xref: /openbmc/linux/fs/xfs/libxfs/xfs_attr_leaf.c (revision c01147d9)
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 	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
694 	hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
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 -EEXIST if attr is found, or -ENOATTR if not
702  * args:  args containing attribute name and namelen
703  * sfep:  If not null, pointer will be set to the last attr entry found on
704 	  -EEXIST.  On -ENOATTR pointer is left at the last entry in the list
705  * basep: If not null, pointer is set to the byte offset of the entry in the
706  *	  list on -EEXIST.  On -ENOATTR, pointer is left at the byte offset of
707  *	  the last entry in the list
708  */
709 int
710 xfs_attr_sf_findname(
711 	struct xfs_da_args	 *args,
712 	struct xfs_attr_sf_entry **sfep,
713 	unsigned int		 *basep)
714 {
715 	struct xfs_attr_shortform *sf;
716 	struct xfs_attr_sf_entry *sfe;
717 	unsigned int		base = sizeof(struct xfs_attr_sf_hdr);
718 	int			size = 0;
719 	int			end;
720 	int			i;
721 
722 	sf = (struct xfs_attr_shortform *)args->dp->i_af.if_u1.if_data;
723 	sfe = &sf->list[0];
724 	end = sf->hdr.count;
725 	for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe),
726 			     base += size, i++) {
727 		size = xfs_attr_sf_entsize(sfe);
728 		if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
729 				    sfe->flags))
730 			continue;
731 		break;
732 	}
733 
734 	if (sfep != NULL)
735 		*sfep = sfe;
736 
737 	if (basep != NULL)
738 		*basep = base;
739 
740 	if (i == end)
741 		return -ENOATTR;
742 	return -EEXIST;
743 }
744 
745 /*
746  * Add a name/value pair to the shortform attribute list.
747  * Overflow from the inode has already been checked for.
748  */
749 void
750 xfs_attr_shortform_add(
751 	struct xfs_da_args		*args,
752 	int				forkoff)
753 {
754 	struct xfs_attr_shortform	*sf;
755 	struct xfs_attr_sf_entry	*sfe;
756 	int				offset, size;
757 	struct xfs_mount		*mp;
758 	struct xfs_inode		*dp;
759 	struct xfs_ifork		*ifp;
760 
761 	trace_xfs_attr_sf_add(args);
762 
763 	dp = args->dp;
764 	mp = dp->i_mount;
765 	dp->i_forkoff = forkoff;
766 
767 	ifp = &dp->i_af;
768 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
769 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
770 	if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
771 		ASSERT(0);
772 
773 	offset = (char *)sfe - (char *)sf;
774 	size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
775 	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
776 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
777 	sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset);
778 
779 	sfe->namelen = args->namelen;
780 	sfe->valuelen = args->valuelen;
781 	sfe->flags = args->attr_filter;
782 	memcpy(sfe->nameval, args->name, args->namelen);
783 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
784 	sf->hdr.count++;
785 	be16_add_cpu(&sf->hdr.totsize, size);
786 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
787 
788 	xfs_sbversion_add_attr2(mp, args->trans);
789 }
790 
791 /*
792  * After the last attribute is removed revert to original inode format,
793  * making all literal area available to the data fork once more.
794  */
795 void
796 xfs_attr_fork_remove(
797 	struct xfs_inode	*ip,
798 	struct xfs_trans	*tp)
799 {
800 	ASSERT(ip->i_af.if_nextents == 0);
801 
802 	xfs_idestroy_fork(&ip->i_af);
803 	xfs_ifork_zap_attr(ip);
804 	ip->i_forkoff = 0;
805 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
806 }
807 
808 /*
809  * Remove an attribute from the shortform attribute list structure.
810  */
811 int
812 xfs_attr_sf_removename(
813 	struct xfs_da_args		*args)
814 {
815 	struct xfs_attr_shortform	*sf;
816 	struct xfs_attr_sf_entry	*sfe;
817 	int				size = 0, end, totsize;
818 	unsigned int			base;
819 	struct xfs_mount		*mp;
820 	struct xfs_inode		*dp;
821 	int				error;
822 
823 	trace_xfs_attr_sf_remove(args);
824 
825 	dp = args->dp;
826 	mp = dp->i_mount;
827 	sf = (struct xfs_attr_shortform *)dp->i_af.if_u1.if_data;
828 
829 	error = xfs_attr_sf_findname(args, &sfe, &base);
830 
831 	/*
832 	 * If we are recovering an operation, finding nothing to
833 	 * remove is not an error - it just means there was nothing
834 	 * to clean up.
835 	 */
836 	if (error == -ENOATTR && (args->op_flags & XFS_DA_OP_RECOVERY))
837 		return 0;
838 	if (error != -EEXIST)
839 		return error;
840 	size = xfs_attr_sf_entsize(sfe);
841 
842 	/*
843 	 * Fix up the attribute fork data, covering the hole
844 	 */
845 	end = base + size;
846 	totsize = be16_to_cpu(sf->hdr.totsize);
847 	if (end != totsize)
848 		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
849 	sf->hdr.count--;
850 	be16_add_cpu(&sf->hdr.totsize, -size);
851 
852 	/*
853 	 * Fix up the start offset of the attribute fork
854 	 */
855 	totsize -= size;
856 	if (totsize == sizeof(xfs_attr_sf_hdr_t) && xfs_has_attr2(mp) &&
857 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
858 	    !(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE))) {
859 		xfs_attr_fork_remove(dp, args->trans);
860 	} else {
861 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
862 		dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
863 		ASSERT(dp->i_forkoff);
864 		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
865 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
866 				!xfs_has_attr2(mp) ||
867 				dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
868 		xfs_trans_log_inode(args->trans, dp,
869 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
870 	}
871 
872 	xfs_sbversion_add_attr2(mp, args->trans);
873 
874 	return 0;
875 }
876 
877 /*
878  * Look up a name in a shortform attribute list structure.
879  */
880 /*ARGSUSED*/
881 int
882 xfs_attr_shortform_lookup(xfs_da_args_t *args)
883 {
884 	struct xfs_attr_shortform *sf;
885 	struct xfs_attr_sf_entry *sfe;
886 	int i;
887 	struct xfs_ifork *ifp;
888 
889 	trace_xfs_attr_sf_lookup(args);
890 
891 	ifp = &args->dp->i_af;
892 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
893 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
894 	sfe = &sf->list[0];
895 	for (i = 0; i < sf->hdr.count;
896 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
897 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
898 				sfe->flags))
899 			return -EEXIST;
900 	}
901 	return -ENOATTR;
902 }
903 
904 /*
905  * Retrieve the attribute value and length.
906  *
907  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
908  * lookup, we only return an error if the attribute does not exist or we can't
909  * retrieve the value.
910  */
911 int
912 xfs_attr_shortform_getvalue(
913 	struct xfs_da_args	*args)
914 {
915 	struct xfs_attr_shortform *sf;
916 	struct xfs_attr_sf_entry *sfe;
917 	int			i;
918 
919 	ASSERT(args->dp->i_af.if_format == XFS_DINODE_FMT_LOCAL);
920 	sf = (struct xfs_attr_shortform *)args->dp->i_af.if_u1.if_data;
921 	sfe = &sf->list[0];
922 	for (i = 0; i < sf->hdr.count;
923 				sfe = xfs_attr_sf_nextentry(sfe), i++) {
924 		if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
925 				sfe->flags))
926 			return xfs_attr_copy_value(args,
927 				&sfe->nameval[args->namelen], sfe->valuelen);
928 	}
929 	return -ENOATTR;
930 }
931 
932 /* Convert from using the shortform to the leaf format. */
933 int
934 xfs_attr_shortform_to_leaf(
935 	struct xfs_da_args		*args)
936 {
937 	struct xfs_inode		*dp;
938 	struct xfs_attr_shortform	*sf;
939 	struct xfs_attr_sf_entry	*sfe;
940 	struct xfs_da_args		nargs;
941 	char				*tmpbuffer;
942 	int				error, i, size;
943 	xfs_dablk_t			blkno;
944 	struct xfs_buf			*bp;
945 	struct xfs_ifork		*ifp;
946 
947 	trace_xfs_attr_sf_to_leaf(args);
948 
949 	dp = args->dp;
950 	ifp = &dp->i_af;
951 	sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
952 	size = be16_to_cpu(sf->hdr.totsize);
953 	tmpbuffer = kmem_alloc(size, 0);
954 	ASSERT(tmpbuffer != NULL);
955 	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
956 	sf = (struct xfs_attr_shortform *)tmpbuffer;
957 
958 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
959 	xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
960 
961 	bp = NULL;
962 	error = xfs_da_grow_inode(args, &blkno);
963 	if (error)
964 		goto out;
965 
966 	ASSERT(blkno == 0);
967 	error = xfs_attr3_leaf_create(args, blkno, &bp);
968 	if (error)
969 		goto out;
970 
971 	memset((char *)&nargs, 0, sizeof(nargs));
972 	nargs.dp = dp;
973 	nargs.geo = args->geo;
974 	nargs.total = args->total;
975 	nargs.whichfork = XFS_ATTR_FORK;
976 	nargs.trans = args->trans;
977 	nargs.op_flags = XFS_DA_OP_OKNOENT;
978 
979 	sfe = &sf->list[0];
980 	for (i = 0; i < sf->hdr.count; i++) {
981 		nargs.name = sfe->nameval;
982 		nargs.namelen = sfe->namelen;
983 		nargs.value = &sfe->nameval[nargs.namelen];
984 		nargs.valuelen = sfe->valuelen;
985 		nargs.hashval = xfs_da_hashname(sfe->nameval,
986 						sfe->namelen);
987 		nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
988 		error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
989 		ASSERT(error == -ENOATTR);
990 		error = xfs_attr3_leaf_add(bp, &nargs);
991 		ASSERT(error != -ENOSPC);
992 		if (error)
993 			goto out;
994 		sfe = xfs_attr_sf_nextentry(sfe);
995 	}
996 	error = 0;
997 out:
998 	kmem_free(tmpbuffer);
999 	return error;
1000 }
1001 
1002 /*
1003  * Check a leaf attribute block to see if all the entries would fit into
1004  * a shortform attribute list.
1005  */
1006 int
1007 xfs_attr_shortform_allfit(
1008 	struct xfs_buf		*bp,
1009 	struct xfs_inode	*dp)
1010 {
1011 	struct xfs_attr_leafblock *leaf;
1012 	struct xfs_attr_leaf_entry *entry;
1013 	xfs_attr_leaf_name_local_t *name_loc;
1014 	struct xfs_attr3_icleaf_hdr leafhdr;
1015 	int			bytes;
1016 	int			i;
1017 	struct xfs_mount	*mp = bp->b_mount;
1018 
1019 	leaf = bp->b_addr;
1020 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
1021 	entry = xfs_attr3_leaf_entryp(leaf);
1022 
1023 	bytes = sizeof(struct xfs_attr_sf_hdr);
1024 	for (i = 0; i < leafhdr.count; entry++, i++) {
1025 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1026 			continue;		/* don't copy partial entries */
1027 		if (!(entry->flags & XFS_ATTR_LOCAL))
1028 			return 0;
1029 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1030 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
1031 			return 0;
1032 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1033 			return 0;
1034 		bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1035 					be16_to_cpu(name_loc->valuelen));
1036 	}
1037 	if (xfs_has_attr2(dp->i_mount) &&
1038 	    (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1039 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
1040 		return -1;
1041 	return xfs_attr_shortform_bytesfit(dp, bytes);
1042 }
1043 
1044 /* Verify the consistency of an inline attribute fork. */
1045 xfs_failaddr_t
1046 xfs_attr_shortform_verify(
1047 	struct xfs_inode		*ip)
1048 {
1049 	struct xfs_attr_shortform	*sfp;
1050 	struct xfs_attr_sf_entry	*sfep;
1051 	struct xfs_attr_sf_entry	*next_sfep;
1052 	char				*endp;
1053 	struct xfs_ifork		*ifp;
1054 	int				i;
1055 	int64_t				size;
1056 
1057 	ASSERT(ip->i_af.if_format == XFS_DINODE_FMT_LOCAL);
1058 	ifp = xfs_ifork_ptr(ip, XFS_ATTR_FORK);
1059 	sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1060 	size = ifp->if_bytes;
1061 
1062 	/*
1063 	 * Give up if the attribute is way too short.
1064 	 */
1065 	if (size < sizeof(struct xfs_attr_sf_hdr))
1066 		return __this_address;
1067 
1068 	endp = (char *)sfp + size;
1069 
1070 	/* Check all reported entries */
1071 	sfep = &sfp->list[0];
1072 	for (i = 0; i < sfp->hdr.count; i++) {
1073 		/*
1074 		 * struct xfs_attr_sf_entry has a variable length.
1075 		 * Check the fixed-offset parts of the structure are
1076 		 * within the data buffer.
1077 		 * xfs_attr_sf_entry is defined with a 1-byte variable
1078 		 * array at the end, so we must subtract that off.
1079 		 */
1080 		if (((char *)sfep + sizeof(*sfep)) >= endp)
1081 			return __this_address;
1082 
1083 		/* Don't allow names with known bad length. */
1084 		if (sfep->namelen == 0)
1085 			return __this_address;
1086 
1087 		/*
1088 		 * Check that the variable-length part of the structure is
1089 		 * within the data buffer.  The next entry starts after the
1090 		 * name component, so nextentry is an acceptable test.
1091 		 */
1092 		next_sfep = xfs_attr_sf_nextentry(sfep);
1093 		if ((char *)next_sfep > endp)
1094 			return __this_address;
1095 
1096 		/*
1097 		 * Check for unknown flags.  Short form doesn't support
1098 		 * the incomplete or local bits, so we can use the namespace
1099 		 * mask here.
1100 		 */
1101 		if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1102 			return __this_address;
1103 
1104 		/*
1105 		 * Check for invalid namespace combinations.  We only allow
1106 		 * one namespace flag per xattr, so we can just count the
1107 		 * bits (i.e. hweight) here.
1108 		 */
1109 		if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1110 			return __this_address;
1111 
1112 		sfep = next_sfep;
1113 	}
1114 	if ((void *)sfep != (void *)endp)
1115 		return __this_address;
1116 
1117 	return NULL;
1118 }
1119 
1120 /*
1121  * Convert a leaf attribute list to shortform attribute list
1122  */
1123 int
1124 xfs_attr3_leaf_to_shortform(
1125 	struct xfs_buf		*bp,
1126 	struct xfs_da_args	*args,
1127 	int			forkoff)
1128 {
1129 	struct xfs_attr_leafblock *leaf;
1130 	struct xfs_attr3_icleaf_hdr ichdr;
1131 	struct xfs_attr_leaf_entry *entry;
1132 	struct xfs_attr_leaf_name_local *name_loc;
1133 	struct xfs_da_args	nargs;
1134 	struct xfs_inode	*dp = args->dp;
1135 	char			*tmpbuffer;
1136 	int			error;
1137 	int			i;
1138 
1139 	trace_xfs_attr_leaf_to_sf(args);
1140 
1141 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1142 	if (!tmpbuffer)
1143 		return -ENOMEM;
1144 
1145 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1146 
1147 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1148 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1149 	entry = xfs_attr3_leaf_entryp(leaf);
1150 
1151 	/* XXX (dgc): buffer is about to be marked stale - why zero it? */
1152 	memset(bp->b_addr, 0, args->geo->blksize);
1153 
1154 	/*
1155 	 * Clean out the prior contents of the attribute list.
1156 	 */
1157 	error = xfs_da_shrink_inode(args, 0, bp);
1158 	if (error)
1159 		goto out;
1160 
1161 	if (forkoff == -1) {
1162 		/*
1163 		 * Don't remove the attr fork if this operation is the first
1164 		 * part of a attr replace operations. We're going to add a new
1165 		 * attr immediately, so we need to keep the attr fork around in
1166 		 * this case.
1167 		 */
1168 		if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1169 			ASSERT(xfs_has_attr2(dp->i_mount));
1170 			ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1171 			xfs_attr_fork_remove(dp, args->trans);
1172 		}
1173 		goto out;
1174 	}
1175 
1176 	xfs_attr_shortform_create(args);
1177 
1178 	/*
1179 	 * Copy the attributes
1180 	 */
1181 	memset((char *)&nargs, 0, sizeof(nargs));
1182 	nargs.geo = args->geo;
1183 	nargs.dp = dp;
1184 	nargs.total = args->total;
1185 	nargs.whichfork = XFS_ATTR_FORK;
1186 	nargs.trans = args->trans;
1187 	nargs.op_flags = XFS_DA_OP_OKNOENT;
1188 
1189 	for (i = 0; i < ichdr.count; entry++, i++) {
1190 		if (entry->flags & XFS_ATTR_INCOMPLETE)
1191 			continue;	/* don't copy partial entries */
1192 		if (!entry->nameidx)
1193 			continue;
1194 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
1195 		name_loc = xfs_attr3_leaf_name_local(leaf, i);
1196 		nargs.name = name_loc->nameval;
1197 		nargs.namelen = name_loc->namelen;
1198 		nargs.value = &name_loc->nameval[nargs.namelen];
1199 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1200 		nargs.hashval = be32_to_cpu(entry->hashval);
1201 		nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1202 		xfs_attr_shortform_add(&nargs, forkoff);
1203 	}
1204 	error = 0;
1205 
1206 out:
1207 	kmem_free(tmpbuffer);
1208 	return error;
1209 }
1210 
1211 /*
1212  * Convert from using a single leaf to a root node and a leaf.
1213  */
1214 int
1215 xfs_attr3_leaf_to_node(
1216 	struct xfs_da_args	*args)
1217 {
1218 	struct xfs_attr_leafblock *leaf;
1219 	struct xfs_attr3_icleaf_hdr icleafhdr;
1220 	struct xfs_attr_leaf_entry *entries;
1221 	struct xfs_da3_icnode_hdr icnodehdr;
1222 	struct xfs_da_intnode	*node;
1223 	struct xfs_inode	*dp = args->dp;
1224 	struct xfs_mount	*mp = dp->i_mount;
1225 	struct xfs_buf		*bp1 = NULL;
1226 	struct xfs_buf		*bp2 = NULL;
1227 	xfs_dablk_t		blkno;
1228 	int			error;
1229 
1230 	trace_xfs_attr_leaf_to_node(args);
1231 
1232 	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1233 		error = -EIO;
1234 		goto out;
1235 	}
1236 
1237 	error = xfs_da_grow_inode(args, &blkno);
1238 	if (error)
1239 		goto out;
1240 	error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1241 	if (error)
1242 		goto out;
1243 
1244 	error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1245 	if (error)
1246 		goto out;
1247 
1248 	/* copy leaf to new buffer, update identifiers */
1249 	xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1250 	bp2->b_ops = bp1->b_ops;
1251 	memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1252 	if (xfs_has_crc(mp)) {
1253 		struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1254 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp2));
1255 	}
1256 	xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1257 
1258 	/*
1259 	 * Set up the new root node.
1260 	 */
1261 	error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1262 	if (error)
1263 		goto out;
1264 	node = bp1->b_addr;
1265 	xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1266 
1267 	leaf = bp2->b_addr;
1268 	xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1269 	entries = xfs_attr3_leaf_entryp(leaf);
1270 
1271 	/* both on-disk, don't endian-flip twice */
1272 	icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1273 	icnodehdr.btree[0].before = cpu_to_be32(blkno);
1274 	icnodehdr.count = 1;
1275 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1276 	xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1277 	error = 0;
1278 out:
1279 	return error;
1280 }
1281 
1282 /*========================================================================
1283  * Routines used for growing the Btree.
1284  *========================================================================*/
1285 
1286 /*
1287  * Create the initial contents of a leaf attribute list
1288  * or a leaf in a node attribute list.
1289  */
1290 STATIC int
1291 xfs_attr3_leaf_create(
1292 	struct xfs_da_args	*args,
1293 	xfs_dablk_t		blkno,
1294 	struct xfs_buf		**bpp)
1295 {
1296 	struct xfs_attr_leafblock *leaf;
1297 	struct xfs_attr3_icleaf_hdr ichdr;
1298 	struct xfs_inode	*dp = args->dp;
1299 	struct xfs_mount	*mp = dp->i_mount;
1300 	struct xfs_buf		*bp;
1301 	int			error;
1302 
1303 	trace_xfs_attr_leaf_create(args);
1304 
1305 	error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1306 					    XFS_ATTR_FORK);
1307 	if (error)
1308 		return error;
1309 	bp->b_ops = &xfs_attr3_leaf_buf_ops;
1310 	xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1311 	leaf = bp->b_addr;
1312 	memset(leaf, 0, args->geo->blksize);
1313 
1314 	memset(&ichdr, 0, sizeof(ichdr));
1315 	ichdr.firstused = args->geo->blksize;
1316 
1317 	if (xfs_has_crc(mp)) {
1318 		struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1319 
1320 		ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1321 
1322 		hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1323 		hdr3->owner = cpu_to_be64(dp->i_ino);
1324 		uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1325 
1326 		ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1327 	} else {
1328 		ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1329 		ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1330 	}
1331 	ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1332 
1333 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1334 	xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1335 
1336 	*bpp = bp;
1337 	return 0;
1338 }
1339 
1340 /*
1341  * Split the leaf node, rebalance, then add the new entry.
1342  */
1343 int
1344 xfs_attr3_leaf_split(
1345 	struct xfs_da_state	*state,
1346 	struct xfs_da_state_blk	*oldblk,
1347 	struct xfs_da_state_blk	*newblk)
1348 {
1349 	xfs_dablk_t blkno;
1350 	int error;
1351 
1352 	trace_xfs_attr_leaf_split(state->args);
1353 
1354 	/*
1355 	 * Allocate space for a new leaf node.
1356 	 */
1357 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1358 	error = xfs_da_grow_inode(state->args, &blkno);
1359 	if (error)
1360 		return error;
1361 	error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1362 	if (error)
1363 		return error;
1364 	newblk->blkno = blkno;
1365 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
1366 
1367 	/*
1368 	 * Rebalance the entries across the two leaves.
1369 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
1370 	 */
1371 	xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1372 	error = xfs_da3_blk_link(state, oldblk, newblk);
1373 	if (error)
1374 		return error;
1375 
1376 	/*
1377 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1378 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1379 	 * "new" attrs info.  Will need the "old" info to remove it later.
1380 	 *
1381 	 * Insert the "new" entry in the correct block.
1382 	 */
1383 	if (state->inleaf) {
1384 		trace_xfs_attr_leaf_add_old(state->args);
1385 		error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1386 	} else {
1387 		trace_xfs_attr_leaf_add_new(state->args);
1388 		error = xfs_attr3_leaf_add(newblk->bp, state->args);
1389 	}
1390 
1391 	/*
1392 	 * Update last hashval in each block since we added the name.
1393 	 */
1394 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1395 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1396 	return error;
1397 }
1398 
1399 /*
1400  * Add a name to the leaf attribute list structure.
1401  */
1402 int
1403 xfs_attr3_leaf_add(
1404 	struct xfs_buf		*bp,
1405 	struct xfs_da_args	*args)
1406 {
1407 	struct xfs_attr_leafblock *leaf;
1408 	struct xfs_attr3_icleaf_hdr ichdr;
1409 	int			tablesize;
1410 	int			entsize;
1411 	int			sum;
1412 	int			tmp;
1413 	int			i;
1414 
1415 	trace_xfs_attr_leaf_add(args);
1416 
1417 	leaf = bp->b_addr;
1418 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1419 	ASSERT(args->index >= 0 && args->index <= ichdr.count);
1420 	entsize = xfs_attr_leaf_newentsize(args, NULL);
1421 
1422 	/*
1423 	 * Search through freemap for first-fit on new name length.
1424 	 * (may need to figure in size of entry struct too)
1425 	 */
1426 	tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1427 					+ xfs_attr3_leaf_hdr_size(leaf);
1428 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1429 		if (tablesize > ichdr.firstused) {
1430 			sum += ichdr.freemap[i].size;
1431 			continue;
1432 		}
1433 		if (!ichdr.freemap[i].size)
1434 			continue;	/* no space in this map */
1435 		tmp = entsize;
1436 		if (ichdr.freemap[i].base < ichdr.firstused)
1437 			tmp += sizeof(xfs_attr_leaf_entry_t);
1438 		if (ichdr.freemap[i].size >= tmp) {
1439 			tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1440 			goto out_log_hdr;
1441 		}
1442 		sum += ichdr.freemap[i].size;
1443 	}
1444 
1445 	/*
1446 	 * If there are no holes in the address space of the block,
1447 	 * and we don't have enough freespace, then compaction will do us
1448 	 * no good and we should just give up.
1449 	 */
1450 	if (!ichdr.holes && sum < entsize)
1451 		return -ENOSPC;
1452 
1453 	/*
1454 	 * Compact the entries to coalesce free space.
1455 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1456 	 */
1457 	xfs_attr3_leaf_compact(args, &ichdr, bp);
1458 
1459 	/*
1460 	 * After compaction, the block is guaranteed to have only one
1461 	 * free region, in freemap[0].  If it is not big enough, give up.
1462 	 */
1463 	if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1464 		tmp = -ENOSPC;
1465 		goto out_log_hdr;
1466 	}
1467 
1468 	tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1469 
1470 out_log_hdr:
1471 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1472 	xfs_trans_log_buf(args->trans, bp,
1473 		XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1474 				xfs_attr3_leaf_hdr_size(leaf)));
1475 	return tmp;
1476 }
1477 
1478 /*
1479  * Add a name to a leaf attribute list structure.
1480  */
1481 STATIC int
1482 xfs_attr3_leaf_add_work(
1483 	struct xfs_buf		*bp,
1484 	struct xfs_attr3_icleaf_hdr *ichdr,
1485 	struct xfs_da_args	*args,
1486 	int			mapindex)
1487 {
1488 	struct xfs_attr_leafblock *leaf;
1489 	struct xfs_attr_leaf_entry *entry;
1490 	struct xfs_attr_leaf_name_local *name_loc;
1491 	struct xfs_attr_leaf_name_remote *name_rmt;
1492 	struct xfs_mount	*mp;
1493 	int			tmp;
1494 	int			i;
1495 
1496 	trace_xfs_attr_leaf_add_work(args);
1497 
1498 	leaf = bp->b_addr;
1499 	ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1500 	ASSERT(args->index >= 0 && args->index <= ichdr->count);
1501 
1502 	/*
1503 	 * Force open some space in the entry array and fill it in.
1504 	 */
1505 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1506 	if (args->index < ichdr->count) {
1507 		tmp  = ichdr->count - args->index;
1508 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1509 		memmove(entry + 1, entry, tmp);
1510 		xfs_trans_log_buf(args->trans, bp,
1511 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1512 	}
1513 	ichdr->count++;
1514 
1515 	/*
1516 	 * Allocate space for the new string (at the end of the run).
1517 	 */
1518 	mp = args->trans->t_mountp;
1519 	ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1520 	ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1521 	ASSERT(ichdr->freemap[mapindex].size >=
1522 		xfs_attr_leaf_newentsize(args, NULL));
1523 	ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1524 	ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1525 
1526 	ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1527 
1528 	entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1529 				     ichdr->freemap[mapindex].size);
1530 	entry->hashval = cpu_to_be32(args->hashval);
1531 	entry->flags = args->attr_filter;
1532 	if (tmp)
1533 		entry->flags |= XFS_ATTR_LOCAL;
1534 	if (args->op_flags & XFS_DA_OP_REPLACE) {
1535 		if (!(args->op_flags & XFS_DA_OP_LOGGED))
1536 			entry->flags |= XFS_ATTR_INCOMPLETE;
1537 		if ((args->blkno2 == args->blkno) &&
1538 		    (args->index2 <= args->index)) {
1539 			args->index2++;
1540 		}
1541 	}
1542 	xfs_trans_log_buf(args->trans, bp,
1543 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1544 	ASSERT((args->index == 0) ||
1545 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1546 	ASSERT((args->index == ichdr->count - 1) ||
1547 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1548 
1549 	/*
1550 	 * For "remote" attribute values, simply note that we need to
1551 	 * allocate space for the "remote" value.  We can't actually
1552 	 * allocate the extents in this transaction, and we can't decide
1553 	 * which blocks they should be as we might allocate more blocks
1554 	 * as part of this transaction (a split operation for example).
1555 	 */
1556 	if (entry->flags & XFS_ATTR_LOCAL) {
1557 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1558 		name_loc->namelen = args->namelen;
1559 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1560 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1561 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1562 				   be16_to_cpu(name_loc->valuelen));
1563 	} else {
1564 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1565 		name_rmt->namelen = args->namelen;
1566 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1567 		entry->flags |= XFS_ATTR_INCOMPLETE;
1568 		/* just in case */
1569 		name_rmt->valuelen = 0;
1570 		name_rmt->valueblk = 0;
1571 		args->rmtblkno = 1;
1572 		args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1573 		args->rmtvaluelen = args->valuelen;
1574 	}
1575 	xfs_trans_log_buf(args->trans, bp,
1576 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1577 				   xfs_attr_leaf_entsize(leaf, args->index)));
1578 
1579 	/*
1580 	 * Update the control info for this leaf node
1581 	 */
1582 	if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1583 		ichdr->firstused = be16_to_cpu(entry->nameidx);
1584 
1585 	ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1586 					+ xfs_attr3_leaf_hdr_size(leaf));
1587 	tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1588 					+ xfs_attr3_leaf_hdr_size(leaf);
1589 
1590 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1591 		if (ichdr->freemap[i].base == tmp) {
1592 			ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1593 			ichdr->freemap[i].size -=
1594 				min_t(uint16_t, ichdr->freemap[i].size,
1595 						sizeof(xfs_attr_leaf_entry_t));
1596 		}
1597 	}
1598 	ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1599 	return 0;
1600 }
1601 
1602 /*
1603  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1604  */
1605 STATIC void
1606 xfs_attr3_leaf_compact(
1607 	struct xfs_da_args	*args,
1608 	struct xfs_attr3_icleaf_hdr *ichdr_dst,
1609 	struct xfs_buf		*bp)
1610 {
1611 	struct xfs_attr_leafblock *leaf_src;
1612 	struct xfs_attr_leafblock *leaf_dst;
1613 	struct xfs_attr3_icleaf_hdr ichdr_src;
1614 	struct xfs_trans	*trans = args->trans;
1615 	char			*tmpbuffer;
1616 
1617 	trace_xfs_attr_leaf_compact(args);
1618 
1619 	tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1620 	memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1621 	memset(bp->b_addr, 0, args->geo->blksize);
1622 	leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1623 	leaf_dst = bp->b_addr;
1624 
1625 	/*
1626 	 * Copy the on-disk header back into the destination buffer to ensure
1627 	 * all the information in the header that is not part of the incore
1628 	 * header structure is preserved.
1629 	 */
1630 	memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1631 
1632 	/* Initialise the incore headers */
1633 	ichdr_src = *ichdr_dst;	/* struct copy */
1634 	ichdr_dst->firstused = args->geo->blksize;
1635 	ichdr_dst->usedbytes = 0;
1636 	ichdr_dst->count = 0;
1637 	ichdr_dst->holes = 0;
1638 	ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1639 	ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1640 						ichdr_dst->freemap[0].base;
1641 
1642 	/* write the header back to initialise the underlying buffer */
1643 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1644 
1645 	/*
1646 	 * Copy all entry's in the same (sorted) order,
1647 	 * but allocate name/value pairs packed and in sequence.
1648 	 */
1649 	xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1650 				leaf_dst, ichdr_dst, 0, ichdr_src.count);
1651 	/*
1652 	 * this logs the entire buffer, but the caller must write the header
1653 	 * back to the buffer when it is finished modifying it.
1654 	 */
1655 	xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1656 
1657 	kmem_free(tmpbuffer);
1658 }
1659 
1660 /*
1661  * Compare two leaf blocks "order".
1662  * Return 0 unless leaf2 should go before leaf1.
1663  */
1664 static int
1665 xfs_attr3_leaf_order(
1666 	struct xfs_buf	*leaf1_bp,
1667 	struct xfs_attr3_icleaf_hdr *leaf1hdr,
1668 	struct xfs_buf	*leaf2_bp,
1669 	struct xfs_attr3_icleaf_hdr *leaf2hdr)
1670 {
1671 	struct xfs_attr_leaf_entry *entries1;
1672 	struct xfs_attr_leaf_entry *entries2;
1673 
1674 	entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1675 	entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1676 	if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1677 	    ((be32_to_cpu(entries2[0].hashval) <
1678 	      be32_to_cpu(entries1[0].hashval)) ||
1679 	     (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1680 	      be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1681 		return 1;
1682 	}
1683 	return 0;
1684 }
1685 
1686 int
1687 xfs_attr_leaf_order(
1688 	struct xfs_buf	*leaf1_bp,
1689 	struct xfs_buf	*leaf2_bp)
1690 {
1691 	struct xfs_attr3_icleaf_hdr ichdr1;
1692 	struct xfs_attr3_icleaf_hdr ichdr2;
1693 	struct xfs_mount *mp = leaf1_bp->b_mount;
1694 
1695 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1696 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1697 	return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1698 }
1699 
1700 /*
1701  * Redistribute the attribute list entries between two leaf nodes,
1702  * taking into account the size of the new entry.
1703  *
1704  * NOTE: if new block is empty, then it will get the upper half of the
1705  * old block.  At present, all (one) callers pass in an empty second block.
1706  *
1707  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1708  * to match what it is doing in splitting the attribute leaf block.  Those
1709  * values are used in "atomic rename" operations on attributes.  Note that
1710  * the "new" and "old" values can end up in different blocks.
1711  */
1712 STATIC void
1713 xfs_attr3_leaf_rebalance(
1714 	struct xfs_da_state	*state,
1715 	struct xfs_da_state_blk	*blk1,
1716 	struct xfs_da_state_blk	*blk2)
1717 {
1718 	struct xfs_da_args	*args;
1719 	struct xfs_attr_leafblock *leaf1;
1720 	struct xfs_attr_leafblock *leaf2;
1721 	struct xfs_attr3_icleaf_hdr ichdr1;
1722 	struct xfs_attr3_icleaf_hdr ichdr2;
1723 	struct xfs_attr_leaf_entry *entries1;
1724 	struct xfs_attr_leaf_entry *entries2;
1725 	int			count;
1726 	int			totallen;
1727 	int			max;
1728 	int			space;
1729 	int			swap;
1730 
1731 	/*
1732 	 * Set up environment.
1733 	 */
1734 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1735 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1736 	leaf1 = blk1->bp->b_addr;
1737 	leaf2 = blk2->bp->b_addr;
1738 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1739 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1740 	ASSERT(ichdr2.count == 0);
1741 	args = state->args;
1742 
1743 	trace_xfs_attr_leaf_rebalance(args);
1744 
1745 	/*
1746 	 * Check ordering of blocks, reverse if it makes things simpler.
1747 	 *
1748 	 * NOTE: Given that all (current) callers pass in an empty
1749 	 * second block, this code should never set "swap".
1750 	 */
1751 	swap = 0;
1752 	if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1753 		swap(blk1, blk2);
1754 
1755 		/* swap structures rather than reconverting them */
1756 		swap(ichdr1, ichdr2);
1757 
1758 		leaf1 = blk1->bp->b_addr;
1759 		leaf2 = blk2->bp->b_addr;
1760 		swap = 1;
1761 	}
1762 
1763 	/*
1764 	 * Examine entries until we reduce the absolute difference in
1765 	 * byte usage between the two blocks to a minimum.  Then get
1766 	 * the direction to copy and the number of elements to move.
1767 	 *
1768 	 * "inleaf" is true if the new entry should be inserted into blk1.
1769 	 * If "swap" is also true, then reverse the sense of "inleaf".
1770 	 */
1771 	state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1772 						      blk2, &ichdr2,
1773 						      &count, &totallen);
1774 	if (swap)
1775 		state->inleaf = !state->inleaf;
1776 
1777 	/*
1778 	 * Move any entries required from leaf to leaf:
1779 	 */
1780 	if (count < ichdr1.count) {
1781 		/*
1782 		 * Figure the total bytes to be added to the destination leaf.
1783 		 */
1784 		/* number entries being moved */
1785 		count = ichdr1.count - count;
1786 		space  = ichdr1.usedbytes - totallen;
1787 		space += count * sizeof(xfs_attr_leaf_entry_t);
1788 
1789 		/*
1790 		 * leaf2 is the destination, compact it if it looks tight.
1791 		 */
1792 		max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1793 		max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1794 		if (space > max)
1795 			xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1796 
1797 		/*
1798 		 * Move high entries from leaf1 to low end of leaf2.
1799 		 */
1800 		xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1801 				ichdr1.count - count, leaf2, &ichdr2, 0, count);
1802 
1803 	} else if (count > ichdr1.count) {
1804 		/*
1805 		 * I assert that since all callers pass in an empty
1806 		 * second buffer, this code should never execute.
1807 		 */
1808 		ASSERT(0);
1809 
1810 		/*
1811 		 * Figure the total bytes to be added to the destination leaf.
1812 		 */
1813 		/* number entries being moved */
1814 		count -= ichdr1.count;
1815 		space  = totallen - ichdr1.usedbytes;
1816 		space += count * sizeof(xfs_attr_leaf_entry_t);
1817 
1818 		/*
1819 		 * leaf1 is the destination, compact it if it looks tight.
1820 		 */
1821 		max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1822 		max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1823 		if (space > max)
1824 			xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1825 
1826 		/*
1827 		 * Move low entries from leaf2 to high end of leaf1.
1828 		 */
1829 		xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1830 					ichdr1.count, count);
1831 	}
1832 
1833 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1834 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1835 	xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1836 	xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1837 
1838 	/*
1839 	 * Copy out last hashval in each block for B-tree code.
1840 	 */
1841 	entries1 = xfs_attr3_leaf_entryp(leaf1);
1842 	entries2 = xfs_attr3_leaf_entryp(leaf2);
1843 	blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1844 	blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1845 
1846 	/*
1847 	 * Adjust the expected index for insertion.
1848 	 * NOTE: this code depends on the (current) situation that the
1849 	 * second block was originally empty.
1850 	 *
1851 	 * If the insertion point moved to the 2nd block, we must adjust
1852 	 * the index.  We must also track the entry just following the
1853 	 * new entry for use in an "atomic rename" operation, that entry
1854 	 * is always the "old" entry and the "new" entry is what we are
1855 	 * inserting.  The index/blkno fields refer to the "old" entry,
1856 	 * while the index2/blkno2 fields refer to the "new" entry.
1857 	 */
1858 	if (blk1->index > ichdr1.count) {
1859 		ASSERT(state->inleaf == 0);
1860 		blk2->index = blk1->index - ichdr1.count;
1861 		args->index = args->index2 = blk2->index;
1862 		args->blkno = args->blkno2 = blk2->blkno;
1863 	} else if (blk1->index == ichdr1.count) {
1864 		if (state->inleaf) {
1865 			args->index = blk1->index;
1866 			args->blkno = blk1->blkno;
1867 			args->index2 = 0;
1868 			args->blkno2 = blk2->blkno;
1869 		} else {
1870 			/*
1871 			 * On a double leaf split, the original attr location
1872 			 * is already stored in blkno2/index2, so don't
1873 			 * overwrite it overwise we corrupt the tree.
1874 			 */
1875 			blk2->index = blk1->index - ichdr1.count;
1876 			args->index = blk2->index;
1877 			args->blkno = blk2->blkno;
1878 			if (!state->extravalid) {
1879 				/*
1880 				 * set the new attr location to match the old
1881 				 * one and let the higher level split code
1882 				 * decide where in the leaf to place it.
1883 				 */
1884 				args->index2 = blk2->index;
1885 				args->blkno2 = blk2->blkno;
1886 			}
1887 		}
1888 	} else {
1889 		ASSERT(state->inleaf == 1);
1890 		args->index = args->index2 = blk1->index;
1891 		args->blkno = args->blkno2 = blk1->blkno;
1892 	}
1893 }
1894 
1895 /*
1896  * Examine entries until we reduce the absolute difference in
1897  * byte usage between the two blocks to a minimum.
1898  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1899  * GROT: there will always be enough room in either block for a new entry.
1900  * GROT: Do a double-split for this case?
1901  */
1902 STATIC int
1903 xfs_attr3_leaf_figure_balance(
1904 	struct xfs_da_state		*state,
1905 	struct xfs_da_state_blk		*blk1,
1906 	struct xfs_attr3_icleaf_hdr	*ichdr1,
1907 	struct xfs_da_state_blk		*blk2,
1908 	struct xfs_attr3_icleaf_hdr	*ichdr2,
1909 	int				*countarg,
1910 	int				*usedbytesarg)
1911 {
1912 	struct xfs_attr_leafblock	*leaf1 = blk1->bp->b_addr;
1913 	struct xfs_attr_leafblock	*leaf2 = blk2->bp->b_addr;
1914 	struct xfs_attr_leaf_entry	*entry;
1915 	int				count;
1916 	int				max;
1917 	int				index;
1918 	int				totallen = 0;
1919 	int				half;
1920 	int				lastdelta;
1921 	int				foundit = 0;
1922 	int				tmp;
1923 
1924 	/*
1925 	 * Examine entries until we reduce the absolute difference in
1926 	 * byte usage between the two blocks to a minimum.
1927 	 */
1928 	max = ichdr1->count + ichdr2->count;
1929 	half = (max + 1) * sizeof(*entry);
1930 	half += ichdr1->usedbytes + ichdr2->usedbytes +
1931 			xfs_attr_leaf_newentsize(state->args, NULL);
1932 	half /= 2;
1933 	lastdelta = state->args->geo->blksize;
1934 	entry = xfs_attr3_leaf_entryp(leaf1);
1935 	for (count = index = 0; count < max; entry++, index++, count++) {
1936 
1937 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1938 		/*
1939 		 * The new entry is in the first block, account for it.
1940 		 */
1941 		if (count == blk1->index) {
1942 			tmp = totallen + sizeof(*entry) +
1943 				xfs_attr_leaf_newentsize(state->args, NULL);
1944 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1945 				break;
1946 			lastdelta = XFS_ATTR_ABS(half - tmp);
1947 			totallen = tmp;
1948 			foundit = 1;
1949 		}
1950 
1951 		/*
1952 		 * Wrap around into the second block if necessary.
1953 		 */
1954 		if (count == ichdr1->count) {
1955 			leaf1 = leaf2;
1956 			entry = xfs_attr3_leaf_entryp(leaf1);
1957 			index = 0;
1958 		}
1959 
1960 		/*
1961 		 * Figure out if next leaf entry would be too much.
1962 		 */
1963 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1964 									index);
1965 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1966 			break;
1967 		lastdelta = XFS_ATTR_ABS(half - tmp);
1968 		totallen = tmp;
1969 #undef XFS_ATTR_ABS
1970 	}
1971 
1972 	/*
1973 	 * Calculate the number of usedbytes that will end up in lower block.
1974 	 * If new entry not in lower block, fix up the count.
1975 	 */
1976 	totallen -= count * sizeof(*entry);
1977 	if (foundit) {
1978 		totallen -= sizeof(*entry) +
1979 				xfs_attr_leaf_newentsize(state->args, NULL);
1980 	}
1981 
1982 	*countarg = count;
1983 	*usedbytesarg = totallen;
1984 	return foundit;
1985 }
1986 
1987 /*========================================================================
1988  * Routines used for shrinking the Btree.
1989  *========================================================================*/
1990 
1991 /*
1992  * Check a leaf block and its neighbors to see if the block should be
1993  * collapsed into one or the other neighbor.  Always keep the block
1994  * with the smaller block number.
1995  * If the current block is over 50% full, don't try to join it, return 0.
1996  * If the block is empty, fill in the state structure and return 2.
1997  * If it can be collapsed, fill in the state structure and return 1.
1998  * If nothing can be done, return 0.
1999  *
2000  * GROT: allow for INCOMPLETE entries in calculation.
2001  */
2002 int
2003 xfs_attr3_leaf_toosmall(
2004 	struct xfs_da_state	*state,
2005 	int			*action)
2006 {
2007 	struct xfs_attr_leafblock *leaf;
2008 	struct xfs_da_state_blk	*blk;
2009 	struct xfs_attr3_icleaf_hdr ichdr;
2010 	struct xfs_buf		*bp;
2011 	xfs_dablk_t		blkno;
2012 	int			bytes;
2013 	int			forward;
2014 	int			error;
2015 	int			retval;
2016 	int			i;
2017 
2018 	trace_xfs_attr_leaf_toosmall(state->args);
2019 
2020 	/*
2021 	 * Check for the degenerate case of the block being over 50% full.
2022 	 * If so, it's not worth even looking to see if we might be able
2023 	 * to coalesce with a sibling.
2024 	 */
2025 	blk = &state->path.blk[ state->path.active-1 ];
2026 	leaf = blk->bp->b_addr;
2027 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
2028 	bytes = xfs_attr3_leaf_hdr_size(leaf) +
2029 		ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2030 		ichdr.usedbytes;
2031 	if (bytes > (state->args->geo->blksize >> 1)) {
2032 		*action = 0;	/* blk over 50%, don't try to join */
2033 		return 0;
2034 	}
2035 
2036 	/*
2037 	 * Check for the degenerate case of the block being empty.
2038 	 * If the block is empty, we'll simply delete it, no need to
2039 	 * coalesce it with a sibling block.  We choose (arbitrarily)
2040 	 * to merge with the forward block unless it is NULL.
2041 	 */
2042 	if (ichdr.count == 0) {
2043 		/*
2044 		 * Make altpath point to the block we want to keep and
2045 		 * path point to the block we want to drop (this one).
2046 		 */
2047 		forward = (ichdr.forw != 0);
2048 		memcpy(&state->altpath, &state->path, sizeof(state->path));
2049 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2050 						 0, &retval);
2051 		if (error)
2052 			return error;
2053 		if (retval) {
2054 			*action = 0;
2055 		} else {
2056 			*action = 2;
2057 		}
2058 		return 0;
2059 	}
2060 
2061 	/*
2062 	 * Examine each sibling block to see if we can coalesce with
2063 	 * at least 25% free space to spare.  We need to figure out
2064 	 * whether to merge with the forward or the backward block.
2065 	 * We prefer coalescing with the lower numbered sibling so as
2066 	 * to shrink an attribute list over time.
2067 	 */
2068 	/* start with smaller blk num */
2069 	forward = ichdr.forw < ichdr.back;
2070 	for (i = 0; i < 2; forward = !forward, i++) {
2071 		struct xfs_attr3_icleaf_hdr ichdr2;
2072 		if (forward)
2073 			blkno = ichdr.forw;
2074 		else
2075 			blkno = ichdr.back;
2076 		if (blkno == 0)
2077 			continue;
2078 		error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2079 					blkno, &bp);
2080 		if (error)
2081 			return error;
2082 
2083 		xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2084 
2085 		bytes = state->args->geo->blksize -
2086 			(state->args->geo->blksize >> 2) -
2087 			ichdr.usedbytes - ichdr2.usedbytes -
2088 			((ichdr.count + ichdr2.count) *
2089 					sizeof(xfs_attr_leaf_entry_t)) -
2090 			xfs_attr3_leaf_hdr_size(leaf);
2091 
2092 		xfs_trans_brelse(state->args->trans, bp);
2093 		if (bytes >= 0)
2094 			break;	/* fits with at least 25% to spare */
2095 	}
2096 	if (i >= 2) {
2097 		*action = 0;
2098 		return 0;
2099 	}
2100 
2101 	/*
2102 	 * Make altpath point to the block we want to keep (the lower
2103 	 * numbered block) and path point to the block we want to drop.
2104 	 */
2105 	memcpy(&state->altpath, &state->path, sizeof(state->path));
2106 	if (blkno < blk->blkno) {
2107 		error = xfs_da3_path_shift(state, &state->altpath, forward,
2108 						 0, &retval);
2109 	} else {
2110 		error = xfs_da3_path_shift(state, &state->path, forward,
2111 						 0, &retval);
2112 	}
2113 	if (error)
2114 		return error;
2115 	if (retval) {
2116 		*action = 0;
2117 	} else {
2118 		*action = 1;
2119 	}
2120 	return 0;
2121 }
2122 
2123 /*
2124  * Remove a name from the leaf attribute list structure.
2125  *
2126  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2127  * If two leaves are 37% full, when combined they will leave 25% free.
2128  */
2129 int
2130 xfs_attr3_leaf_remove(
2131 	struct xfs_buf		*bp,
2132 	struct xfs_da_args	*args)
2133 {
2134 	struct xfs_attr_leafblock *leaf;
2135 	struct xfs_attr3_icleaf_hdr ichdr;
2136 	struct xfs_attr_leaf_entry *entry;
2137 	int			before;
2138 	int			after;
2139 	int			smallest;
2140 	int			entsize;
2141 	int			tablesize;
2142 	int			tmp;
2143 	int			i;
2144 
2145 	trace_xfs_attr_leaf_remove(args);
2146 
2147 	leaf = bp->b_addr;
2148 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2149 
2150 	ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2151 	ASSERT(args->index >= 0 && args->index < ichdr.count);
2152 	ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2153 					xfs_attr3_leaf_hdr_size(leaf));
2154 
2155 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2156 
2157 	ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2158 	ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2159 
2160 	/*
2161 	 * Scan through free region table:
2162 	 *    check for adjacency of free'd entry with an existing one,
2163 	 *    find smallest free region in case we need to replace it,
2164 	 *    adjust any map that borders the entry table,
2165 	 */
2166 	tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2167 					+ xfs_attr3_leaf_hdr_size(leaf);
2168 	tmp = ichdr.freemap[0].size;
2169 	before = after = -1;
2170 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2171 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
2172 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2173 		ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2174 		ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2175 		if (ichdr.freemap[i].base == tablesize) {
2176 			ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2177 			ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2178 		}
2179 
2180 		if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2181 				be16_to_cpu(entry->nameidx)) {
2182 			before = i;
2183 		} else if (ichdr.freemap[i].base ==
2184 				(be16_to_cpu(entry->nameidx) + entsize)) {
2185 			after = i;
2186 		} else if (ichdr.freemap[i].size < tmp) {
2187 			tmp = ichdr.freemap[i].size;
2188 			smallest = i;
2189 		}
2190 	}
2191 
2192 	/*
2193 	 * Coalesce adjacent freemap regions,
2194 	 * or replace the smallest region.
2195 	 */
2196 	if ((before >= 0) || (after >= 0)) {
2197 		if ((before >= 0) && (after >= 0)) {
2198 			ichdr.freemap[before].size += entsize;
2199 			ichdr.freemap[before].size += ichdr.freemap[after].size;
2200 			ichdr.freemap[after].base = 0;
2201 			ichdr.freemap[after].size = 0;
2202 		} else if (before >= 0) {
2203 			ichdr.freemap[before].size += entsize;
2204 		} else {
2205 			ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2206 			ichdr.freemap[after].size += entsize;
2207 		}
2208 	} else {
2209 		/*
2210 		 * Replace smallest region (if it is smaller than free'd entry)
2211 		 */
2212 		if (ichdr.freemap[smallest].size < entsize) {
2213 			ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2214 			ichdr.freemap[smallest].size = entsize;
2215 		}
2216 	}
2217 
2218 	/*
2219 	 * Did we remove the first entry?
2220 	 */
2221 	if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2222 		smallest = 1;
2223 	else
2224 		smallest = 0;
2225 
2226 	/*
2227 	 * Compress the remaining entries and zero out the removed stuff.
2228 	 */
2229 	memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2230 	ichdr.usedbytes -= entsize;
2231 	xfs_trans_log_buf(args->trans, bp,
2232 	     XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2233 				   entsize));
2234 
2235 	tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2236 	memmove(entry, entry + 1, tmp);
2237 	ichdr.count--;
2238 	xfs_trans_log_buf(args->trans, bp,
2239 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2240 
2241 	entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2242 	memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2243 
2244 	/*
2245 	 * If we removed the first entry, re-find the first used byte
2246 	 * in the name area.  Note that if the entry was the "firstused",
2247 	 * then we don't have a "hole" in our block resulting from
2248 	 * removing the name.
2249 	 */
2250 	if (smallest) {
2251 		tmp = args->geo->blksize;
2252 		entry = xfs_attr3_leaf_entryp(leaf);
2253 		for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2254 			ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2255 			ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2256 
2257 			if (be16_to_cpu(entry->nameidx) < tmp)
2258 				tmp = be16_to_cpu(entry->nameidx);
2259 		}
2260 		ichdr.firstused = tmp;
2261 		ASSERT(ichdr.firstused != 0);
2262 	} else {
2263 		ichdr.holes = 1;	/* mark as needing compaction */
2264 	}
2265 	xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2266 	xfs_trans_log_buf(args->trans, bp,
2267 			  XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2268 					  xfs_attr3_leaf_hdr_size(leaf)));
2269 
2270 	/*
2271 	 * Check if leaf is less than 50% full, caller may want to
2272 	 * "join" the leaf with a sibling if so.
2273 	 */
2274 	tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2275 	      ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2276 
2277 	return tmp < args->geo->magicpct; /* leaf is < 37% full */
2278 }
2279 
2280 /*
2281  * Move all the attribute list entries from drop_leaf into save_leaf.
2282  */
2283 void
2284 xfs_attr3_leaf_unbalance(
2285 	struct xfs_da_state	*state,
2286 	struct xfs_da_state_blk	*drop_blk,
2287 	struct xfs_da_state_blk	*save_blk)
2288 {
2289 	struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2290 	struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2291 	struct xfs_attr3_icleaf_hdr drophdr;
2292 	struct xfs_attr3_icleaf_hdr savehdr;
2293 	struct xfs_attr_leaf_entry *entry;
2294 
2295 	trace_xfs_attr_leaf_unbalance(state->args);
2296 
2297 	drop_leaf = drop_blk->bp->b_addr;
2298 	save_leaf = save_blk->bp->b_addr;
2299 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2300 	xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2301 	entry = xfs_attr3_leaf_entryp(drop_leaf);
2302 
2303 	/*
2304 	 * Save last hashval from dying block for later Btree fixup.
2305 	 */
2306 	drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2307 
2308 	/*
2309 	 * Check if we need a temp buffer, or can we do it in place.
2310 	 * Note that we don't check "leaf" for holes because we will
2311 	 * always be dropping it, toosmall() decided that for us already.
2312 	 */
2313 	if (savehdr.holes == 0) {
2314 		/*
2315 		 * dest leaf has no holes, so we add there.  May need
2316 		 * to make some room in the entry array.
2317 		 */
2318 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2319 					 drop_blk->bp, &drophdr)) {
2320 			xfs_attr3_leaf_moveents(state->args,
2321 						drop_leaf, &drophdr, 0,
2322 						save_leaf, &savehdr, 0,
2323 						drophdr.count);
2324 		} else {
2325 			xfs_attr3_leaf_moveents(state->args,
2326 						drop_leaf, &drophdr, 0,
2327 						save_leaf, &savehdr,
2328 						savehdr.count, drophdr.count);
2329 		}
2330 	} else {
2331 		/*
2332 		 * Destination has holes, so we make a temporary copy
2333 		 * of the leaf and add them both to that.
2334 		 */
2335 		struct xfs_attr_leafblock *tmp_leaf;
2336 		struct xfs_attr3_icleaf_hdr tmphdr;
2337 
2338 		tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2339 
2340 		/*
2341 		 * Copy the header into the temp leaf so that all the stuff
2342 		 * not in the incore header is present and gets copied back in
2343 		 * once we've moved all the entries.
2344 		 */
2345 		memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2346 
2347 		memset(&tmphdr, 0, sizeof(tmphdr));
2348 		tmphdr.magic = savehdr.magic;
2349 		tmphdr.forw = savehdr.forw;
2350 		tmphdr.back = savehdr.back;
2351 		tmphdr.firstused = state->args->geo->blksize;
2352 
2353 		/* write the header to the temp buffer to initialise it */
2354 		xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2355 
2356 		if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2357 					 drop_blk->bp, &drophdr)) {
2358 			xfs_attr3_leaf_moveents(state->args,
2359 						drop_leaf, &drophdr, 0,
2360 						tmp_leaf, &tmphdr, 0,
2361 						drophdr.count);
2362 			xfs_attr3_leaf_moveents(state->args,
2363 						save_leaf, &savehdr, 0,
2364 						tmp_leaf, &tmphdr, tmphdr.count,
2365 						savehdr.count);
2366 		} else {
2367 			xfs_attr3_leaf_moveents(state->args,
2368 						save_leaf, &savehdr, 0,
2369 						tmp_leaf, &tmphdr, 0,
2370 						savehdr.count);
2371 			xfs_attr3_leaf_moveents(state->args,
2372 						drop_leaf, &drophdr, 0,
2373 						tmp_leaf, &tmphdr, tmphdr.count,
2374 						drophdr.count);
2375 		}
2376 		memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2377 		savehdr = tmphdr; /* struct copy */
2378 		kmem_free(tmp_leaf);
2379 	}
2380 
2381 	xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2382 	xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2383 					   state->args->geo->blksize - 1);
2384 
2385 	/*
2386 	 * Copy out last hashval in each block for B-tree code.
2387 	 */
2388 	entry = xfs_attr3_leaf_entryp(save_leaf);
2389 	save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2390 }
2391 
2392 /*========================================================================
2393  * Routines used for finding things in the Btree.
2394  *========================================================================*/
2395 
2396 /*
2397  * Look up a name in a leaf attribute list structure.
2398  * This is the internal routine, it uses the caller's buffer.
2399  *
2400  * Note that duplicate keys are allowed, but only check within the
2401  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2402  *
2403  * Return in args->index the index into the entry[] array of either
2404  * the found entry, or where the entry should have been (insert before
2405  * that entry).
2406  *
2407  * Don't change the args->value unless we find the attribute.
2408  */
2409 int
2410 xfs_attr3_leaf_lookup_int(
2411 	struct xfs_buf		*bp,
2412 	struct xfs_da_args	*args)
2413 {
2414 	struct xfs_attr_leafblock *leaf;
2415 	struct xfs_attr3_icleaf_hdr ichdr;
2416 	struct xfs_attr_leaf_entry *entry;
2417 	struct xfs_attr_leaf_entry *entries;
2418 	struct xfs_attr_leaf_name_local *name_loc;
2419 	struct xfs_attr_leaf_name_remote *name_rmt;
2420 	xfs_dahash_t		hashval;
2421 	int			probe;
2422 	int			span;
2423 
2424 	trace_xfs_attr_leaf_lookup(args);
2425 
2426 	leaf = bp->b_addr;
2427 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2428 	entries = xfs_attr3_leaf_entryp(leaf);
2429 	if (ichdr.count >= args->geo->blksize / 8) {
2430 		xfs_buf_mark_corrupt(bp);
2431 		return -EFSCORRUPTED;
2432 	}
2433 
2434 	/*
2435 	 * Binary search.  (note: small blocks will skip this loop)
2436 	 */
2437 	hashval = args->hashval;
2438 	probe = span = ichdr.count / 2;
2439 	for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2440 		span /= 2;
2441 		if (be32_to_cpu(entry->hashval) < hashval)
2442 			probe += span;
2443 		else if (be32_to_cpu(entry->hashval) > hashval)
2444 			probe -= span;
2445 		else
2446 			break;
2447 	}
2448 	if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2449 		xfs_buf_mark_corrupt(bp);
2450 		return -EFSCORRUPTED;
2451 	}
2452 	if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2453 		xfs_buf_mark_corrupt(bp);
2454 		return -EFSCORRUPTED;
2455 	}
2456 
2457 	/*
2458 	 * Since we may have duplicate hashval's, find the first matching
2459 	 * hashval in the leaf.
2460 	 */
2461 	while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2462 		entry--;
2463 		probe--;
2464 	}
2465 	while (probe < ichdr.count &&
2466 	       be32_to_cpu(entry->hashval) < hashval) {
2467 		entry++;
2468 		probe++;
2469 	}
2470 	if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2471 		args->index = probe;
2472 		return -ENOATTR;
2473 	}
2474 
2475 	/*
2476 	 * Duplicate keys may be present, so search all of them for a match.
2477 	 */
2478 	for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2479 			entry++, probe++) {
2480 /*
2481  * GROT: Add code to remove incomplete entries.
2482  */
2483 		if (entry->flags & XFS_ATTR_LOCAL) {
2484 			name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2485 			if (!xfs_attr_match(args, name_loc->namelen,
2486 					name_loc->nameval, entry->flags))
2487 				continue;
2488 			args->index = probe;
2489 			return -EEXIST;
2490 		} else {
2491 			name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2492 			if (!xfs_attr_match(args, name_rmt->namelen,
2493 					name_rmt->name, entry->flags))
2494 				continue;
2495 			args->index = probe;
2496 			args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2497 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2498 			args->rmtblkcnt = xfs_attr3_rmt_blocks(
2499 							args->dp->i_mount,
2500 							args->rmtvaluelen);
2501 			return -EEXIST;
2502 		}
2503 	}
2504 	args->index = probe;
2505 	return -ENOATTR;
2506 }
2507 
2508 /*
2509  * Get the value associated with an attribute name from a leaf attribute
2510  * list structure.
2511  *
2512  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2513  * lookup, we only return an error if the attribute does not exist or we can't
2514  * retrieve the value.
2515  */
2516 int
2517 xfs_attr3_leaf_getvalue(
2518 	struct xfs_buf		*bp,
2519 	struct xfs_da_args	*args)
2520 {
2521 	struct xfs_attr_leafblock *leaf;
2522 	struct xfs_attr3_icleaf_hdr ichdr;
2523 	struct xfs_attr_leaf_entry *entry;
2524 	struct xfs_attr_leaf_name_local *name_loc;
2525 	struct xfs_attr_leaf_name_remote *name_rmt;
2526 
2527 	leaf = bp->b_addr;
2528 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2529 	ASSERT(ichdr.count < args->geo->blksize / 8);
2530 	ASSERT(args->index < ichdr.count);
2531 
2532 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2533 	if (entry->flags & XFS_ATTR_LOCAL) {
2534 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2535 		ASSERT(name_loc->namelen == args->namelen);
2536 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2537 		return xfs_attr_copy_value(args,
2538 					&name_loc->nameval[args->namelen],
2539 					be16_to_cpu(name_loc->valuelen));
2540 	}
2541 
2542 	name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2543 	ASSERT(name_rmt->namelen == args->namelen);
2544 	ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2545 	args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2546 	args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2547 	args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2548 					       args->rmtvaluelen);
2549 	return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2550 }
2551 
2552 /*========================================================================
2553  * Utility routines.
2554  *========================================================================*/
2555 
2556 /*
2557  * Move the indicated entries from one leaf to another.
2558  * NOTE: this routine modifies both source and destination leaves.
2559  */
2560 /*ARGSUSED*/
2561 STATIC void
2562 xfs_attr3_leaf_moveents(
2563 	struct xfs_da_args		*args,
2564 	struct xfs_attr_leafblock	*leaf_s,
2565 	struct xfs_attr3_icleaf_hdr	*ichdr_s,
2566 	int				start_s,
2567 	struct xfs_attr_leafblock	*leaf_d,
2568 	struct xfs_attr3_icleaf_hdr	*ichdr_d,
2569 	int				start_d,
2570 	int				count)
2571 {
2572 	struct xfs_attr_leaf_entry	*entry_s;
2573 	struct xfs_attr_leaf_entry	*entry_d;
2574 	int				desti;
2575 	int				tmp;
2576 	int				i;
2577 
2578 	/*
2579 	 * Check for nothing to do.
2580 	 */
2581 	if (count == 0)
2582 		return;
2583 
2584 	/*
2585 	 * Set up environment.
2586 	 */
2587 	ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2588 	       ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2589 	ASSERT(ichdr_s->magic == ichdr_d->magic);
2590 	ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2591 	ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2592 					+ xfs_attr3_leaf_hdr_size(leaf_s));
2593 	ASSERT(ichdr_d->count < args->geo->blksize / 8);
2594 	ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2595 					+ xfs_attr3_leaf_hdr_size(leaf_d));
2596 
2597 	ASSERT(start_s < ichdr_s->count);
2598 	ASSERT(start_d <= ichdr_d->count);
2599 	ASSERT(count <= ichdr_s->count);
2600 
2601 
2602 	/*
2603 	 * Move the entries in the destination leaf up to make a hole?
2604 	 */
2605 	if (start_d < ichdr_d->count) {
2606 		tmp  = ichdr_d->count - start_d;
2607 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2608 		entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2609 		entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2610 		memmove(entry_d, entry_s, tmp);
2611 	}
2612 
2613 	/*
2614 	 * Copy all entry's in the same (sorted) order,
2615 	 * but allocate attribute info packed and in sequence.
2616 	 */
2617 	entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2618 	entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2619 	desti = start_d;
2620 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2621 		ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2622 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2623 #ifdef GROT
2624 		/*
2625 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2626 		 * may also need to change the insertion index.  Code turned
2627 		 * off for 6.2, should be revisited later.
2628 		 */
2629 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2630 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2631 			ichdr_s->usedbytes -= tmp;
2632 			ichdr_s->count -= 1;
2633 			entry_d--;	/* to compensate for ++ in loop hdr */
2634 			desti--;
2635 			if ((start_s + i) < offset)
2636 				result++;	/* insertion index adjustment */
2637 		} else {
2638 #endif /* GROT */
2639 			ichdr_d->firstused -= tmp;
2640 			/* both on-disk, don't endian flip twice */
2641 			entry_d->hashval = entry_s->hashval;
2642 			entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2643 			entry_d->flags = entry_s->flags;
2644 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2645 							<= args->geo->blksize);
2646 			memmove(xfs_attr3_leaf_name(leaf_d, desti),
2647 				xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2648 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2649 							<= args->geo->blksize);
2650 			memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2651 			ichdr_s->usedbytes -= tmp;
2652 			ichdr_d->usedbytes += tmp;
2653 			ichdr_s->count -= 1;
2654 			ichdr_d->count += 1;
2655 			tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2656 					+ xfs_attr3_leaf_hdr_size(leaf_d);
2657 			ASSERT(ichdr_d->firstused >= tmp);
2658 #ifdef GROT
2659 		}
2660 #endif /* GROT */
2661 	}
2662 
2663 	/*
2664 	 * Zero out the entries we just copied.
2665 	 */
2666 	if (start_s == ichdr_s->count) {
2667 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2668 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2669 		ASSERT(((char *)entry_s + tmp) <=
2670 		       ((char *)leaf_s + args->geo->blksize));
2671 		memset(entry_s, 0, tmp);
2672 	} else {
2673 		/*
2674 		 * Move the remaining entries down to fill the hole,
2675 		 * then zero the entries at the top.
2676 		 */
2677 		tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2678 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2679 		entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2680 		memmove(entry_d, entry_s, tmp);
2681 
2682 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2683 		entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2684 		ASSERT(((char *)entry_s + tmp) <=
2685 		       ((char *)leaf_s + args->geo->blksize));
2686 		memset(entry_s, 0, tmp);
2687 	}
2688 
2689 	/*
2690 	 * Fill in the freemap information
2691 	 */
2692 	ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2693 	ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2694 	ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2695 	ichdr_d->freemap[1].base = 0;
2696 	ichdr_d->freemap[2].base = 0;
2697 	ichdr_d->freemap[1].size = 0;
2698 	ichdr_d->freemap[2].size = 0;
2699 	ichdr_s->holes = 1;	/* leaf may not be compact */
2700 }
2701 
2702 /*
2703  * Pick up the last hashvalue from a leaf block.
2704  */
2705 xfs_dahash_t
2706 xfs_attr_leaf_lasthash(
2707 	struct xfs_buf	*bp,
2708 	int		*count)
2709 {
2710 	struct xfs_attr3_icleaf_hdr ichdr;
2711 	struct xfs_attr_leaf_entry *entries;
2712 	struct xfs_mount *mp = bp->b_mount;
2713 
2714 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2715 	entries = xfs_attr3_leaf_entryp(bp->b_addr);
2716 	if (count)
2717 		*count = ichdr.count;
2718 	if (!ichdr.count)
2719 		return 0;
2720 	return be32_to_cpu(entries[ichdr.count - 1].hashval);
2721 }
2722 
2723 /*
2724  * Calculate the number of bytes used to store the indicated attribute
2725  * (whether local or remote only calculate bytes in this block).
2726  */
2727 STATIC int
2728 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2729 {
2730 	struct xfs_attr_leaf_entry *entries;
2731 	xfs_attr_leaf_name_local_t *name_loc;
2732 	xfs_attr_leaf_name_remote_t *name_rmt;
2733 	int size;
2734 
2735 	entries = xfs_attr3_leaf_entryp(leaf);
2736 	if (entries[index].flags & XFS_ATTR_LOCAL) {
2737 		name_loc = xfs_attr3_leaf_name_local(leaf, index);
2738 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2739 						   be16_to_cpu(name_loc->valuelen));
2740 	} else {
2741 		name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2742 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2743 	}
2744 	return size;
2745 }
2746 
2747 /*
2748  * Calculate the number of bytes that would be required to store the new
2749  * attribute (whether local or remote only calculate bytes in this block).
2750  * This routine decides as a side effect whether the attribute will be
2751  * a "local" or a "remote" attribute.
2752  */
2753 int
2754 xfs_attr_leaf_newentsize(
2755 	struct xfs_da_args	*args,
2756 	int			*local)
2757 {
2758 	int			size;
2759 
2760 	size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2761 	if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2762 		if (local)
2763 			*local = 1;
2764 		return size;
2765 	}
2766 	if (local)
2767 		*local = 0;
2768 	return xfs_attr_leaf_entsize_remote(args->namelen);
2769 }
2770 
2771 
2772 /*========================================================================
2773  * Manage the INCOMPLETE flag in a leaf entry
2774  *========================================================================*/
2775 
2776 /*
2777  * Clear the INCOMPLETE flag on an entry in a leaf block.
2778  */
2779 int
2780 xfs_attr3_leaf_clearflag(
2781 	struct xfs_da_args	*args)
2782 {
2783 	struct xfs_attr_leafblock *leaf;
2784 	struct xfs_attr_leaf_entry *entry;
2785 	struct xfs_attr_leaf_name_remote *name_rmt;
2786 	struct xfs_buf		*bp;
2787 	int			error;
2788 #ifdef DEBUG
2789 	struct xfs_attr3_icleaf_hdr ichdr;
2790 	xfs_attr_leaf_name_local_t *name_loc;
2791 	int namelen;
2792 	char *name;
2793 #endif /* DEBUG */
2794 
2795 	trace_xfs_attr_leaf_clearflag(args);
2796 	/*
2797 	 * Set up the operation.
2798 	 */
2799 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2800 	if (error)
2801 		return error;
2802 
2803 	leaf = bp->b_addr;
2804 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2805 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2806 
2807 #ifdef DEBUG
2808 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2809 	ASSERT(args->index < ichdr.count);
2810 	ASSERT(args->index >= 0);
2811 
2812 	if (entry->flags & XFS_ATTR_LOCAL) {
2813 		name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2814 		namelen = name_loc->namelen;
2815 		name = (char *)name_loc->nameval;
2816 	} else {
2817 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2818 		namelen = name_rmt->namelen;
2819 		name = (char *)name_rmt->name;
2820 	}
2821 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2822 	ASSERT(namelen == args->namelen);
2823 	ASSERT(memcmp(name, args->name, namelen) == 0);
2824 #endif /* DEBUG */
2825 
2826 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2827 	xfs_trans_log_buf(args->trans, bp,
2828 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2829 
2830 	if (args->rmtblkno) {
2831 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2832 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2833 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2834 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2835 		xfs_trans_log_buf(args->trans, bp,
2836 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2837 	}
2838 
2839 	return 0;
2840 }
2841 
2842 /*
2843  * Set the INCOMPLETE flag on an entry in a leaf block.
2844  */
2845 int
2846 xfs_attr3_leaf_setflag(
2847 	struct xfs_da_args	*args)
2848 {
2849 	struct xfs_attr_leafblock *leaf;
2850 	struct xfs_attr_leaf_entry *entry;
2851 	struct xfs_attr_leaf_name_remote *name_rmt;
2852 	struct xfs_buf		*bp;
2853 	int error;
2854 #ifdef DEBUG
2855 	struct xfs_attr3_icleaf_hdr ichdr;
2856 #endif
2857 
2858 	trace_xfs_attr_leaf_setflag(args);
2859 
2860 	/*
2861 	 * Set up the operation.
2862 	 */
2863 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2864 	if (error)
2865 		return error;
2866 
2867 	leaf = bp->b_addr;
2868 #ifdef DEBUG
2869 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2870 	ASSERT(args->index < ichdr.count);
2871 	ASSERT(args->index >= 0);
2872 #endif
2873 	entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2874 
2875 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2876 	entry->flags |= XFS_ATTR_INCOMPLETE;
2877 	xfs_trans_log_buf(args->trans, bp,
2878 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2879 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2880 		name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2881 		name_rmt->valueblk = 0;
2882 		name_rmt->valuelen = 0;
2883 		xfs_trans_log_buf(args->trans, bp,
2884 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2885 	}
2886 
2887 	return 0;
2888 }
2889 
2890 /*
2891  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2892  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2893  * entry given by args->blkno2/index2.
2894  *
2895  * Note that they could be in different blocks, or in the same block.
2896  */
2897 int
2898 xfs_attr3_leaf_flipflags(
2899 	struct xfs_da_args	*args)
2900 {
2901 	struct xfs_attr_leafblock *leaf1;
2902 	struct xfs_attr_leafblock *leaf2;
2903 	struct xfs_attr_leaf_entry *entry1;
2904 	struct xfs_attr_leaf_entry *entry2;
2905 	struct xfs_attr_leaf_name_remote *name_rmt;
2906 	struct xfs_buf		*bp1;
2907 	struct xfs_buf		*bp2;
2908 	int error;
2909 #ifdef DEBUG
2910 	struct xfs_attr3_icleaf_hdr ichdr1;
2911 	struct xfs_attr3_icleaf_hdr ichdr2;
2912 	xfs_attr_leaf_name_local_t *name_loc;
2913 	int namelen1, namelen2;
2914 	char *name1, *name2;
2915 #endif /* DEBUG */
2916 
2917 	trace_xfs_attr_leaf_flipflags(args);
2918 
2919 	/*
2920 	 * Read the block containing the "old" attr
2921 	 */
2922 	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2923 	if (error)
2924 		return error;
2925 
2926 	/*
2927 	 * Read the block containing the "new" attr, if it is different
2928 	 */
2929 	if (args->blkno2 != args->blkno) {
2930 		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2931 					   &bp2);
2932 		if (error)
2933 			return error;
2934 	} else {
2935 		bp2 = bp1;
2936 	}
2937 
2938 	leaf1 = bp1->b_addr;
2939 	entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2940 
2941 	leaf2 = bp2->b_addr;
2942 	entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2943 
2944 #ifdef DEBUG
2945 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2946 	ASSERT(args->index < ichdr1.count);
2947 	ASSERT(args->index >= 0);
2948 
2949 	xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2950 	ASSERT(args->index2 < ichdr2.count);
2951 	ASSERT(args->index2 >= 0);
2952 
2953 	if (entry1->flags & XFS_ATTR_LOCAL) {
2954 		name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2955 		namelen1 = name_loc->namelen;
2956 		name1 = (char *)name_loc->nameval;
2957 	} else {
2958 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2959 		namelen1 = name_rmt->namelen;
2960 		name1 = (char *)name_rmt->name;
2961 	}
2962 	if (entry2->flags & XFS_ATTR_LOCAL) {
2963 		name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2964 		namelen2 = name_loc->namelen;
2965 		name2 = (char *)name_loc->nameval;
2966 	} else {
2967 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2968 		namelen2 = name_rmt->namelen;
2969 		name2 = (char *)name_rmt->name;
2970 	}
2971 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2972 	ASSERT(namelen1 == namelen2);
2973 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2974 #endif /* DEBUG */
2975 
2976 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2977 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2978 
2979 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2980 	xfs_trans_log_buf(args->trans, bp1,
2981 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2982 	if (args->rmtblkno) {
2983 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2984 		name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2985 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2986 		name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2987 		xfs_trans_log_buf(args->trans, bp1,
2988 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2989 	}
2990 
2991 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2992 	xfs_trans_log_buf(args->trans, bp2,
2993 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2994 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2995 		name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2996 		name_rmt->valueblk = 0;
2997 		name_rmt->valuelen = 0;
2998 		xfs_trans_log_buf(args->trans, bp2,
2999 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
3000 	}
3001 
3002 	return 0;
3003 }
3004