	struct xfs_attr_leafblock *leaf1;
2891 	struct xfs_attr_leafblock *leaf2;
2892 	struct xfs_attr_leaf_entry *entry1;
2893 	struct xfs_attr_leaf_entry *entry2;
2894 	struct xfs_attr_leaf_name_remote *name_rmt;
2895 	struct xfs_buf		*bp1;
2896 	struct xfs_buf		*bp2;
2897 	int error;
2898 #ifdef DEBUG
2899 	struct xfs_attr3_icleaf_hdr ichdr1;
2900 	struct xfs_attr3_icleaf_hdr ichdr2;
2901 	xfs_attr_leaf_name_local_t *name_loc;
2902 	int namelen1, namelen2;
2903 	char *name1, *name2;
2904 #endif /* DEBUG */
2905 
2906 	trace_xfs_attr_leaf_flipflags(args);
2907 
2908 	/*
2909 	 * Read the block containing the "old" attr
2910 	 */
2911 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2912 	if (error)
2913 		return error;
2914 
2915 	/*
2916 	 * Read the block containing the "new" attr, if it is different
2917 	 */
2918 	if (args->blkno2 != args->blkno) {
2919 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2920 					   &bp2);
2921 		if (error)
2922 			return error;
2923 	} else {
2924 		bp2 = bp1;
2925 	}
2926 
2927 	leaf1 = bp1->b_addr;
2928 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2929 
2930 	leaf2 = bp2->b_addr;
2931 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2932 
2933 #ifdef DEBUG
2934 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2935 	ASSERT(args->index < ichdr1.count);
2936 	ASSERT(args->index >= 0);
2937 
2938 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2939 	ASSERT(args->index2 < ichdr2.count);
2940 	ASSERT(args->index2 >= 0);
2941 
2942 	if (entry1->flags & XFS_ATTR_LOCAL) {
2943 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2944 		namelen1 = name_loc->namelen;
2945 		name1 = (char *)name_loc->nameval;
2946 	} else {
2947 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2948 		namelen1 = name_rmt->namelen;
2949 		name1 = (char *)name_rmt->name;
2950 	}
2951 	if (entry2->flags & XFS_ATTR_LOCAL) {
2952 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2953 		namelen2 = name_loc->namelen;
2954 		name2 = (char *)name_loc->nameval;
2955 	} else {
2956 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2957 		namelen2 = name_rmt->namelen;
2958 		name2 = (char *)name_rmt->name;
2959 	}
2960 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2961 	ASSERT(namelen1 == namelen2);
2962 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2963 #endif /* DEBUG */
2964 
2965 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2966 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2967 
2968 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2969 	xfs_trans_log_buf(args->trans, bp1,
2970 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2971 	if (args->rmtblkno) {
2972 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2973 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2974 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2975 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2976 		xfs_trans_log_buf(args->trans, bp1,
2977 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2978 	}
2979 
2980 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2981 	xfs_trans_log_buf(args->trans, bp2,
2982 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2983 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2984 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2985 		name_rmt->valueblk = 0;
2986 		name_rmt->valuelen = 0;
2987 		xfs_trans_log_buf(args->trans, bp2,
2988 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2989 	}
2990 
2991 	return 0;
2992 }
2993