xref: /linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision 6179d4a213006491ff0d50073256f21fad22149b)
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 /*
293  * Validate an attribute leaf block.
294  *
295  * Empty leaf blocks can occur under the following circumstances:
296  *
297  * 1. setxattr adds a new extended attribute to a file;
298  * 2. The file has zero existing attributes;
299  * 3. The attribute is too large to fit in the attribute fork;
300  * 4. The attribute is small enough to fit in a leaf block;
301  * 5. A log flush occurs after committing the transaction that creates
302  *    the (empty) leaf block; and
303  * 6. The filesystem goes down after the log flush but before the new
304  *    attribute can be committed to the leaf block.
305  *
306  * Hence we need to ensure that we don't fail the validation purely
307  * because the leaf is empty.
308  */
309 static xfs_failaddr_t
310 xfs_attr3_leaf_verify(
311 	struct xfs_buf			*bp)
312 {
313 	struct xfs_attr3_icleaf_hdr	ichdr;
314 	struct xfs_mount		*mp = bp->b_mount;
315 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
316 	struct xfs_attr_leaf_entry	*entries;
317 	struct xfs_attr_leaf_entry	*ent;
318 	char				*buf_end;
319 	uint32_t			end;	/* must be 32bit - see below */
320 	__u32				last_hashval = 0;
321 	int				i;
322 	xfs_failaddr_t			fa;
323 
324 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
325 
326 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
327 	if (fa)
328 		return fa;
329 
330 	/*
331 	 * firstused is the block offset of the first name info structure.
332 	 * Make sure it doesn't go off the block or crash into the header.
333 	 */
334 	if (ichdr.firstused > mp->m_attr_geo->blksize)
335 		return __this_address;
336 	if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
337 		return __this_address;
338 
339 	/* Make sure the entries array doesn't crash into the name info. */
340 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
341 	if ((char *)&entries[ichdr.count] >
342 	    (char *)bp->b_addr + ichdr.firstused)
343 		return __this_address;
344 
345 	/*
346 	 * NOTE: This verifier historically failed empty leaf buffers because
347 	 * we expect the fork to be in another format. Empty attr fork format
348 	 * conversions are possible during xattr set, however, and format
349 	 * conversion is not atomic with the xattr set that triggers it. We
350 	 * cannot assume leaf blocks are non-empty until that is addressed.
351 	*/
352 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
353 	for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
354 		fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
355 				ent, i, &last_hashval);
356 		if (fa)
357 			return fa;
358 	}
359 
360 	/*
361 	 * Quickly check the freemap information.  Attribute data has to be
362 	 * aligned to 4-byte boundaries, and likewise for the free space.
363 	 *
364 	 * Note that for 64k block size filesystems, the freemap entries cannot
365 	 * overflow as they are only be16 fields. However, when checking end
366 	 * pointer of the freemap, we have to be careful to detect overflows and
367 	 * so use uint32_t for those checks.
368 	 */
369 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
370 		if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
371 			return __this_address;
372 		if (ichdr.freemap[i].base & 0x3)
373 			return __this_address;
374 		if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
375 			return __this_address;
376 		if (ichdr.freemap[i].size & 0x3)
377 			return __this_address;
378 
379 		/* be care of 16 bit overflows here */
380 		end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
381 		if (end < ichdr.freemap[i].base)
382 			return __this_address;
383 		if (end > mp->m_attr_geo->blksize)
384 			return __this_address;
385 	}
386 
387 	return NULL;
388 }
389 
390 static void
391 xfs_attr3_leaf_write_verify(
392 	struct xfs_buf	*bp)
393 {
394 	struct xfs_mount	*mp = bp->b_mount;
395 	struct xfs_buf_log_item	*bip = bp->b_log_item;
396 	struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
397 	xfs_failaddr_t		fa;
398 
399 	fa = xfs_attr3_leaf_verify(bp);
400 	if (fa) {
401 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
402 		return;
403 	}
404 
405 	if (!xfs_has_crc(mp))
406 		return;
407 
408 	if (bip)
409 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
410 
411 	xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
412 }
413 
414 /*
415  * leaf/node format detection on trees is sketchy, so a node read can be done on
416  * leaf level blocks when detection identifies the tree as a node format tree
417  * incorrectly. In this case, we need to swap the verifier to match the correct
418  * format of the block being read.
419  */
420 static void
421 xfs_attr3_leaf_read_verify(
422 	struct xfs_buf		*bp)
423 {
424 	struct xfs_mount	*mp = bp->b_mount;
425 	xfs_failaddr_t		fa;
426 
427 	if (xfs_has_crc(mp) &&
428 	     !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
429 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
430 	else {
431 		fa = xfs_attr3_leaf_verify(bp);
432 		if (fa)
433 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
434 	}
435 }
436 
437 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
438 	.name = "xfs_attr3_leaf",
439 	.magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
440 		     cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
441 	.verify_read = xfs_attr3_leaf_read_verify,
442 	.verify_write = xfs_attr3_leaf_write_verify,
443 	.verify_struct = xfs_attr3_leaf_verify,
444 };
445 
446 int
447 xfs_attr3_leaf_read(
448 	struct xfs_trans	*tp,
449 	struct xfs_inode	*dp,
450 	xfs_dablk_t		bno,
451 	struct xfs_buf		**bpp)
452 {
453 	int			err;
454 
455 	err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
456 			&xfs_attr3_leaf_buf_ops);
457 	if (!err && tp && *bpp)
458 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
459 	return err;
460 }
461 
462 /*========================================================================
463  * Namespace helper routines
464  *========================================================================*/
465 
466 /*
467  * If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
468  * flag on disk - if there's an incomplete attr then recovery needs to tear it
469  * down. If there's no incomplete attr, then recovery needs to tear that attr
470  * down to replace it with the attr that has been logged. In this case, the
471  * INCOMPLETE flag will not be set in attr->attr_filter, but rather
472  * XFS_DA_OP_RECOVERY will be set in args->op_flags.
473  */
474 static bool
475 xfs_attr_match(
476 	struct xfs_da_args	*args,
477 	uint8_t			namelen,
478 	unsigned char		*name,
479 	int			flags)
480 {
481 
482 	if (args->namelen != namelen)
483 		return false;
484 	if (memcmp(args->name, name, namelen) != 0)
485 		return false;
486 
487 	/* Recovery ignores the INCOMPLETE flag. */
488 	if ((args->op_flags & XFS_DA_OP_RECOVERY) &&
489 	    args->attr_filter == (flags & XFS_ATTR_NSP_ONDISK_MASK))
490 		return true;
491 
492 	/* All remaining matches need to be filtered by INCOMPLETE state. */
493 	if (args->attr_filter !=
494 	    (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
495 		return false;
496 	return true;
497 }
498 
499 static int
500 xfs_attr_copy_value(
501 	struct xfs_da_args	*args,
502 	unsigned char		*value,
503 	int			valuelen)
504 {
505 	/*
506 	 * No copy if all we have to do is get the length
507 	 */
508 	if (!args->valuelen) {
509 		args->valuelen = valuelen;
510 		return 0;
511 	}
512 
513 	/*
514 	 * No copy if the length of the existing buffer is too small
515 	 */
516 	if (args->valuelen < valuelen) {
517 		args->valuelen = valuelen;
518 		return -ERANGE;
519 	}
520 
521 	if (!args->value) {
522 		args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
523 		if (!args->value)
524 			return -ENOMEM;
525 	}
526 	args->valuelen = valuelen;
527 
528 	/* remote block xattr requires IO for copy-in */
529 	if (args->rmtblkno)
530 		return xfs_attr_rmtval_get(args);
531 
532 	/*
533 	 * This is to prevent a GCC warning because the remote xattr case
534 	 * doesn't have a value to pass in. In that case, we never reach here,
535 	 * but GCC can't work that out and so throws a "passing NULL to
536 	 * memcpy" warning.
537 	 */
538 	if (!value)
539 		return -EINVAL;
540 	memcpy(args->value, value, valuelen);
541 	return 0;
542 }
543 
544 /*========================================================================
545  * External routines when attribute fork size < XFS_LITINO(mp).
546  *========================================================================*/
547 
548 /*
549  * Query whether the total requested number of attr fork bytes of extended
550  * attribute space will be able to fit inline.
551  *
552  * Returns zero if not, else the i_forkoff fork offset to be used in the
553  * literal area for attribute data once the new bytes have been added.
554  *
555  * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
556  * special case for dev/uuid inodes, they have fixed size data forks.
557  */
558 int
559 xfs_attr_shortform_bytesfit(
560 	struct xfs_inode	*dp,
561 	int			bytes)
562 {
563 	struct xfs_mount	*mp = dp->i_mount;
564 	int64_t			dsize;
565 	int			minforkoff;
566 	int			maxforkoff;
567 	int			offset;
568 
569 	/*
570 	 * Check if the new size could fit at all first:
571 	 */
572 	if (bytes > XFS_LITINO(mp))
573 		return 0;
574 
575 	/* rounded down */
576 	offset = (XFS_LITINO(mp) - bytes) >> 3;
577 
578 	if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
579 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
580 		return (offset >= minforkoff) ? minforkoff : 0;
581 	}
582 
583 	/*
584 	 * If the requested numbers of bytes is smaller or equal to the
585 	 * current attribute fork size we can always proceed.
586 	 *
587 	 * Note that if_bytes in the data fork might actually be larger than
588 	 * the current data fork size is due to delalloc extents. In that
589 	 * case either the extent count will go down when they are converted
590 	 * to real extents, or the delalloc conversion will take care of the
591 	 * literal area rebalancing.
592 	 */
593 	if (bytes <= xfs_inode_attr_fork_size(dp))
594 		return dp->i_forkoff;
595 
596 	/*
597 	 * For attr2 we can try to move the forkoff if there is space in the
598 	 * literal area, but for the old format we are done if there is no
599 	 * space in the fixed attribute fork.
600 	 */
601 	if (!xfs_has_attr2(mp))
602 		return 0;
603 
604 	dsize = dp->i_df.if_bytes;
605 
606 	switch (dp->i_df.if_format) {
607 	case XFS_DINODE_FMT_EXTENTS:
608 		/*
609 		 * If there is no attr fork and the data fork is extents,
610 		 * determine if creating the default attr fork will result
611 		 * in the extents form migrating to btree. If so, the
612 		 * minimum offset only needs to be the space required for
613 		 * the btree root.
614 		 */
615 		if (!dp->i_forkoff && dp->i_df.if_bytes >
616 		    xfs_default_attroffset(dp))
617 			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
618 		break;
619 	case XFS_DINODE_FMT_BTREE:
620 		/*
621 		 * If we have a data btree then keep forkoff if we have one,
622 		 * otherwise we are adding a new attr, so then we set
623 		 * minforkoff to where the btree root can finish so we have
624 		 * plenty of room for attrs
625 		 */
626 		if (dp->i_forkoff) {
627 			if (offset < dp->i_forkoff)
628 				return 0;
629 			return dp->i_forkoff;
630 		}
631 		dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
632 		break;
633 	}
634 
635 	/*
636 	 * A data fork btree root must have space for at least
637 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
638 	 */
639 	minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
640 	minforkoff = roundup(minforkoff, 8) >> 3;
641 
642 	/* attr fork btree root can have at least this many key/ptr pairs */
643 	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
644 	maxforkoff = maxforkoff >> 3;	/* rounded down */
645 
646 	if (offset >= maxforkoff)
647 		return maxforkoff;
648 	if (offset >= minforkoff)
649 		return offset;
650 	return 0;
651 }
652 
653 /*
654  * Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
655  * - noattr2 mount option is set,
656  * - on-disk version bit says it is already set, or
657  * - the attr2 mount option is not set to enable automatic upgrade from attr1.
658  */
659 STATIC void
660 xfs_sbversion_add_attr2(
661 	struct xfs_mount	*mp,
662 	struct xfs_trans	*tp)
663 {
664 	if (xfs_has_noattr2(mp))
665 		return;
666 	if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
667 		return;
668 	if (!xfs_has_attr2(mp))
669 		return;
670 
671 	spin_lock(&mp->m_sb_lock);
672 	xfs_add_attr2(mp);
673 	spin_unlock(&mp->m_sb_lock);
674 	xfs_log_sb(tp);
675 }
676 
677 /*
678  * Create the initial contents of a shortform attribute list.
679  */
680 void
681 xfs_attr_shortform_create(
682 	struct xfs_da_args	*args)
683 {
684 	struct xfs_inode	*dp = args->dp;
685 	struct xfs_ifork	*ifp = &dp->i_af;
686 	struct xfs_attr_sf_hdr	*hdr;
687 
688 	trace_xfs_attr_sf_create(args);
689 
690 	ASSERT(ifp->if_bytes == 0);
691 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
692 		ifp->if_format = XFS_DINODE_FMT_LOCAL;
693 
694 	hdr = xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
695 	memset(hdr, 0, sizeof(*hdr));
696 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
697 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
698 }
699 
700 /*
701  * Return the entry if the attr in args is found, or NULL if not.
702  */
703 struct xfs_attr_sf_entry *
704 xfs_attr_sf_findname(
705 	struct xfs_da_args		*args)
706 {
707 	struct xfs_attr_sf_hdr		*sf = args->dp->i_af.if_data;
708 	struct xfs_attr_sf_entry	*sfe;
709 
710 	for (sfe = xfs_attr_sf_firstentry(sf);
711 	     sfe < xfs_attr_sf_endptr(sf);
712 	     sfe = xfs_attr_sf_nextentry(sfe)) {
713 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
714 				sfe->flags))
715 			return sfe;
716 	}
717 
718 	return NULL;
719 }
720 
721 /*
722  * Add a name/value pair to the shortform attribute list.
723  * Overflow from the inode has already been checked for.
724  */
725 void
726 xfs_attr_shortform_add(
727 	struct xfs_da_args		*args,
728 	int				forkoff)
729 {
730 	struct xfs_inode		*dp = args->dp;
731 	struct xfs_mount		*mp = dp->i_mount;
732 	struct xfs_ifork		*ifp = &dp->i_af;
733 	struct xfs_attr_sf_hdr		*sf = ifp->if_data;
734 	struct xfs_attr_sf_entry	*sfe;
735 	int				size;
736 
737 	trace_xfs_attr_sf_add(args);
738 
739 	dp->i_forkoff = forkoff;
740 
741 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
742 	ASSERT(!xfs_attr_sf_findname(args));
743 
744 	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
745 	sf = xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
746 
747 	sfe = xfs_attr_sf_endptr(sf);
748 	sfe->namelen = args->namelen;
749 	sfe->valuelen = args->valuelen;
750 	sfe->flags = args->attr_filter;
751 	memcpy(sfe->nameval, args->name, args->namelen);
752 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
753 	sf->count++;
754 	be16_add_cpu(&sf->totsize, size);
755 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
756 
757 	xfs_sbversion_add_attr2(mp, args->trans);
758 }
759 
760 /*
761  * After the last attribute is removed revert to original inode format,
762  * making all literal area available to the data fork once more.
763  */
764 void
765 xfs_attr_fork_remove(
766 	struct xfs_inode	*ip,
767 	struct xfs_trans	*tp)
768 {
769 	ASSERT(ip->i_af.if_nextents == 0);
770 
771 	xfs_ifork_zap_attr(ip);
772 	ip->i_forkoff = 0;
773 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
774 }
775 
776 /*
777  * Remove an attribute from the shortform attribute list structure.
778  */
779 int
780 xfs_attr_sf_removename(
781 	struct xfs_da_args		*args)
782 {
783 	struct xfs_inode		*dp = args->dp;
784 	struct xfs_mount		*mp = dp->i_mount;
785 	struct xfs_attr_sf_hdr		*sf = dp->i_af.if_data;
786 	struct xfs_attr_sf_entry	*sfe;
787 	uint16_t			totsize = be16_to_cpu(sf->totsize);
788 	void				*next, *end;
789 	int				size = 0;
790 
791 	trace_xfs_attr_sf_remove(args);
792 
793 	sfe = xfs_attr_sf_findname(args);
794 	if (!sfe) {
795 		/*
796 		 * If we are recovering an operation, finding nothing to remove
797 		 * is not an error, it just means there was nothing to clean up.
798 		 */
799 		if (args->op_flags & XFS_DA_OP_RECOVERY)
800 			return 0;
801 		return -ENOATTR;
802 	}
803 
804 	/*
805 	 * Fix up the attribute fork data, covering the hole
806 	 */
807 	size = xfs_attr_sf_entsize(sfe);
808 	next = xfs_attr_sf_nextentry(sfe);
809 	end = xfs_attr_sf_endptr(sf);
810 	if (next < end)
811 		memmove(sfe, next, end - next);
812 	sf->count--;
813 	totsize -= size;
814 	sf->totsize = cpu_to_be16(totsize);
815 
816 	/*
817 	 * Fix up the start offset of the attribute fork
818 	 */
819 	if (totsize == sizeof(struct xfs_attr_sf_hdr) && xfs_has_attr2(mp) &&
820 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
821 	    !(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE))) {
822 		xfs_attr_fork_remove(dp, args->trans);
823 	} else {
824 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
825 		dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
826 		ASSERT(dp->i_forkoff);
827 		ASSERT(totsize > sizeof(struct xfs_attr_sf_hdr) ||
828 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
829 				!xfs_has_attr2(mp) ||
830 				dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
831 		xfs_trans_log_inode(args->trans, dp,
832 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
833 	}
834 
835 	xfs_sbversion_add_attr2(mp, args->trans);
836 
837 	return 0;
838 }
839 
840 /*
841  * Retrieve the attribute value and length.
842  *
843  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
844  * lookup, we only return an error if the attribute does not exist or we can't
845  * retrieve the value.
846  */
847 int
848 xfs_attr_shortform_getvalue(
849 	struct xfs_da_args		*args)
850 {
851 	struct xfs_attr_sf_entry	*sfe;
852 
853 	ASSERT(args->dp->i_af.if_format == XFS_DINODE_FMT_LOCAL);
854 
855 	trace_xfs_attr_sf_lookup(args);
856 
857 	sfe = xfs_attr_sf_findname(args);
858 	if (!sfe)
859 		return -ENOATTR;
860 	return xfs_attr_copy_value(args, &sfe->nameval[args->namelen],
861 			sfe->valuelen);
862 }
863 
864 /* Convert from using the shortform to the leaf format. */
865 int
866 xfs_attr_shortform_to_leaf(
867 	struct xfs_da_args		*args)
868 {
869 	struct xfs_inode		*dp = args->dp;
870 	struct xfs_ifork		*ifp = &dp->i_af;
871 	struct xfs_attr_sf_hdr		*sf = ifp->if_data;
872 	struct xfs_attr_sf_entry	*sfe;
873 	int				size = be16_to_cpu(sf->totsize);
874 	struct xfs_da_args		nargs;
875 	char				*tmpbuffer;
876 	int				error, i;
877 	xfs_dablk_t			blkno;
878 	struct xfs_buf			*bp;
879 
880 	trace_xfs_attr_sf_to_leaf(args);
881 
882 	tmpbuffer = kmem_alloc(size, 0);
883 	ASSERT(tmpbuffer != NULL);
884 	memcpy(tmpbuffer, ifp->if_data, size);
885 	sf = (struct xfs_attr_sf_hdr *)tmpbuffer;
886 
887 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
888 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
889 
890 	bp = NULL;
891 	error = xfs_da_grow_inode(args, &blkno);
892 	if (error)
893 		goto out;
894 
895 	ASSERT(blkno == 0);
896 	error = xfs_attr3_leaf_create(args, blkno, &bp);
897 	if (error)
898 		goto out;
899 
900 	memset((char *)&nargs, 0, sizeof(nargs));
901 	nargs.dp = dp;
902 	nargs.geo = args->geo;
903 	nargs.total = args->total;
904 	nargs.whichfork = XFS_ATTR_FORK;
905 	nargs.trans = args->trans;
906 	nargs.op_flags = XFS_DA_OP_OKNOENT;
907 
908 	sfe = xfs_attr_sf_firstentry(sf);
909 	for (i = 0; i < sf->count; i++) {
910 		nargs.name = sfe->nameval;
911 		nargs.namelen = sfe->namelen;
912 		nargs.value = &sfe->nameval[nargs.namelen];
913 		nargs.valuelen = sfe->valuelen;
914 		nargs.hashval = xfs_da_hashname(sfe->nameval,
915 						sfe->namelen);
916 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
917 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
918 		ASSERT(error == -ENOATTR);
919 		error = xfs_attr3_leaf_add(bp, &nargs);
920 		ASSERT(error != -ENOSPC);
921 		if (error)
922 			goto out;
923 		sfe = xfs_attr_sf_nextentry(sfe);
924 	}
925 	error = 0;
926 out:
927 	kmem_free(tmpbuffer);
928 	return error;
929 }
930 
931 /*
932  * Check a leaf attribute block to see if all the entries would fit into
933  * a shortform attribute list.
934  */
935 int
936 xfs_attr_shortform_allfit(
937 	struct xfs_buf		*bp,
938 	struct xfs_inode	*dp)
939 {
940 	struct xfs_attr_leafblock *leaf;
941 	struct xfs_attr_leaf_entry *entry;
942 	xfs_attr_leaf_name_local_t *name_loc;
943 	struct xfs_attr3_icleaf_hdr leafhdr;
944 	int			bytes;
945 	int			i;
946 	struct xfs_mount	*mp = bp->b_mount;
947 
948 	leaf = bp->b_addr;
949 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
950 	entry = xfs_attr3_leaf_entryp(leaf);
951 
952 	bytes = sizeof(struct xfs_attr_sf_hdr);
953 	for (i = 0; i < leafhdr.count; entry++, i++) {
954 		if (entry->flags & XFS_ATTR_INCOMPLETE)
955 			continue;		/* don't copy partial entries */
956 		if (!(entry->flags & XFS_ATTR_LOCAL))
957 			return 0;
958 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
959 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
960 			return 0;
961 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
962 			return 0;
963 		bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
964 					be16_to_cpu(name_loc->valuelen));
965 	}
966 	if (xfs_has_attr2(dp->i_mount) &&
967 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
968 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
969 		return -1;
970 	return xfs_attr_shortform_bytesfit(dp, bytes);
971 }
972 
973 /* Verify the consistency of a raw inline attribute fork. */
974 xfs_failaddr_t
975 xfs_attr_shortform_verify(
976 	struct xfs_attr_sf_hdr		*sfp,
977 	size_t				size)
978 {
979 	struct xfs_attr_sf_entry	*sfep = xfs_attr_sf_firstentry(sfp);
980 	struct xfs_attr_sf_entry	*next_sfep;
981 	char				*endp;
982 	int				i;
983 
984 	/*
985 	 * Give up if the attribute is way too short.
986 	 */
987 	if (size < sizeof(struct xfs_attr_sf_hdr))
988 		return __this_address;
989 
990 	endp = (char *)sfp + size;
991 
992 	/* Check all reported entries */
993 	for (i = 0; i < sfp->count; i++) {
994 		/*
995 		 * struct xfs_attr_sf_entry has a variable length.
996 		 * Check the fixed-offset parts of the structure are
997 		 * within the data buffer.
998 		 * xfs_attr_sf_entry is defined with a 1-byte variable
999 		 * array at the end, so we must subtract that off.
1000 		 */
1001 		if (((char *)sfep + sizeof(*sfep)) >= endp)
1002 			return __this_address;
1003 
1004 		/* Don't allow names with known bad length. */
1005 		if (sfep->namelen == 0)
1006 			return __this_address;
1007 
1008 		/*
1009 		 * Check that the variable-length part of the structure is
1010 		 * within the data buffer.  The next entry starts after the
1011 		 * name component, so nextentry is an acceptable test.
1012 		 */
1013 		next_sfep = xfs_attr_sf_nextentry(sfep);
1014 		if ((char *)next_sfep > endp)
1015 			return __this_address;
1016 
1017 		/*
1018 		 * Check for unknown flags.  Short form doesn't support
1019 		 * the incomplete or local bits, so we can use the namespace
1020 		 * mask here.
1021 		 */
1022 		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1023 			return __this_address;
1024 
1025 		/*
1026 		 * Check for invalid namespace combinations.  We only allow
1027 		 * one namespace flag per xattr, so we can just count the
1028 		 * bits (i.e. hweight) here.
1029 		 */
1030 		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1031 			return __this_address;
1032 
1033 		sfep = next_sfep;
1034 	}
1035 	if ((void *)sfep != (void *)endp)
1036 		return __this_address;
1037 
1038 	return NULL;
1039 }
1040 
1041 /*
1042  * Convert a leaf attribute list to shortform attribute list
1043  */
1044 int
1045 xfs_attr3_leaf_to_shortform(
1046 	struct xfs_buf		*bp,
1047 	struct xfs_da_args	*args,
1048 	int			forkoff)
1049 {
1050 	struct xfs_attr_leafblock *leaf;
1051 	struct xfs_attr3_icleaf_hdr ichdr;
1052 	struct xfs_attr_leaf_entry *entry;
1053 	struct xfs_attr_leaf_name_local *name_loc;
1054 	struct xfs_da_args	nargs;
1055 	struct xfs_inode	*dp = args->dp;
1056 	char			*tmpbuffer;
1057 	int			error;
1058 	int			i;
1059 
1060 	trace_xfs_attr_leaf_to_sf(args);
1061 
1062 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1063 	if (!tmpbuffer)
1064 		return -ENOMEM;
1065 
1066 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1067 
1068 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1069 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1070 	entry = xfs_attr3_leaf_entryp(leaf);
1071 
1072 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1073 	memset(bp->b_addr, 0, args->geo->blksize);
1074 
1075 	/*
1076 	 * Clean out the prior contents of the attribute list.
1077 	 */
1078 	error = xfs_da_shrink_inode(args, 0, bp);
1079 	if (error)
1080 		goto out;
1081 
1082 	if (forkoff == -1) {
1083 		/*
1084 		 * Don't remove the attr fork if this operation is the first
1085 		 * part of a attr replace operations. We're going to add a new
1086 		 * attr immediately, so we need to keep the attr fork around in
1087 		 * this case.
1088 		 */
1089 		if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1090 			ASSERT(xfs_has_attr2(dp->i_mount));
1091 			ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1092 			xfs_attr_fork_remove(dp, args->trans);
1093 		}
1094 		goto out;
1095 	}
1096 
1097 	xfs_attr_shortform_create(args);
1098 
1099 	/*
1100 	 * Copy the attributes
1101 	 */
1102 	memset((char *)&nargs, 0, sizeof(nargs));
1103 	nargs.geo = args->geo;
1104 	nargs.dp = dp;
1105 	nargs.total = args->total;
1106 	nargs.whichfork = XFS_ATTR_FORK;
1107 	nargs.trans = args->trans;
1108 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1109 
1110 	for (i = 0; i < ichdr.count; entry++, i++) {
1111 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1112 			continue;	/* don't copy partial entries */
1113 		if (!entry->nameidx)
1114 			continue;
1115 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1116 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1117 		nargs.name = name_loc->nameval;
1118 		nargs.namelen = name_loc->namelen;
1119 		nargs.value = &name_loc->nameval[nargs.namelen];
1120 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1121 		nargs.hashval = be32_to_cpu(entry->hashval);
1122 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1123 		xfs_attr_shortform_add(&nargs, forkoff);
1124 	}
1125 	error = 0;
1126 
1127 out:
1128 	kmem_free(tmpbuffer);
1129 	return error;
1130 }
1131 
1132 /*
1133  * Convert from using a single leaf to a root node and a leaf.
1134  */
1135 int
1136 xfs_attr3_leaf_to_node(
1137 	struct xfs_da_args	*args)
1138 {
1139 	struct xfs_attr_leafblock *leaf;
1140 	struct xfs_attr3_icleaf_hdr icleafhdr;
1141 	struct xfs_attr_leaf_entry *entries;
1142 	struct xfs_da3_icnode_hdr icnodehdr;
1143 	struct xfs_da_intnode	*node;
1144 	struct xfs_inode	*dp = args->dp;
1145 	struct xfs_mount	*mp = dp->i_mount;
1146 	struct xfs_buf		*bp1 = NULL;
1147 	struct xfs_buf		*bp2 = NULL;
1148 	xfs_dablk_t		blkno;
1149 	int			error;
1150 
1151 	trace_xfs_attr_leaf_to_node(args);
1152 
1153 	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1154 		error = -EIO;
1155 		goto out;
1156 	}
1157 
1158 	error = xfs_da_grow_inode(args, &blkno);
1159 	if (error)
1160 		goto out;
1161 	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1162 	if (error)
1163 		goto out;
1164 
1165 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1166 	if (error)
1167 		goto out;
1168 
1169 	/*
1170 	 * Copy leaf to new buffer and log it.
1171 	 */
1172 	xfs_da_buf_copy(bp2, bp1, args->geo->blksize);
1173 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1174 
1175 	/*
1176 	 * Set up the new root node.
1177 	 */
1178 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1179 	if (error)
1180 		goto out;
1181 	node = bp1->b_addr;
1182 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1183 
1184 	leaf = bp2->b_addr;
1185 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1186 	entries = xfs_attr3_leaf_entryp(leaf);
1187 
1188 	/* both on-disk, don't endian-flip twice */
1189 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1190 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1191 	icnodehdr.count = 1;
1192 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1193 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1194 	error = 0;
1195 out:
1196 	return error;
1197 }
1198 
1199 /*========================================================================
1200  * Routines used for growing the Btree.
1201  *========================================================================*/
1202 
1203 /*
1204  * Create the initial contents of a leaf attribute list
1205  * or a leaf in a node attribute list.
1206  */
1207 STATIC int
1208 xfs_attr3_leaf_create(
1209 	struct xfs_da_args	*args,
1210 	xfs_dablk_t		blkno,
1211 	struct xfs_buf		**bpp)
1212 {
1213 	struct xfs_attr_leafblock *leaf;
1214 	struct xfs_attr3_icleaf_hdr ichdr;
1215 	struct xfs_inode	*dp = args->dp;
1216 	struct xfs_mount	*mp = dp->i_mount;
1217 	struct xfs_buf		*bp;
1218 	int			error;
1219 
1220 	trace_xfs_attr_leaf_create(args);
1221 
1222 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1223 					    XFS_ATTR_FORK);
1224 	if (error)
1225 		return error;
1226 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1227 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1228 	leaf = bp->b_addr;
1229 	memset(leaf, 0, args->geo->blksize);
1230 
1231 	memset(&ichdr, 0, sizeof(ichdr));
1232 	ichdr.firstused = args->geo->blksize;
1233 
1234 	if (xfs_has_crc(mp)) {
1235 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1236 
1237 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1238 
1239 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1240 		hdr3->owner = cpu_to_be64(dp->i_ino);
1241 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1242 
1243 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1244 	} else {
1245 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1246 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1247 	}
1248 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1249 
1250 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1251 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1252 
1253 	*bpp = bp;
1254 	return 0;
1255 }
1256 
1257 /*
1258  * Split the leaf node, rebalance, then add the new entry.
1259  */
1260 int
1261 xfs_attr3_leaf_split(
1262 	struct xfs_da_state	*state,
1263 	struct xfs_da_state_blk	*oldblk,
1264 	struct xfs_da_state_blk	*newblk)
1265 {
1266 	xfs_dablk_t blkno;
1267 	int error;
1268 
1269 	trace_xfs_attr_leaf_split(state->args);
1270 
1271 	/*
1272 	 * Allocate space for a new leaf node.
1273 	 */
1274 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1275 	error = xfs_da_grow_inode(state->args, &blkno);
1276 	if (error)
1277 		return error;
1278 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1279 	if (error)
1280 		return error;
1281 	newblk->blkno = blkno;
1282 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1283 
1284 	/*
1285 	 * Rebalance the entries across the two leaves.
1286 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1287 	 */
1288 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1289 	error = xfs_da3_blk_link(state, oldblk, newblk);
1290 	if (error)
1291 		return error;
1292 
1293 	/*
1294 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1295 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1296 	 * "new" attrs info.  Will need the "old" info to remove it later.
1297 	 *
1298 	 * Insert the "new" entry in the correct block.
1299 	 */
1300 	if (state->inleaf) {
1301 		trace_xfs_attr_leaf_add_old(state->args);
1302 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1303 	} else {
1304 		trace_xfs_attr_leaf_add_new(state->args);
1305 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1306 	}
1307 
1308 	/*
1309 	 * Update last hashval in each block since we added the name.
1310 	 */
1311 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1312 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1313 	return error;
1314 }
1315 
1316 /*
1317  * Add a name to the leaf attribute list structure.
1318  */
1319 int
1320 xfs_attr3_leaf_add(
1321 	struct xfs_buf		*bp,
1322 	struct xfs_da_args	*args)
1323 {
1324 	struct xfs_attr_leafblock *leaf;
1325 	struct xfs_attr3_icleaf_hdr ichdr;
1326 	int			tablesize;
1327 	int			entsize;
1328 	int			sum;
1329 	int			tmp;
1330 	int			i;
1331 
1332 	trace_xfs_attr_leaf_add(args);
1333 
1334 	leaf = bp->b_addr;
1335 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1336 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1337 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1338 
1339 	/*
1340 	 * Search through freemap for first-fit on new name length.
1341 	 * (may need to figure in size of entry struct too)
1342 	 */
1343 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1344 					+ xfs_attr3_leaf_hdr_size(leaf);
1345 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1346 		if (tablesize > ichdr.firstused) {
1347 			sum += ichdr.freemap[i].size;
1348 			continue;
1349 		}
1350 		if (!ichdr.freemap[i].size)
1351 			continue;	/* no space in this map */
1352 		tmp = entsize;
1353 		if (ichdr.freemap[i].base < ichdr.firstused)
1354 			tmp += sizeof(xfs_attr_leaf_entry_t);
1355 		if (ichdr.freemap[i].size >= tmp) {
1356 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1357 			goto out_log_hdr;
1358 		}
1359 		sum += ichdr.freemap[i].size;
1360 	}
1361 
1362 	/*
1363 	 * If there are no holes in the address space of the block,
1364 	 * and we don't have enough freespace, then compaction will do us
1365 	 * no good and we should just give up.
1366 	 */
1367 	if (!ichdr.holes && sum < entsize)
1368 		return -ENOSPC;
1369 
1370 	/*
1371 	 * Compact the entries to coalesce free space.
1372 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1373 	 */
1374 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1375 
1376 	/*
1377 	 * After compaction, the block is guaranteed to have only one
1378 	 * free region, in freemap[0].  If it is not big enough, give up.
1379 	 */
1380 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1381 		tmp = -ENOSPC;
1382 		goto out_log_hdr;
1383 	}
1384 
1385 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1386 
1387 out_log_hdr:
1388 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1389 	xfs_trans_log_buf(args->trans, bp,
1390 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1391 				xfs_attr3_leaf_hdr_size(leaf)));
1392 	return tmp;
1393 }
1394 
1395 /*
1396  * Add a name to a leaf attribute list structure.
1397  */
1398 STATIC int
1399 xfs_attr3_leaf_add_work(
1400 	struct xfs_buf		*bp,
1401 	struct xfs_attr3_icleaf_hdr *ichdr,
1402 	struct xfs_da_args	*args,
1403 	int			mapindex)
1404 {
1405 	struct xfs_attr_leafblock *leaf;
1406 	struct xfs_attr_leaf_entry *entry;
1407 	struct xfs_attr_leaf_name_local *name_loc;
1408 	struct xfs_attr_leaf_name_remote *name_rmt;
1409 	struct xfs_mount	*mp;
1410 	int			tmp;
1411 	int			i;
1412 
1413 	trace_xfs_attr_leaf_add_work(args);
1414 
1415 	leaf = bp->b_addr;
1416 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1417 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1418 
1419 	/*
1420 	 * Force open some space in the entry array and fill it in.
1421 	 */
1422 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1423 	if (args->index < ichdr->count) {
1424 		tmp  = ichdr->count - args->index;
1425 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1426 		memmove(entry + 1, entry, tmp);
1427 		xfs_trans_log_buf(args->trans, bp,
1428 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1429 	}
1430 	ichdr->count++;
1431 
1432 	/*
1433 	 * Allocate space for the new string (at the end of the run).
1434 	 */
1435 	mp = args->trans->t_mountp;
1436 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1437 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1438 	ASSERT(ichdr->freemap[mapindex].size >=
1439 		xfs_attr_leaf_newentsize(args, NULL));
1440 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1441 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1442 
1443 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1444 
1445 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1446 				     ichdr->freemap[mapindex].size);
1447 	entry->hashval = cpu_to_be32(args->hashval);
1448 	entry->flags = args->attr_filter;
1449 	if (tmp)
1450 		entry->flags |= XFS_ATTR_LOCAL;
1451 	if (args->op_flags & XFS_DA_OP_REPLACE) {
1452 		if (!(args->op_flags & XFS_DA_OP_LOGGED))
1453 			entry->flags |= XFS_ATTR_INCOMPLETE;
1454 		if ((args->blkno2 == args->blkno) &&
1455 		    (args->index2 <= args->index)) {
1456 			args->index2++;
1457 		}
1458 	}
1459 	xfs_trans_log_buf(args->trans, bp,
1460 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1461 	ASSERT((args->index == 0) ||
1462 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1463 	ASSERT((args->index == ichdr->count - 1) ||
1464 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1465 
1466 	/*
1467 	 * For "remote" attribute values, simply note that we need to
1468 	 * allocate space for the "remote" value.  We can't actually
1469 	 * allocate the extents in this transaction, and we can't decide
1470 	 * which blocks they should be as we might allocate more blocks
1471 	 * as part of this transaction (a split operation for example).
1472 	 */
1473 	if (entry->flags & XFS_ATTR_LOCAL) {
1474 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1475 		name_loc->namelen = args->namelen;
1476 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1477 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1478 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1479 				   be16_to_cpu(name_loc->valuelen));
1480 	} else {
1481 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1482 		name_rmt->namelen = args->namelen;
1483 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1484 		entry->flags |= XFS_ATTR_INCOMPLETE;
1485 		/* just in case */
1486 		name_rmt->valuelen = 0;
1487 		name_rmt->valueblk = 0;
1488 		args->rmtblkno = 1;
1489 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1490 		args->rmtvaluelen = args->valuelen;
1491 	}
1492 	xfs_trans_log_buf(args->trans, bp,
1493 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1494 				   xfs_attr_leaf_entsize(leaf, args->index)));
1495 
1496 	/*
1497 	 * Update the control info for this leaf node
1498 	 */
1499 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1500 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1501 
1502 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1503 					+ xfs_attr3_leaf_hdr_size(leaf));
1504 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1505 					+ xfs_attr3_leaf_hdr_size(leaf);
1506 
1507 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1508 		if (ichdr->freemap[i].base == tmp) {
1509 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1510 			ichdr->freemap[i].size -=
1511 				min_t(uint16_t, ichdr->freemap[i].size,
1512 						sizeof(xfs_attr_leaf_entry_t));
1513 		}
1514 	}
1515 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1516 	return 0;
1517 }
1518 
1519 /*
1520  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1521  */
1522 STATIC void
1523 xfs_attr3_leaf_compact(
1524 	struct xfs_da_args	*args,
1525 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1526 	struct xfs_buf		*bp)
1527 {
1528 	struct xfs_attr_leafblock *leaf_src;
1529 	struct xfs_attr_leafblock *leaf_dst;
1530 	struct xfs_attr3_icleaf_hdr ichdr_src;
1531 	struct xfs_trans	*trans = args->trans;
1532 	char			*tmpbuffer;
1533 
1534 	trace_xfs_attr_leaf_compact(args);
1535 
1536 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1537 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1538 	memset(bp->b_addr, 0, args->geo->blksize);
1539 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1540 	leaf_dst = bp->b_addr;
1541 
1542 	/*
1543 	 * Copy the on-disk header back into the destination buffer to ensure
1544 	 * all the information in the header that is not part of the incore
1545 	 * header structure is preserved.
1546 	 */
1547 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1548 
1549 	/* Initialise the incore headers */
1550 	ichdr_src = *ichdr_dst;	/* struct copy */
1551 	ichdr_dst->firstused = args->geo->blksize;
1552 	ichdr_dst->usedbytes = 0;
1553 	ichdr_dst->count = 0;
1554 	ichdr_dst->holes = 0;
1555 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1556 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1557 						ichdr_dst->freemap[0].base;
1558 
1559 	/* write the header back to initialise the underlying buffer */
1560 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1561 
1562 	/*
1563 	 * Copy all entry's in the same (sorted) order,
1564 	 * but allocate name/value pairs packed and in sequence.
1565 	 */
1566 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1567 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1568 	/*
1569 	 * this logs the entire buffer, but the caller must write the header
1570 	 * back to the buffer when it is finished modifying it.
1571 	 */
1572 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1573 
1574 	kmem_free(tmpbuffer);
1575 }
1576 
1577 /*
1578  * Compare two leaf blocks "order".
1579  * Return 0 unless leaf2 should go before leaf1.
1580  */
1581 static int
1582 xfs_attr3_leaf_order(
1583 	struct xfs_buf	*leaf1_bp,
1584 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1585 	struct xfs_buf	*leaf2_bp,
1586 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1587 {
1588 	struct xfs_attr_leaf_entry *entries1;
1589 	struct xfs_attr_leaf_entry *entries2;
1590 
1591 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1592 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1593 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1594 	    ((be32_to_cpu(entries2[0].hashval) <
1595 	      be32_to_cpu(entries1[0].hashval)) ||
1596 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1597 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1598 		return 1;
1599 	}
1600 	return 0;
1601 }
1602 
1603 int
1604 xfs_attr_leaf_order(
1605 	struct xfs_buf	*leaf1_bp,
1606 	struct xfs_buf	*leaf2_bp)
1607 {
1608 	struct xfs_attr3_icleaf_hdr ichdr1;
1609 	struct xfs_attr3_icleaf_hdr ichdr2;
1610 	struct xfs_mount *mp = leaf1_bp->b_mount;
1611 
1612 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1613 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1614 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1615 }
1616 
1617 /*
1618  * Redistribute the attribute list entries between two leaf nodes,
1619  * taking into account the size of the new entry.
1620  *
1621  * NOTE: if new block is empty, then it will get the upper half of the
1622  * old block.  At present, all (one) callers pass in an empty second block.
1623  *
1624  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1625  * to match what it is doing in splitting the attribute leaf block.  Those
1626  * values are used in "atomic rename" operations on attributes.  Note that
1627  * the "new" and "old" values can end up in different blocks.
1628  */
1629 STATIC void
1630 xfs_attr3_leaf_rebalance(
1631 	struct xfs_da_state	*state,
1632 	struct xfs_da_state_blk	*blk1,
1633 	struct xfs_da_state_blk	*blk2)
1634 {
1635 	struct xfs_da_args	*args;
1636 	struct xfs_attr_leafblock *leaf1;
1637 	struct xfs_attr_leafblock *leaf2;
1638 	struct xfs_attr3_icleaf_hdr ichdr1;
1639 	struct xfs_attr3_icleaf_hdr ichdr2;
1640 	struct xfs_attr_leaf_entry *entries1;
1641 	struct xfs_attr_leaf_entry *entries2;
1642 	int			count;
1643 	int			totallen;
1644 	int			max;
1645 	int			space;
1646 	int			swap;
1647 
1648 	/*
1649 	 * Set up environment.
1650 	 */
1651 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1652 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1653 	leaf1 = blk1->bp->b_addr;
1654 	leaf2 = blk2->bp->b_addr;
1655 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1656 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1657 	ASSERT(ichdr2.count == 0);
1658 	args = state->args;
1659 
1660 	trace_xfs_attr_leaf_rebalance(args);
1661 
1662 	/*
1663 	 * Check ordering of blocks, reverse if it makes things simpler.
1664 	 *
1665 	 * NOTE: Given that all (current) callers pass in an empty
1666 	 * second block, this code should never set "swap".
1667 	 */
1668 	swap = 0;
1669 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1670 		swap(blk1, blk2);
1671 
1672 		/* swap structures rather than reconverting them */
1673 		swap(ichdr1, ichdr2);
1674 
1675 		leaf1 = blk1->bp->b_addr;
1676 		leaf2 = blk2->bp->b_addr;
1677 		swap = 1;
1678 	}
1679 
1680 	/*
1681 	 * Examine entries until we reduce the absolute difference in
1682 	 * byte usage between the two blocks to a minimum.  Then get
1683 	 * the direction to copy and the number of elements to move.
1684 	 *
1685 	 * "inleaf" is true if the new entry should be inserted into blk1.
1686 	 * If "swap" is also true, then reverse the sense of "inleaf".
1687 	 */
1688 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1689 						      blk2, &ichdr2,
1690 						      &count, &totallen);
1691 	if (swap)
1692 		state->inleaf = !state->inleaf;
1693 
1694 	/*
1695 	 * Move any entries required from leaf to leaf:
1696 	 */
1697 	if (count < ichdr1.count) {
1698 		/*
1699 		 * Figure the total bytes to be added to the destination leaf.
1700 		 */
1701 		/* number entries being moved */
1702 		count = ichdr1.count - count;
1703 		space  = ichdr1.usedbytes - totallen;
1704 		space += count * sizeof(xfs_attr_leaf_entry_t);
1705 
1706 		/*
1707 		 * leaf2 is the destination, compact it if it looks tight.
1708 		 */
1709 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1710 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1711 		if (space > max)
1712 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1713 
1714 		/*
1715 		 * Move high entries from leaf1 to low end of leaf2.
1716 		 */
1717 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1718 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1719 
1720 	} else if (count > ichdr1.count) {
1721 		/*
1722 		 * I assert that since all callers pass in an empty
1723 		 * second buffer, this code should never execute.
1724 		 */
1725 		ASSERT(0);
1726 
1727 		/*
1728 		 * Figure the total bytes to be added to the destination leaf.
1729 		 */
1730 		/* number entries being moved */
1731 		count -= ichdr1.count;
1732 		space  = totallen - ichdr1.usedbytes;
1733 		space += count * sizeof(xfs_attr_leaf_entry_t);
1734 
1735 		/*
1736 		 * leaf1 is the destination, compact it if it looks tight.
1737 		 */
1738 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1739 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1740 		if (space > max)
1741 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1742 
1743 		/*
1744 		 * Move low entries from leaf2 to high end of leaf1.
1745 		 */
1746 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1747 					ichdr1.count, count);
1748 	}
1749 
1750 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1751 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1752 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1753 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1754 
1755 	/*
1756 	 * Copy out last hashval in each block for B-tree code.
1757 	 */
1758 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1759 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1760 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1761 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1762 
1763 	/*
1764 	 * Adjust the expected index for insertion.
1765 	 * NOTE: this code depends on the (current) situation that the
1766 	 * second block was originally empty.
1767 	 *
1768 	 * If the insertion point moved to the 2nd block, we must adjust
1769 	 * the index.  We must also track the entry just following the
1770 	 * new entry for use in an "atomic rename" operation, that entry
1771 	 * is always the "old" entry and the "new" entry is what we are
1772 	 * inserting.  The index/blkno fields refer to the "old" entry,
1773 	 * while the index2/blkno2 fields refer to the "new" entry.
1774 	 */
1775 	if (blk1->index > ichdr1.count) {
1776 		ASSERT(state->inleaf == 0);
1777 		blk2->index = blk1->index - ichdr1.count;
1778 		args->index = args->index2 = blk2->index;
1779 		args->blkno = args->blkno2 = blk2->blkno;
1780 	} else if (blk1->index == ichdr1.count) {
1781 		if (state->inleaf) {
1782 			args->index = blk1->index;
1783 			args->blkno = blk1->blkno;
1784 			args->index2 = 0;
1785 			args->blkno2 = blk2->blkno;
1786 		} else {
1787 			/*
1788 			 * On a double leaf split, the original attr location
1789 			 * is already stored in blkno2/index2, so don't
1790 			 * overwrite it overwise we corrupt the tree.
1791 			 */
1792 			blk2->index = blk1->index - ichdr1.count;
1793 			args->index = blk2->index;
1794 			args->blkno = blk2->blkno;
1795 			if (!state->extravalid) {
1796 				/*
1797 				 * set the new attr location to match the old
1798 				 * one and let the higher level split code
1799 				 * decide where in the leaf to place it.
1800 				 */
1801 				args->index2 = blk2->index;
1802 				args->blkno2 = blk2->blkno;
1803 			}
1804 		}
1805 	} else {
1806 		ASSERT(state->inleaf == 1);
1807 		args->index = args->index2 = blk1->index;
1808 		args->blkno = args->blkno2 = blk1->blkno;
1809 	}
1810 }
1811 
1812 /*
1813  * Examine entries until we reduce the absolute difference in
1814  * byte usage between the two blocks to a minimum.
1815  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1816  * GROT: there will always be enough room in either block for a new entry.
1817  * GROT: Do a double-split for this case?
1818  */
1819 STATIC int
1820 xfs_attr3_leaf_figure_balance(
1821 	struct xfs_da_state		*state,
1822 	struct xfs_da_state_blk		*blk1,
1823 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1824 	struct xfs_da_state_blk		*blk2,
1825 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1826 	int				*countarg,
1827 	int				*usedbytesarg)
1828 {
1829 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1830 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1831 	struct xfs_attr_leaf_entry	*entry;
1832 	int				count;
1833 	int				max;
1834 	int				index;
1835 	int				totallen = 0;
1836 	int				half;
1837 	int				lastdelta;
1838 	int				foundit = 0;
1839 	int				tmp;
1840 
1841 	/*
1842 	 * Examine entries until we reduce the absolute difference in
1843 	 * byte usage between the two blocks to a minimum.
1844 	 */
1845 	max = ichdr1->count + ichdr2->count;
1846 	half = (max + 1) * sizeof(*entry);
1847 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1848 			xfs_attr_leaf_newentsize(state->args, NULL);
1849 	half /= 2;
1850 	lastdelta = state->args->geo->blksize;
1851 	entry = xfs_attr3_leaf_entryp(leaf1);
1852 	for (count = index = 0; count < max; entry++, index++, count++) {
1853 
1854 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1855 		/*
1856 		 * The new entry is in the first block, account for it.
1857 		 */
1858 		if (count == blk1->index) {
1859 			tmp = totallen + sizeof(*entry) +
1860 				xfs_attr_leaf_newentsize(state->args, NULL);
1861 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1862 				break;
1863 			lastdelta = XFS_ATTR_ABS(half - tmp);
1864 			totallen = tmp;
1865 			foundit = 1;
1866 		}
1867 
1868 		/*
1869 		 * Wrap around into the second block if necessary.
1870 		 */
1871 		if (count == ichdr1->count) {
1872 			leaf1 = leaf2;
1873 			entry = xfs_attr3_leaf_entryp(leaf1);
1874 			index = 0;
1875 		}
1876 
1877 		/*
1878 		 * Figure out if next leaf entry would be too much.
1879 		 */
1880 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1881 									index);
1882 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1883 			break;
1884 		lastdelta = XFS_ATTR_ABS(half - tmp);
1885 		totallen = tmp;
1886 #undef XFS_ATTR_ABS
1887 	}
1888 
1889 	/*
1890 	 * Calculate the number of usedbytes that will end up in lower block.
1891 	 * If new entry not in lower block, fix up the count.
1892 	 */
1893 	totallen -= count * sizeof(*entry);
1894 	if (foundit) {
1895 		totallen -= sizeof(*entry) +
1896 				xfs_attr_leaf_newentsize(state->args, NULL);
1897 	}
1898 
1899 	*countarg = count;
1900 	*usedbytesarg = totallen;
1901 	return foundit;
1902 }
1903 
1904 /*========================================================================
1905  * Routines used for shrinking the Btree.
1906  *========================================================================*/
1907 
1908 /*
1909  * Check a leaf block and its neighbors to see if the block should be
1910  * collapsed into one or the other neighbor.  Always keep the block
1911  * with the smaller block number.
1912  * If the current block is over 50% full, don't try to join it, return 0.
1913  * If the block is empty, fill in the state structure and return 2.
1914  * If it can be collapsed, fill in the state structure and return 1.
1915  * If nothing can be done, return 0.
1916  *
1917  * GROT: allow for INCOMPLETE entries in calculation.
1918  */
1919 int
1920 xfs_attr3_leaf_toosmall(
1921 	struct xfs_da_state	*state,
1922 	int			*action)
1923 {
1924 	struct xfs_attr_leafblock *leaf;
1925 	struct xfs_da_state_blk	*blk;
1926 	struct xfs_attr3_icleaf_hdr ichdr;
1927 	struct xfs_buf		*bp;
1928 	xfs_dablk_t		blkno;
1929 	int			bytes;
1930 	int			forward;
1931 	int			error;
1932 	int			retval;
1933 	int			i;
1934 
1935 	trace_xfs_attr_leaf_toosmall(state->args);
1936 
1937 	/*
1938 	 * Check for the degenerate case of the block being over 50% full.
1939 	 * If so, it's not worth even looking to see if we might be able
1940 	 * to coalesce with a sibling.
1941 	 */
1942 	blk = &state->path.blk[ state->path.active-1 ];
1943 	leaf = blk->bp->b_addr;
1944 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1945 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
1946 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1947 		ichdr.usedbytes;
1948 	if (bytes > (state->args->geo->blksize >> 1)) {
1949 		*action = 0;	/* blk over 50%, don't try to join */
1950 		return 0;
1951 	}
1952 
1953 	/*
1954 	 * Check for the degenerate case of the block being empty.
1955 	 * If the block is empty, we'll simply delete it, no need to
1956 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1957 	 * to merge with the forward block unless it is NULL.
1958 	 */
1959 	if (ichdr.count == 0) {
1960 		/*
1961 		 * Make altpath point to the block we want to keep and
1962 		 * path point to the block we want to drop (this one).
1963 		 */
1964 		forward = (ichdr.forw != 0);
1965 		memcpy(&state->altpath, &state->path, sizeof(state->path));
1966 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1967 						 0, &retval);
1968 		if (error)
1969 			return error;
1970 		if (retval) {
1971 			*action = 0;
1972 		} else {
1973 			*action = 2;
1974 		}
1975 		return 0;
1976 	}
1977 
1978 	/*
1979 	 * Examine each sibling block to see if we can coalesce with
1980 	 * at least 25% free space to spare.  We need to figure out
1981 	 * whether to merge with the forward or the backward block.
1982 	 * We prefer coalescing with the lower numbered sibling so as
1983 	 * to shrink an attribute list over time.
1984 	 */
1985 	/* start with smaller blk num */
1986 	forward = ichdr.forw < ichdr.back;
1987 	for (i = 0; i < 2; forward = !forward, i++) {
1988 		struct xfs_attr3_icleaf_hdr ichdr2;
1989 		if (forward)
1990 			blkno = ichdr.forw;
1991 		else
1992 			blkno = ichdr.back;
1993 		if (blkno == 0)
1994 			continue;
1995 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1996 					blkno, &bp);
1997 		if (error)
1998 			return error;
1999 
2000 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2001 
2002 		bytes = state->args->geo->blksize -
2003 			(state->args->geo->blksize >> 2) -
2004 			ichdr.usedbytes - ichdr2.usedbytes -
2005 			((ichdr.count + ichdr2.count) *
2006 					sizeof(xfs_attr_leaf_entry_t)) -
2007 			xfs_attr3_leaf_hdr_size(leaf);
2008 
2009 		xfs_trans_brelse(state->args->trans, bp);
2010 		if (bytes >= 0)
2011 			break;	/* fits with at least 25% to spare */
2012 	}
2013 	if (i >= 2) {
2014 		*action = 0;
2015 		return 0;
2016 	}
2017 
2018 	/*
2019 	 * Make altpath point to the block we want to keep (the lower
2020 	 * numbered block) and path point to the block we want to drop.
2021 	 */
2022 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2023 	if (blkno < blk->blkno) {
2024 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2025 						 0, &retval);
2026 	} else {
2027 		error = xfs_da3_path_shift(state, &state->path, forward,
2028 						 0, &retval);
2029 	}
2030 	if (error)
2031 		return error;
2032 	if (retval) {
2033 		*action = 0;
2034 	} else {
2035 		*action = 1;
2036 	}
2037 	return 0;
2038 }
2039 
2040 /*
2041  * Remove a name from the leaf attribute list structure.
2042  *
2043  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2044  * If two leaves are 37% full, when combined they will leave 25% free.
2045  */
2046 int
2047 xfs_attr3_leaf_remove(
2048 	struct xfs_buf		*bp,
2049 	struct xfs_da_args	*args)
2050 {
2051 	struct xfs_attr_leafblock *leaf;
2052 	struct xfs_attr3_icleaf_hdr ichdr;
2053 	struct xfs_attr_leaf_entry *entry;
2054 	int			before;
2055 	int			after;
2056 	int			smallest;
2057 	int			entsize;
2058 	int			tablesize;
2059 	int			tmp;
2060 	int			i;
2061 
2062 	trace_xfs_attr_leaf_remove(args);
2063 
2064 	leaf = bp->b_addr;
2065 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2066 
2067 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2068 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2069 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2070 					xfs_attr3_leaf_hdr_size(leaf));
2071 
2072 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2073 
2074 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2075 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2076 
2077 	/*
2078 	 * Scan through free region table:
2079 	 *    check for adjacency of free'd entry with an existing one,
2080 	 *    find smallest free region in case we need to replace it,
2081 	 *    adjust any map that borders the entry table,
2082 	 */
2083 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2084 					+ xfs_attr3_leaf_hdr_size(leaf);
2085 	tmp = ichdr.freemap[0].size;
2086 	before = after = -1;
2087 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2088 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2089 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2090 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2091 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2092 		if (ichdr.freemap[i].base == tablesize) {
2093 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2094 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2095 		}
2096 
2097 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2098 				be16_to_cpu(entry->nameidx)) {
2099 			before = i;
2100 		} else if (ichdr.freemap[i].base ==
2101 				(be16_to_cpu(entry->nameidx) + entsize)) {
2102 			after = i;
2103 		} else if (ichdr.freemap[i].size < tmp) {
2104 			tmp = ichdr.freemap[i].size;
2105 			smallest = i;
2106 		}
2107 	}
2108 
2109 	/*
2110 	 * Coalesce adjacent freemap regions,
2111 	 * or replace the smallest region.
2112 	 */
2113 	if ((before >= 0) || (after >= 0)) {
2114 		if ((before >= 0) && (after >= 0)) {
2115 			ichdr.freemap[before].size += entsize;
2116 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2117 			ichdr.freemap[after].base = 0;
2118 			ichdr.freemap[after].size = 0;
2119 		} else if (before >= 0) {
2120 			ichdr.freemap[before].size += entsize;
2121 		} else {
2122 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2123 			ichdr.freemap[after].size += entsize;
2124 		}
2125 	} else {
2126 		/*
2127 		 * Replace smallest region (if it is smaller than free'd entry)
2128 		 */
2129 		if (ichdr.freemap[smallest].size < entsize) {
2130 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2131 			ichdr.freemap[smallest].size = entsize;
2132 		}
2133 	}
2134 
2135 	/*
2136 	 * Did we remove the first entry?
2137 	 */
2138 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2139 		smallest = 1;
2140 	else
2141 		smallest = 0;
2142 
2143 	/*
2144 	 * Compress the remaining entries and zero out the removed stuff.
2145 	 */
2146 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2147 	ichdr.usedbytes -= entsize;
2148 	xfs_trans_log_buf(args->trans, bp,
2149 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2150 				   entsize));
2151 
2152 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2153 	memmove(entry, entry + 1, tmp);
2154 	ichdr.count--;
2155 	xfs_trans_log_buf(args->trans, bp,
2156 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2157 
2158 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2159 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2160 
2161 	/*
2162 	 * If we removed the first entry, re-find the first used byte
2163 	 * in the name area.  Note that if the entry was the "firstused",
2164 	 * then we don't have a "hole" in our block resulting from
2165 	 * removing the name.
2166 	 */
2167 	if (smallest) {
2168 		tmp = args->geo->blksize;
2169 		entry = xfs_attr3_leaf_entryp(leaf);
2170 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2171 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2172 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2173 
2174 			if (be16_to_cpu(entry->nameidx) < tmp)
2175 				tmp = be16_to_cpu(entry->nameidx);
2176 		}
2177 		ichdr.firstused = tmp;
2178 		ASSERT(ichdr.firstused != 0);
2179 	} else {
2180 		ichdr.holes = 1;	/* mark as needing compaction */
2181 	}
2182 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2183 	xfs_trans_log_buf(args->trans, bp,
2184 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2185 					  xfs_attr3_leaf_hdr_size(leaf)));
2186 
2187 	/*
2188 	 * Check if leaf is less than 50% full, caller may want to
2189 	 * "join" the leaf with a sibling if so.
2190 	 */
2191 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2192 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2193 
2194 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2195 }
2196 
2197 /*
2198  * Move all the attribute list entries from drop_leaf into save_leaf.
2199  */
2200 void
2201 xfs_attr3_leaf_unbalance(
2202 	struct xfs_da_state	*state,
2203 	struct xfs_da_state_blk	*drop_blk,
2204 	struct xfs_da_state_blk	*save_blk)
2205 {
2206 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2207 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2208 	struct xfs_attr3_icleaf_hdr drophdr;
2209 	struct xfs_attr3_icleaf_hdr savehdr;
2210 	struct xfs_attr_leaf_entry *entry;
2211 
2212 	trace_xfs_attr_leaf_unbalance(state->args);
2213 
2214 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2215 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2216 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2217 
2218 	/*
2219 	 * Save last hashval from dying block for later Btree fixup.
2220 	 */
2221 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2222 
2223 	/*
2224 	 * Check if we need a temp buffer, or can we do it in place.
2225 	 * Note that we don't check "leaf" for holes because we will
2226 	 * always be dropping it, toosmall() decided that for us already.
2227 	 */
2228 	if (savehdr.holes == 0) {
2229 		/*
2230 		 * dest leaf has no holes, so we add there.  May need
2231 		 * to make some room in the entry array.
2232 		 */
2233 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2234 					 drop_blk->bp, &drophdr)) {
2235 			xfs_attr3_leaf_moveents(state->args,
2236 						drop_leaf, &drophdr, 0,
2237 						save_leaf, &savehdr, 0,
2238 						drophdr.count);
2239 		} else {
2240 			xfs_attr3_leaf_moveents(state->args,
2241 						drop_leaf, &drophdr, 0,
2242 						save_leaf, &savehdr,
2243 						savehdr.count, drophdr.count);
2244 		}
2245 	} else {
2246 		/*
2247 		 * Destination has holes, so we make a temporary copy
2248 		 * of the leaf and add them both to that.
2249 		 */
2250 		struct xfs_attr_leafblock *tmp_leaf;
2251 		struct xfs_attr3_icleaf_hdr tmphdr;
2252 
2253 		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2254 
2255 		/*
2256 		 * Copy the header into the temp leaf so that all the stuff
2257 		 * not in the incore header is present and gets copied back in
2258 		 * once we've moved all the entries.
2259 		 */
2260 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2261 
2262 		memset(&tmphdr, 0, sizeof(tmphdr));
2263 		tmphdr.magic = savehdr.magic;
2264 		tmphdr.forw = savehdr.forw;
2265 		tmphdr.back = savehdr.back;
2266 		tmphdr.firstused = state->args->geo->blksize;
2267 
2268 		/* write the header to the temp buffer to initialise it */
2269 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2270 
2271 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2272 					 drop_blk->bp, &drophdr)) {
2273 			xfs_attr3_leaf_moveents(state->args,
2274 						drop_leaf, &drophdr, 0,
2275 						tmp_leaf, &tmphdr, 0,
2276 						drophdr.count);
2277 			xfs_attr3_leaf_moveents(state->args,
2278 						save_leaf, &savehdr, 0,
2279 						tmp_leaf, &tmphdr, tmphdr.count,
2280 						savehdr.count);
2281 		} else {
2282 			xfs_attr3_leaf_moveents(state->args,
2283 						save_leaf, &savehdr, 0,
2284 						tmp_leaf, &tmphdr, 0,
2285 						savehdr.count);
2286 			xfs_attr3_leaf_moveents(state->args,
2287 						drop_leaf, &drophdr, 0,
2288 						tmp_leaf, &tmphdr, tmphdr.count,
2289 						drophdr.count);
2290 		}
2291 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2292 		savehdr = tmphdr; /* struct copy */
2293 		kmem_free(tmp_leaf);
2294 	}
2295 
2296 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2297 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2298 					   state->args->geo->blksize - 1);
2299 
2300 	/*
2301 	 * Copy out last hashval in each block for B-tree code.
2302 	 */
2303 	entry = xfs_attr3_leaf_entryp(save_leaf);
2304 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2305 }
2306 
2307 /*========================================================================
2308  * Routines used for finding things in the Btree.
2309  *========================================================================*/
2310 
2311 /*
2312  * Look up a name in a leaf attribute list structure.
2313  * This is the internal routine, it uses the caller's buffer.
2314  *
2315  * Note that duplicate keys are allowed, but only check within the
2316  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2317  *
2318  * Return in args->index the index into the entry[] array of either
2319  * the found entry, or where the entry should have been (insert before
2320  * that entry).
2321  *
2322  * Don't change the args->value unless we find the attribute.
2323  */
2324 int
2325 xfs_attr3_leaf_lookup_int(
2326 	struct xfs_buf		*bp,
2327 	struct xfs_da_args	*args)
2328 {
2329 	struct xfs_attr_leafblock *leaf;
2330 	struct xfs_attr3_icleaf_hdr ichdr;
2331 	struct xfs_attr_leaf_entry *entry;
2332 	struct xfs_attr_leaf_entry *entries;
2333 	struct xfs_attr_leaf_name_local *name_loc;
2334 	struct xfs_attr_leaf_name_remote *name_rmt;
2335 	xfs_dahash_t		hashval;
2336 	int			probe;
2337 	int			span;
2338 
2339 	trace_xfs_attr_leaf_lookup(args);
2340 
2341 	leaf = bp->b_addr;
2342 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2343 	entries = xfs_attr3_leaf_entryp(leaf);
2344 	if (ichdr.count >= args->geo->blksize / 8) {
2345 		xfs_buf_mark_corrupt(bp);
2346 		return -EFSCORRUPTED;
2347 	}
2348 
2349 	/*
2350 	 * Binary search.  (note: small blocks will skip this loop)
2351 	 */
2352 	hashval = args->hashval;
2353 	probe = span = ichdr.count / 2;
2354 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2355 		span /= 2;
2356 		if (be32_to_cpu(entry->hashval) < hashval)
2357 			probe += span;
2358 		else if (be32_to_cpu(entry->hashval) > hashval)
2359 			probe -= span;
2360 		else
2361 			break;
2362 	}
2363 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2364 		xfs_buf_mark_corrupt(bp);
2365 		return -EFSCORRUPTED;
2366 	}
2367 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2368 		xfs_buf_mark_corrupt(bp);
2369 		return -EFSCORRUPTED;
2370 	}
2371 
2372 	/*
2373 	 * Since we may have duplicate hashval's, find the first matching
2374 	 * hashval in the leaf.
2375 	 */
2376 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2377 		entry--;
2378 		probe--;
2379 	}
2380 	while (probe < ichdr.count &&
2381 	       be32_to_cpu(entry->hashval) < hashval) {
2382 		entry++;
2383 		probe++;
2384 	}
2385 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2386 		args->index = probe;
2387 		return -ENOATTR;
2388 	}
2389 
2390 	/*
2391 	 * Duplicate keys may be present, so search all of them for a match.
2392 	 */
2393 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2394 			entry++, probe++) {
2395 /*
2396  * GROT: Add code to remove incomplete entries.
2397  */
2398 		if (entry->flags & XFS_ATTR_LOCAL) {
2399 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2400 			if (!xfs_attr_match(args, name_loc->namelen,
2401 					name_loc->nameval, entry->flags))
2402 				continue;
2403 			args->index = probe;
2404 			return -EEXIST;
2405 		} else {
2406 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2407 			if (!xfs_attr_match(args, name_rmt->namelen,
2408 					name_rmt->name, entry->flags))
2409 				continue;
2410 			args->index = probe;
2411 			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2412 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2413 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2414 							args->dp->i_mount,
2415 							args->rmtvaluelen);
2416 			return -EEXIST;
2417 		}
2418 	}
2419 	args->index = probe;
2420 	return -ENOATTR;
2421 }
2422 
2423 /*
2424  * Get the value associated with an attribute name from a leaf attribute
2425  * list structure.
2426  *
2427  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2428  * lookup, we only return an error if the attribute does not exist or we can't
2429  * retrieve the value.
2430  */
2431 int
2432 xfs_attr3_leaf_getvalue(
2433 	struct xfs_buf		*bp,
2434 	struct xfs_da_args	*args)
2435 {
2436 	struct xfs_attr_leafblock *leaf;
2437 	struct xfs_attr3_icleaf_hdr ichdr;
2438 	struct xfs_attr_leaf_entry *entry;
2439 	struct xfs_attr_leaf_name_local *name_loc;
2440 	struct xfs_attr_leaf_name_remote *name_rmt;
2441 
2442 	leaf = bp->b_addr;
2443 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2444 	ASSERT(ichdr.count < args->geo->blksize / 8);
2445 	ASSERT(args->index < ichdr.count);
2446 
2447 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2448 	if (entry->flags & XFS_ATTR_LOCAL) {
2449 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2450 		ASSERT(name_loc->namelen == args->namelen);
2451 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2452 		return xfs_attr_copy_value(args,
2453 					&name_loc->nameval[args->namelen],
2454 					be16_to_cpu(name_loc->valuelen));
2455 	}
2456 
2457 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2458 	ASSERT(name_rmt->namelen == args->namelen);
2459 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2460 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2461 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2462 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2463 					       args->rmtvaluelen);
2464 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2465 }
2466 
2467 /*========================================================================
2468  * Utility routines.
2469  *========================================================================*/
2470 
2471 /*
2472  * Move the indicated entries from one leaf to another.
2473  * NOTE: this routine modifies both source and destination leaves.
2474  */
2475 /*ARGSUSED*/
2476 STATIC void
2477 xfs_attr3_leaf_moveents(
2478 	struct xfs_da_args		*args,
2479 	struct xfs_attr_leafblock	*leaf_s,
2480 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2481 	int				start_s,
2482 	struct xfs_attr_leafblock	*leaf_d,
2483 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2484 	int				start_d,
2485 	int				count)
2486 {
2487 	struct xfs_attr_leaf_entry	*entry_s;
2488 	struct xfs_attr_leaf_entry	*entry_d;
2489 	int				desti;
2490 	int				tmp;
2491 	int				i;
2492 
2493 	/*
2494 	 * Check for nothing to do.
2495 	 */
2496 	if (count == 0)
2497 		return;
2498 
2499 	/*
2500 	 * Set up environment.
2501 	 */
2502 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2503 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2504 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2505 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2506 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2507 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2508 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2509 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2510 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2511 
2512 	ASSERT(start_s < ichdr_s->count);
2513 	ASSERT(start_d <= ichdr_d->count);
2514 	ASSERT(count <= ichdr_s->count);
2515 
2516 
2517 	/*
2518 	 * Move the entries in the destination leaf up to make a hole?
2519 	 */
2520 	if (start_d < ichdr_d->count) {
2521 		tmp  = ichdr_d->count - start_d;
2522 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2523 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2524 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2525 		memmove(entry_d, entry_s, tmp);
2526 	}
2527 
2528 	/*
2529 	 * Copy all entry's in the same (sorted) order,
2530 	 * but allocate attribute info packed and in sequence.
2531 	 */
2532 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2533 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2534 	desti = start_d;
2535 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2536 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2537 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2538 #ifdef GROT
2539 		/*
2540 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2541 		 * may also need to change the insertion index.  Code turned
2542 		 * off for 6.2, should be revisited later.
2543 		 */
2544 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2545 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2546 			ichdr_s->usedbytes -= tmp;
2547 			ichdr_s->count -= 1;
2548 			entry_d--;	/* to compensate for ++ in loop hdr */
2549 			desti--;
2550 			if ((start_s + i) < offset)
2551 				result++;	/* insertion index adjustment */
2552 		} else {
2553 #endif /* GROT */
2554 			ichdr_d->firstused -= tmp;
2555 			/* both on-disk, don't endian flip twice */
2556 			entry_d->hashval = entry_s->hashval;
2557 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2558 			entry_d->flags = entry_s->flags;
2559 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2560 							<= args->geo->blksize);
2561 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2562 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2563 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2564 							<= args->geo->blksize);
2565 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2566 			ichdr_s->usedbytes -= tmp;
2567 			ichdr_d->usedbytes += tmp;
2568 			ichdr_s->count -= 1;
2569 			ichdr_d->count += 1;
2570 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2571 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2572 			ASSERT(ichdr_d->firstused >= tmp);
2573 #ifdef GROT
2574 		}
2575 #endif /* GROT */
2576 	}
2577 
2578 	/*
2579 	 * Zero out the entries we just copied.
2580 	 */
2581 	if (start_s == ichdr_s->count) {
2582 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2583 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2584 		ASSERT(((char *)entry_s + tmp) <=
2585 		       ((char *)leaf_s + args->geo->blksize));
2586 		memset(entry_s, 0, tmp);
2587 	} else {
2588 		/*
2589 		 * Move the remaining entries down to fill the hole,
2590 		 * then zero the entries at the top.
2591 		 */
2592 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2593 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2594 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2595 		memmove(entry_d, entry_s, tmp);
2596 
2597 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2598 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2599 		ASSERT(((char *)entry_s + tmp) <=
2600 		       ((char *)leaf_s + args->geo->blksize));
2601 		memset(entry_s, 0, tmp);
2602 	}
2603 
2604 	/*
2605 	 * Fill in the freemap information
2606 	 */
2607 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2608 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2609 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2610 	ichdr_d->freemap[1].base = 0;
2611 	ichdr_d->freemap[2].base = 0;
2612 	ichdr_d->freemap[1].size = 0;
2613 	ichdr_d->freemap[2].size = 0;
2614 	ichdr_s->holes = 1;	/* leaf may not be compact */
2615 }
2616 
2617 /*
2618  * Pick up the last hashvalue from a leaf block.
2619  */
2620 xfs_dahash_t
2621 xfs_attr_leaf_lasthash(
2622 	struct xfs_buf	*bp,
2623 	int		*count)
2624 {
2625 	struct xfs_attr3_icleaf_hdr ichdr;
2626 	struct xfs_attr_leaf_entry *entries;
2627 	struct xfs_mount *mp = bp->b_mount;
2628 
2629 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2630 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2631 	if (count)
2632 		*count = ichdr.count;
2633 	if (!ichdr.count)
2634 		return 0;
2635 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2636 }
2637 
2638 /*
2639  * Calculate the number of bytes used to store the indicated attribute
2640  * (whether local or remote only calculate bytes in this block).
2641  */
2642 STATIC int
2643 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2644 {
2645 	struct xfs_attr_leaf_entry *entries;
2646 	xfs_attr_leaf_name_local_t *name_loc;
2647 	xfs_attr_leaf_name_remote_t *name_rmt;
2648 	int size;
2649 
2650 	entries = xfs_attr3_leaf_entryp(leaf);
2651 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2652 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2653 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2654 						   be16_to_cpu(name_loc->valuelen));
2655 	} else {
2656 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2657 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2658 	}
2659 	return size;
2660 }
2661 
2662 /*
2663  * Calculate the number of bytes that would be required to store the new
2664  * attribute (whether local or remote only calculate bytes in this block).
2665  * This routine decides as a side effect whether the attribute will be
2666  * a "local" or a "remote" attribute.
2667  */
2668 int
2669 xfs_attr_leaf_newentsize(
2670 	struct xfs_da_args	*args,
2671 	int			*local)
2672 {
2673 	int			size;
2674 
2675 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2676 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2677 		if (local)
2678 			*local = 1;
2679 		return size;
2680 	}
2681 	if (local)
2682 		*local = 0;
2683 	return xfs_attr_leaf_entsize_remote(args->namelen);
2684 }
2685 
2686 
2687 /*========================================================================
2688  * Manage the INCOMPLETE flag in a leaf entry
2689  *========================================================================*/
2690 
2691 /*
2692  * Clear the INCOMPLETE flag on an entry in a leaf block.
2693  */
2694 int
2695 xfs_attr3_leaf_clearflag(
2696 	struct xfs_da_args	*args)
2697 {
2698 	struct xfs_attr_leafblock *leaf;
2699 	struct xfs_attr_leaf_entry *entry;
2700 	struct xfs_attr_leaf_name_remote *name_rmt;
2701 	struct xfs_buf		*bp;
2702 	int			error;
2703 #ifdef DEBUG
2704 	struct xfs_attr3_icleaf_hdr ichdr;
2705 	xfs_attr_leaf_name_local_t *name_loc;
2706 	int namelen;
2707 	char *name;
2708 #endif /* DEBUG */
2709 
2710 	trace_xfs_attr_leaf_clearflag(args);
2711 	/*
2712 	 * Set up the operation.
2713 	 */
2714 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2715 	if (error)
2716 		return error;
2717 
2718 	leaf = bp->b_addr;
2719 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2720 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2721 
2722 #ifdef DEBUG
2723 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2724 	ASSERT(args->index < ichdr.count);
2725 	ASSERT(args->index >= 0);
2726 
2727 	if (entry->flags & XFS_ATTR_LOCAL) {
2728 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2729 		namelen = name_loc->namelen;
2730 		name = (char *)name_loc->nameval;
2731 	} else {
2732 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2733 		namelen = name_rmt->namelen;
2734 		name = (char *)name_rmt->name;
2735 	}
2736 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2737 	ASSERT(namelen == args->namelen);
2738 	ASSERT(memcmp(name, args->name, namelen) == 0);
2739 #endif /* DEBUG */
2740 
2741 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2742 	xfs_trans_log_buf(args->trans, bp,
2743 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2744 
2745 	if (args->rmtblkno) {
2746 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2747 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2748 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2749 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2750 		xfs_trans_log_buf(args->trans, bp,
2751 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2752 	}
2753 
2754 	return 0;
2755 }
2756 
2757 /*
2758  * Set the INCOMPLETE flag on an entry in a leaf block.
2759  */
2760 int
2761 xfs_attr3_leaf_setflag(
2762 	struct xfs_da_args	*args)
2763 {
2764 	struct xfs_attr_leafblock *leaf;
2765 	struct xfs_attr_leaf_entry *entry;
2766 	struct xfs_attr_leaf_name_remote *name_rmt;
2767 	struct xfs_buf		*bp;
2768 	int error;
2769 #ifdef DEBUG
2770 	struct xfs_attr3_icleaf_hdr ichdr;
2771 #endif
2772 
2773 	trace_xfs_attr_leaf_setflag(args);
2774 
2775 	/*
2776 	 * Set up the operation.
2777 	 */
2778 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2779 	if (error)
2780 		return error;
2781 
2782 	leaf = bp->b_addr;
2783 #ifdef DEBUG
2784 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2785 	ASSERT(args->index < ichdr.count);
2786 	ASSERT(args->index >= 0);
2787 #endif
2788 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2789 
2790 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2791 	entry->flags |= XFS_ATTR_INCOMPLETE;
2792 	xfs_trans_log_buf(args->trans, bp,
2793 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2794 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2795 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2796 		name_rmt->valueblk = 0;
2797 		name_rmt->valuelen = 0;
2798 		xfs_trans_log_buf(args->trans, bp,
2799 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2800 	}
2801 
2802 	return 0;
2803 }
2804 
2805 /*
2806  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2807  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2808  * entry given by args->blkno2/index2.
2809  *
2810  * Note that they could be in different blocks, or in the same block.
2811  */
2812 int
2813 xfs_attr3_leaf_flipflags(
2814 	struct xfs_da_args	*args)
2815 {
2816 	struct xfs_attr_leafblock *leaf1;
2817 	struct xfs_attr_leafblock *leaf2;
2818 	struct xfs_attr_leaf_entry *entry1;
2819 	struct xfs_attr_leaf_entry *entry2;
2820 	struct xfs_attr_leaf_name_remote *name_rmt;
2821 	struct xfs_buf		*bp1;
2822 	struct xfs_buf		*bp2;
2823 	int error;
2824 #ifdef DEBUG
2825 	struct xfs_attr3_icleaf_hdr ichdr1;
2826 	struct xfs_attr3_icleaf_hdr ichdr2;
2827 	xfs_attr_leaf_name_local_t *name_loc;
2828 	int namelen1, namelen2;
2829 	char *name1, *name2;
2830 #endif /* DEBUG */
2831 
2832 	trace_xfs_attr_leaf_flipflags(args);
2833 
2834 	/*
2835 	 * Read the block containing the "old" attr
2836 	 */
2837 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2838 	if (error)
2839 		return error;
2840 
2841 	/*
2842 	 * Read the block containing the "new" attr, if it is different
2843 	 */
2844 	if (args->blkno2 != args->blkno) {
2845 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2846 					   &bp2);
2847 		if (error)
2848 			return error;
2849 	} else {
2850 		bp2 = bp1;
2851 	}
2852 
2853 	leaf1 = bp1->b_addr;
2854 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2855 
2856 	leaf2 = bp2->b_addr;
2857 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2858 
2859 #ifdef DEBUG
2860 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2861 	ASSERT(args->index < ichdr1.count);
2862 	ASSERT(args->index >= 0);
2863 
2864 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2865 	ASSERT(args->index2 < ichdr2.count);
2866 	ASSERT(args->index2 >= 0);
2867 
2868 	if (entry1->flags & XFS_ATTR_LOCAL) {
2869 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2870 		namelen1 = name_loc->namelen;
2871 		name1 = (char *)name_loc->nameval;
2872 	} else {
2873 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2874 		namelen1 = name_rmt->namelen;
2875 		name1 = (char *)name_rmt->name;
2876 	}
2877 	if (entry2->flags & XFS_ATTR_LOCAL) {
2878 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2879 		namelen2 = name_loc->namelen;
2880 		name2 = (char *)name_loc->nameval;
2881 	} else {
2882 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2883 		namelen2 = name_rmt->namelen;
2884 		name2 = (char *)name_rmt->name;
2885 	}
2886 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2887 	ASSERT(namelen1 == namelen2);
2888 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2889 #endif /* DEBUG */
2890 
2891 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2892 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2893 
2894 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2895 	xfs_trans_log_buf(args->trans, bp1,
2896 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2897 	if (args->rmtblkno) {
2898 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2899 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2900 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2901 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2902 		xfs_trans_log_buf(args->trans, bp1,
2903 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2904 	}
2905 
2906 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2907 	xfs_trans_log_buf(args->trans, bp2,
2908 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2909 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2910 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2911 		name_rmt->valueblk = 0;
2912 		name_rmt->valuelen = 0;
2913 		xfs_trans_log_buf(args->trans, bp2,
2914 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2915 	}
2916 
2917 	return 0;
2918 }
2919